]>
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> | |
1da177e4 | 73 | |
5517d86b | 74 | #include <asm/tlb.h> |
838225b4 | 75 | #include <asm/irq_regs.h> |
1da177e4 | 76 | |
6e0534f2 GH |
77 | #include "sched_cpupri.h" |
78 | ||
1da177e4 LT |
79 | /* |
80 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
81 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
82 | * and back. | |
83 | */ | |
84 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
85 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
86 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
87 | ||
88 | /* | |
89 | * 'User priority' is the nice value converted to something we | |
90 | * can work with better when scaling various scheduler parameters, | |
91 | * it's a [ 0 ... 39 ] range. | |
92 | */ | |
93 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
94 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
95 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
96 | ||
97 | /* | |
d7876a08 | 98 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 99 | */ |
d6322faf | 100 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 101 | |
6aa645ea IM |
102 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
103 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
104 | ||
1da177e4 LT |
105 | /* |
106 | * These are the 'tuning knobs' of the scheduler: | |
107 | * | |
a4ec24b4 | 108 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
109 | * Timeslices get refilled after they expire. |
110 | */ | |
1da177e4 | 111 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 112 | |
d0b27fa7 PZ |
113 | /* |
114 | * single value that denotes runtime == period, ie unlimited time. | |
115 | */ | |
116 | #define RUNTIME_INF ((u64)~0ULL) | |
117 | ||
5517d86b ED |
118 | #ifdef CONFIG_SMP |
119 | /* | |
120 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
121 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
122 | */ | |
123 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
124 | { | |
125 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
126 | } | |
127 | ||
128 | /* | |
129 | * Each time a sched group cpu_power is changed, | |
130 | * we must compute its reciprocal value | |
131 | */ | |
132 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
133 | { | |
134 | sg->__cpu_power += val; | |
135 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
136 | } | |
137 | #endif | |
138 | ||
e05606d3 IM |
139 | static inline int rt_policy(int policy) |
140 | { | |
3f33a7ce | 141 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
142 | return 1; |
143 | return 0; | |
144 | } | |
145 | ||
146 | static inline int task_has_rt_policy(struct task_struct *p) | |
147 | { | |
148 | return rt_policy(p->policy); | |
149 | } | |
150 | ||
1da177e4 | 151 | /* |
6aa645ea | 152 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 153 | */ |
6aa645ea IM |
154 | struct rt_prio_array { |
155 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
20b6331b | 156 | struct list_head queue[MAX_RT_PRIO]; |
6aa645ea IM |
157 | }; |
158 | ||
d0b27fa7 | 159 | struct rt_bandwidth { |
ea736ed5 IM |
160 | /* nests inside the rq lock: */ |
161 | spinlock_t rt_runtime_lock; | |
162 | ktime_t rt_period; | |
163 | u64 rt_runtime; | |
164 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
165 | }; |
166 | ||
167 | static struct rt_bandwidth def_rt_bandwidth; | |
168 | ||
169 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
170 | ||
171 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
172 | { | |
173 | struct rt_bandwidth *rt_b = | |
174 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
175 | ktime_t now; | |
176 | int overrun; | |
177 | int idle = 0; | |
178 | ||
179 | for (;;) { | |
180 | now = hrtimer_cb_get_time(timer); | |
181 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
182 | ||
183 | if (!overrun) | |
184 | break; | |
185 | ||
186 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
187 | } | |
188 | ||
189 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
190 | } | |
191 | ||
192 | static | |
193 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
194 | { | |
195 | rt_b->rt_period = ns_to_ktime(period); | |
196 | rt_b->rt_runtime = runtime; | |
197 | ||
ac086bc2 PZ |
198 | spin_lock_init(&rt_b->rt_runtime_lock); |
199 | ||
d0b27fa7 PZ |
200 | hrtimer_init(&rt_b->rt_period_timer, |
201 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
202 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
203 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
204 | } | |
205 | ||
206 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
207 | { | |
208 | ktime_t now; | |
209 | ||
210 | if (rt_b->rt_runtime == RUNTIME_INF) | |
211 | return; | |
212 | ||
213 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
214 | return; | |
215 | ||
216 | spin_lock(&rt_b->rt_runtime_lock); | |
217 | for (;;) { | |
218 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
219 | break; | |
220 | ||
221 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
222 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
223 | hrtimer_start(&rt_b->rt_period_timer, | |
224 | rt_b->rt_period_timer.expires, | |
225 | HRTIMER_MODE_ABS); | |
226 | } | |
227 | spin_unlock(&rt_b->rt_runtime_lock); | |
228 | } | |
229 | ||
230 | #ifdef CONFIG_RT_GROUP_SCHED | |
231 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
232 | { | |
233 | hrtimer_cancel(&rt_b->rt_period_timer); | |
234 | } | |
235 | #endif | |
236 | ||
712555ee HC |
237 | /* |
238 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
239 | * detach_destroy_domains and partition_sched_domains. | |
240 | */ | |
241 | static DEFINE_MUTEX(sched_domains_mutex); | |
242 | ||
052f1dc7 | 243 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 244 | |
68318b8e SV |
245 | #include <linux/cgroup.h> |
246 | ||
29f59db3 SV |
247 | struct cfs_rq; |
248 | ||
6f505b16 PZ |
249 | static LIST_HEAD(task_groups); |
250 | ||
29f59db3 | 251 | /* task group related information */ |
4cf86d77 | 252 | struct task_group { |
052f1dc7 | 253 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
254 | struct cgroup_subsys_state css; |
255 | #endif | |
052f1dc7 PZ |
256 | |
257 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
258 | /* schedulable entities of this group on each cpu */ |
259 | struct sched_entity **se; | |
260 | /* runqueue "owned" by this group on each cpu */ | |
261 | struct cfs_rq **cfs_rq; | |
262 | unsigned long shares; | |
052f1dc7 PZ |
263 | #endif |
264 | ||
265 | #ifdef CONFIG_RT_GROUP_SCHED | |
266 | struct sched_rt_entity **rt_se; | |
267 | struct rt_rq **rt_rq; | |
268 | ||
d0b27fa7 | 269 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 270 | #endif |
6b2d7700 | 271 | |
ae8393e5 | 272 | struct rcu_head rcu; |
6f505b16 | 273 | struct list_head list; |
f473aa5e PZ |
274 | |
275 | struct task_group *parent; | |
276 | struct list_head siblings; | |
277 | struct list_head children; | |
29f59db3 SV |
278 | }; |
279 | ||
354d60c2 | 280 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
281 | |
282 | /* | |
283 | * Root task group. | |
284 | * Every UID task group (including init_task_group aka UID-0) will | |
285 | * be a child to this group. | |
286 | */ | |
287 | struct task_group root_task_group; | |
288 | ||
052f1dc7 | 289 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
290 | /* Default task group's sched entity on each cpu */ |
291 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
292 | /* Default task group's cfs_rq on each cpu */ | |
293 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 294 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
295 | |
296 | #ifdef CONFIG_RT_GROUP_SCHED | |
297 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
298 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad DG |
299 | #endif /* CONFIG_RT_GROUP_SCHED */ |
300 | #else /* !CONFIG_FAIR_GROUP_SCHED */ | |
eff766a6 | 301 | #define root_task_group init_task_group |
6d6bc0ad | 302 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
6f505b16 | 303 | |
8ed36996 | 304 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
305 | * a task group's cpu shares. |
306 | */ | |
8ed36996 | 307 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 308 | |
052f1dc7 | 309 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
310 | #ifdef CONFIG_USER_SCHED |
311 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 312 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 313 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 314 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 315 | |
cb4ad1ff | 316 | /* |
2e084786 LJ |
317 | * A weight of 0 or 1 can cause arithmetics problems. |
318 | * A weight of a cfs_rq is the sum of weights of which entities | |
319 | * are queued on this cfs_rq, so a weight of a entity should not be | |
320 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
321 | * (The default weight is 1024 - so there's no practical |
322 | * limitation from this.) | |
323 | */ | |
18d95a28 | 324 | #define MIN_SHARES 2 |
2e084786 | 325 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 326 | |
052f1dc7 PZ |
327 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
328 | #endif | |
329 | ||
29f59db3 | 330 | /* Default task group. |
3a252015 | 331 | * Every task in system belong to this group at bootup. |
29f59db3 | 332 | */ |
434d53b0 | 333 | struct task_group init_task_group; |
29f59db3 SV |
334 | |
335 | /* return group to which a task belongs */ | |
4cf86d77 | 336 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 337 | { |
4cf86d77 | 338 | struct task_group *tg; |
9b5b7751 | 339 | |
052f1dc7 | 340 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 341 | tg = p->user->tg; |
052f1dc7 | 342 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
343 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
344 | struct task_group, css); | |
24e377a8 | 345 | #else |
41a2d6cf | 346 | tg = &init_task_group; |
24e377a8 | 347 | #endif |
9b5b7751 | 348 | return tg; |
29f59db3 SV |
349 | } |
350 | ||
351 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 352 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 353 | { |
052f1dc7 | 354 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
355 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
356 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 357 | #endif |
6f505b16 | 358 | |
052f1dc7 | 359 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
360 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
361 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 362 | #endif |
29f59db3 SV |
363 | } |
364 | ||
365 | #else | |
366 | ||
6f505b16 | 367 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
29f59db3 | 368 | |
052f1dc7 | 369 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 370 | |
6aa645ea IM |
371 | /* CFS-related fields in a runqueue */ |
372 | struct cfs_rq { | |
373 | struct load_weight load; | |
374 | unsigned long nr_running; | |
375 | ||
6aa645ea | 376 | u64 exec_clock; |
e9acbff6 | 377 | u64 min_vruntime; |
103638d9 | 378 | u64 pair_start; |
6aa645ea IM |
379 | |
380 | struct rb_root tasks_timeline; | |
381 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
382 | |
383 | struct list_head tasks; | |
384 | struct list_head *balance_iterator; | |
385 | ||
386 | /* | |
387 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
388 | * It is set to NULL otherwise (i.e when none are currently running). |
389 | */ | |
aa2ac252 | 390 | struct sched_entity *curr, *next; |
ddc97297 PZ |
391 | |
392 | unsigned long nr_spread_over; | |
393 | ||
62160e3f | 394 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
395 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
396 | ||
41a2d6cf IM |
397 | /* |
398 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
399 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
400 | * (like users, containers etc.) | |
401 | * | |
402 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
403 | * list is used during load balance. | |
404 | */ | |
41a2d6cf IM |
405 | struct list_head leaf_cfs_rq_list; |
406 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
407 | |
408 | #ifdef CONFIG_SMP | |
c09595f6 | 409 | /* |
c8cba857 | 410 | * the part of load.weight contributed by tasks |
c09595f6 | 411 | */ |
c8cba857 | 412 | unsigned long task_weight; |
c09595f6 | 413 | |
c8cba857 PZ |
414 | /* |
415 | * h_load = weight * f(tg) | |
416 | * | |
417 | * Where f(tg) is the recursive weight fraction assigned to | |
418 | * this group. | |
419 | */ | |
420 | unsigned long h_load; | |
c09595f6 | 421 | |
c8cba857 PZ |
422 | /* |
423 | * this cpu's part of tg->shares | |
424 | */ | |
425 | unsigned long shares; | |
c09595f6 | 426 | #endif |
6aa645ea IM |
427 | #endif |
428 | }; | |
1da177e4 | 429 | |
6aa645ea IM |
430 | /* Real-Time classes' related field in a runqueue: */ |
431 | struct rt_rq { | |
432 | struct rt_prio_array active; | |
63489e45 | 433 | unsigned long rt_nr_running; |
052f1dc7 | 434 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
435 | int highest_prio; /* highest queued rt task prio */ |
436 | #endif | |
fa85ae24 | 437 | #ifdef CONFIG_SMP |
73fe6aae | 438 | unsigned long rt_nr_migratory; |
a22d7fc1 | 439 | int overloaded; |
fa85ae24 | 440 | #endif |
6f505b16 | 441 | int rt_throttled; |
fa85ae24 | 442 | u64 rt_time; |
ac086bc2 | 443 | u64 rt_runtime; |
ea736ed5 | 444 | /* Nests inside the rq lock: */ |
ac086bc2 | 445 | spinlock_t rt_runtime_lock; |
6f505b16 | 446 | |
052f1dc7 | 447 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
448 | unsigned long rt_nr_boosted; |
449 | ||
6f505b16 PZ |
450 | struct rq *rq; |
451 | struct list_head leaf_rt_rq_list; | |
452 | struct task_group *tg; | |
453 | struct sched_rt_entity *rt_se; | |
454 | #endif | |
6aa645ea IM |
455 | }; |
456 | ||
57d885fe GH |
457 | #ifdef CONFIG_SMP |
458 | ||
459 | /* | |
460 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
461 | * variables. Each exclusive cpuset essentially defines an island domain by |
462 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
463 | * exclusive cpuset is created, we also create and attach a new root-domain |
464 | * object. | |
465 | * | |
57d885fe GH |
466 | */ |
467 | struct root_domain { | |
468 | atomic_t refcount; | |
469 | cpumask_t span; | |
470 | cpumask_t online; | |
637f5085 | 471 | |
0eab9146 | 472 | /* |
637f5085 GH |
473 | * The "RT overload" flag: it gets set if a CPU has more than |
474 | * one runnable RT task. | |
475 | */ | |
476 | cpumask_t rto_mask; | |
0eab9146 | 477 | atomic_t rto_count; |
6e0534f2 GH |
478 | #ifdef CONFIG_SMP |
479 | struct cpupri cpupri; | |
480 | #endif | |
57d885fe GH |
481 | }; |
482 | ||
dc938520 GH |
483 | /* |
484 | * By default the system creates a single root-domain with all cpus as | |
485 | * members (mimicking the global state we have today). | |
486 | */ | |
57d885fe GH |
487 | static struct root_domain def_root_domain; |
488 | ||
489 | #endif | |
490 | ||
1da177e4 LT |
491 | /* |
492 | * This is the main, per-CPU runqueue data structure. | |
493 | * | |
494 | * Locking rule: those places that want to lock multiple runqueues | |
495 | * (such as the load balancing or the thread migration code), lock | |
496 | * acquire operations must be ordered by ascending &runqueue. | |
497 | */ | |
70b97a7f | 498 | struct rq { |
d8016491 IM |
499 | /* runqueue lock: */ |
500 | spinlock_t lock; | |
1da177e4 LT |
501 | |
502 | /* | |
503 | * nr_running and cpu_load should be in the same cacheline because | |
504 | * remote CPUs use both these fields when doing load calculation. | |
505 | */ | |
506 | unsigned long nr_running; | |
6aa645ea IM |
507 | #define CPU_LOAD_IDX_MAX 5 |
508 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 509 | unsigned char idle_at_tick; |
46cb4b7c | 510 | #ifdef CONFIG_NO_HZ |
15934a37 | 511 | unsigned long last_tick_seen; |
46cb4b7c SS |
512 | unsigned char in_nohz_recently; |
513 | #endif | |
d8016491 IM |
514 | /* capture load from *all* tasks on this cpu: */ |
515 | struct load_weight load; | |
6aa645ea IM |
516 | unsigned long nr_load_updates; |
517 | u64 nr_switches; | |
518 | ||
519 | struct cfs_rq cfs; | |
6f505b16 | 520 | struct rt_rq rt; |
6f505b16 | 521 | |
6aa645ea | 522 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
523 | /* list of leaf cfs_rq on this cpu: */ |
524 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
525 | #endif |
526 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 527 | struct list_head leaf_rt_rq_list; |
1da177e4 | 528 | #endif |
1da177e4 LT |
529 | |
530 | /* | |
531 | * This is part of a global counter where only the total sum | |
532 | * over all CPUs matters. A task can increase this counter on | |
533 | * one CPU and if it got migrated afterwards it may decrease | |
534 | * it on another CPU. Always updated under the runqueue lock: | |
535 | */ | |
536 | unsigned long nr_uninterruptible; | |
537 | ||
36c8b586 | 538 | struct task_struct *curr, *idle; |
c9819f45 | 539 | unsigned long next_balance; |
1da177e4 | 540 | struct mm_struct *prev_mm; |
6aa645ea | 541 | |
3e51f33f | 542 | u64 clock; |
6aa645ea | 543 | |
1da177e4 LT |
544 | atomic_t nr_iowait; |
545 | ||
546 | #ifdef CONFIG_SMP | |
0eab9146 | 547 | struct root_domain *rd; |
1da177e4 LT |
548 | struct sched_domain *sd; |
549 | ||
550 | /* For active balancing */ | |
551 | int active_balance; | |
552 | int push_cpu; | |
d8016491 IM |
553 | /* cpu of this runqueue: */ |
554 | int cpu; | |
1f11eb6a | 555 | int online; |
1da177e4 | 556 | |
a8a51d5e PZ |
557 | unsigned long avg_load_per_task; |
558 | ||
36c8b586 | 559 | struct task_struct *migration_thread; |
1da177e4 LT |
560 | struct list_head migration_queue; |
561 | #endif | |
562 | ||
8f4d37ec PZ |
563 | #ifdef CONFIG_SCHED_HRTICK |
564 | unsigned long hrtick_flags; | |
565 | ktime_t hrtick_expire; | |
566 | struct hrtimer hrtick_timer; | |
567 | #endif | |
568 | ||
1da177e4 LT |
569 | #ifdef CONFIG_SCHEDSTATS |
570 | /* latency stats */ | |
571 | struct sched_info rq_sched_info; | |
572 | ||
573 | /* sys_sched_yield() stats */ | |
480b9434 KC |
574 | unsigned int yld_exp_empty; |
575 | unsigned int yld_act_empty; | |
576 | unsigned int yld_both_empty; | |
577 | unsigned int yld_count; | |
1da177e4 LT |
578 | |
579 | /* schedule() stats */ | |
480b9434 KC |
580 | unsigned int sched_switch; |
581 | unsigned int sched_count; | |
582 | unsigned int sched_goidle; | |
1da177e4 LT |
583 | |
584 | /* try_to_wake_up() stats */ | |
480b9434 KC |
585 | unsigned int ttwu_count; |
586 | unsigned int ttwu_local; | |
b8efb561 IM |
587 | |
588 | /* BKL stats */ | |
480b9434 | 589 | unsigned int bkl_count; |
1da177e4 | 590 | #endif |
fcb99371 | 591 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
592 | }; |
593 | ||
f34e3b61 | 594 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 595 | |
dd41f596 IM |
596 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
597 | { | |
598 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
599 | } | |
600 | ||
0a2966b4 CL |
601 | static inline int cpu_of(struct rq *rq) |
602 | { | |
603 | #ifdef CONFIG_SMP | |
604 | return rq->cpu; | |
605 | #else | |
606 | return 0; | |
607 | #endif | |
608 | } | |
609 | ||
674311d5 NP |
610 | /* |
611 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 612 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
613 | * |
614 | * The domain tree of any CPU may only be accessed from within | |
615 | * preempt-disabled sections. | |
616 | */ | |
48f24c4d IM |
617 | #define for_each_domain(cpu, __sd) \ |
618 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
619 | |
620 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
621 | #define this_rq() (&__get_cpu_var(runqueues)) | |
622 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
623 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
624 | ||
3e51f33f PZ |
625 | static inline void update_rq_clock(struct rq *rq) |
626 | { | |
627 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
628 | } | |
629 | ||
bf5c91ba IM |
630 | /* |
631 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
632 | */ | |
633 | #ifdef CONFIG_SCHED_DEBUG | |
634 | # define const_debug __read_mostly | |
635 | #else | |
636 | # define const_debug static const | |
637 | #endif | |
638 | ||
639 | /* | |
640 | * Debugging: various feature bits | |
641 | */ | |
f00b45c1 PZ |
642 | |
643 | #define SCHED_FEAT(name, enabled) \ | |
644 | __SCHED_FEAT_##name , | |
645 | ||
bf5c91ba | 646 | enum { |
f00b45c1 | 647 | #include "sched_features.h" |
bf5c91ba IM |
648 | }; |
649 | ||
f00b45c1 PZ |
650 | #undef SCHED_FEAT |
651 | ||
652 | #define SCHED_FEAT(name, enabled) \ | |
653 | (1UL << __SCHED_FEAT_##name) * enabled | | |
654 | ||
bf5c91ba | 655 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
656 | #include "sched_features.h" |
657 | 0; | |
658 | ||
659 | #undef SCHED_FEAT | |
660 | ||
661 | #ifdef CONFIG_SCHED_DEBUG | |
662 | #define SCHED_FEAT(name, enabled) \ | |
663 | #name , | |
664 | ||
983ed7a6 | 665 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
666 | #include "sched_features.h" |
667 | NULL | |
668 | }; | |
669 | ||
670 | #undef SCHED_FEAT | |
671 | ||
983ed7a6 | 672 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
673 | { |
674 | filp->private_data = inode->i_private; | |
675 | return 0; | |
676 | } | |
677 | ||
678 | static ssize_t | |
679 | sched_feat_read(struct file *filp, char __user *ubuf, | |
680 | size_t cnt, loff_t *ppos) | |
681 | { | |
682 | char *buf; | |
683 | int r = 0; | |
684 | int len = 0; | |
685 | int i; | |
686 | ||
687 | for (i = 0; sched_feat_names[i]; i++) { | |
688 | len += strlen(sched_feat_names[i]); | |
689 | len += 4; | |
690 | } | |
691 | ||
692 | buf = kmalloc(len + 2, GFP_KERNEL); | |
693 | if (!buf) | |
694 | return -ENOMEM; | |
695 | ||
696 | for (i = 0; sched_feat_names[i]; i++) { | |
697 | if (sysctl_sched_features & (1UL << i)) | |
698 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
699 | else | |
c24b7c52 | 700 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
701 | } |
702 | ||
703 | r += sprintf(buf + r, "\n"); | |
704 | WARN_ON(r >= len + 2); | |
705 | ||
706 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
707 | ||
708 | kfree(buf); | |
709 | ||
710 | return r; | |
711 | } | |
712 | ||
713 | static ssize_t | |
714 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
715 | size_t cnt, loff_t *ppos) | |
716 | { | |
717 | char buf[64]; | |
718 | char *cmp = buf; | |
719 | int neg = 0; | |
720 | int i; | |
721 | ||
722 | if (cnt > 63) | |
723 | cnt = 63; | |
724 | ||
725 | if (copy_from_user(&buf, ubuf, cnt)) | |
726 | return -EFAULT; | |
727 | ||
728 | buf[cnt] = 0; | |
729 | ||
c24b7c52 | 730 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
731 | neg = 1; |
732 | cmp += 3; | |
733 | } | |
734 | ||
735 | for (i = 0; sched_feat_names[i]; i++) { | |
736 | int len = strlen(sched_feat_names[i]); | |
737 | ||
738 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
739 | if (neg) | |
740 | sysctl_sched_features &= ~(1UL << i); | |
741 | else | |
742 | sysctl_sched_features |= (1UL << i); | |
743 | break; | |
744 | } | |
745 | } | |
746 | ||
747 | if (!sched_feat_names[i]) | |
748 | return -EINVAL; | |
749 | ||
750 | filp->f_pos += cnt; | |
751 | ||
752 | return cnt; | |
753 | } | |
754 | ||
755 | static struct file_operations sched_feat_fops = { | |
756 | .open = sched_feat_open, | |
757 | .read = sched_feat_read, | |
758 | .write = sched_feat_write, | |
759 | }; | |
760 | ||
761 | static __init int sched_init_debug(void) | |
762 | { | |
f00b45c1 PZ |
763 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
764 | &sched_feat_fops); | |
765 | ||
766 | return 0; | |
767 | } | |
768 | late_initcall(sched_init_debug); | |
769 | ||
770 | #endif | |
771 | ||
772 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 773 | |
b82d9fdd PZ |
774 | /* |
775 | * Number of tasks to iterate in a single balance run. | |
776 | * Limited because this is done with IRQs disabled. | |
777 | */ | |
778 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
779 | ||
fa85ae24 | 780 | /* |
9f0c1e56 | 781 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
782 | * default: 1s |
783 | */ | |
9f0c1e56 | 784 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 785 | |
6892b75e IM |
786 | static __read_mostly int scheduler_running; |
787 | ||
9f0c1e56 PZ |
788 | /* |
789 | * part of the period that we allow rt tasks to run in us. | |
790 | * default: 0.95s | |
791 | */ | |
792 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 793 | |
d0b27fa7 PZ |
794 | static inline u64 global_rt_period(void) |
795 | { | |
796 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
797 | } | |
798 | ||
799 | static inline u64 global_rt_runtime(void) | |
800 | { | |
801 | if (sysctl_sched_rt_period < 0) | |
802 | return RUNTIME_INF; | |
803 | ||
804 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
805 | } | |
fa85ae24 | 806 | |
1da177e4 | 807 | #ifndef prepare_arch_switch |
4866cde0 NP |
808 | # define prepare_arch_switch(next) do { } while (0) |
809 | #endif | |
810 | #ifndef finish_arch_switch | |
811 | # define finish_arch_switch(prev) do { } while (0) | |
812 | #endif | |
813 | ||
051a1d1a DA |
814 | static inline int task_current(struct rq *rq, struct task_struct *p) |
815 | { | |
816 | return rq->curr == p; | |
817 | } | |
818 | ||
4866cde0 | 819 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 820 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 821 | { |
051a1d1a | 822 | return task_current(rq, p); |
4866cde0 NP |
823 | } |
824 | ||
70b97a7f | 825 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
826 | { |
827 | } | |
828 | ||
70b97a7f | 829 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 830 | { |
da04c035 IM |
831 | #ifdef CONFIG_DEBUG_SPINLOCK |
832 | /* this is a valid case when another task releases the spinlock */ | |
833 | rq->lock.owner = current; | |
834 | #endif | |
8a25d5de IM |
835 | /* |
836 | * If we are tracking spinlock dependencies then we have to | |
837 | * fix up the runqueue lock - which gets 'carried over' from | |
838 | * prev into current: | |
839 | */ | |
840 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
841 | ||
4866cde0 NP |
842 | spin_unlock_irq(&rq->lock); |
843 | } | |
844 | ||
845 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 846 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
847 | { |
848 | #ifdef CONFIG_SMP | |
849 | return p->oncpu; | |
850 | #else | |
051a1d1a | 851 | return task_current(rq, p); |
4866cde0 NP |
852 | #endif |
853 | } | |
854 | ||
70b97a7f | 855 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
856 | { |
857 | #ifdef CONFIG_SMP | |
858 | /* | |
859 | * We can optimise this out completely for !SMP, because the | |
860 | * SMP rebalancing from interrupt is the only thing that cares | |
861 | * here. | |
862 | */ | |
863 | next->oncpu = 1; | |
864 | #endif | |
865 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
866 | spin_unlock_irq(&rq->lock); | |
867 | #else | |
868 | spin_unlock(&rq->lock); | |
869 | #endif | |
870 | } | |
871 | ||
70b97a7f | 872 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
873 | { |
874 | #ifdef CONFIG_SMP | |
875 | /* | |
876 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
877 | * We must ensure this doesn't happen until the switch is completely | |
878 | * finished. | |
879 | */ | |
880 | smp_wmb(); | |
881 | prev->oncpu = 0; | |
882 | #endif | |
883 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
884 | local_irq_enable(); | |
1da177e4 | 885 | #endif |
4866cde0 NP |
886 | } |
887 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 888 | |
b29739f9 IM |
889 | /* |
890 | * __task_rq_lock - lock the runqueue a given task resides on. | |
891 | * Must be called interrupts disabled. | |
892 | */ | |
70b97a7f | 893 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
894 | __acquires(rq->lock) |
895 | { | |
3a5c359a AK |
896 | for (;;) { |
897 | struct rq *rq = task_rq(p); | |
898 | spin_lock(&rq->lock); | |
899 | if (likely(rq == task_rq(p))) | |
900 | return rq; | |
b29739f9 | 901 | spin_unlock(&rq->lock); |
b29739f9 | 902 | } |
b29739f9 IM |
903 | } |
904 | ||
1da177e4 LT |
905 | /* |
906 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 907 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
908 | * explicitly disabling preemption. |
909 | */ | |
70b97a7f | 910 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
911 | __acquires(rq->lock) |
912 | { | |
70b97a7f | 913 | struct rq *rq; |
1da177e4 | 914 | |
3a5c359a AK |
915 | for (;;) { |
916 | local_irq_save(*flags); | |
917 | rq = task_rq(p); | |
918 | spin_lock(&rq->lock); | |
919 | if (likely(rq == task_rq(p))) | |
920 | return rq; | |
1da177e4 | 921 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 922 | } |
1da177e4 LT |
923 | } |
924 | ||
a9957449 | 925 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
926 | __releases(rq->lock) |
927 | { | |
928 | spin_unlock(&rq->lock); | |
929 | } | |
930 | ||
70b97a7f | 931 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
932 | __releases(rq->lock) |
933 | { | |
934 | spin_unlock_irqrestore(&rq->lock, *flags); | |
935 | } | |
936 | ||
1da177e4 | 937 | /* |
cc2a73b5 | 938 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 939 | */ |
a9957449 | 940 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
941 | __acquires(rq->lock) |
942 | { | |
70b97a7f | 943 | struct rq *rq; |
1da177e4 LT |
944 | |
945 | local_irq_disable(); | |
946 | rq = this_rq(); | |
947 | spin_lock(&rq->lock); | |
948 | ||
949 | return rq; | |
950 | } | |
951 | ||
8f4d37ec PZ |
952 | static void __resched_task(struct task_struct *p, int tif_bit); |
953 | ||
954 | static inline void resched_task(struct task_struct *p) | |
955 | { | |
956 | __resched_task(p, TIF_NEED_RESCHED); | |
957 | } | |
958 | ||
959 | #ifdef CONFIG_SCHED_HRTICK | |
960 | /* | |
961 | * Use HR-timers to deliver accurate preemption points. | |
962 | * | |
963 | * Its all a bit involved since we cannot program an hrt while holding the | |
964 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
965 | * reschedule event. | |
966 | * | |
967 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
968 | * rq->lock. | |
969 | */ | |
970 | static inline void resched_hrt(struct task_struct *p) | |
971 | { | |
972 | __resched_task(p, TIF_HRTICK_RESCHED); | |
973 | } | |
974 | ||
975 | static inline void resched_rq(struct rq *rq) | |
976 | { | |
977 | unsigned long flags; | |
978 | ||
979 | spin_lock_irqsave(&rq->lock, flags); | |
980 | resched_task(rq->curr); | |
981 | spin_unlock_irqrestore(&rq->lock, flags); | |
982 | } | |
983 | ||
984 | enum { | |
985 | HRTICK_SET, /* re-programm hrtick_timer */ | |
986 | HRTICK_RESET, /* not a new slice */ | |
b328ca18 | 987 | HRTICK_BLOCK, /* stop hrtick operations */ |
8f4d37ec PZ |
988 | }; |
989 | ||
990 | /* | |
991 | * Use hrtick when: | |
992 | * - enabled by features | |
993 | * - hrtimer is actually high res | |
994 | */ | |
995 | static inline int hrtick_enabled(struct rq *rq) | |
996 | { | |
997 | if (!sched_feat(HRTICK)) | |
998 | return 0; | |
b328ca18 PZ |
999 | if (unlikely(test_bit(HRTICK_BLOCK, &rq->hrtick_flags))) |
1000 | return 0; | |
8f4d37ec PZ |
1001 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1002 | } | |
1003 | ||
1004 | /* | |
1005 | * Called to set the hrtick timer state. | |
1006 | * | |
1007 | * called with rq->lock held and irqs disabled | |
1008 | */ | |
1009 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
1010 | { | |
1011 | assert_spin_locked(&rq->lock); | |
1012 | ||
1013 | /* | |
1014 | * preempt at: now + delay | |
1015 | */ | |
1016 | rq->hrtick_expire = | |
1017 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
1018 | /* | |
1019 | * indicate we need to program the timer | |
1020 | */ | |
1021 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
1022 | if (reset) | |
1023 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1024 | ||
1025 | /* | |
1026 | * New slices are called from the schedule path and don't need a | |
1027 | * forced reschedule. | |
1028 | */ | |
1029 | if (reset) | |
1030 | resched_hrt(rq->curr); | |
1031 | } | |
1032 | ||
1033 | static void hrtick_clear(struct rq *rq) | |
1034 | { | |
1035 | if (hrtimer_active(&rq->hrtick_timer)) | |
1036 | hrtimer_cancel(&rq->hrtick_timer); | |
1037 | } | |
1038 | ||
1039 | /* | |
1040 | * Update the timer from the possible pending state. | |
1041 | */ | |
1042 | static void hrtick_set(struct rq *rq) | |
1043 | { | |
1044 | ktime_t time; | |
1045 | int set, reset; | |
1046 | unsigned long flags; | |
1047 | ||
1048 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1049 | ||
1050 | spin_lock_irqsave(&rq->lock, flags); | |
1051 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
1052 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1053 | time = rq->hrtick_expire; | |
1054 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
1055 | spin_unlock_irqrestore(&rq->lock, flags); | |
1056 | ||
1057 | if (set) { | |
1058 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
1059 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
1060 | resched_rq(rq); | |
1061 | } else | |
1062 | hrtick_clear(rq); | |
1063 | } | |
1064 | ||
1065 | /* | |
1066 | * High-resolution timer tick. | |
1067 | * Runs from hardirq context with interrupts disabled. | |
1068 | */ | |
1069 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1070 | { | |
1071 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1072 | ||
1073 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1074 | ||
1075 | spin_lock(&rq->lock); | |
3e51f33f | 1076 | update_rq_clock(rq); |
8f4d37ec PZ |
1077 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1078 | spin_unlock(&rq->lock); | |
1079 | ||
1080 | return HRTIMER_NORESTART; | |
1081 | } | |
1082 | ||
81d41d7e | 1083 | #ifdef CONFIG_SMP |
b328ca18 PZ |
1084 | static void hotplug_hrtick_disable(int cpu) |
1085 | { | |
1086 | struct rq *rq = cpu_rq(cpu); | |
1087 | unsigned long flags; | |
1088 | ||
1089 | spin_lock_irqsave(&rq->lock, flags); | |
1090 | rq->hrtick_flags = 0; | |
1091 | __set_bit(HRTICK_BLOCK, &rq->hrtick_flags); | |
1092 | spin_unlock_irqrestore(&rq->lock, flags); | |
1093 | ||
1094 | hrtick_clear(rq); | |
1095 | } | |
1096 | ||
1097 | static void hotplug_hrtick_enable(int cpu) | |
1098 | { | |
1099 | struct rq *rq = cpu_rq(cpu); | |
1100 | unsigned long flags; | |
1101 | ||
1102 | spin_lock_irqsave(&rq->lock, flags); | |
1103 | __clear_bit(HRTICK_BLOCK, &rq->hrtick_flags); | |
1104 | spin_unlock_irqrestore(&rq->lock, flags); | |
1105 | } | |
1106 | ||
1107 | static int | |
1108 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1109 | { | |
1110 | int cpu = (int)(long)hcpu; | |
1111 | ||
1112 | switch (action) { | |
1113 | case CPU_UP_CANCELED: | |
1114 | case CPU_UP_CANCELED_FROZEN: | |
1115 | case CPU_DOWN_PREPARE: | |
1116 | case CPU_DOWN_PREPARE_FROZEN: | |
1117 | case CPU_DEAD: | |
1118 | case CPU_DEAD_FROZEN: | |
1119 | hotplug_hrtick_disable(cpu); | |
1120 | return NOTIFY_OK; | |
1121 | ||
1122 | case CPU_UP_PREPARE: | |
1123 | case CPU_UP_PREPARE_FROZEN: | |
1124 | case CPU_DOWN_FAILED: | |
1125 | case CPU_DOWN_FAILED_FROZEN: | |
1126 | case CPU_ONLINE: | |
1127 | case CPU_ONLINE_FROZEN: | |
1128 | hotplug_hrtick_enable(cpu); | |
1129 | return NOTIFY_OK; | |
1130 | } | |
1131 | ||
1132 | return NOTIFY_DONE; | |
1133 | } | |
1134 | ||
1135 | static void init_hrtick(void) | |
1136 | { | |
1137 | hotcpu_notifier(hotplug_hrtick, 0); | |
1138 | } | |
81d41d7e | 1139 | #endif /* CONFIG_SMP */ |
b328ca18 PZ |
1140 | |
1141 | static void init_rq_hrtick(struct rq *rq) | |
8f4d37ec PZ |
1142 | { |
1143 | rq->hrtick_flags = 0; | |
1144 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1145 | rq->hrtick_timer.function = hrtick; | |
1146 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
1147 | } | |
1148 | ||
1149 | void hrtick_resched(void) | |
1150 | { | |
1151 | struct rq *rq; | |
1152 | unsigned long flags; | |
1153 | ||
1154 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
1155 | return; | |
1156 | ||
1157 | local_irq_save(flags); | |
1158 | rq = cpu_rq(smp_processor_id()); | |
1159 | hrtick_set(rq); | |
1160 | local_irq_restore(flags); | |
1161 | } | |
1162 | #else | |
1163 | static inline void hrtick_clear(struct rq *rq) | |
1164 | { | |
1165 | } | |
1166 | ||
1167 | static inline void hrtick_set(struct rq *rq) | |
1168 | { | |
1169 | } | |
1170 | ||
1171 | static inline void init_rq_hrtick(struct rq *rq) | |
1172 | { | |
1173 | } | |
1174 | ||
1175 | void hrtick_resched(void) | |
1176 | { | |
1177 | } | |
b328ca18 PZ |
1178 | |
1179 | static inline void init_hrtick(void) | |
1180 | { | |
1181 | } | |
8f4d37ec PZ |
1182 | #endif |
1183 | ||
c24d20db IM |
1184 | /* |
1185 | * resched_task - mark a task 'to be rescheduled now'. | |
1186 | * | |
1187 | * On UP this means the setting of the need_resched flag, on SMP it | |
1188 | * might also involve a cross-CPU call to trigger the scheduler on | |
1189 | * the target CPU. | |
1190 | */ | |
1191 | #ifdef CONFIG_SMP | |
1192 | ||
1193 | #ifndef tsk_is_polling | |
1194 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1195 | #endif | |
1196 | ||
8f4d37ec | 1197 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1198 | { |
1199 | int cpu; | |
1200 | ||
1201 | assert_spin_locked(&task_rq(p)->lock); | |
1202 | ||
8f4d37ec | 1203 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1204 | return; |
1205 | ||
8f4d37ec | 1206 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1207 | |
1208 | cpu = task_cpu(p); | |
1209 | if (cpu == smp_processor_id()) | |
1210 | return; | |
1211 | ||
1212 | /* NEED_RESCHED must be visible before we test polling */ | |
1213 | smp_mb(); | |
1214 | if (!tsk_is_polling(p)) | |
1215 | smp_send_reschedule(cpu); | |
1216 | } | |
1217 | ||
1218 | static void resched_cpu(int cpu) | |
1219 | { | |
1220 | struct rq *rq = cpu_rq(cpu); | |
1221 | unsigned long flags; | |
1222 | ||
1223 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1224 | return; | |
1225 | resched_task(cpu_curr(cpu)); | |
1226 | spin_unlock_irqrestore(&rq->lock, flags); | |
1227 | } | |
06d8308c TG |
1228 | |
1229 | #ifdef CONFIG_NO_HZ | |
1230 | /* | |
1231 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1232 | * idle CPU then this timer might expire before the next timer event | |
1233 | * which is scheduled to wake up that CPU. In case of a completely | |
1234 | * idle system the next event might even be infinite time into the | |
1235 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1236 | * leaves the inner idle loop so the newly added timer is taken into | |
1237 | * account when the CPU goes back to idle and evaluates the timer | |
1238 | * wheel for the next timer event. | |
1239 | */ | |
1240 | void wake_up_idle_cpu(int cpu) | |
1241 | { | |
1242 | struct rq *rq = cpu_rq(cpu); | |
1243 | ||
1244 | if (cpu == smp_processor_id()) | |
1245 | return; | |
1246 | ||
1247 | /* | |
1248 | * This is safe, as this function is called with the timer | |
1249 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1250 | * to idle and has not yet set rq->curr to idle then it will | |
1251 | * be serialized on the timer wheel base lock and take the new | |
1252 | * timer into account automatically. | |
1253 | */ | |
1254 | if (rq->curr != rq->idle) | |
1255 | return; | |
1256 | ||
1257 | /* | |
1258 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1259 | * lockless. The worst case is that the other CPU runs the | |
1260 | * idle task through an additional NOOP schedule() | |
1261 | */ | |
1262 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1263 | ||
1264 | /* NEED_RESCHED must be visible before we test polling */ | |
1265 | smp_mb(); | |
1266 | if (!tsk_is_polling(rq->idle)) | |
1267 | smp_send_reschedule(cpu); | |
1268 | } | |
6d6bc0ad | 1269 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1270 | |
6d6bc0ad | 1271 | #else /* !CONFIG_SMP */ |
8f4d37ec | 1272 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1273 | { |
1274 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1275 | set_tsk_thread_flag(p, tif_bit); |
c24d20db | 1276 | } |
6d6bc0ad | 1277 | #endif /* CONFIG_SMP */ |
c24d20db | 1278 | |
45bf76df IM |
1279 | #if BITS_PER_LONG == 32 |
1280 | # define WMULT_CONST (~0UL) | |
1281 | #else | |
1282 | # define WMULT_CONST (1UL << 32) | |
1283 | #endif | |
1284 | ||
1285 | #define WMULT_SHIFT 32 | |
1286 | ||
194081eb IM |
1287 | /* |
1288 | * Shift right and round: | |
1289 | */ | |
cf2ab469 | 1290 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1291 | |
a7be37ac PZ |
1292 | /* |
1293 | * delta *= weight / lw | |
1294 | */ | |
cb1c4fc9 | 1295 | static unsigned long |
45bf76df IM |
1296 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1297 | struct load_weight *lw) | |
1298 | { | |
1299 | u64 tmp; | |
1300 | ||
7a232e03 LJ |
1301 | if (!lw->inv_weight) { |
1302 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1303 | lw->inv_weight = 1; | |
1304 | else | |
1305 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1306 | / (lw->weight+1); | |
1307 | } | |
45bf76df IM |
1308 | |
1309 | tmp = (u64)delta_exec * weight; | |
1310 | /* | |
1311 | * Check whether we'd overflow the 64-bit multiplication: | |
1312 | */ | |
194081eb | 1313 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1314 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1315 | WMULT_SHIFT/2); |
1316 | else | |
cf2ab469 | 1317 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1318 | |
ecf691da | 1319 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1320 | } |
1321 | ||
1091985b | 1322 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1323 | { |
1324 | lw->weight += inc; | |
e89996ae | 1325 | lw->inv_weight = 0; |
45bf76df IM |
1326 | } |
1327 | ||
1091985b | 1328 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1329 | { |
1330 | lw->weight -= dec; | |
e89996ae | 1331 | lw->inv_weight = 0; |
45bf76df IM |
1332 | } |
1333 | ||
2dd73a4f PW |
1334 | /* |
1335 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1336 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1337 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1338 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1339 | * scaled version of the new time slice allocation that they receive on time |
1340 | * slice expiry etc. | |
1341 | */ | |
1342 | ||
dd41f596 IM |
1343 | #define WEIGHT_IDLEPRIO 2 |
1344 | #define WMULT_IDLEPRIO (1 << 31) | |
1345 | ||
1346 | /* | |
1347 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1348 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1349 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1350 | * that remained on nice 0. | |
1351 | * | |
1352 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1353 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1354 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1355 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1356 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1357 | */ |
1358 | static const int prio_to_weight[40] = { | |
254753dc IM |
1359 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1360 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1361 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1362 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1363 | /* 0 */ 1024, 820, 655, 526, 423, | |
1364 | /* 5 */ 335, 272, 215, 172, 137, | |
1365 | /* 10 */ 110, 87, 70, 56, 45, | |
1366 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1367 | }; |
1368 | ||
5714d2de IM |
1369 | /* |
1370 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1371 | * | |
1372 | * In cases where the weight does not change often, we can use the | |
1373 | * precalculated inverse to speed up arithmetics by turning divisions | |
1374 | * into multiplications: | |
1375 | */ | |
dd41f596 | 1376 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1377 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1378 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1379 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1380 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1381 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1382 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1383 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1384 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1385 | }; |
2dd73a4f | 1386 | |
dd41f596 IM |
1387 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1388 | ||
1389 | /* | |
1390 | * runqueue iterator, to support SMP load-balancing between different | |
1391 | * scheduling classes, without having to expose their internal data | |
1392 | * structures to the load-balancing proper: | |
1393 | */ | |
1394 | struct rq_iterator { | |
1395 | void *arg; | |
1396 | struct task_struct *(*start)(void *); | |
1397 | struct task_struct *(*next)(void *); | |
1398 | }; | |
1399 | ||
e1d1484f PW |
1400 | #ifdef CONFIG_SMP |
1401 | static unsigned long | |
1402 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1403 | unsigned long max_load_move, struct sched_domain *sd, | |
1404 | enum cpu_idle_type idle, int *all_pinned, | |
1405 | int *this_best_prio, struct rq_iterator *iterator); | |
1406 | ||
1407 | static int | |
1408 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1409 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1410 | struct rq_iterator *iterator); | |
e1d1484f | 1411 | #endif |
dd41f596 | 1412 | |
d842de87 SV |
1413 | #ifdef CONFIG_CGROUP_CPUACCT |
1414 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1415 | #else | |
1416 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1417 | #endif | |
1418 | ||
18d95a28 PZ |
1419 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1420 | { | |
1421 | update_load_add(&rq->load, load); | |
1422 | } | |
1423 | ||
1424 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1425 | { | |
1426 | update_load_sub(&rq->load, load); | |
1427 | } | |
1428 | ||
e7693a36 GH |
1429 | #ifdef CONFIG_SMP |
1430 | static unsigned long source_load(int cpu, int type); | |
1431 | static unsigned long target_load(int cpu, int type); | |
e7693a36 | 1432 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
c09595f6 | 1433 | |
a8a51d5e PZ |
1434 | static unsigned long cpu_avg_load_per_task(int cpu) |
1435 | { | |
1436 | struct rq *rq = cpu_rq(cpu); | |
1437 | ||
1438 | if (rq->nr_running) | |
1439 | rq->avg_load_per_task = rq->load.weight / rq->nr_running; | |
1440 | ||
1441 | return rq->avg_load_per_task; | |
1442 | } | |
1443 | ||
c09595f6 PZ |
1444 | #ifdef CONFIG_FAIR_GROUP_SCHED |
1445 | ||
c8cba857 | 1446 | typedef void (*tg_visitor)(struct task_group *, int, struct sched_domain *); |
c09595f6 PZ |
1447 | |
1448 | /* | |
1449 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1450 | * leaving it for the final time. | |
1451 | */ | |
c8cba857 PZ |
1452 | static void |
1453 | walk_tg_tree(tg_visitor down, tg_visitor up, int cpu, struct sched_domain *sd) | |
c09595f6 PZ |
1454 | { |
1455 | struct task_group *parent, *child; | |
1456 | ||
1457 | rcu_read_lock(); | |
1458 | parent = &root_task_group; | |
1459 | down: | |
b6a86c74 | 1460 | (*down)(parent, cpu, sd); |
c09595f6 PZ |
1461 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1462 | parent = child; | |
1463 | goto down; | |
1464 | ||
1465 | up: | |
1466 | continue; | |
1467 | } | |
b6a86c74 | 1468 | (*up)(parent, cpu, sd); |
c09595f6 PZ |
1469 | |
1470 | child = parent; | |
1471 | parent = parent->parent; | |
1472 | if (parent) | |
1473 | goto up; | |
1474 | rcu_read_unlock(); | |
1475 | } | |
1476 | ||
c09595f6 PZ |
1477 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1478 | ||
1479 | /* | |
1480 | * Calculate and set the cpu's group shares. | |
1481 | */ | |
1482 | static void | |
b6a86c74 | 1483 | __update_group_shares_cpu(struct task_group *tg, int cpu, |
c8cba857 | 1484 | unsigned long sd_shares, unsigned long sd_rq_weight) |
c09595f6 PZ |
1485 | { |
1486 | int boost = 0; | |
1487 | unsigned long shares; | |
1488 | unsigned long rq_weight; | |
1489 | ||
c8cba857 | 1490 | if (!tg->se[cpu]) |
c09595f6 PZ |
1491 | return; |
1492 | ||
c8cba857 | 1493 | rq_weight = tg->cfs_rq[cpu]->load.weight; |
c09595f6 PZ |
1494 | |
1495 | /* | |
1496 | * If there are currently no tasks on the cpu pretend there is one of | |
1497 | * average load so that when a new task gets to run here it will not | |
1498 | * get delayed by group starvation. | |
1499 | */ | |
1500 | if (!rq_weight) { | |
1501 | boost = 1; | |
1502 | rq_weight = NICE_0_LOAD; | |
1503 | } | |
1504 | ||
c8cba857 PZ |
1505 | if (unlikely(rq_weight > sd_rq_weight)) |
1506 | rq_weight = sd_rq_weight; | |
1507 | ||
c09595f6 PZ |
1508 | /* |
1509 | * \Sum shares * rq_weight | |
1510 | * shares = ----------------------- | |
1511 | * \Sum rq_weight | |
1512 | * | |
1513 | */ | |
c8cba857 | 1514 | shares = (sd_shares * rq_weight) / (sd_rq_weight + 1); |
c09595f6 PZ |
1515 | |
1516 | /* | |
1517 | * record the actual number of shares, not the boosted amount. | |
1518 | */ | |
c8cba857 | 1519 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
c09595f6 PZ |
1520 | |
1521 | if (shares < MIN_SHARES) | |
1522 | shares = MIN_SHARES; | |
1523 | else if (shares > MAX_SHARES) | |
1524 | shares = MAX_SHARES; | |
1525 | ||
c8cba857 | 1526 | __set_se_shares(tg->se[cpu], shares); |
c09595f6 PZ |
1527 | } |
1528 | ||
1529 | /* | |
c8cba857 PZ |
1530 | * Re-compute the task group their per cpu shares over the given domain. |
1531 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1532 | * parent group depends on the shares of its child groups. | |
c09595f6 PZ |
1533 | */ |
1534 | static void | |
c8cba857 | 1535 | tg_shares_up(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1536 | { |
c8cba857 PZ |
1537 | unsigned long rq_weight = 0; |
1538 | unsigned long shares = 0; | |
1539 | int i; | |
c09595f6 | 1540 | |
c8cba857 PZ |
1541 | for_each_cpu_mask(i, sd->span) { |
1542 | rq_weight += tg->cfs_rq[i]->load.weight; | |
1543 | shares += tg->cfs_rq[i]->shares; | |
c09595f6 | 1544 | } |
c09595f6 | 1545 | |
c8cba857 PZ |
1546 | if ((!shares && rq_weight) || shares > tg->shares) |
1547 | shares = tg->shares; | |
1548 | ||
1549 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1550 | shares = tg->shares; | |
c09595f6 PZ |
1551 | |
1552 | for_each_cpu_mask(i, sd->span) { | |
1553 | struct rq *rq = cpu_rq(i); | |
1554 | unsigned long flags; | |
1555 | ||
1556 | spin_lock_irqsave(&rq->lock, flags); | |
c8cba857 | 1557 | __update_group_shares_cpu(tg, i, shares, rq_weight); |
c09595f6 PZ |
1558 | spin_unlock_irqrestore(&rq->lock, flags); |
1559 | } | |
c09595f6 PZ |
1560 | } |
1561 | ||
1562 | /* | |
c8cba857 PZ |
1563 | * Compute the cpu's hierarchical load factor for each task group. |
1564 | * This needs to be done in a top-down fashion because the load of a child | |
1565 | * group is a fraction of its parents load. | |
c09595f6 | 1566 | */ |
b6a86c74 | 1567 | static void |
c8cba857 | 1568 | tg_load_down(struct task_group *tg, int cpu, struct sched_domain *sd) |
c09595f6 | 1569 | { |
c8cba857 | 1570 | unsigned long load; |
c09595f6 | 1571 | |
c8cba857 PZ |
1572 | if (!tg->parent) { |
1573 | load = cpu_rq(cpu)->load.weight; | |
1574 | } else { | |
1575 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1576 | load *= tg->cfs_rq[cpu]->shares; | |
1577 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1578 | } | |
c09595f6 | 1579 | |
c8cba857 | 1580 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 PZ |
1581 | } |
1582 | ||
c8cba857 PZ |
1583 | static void |
1584 | tg_nop(struct task_group *tg, int cpu, struct sched_domain *sd) | |
c09595f6 | 1585 | { |
c09595f6 PZ |
1586 | } |
1587 | ||
c8cba857 | 1588 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1589 | { |
c8cba857 | 1590 | walk_tg_tree(tg_nop, tg_shares_up, 0, sd); |
4d8d595d PZ |
1591 | } |
1592 | ||
3e5459b4 PZ |
1593 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1594 | { | |
1595 | spin_unlock(&rq->lock); | |
1596 | update_shares(sd); | |
1597 | spin_lock(&rq->lock); | |
1598 | } | |
1599 | ||
c8cba857 | 1600 | static void update_h_load(int cpu) |
c09595f6 | 1601 | { |
c8cba857 | 1602 | walk_tg_tree(tg_load_down, tg_nop, cpu, NULL); |
c09595f6 PZ |
1603 | } |
1604 | ||
1605 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) | |
1606 | { | |
1607 | cfs_rq->shares = shares; | |
1608 | } | |
1609 | ||
1610 | #else | |
1611 | ||
c8cba857 | 1612 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1613 | { |
1614 | } | |
1615 | ||
3e5459b4 PZ |
1616 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1617 | { | |
1618 | } | |
1619 | ||
c09595f6 PZ |
1620 | #endif |
1621 | ||
18d95a28 PZ |
1622 | #endif |
1623 | ||
dd41f596 | 1624 | #include "sched_stats.h" |
dd41f596 | 1625 | #include "sched_idletask.c" |
5522d5d5 IM |
1626 | #include "sched_fair.c" |
1627 | #include "sched_rt.c" | |
dd41f596 IM |
1628 | #ifdef CONFIG_SCHED_DEBUG |
1629 | # include "sched_debug.c" | |
1630 | #endif | |
1631 | ||
1632 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1633 | #define for_each_class(class) \ |
1634 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1635 | |
c09595f6 | 1636 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1637 | { |
1638 | rq->nr_running++; | |
9c217245 IM |
1639 | } |
1640 | ||
c09595f6 | 1641 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1642 | { |
1643 | rq->nr_running--; | |
9c217245 IM |
1644 | } |
1645 | ||
45bf76df IM |
1646 | static void set_load_weight(struct task_struct *p) |
1647 | { | |
1648 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1649 | p->se.load.weight = prio_to_weight[0] * 2; |
1650 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1651 | return; | |
1652 | } | |
45bf76df | 1653 | |
dd41f596 IM |
1654 | /* |
1655 | * SCHED_IDLE tasks get minimal weight: | |
1656 | */ | |
1657 | if (p->policy == SCHED_IDLE) { | |
1658 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1659 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1660 | return; | |
1661 | } | |
71f8bd46 | 1662 | |
dd41f596 IM |
1663 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1664 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1665 | } |
1666 | ||
8159f87e | 1667 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1668 | { |
dd41f596 | 1669 | sched_info_queued(p); |
fd390f6a | 1670 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1671 | p->se.on_rq = 1; |
71f8bd46 IM |
1672 | } |
1673 | ||
69be72c1 | 1674 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1675 | { |
f02231e5 | 1676 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1677 | p->se.on_rq = 0; |
71f8bd46 IM |
1678 | } |
1679 | ||
14531189 | 1680 | /* |
dd41f596 | 1681 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1682 | */ |
14531189 IM |
1683 | static inline int __normal_prio(struct task_struct *p) |
1684 | { | |
dd41f596 | 1685 | return p->static_prio; |
14531189 IM |
1686 | } |
1687 | ||
b29739f9 IM |
1688 | /* |
1689 | * Calculate the expected normal priority: i.e. priority | |
1690 | * without taking RT-inheritance into account. Might be | |
1691 | * boosted by interactivity modifiers. Changes upon fork, | |
1692 | * setprio syscalls, and whenever the interactivity | |
1693 | * estimator recalculates. | |
1694 | */ | |
36c8b586 | 1695 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1696 | { |
1697 | int prio; | |
1698 | ||
e05606d3 | 1699 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1700 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1701 | else | |
1702 | prio = __normal_prio(p); | |
1703 | return prio; | |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * Calculate the current priority, i.e. the priority | |
1708 | * taken into account by the scheduler. This value might | |
1709 | * be boosted by RT tasks, or might be boosted by | |
1710 | * interactivity modifiers. Will be RT if the task got | |
1711 | * RT-boosted. If not then it returns p->normal_prio. | |
1712 | */ | |
36c8b586 | 1713 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1714 | { |
1715 | p->normal_prio = normal_prio(p); | |
1716 | /* | |
1717 | * If we are RT tasks or we were boosted to RT priority, | |
1718 | * keep the priority unchanged. Otherwise, update priority | |
1719 | * to the normal priority: | |
1720 | */ | |
1721 | if (!rt_prio(p->prio)) | |
1722 | return p->normal_prio; | |
1723 | return p->prio; | |
1724 | } | |
1725 | ||
1da177e4 | 1726 | /* |
dd41f596 | 1727 | * activate_task - move a task to the runqueue. |
1da177e4 | 1728 | */ |
dd41f596 | 1729 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1730 | { |
d9514f6c | 1731 | if (task_contributes_to_load(p)) |
dd41f596 | 1732 | rq->nr_uninterruptible--; |
1da177e4 | 1733 | |
8159f87e | 1734 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1735 | inc_nr_running(rq); |
1da177e4 LT |
1736 | } |
1737 | ||
1da177e4 LT |
1738 | /* |
1739 | * deactivate_task - remove a task from the runqueue. | |
1740 | */ | |
2e1cb74a | 1741 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1742 | { |
d9514f6c | 1743 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1744 | rq->nr_uninterruptible++; |
1745 | ||
69be72c1 | 1746 | dequeue_task(rq, p, sleep); |
c09595f6 | 1747 | dec_nr_running(rq); |
1da177e4 LT |
1748 | } |
1749 | ||
1da177e4 LT |
1750 | /** |
1751 | * task_curr - is this task currently executing on a CPU? | |
1752 | * @p: the task in question. | |
1753 | */ | |
36c8b586 | 1754 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1755 | { |
1756 | return cpu_curr(task_cpu(p)) == p; | |
1757 | } | |
1758 | ||
dd41f596 IM |
1759 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1760 | { | |
6f505b16 | 1761 | set_task_rq(p, cpu); |
dd41f596 | 1762 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1763 | /* |
1764 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1765 | * successfuly executed on another CPU. We must ensure that updates of | |
1766 | * per-task data have been completed by this moment. | |
1767 | */ | |
1768 | smp_wmb(); | |
dd41f596 | 1769 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1770 | #endif |
2dd73a4f PW |
1771 | } |
1772 | ||
cb469845 SR |
1773 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1774 | const struct sched_class *prev_class, | |
1775 | int oldprio, int running) | |
1776 | { | |
1777 | if (prev_class != p->sched_class) { | |
1778 | if (prev_class->switched_from) | |
1779 | prev_class->switched_from(rq, p, running); | |
1780 | p->sched_class->switched_to(rq, p, running); | |
1781 | } else | |
1782 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1783 | } | |
1784 | ||
1da177e4 | 1785 | #ifdef CONFIG_SMP |
c65cc870 | 1786 | |
e958b360 TG |
1787 | /* Used instead of source_load when we know the type == 0 */ |
1788 | static unsigned long weighted_cpuload(const int cpu) | |
1789 | { | |
1790 | return cpu_rq(cpu)->load.weight; | |
1791 | } | |
1792 | ||
cc367732 IM |
1793 | /* |
1794 | * Is this task likely cache-hot: | |
1795 | */ | |
e7693a36 | 1796 | static int |
cc367732 IM |
1797 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1798 | { | |
1799 | s64 delta; | |
1800 | ||
f540a608 IM |
1801 | /* |
1802 | * Buddy candidates are cache hot: | |
1803 | */ | |
d25ce4cd | 1804 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1805 | return 1; |
1806 | ||
cc367732 IM |
1807 | if (p->sched_class != &fair_sched_class) |
1808 | return 0; | |
1809 | ||
6bc1665b IM |
1810 | if (sysctl_sched_migration_cost == -1) |
1811 | return 1; | |
1812 | if (sysctl_sched_migration_cost == 0) | |
1813 | return 0; | |
1814 | ||
cc367732 IM |
1815 | delta = now - p->se.exec_start; |
1816 | ||
1817 | return delta < (s64)sysctl_sched_migration_cost; | |
1818 | } | |
1819 | ||
1820 | ||
dd41f596 | 1821 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1822 | { |
dd41f596 IM |
1823 | int old_cpu = task_cpu(p); |
1824 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1825 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1826 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1827 | u64 clock_offset; |
dd41f596 IM |
1828 | |
1829 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1830 | |
1831 | #ifdef CONFIG_SCHEDSTATS | |
1832 | if (p->se.wait_start) | |
1833 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1834 | if (p->se.sleep_start) |
1835 | p->se.sleep_start -= clock_offset; | |
1836 | if (p->se.block_start) | |
1837 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1838 | if (old_cpu != new_cpu) { |
1839 | schedstat_inc(p, se.nr_migrations); | |
1840 | if (task_hot(p, old_rq->clock, NULL)) | |
1841 | schedstat_inc(p, se.nr_forced2_migrations); | |
1842 | } | |
6cfb0d5d | 1843 | #endif |
2830cf8c SV |
1844 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1845 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1846 | |
1847 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1848 | } |
1849 | ||
70b97a7f | 1850 | struct migration_req { |
1da177e4 | 1851 | struct list_head list; |
1da177e4 | 1852 | |
36c8b586 | 1853 | struct task_struct *task; |
1da177e4 LT |
1854 | int dest_cpu; |
1855 | ||
1da177e4 | 1856 | struct completion done; |
70b97a7f | 1857 | }; |
1da177e4 LT |
1858 | |
1859 | /* | |
1860 | * The task's runqueue lock must be held. | |
1861 | * Returns true if you have to wait for migration thread. | |
1862 | */ | |
36c8b586 | 1863 | static int |
70b97a7f | 1864 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1865 | { |
70b97a7f | 1866 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1867 | |
1868 | /* | |
1869 | * If the task is not on a runqueue (and not running), then | |
1870 | * it is sufficient to simply update the task's cpu field. | |
1871 | */ | |
dd41f596 | 1872 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1873 | set_task_cpu(p, dest_cpu); |
1874 | return 0; | |
1875 | } | |
1876 | ||
1877 | init_completion(&req->done); | |
1da177e4 LT |
1878 | req->task = p; |
1879 | req->dest_cpu = dest_cpu; | |
1880 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1881 | |
1da177e4 LT |
1882 | return 1; |
1883 | } | |
1884 | ||
1885 | /* | |
1886 | * wait_task_inactive - wait for a thread to unschedule. | |
1887 | * | |
1888 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1889 | * else this function might spin for a *long* time. This function can't | |
1890 | * be called with interrupts off, or it may introduce deadlock with | |
1891 | * smp_call_function() if an IPI is sent by the same process we are | |
1892 | * waiting to become inactive. | |
1893 | */ | |
36c8b586 | 1894 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1895 | { |
1896 | unsigned long flags; | |
dd41f596 | 1897 | int running, on_rq; |
70b97a7f | 1898 | struct rq *rq; |
1da177e4 | 1899 | |
3a5c359a AK |
1900 | for (;;) { |
1901 | /* | |
1902 | * We do the initial early heuristics without holding | |
1903 | * any task-queue locks at all. We'll only try to get | |
1904 | * the runqueue lock when things look like they will | |
1905 | * work out! | |
1906 | */ | |
1907 | rq = task_rq(p); | |
fa490cfd | 1908 | |
3a5c359a AK |
1909 | /* |
1910 | * If the task is actively running on another CPU | |
1911 | * still, just relax and busy-wait without holding | |
1912 | * any locks. | |
1913 | * | |
1914 | * NOTE! Since we don't hold any locks, it's not | |
1915 | * even sure that "rq" stays as the right runqueue! | |
1916 | * But we don't care, since "task_running()" will | |
1917 | * return false if the runqueue has changed and p | |
1918 | * is actually now running somewhere else! | |
1919 | */ | |
1920 | while (task_running(rq, p)) | |
1921 | cpu_relax(); | |
fa490cfd | 1922 | |
3a5c359a AK |
1923 | /* |
1924 | * Ok, time to look more closely! We need the rq | |
1925 | * lock now, to be *sure*. If we're wrong, we'll | |
1926 | * just go back and repeat. | |
1927 | */ | |
1928 | rq = task_rq_lock(p, &flags); | |
1929 | running = task_running(rq, p); | |
1930 | on_rq = p->se.on_rq; | |
1931 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1932 | |
3a5c359a AK |
1933 | /* |
1934 | * Was it really running after all now that we | |
1935 | * checked with the proper locks actually held? | |
1936 | * | |
1937 | * Oops. Go back and try again.. | |
1938 | */ | |
1939 | if (unlikely(running)) { | |
1940 | cpu_relax(); | |
1941 | continue; | |
1942 | } | |
fa490cfd | 1943 | |
3a5c359a AK |
1944 | /* |
1945 | * It's not enough that it's not actively running, | |
1946 | * it must be off the runqueue _entirely_, and not | |
1947 | * preempted! | |
1948 | * | |
1949 | * So if it wa still runnable (but just not actively | |
1950 | * running right now), it's preempted, and we should | |
1951 | * yield - it could be a while. | |
1952 | */ | |
1953 | if (unlikely(on_rq)) { | |
1954 | schedule_timeout_uninterruptible(1); | |
1955 | continue; | |
1956 | } | |
fa490cfd | 1957 | |
3a5c359a AK |
1958 | /* |
1959 | * Ahh, all good. It wasn't running, and it wasn't | |
1960 | * runnable, which means that it will never become | |
1961 | * running in the future either. We're all done! | |
1962 | */ | |
1963 | break; | |
1964 | } | |
1da177e4 LT |
1965 | } |
1966 | ||
1967 | /*** | |
1968 | * kick_process - kick a running thread to enter/exit the kernel | |
1969 | * @p: the to-be-kicked thread | |
1970 | * | |
1971 | * Cause a process which is running on another CPU to enter | |
1972 | * kernel-mode, without any delay. (to get signals handled.) | |
1973 | * | |
1974 | * NOTE: this function doesnt have to take the runqueue lock, | |
1975 | * because all it wants to ensure is that the remote task enters | |
1976 | * the kernel. If the IPI races and the task has been migrated | |
1977 | * to another CPU then no harm is done and the purpose has been | |
1978 | * achieved as well. | |
1979 | */ | |
36c8b586 | 1980 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1981 | { |
1982 | int cpu; | |
1983 | ||
1984 | preempt_disable(); | |
1985 | cpu = task_cpu(p); | |
1986 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1987 | smp_send_reschedule(cpu); | |
1988 | preempt_enable(); | |
1989 | } | |
1990 | ||
1991 | /* | |
2dd73a4f PW |
1992 | * Return a low guess at the load of a migration-source cpu weighted |
1993 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1994 | * |
1995 | * We want to under-estimate the load of migration sources, to | |
1996 | * balance conservatively. | |
1997 | */ | |
a9957449 | 1998 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1999 | { |
70b97a7f | 2000 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2001 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2002 | |
93b75217 | 2003 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2004 | return total; |
b910472d | 2005 | |
dd41f596 | 2006 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2007 | } |
2008 | ||
2009 | /* | |
2dd73a4f PW |
2010 | * Return a high guess at the load of a migration-target cpu weighted |
2011 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2012 | */ |
a9957449 | 2013 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2014 | { |
70b97a7f | 2015 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2016 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2017 | |
93b75217 | 2018 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2019 | return total; |
3b0bd9bc | 2020 | |
dd41f596 | 2021 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2022 | } |
2023 | ||
147cbb4b NP |
2024 | /* |
2025 | * find_idlest_group finds and returns the least busy CPU group within the | |
2026 | * domain. | |
2027 | */ | |
2028 | static struct sched_group * | |
2029 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2030 | { | |
2031 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2032 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2033 | int load_idx = sd->forkexec_idx; | |
2034 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2035 | ||
2036 | do { | |
2037 | unsigned long load, avg_load; | |
2038 | int local_group; | |
2039 | int i; | |
2040 | ||
da5a5522 BD |
2041 | /* Skip over this group if it has no CPUs allowed */ |
2042 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 2043 | continue; |
da5a5522 | 2044 | |
147cbb4b | 2045 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
2046 | |
2047 | /* Tally up the load of all CPUs in the group */ | |
2048 | avg_load = 0; | |
2049 | ||
2050 | for_each_cpu_mask(i, group->cpumask) { | |
2051 | /* Bias balancing toward cpus of our domain */ | |
2052 | if (local_group) | |
2053 | load = source_load(i, load_idx); | |
2054 | else | |
2055 | load = target_load(i, load_idx); | |
2056 | ||
2057 | avg_load += load; | |
2058 | } | |
2059 | ||
2060 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2061 | avg_load = sg_div_cpu_power(group, |
2062 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2063 | |
2064 | if (local_group) { | |
2065 | this_load = avg_load; | |
2066 | this = group; | |
2067 | } else if (avg_load < min_load) { | |
2068 | min_load = avg_load; | |
2069 | idlest = group; | |
2070 | } | |
3a5c359a | 2071 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2072 | |
2073 | if (!idlest || 100*this_load < imbalance*min_load) | |
2074 | return NULL; | |
2075 | return idlest; | |
2076 | } | |
2077 | ||
2078 | /* | |
0feaece9 | 2079 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2080 | */ |
95cdf3b7 | 2081 | static int |
7c16ec58 MT |
2082 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
2083 | cpumask_t *tmp) | |
147cbb4b NP |
2084 | { |
2085 | unsigned long load, min_load = ULONG_MAX; | |
2086 | int idlest = -1; | |
2087 | int i; | |
2088 | ||
da5a5522 | 2089 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2090 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2091 | |
7c16ec58 | 2092 | for_each_cpu_mask(i, *tmp) { |
2dd73a4f | 2093 | load = weighted_cpuload(i); |
147cbb4b NP |
2094 | |
2095 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2096 | min_load = load; | |
2097 | idlest = i; | |
2098 | } | |
2099 | } | |
2100 | ||
2101 | return idlest; | |
2102 | } | |
2103 | ||
476d139c NP |
2104 | /* |
2105 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2106 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2107 | * SD_BALANCE_EXEC. | |
2108 | * | |
2109 | * Balance, ie. select the least loaded group. | |
2110 | * | |
2111 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2112 | * | |
2113 | * preempt must be disabled. | |
2114 | */ | |
2115 | static int sched_balance_self(int cpu, int flag) | |
2116 | { | |
2117 | struct task_struct *t = current; | |
2118 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2119 | |
c96d145e | 2120 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2121 | /* |
2122 | * If power savings logic is enabled for a domain, stop there. | |
2123 | */ | |
5c45bf27 SS |
2124 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2125 | break; | |
476d139c NP |
2126 | if (tmp->flags & flag) |
2127 | sd = tmp; | |
c96d145e | 2128 | } |
476d139c | 2129 | |
039a1c41 PZ |
2130 | if (sd) |
2131 | update_shares(sd); | |
2132 | ||
476d139c | 2133 | while (sd) { |
7c16ec58 | 2134 | cpumask_t span, tmpmask; |
476d139c | 2135 | struct sched_group *group; |
1a848870 SS |
2136 | int new_cpu, weight; |
2137 | ||
2138 | if (!(sd->flags & flag)) { | |
2139 | sd = sd->child; | |
2140 | continue; | |
2141 | } | |
476d139c NP |
2142 | |
2143 | span = sd->span; | |
2144 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2145 | if (!group) { |
2146 | sd = sd->child; | |
2147 | continue; | |
2148 | } | |
476d139c | 2149 | |
7c16ec58 | 2150 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2151 | if (new_cpu == -1 || new_cpu == cpu) { |
2152 | /* Now try balancing at a lower domain level of cpu */ | |
2153 | sd = sd->child; | |
2154 | continue; | |
2155 | } | |
476d139c | 2156 | |
1a848870 | 2157 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2158 | cpu = new_cpu; |
476d139c NP |
2159 | sd = NULL; |
2160 | weight = cpus_weight(span); | |
2161 | for_each_domain(cpu, tmp) { | |
2162 | if (weight <= cpus_weight(tmp->span)) | |
2163 | break; | |
2164 | if (tmp->flags & flag) | |
2165 | sd = tmp; | |
2166 | } | |
2167 | /* while loop will break here if sd == NULL */ | |
2168 | } | |
2169 | ||
2170 | return cpu; | |
2171 | } | |
2172 | ||
2173 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2174 | |
1da177e4 LT |
2175 | /*** |
2176 | * try_to_wake_up - wake up a thread | |
2177 | * @p: the to-be-woken-up thread | |
2178 | * @state: the mask of task states that can be woken | |
2179 | * @sync: do a synchronous wakeup? | |
2180 | * | |
2181 | * Put it on the run-queue if it's not already there. The "current" | |
2182 | * thread is always on the run-queue (except when the actual | |
2183 | * re-schedule is in progress), and as such you're allowed to do | |
2184 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2185 | * runnable without the overhead of this. | |
2186 | * | |
2187 | * returns failure only if the task is already active. | |
2188 | */ | |
36c8b586 | 2189 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2190 | { |
cc367732 | 2191 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2192 | unsigned long flags; |
2193 | long old_state; | |
70b97a7f | 2194 | struct rq *rq; |
1da177e4 | 2195 | |
b85d0667 IM |
2196 | if (!sched_feat(SYNC_WAKEUPS)) |
2197 | sync = 0; | |
2198 | ||
04e2f174 | 2199 | smp_wmb(); |
1da177e4 LT |
2200 | rq = task_rq_lock(p, &flags); |
2201 | old_state = p->state; | |
2202 | if (!(old_state & state)) | |
2203 | goto out; | |
2204 | ||
dd41f596 | 2205 | if (p->se.on_rq) |
1da177e4 LT |
2206 | goto out_running; |
2207 | ||
2208 | cpu = task_cpu(p); | |
cc367732 | 2209 | orig_cpu = cpu; |
1da177e4 LT |
2210 | this_cpu = smp_processor_id(); |
2211 | ||
2212 | #ifdef CONFIG_SMP | |
2213 | if (unlikely(task_running(rq, p))) | |
2214 | goto out_activate; | |
2215 | ||
5d2f5a61 DA |
2216 | cpu = p->sched_class->select_task_rq(p, sync); |
2217 | if (cpu != orig_cpu) { | |
2218 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2219 | task_rq_unlock(rq, &flags); |
2220 | /* might preempt at this point */ | |
2221 | rq = task_rq_lock(p, &flags); | |
2222 | old_state = p->state; | |
2223 | if (!(old_state & state)) | |
2224 | goto out; | |
dd41f596 | 2225 | if (p->se.on_rq) |
1da177e4 LT |
2226 | goto out_running; |
2227 | ||
2228 | this_cpu = smp_processor_id(); | |
2229 | cpu = task_cpu(p); | |
2230 | } | |
2231 | ||
e7693a36 GH |
2232 | #ifdef CONFIG_SCHEDSTATS |
2233 | schedstat_inc(rq, ttwu_count); | |
2234 | if (cpu == this_cpu) | |
2235 | schedstat_inc(rq, ttwu_local); | |
2236 | else { | |
2237 | struct sched_domain *sd; | |
2238 | for_each_domain(this_cpu, sd) { | |
2239 | if (cpu_isset(cpu, sd->span)) { | |
2240 | schedstat_inc(sd, ttwu_wake_remote); | |
2241 | break; | |
2242 | } | |
2243 | } | |
2244 | } | |
6d6bc0ad | 2245 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2246 | |
1da177e4 LT |
2247 | out_activate: |
2248 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2249 | schedstat_inc(p, se.nr_wakeups); |
2250 | if (sync) | |
2251 | schedstat_inc(p, se.nr_wakeups_sync); | |
2252 | if (orig_cpu != cpu) | |
2253 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2254 | if (cpu == this_cpu) | |
2255 | schedstat_inc(p, se.nr_wakeups_local); | |
2256 | else | |
2257 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2258 | update_rq_clock(rq); |
dd41f596 | 2259 | activate_task(rq, p, 1); |
1da177e4 LT |
2260 | success = 1; |
2261 | ||
2262 | out_running: | |
4ae7d5ce IM |
2263 | check_preempt_curr(rq, p); |
2264 | ||
1da177e4 | 2265 | p->state = TASK_RUNNING; |
9a897c5a SR |
2266 | #ifdef CONFIG_SMP |
2267 | if (p->sched_class->task_wake_up) | |
2268 | p->sched_class->task_wake_up(rq, p); | |
2269 | #endif | |
1da177e4 LT |
2270 | out: |
2271 | task_rq_unlock(rq, &flags); | |
2272 | ||
2273 | return success; | |
2274 | } | |
2275 | ||
7ad5b3a5 | 2276 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2277 | { |
d9514f6c | 2278 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2279 | } |
1da177e4 LT |
2280 | EXPORT_SYMBOL(wake_up_process); |
2281 | ||
7ad5b3a5 | 2282 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2283 | { |
2284 | return try_to_wake_up(p, state, 0); | |
2285 | } | |
2286 | ||
1da177e4 LT |
2287 | /* |
2288 | * Perform scheduler related setup for a newly forked process p. | |
2289 | * p is forked by current. | |
dd41f596 IM |
2290 | * |
2291 | * __sched_fork() is basic setup used by init_idle() too: | |
2292 | */ | |
2293 | static void __sched_fork(struct task_struct *p) | |
2294 | { | |
dd41f596 IM |
2295 | p->se.exec_start = 0; |
2296 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2297 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2298 | p->se.last_wakeup = 0; |
2299 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2300 | |
2301 | #ifdef CONFIG_SCHEDSTATS | |
2302 | p->se.wait_start = 0; | |
dd41f596 IM |
2303 | p->se.sum_sleep_runtime = 0; |
2304 | p->se.sleep_start = 0; | |
dd41f596 IM |
2305 | p->se.block_start = 0; |
2306 | p->se.sleep_max = 0; | |
2307 | p->se.block_max = 0; | |
2308 | p->se.exec_max = 0; | |
eba1ed4b | 2309 | p->se.slice_max = 0; |
dd41f596 | 2310 | p->se.wait_max = 0; |
6cfb0d5d | 2311 | #endif |
476d139c | 2312 | |
fa717060 | 2313 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2314 | p->se.on_rq = 0; |
4a55bd5e | 2315 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2316 | |
e107be36 AK |
2317 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2318 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2319 | #endif | |
2320 | ||
1da177e4 LT |
2321 | /* |
2322 | * We mark the process as running here, but have not actually | |
2323 | * inserted it onto the runqueue yet. This guarantees that | |
2324 | * nobody will actually run it, and a signal or other external | |
2325 | * event cannot wake it up and insert it on the runqueue either. | |
2326 | */ | |
2327 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2328 | } |
2329 | ||
2330 | /* | |
2331 | * fork()/clone()-time setup: | |
2332 | */ | |
2333 | void sched_fork(struct task_struct *p, int clone_flags) | |
2334 | { | |
2335 | int cpu = get_cpu(); | |
2336 | ||
2337 | __sched_fork(p); | |
2338 | ||
2339 | #ifdef CONFIG_SMP | |
2340 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2341 | #endif | |
02e4bac2 | 2342 | set_task_cpu(p, cpu); |
b29739f9 IM |
2343 | |
2344 | /* | |
2345 | * Make sure we do not leak PI boosting priority to the child: | |
2346 | */ | |
2347 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2348 | if (!rt_prio(p->prio)) |
2349 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2350 | |
52f17b6c | 2351 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2352 | if (likely(sched_info_on())) |
52f17b6c | 2353 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2354 | #endif |
d6077cb8 | 2355 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2356 | p->oncpu = 0; |
2357 | #endif | |
1da177e4 | 2358 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2359 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2360 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2361 | #endif |
476d139c | 2362 | put_cpu(); |
1da177e4 LT |
2363 | } |
2364 | ||
2365 | /* | |
2366 | * wake_up_new_task - wake up a newly created task for the first time. | |
2367 | * | |
2368 | * This function will do some initial scheduler statistics housekeeping | |
2369 | * that must be done for every newly created context, then puts the task | |
2370 | * on the runqueue and wakes it. | |
2371 | */ | |
7ad5b3a5 | 2372 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2373 | { |
2374 | unsigned long flags; | |
dd41f596 | 2375 | struct rq *rq; |
1da177e4 LT |
2376 | |
2377 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2378 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2379 | update_rq_clock(rq); |
1da177e4 LT |
2380 | |
2381 | p->prio = effective_prio(p); | |
2382 | ||
b9dca1e0 | 2383 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2384 | activate_task(rq, p, 0); |
1da177e4 | 2385 | } else { |
1da177e4 | 2386 | /* |
dd41f596 IM |
2387 | * Let the scheduling class do new task startup |
2388 | * management (if any): | |
1da177e4 | 2389 | */ |
ee0827d8 | 2390 | p->sched_class->task_new(rq, p); |
c09595f6 | 2391 | inc_nr_running(rq); |
1da177e4 | 2392 | } |
dd41f596 | 2393 | check_preempt_curr(rq, p); |
9a897c5a SR |
2394 | #ifdef CONFIG_SMP |
2395 | if (p->sched_class->task_wake_up) | |
2396 | p->sched_class->task_wake_up(rq, p); | |
2397 | #endif | |
dd41f596 | 2398 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2399 | } |
2400 | ||
e107be36 AK |
2401 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2402 | ||
2403 | /** | |
421cee29 RD |
2404 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2405 | * @notifier: notifier struct to register | |
e107be36 AK |
2406 | */ |
2407 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2408 | { | |
2409 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2410 | } | |
2411 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2412 | ||
2413 | /** | |
2414 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2415 | * @notifier: notifier struct to unregister |
e107be36 AK |
2416 | * |
2417 | * This is safe to call from within a preemption notifier. | |
2418 | */ | |
2419 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2420 | { | |
2421 | hlist_del(¬ifier->link); | |
2422 | } | |
2423 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2424 | ||
2425 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2426 | { | |
2427 | struct preempt_notifier *notifier; | |
2428 | struct hlist_node *node; | |
2429 | ||
2430 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2431 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2432 | } | |
2433 | ||
2434 | static void | |
2435 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2436 | struct task_struct *next) | |
2437 | { | |
2438 | struct preempt_notifier *notifier; | |
2439 | struct hlist_node *node; | |
2440 | ||
2441 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2442 | notifier->ops->sched_out(notifier, next); | |
2443 | } | |
2444 | ||
6d6bc0ad | 2445 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2446 | |
2447 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2448 | { | |
2449 | } | |
2450 | ||
2451 | static void | |
2452 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2453 | struct task_struct *next) | |
2454 | { | |
2455 | } | |
2456 | ||
6d6bc0ad | 2457 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2458 | |
4866cde0 NP |
2459 | /** |
2460 | * prepare_task_switch - prepare to switch tasks | |
2461 | * @rq: the runqueue preparing to switch | |
421cee29 | 2462 | * @prev: the current task that is being switched out |
4866cde0 NP |
2463 | * @next: the task we are going to switch to. |
2464 | * | |
2465 | * This is called with the rq lock held and interrupts off. It must | |
2466 | * be paired with a subsequent finish_task_switch after the context | |
2467 | * switch. | |
2468 | * | |
2469 | * prepare_task_switch sets up locking and calls architecture specific | |
2470 | * hooks. | |
2471 | */ | |
e107be36 AK |
2472 | static inline void |
2473 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2474 | struct task_struct *next) | |
4866cde0 | 2475 | { |
e107be36 | 2476 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2477 | prepare_lock_switch(rq, next); |
2478 | prepare_arch_switch(next); | |
2479 | } | |
2480 | ||
1da177e4 LT |
2481 | /** |
2482 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2483 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2484 | * @prev: the thread we just switched away from. |
2485 | * | |
4866cde0 NP |
2486 | * finish_task_switch must be called after the context switch, paired |
2487 | * with a prepare_task_switch call before the context switch. | |
2488 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2489 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2490 | * |
2491 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2492 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2493 | * with the lock held can cause deadlocks; see schedule() for |
2494 | * details.) | |
2495 | */ | |
a9957449 | 2496 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2497 | __releases(rq->lock) |
2498 | { | |
1da177e4 | 2499 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2500 | long prev_state; |
1da177e4 LT |
2501 | |
2502 | rq->prev_mm = NULL; | |
2503 | ||
2504 | /* | |
2505 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2506 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2507 | * schedule one last time. The schedule call will never return, and |
2508 | * the scheduled task must drop that reference. | |
c394cc9f | 2509 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2510 | * still held, otherwise prev could be scheduled on another cpu, die |
2511 | * there before we look at prev->state, and then the reference would | |
2512 | * be dropped twice. | |
2513 | * Manfred Spraul <manfred@colorfullife.com> | |
2514 | */ | |
55a101f8 | 2515 | prev_state = prev->state; |
4866cde0 NP |
2516 | finish_arch_switch(prev); |
2517 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2518 | #ifdef CONFIG_SMP |
2519 | if (current->sched_class->post_schedule) | |
2520 | current->sched_class->post_schedule(rq); | |
2521 | #endif | |
e8fa1362 | 2522 | |
e107be36 | 2523 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2524 | if (mm) |
2525 | mmdrop(mm); | |
c394cc9f | 2526 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2527 | /* |
2528 | * Remove function-return probe instances associated with this | |
2529 | * task and put them back on the free list. | |
9761eea8 | 2530 | */ |
c6fd91f0 | 2531 | kprobe_flush_task(prev); |
1da177e4 | 2532 | put_task_struct(prev); |
c6fd91f0 | 2533 | } |
1da177e4 LT |
2534 | } |
2535 | ||
2536 | /** | |
2537 | * schedule_tail - first thing a freshly forked thread must call. | |
2538 | * @prev: the thread we just switched away from. | |
2539 | */ | |
36c8b586 | 2540 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2541 | __releases(rq->lock) |
2542 | { | |
70b97a7f IM |
2543 | struct rq *rq = this_rq(); |
2544 | ||
4866cde0 NP |
2545 | finish_task_switch(rq, prev); |
2546 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2547 | /* In this case, finish_task_switch does not reenable preemption */ | |
2548 | preempt_enable(); | |
2549 | #endif | |
1da177e4 | 2550 | if (current->set_child_tid) |
b488893a | 2551 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2552 | } |
2553 | ||
2554 | /* | |
2555 | * context_switch - switch to the new MM and the new | |
2556 | * thread's register state. | |
2557 | */ | |
dd41f596 | 2558 | static inline void |
70b97a7f | 2559 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2560 | struct task_struct *next) |
1da177e4 | 2561 | { |
dd41f596 | 2562 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2563 | |
e107be36 | 2564 | prepare_task_switch(rq, prev, next); |
dd41f596 IM |
2565 | mm = next->mm; |
2566 | oldmm = prev->active_mm; | |
9226d125 ZA |
2567 | /* |
2568 | * For paravirt, this is coupled with an exit in switch_to to | |
2569 | * combine the page table reload and the switch backend into | |
2570 | * one hypercall. | |
2571 | */ | |
2572 | arch_enter_lazy_cpu_mode(); | |
2573 | ||
dd41f596 | 2574 | if (unlikely(!mm)) { |
1da177e4 LT |
2575 | next->active_mm = oldmm; |
2576 | atomic_inc(&oldmm->mm_count); | |
2577 | enter_lazy_tlb(oldmm, next); | |
2578 | } else | |
2579 | switch_mm(oldmm, mm, next); | |
2580 | ||
dd41f596 | 2581 | if (unlikely(!prev->mm)) { |
1da177e4 | 2582 | prev->active_mm = NULL; |
1da177e4 LT |
2583 | rq->prev_mm = oldmm; |
2584 | } | |
3a5f5e48 IM |
2585 | /* |
2586 | * Since the runqueue lock will be released by the next | |
2587 | * task (which is an invalid locking op but in the case | |
2588 | * of the scheduler it's an obvious special-case), so we | |
2589 | * do an early lockdep release here: | |
2590 | */ | |
2591 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2592 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2593 | #endif |
1da177e4 LT |
2594 | |
2595 | /* Here we just switch the register state and the stack. */ | |
2596 | switch_to(prev, next, prev); | |
2597 | ||
dd41f596 IM |
2598 | barrier(); |
2599 | /* | |
2600 | * this_rq must be evaluated again because prev may have moved | |
2601 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2602 | * frame will be invalid. | |
2603 | */ | |
2604 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2605 | } |
2606 | ||
2607 | /* | |
2608 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2609 | * | |
2610 | * externally visible scheduler statistics: current number of runnable | |
2611 | * threads, current number of uninterruptible-sleeping threads, total | |
2612 | * number of context switches performed since bootup. | |
2613 | */ | |
2614 | unsigned long nr_running(void) | |
2615 | { | |
2616 | unsigned long i, sum = 0; | |
2617 | ||
2618 | for_each_online_cpu(i) | |
2619 | sum += cpu_rq(i)->nr_running; | |
2620 | ||
2621 | return sum; | |
2622 | } | |
2623 | ||
2624 | unsigned long nr_uninterruptible(void) | |
2625 | { | |
2626 | unsigned long i, sum = 0; | |
2627 | ||
0a945022 | 2628 | for_each_possible_cpu(i) |
1da177e4 LT |
2629 | sum += cpu_rq(i)->nr_uninterruptible; |
2630 | ||
2631 | /* | |
2632 | * Since we read the counters lockless, it might be slightly | |
2633 | * inaccurate. Do not allow it to go below zero though: | |
2634 | */ | |
2635 | if (unlikely((long)sum < 0)) | |
2636 | sum = 0; | |
2637 | ||
2638 | return sum; | |
2639 | } | |
2640 | ||
2641 | unsigned long long nr_context_switches(void) | |
2642 | { | |
cc94abfc SR |
2643 | int i; |
2644 | unsigned long long sum = 0; | |
1da177e4 | 2645 | |
0a945022 | 2646 | for_each_possible_cpu(i) |
1da177e4 LT |
2647 | sum += cpu_rq(i)->nr_switches; |
2648 | ||
2649 | return sum; | |
2650 | } | |
2651 | ||
2652 | unsigned long nr_iowait(void) | |
2653 | { | |
2654 | unsigned long i, sum = 0; | |
2655 | ||
0a945022 | 2656 | for_each_possible_cpu(i) |
1da177e4 LT |
2657 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2658 | ||
2659 | return sum; | |
2660 | } | |
2661 | ||
db1b1fef JS |
2662 | unsigned long nr_active(void) |
2663 | { | |
2664 | unsigned long i, running = 0, uninterruptible = 0; | |
2665 | ||
2666 | for_each_online_cpu(i) { | |
2667 | running += cpu_rq(i)->nr_running; | |
2668 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2669 | } | |
2670 | ||
2671 | if (unlikely((long)uninterruptible < 0)) | |
2672 | uninterruptible = 0; | |
2673 | ||
2674 | return running + uninterruptible; | |
2675 | } | |
2676 | ||
48f24c4d | 2677 | /* |
dd41f596 IM |
2678 | * Update rq->cpu_load[] statistics. This function is usually called every |
2679 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2680 | */ |
dd41f596 | 2681 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2682 | { |
495eca49 | 2683 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2684 | int i, scale; |
2685 | ||
2686 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2687 | |
2688 | /* Update our load: */ | |
2689 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2690 | unsigned long old_load, new_load; | |
2691 | ||
2692 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2693 | ||
2694 | old_load = this_rq->cpu_load[i]; | |
2695 | new_load = this_load; | |
a25707f3 IM |
2696 | /* |
2697 | * Round up the averaging division if load is increasing. This | |
2698 | * prevents us from getting stuck on 9 if the load is 10, for | |
2699 | * example. | |
2700 | */ | |
2701 | if (new_load > old_load) | |
2702 | new_load += scale-1; | |
dd41f596 IM |
2703 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2704 | } | |
48f24c4d IM |
2705 | } |
2706 | ||
dd41f596 IM |
2707 | #ifdef CONFIG_SMP |
2708 | ||
1da177e4 LT |
2709 | /* |
2710 | * double_rq_lock - safely lock two runqueues | |
2711 | * | |
2712 | * Note this does not disable interrupts like task_rq_lock, | |
2713 | * you need to do so manually before calling. | |
2714 | */ | |
70b97a7f | 2715 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2716 | __acquires(rq1->lock) |
2717 | __acquires(rq2->lock) | |
2718 | { | |
054b9108 | 2719 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2720 | if (rq1 == rq2) { |
2721 | spin_lock(&rq1->lock); | |
2722 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2723 | } else { | |
c96d145e | 2724 | if (rq1 < rq2) { |
1da177e4 LT |
2725 | spin_lock(&rq1->lock); |
2726 | spin_lock(&rq2->lock); | |
2727 | } else { | |
2728 | spin_lock(&rq2->lock); | |
2729 | spin_lock(&rq1->lock); | |
2730 | } | |
2731 | } | |
6e82a3be IM |
2732 | update_rq_clock(rq1); |
2733 | update_rq_clock(rq2); | |
1da177e4 LT |
2734 | } |
2735 | ||
2736 | /* | |
2737 | * double_rq_unlock - safely unlock two runqueues | |
2738 | * | |
2739 | * Note this does not restore interrupts like task_rq_unlock, | |
2740 | * you need to do so manually after calling. | |
2741 | */ | |
70b97a7f | 2742 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2743 | __releases(rq1->lock) |
2744 | __releases(rq2->lock) | |
2745 | { | |
2746 | spin_unlock(&rq1->lock); | |
2747 | if (rq1 != rq2) | |
2748 | spin_unlock(&rq2->lock); | |
2749 | else | |
2750 | __release(rq2->lock); | |
2751 | } | |
2752 | ||
2753 | /* | |
2754 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2755 | */ | |
e8fa1362 | 2756 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2757 | __releases(this_rq->lock) |
2758 | __acquires(busiest->lock) | |
2759 | __acquires(this_rq->lock) | |
2760 | { | |
e8fa1362 SR |
2761 | int ret = 0; |
2762 | ||
054b9108 KK |
2763 | if (unlikely(!irqs_disabled())) { |
2764 | /* printk() doesn't work good under rq->lock */ | |
2765 | spin_unlock(&this_rq->lock); | |
2766 | BUG_ON(1); | |
2767 | } | |
1da177e4 | 2768 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2769 | if (busiest < this_rq) { |
1da177e4 LT |
2770 | spin_unlock(&this_rq->lock); |
2771 | spin_lock(&busiest->lock); | |
2772 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2773 | ret = 1; |
1da177e4 LT |
2774 | } else |
2775 | spin_lock(&busiest->lock); | |
2776 | } | |
e8fa1362 | 2777 | return ret; |
1da177e4 LT |
2778 | } |
2779 | ||
1da177e4 LT |
2780 | /* |
2781 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2782 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2783 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2784 | * the cpu_allowed mask is restored. |
2785 | */ | |
36c8b586 | 2786 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2787 | { |
70b97a7f | 2788 | struct migration_req req; |
1da177e4 | 2789 | unsigned long flags; |
70b97a7f | 2790 | struct rq *rq; |
1da177e4 LT |
2791 | |
2792 | rq = task_rq_lock(p, &flags); | |
2793 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2794 | || unlikely(cpu_is_offline(dest_cpu))) | |
2795 | goto out; | |
2796 | ||
2797 | /* force the process onto the specified CPU */ | |
2798 | if (migrate_task(p, dest_cpu, &req)) { | |
2799 | /* Need to wait for migration thread (might exit: take ref). */ | |
2800 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2801 | |
1da177e4 LT |
2802 | get_task_struct(mt); |
2803 | task_rq_unlock(rq, &flags); | |
2804 | wake_up_process(mt); | |
2805 | put_task_struct(mt); | |
2806 | wait_for_completion(&req.done); | |
36c8b586 | 2807 | |
1da177e4 LT |
2808 | return; |
2809 | } | |
2810 | out: | |
2811 | task_rq_unlock(rq, &flags); | |
2812 | } | |
2813 | ||
2814 | /* | |
476d139c NP |
2815 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2816 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2817 | */ |
2818 | void sched_exec(void) | |
2819 | { | |
1da177e4 | 2820 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2821 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2822 | put_cpu(); |
476d139c NP |
2823 | if (new_cpu != this_cpu) |
2824 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2825 | } |
2826 | ||
2827 | /* | |
2828 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2829 | * Both runqueues must be locked. | |
2830 | */ | |
dd41f596 IM |
2831 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2832 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2833 | { |
2e1cb74a | 2834 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2835 | set_task_cpu(p, this_cpu); |
dd41f596 | 2836 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2837 | /* |
2838 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2839 | * to be always true for them. | |
2840 | */ | |
dd41f596 | 2841 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2842 | } |
2843 | ||
2844 | /* | |
2845 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2846 | */ | |
858119e1 | 2847 | static |
70b97a7f | 2848 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2849 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2850 | int *all_pinned) |
1da177e4 LT |
2851 | { |
2852 | /* | |
2853 | * We do not migrate tasks that are: | |
2854 | * 1) running (obviously), or | |
2855 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2856 | * 3) are cache-hot on their current CPU. | |
2857 | */ | |
cc367732 IM |
2858 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2859 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2860 | return 0; |
cc367732 | 2861 | } |
81026794 NP |
2862 | *all_pinned = 0; |
2863 | ||
cc367732 IM |
2864 | if (task_running(rq, p)) { |
2865 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2866 | return 0; |
cc367732 | 2867 | } |
1da177e4 | 2868 | |
da84d961 IM |
2869 | /* |
2870 | * Aggressive migration if: | |
2871 | * 1) task is cache cold, or | |
2872 | * 2) too many balance attempts have failed. | |
2873 | */ | |
2874 | ||
6bc1665b IM |
2875 | if (!task_hot(p, rq->clock, sd) || |
2876 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2877 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2878 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2879 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2880 | schedstat_inc(p, se.nr_forced_migrations); |
2881 | } | |
da84d961 IM |
2882 | #endif |
2883 | return 1; | |
2884 | } | |
2885 | ||
cc367732 IM |
2886 | if (task_hot(p, rq->clock, sd)) { |
2887 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2888 | return 0; |
cc367732 | 2889 | } |
1da177e4 LT |
2890 | return 1; |
2891 | } | |
2892 | ||
e1d1484f PW |
2893 | static unsigned long |
2894 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2895 | unsigned long max_load_move, struct sched_domain *sd, | |
2896 | enum cpu_idle_type idle, int *all_pinned, | |
2897 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2898 | { |
051c6764 | 2899 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2900 | struct task_struct *p; |
2901 | long rem_load_move = max_load_move; | |
1da177e4 | 2902 | |
e1d1484f | 2903 | if (max_load_move == 0) |
1da177e4 LT |
2904 | goto out; |
2905 | ||
81026794 NP |
2906 | pinned = 1; |
2907 | ||
1da177e4 | 2908 | /* |
dd41f596 | 2909 | * Start the load-balancing iterator: |
1da177e4 | 2910 | */ |
dd41f596 IM |
2911 | p = iterator->start(iterator->arg); |
2912 | next: | |
b82d9fdd | 2913 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2914 | goto out; |
051c6764 PZ |
2915 | |
2916 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2917 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2918 | p = iterator->next(iterator->arg); |
2919 | goto next; | |
1da177e4 LT |
2920 | } |
2921 | ||
dd41f596 | 2922 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2923 | pulled++; |
dd41f596 | 2924 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2925 | |
2dd73a4f | 2926 | /* |
b82d9fdd | 2927 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2928 | */ |
e1d1484f | 2929 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2930 | if (p->prio < *this_best_prio) |
2931 | *this_best_prio = p->prio; | |
dd41f596 IM |
2932 | p = iterator->next(iterator->arg); |
2933 | goto next; | |
1da177e4 LT |
2934 | } |
2935 | out: | |
2936 | /* | |
e1d1484f | 2937 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2938 | * so we can safely collect pull_task() stats here rather than |
2939 | * inside pull_task(). | |
2940 | */ | |
2941 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2942 | |
2943 | if (all_pinned) | |
2944 | *all_pinned = pinned; | |
e1d1484f PW |
2945 | |
2946 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2947 | } |
2948 | ||
dd41f596 | 2949 | /* |
43010659 PW |
2950 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2951 | * this_rq, as part of a balancing operation within domain "sd". | |
2952 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2953 | * |
2954 | * Called with both runqueues locked. | |
2955 | */ | |
2956 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2957 | unsigned long max_load_move, |
dd41f596 IM |
2958 | struct sched_domain *sd, enum cpu_idle_type idle, |
2959 | int *all_pinned) | |
2960 | { | |
5522d5d5 | 2961 | const struct sched_class *class = sched_class_highest; |
43010659 | 2962 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2963 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2964 | |
2965 | do { | |
43010659 PW |
2966 | total_load_moved += |
2967 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2968 | max_load_move - total_load_moved, |
a4ac01c3 | 2969 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2970 | class = class->next; |
43010659 | 2971 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2972 | |
43010659 PW |
2973 | return total_load_moved > 0; |
2974 | } | |
2975 | ||
e1d1484f PW |
2976 | static int |
2977 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2978 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2979 | struct rq_iterator *iterator) | |
2980 | { | |
2981 | struct task_struct *p = iterator->start(iterator->arg); | |
2982 | int pinned = 0; | |
2983 | ||
2984 | while (p) { | |
2985 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2986 | pull_task(busiest, p, this_rq, this_cpu); | |
2987 | /* | |
2988 | * Right now, this is only the second place pull_task() | |
2989 | * is called, so we can safely collect pull_task() | |
2990 | * stats here rather than inside pull_task(). | |
2991 | */ | |
2992 | schedstat_inc(sd, lb_gained[idle]); | |
2993 | ||
2994 | return 1; | |
2995 | } | |
2996 | p = iterator->next(iterator->arg); | |
2997 | } | |
2998 | ||
2999 | return 0; | |
3000 | } | |
3001 | ||
43010659 PW |
3002 | /* |
3003 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3004 | * part of active balancing operations within "domain". | |
3005 | * Returns 1 if successful and 0 otherwise. | |
3006 | * | |
3007 | * Called with both runqueues locked. | |
3008 | */ | |
3009 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3010 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3011 | { | |
5522d5d5 | 3012 | const struct sched_class *class; |
43010659 PW |
3013 | |
3014 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3015 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3016 | return 1; |
3017 | ||
3018 | return 0; | |
dd41f596 IM |
3019 | } |
3020 | ||
1da177e4 LT |
3021 | /* |
3022 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3023 | * domain. It calculates and returns the amount of weighted load which |
3024 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3025 | */ |
3026 | static struct sched_group * | |
3027 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3028 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 3029 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
3030 | { |
3031 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3032 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3033 | unsigned long max_pull; |
2dd73a4f PW |
3034 | unsigned long busiest_load_per_task, busiest_nr_running; |
3035 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3036 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3037 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3038 | int power_savings_balance = 1; | |
3039 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3040 | unsigned long min_nr_running = ULONG_MAX; | |
3041 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3042 | #endif | |
1da177e4 LT |
3043 | |
3044 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3045 | busiest_load_per_task = busiest_nr_running = 0; |
3046 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3047 | |
d15bcfdb | 3048 | if (idle == CPU_NOT_IDLE) |
7897986b | 3049 | load_idx = sd->busy_idx; |
d15bcfdb | 3050 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3051 | load_idx = sd->newidle_idx; |
3052 | else | |
3053 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3054 | |
3055 | do { | |
908a7c1b | 3056 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3057 | int local_group; |
3058 | int i; | |
908a7c1b | 3059 | int __group_imb = 0; |
783609c6 | 3060 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3061 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3062 | unsigned long sum_avg_load_per_task; |
3063 | unsigned long avg_load_per_task; | |
1da177e4 LT |
3064 | |
3065 | local_group = cpu_isset(this_cpu, group->cpumask); | |
3066 | ||
783609c6 SS |
3067 | if (local_group) |
3068 | balance_cpu = first_cpu(group->cpumask); | |
3069 | ||
1da177e4 | 3070 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3071 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3072 | sum_avg_load_per_task = avg_load_per_task = 0; |
3073 | ||
908a7c1b KC |
3074 | max_cpu_load = 0; |
3075 | min_cpu_load = ~0UL; | |
1da177e4 LT |
3076 | |
3077 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
3078 | struct rq *rq; |
3079 | ||
3080 | if (!cpu_isset(i, *cpus)) | |
3081 | continue; | |
3082 | ||
3083 | rq = cpu_rq(i); | |
2dd73a4f | 3084 | |
9439aab8 | 3085 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3086 | *sd_idle = 0; |
3087 | ||
1da177e4 | 3088 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3089 | if (local_group) { |
3090 | if (idle_cpu(i) && !first_idle_cpu) { | |
3091 | first_idle_cpu = 1; | |
3092 | balance_cpu = i; | |
3093 | } | |
3094 | ||
a2000572 | 3095 | load = target_load(i, load_idx); |
908a7c1b | 3096 | } else { |
a2000572 | 3097 | load = source_load(i, load_idx); |
908a7c1b KC |
3098 | if (load > max_cpu_load) |
3099 | max_cpu_load = load; | |
3100 | if (min_cpu_load > load) | |
3101 | min_cpu_load = load; | |
3102 | } | |
1da177e4 LT |
3103 | |
3104 | avg_load += load; | |
2dd73a4f | 3105 | sum_nr_running += rq->nr_running; |
dd41f596 | 3106 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3107 | |
3108 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3109 | } |
3110 | ||
783609c6 SS |
3111 | /* |
3112 | * First idle cpu or the first cpu(busiest) in this sched group | |
3113 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3114 | * domains. In the newly idle case, we will allow all the cpu's |
3115 | * to do the newly idle load balance. | |
783609c6 | 3116 | */ |
9439aab8 SS |
3117 | if (idle != CPU_NEWLY_IDLE && local_group && |
3118 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3119 | *balance = 0; |
3120 | goto ret; | |
3121 | } | |
3122 | ||
1da177e4 | 3123 | total_load += avg_load; |
5517d86b | 3124 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3125 | |
3126 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3127 | avg_load = sg_div_cpu_power(group, |
3128 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3129 | |
408ed066 PZ |
3130 | |
3131 | /* | |
3132 | * Consider the group unbalanced when the imbalance is larger | |
3133 | * than the average weight of two tasks. | |
3134 | * | |
3135 | * APZ: with cgroup the avg task weight can vary wildly and | |
3136 | * might not be a suitable number - should we keep a | |
3137 | * normalized nr_running number somewhere that negates | |
3138 | * the hierarchy? | |
3139 | */ | |
3140 | avg_load_per_task = sg_div_cpu_power(group, | |
3141 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3142 | ||
3143 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3144 | __group_imb = 1; |
3145 | ||
5517d86b | 3146 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3147 | |
1da177e4 LT |
3148 | if (local_group) { |
3149 | this_load = avg_load; | |
3150 | this = group; | |
2dd73a4f PW |
3151 | this_nr_running = sum_nr_running; |
3152 | this_load_per_task = sum_weighted_load; | |
3153 | } else if (avg_load > max_load && | |
908a7c1b | 3154 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3155 | max_load = avg_load; |
3156 | busiest = group; | |
2dd73a4f PW |
3157 | busiest_nr_running = sum_nr_running; |
3158 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3159 | group_imb = __group_imb; |
1da177e4 | 3160 | } |
5c45bf27 SS |
3161 | |
3162 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3163 | /* | |
3164 | * Busy processors will not participate in power savings | |
3165 | * balance. | |
3166 | */ | |
dd41f596 IM |
3167 | if (idle == CPU_NOT_IDLE || |
3168 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3169 | goto group_next; | |
5c45bf27 SS |
3170 | |
3171 | /* | |
3172 | * If the local group is idle or completely loaded | |
3173 | * no need to do power savings balance at this domain | |
3174 | */ | |
3175 | if (local_group && (this_nr_running >= group_capacity || | |
3176 | !this_nr_running)) | |
3177 | power_savings_balance = 0; | |
3178 | ||
dd41f596 | 3179 | /* |
5c45bf27 SS |
3180 | * If a group is already running at full capacity or idle, |
3181 | * don't include that group in power savings calculations | |
dd41f596 IM |
3182 | */ |
3183 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3184 | || !sum_nr_running) |
dd41f596 | 3185 | goto group_next; |
5c45bf27 | 3186 | |
dd41f596 | 3187 | /* |
5c45bf27 | 3188 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3189 | * This is the group from where we need to pick up the load |
3190 | * for saving power | |
3191 | */ | |
3192 | if ((sum_nr_running < min_nr_running) || | |
3193 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3194 | first_cpu(group->cpumask) < |
3195 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3196 | group_min = group; |
3197 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3198 | min_load_per_task = sum_weighted_load / |
3199 | sum_nr_running; | |
dd41f596 | 3200 | } |
5c45bf27 | 3201 | |
dd41f596 | 3202 | /* |
5c45bf27 | 3203 | * Calculate the group which is almost near its |
dd41f596 IM |
3204 | * capacity but still has some space to pick up some load |
3205 | * from other group and save more power | |
3206 | */ | |
3207 | if (sum_nr_running <= group_capacity - 1) { | |
3208 | if (sum_nr_running > leader_nr_running || | |
3209 | (sum_nr_running == leader_nr_running && | |
3210 | first_cpu(group->cpumask) > | |
3211 | first_cpu(group_leader->cpumask))) { | |
3212 | group_leader = group; | |
3213 | leader_nr_running = sum_nr_running; | |
3214 | } | |
48f24c4d | 3215 | } |
5c45bf27 SS |
3216 | group_next: |
3217 | #endif | |
1da177e4 LT |
3218 | group = group->next; |
3219 | } while (group != sd->groups); | |
3220 | ||
2dd73a4f | 3221 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3222 | goto out_balanced; |
3223 | ||
3224 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3225 | ||
3226 | if (this_load >= avg_load || | |
3227 | 100*max_load <= sd->imbalance_pct*this_load) | |
3228 | goto out_balanced; | |
3229 | ||
2dd73a4f | 3230 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3231 | if (group_imb) |
3232 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3233 | ||
1da177e4 LT |
3234 | /* |
3235 | * We're trying to get all the cpus to the average_load, so we don't | |
3236 | * want to push ourselves above the average load, nor do we wish to | |
3237 | * reduce the max loaded cpu below the average load, as either of these | |
3238 | * actions would just result in more rebalancing later, and ping-pong | |
3239 | * tasks around. Thus we look for the minimum possible imbalance. | |
3240 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3241 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3242 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3243 | * appear as very large values with unsigned longs. |
3244 | */ | |
2dd73a4f PW |
3245 | if (max_load <= busiest_load_per_task) |
3246 | goto out_balanced; | |
3247 | ||
3248 | /* | |
3249 | * In the presence of smp nice balancing, certain scenarios can have | |
3250 | * max load less than avg load(as we skip the groups at or below | |
3251 | * its cpu_power, while calculating max_load..) | |
3252 | */ | |
3253 | if (max_load < avg_load) { | |
3254 | *imbalance = 0; | |
3255 | goto small_imbalance; | |
3256 | } | |
0c117f1b SS |
3257 | |
3258 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3259 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3260 | |
1da177e4 | 3261 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3262 | *imbalance = min(max_pull * busiest->__cpu_power, |
3263 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3264 | / SCHED_LOAD_SCALE; |
3265 | ||
2dd73a4f PW |
3266 | /* |
3267 | * if *imbalance is less than the average load per runnable task | |
3268 | * there is no gaurantee that any tasks will be moved so we'll have | |
3269 | * a think about bumping its value to force at least one task to be | |
3270 | * moved | |
3271 | */ | |
7fd0d2dd | 3272 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3273 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3274 | unsigned int imbn; |
3275 | ||
3276 | small_imbalance: | |
3277 | pwr_move = pwr_now = 0; | |
3278 | imbn = 2; | |
3279 | if (this_nr_running) { | |
3280 | this_load_per_task /= this_nr_running; | |
3281 | if (busiest_load_per_task > this_load_per_task) | |
3282 | imbn = 1; | |
3283 | } else | |
408ed066 | 3284 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3285 | |
408ed066 | 3286 | if (max_load - this_load + 2*busiest_load_per_task >= |
dd41f596 | 3287 | busiest_load_per_task * imbn) { |
2dd73a4f | 3288 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3289 | return busiest; |
3290 | } | |
3291 | ||
3292 | /* | |
3293 | * OK, we don't have enough imbalance to justify moving tasks, | |
3294 | * however we may be able to increase total CPU power used by | |
3295 | * moving them. | |
3296 | */ | |
3297 | ||
5517d86b ED |
3298 | pwr_now += busiest->__cpu_power * |
3299 | min(busiest_load_per_task, max_load); | |
3300 | pwr_now += this->__cpu_power * | |
3301 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3302 | pwr_now /= SCHED_LOAD_SCALE; |
3303 | ||
3304 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3305 | tmp = sg_div_cpu_power(busiest, |
3306 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3307 | if (max_load > tmp) |
5517d86b | 3308 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3309 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3310 | |
3311 | /* Amount of load we'd add */ | |
5517d86b | 3312 | if (max_load * busiest->__cpu_power < |
33859f7f | 3313 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3314 | tmp = sg_div_cpu_power(this, |
3315 | max_load * busiest->__cpu_power); | |
1da177e4 | 3316 | else |
5517d86b ED |
3317 | tmp = sg_div_cpu_power(this, |
3318 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3319 | pwr_move += this->__cpu_power * | |
3320 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3321 | pwr_move /= SCHED_LOAD_SCALE; |
3322 | ||
3323 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3324 | if (pwr_move > pwr_now) |
3325 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3326 | } |
3327 | ||
1da177e4 LT |
3328 | return busiest; |
3329 | ||
3330 | out_balanced: | |
5c45bf27 | 3331 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3332 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3333 | goto ret; |
1da177e4 | 3334 | |
5c45bf27 SS |
3335 | if (this == group_leader && group_leader != group_min) { |
3336 | *imbalance = min_load_per_task; | |
3337 | return group_min; | |
3338 | } | |
5c45bf27 | 3339 | #endif |
783609c6 | 3340 | ret: |
1da177e4 LT |
3341 | *imbalance = 0; |
3342 | return NULL; | |
3343 | } | |
3344 | ||
3345 | /* | |
3346 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3347 | */ | |
70b97a7f | 3348 | static struct rq * |
d15bcfdb | 3349 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3350 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3351 | { |
70b97a7f | 3352 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3353 | unsigned long max_load = 0; |
1da177e4 LT |
3354 | int i; |
3355 | ||
3356 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 3357 | unsigned long wl; |
0a2966b4 CL |
3358 | |
3359 | if (!cpu_isset(i, *cpus)) | |
3360 | continue; | |
3361 | ||
48f24c4d | 3362 | rq = cpu_rq(i); |
dd41f596 | 3363 | wl = weighted_cpuload(i); |
2dd73a4f | 3364 | |
dd41f596 | 3365 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3366 | continue; |
1da177e4 | 3367 | |
dd41f596 IM |
3368 | if (wl > max_load) { |
3369 | max_load = wl; | |
48f24c4d | 3370 | busiest = rq; |
1da177e4 LT |
3371 | } |
3372 | } | |
3373 | ||
3374 | return busiest; | |
3375 | } | |
3376 | ||
77391d71 NP |
3377 | /* |
3378 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3379 | * so long as it is large enough. | |
3380 | */ | |
3381 | #define MAX_PINNED_INTERVAL 512 | |
3382 | ||
1da177e4 LT |
3383 | /* |
3384 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3385 | * tasks if there is an imbalance. | |
1da177e4 | 3386 | */ |
70b97a7f | 3387 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3388 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3389 | int *balance, cpumask_t *cpus) |
1da177e4 | 3390 | { |
43010659 | 3391 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3392 | struct sched_group *group; |
1da177e4 | 3393 | unsigned long imbalance; |
70b97a7f | 3394 | struct rq *busiest; |
fe2eea3f | 3395 | unsigned long flags; |
5969fe06 | 3396 | |
7c16ec58 MT |
3397 | cpus_setall(*cpus); |
3398 | ||
89c4710e SS |
3399 | /* |
3400 | * When power savings policy is enabled for the parent domain, idle | |
3401 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3402 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3403 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3404 | */ |
d15bcfdb | 3405 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3406 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3407 | sd_idle = 1; |
1da177e4 | 3408 | |
2d72376b | 3409 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3410 | |
0a2966b4 | 3411 | redo: |
c8cba857 | 3412 | update_shares(sd); |
0a2966b4 | 3413 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3414 | cpus, balance); |
783609c6 | 3415 | |
06066714 | 3416 | if (*balance == 0) |
783609c6 | 3417 | goto out_balanced; |
783609c6 | 3418 | |
1da177e4 LT |
3419 | if (!group) { |
3420 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3421 | goto out_balanced; | |
3422 | } | |
3423 | ||
7c16ec58 | 3424 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3425 | if (!busiest) { |
3426 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3427 | goto out_balanced; | |
3428 | } | |
3429 | ||
db935dbd | 3430 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3431 | |
3432 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3433 | ||
43010659 | 3434 | ld_moved = 0; |
1da177e4 LT |
3435 | if (busiest->nr_running > 1) { |
3436 | /* | |
3437 | * Attempt to move tasks. If find_busiest_group has found | |
3438 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3439 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3440 | * correctly treated as an imbalance. |
3441 | */ | |
fe2eea3f | 3442 | local_irq_save(flags); |
e17224bf | 3443 | double_rq_lock(this_rq, busiest); |
43010659 | 3444 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3445 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3446 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3447 | local_irq_restore(flags); |
81026794 | 3448 | |
46cb4b7c SS |
3449 | /* |
3450 | * some other cpu did the load balance for us. | |
3451 | */ | |
43010659 | 3452 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3453 | resched_cpu(this_cpu); |
3454 | ||
81026794 | 3455 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3456 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3457 | cpu_clear(cpu_of(busiest), *cpus); |
3458 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3459 | goto redo; |
81026794 | 3460 | goto out_balanced; |
0a2966b4 | 3461 | } |
1da177e4 | 3462 | } |
81026794 | 3463 | |
43010659 | 3464 | if (!ld_moved) { |
1da177e4 LT |
3465 | schedstat_inc(sd, lb_failed[idle]); |
3466 | sd->nr_balance_failed++; | |
3467 | ||
3468 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3469 | |
fe2eea3f | 3470 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3471 | |
3472 | /* don't kick the migration_thread, if the curr | |
3473 | * task on busiest cpu can't be moved to this_cpu | |
3474 | */ | |
3475 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3476 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3477 | all_pinned = 1; |
3478 | goto out_one_pinned; | |
3479 | } | |
3480 | ||
1da177e4 LT |
3481 | if (!busiest->active_balance) { |
3482 | busiest->active_balance = 1; | |
3483 | busiest->push_cpu = this_cpu; | |
81026794 | 3484 | active_balance = 1; |
1da177e4 | 3485 | } |
fe2eea3f | 3486 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3487 | if (active_balance) |
1da177e4 LT |
3488 | wake_up_process(busiest->migration_thread); |
3489 | ||
3490 | /* | |
3491 | * We've kicked active balancing, reset the failure | |
3492 | * counter. | |
3493 | */ | |
39507451 | 3494 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3495 | } |
81026794 | 3496 | } else |
1da177e4 LT |
3497 | sd->nr_balance_failed = 0; |
3498 | ||
81026794 | 3499 | if (likely(!active_balance)) { |
1da177e4 LT |
3500 | /* We were unbalanced, so reset the balancing interval */ |
3501 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3502 | } else { |
3503 | /* | |
3504 | * If we've begun active balancing, start to back off. This | |
3505 | * case may not be covered by the all_pinned logic if there | |
3506 | * is only 1 task on the busy runqueue (because we don't call | |
3507 | * move_tasks). | |
3508 | */ | |
3509 | if (sd->balance_interval < sd->max_interval) | |
3510 | sd->balance_interval *= 2; | |
1da177e4 LT |
3511 | } |
3512 | ||
43010659 | 3513 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3514 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3515 | ld_moved = -1; |
3516 | ||
3517 | goto out; | |
1da177e4 LT |
3518 | |
3519 | out_balanced: | |
1da177e4 LT |
3520 | schedstat_inc(sd, lb_balanced[idle]); |
3521 | ||
16cfb1c0 | 3522 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3523 | |
3524 | out_one_pinned: | |
1da177e4 | 3525 | /* tune up the balancing interval */ |
77391d71 NP |
3526 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3527 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3528 | sd->balance_interval *= 2; |
3529 | ||
48f24c4d | 3530 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3531 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3532 | ld_moved = -1; |
3533 | else | |
3534 | ld_moved = 0; | |
3535 | out: | |
c8cba857 PZ |
3536 | if (ld_moved) |
3537 | update_shares(sd); | |
c09595f6 | 3538 | return ld_moved; |
1da177e4 LT |
3539 | } |
3540 | ||
3541 | /* | |
3542 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3543 | * tasks if there is an imbalance. | |
3544 | * | |
d15bcfdb | 3545 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3546 | * this_rq is locked. |
3547 | */ | |
48f24c4d | 3548 | static int |
7c16ec58 MT |
3549 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3550 | cpumask_t *cpus) | |
1da177e4 LT |
3551 | { |
3552 | struct sched_group *group; | |
70b97a7f | 3553 | struct rq *busiest = NULL; |
1da177e4 | 3554 | unsigned long imbalance; |
43010659 | 3555 | int ld_moved = 0; |
5969fe06 | 3556 | int sd_idle = 0; |
969bb4e4 | 3557 | int all_pinned = 0; |
7c16ec58 MT |
3558 | |
3559 | cpus_setall(*cpus); | |
5969fe06 | 3560 | |
89c4710e SS |
3561 | /* |
3562 | * When power savings policy is enabled for the parent domain, idle | |
3563 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3564 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3565 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3566 | */ |
3567 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3568 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3569 | sd_idle = 1; |
1da177e4 | 3570 | |
2d72376b | 3571 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3572 | redo: |
3e5459b4 | 3573 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3574 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3575 | &sd_idle, cpus, NULL); |
1da177e4 | 3576 | if (!group) { |
d15bcfdb | 3577 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3578 | goto out_balanced; |
1da177e4 LT |
3579 | } |
3580 | ||
7c16ec58 | 3581 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3582 | if (!busiest) { |
d15bcfdb | 3583 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3584 | goto out_balanced; |
1da177e4 LT |
3585 | } |
3586 | ||
db935dbd NP |
3587 | BUG_ON(busiest == this_rq); |
3588 | ||
d15bcfdb | 3589 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3590 | |
43010659 | 3591 | ld_moved = 0; |
d6d5cfaf NP |
3592 | if (busiest->nr_running > 1) { |
3593 | /* Attempt to move tasks */ | |
3594 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3595 | /* this_rq->clock is already updated */ |
3596 | update_rq_clock(busiest); | |
43010659 | 3597 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3598 | imbalance, sd, CPU_NEWLY_IDLE, |
3599 | &all_pinned); | |
d6d5cfaf | 3600 | spin_unlock(&busiest->lock); |
0a2966b4 | 3601 | |
969bb4e4 | 3602 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3603 | cpu_clear(cpu_of(busiest), *cpus); |
3604 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3605 | goto redo; |
3606 | } | |
d6d5cfaf NP |
3607 | } |
3608 | ||
43010659 | 3609 | if (!ld_moved) { |
d15bcfdb | 3610 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3611 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3612 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3613 | return -1; |
3614 | } else | |
16cfb1c0 | 3615 | sd->nr_balance_failed = 0; |
1da177e4 | 3616 | |
3e5459b4 | 3617 | update_shares_locked(this_rq, sd); |
43010659 | 3618 | return ld_moved; |
16cfb1c0 NP |
3619 | |
3620 | out_balanced: | |
d15bcfdb | 3621 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3622 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3623 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3624 | return -1; |
16cfb1c0 | 3625 | sd->nr_balance_failed = 0; |
48f24c4d | 3626 | |
16cfb1c0 | 3627 | return 0; |
1da177e4 LT |
3628 | } |
3629 | ||
3630 | /* | |
3631 | * idle_balance is called by schedule() if this_cpu is about to become | |
3632 | * idle. Attempts to pull tasks from other CPUs. | |
3633 | */ | |
70b97a7f | 3634 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3635 | { |
3636 | struct sched_domain *sd; | |
dd41f596 IM |
3637 | int pulled_task = -1; |
3638 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3639 | cpumask_t tmpmask; |
1da177e4 LT |
3640 | |
3641 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3642 | unsigned long interval; |
3643 | ||
3644 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3645 | continue; | |
3646 | ||
3647 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3648 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3649 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3650 | sd, &tmpmask); | |
92c4ca5c CL |
3651 | |
3652 | interval = msecs_to_jiffies(sd->balance_interval); | |
3653 | if (time_after(next_balance, sd->last_balance + interval)) | |
3654 | next_balance = sd->last_balance + interval; | |
3655 | if (pulled_task) | |
3656 | break; | |
1da177e4 | 3657 | } |
dd41f596 | 3658 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3659 | /* |
3660 | * We are going idle. next_balance may be set based on | |
3661 | * a busy processor. So reset next_balance. | |
3662 | */ | |
3663 | this_rq->next_balance = next_balance; | |
dd41f596 | 3664 | } |
1da177e4 LT |
3665 | } |
3666 | ||
3667 | /* | |
3668 | * active_load_balance is run by migration threads. It pushes running tasks | |
3669 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3670 | * running on each physical CPU where possible, and avoids physical / | |
3671 | * logical imbalances. | |
3672 | * | |
3673 | * Called with busiest_rq locked. | |
3674 | */ | |
70b97a7f | 3675 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3676 | { |
39507451 | 3677 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3678 | struct sched_domain *sd; |
3679 | struct rq *target_rq; | |
39507451 | 3680 | |
48f24c4d | 3681 | /* Is there any task to move? */ |
39507451 | 3682 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3683 | return; |
3684 | ||
3685 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3686 | |
3687 | /* | |
39507451 | 3688 | * This condition is "impossible", if it occurs |
41a2d6cf | 3689 | * we need to fix it. Originally reported by |
39507451 | 3690 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3691 | */ |
39507451 | 3692 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3693 | |
39507451 NP |
3694 | /* move a task from busiest_rq to target_rq */ |
3695 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3696 | update_rq_clock(busiest_rq); |
3697 | update_rq_clock(target_rq); | |
39507451 NP |
3698 | |
3699 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3700 | for_each_domain(target_cpu, sd) { |
39507451 | 3701 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3702 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3703 | break; |
c96d145e | 3704 | } |
39507451 | 3705 | |
48f24c4d | 3706 | if (likely(sd)) { |
2d72376b | 3707 | schedstat_inc(sd, alb_count); |
39507451 | 3708 | |
43010659 PW |
3709 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3710 | sd, CPU_IDLE)) | |
48f24c4d IM |
3711 | schedstat_inc(sd, alb_pushed); |
3712 | else | |
3713 | schedstat_inc(sd, alb_failed); | |
3714 | } | |
39507451 | 3715 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3716 | } |
3717 | ||
46cb4b7c SS |
3718 | #ifdef CONFIG_NO_HZ |
3719 | static struct { | |
3720 | atomic_t load_balancer; | |
41a2d6cf | 3721 | cpumask_t cpu_mask; |
46cb4b7c SS |
3722 | } nohz ____cacheline_aligned = { |
3723 | .load_balancer = ATOMIC_INIT(-1), | |
3724 | .cpu_mask = CPU_MASK_NONE, | |
3725 | }; | |
3726 | ||
7835b98b | 3727 | /* |
46cb4b7c SS |
3728 | * This routine will try to nominate the ilb (idle load balancing) |
3729 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3730 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3731 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3732 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3733 | * arrives... | |
3734 | * | |
3735 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3736 | * for idle load balancing. ilb owner will still be part of | |
3737 | * nohz.cpu_mask.. | |
7835b98b | 3738 | * |
46cb4b7c SS |
3739 | * While stopping the tick, this cpu will become the ilb owner if there |
3740 | * is no other owner. And will be the owner till that cpu becomes busy | |
3741 | * or if all cpus in the system stop their ticks at which point | |
3742 | * there is no need for ilb owner. | |
3743 | * | |
3744 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3745 | * next busy scheduler_tick() | |
3746 | */ | |
3747 | int select_nohz_load_balancer(int stop_tick) | |
3748 | { | |
3749 | int cpu = smp_processor_id(); | |
3750 | ||
3751 | if (stop_tick) { | |
3752 | cpu_set(cpu, nohz.cpu_mask); | |
3753 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3754 | ||
3755 | /* | |
3756 | * If we are going offline and still the leader, give up! | |
3757 | */ | |
3758 | if (cpu_is_offline(cpu) && | |
3759 | atomic_read(&nohz.load_balancer) == cpu) { | |
3760 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3761 | BUG(); | |
3762 | return 0; | |
3763 | } | |
3764 | ||
3765 | /* time for ilb owner also to sleep */ | |
3766 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3767 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3768 | atomic_set(&nohz.load_balancer, -1); | |
3769 | return 0; | |
3770 | } | |
3771 | ||
3772 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3773 | /* make me the ilb owner */ | |
3774 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3775 | return 1; | |
3776 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3777 | return 1; | |
3778 | } else { | |
3779 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3780 | return 0; | |
3781 | ||
3782 | cpu_clear(cpu, nohz.cpu_mask); | |
3783 | ||
3784 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3785 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3786 | BUG(); | |
3787 | } | |
3788 | return 0; | |
3789 | } | |
3790 | #endif | |
3791 | ||
3792 | static DEFINE_SPINLOCK(balancing); | |
3793 | ||
3794 | /* | |
7835b98b CL |
3795 | * It checks each scheduling domain to see if it is due to be balanced, |
3796 | * and initiates a balancing operation if so. | |
3797 | * | |
3798 | * Balancing parameters are set up in arch_init_sched_domains. | |
3799 | */ | |
a9957449 | 3800 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3801 | { |
46cb4b7c SS |
3802 | int balance = 1; |
3803 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3804 | unsigned long interval; |
3805 | struct sched_domain *sd; | |
46cb4b7c | 3806 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3807 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3808 | int update_next_balance = 0; |
d07355f5 | 3809 | int need_serialize; |
7c16ec58 | 3810 | cpumask_t tmp; |
1da177e4 | 3811 | |
46cb4b7c | 3812 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3813 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3814 | continue; | |
3815 | ||
3816 | interval = sd->balance_interval; | |
d15bcfdb | 3817 | if (idle != CPU_IDLE) |
1da177e4 LT |
3818 | interval *= sd->busy_factor; |
3819 | ||
3820 | /* scale ms to jiffies */ | |
3821 | interval = msecs_to_jiffies(interval); | |
3822 | if (unlikely(!interval)) | |
3823 | interval = 1; | |
dd41f596 IM |
3824 | if (interval > HZ*NR_CPUS/10) |
3825 | interval = HZ*NR_CPUS/10; | |
3826 | ||
d07355f5 | 3827 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3828 | |
d07355f5 | 3829 | if (need_serialize) { |
08c183f3 CL |
3830 | if (!spin_trylock(&balancing)) |
3831 | goto out; | |
3832 | } | |
3833 | ||
c9819f45 | 3834 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3835 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3836 | /* |
3837 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3838 | * longer idle, or one of our SMT siblings is |
3839 | * not idle. | |
3840 | */ | |
d15bcfdb | 3841 | idle = CPU_NOT_IDLE; |
1da177e4 | 3842 | } |
1bd77f2d | 3843 | sd->last_balance = jiffies; |
1da177e4 | 3844 | } |
d07355f5 | 3845 | if (need_serialize) |
08c183f3 CL |
3846 | spin_unlock(&balancing); |
3847 | out: | |
f549da84 | 3848 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3849 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3850 | update_next_balance = 1; |
3851 | } | |
783609c6 SS |
3852 | |
3853 | /* | |
3854 | * Stop the load balance at this level. There is another | |
3855 | * CPU in our sched group which is doing load balancing more | |
3856 | * actively. | |
3857 | */ | |
3858 | if (!balance) | |
3859 | break; | |
1da177e4 | 3860 | } |
f549da84 SS |
3861 | |
3862 | /* | |
3863 | * next_balance will be updated only when there is a need. | |
3864 | * When the cpu is attached to null domain for ex, it will not be | |
3865 | * updated. | |
3866 | */ | |
3867 | if (likely(update_next_balance)) | |
3868 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3869 | } |
3870 | ||
3871 | /* | |
3872 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3873 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3874 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3875 | */ | |
3876 | static void run_rebalance_domains(struct softirq_action *h) | |
3877 | { | |
dd41f596 IM |
3878 | int this_cpu = smp_processor_id(); |
3879 | struct rq *this_rq = cpu_rq(this_cpu); | |
3880 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3881 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3882 | |
dd41f596 | 3883 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3884 | |
3885 | #ifdef CONFIG_NO_HZ | |
3886 | /* | |
3887 | * If this cpu is the owner for idle load balancing, then do the | |
3888 | * balancing on behalf of the other idle cpus whose ticks are | |
3889 | * stopped. | |
3890 | */ | |
dd41f596 IM |
3891 | if (this_rq->idle_at_tick && |
3892 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3893 | cpumask_t cpus = nohz.cpu_mask; |
3894 | struct rq *rq; | |
3895 | int balance_cpu; | |
3896 | ||
dd41f596 | 3897 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3898 | for_each_cpu_mask(balance_cpu, cpus) { |
3899 | /* | |
3900 | * If this cpu gets work to do, stop the load balancing | |
3901 | * work being done for other cpus. Next load | |
3902 | * balancing owner will pick it up. | |
3903 | */ | |
3904 | if (need_resched()) | |
3905 | break; | |
3906 | ||
de0cf899 | 3907 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3908 | |
3909 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3910 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3911 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3912 | } |
3913 | } | |
3914 | #endif | |
3915 | } | |
3916 | ||
3917 | /* | |
3918 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3919 | * | |
3920 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3921 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3922 | * if the whole system is idle. | |
3923 | */ | |
dd41f596 | 3924 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3925 | { |
46cb4b7c SS |
3926 | #ifdef CONFIG_NO_HZ |
3927 | /* | |
3928 | * If we were in the nohz mode recently and busy at the current | |
3929 | * scheduler tick, then check if we need to nominate new idle | |
3930 | * load balancer. | |
3931 | */ | |
3932 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3933 | rq->in_nohz_recently = 0; | |
3934 | ||
3935 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3936 | cpu_clear(cpu, nohz.cpu_mask); | |
3937 | atomic_set(&nohz.load_balancer, -1); | |
3938 | } | |
3939 | ||
3940 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3941 | /* | |
3942 | * simple selection for now: Nominate the | |
3943 | * first cpu in the nohz list to be the next | |
3944 | * ilb owner. | |
3945 | * | |
3946 | * TBD: Traverse the sched domains and nominate | |
3947 | * the nearest cpu in the nohz.cpu_mask. | |
3948 | */ | |
3949 | int ilb = first_cpu(nohz.cpu_mask); | |
3950 | ||
434d53b0 | 3951 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3952 | resched_cpu(ilb); |
3953 | } | |
3954 | } | |
3955 | ||
3956 | /* | |
3957 | * If this cpu is idle and doing idle load balancing for all the | |
3958 | * cpus with ticks stopped, is it time for that to stop? | |
3959 | */ | |
3960 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3961 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3962 | resched_cpu(cpu); | |
3963 | return; | |
3964 | } | |
3965 | ||
3966 | /* | |
3967 | * If this cpu is idle and the idle load balancing is done by | |
3968 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3969 | */ | |
3970 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3971 | cpu_isset(cpu, nohz.cpu_mask)) | |
3972 | return; | |
3973 | #endif | |
3974 | if (time_after_eq(jiffies, rq->next_balance)) | |
3975 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3976 | } |
dd41f596 IM |
3977 | |
3978 | #else /* CONFIG_SMP */ | |
3979 | ||
1da177e4 LT |
3980 | /* |
3981 | * on UP we do not need to balance between CPUs: | |
3982 | */ | |
70b97a7f | 3983 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3984 | { |
3985 | } | |
dd41f596 | 3986 | |
1da177e4 LT |
3987 | #endif |
3988 | ||
1da177e4 LT |
3989 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3990 | ||
3991 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3992 | ||
3993 | /* | |
41b86e9c IM |
3994 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3995 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3996 | */ |
41b86e9c | 3997 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3998 | { |
1da177e4 | 3999 | unsigned long flags; |
41b86e9c IM |
4000 | u64 ns, delta_exec; |
4001 | struct rq *rq; | |
48f24c4d | 4002 | |
41b86e9c IM |
4003 | rq = task_rq_lock(p, &flags); |
4004 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 4005 | if (task_current(rq, p)) { |
a8e504d2 IM |
4006 | update_rq_clock(rq); |
4007 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
4008 | if ((s64)delta_exec > 0) |
4009 | ns += delta_exec; | |
4010 | } | |
4011 | task_rq_unlock(rq, &flags); | |
48f24c4d | 4012 | |
1da177e4 LT |
4013 | return ns; |
4014 | } | |
4015 | ||
1da177e4 LT |
4016 | /* |
4017 | * Account user cpu time to a process. | |
4018 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4019 | * @cputime: the cpu time spent in user space since the last update |
4020 | */ | |
4021 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4022 | { | |
4023 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4024 | cputime64_t tmp; | |
4025 | ||
4026 | p->utime = cputime_add(p->utime, cputime); | |
4027 | ||
4028 | /* Add user time to cpustat. */ | |
4029 | tmp = cputime_to_cputime64(cputime); | |
4030 | if (TASK_NICE(p) > 0) | |
4031 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4032 | else | |
4033 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4034 | } | |
4035 | ||
94886b84 LV |
4036 | /* |
4037 | * Account guest cpu time to a process. | |
4038 | * @p: the process that the cpu time gets accounted to | |
4039 | * @cputime: the cpu time spent in virtual machine since the last update | |
4040 | */ | |
f7402e03 | 4041 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4042 | { |
4043 | cputime64_t tmp; | |
4044 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4045 | ||
4046 | tmp = cputime_to_cputime64(cputime); | |
4047 | ||
4048 | p->utime = cputime_add(p->utime, cputime); | |
4049 | p->gtime = cputime_add(p->gtime, cputime); | |
4050 | ||
4051 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4052 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4053 | } | |
4054 | ||
c66f08be MN |
4055 | /* |
4056 | * Account scaled user cpu time to a process. | |
4057 | * @p: the process that the cpu time gets accounted to | |
4058 | * @cputime: the cpu time spent in user space since the last update | |
4059 | */ | |
4060 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4061 | { | |
4062 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4063 | } | |
4064 | ||
1da177e4 LT |
4065 | /* |
4066 | * Account system cpu time to a process. | |
4067 | * @p: the process that the cpu time gets accounted to | |
4068 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4069 | * @cputime: the cpu time spent in kernel space since the last update | |
4070 | */ | |
4071 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4072 | cputime_t cputime) | |
4073 | { | |
4074 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4075 | struct rq *rq = this_rq(); |
1da177e4 LT |
4076 | cputime64_t tmp; |
4077 | ||
983ed7a6 HH |
4078 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4079 | account_guest_time(p, cputime); | |
4080 | return; | |
4081 | } | |
94886b84 | 4082 | |
1da177e4 LT |
4083 | p->stime = cputime_add(p->stime, cputime); |
4084 | ||
4085 | /* Add system time to cpustat. */ | |
4086 | tmp = cputime_to_cputime64(cputime); | |
4087 | if (hardirq_count() - hardirq_offset) | |
4088 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4089 | else if (softirq_count()) | |
4090 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4091 | else if (p != rq->idle) |
1da177e4 | 4092 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4093 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4094 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4095 | else | |
4096 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4097 | /* Account for system time used */ | |
4098 | acct_update_integrals(p); | |
1da177e4 LT |
4099 | } |
4100 | ||
c66f08be MN |
4101 | /* |
4102 | * Account scaled system cpu time to a process. | |
4103 | * @p: the process that the cpu time gets accounted to | |
4104 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4105 | * @cputime: the cpu time spent in kernel space since the last update | |
4106 | */ | |
4107 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4108 | { | |
4109 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4110 | } | |
4111 | ||
1da177e4 LT |
4112 | /* |
4113 | * Account for involuntary wait time. | |
4114 | * @p: the process from which the cpu time has been stolen | |
4115 | * @steal: the cpu time spent in involuntary wait | |
4116 | */ | |
4117 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4118 | { | |
4119 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4120 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4121 | struct rq *rq = this_rq(); |
1da177e4 LT |
4122 | |
4123 | if (p == rq->idle) { | |
4124 | p->stime = cputime_add(p->stime, steal); | |
4125 | if (atomic_read(&rq->nr_iowait) > 0) | |
4126 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4127 | else | |
4128 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4129 | } else |
1da177e4 LT |
4130 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4131 | } | |
4132 | ||
7835b98b CL |
4133 | /* |
4134 | * This function gets called by the timer code, with HZ frequency. | |
4135 | * We call it with interrupts disabled. | |
4136 | * | |
4137 | * It also gets called by the fork code, when changing the parent's | |
4138 | * timeslices. | |
4139 | */ | |
4140 | void scheduler_tick(void) | |
4141 | { | |
7835b98b CL |
4142 | int cpu = smp_processor_id(); |
4143 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4144 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4145 | |
4146 | sched_clock_tick(); | |
dd41f596 IM |
4147 | |
4148 | spin_lock(&rq->lock); | |
3e51f33f | 4149 | update_rq_clock(rq); |
f1a438d8 | 4150 | update_cpu_load(rq); |
fa85ae24 | 4151 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4152 | spin_unlock(&rq->lock); |
7835b98b | 4153 | |
e418e1c2 | 4154 | #ifdef CONFIG_SMP |
dd41f596 IM |
4155 | rq->idle_at_tick = idle_cpu(cpu); |
4156 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4157 | #endif |
1da177e4 LT |
4158 | } |
4159 | ||
1da177e4 LT |
4160 | #if defined(CONFIG_PREEMPT) && defined(CONFIG_DEBUG_PREEMPT) |
4161 | ||
43627582 | 4162 | void __kprobes add_preempt_count(int val) |
1da177e4 LT |
4163 | { |
4164 | /* | |
4165 | * Underflow? | |
4166 | */ | |
9a11b49a IM |
4167 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4168 | return; | |
1da177e4 LT |
4169 | preempt_count() += val; |
4170 | /* | |
4171 | * Spinlock count overflowing soon? | |
4172 | */ | |
33859f7f MOS |
4173 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4174 | PREEMPT_MASK - 10); | |
1da177e4 LT |
4175 | } |
4176 | EXPORT_SYMBOL(add_preempt_count); | |
4177 | ||
43627582 | 4178 | void __kprobes sub_preempt_count(int val) |
1da177e4 LT |
4179 | { |
4180 | /* | |
4181 | * Underflow? | |
4182 | */ | |
9a11b49a IM |
4183 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4184 | return; | |
1da177e4 LT |
4185 | /* |
4186 | * Is the spinlock portion underflowing? | |
4187 | */ | |
9a11b49a IM |
4188 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4189 | !(preempt_count() & PREEMPT_MASK))) | |
4190 | return; | |
4191 | ||
1da177e4 LT |
4192 | preempt_count() -= val; |
4193 | } | |
4194 | EXPORT_SYMBOL(sub_preempt_count); | |
4195 | ||
4196 | #endif | |
4197 | ||
4198 | /* | |
dd41f596 | 4199 | * Print scheduling while atomic bug: |
1da177e4 | 4200 | */ |
dd41f596 | 4201 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4202 | { |
838225b4 SS |
4203 | struct pt_regs *regs = get_irq_regs(); |
4204 | ||
4205 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4206 | prev->comm, prev->pid, preempt_count()); | |
4207 | ||
dd41f596 | 4208 | debug_show_held_locks(prev); |
e21f5b15 | 4209 | print_modules(); |
dd41f596 IM |
4210 | if (irqs_disabled()) |
4211 | print_irqtrace_events(prev); | |
838225b4 SS |
4212 | |
4213 | if (regs) | |
4214 | show_regs(regs); | |
4215 | else | |
4216 | dump_stack(); | |
dd41f596 | 4217 | } |
1da177e4 | 4218 | |
dd41f596 IM |
4219 | /* |
4220 | * Various schedule()-time debugging checks and statistics: | |
4221 | */ | |
4222 | static inline void schedule_debug(struct task_struct *prev) | |
4223 | { | |
1da177e4 | 4224 | /* |
41a2d6cf | 4225 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4226 | * schedule() atomically, we ignore that path for now. |
4227 | * Otherwise, whine if we are scheduling when we should not be. | |
4228 | */ | |
3f33a7ce | 4229 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4230 | __schedule_bug(prev); |
4231 | ||
1da177e4 LT |
4232 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4233 | ||
2d72376b | 4234 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4235 | #ifdef CONFIG_SCHEDSTATS |
4236 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4237 | schedstat_inc(this_rq(), bkl_count); |
4238 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4239 | } |
4240 | #endif | |
dd41f596 IM |
4241 | } |
4242 | ||
4243 | /* | |
4244 | * Pick up the highest-prio task: | |
4245 | */ | |
4246 | static inline struct task_struct * | |
ff95f3df | 4247 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4248 | { |
5522d5d5 | 4249 | const struct sched_class *class; |
dd41f596 | 4250 | struct task_struct *p; |
1da177e4 LT |
4251 | |
4252 | /* | |
dd41f596 IM |
4253 | * Optimization: we know that if all tasks are in |
4254 | * the fair class we can call that function directly: | |
1da177e4 | 4255 | */ |
dd41f596 | 4256 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4257 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4258 | if (likely(p)) |
4259 | return p; | |
1da177e4 LT |
4260 | } |
4261 | ||
dd41f596 IM |
4262 | class = sched_class_highest; |
4263 | for ( ; ; ) { | |
fb8d4724 | 4264 | p = class->pick_next_task(rq); |
dd41f596 IM |
4265 | if (p) |
4266 | return p; | |
4267 | /* | |
4268 | * Will never be NULL as the idle class always | |
4269 | * returns a non-NULL p: | |
4270 | */ | |
4271 | class = class->next; | |
4272 | } | |
4273 | } | |
1da177e4 | 4274 | |
dd41f596 IM |
4275 | /* |
4276 | * schedule() is the main scheduler function. | |
4277 | */ | |
4278 | asmlinkage void __sched schedule(void) | |
4279 | { | |
4280 | struct task_struct *prev, *next; | |
67ca7bde | 4281 | unsigned long *switch_count; |
dd41f596 | 4282 | struct rq *rq; |
f333fdc9 | 4283 | int cpu, hrtick = sched_feat(HRTICK); |
dd41f596 IM |
4284 | |
4285 | need_resched: | |
4286 | preempt_disable(); | |
4287 | cpu = smp_processor_id(); | |
4288 | rq = cpu_rq(cpu); | |
4289 | rcu_qsctr_inc(cpu); | |
4290 | prev = rq->curr; | |
4291 | switch_count = &prev->nivcsw; | |
4292 | ||
4293 | release_kernel_lock(prev); | |
4294 | need_resched_nonpreemptible: | |
4295 | ||
4296 | schedule_debug(prev); | |
1da177e4 | 4297 | |
f333fdc9 MG |
4298 | if (hrtick) |
4299 | hrtick_clear(rq); | |
8f4d37ec | 4300 | |
1e819950 IM |
4301 | /* |
4302 | * Do the rq-clock update outside the rq lock: | |
4303 | */ | |
4304 | local_irq_disable(); | |
3e51f33f | 4305 | update_rq_clock(rq); |
1e819950 IM |
4306 | spin_lock(&rq->lock); |
4307 | clear_tsk_need_resched(prev); | |
1da177e4 | 4308 | |
1da177e4 | 4309 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4310 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4311 | prev->state = TASK_RUNNING; |
16882c1e | 4312 | else |
2e1cb74a | 4313 | deactivate_task(rq, prev, 1); |
dd41f596 | 4314 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4315 | } |
4316 | ||
9a897c5a SR |
4317 | #ifdef CONFIG_SMP |
4318 | if (prev->sched_class->pre_schedule) | |
4319 | prev->sched_class->pre_schedule(rq, prev); | |
4320 | #endif | |
f65eda4f | 4321 | |
dd41f596 | 4322 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4323 | idle_balance(cpu, rq); |
1da177e4 | 4324 | |
31ee529c | 4325 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4326 | next = pick_next_task(rq, prev); |
1da177e4 | 4327 | |
1da177e4 | 4328 | if (likely(prev != next)) { |
673a90a1 DS |
4329 | sched_info_switch(prev, next); |
4330 | ||
1da177e4 LT |
4331 | rq->nr_switches++; |
4332 | rq->curr = next; | |
4333 | ++*switch_count; | |
4334 | ||
dd41f596 | 4335 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4336 | /* |
4337 | * the context switch might have flipped the stack from under | |
4338 | * us, hence refresh the local variables. | |
4339 | */ | |
4340 | cpu = smp_processor_id(); | |
4341 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4342 | } else |
4343 | spin_unlock_irq(&rq->lock); | |
4344 | ||
f333fdc9 MG |
4345 | if (hrtick) |
4346 | hrtick_set(rq); | |
8f4d37ec PZ |
4347 | |
4348 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 4349 | goto need_resched_nonpreemptible; |
8f4d37ec | 4350 | |
1da177e4 LT |
4351 | preempt_enable_no_resched(); |
4352 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4353 | goto need_resched; | |
4354 | } | |
1da177e4 LT |
4355 | EXPORT_SYMBOL(schedule); |
4356 | ||
4357 | #ifdef CONFIG_PREEMPT | |
4358 | /* | |
2ed6e34f | 4359 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4360 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4361 | * occur there and call schedule directly. |
4362 | */ | |
4363 | asmlinkage void __sched preempt_schedule(void) | |
4364 | { | |
4365 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4366 | |
1da177e4 LT |
4367 | /* |
4368 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4369 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4370 | */ |
beed33a8 | 4371 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4372 | return; |
4373 | ||
3a5c359a AK |
4374 | do { |
4375 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4376 | schedule(); |
3a5c359a | 4377 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4378 | |
3a5c359a AK |
4379 | /* |
4380 | * Check again in case we missed a preemption opportunity | |
4381 | * between schedule and now. | |
4382 | */ | |
4383 | barrier(); | |
4384 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4385 | } |
1da177e4 LT |
4386 | EXPORT_SYMBOL(preempt_schedule); |
4387 | ||
4388 | /* | |
2ed6e34f | 4389 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4390 | * off of irq context. |
4391 | * Note, that this is called and return with irqs disabled. This will | |
4392 | * protect us against recursive calling from irq. | |
4393 | */ | |
4394 | asmlinkage void __sched preempt_schedule_irq(void) | |
4395 | { | |
4396 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4397 | |
2ed6e34f | 4398 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4399 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4400 | ||
3a5c359a AK |
4401 | do { |
4402 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4403 | local_irq_enable(); |
4404 | schedule(); | |
4405 | local_irq_disable(); | |
3a5c359a | 4406 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4407 | |
3a5c359a AK |
4408 | /* |
4409 | * Check again in case we missed a preemption opportunity | |
4410 | * between schedule and now. | |
4411 | */ | |
4412 | barrier(); | |
4413 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4414 | } |
4415 | ||
4416 | #endif /* CONFIG_PREEMPT */ | |
4417 | ||
95cdf3b7 IM |
4418 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4419 | void *key) | |
1da177e4 | 4420 | { |
48f24c4d | 4421 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4422 | } |
1da177e4 LT |
4423 | EXPORT_SYMBOL(default_wake_function); |
4424 | ||
4425 | /* | |
41a2d6cf IM |
4426 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4427 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4428 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4429 | * | |
4430 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4431 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4432 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4433 | */ | |
4434 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4435 | int nr_exclusive, int sync, void *key) | |
4436 | { | |
2e45874c | 4437 | wait_queue_t *curr, *next; |
1da177e4 | 4438 | |
2e45874c | 4439 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4440 | unsigned flags = curr->flags; |
4441 | ||
1da177e4 | 4442 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4443 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4444 | break; |
4445 | } | |
4446 | } | |
4447 | ||
4448 | /** | |
4449 | * __wake_up - wake up threads blocked on a waitqueue. | |
4450 | * @q: the waitqueue | |
4451 | * @mode: which threads | |
4452 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4453 | * @key: is directly passed to the wakeup function |
1da177e4 | 4454 | */ |
7ad5b3a5 | 4455 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4456 | int nr_exclusive, void *key) |
1da177e4 LT |
4457 | { |
4458 | unsigned long flags; | |
4459 | ||
4460 | spin_lock_irqsave(&q->lock, flags); | |
4461 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4462 | spin_unlock_irqrestore(&q->lock, flags); | |
4463 | } | |
1da177e4 LT |
4464 | EXPORT_SYMBOL(__wake_up); |
4465 | ||
4466 | /* | |
4467 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4468 | */ | |
7ad5b3a5 | 4469 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4470 | { |
4471 | __wake_up_common(q, mode, 1, 0, NULL); | |
4472 | } | |
4473 | ||
4474 | /** | |
67be2dd1 | 4475 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4476 | * @q: the waitqueue |
4477 | * @mode: which threads | |
4478 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4479 | * | |
4480 | * The sync wakeup differs that the waker knows that it will schedule | |
4481 | * away soon, so while the target thread will be woken up, it will not | |
4482 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4483 | * with each other. This can prevent needless bouncing between CPUs. | |
4484 | * | |
4485 | * On UP it can prevent extra preemption. | |
4486 | */ | |
7ad5b3a5 | 4487 | void |
95cdf3b7 | 4488 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4489 | { |
4490 | unsigned long flags; | |
4491 | int sync = 1; | |
4492 | ||
4493 | if (unlikely(!q)) | |
4494 | return; | |
4495 | ||
4496 | if (unlikely(!nr_exclusive)) | |
4497 | sync = 0; | |
4498 | ||
4499 | spin_lock_irqsave(&q->lock, flags); | |
4500 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4501 | spin_unlock_irqrestore(&q->lock, flags); | |
4502 | } | |
4503 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4504 | ||
b15136e9 | 4505 | void complete(struct completion *x) |
1da177e4 LT |
4506 | { |
4507 | unsigned long flags; | |
4508 | ||
4509 | spin_lock_irqsave(&x->wait.lock, flags); | |
4510 | x->done++; | |
d9514f6c | 4511 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4512 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4513 | } | |
4514 | EXPORT_SYMBOL(complete); | |
4515 | ||
b15136e9 | 4516 | void complete_all(struct completion *x) |
1da177e4 LT |
4517 | { |
4518 | unsigned long flags; | |
4519 | ||
4520 | spin_lock_irqsave(&x->wait.lock, flags); | |
4521 | x->done += UINT_MAX/2; | |
d9514f6c | 4522 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4523 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4524 | } | |
4525 | EXPORT_SYMBOL(complete_all); | |
4526 | ||
8cbbe86d AK |
4527 | static inline long __sched |
4528 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4529 | { |
1da177e4 LT |
4530 | if (!x->done) { |
4531 | DECLARE_WAITQUEUE(wait, current); | |
4532 | ||
4533 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4534 | __add_wait_queue_tail(&x->wait, &wait); | |
4535 | do { | |
009e577e MW |
4536 | if ((state == TASK_INTERRUPTIBLE && |
4537 | signal_pending(current)) || | |
4538 | (state == TASK_KILLABLE && | |
4539 | fatal_signal_pending(current))) { | |
ea71a546 ON |
4540 | timeout = -ERESTARTSYS; |
4541 | break; | |
8cbbe86d AK |
4542 | } |
4543 | __set_current_state(state); | |
1da177e4 LT |
4544 | spin_unlock_irq(&x->wait.lock); |
4545 | timeout = schedule_timeout(timeout); | |
4546 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4547 | } while (!x->done && timeout); |
1da177e4 | 4548 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4549 | if (!x->done) |
4550 | return timeout; | |
1da177e4 LT |
4551 | } |
4552 | x->done--; | |
ea71a546 | 4553 | return timeout ?: 1; |
1da177e4 | 4554 | } |
1da177e4 | 4555 | |
8cbbe86d AK |
4556 | static long __sched |
4557 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4558 | { |
1da177e4 LT |
4559 | might_sleep(); |
4560 | ||
4561 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4562 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4563 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4564 | return timeout; |
4565 | } | |
1da177e4 | 4566 | |
b15136e9 | 4567 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4568 | { |
4569 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4570 | } |
8cbbe86d | 4571 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4572 | |
b15136e9 | 4573 | unsigned long __sched |
8cbbe86d | 4574 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4575 | { |
8cbbe86d | 4576 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4577 | } |
8cbbe86d | 4578 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4579 | |
8cbbe86d | 4580 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4581 | { |
51e97990 AK |
4582 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4583 | if (t == -ERESTARTSYS) | |
4584 | return t; | |
4585 | return 0; | |
0fec171c | 4586 | } |
8cbbe86d | 4587 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4588 | |
b15136e9 | 4589 | unsigned long __sched |
8cbbe86d AK |
4590 | wait_for_completion_interruptible_timeout(struct completion *x, |
4591 | unsigned long timeout) | |
0fec171c | 4592 | { |
8cbbe86d | 4593 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4594 | } |
8cbbe86d | 4595 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4596 | |
009e577e MW |
4597 | int __sched wait_for_completion_killable(struct completion *x) |
4598 | { | |
4599 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4600 | if (t == -ERESTARTSYS) | |
4601 | return t; | |
4602 | return 0; | |
4603 | } | |
4604 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4605 | ||
8cbbe86d AK |
4606 | static long __sched |
4607 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4608 | { |
0fec171c IM |
4609 | unsigned long flags; |
4610 | wait_queue_t wait; | |
4611 | ||
4612 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4613 | |
8cbbe86d | 4614 | __set_current_state(state); |
1da177e4 | 4615 | |
8cbbe86d AK |
4616 | spin_lock_irqsave(&q->lock, flags); |
4617 | __add_wait_queue(q, &wait); | |
4618 | spin_unlock(&q->lock); | |
4619 | timeout = schedule_timeout(timeout); | |
4620 | spin_lock_irq(&q->lock); | |
4621 | __remove_wait_queue(q, &wait); | |
4622 | spin_unlock_irqrestore(&q->lock, flags); | |
4623 | ||
4624 | return timeout; | |
4625 | } | |
4626 | ||
4627 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4628 | { | |
4629 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4630 | } |
1da177e4 LT |
4631 | EXPORT_SYMBOL(interruptible_sleep_on); |
4632 | ||
0fec171c | 4633 | long __sched |
95cdf3b7 | 4634 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4635 | { |
8cbbe86d | 4636 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4637 | } |
1da177e4 LT |
4638 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4639 | ||
0fec171c | 4640 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4641 | { |
8cbbe86d | 4642 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4643 | } |
1da177e4 LT |
4644 | EXPORT_SYMBOL(sleep_on); |
4645 | ||
0fec171c | 4646 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4647 | { |
8cbbe86d | 4648 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4649 | } |
1da177e4 LT |
4650 | EXPORT_SYMBOL(sleep_on_timeout); |
4651 | ||
b29739f9 IM |
4652 | #ifdef CONFIG_RT_MUTEXES |
4653 | ||
4654 | /* | |
4655 | * rt_mutex_setprio - set the current priority of a task | |
4656 | * @p: task | |
4657 | * @prio: prio value (kernel-internal form) | |
4658 | * | |
4659 | * This function changes the 'effective' priority of a task. It does | |
4660 | * not touch ->normal_prio like __setscheduler(). | |
4661 | * | |
4662 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4663 | */ | |
36c8b586 | 4664 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4665 | { |
4666 | unsigned long flags; | |
83b699ed | 4667 | int oldprio, on_rq, running; |
70b97a7f | 4668 | struct rq *rq; |
cb469845 | 4669 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4670 | |
4671 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4672 | ||
4673 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4674 | update_rq_clock(rq); |
b29739f9 | 4675 | |
d5f9f942 | 4676 | oldprio = p->prio; |
dd41f596 | 4677 | on_rq = p->se.on_rq; |
051a1d1a | 4678 | running = task_current(rq, p); |
0e1f3483 | 4679 | if (on_rq) |
69be72c1 | 4680 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4681 | if (running) |
4682 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4683 | |
4684 | if (rt_prio(prio)) | |
4685 | p->sched_class = &rt_sched_class; | |
4686 | else | |
4687 | p->sched_class = &fair_sched_class; | |
4688 | ||
b29739f9 IM |
4689 | p->prio = prio; |
4690 | ||
0e1f3483 HS |
4691 | if (running) |
4692 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4693 | if (on_rq) { |
8159f87e | 4694 | enqueue_task(rq, p, 0); |
cb469845 SR |
4695 | |
4696 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4697 | } |
4698 | task_rq_unlock(rq, &flags); | |
4699 | } | |
4700 | ||
4701 | #endif | |
4702 | ||
36c8b586 | 4703 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4704 | { |
dd41f596 | 4705 | int old_prio, delta, on_rq; |
1da177e4 | 4706 | unsigned long flags; |
70b97a7f | 4707 | struct rq *rq; |
1da177e4 LT |
4708 | |
4709 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4710 | return; | |
4711 | /* | |
4712 | * We have to be careful, if called from sys_setpriority(), | |
4713 | * the task might be in the middle of scheduling on another CPU. | |
4714 | */ | |
4715 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4716 | update_rq_clock(rq); |
1da177e4 LT |
4717 | /* |
4718 | * The RT priorities are set via sched_setscheduler(), but we still | |
4719 | * allow the 'normal' nice value to be set - but as expected | |
4720 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4721 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4722 | */ |
e05606d3 | 4723 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4724 | p->static_prio = NICE_TO_PRIO(nice); |
4725 | goto out_unlock; | |
4726 | } | |
dd41f596 | 4727 | on_rq = p->se.on_rq; |
c09595f6 | 4728 | if (on_rq) |
69be72c1 | 4729 | dequeue_task(rq, p, 0); |
1da177e4 | 4730 | |
1da177e4 | 4731 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4732 | set_load_weight(p); |
b29739f9 IM |
4733 | old_prio = p->prio; |
4734 | p->prio = effective_prio(p); | |
4735 | delta = p->prio - old_prio; | |
1da177e4 | 4736 | |
dd41f596 | 4737 | if (on_rq) { |
8159f87e | 4738 | enqueue_task(rq, p, 0); |
1da177e4 | 4739 | /* |
d5f9f942 AM |
4740 | * If the task increased its priority or is running and |
4741 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4742 | */ |
d5f9f942 | 4743 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4744 | resched_task(rq->curr); |
4745 | } | |
4746 | out_unlock: | |
4747 | task_rq_unlock(rq, &flags); | |
4748 | } | |
1da177e4 LT |
4749 | EXPORT_SYMBOL(set_user_nice); |
4750 | ||
e43379f1 MM |
4751 | /* |
4752 | * can_nice - check if a task can reduce its nice value | |
4753 | * @p: task | |
4754 | * @nice: nice value | |
4755 | */ | |
36c8b586 | 4756 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4757 | { |
024f4747 MM |
4758 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4759 | int nice_rlim = 20 - nice; | |
48f24c4d | 4760 | |
e43379f1 MM |
4761 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4762 | capable(CAP_SYS_NICE)); | |
4763 | } | |
4764 | ||
1da177e4 LT |
4765 | #ifdef __ARCH_WANT_SYS_NICE |
4766 | ||
4767 | /* | |
4768 | * sys_nice - change the priority of the current process. | |
4769 | * @increment: priority increment | |
4770 | * | |
4771 | * sys_setpriority is a more generic, but much slower function that | |
4772 | * does similar things. | |
4773 | */ | |
4774 | asmlinkage long sys_nice(int increment) | |
4775 | { | |
48f24c4d | 4776 | long nice, retval; |
1da177e4 LT |
4777 | |
4778 | /* | |
4779 | * Setpriority might change our priority at the same moment. | |
4780 | * We don't have to worry. Conceptually one call occurs first | |
4781 | * and we have a single winner. | |
4782 | */ | |
e43379f1 MM |
4783 | if (increment < -40) |
4784 | increment = -40; | |
1da177e4 LT |
4785 | if (increment > 40) |
4786 | increment = 40; | |
4787 | ||
4788 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4789 | if (nice < -20) | |
4790 | nice = -20; | |
4791 | if (nice > 19) | |
4792 | nice = 19; | |
4793 | ||
e43379f1 MM |
4794 | if (increment < 0 && !can_nice(current, nice)) |
4795 | return -EPERM; | |
4796 | ||
1da177e4 LT |
4797 | retval = security_task_setnice(current, nice); |
4798 | if (retval) | |
4799 | return retval; | |
4800 | ||
4801 | set_user_nice(current, nice); | |
4802 | return 0; | |
4803 | } | |
4804 | ||
4805 | #endif | |
4806 | ||
4807 | /** | |
4808 | * task_prio - return the priority value of a given task. | |
4809 | * @p: the task in question. | |
4810 | * | |
4811 | * This is the priority value as seen by users in /proc. | |
4812 | * RT tasks are offset by -200. Normal tasks are centered | |
4813 | * around 0, value goes from -16 to +15. | |
4814 | */ | |
36c8b586 | 4815 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4816 | { |
4817 | return p->prio - MAX_RT_PRIO; | |
4818 | } | |
4819 | ||
4820 | /** | |
4821 | * task_nice - return the nice value of a given task. | |
4822 | * @p: the task in question. | |
4823 | */ | |
36c8b586 | 4824 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4825 | { |
4826 | return TASK_NICE(p); | |
4827 | } | |
150d8bed | 4828 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4829 | |
4830 | /** | |
4831 | * idle_cpu - is a given cpu idle currently? | |
4832 | * @cpu: the processor in question. | |
4833 | */ | |
4834 | int idle_cpu(int cpu) | |
4835 | { | |
4836 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4837 | } | |
4838 | ||
1da177e4 LT |
4839 | /** |
4840 | * idle_task - return the idle task for a given cpu. | |
4841 | * @cpu: the processor in question. | |
4842 | */ | |
36c8b586 | 4843 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4844 | { |
4845 | return cpu_rq(cpu)->idle; | |
4846 | } | |
4847 | ||
4848 | /** | |
4849 | * find_process_by_pid - find a process with a matching PID value. | |
4850 | * @pid: the pid in question. | |
4851 | */ | |
a9957449 | 4852 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4853 | { |
228ebcbe | 4854 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4855 | } |
4856 | ||
4857 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4858 | static void |
4859 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4860 | { |
dd41f596 | 4861 | BUG_ON(p->se.on_rq); |
48f24c4d | 4862 | |
1da177e4 | 4863 | p->policy = policy; |
dd41f596 IM |
4864 | switch (p->policy) { |
4865 | case SCHED_NORMAL: | |
4866 | case SCHED_BATCH: | |
4867 | case SCHED_IDLE: | |
4868 | p->sched_class = &fair_sched_class; | |
4869 | break; | |
4870 | case SCHED_FIFO: | |
4871 | case SCHED_RR: | |
4872 | p->sched_class = &rt_sched_class; | |
4873 | break; | |
4874 | } | |
4875 | ||
1da177e4 | 4876 | p->rt_priority = prio; |
b29739f9 IM |
4877 | p->normal_prio = normal_prio(p); |
4878 | /* we are holding p->pi_lock already */ | |
4879 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4880 | set_load_weight(p); |
1da177e4 LT |
4881 | } |
4882 | ||
4883 | /** | |
72fd4a35 | 4884 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4885 | * @p: the task in question. |
4886 | * @policy: new policy. | |
4887 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4888 | * |
72fd4a35 | 4889 | * NOTE that the task may be already dead. |
1da177e4 | 4890 | */ |
95cdf3b7 IM |
4891 | int sched_setscheduler(struct task_struct *p, int policy, |
4892 | struct sched_param *param) | |
1da177e4 | 4893 | { |
83b699ed | 4894 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4895 | unsigned long flags; |
cb469845 | 4896 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4897 | struct rq *rq; |
1da177e4 | 4898 | |
66e5393a SR |
4899 | /* may grab non-irq protected spin_locks */ |
4900 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4901 | recheck: |
4902 | /* double check policy once rq lock held */ | |
4903 | if (policy < 0) | |
4904 | policy = oldpolicy = p->policy; | |
4905 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4906 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4907 | policy != SCHED_IDLE) | |
b0a9499c | 4908 | return -EINVAL; |
1da177e4 LT |
4909 | /* |
4910 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4911 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4912 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4913 | */ |
4914 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4915 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4916 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4917 | return -EINVAL; |
e05606d3 | 4918 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4919 | return -EINVAL; |
4920 | ||
37e4ab3f OC |
4921 | /* |
4922 | * Allow unprivileged RT tasks to decrease priority: | |
4923 | */ | |
4924 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4925 | if (rt_policy(policy)) { |
8dc3e909 | 4926 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4927 | |
4928 | if (!lock_task_sighand(p, &flags)) | |
4929 | return -ESRCH; | |
4930 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4931 | unlock_task_sighand(p, &flags); | |
4932 | ||
4933 | /* can't set/change the rt policy */ | |
4934 | if (policy != p->policy && !rlim_rtprio) | |
4935 | return -EPERM; | |
4936 | ||
4937 | /* can't increase priority */ | |
4938 | if (param->sched_priority > p->rt_priority && | |
4939 | param->sched_priority > rlim_rtprio) | |
4940 | return -EPERM; | |
4941 | } | |
dd41f596 IM |
4942 | /* |
4943 | * Like positive nice levels, dont allow tasks to | |
4944 | * move out of SCHED_IDLE either: | |
4945 | */ | |
4946 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4947 | return -EPERM; | |
5fe1d75f | 4948 | |
37e4ab3f OC |
4949 | /* can't change other user's priorities */ |
4950 | if ((current->euid != p->euid) && | |
4951 | (current->euid != p->uid)) | |
4952 | return -EPERM; | |
4953 | } | |
1da177e4 | 4954 | |
b68aa230 PZ |
4955 | #ifdef CONFIG_RT_GROUP_SCHED |
4956 | /* | |
4957 | * Do not allow realtime tasks into groups that have no runtime | |
4958 | * assigned. | |
4959 | */ | |
d0b27fa7 | 4960 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
4961 | return -EPERM; |
4962 | #endif | |
4963 | ||
1da177e4 LT |
4964 | retval = security_task_setscheduler(p, policy, param); |
4965 | if (retval) | |
4966 | return retval; | |
b29739f9 IM |
4967 | /* |
4968 | * make sure no PI-waiters arrive (or leave) while we are | |
4969 | * changing the priority of the task: | |
4970 | */ | |
4971 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4972 | /* |
4973 | * To be able to change p->policy safely, the apropriate | |
4974 | * runqueue lock must be held. | |
4975 | */ | |
b29739f9 | 4976 | rq = __task_rq_lock(p); |
1da177e4 LT |
4977 | /* recheck policy now with rq lock held */ |
4978 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4979 | policy = oldpolicy = -1; | |
b29739f9 IM |
4980 | __task_rq_unlock(rq); |
4981 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4982 | goto recheck; |
4983 | } | |
2daa3577 | 4984 | update_rq_clock(rq); |
dd41f596 | 4985 | on_rq = p->se.on_rq; |
051a1d1a | 4986 | running = task_current(rq, p); |
0e1f3483 | 4987 | if (on_rq) |
2e1cb74a | 4988 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4989 | if (running) |
4990 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4991 | |
1da177e4 | 4992 | oldprio = p->prio; |
dd41f596 | 4993 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4994 | |
0e1f3483 HS |
4995 | if (running) |
4996 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4997 | if (on_rq) { |
4998 | activate_task(rq, p, 0); | |
cb469845 SR |
4999 | |
5000 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5001 | } |
b29739f9 IM |
5002 | __task_rq_unlock(rq); |
5003 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5004 | ||
95e02ca9 TG |
5005 | rt_mutex_adjust_pi(p); |
5006 | ||
1da177e4 LT |
5007 | return 0; |
5008 | } | |
5009 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
5010 | ||
95cdf3b7 IM |
5011 | static int |
5012 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5013 | { |
1da177e4 LT |
5014 | struct sched_param lparam; |
5015 | struct task_struct *p; | |
36c8b586 | 5016 | int retval; |
1da177e4 LT |
5017 | |
5018 | if (!param || pid < 0) | |
5019 | return -EINVAL; | |
5020 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5021 | return -EFAULT; | |
5fe1d75f ON |
5022 | |
5023 | rcu_read_lock(); | |
5024 | retval = -ESRCH; | |
1da177e4 | 5025 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5026 | if (p != NULL) |
5027 | retval = sched_setscheduler(p, policy, &lparam); | |
5028 | rcu_read_unlock(); | |
36c8b586 | 5029 | |
1da177e4 LT |
5030 | return retval; |
5031 | } | |
5032 | ||
5033 | /** | |
5034 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5035 | * @pid: the pid in question. | |
5036 | * @policy: new policy. | |
5037 | * @param: structure containing the new RT priority. | |
5038 | */ | |
41a2d6cf IM |
5039 | asmlinkage long |
5040 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5041 | { |
c21761f1 JB |
5042 | /* negative values for policy are not valid */ |
5043 | if (policy < 0) | |
5044 | return -EINVAL; | |
5045 | ||
1da177e4 LT |
5046 | return do_sched_setscheduler(pid, policy, param); |
5047 | } | |
5048 | ||
5049 | /** | |
5050 | * sys_sched_setparam - set/change the RT priority of a thread | |
5051 | * @pid: the pid in question. | |
5052 | * @param: structure containing the new RT priority. | |
5053 | */ | |
5054 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5055 | { | |
5056 | return do_sched_setscheduler(pid, -1, param); | |
5057 | } | |
5058 | ||
5059 | /** | |
5060 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5061 | * @pid: the pid in question. | |
5062 | */ | |
5063 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5064 | { | |
36c8b586 | 5065 | struct task_struct *p; |
3a5c359a | 5066 | int retval; |
1da177e4 LT |
5067 | |
5068 | if (pid < 0) | |
3a5c359a | 5069 | return -EINVAL; |
1da177e4 LT |
5070 | |
5071 | retval = -ESRCH; | |
5072 | read_lock(&tasklist_lock); | |
5073 | p = find_process_by_pid(pid); | |
5074 | if (p) { | |
5075 | retval = security_task_getscheduler(p); | |
5076 | if (!retval) | |
5077 | retval = p->policy; | |
5078 | } | |
5079 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5080 | return retval; |
5081 | } | |
5082 | ||
5083 | /** | |
5084 | * sys_sched_getscheduler - get the RT priority of a thread | |
5085 | * @pid: the pid in question. | |
5086 | * @param: structure containing the RT priority. | |
5087 | */ | |
5088 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5089 | { | |
5090 | struct sched_param lp; | |
36c8b586 | 5091 | struct task_struct *p; |
3a5c359a | 5092 | int retval; |
1da177e4 LT |
5093 | |
5094 | if (!param || pid < 0) | |
3a5c359a | 5095 | return -EINVAL; |
1da177e4 LT |
5096 | |
5097 | read_lock(&tasklist_lock); | |
5098 | p = find_process_by_pid(pid); | |
5099 | retval = -ESRCH; | |
5100 | if (!p) | |
5101 | goto out_unlock; | |
5102 | ||
5103 | retval = security_task_getscheduler(p); | |
5104 | if (retval) | |
5105 | goto out_unlock; | |
5106 | ||
5107 | lp.sched_priority = p->rt_priority; | |
5108 | read_unlock(&tasklist_lock); | |
5109 | ||
5110 | /* | |
5111 | * This one might sleep, we cannot do it with a spinlock held ... | |
5112 | */ | |
5113 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5114 | ||
1da177e4 LT |
5115 | return retval; |
5116 | ||
5117 | out_unlock: | |
5118 | read_unlock(&tasklist_lock); | |
5119 | return retval; | |
5120 | } | |
5121 | ||
b53e921b | 5122 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5123 | { |
1da177e4 | 5124 | cpumask_t cpus_allowed; |
b53e921b | 5125 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5126 | struct task_struct *p; |
5127 | int retval; | |
1da177e4 | 5128 | |
95402b38 | 5129 | get_online_cpus(); |
1da177e4 LT |
5130 | read_lock(&tasklist_lock); |
5131 | ||
5132 | p = find_process_by_pid(pid); | |
5133 | if (!p) { | |
5134 | read_unlock(&tasklist_lock); | |
95402b38 | 5135 | put_online_cpus(); |
1da177e4 LT |
5136 | return -ESRCH; |
5137 | } | |
5138 | ||
5139 | /* | |
5140 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5141 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5142 | * usage count and then drop tasklist_lock. |
5143 | */ | |
5144 | get_task_struct(p); | |
5145 | read_unlock(&tasklist_lock); | |
5146 | ||
5147 | retval = -EPERM; | |
5148 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5149 | !capable(CAP_SYS_NICE)) | |
5150 | goto out_unlock; | |
5151 | ||
e7834f8f DQ |
5152 | retval = security_task_setscheduler(p, 0, NULL); |
5153 | if (retval) | |
5154 | goto out_unlock; | |
5155 | ||
f9a86fcb | 5156 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5157 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5158 | again: |
7c16ec58 | 5159 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5160 | |
8707d8b8 | 5161 | if (!retval) { |
f9a86fcb | 5162 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5163 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5164 | /* | |
5165 | * We must have raced with a concurrent cpuset | |
5166 | * update. Just reset the cpus_allowed to the | |
5167 | * cpuset's cpus_allowed | |
5168 | */ | |
5169 | new_mask = cpus_allowed; | |
5170 | goto again; | |
5171 | } | |
5172 | } | |
1da177e4 LT |
5173 | out_unlock: |
5174 | put_task_struct(p); | |
95402b38 | 5175 | put_online_cpus(); |
1da177e4 LT |
5176 | return retval; |
5177 | } | |
5178 | ||
5179 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5180 | cpumask_t *new_mask) | |
5181 | { | |
5182 | if (len < sizeof(cpumask_t)) { | |
5183 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5184 | } else if (len > sizeof(cpumask_t)) { | |
5185 | len = sizeof(cpumask_t); | |
5186 | } | |
5187 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5188 | } | |
5189 | ||
5190 | /** | |
5191 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5192 | * @pid: pid of the process | |
5193 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5194 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5195 | */ | |
5196 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5197 | unsigned long __user *user_mask_ptr) | |
5198 | { | |
5199 | cpumask_t new_mask; | |
5200 | int retval; | |
5201 | ||
5202 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5203 | if (retval) | |
5204 | return retval; | |
5205 | ||
b53e921b | 5206 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5207 | } |
5208 | ||
1da177e4 LT |
5209 | long sched_getaffinity(pid_t pid, cpumask_t *mask) |
5210 | { | |
36c8b586 | 5211 | struct task_struct *p; |
1da177e4 | 5212 | int retval; |
1da177e4 | 5213 | |
95402b38 | 5214 | get_online_cpus(); |
1da177e4 LT |
5215 | read_lock(&tasklist_lock); |
5216 | ||
5217 | retval = -ESRCH; | |
5218 | p = find_process_by_pid(pid); | |
5219 | if (!p) | |
5220 | goto out_unlock; | |
5221 | ||
e7834f8f DQ |
5222 | retval = security_task_getscheduler(p); |
5223 | if (retval) | |
5224 | goto out_unlock; | |
5225 | ||
2f7016d9 | 5226 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5227 | |
5228 | out_unlock: | |
5229 | read_unlock(&tasklist_lock); | |
95402b38 | 5230 | put_online_cpus(); |
1da177e4 | 5231 | |
9531b62f | 5232 | return retval; |
1da177e4 LT |
5233 | } |
5234 | ||
5235 | /** | |
5236 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5237 | * @pid: pid of the process | |
5238 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5239 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5240 | */ | |
5241 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5242 | unsigned long __user *user_mask_ptr) | |
5243 | { | |
5244 | int ret; | |
5245 | cpumask_t mask; | |
5246 | ||
5247 | if (len < sizeof(cpumask_t)) | |
5248 | return -EINVAL; | |
5249 | ||
5250 | ret = sched_getaffinity(pid, &mask); | |
5251 | if (ret < 0) | |
5252 | return ret; | |
5253 | ||
5254 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5255 | return -EFAULT; | |
5256 | ||
5257 | return sizeof(cpumask_t); | |
5258 | } | |
5259 | ||
5260 | /** | |
5261 | * sys_sched_yield - yield the current processor to other threads. | |
5262 | * | |
dd41f596 IM |
5263 | * This function yields the current CPU to other tasks. If there are no |
5264 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5265 | */ |
5266 | asmlinkage long sys_sched_yield(void) | |
5267 | { | |
70b97a7f | 5268 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5269 | |
2d72376b | 5270 | schedstat_inc(rq, yld_count); |
4530d7ab | 5271 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5272 | |
5273 | /* | |
5274 | * Since we are going to call schedule() anyway, there's | |
5275 | * no need to preempt or enable interrupts: | |
5276 | */ | |
5277 | __release(rq->lock); | |
8a25d5de | 5278 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5279 | _raw_spin_unlock(&rq->lock); |
5280 | preempt_enable_no_resched(); | |
5281 | ||
5282 | schedule(); | |
5283 | ||
5284 | return 0; | |
5285 | } | |
5286 | ||
e7b38404 | 5287 | static void __cond_resched(void) |
1da177e4 | 5288 | { |
8e0a43d8 IM |
5289 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5290 | __might_sleep(__FILE__, __LINE__); | |
5291 | #endif | |
5bbcfd90 IM |
5292 | /* |
5293 | * The BKS might be reacquired before we have dropped | |
5294 | * PREEMPT_ACTIVE, which could trigger a second | |
5295 | * cond_resched() call. | |
5296 | */ | |
1da177e4 LT |
5297 | do { |
5298 | add_preempt_count(PREEMPT_ACTIVE); | |
5299 | schedule(); | |
5300 | sub_preempt_count(PREEMPT_ACTIVE); | |
5301 | } while (need_resched()); | |
5302 | } | |
5303 | ||
02b67cc3 | 5304 | int __sched _cond_resched(void) |
1da177e4 | 5305 | { |
9414232f IM |
5306 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5307 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5308 | __cond_resched(); |
5309 | return 1; | |
5310 | } | |
5311 | return 0; | |
5312 | } | |
02b67cc3 | 5313 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5314 | |
5315 | /* | |
5316 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5317 | * call schedule, and on return reacquire the lock. | |
5318 | * | |
41a2d6cf | 5319 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5320 | * operations here to prevent schedule() from being called twice (once via |
5321 | * spin_unlock(), once by hand). | |
5322 | */ | |
95cdf3b7 | 5323 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5324 | { |
95c354fe | 5325 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5326 | int ret = 0; |
5327 | ||
95c354fe | 5328 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5329 | spin_unlock(lock); |
95c354fe NP |
5330 | if (resched && need_resched()) |
5331 | __cond_resched(); | |
5332 | else | |
5333 | cpu_relax(); | |
6df3cecb | 5334 | ret = 1; |
1da177e4 | 5335 | spin_lock(lock); |
1da177e4 | 5336 | } |
6df3cecb | 5337 | return ret; |
1da177e4 | 5338 | } |
1da177e4 LT |
5339 | EXPORT_SYMBOL(cond_resched_lock); |
5340 | ||
5341 | int __sched cond_resched_softirq(void) | |
5342 | { | |
5343 | BUG_ON(!in_softirq()); | |
5344 | ||
9414232f | 5345 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5346 | local_bh_enable(); |
1da177e4 LT |
5347 | __cond_resched(); |
5348 | local_bh_disable(); | |
5349 | return 1; | |
5350 | } | |
5351 | return 0; | |
5352 | } | |
1da177e4 LT |
5353 | EXPORT_SYMBOL(cond_resched_softirq); |
5354 | ||
1da177e4 LT |
5355 | /** |
5356 | * yield - yield the current processor to other threads. | |
5357 | * | |
72fd4a35 | 5358 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5359 | * thread runnable and calls sys_sched_yield(). |
5360 | */ | |
5361 | void __sched yield(void) | |
5362 | { | |
5363 | set_current_state(TASK_RUNNING); | |
5364 | sys_sched_yield(); | |
5365 | } | |
1da177e4 LT |
5366 | EXPORT_SYMBOL(yield); |
5367 | ||
5368 | /* | |
41a2d6cf | 5369 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5370 | * that process accounting knows that this is a task in IO wait state. |
5371 | * | |
5372 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5373 | * has set its backing_dev_info: the queue against which it should throttle) | |
5374 | */ | |
5375 | void __sched io_schedule(void) | |
5376 | { | |
70b97a7f | 5377 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5378 | |
0ff92245 | 5379 | delayacct_blkio_start(); |
1da177e4 LT |
5380 | atomic_inc(&rq->nr_iowait); |
5381 | schedule(); | |
5382 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5383 | delayacct_blkio_end(); |
1da177e4 | 5384 | } |
1da177e4 LT |
5385 | EXPORT_SYMBOL(io_schedule); |
5386 | ||
5387 | long __sched io_schedule_timeout(long timeout) | |
5388 | { | |
70b97a7f | 5389 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5390 | long ret; |
5391 | ||
0ff92245 | 5392 | delayacct_blkio_start(); |
1da177e4 LT |
5393 | atomic_inc(&rq->nr_iowait); |
5394 | ret = schedule_timeout(timeout); | |
5395 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5396 | delayacct_blkio_end(); |
1da177e4 LT |
5397 | return ret; |
5398 | } | |
5399 | ||
5400 | /** | |
5401 | * sys_sched_get_priority_max - return maximum RT priority. | |
5402 | * @policy: scheduling class. | |
5403 | * | |
5404 | * this syscall returns the maximum rt_priority that can be used | |
5405 | * by a given scheduling class. | |
5406 | */ | |
5407 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5408 | { | |
5409 | int ret = -EINVAL; | |
5410 | ||
5411 | switch (policy) { | |
5412 | case SCHED_FIFO: | |
5413 | case SCHED_RR: | |
5414 | ret = MAX_USER_RT_PRIO-1; | |
5415 | break; | |
5416 | case SCHED_NORMAL: | |
b0a9499c | 5417 | case SCHED_BATCH: |
dd41f596 | 5418 | case SCHED_IDLE: |
1da177e4 LT |
5419 | ret = 0; |
5420 | break; | |
5421 | } | |
5422 | return ret; | |
5423 | } | |
5424 | ||
5425 | /** | |
5426 | * sys_sched_get_priority_min - return minimum RT priority. | |
5427 | * @policy: scheduling class. | |
5428 | * | |
5429 | * this syscall returns the minimum rt_priority that can be used | |
5430 | * by a given scheduling class. | |
5431 | */ | |
5432 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5433 | { | |
5434 | int ret = -EINVAL; | |
5435 | ||
5436 | switch (policy) { | |
5437 | case SCHED_FIFO: | |
5438 | case SCHED_RR: | |
5439 | ret = 1; | |
5440 | break; | |
5441 | case SCHED_NORMAL: | |
b0a9499c | 5442 | case SCHED_BATCH: |
dd41f596 | 5443 | case SCHED_IDLE: |
1da177e4 LT |
5444 | ret = 0; |
5445 | } | |
5446 | return ret; | |
5447 | } | |
5448 | ||
5449 | /** | |
5450 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5451 | * @pid: pid of the process. | |
5452 | * @interval: userspace pointer to the timeslice value. | |
5453 | * | |
5454 | * this syscall writes the default timeslice value of a given process | |
5455 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5456 | */ | |
5457 | asmlinkage | |
5458 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5459 | { | |
36c8b586 | 5460 | struct task_struct *p; |
a4ec24b4 | 5461 | unsigned int time_slice; |
3a5c359a | 5462 | int retval; |
1da177e4 | 5463 | struct timespec t; |
1da177e4 LT |
5464 | |
5465 | if (pid < 0) | |
3a5c359a | 5466 | return -EINVAL; |
1da177e4 LT |
5467 | |
5468 | retval = -ESRCH; | |
5469 | read_lock(&tasklist_lock); | |
5470 | p = find_process_by_pid(pid); | |
5471 | if (!p) | |
5472 | goto out_unlock; | |
5473 | ||
5474 | retval = security_task_getscheduler(p); | |
5475 | if (retval) | |
5476 | goto out_unlock; | |
5477 | ||
77034937 IM |
5478 | /* |
5479 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5480 | * tasks that are on an otherwise idle runqueue: | |
5481 | */ | |
5482 | time_slice = 0; | |
5483 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5484 | time_slice = DEF_TIMESLICE; |
1868f958 | 5485 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5486 | struct sched_entity *se = &p->se; |
5487 | unsigned long flags; | |
5488 | struct rq *rq; | |
5489 | ||
5490 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5491 | if (rq->cfs.load.weight) |
5492 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5493 | task_rq_unlock(rq, &flags); |
5494 | } | |
1da177e4 | 5495 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5496 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5497 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5498 | return retval; |
3a5c359a | 5499 | |
1da177e4 LT |
5500 | out_unlock: |
5501 | read_unlock(&tasklist_lock); | |
5502 | return retval; | |
5503 | } | |
5504 | ||
2ed6e34f | 5505 | static const char stat_nam[] = "RSDTtZX"; |
36c8b586 | 5506 | |
82a1fcb9 | 5507 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5508 | { |
1da177e4 | 5509 | unsigned long free = 0; |
36c8b586 | 5510 | unsigned state; |
1da177e4 | 5511 | |
1da177e4 | 5512 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5513 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5514 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5515 | #if BITS_PER_LONG == 32 |
1da177e4 | 5516 | if (state == TASK_RUNNING) |
cc4ea795 | 5517 | printk(KERN_CONT " running "); |
1da177e4 | 5518 | else |
cc4ea795 | 5519 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5520 | #else |
5521 | if (state == TASK_RUNNING) | |
cc4ea795 | 5522 | printk(KERN_CONT " running task "); |
1da177e4 | 5523 | else |
cc4ea795 | 5524 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5525 | #endif |
5526 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5527 | { | |
10ebffde | 5528 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5529 | while (!*n) |
5530 | n++; | |
10ebffde | 5531 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5532 | } |
5533 | #endif | |
ba25f9dc | 5534 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5535 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5536 | |
5fb5e6de | 5537 | show_stack(p, NULL); |
1da177e4 LT |
5538 | } |
5539 | ||
e59e2ae2 | 5540 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5541 | { |
36c8b586 | 5542 | struct task_struct *g, *p; |
1da177e4 | 5543 | |
4bd77321 IM |
5544 | #if BITS_PER_LONG == 32 |
5545 | printk(KERN_INFO | |
5546 | " task PC stack pid father\n"); | |
1da177e4 | 5547 | #else |
4bd77321 IM |
5548 | printk(KERN_INFO |
5549 | " task PC stack pid father\n"); | |
1da177e4 LT |
5550 | #endif |
5551 | read_lock(&tasklist_lock); | |
5552 | do_each_thread(g, p) { | |
5553 | /* | |
5554 | * reset the NMI-timeout, listing all files on a slow | |
5555 | * console might take alot of time: | |
5556 | */ | |
5557 | touch_nmi_watchdog(); | |
39bc89fd | 5558 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5559 | sched_show_task(p); |
1da177e4 LT |
5560 | } while_each_thread(g, p); |
5561 | ||
04c9167f JF |
5562 | touch_all_softlockup_watchdogs(); |
5563 | ||
dd41f596 IM |
5564 | #ifdef CONFIG_SCHED_DEBUG |
5565 | sysrq_sched_debug_show(); | |
5566 | #endif | |
1da177e4 | 5567 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5568 | /* |
5569 | * Only show locks if all tasks are dumped: | |
5570 | */ | |
5571 | if (state_filter == -1) | |
5572 | debug_show_all_locks(); | |
1da177e4 LT |
5573 | } |
5574 | ||
1df21055 IM |
5575 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5576 | { | |
dd41f596 | 5577 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5578 | } |
5579 | ||
f340c0d1 IM |
5580 | /** |
5581 | * init_idle - set up an idle thread for a given CPU | |
5582 | * @idle: task in question | |
5583 | * @cpu: cpu the idle task belongs to | |
5584 | * | |
5585 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5586 | * flag, to make booting more robust. | |
5587 | */ | |
5c1e1767 | 5588 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5589 | { |
70b97a7f | 5590 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5591 | unsigned long flags; |
5592 | ||
dd41f596 IM |
5593 | __sched_fork(idle); |
5594 | idle->se.exec_start = sched_clock(); | |
5595 | ||
b29739f9 | 5596 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5597 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5598 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5599 | |
5600 | spin_lock_irqsave(&rq->lock, flags); | |
5601 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5602 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5603 | idle->oncpu = 1; | |
5604 | #endif | |
1da177e4 LT |
5605 | spin_unlock_irqrestore(&rq->lock, flags); |
5606 | ||
5607 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5608 | #if defined(CONFIG_PREEMPT) |
5609 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5610 | #else | |
a1261f54 | 5611 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5612 | #endif |
dd41f596 IM |
5613 | /* |
5614 | * The idle tasks have their own, simple scheduling class: | |
5615 | */ | |
5616 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5617 | } |
5618 | ||
5619 | /* | |
5620 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5621 | * indicates which cpus entered this state. This is used | |
5622 | * in the rcu update to wait only for active cpus. For system | |
5623 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5624 | * always be CPU_MASK_NONE. | |
5625 | */ | |
5626 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5627 | ||
19978ca6 IM |
5628 | /* |
5629 | * Increase the granularity value when there are more CPUs, | |
5630 | * because with more CPUs the 'effective latency' as visible | |
5631 | * to users decreases. But the relationship is not linear, | |
5632 | * so pick a second-best guess by going with the log2 of the | |
5633 | * number of CPUs. | |
5634 | * | |
5635 | * This idea comes from the SD scheduler of Con Kolivas: | |
5636 | */ | |
5637 | static inline void sched_init_granularity(void) | |
5638 | { | |
5639 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5640 | const unsigned long limit = 200000000; | |
5641 | ||
5642 | sysctl_sched_min_granularity *= factor; | |
5643 | if (sysctl_sched_min_granularity > limit) | |
5644 | sysctl_sched_min_granularity = limit; | |
5645 | ||
5646 | sysctl_sched_latency *= factor; | |
5647 | if (sysctl_sched_latency > limit) | |
5648 | sysctl_sched_latency = limit; | |
5649 | ||
5650 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5651 | } |
5652 | ||
1da177e4 LT |
5653 | #ifdef CONFIG_SMP |
5654 | /* | |
5655 | * This is how migration works: | |
5656 | * | |
70b97a7f | 5657 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5658 | * runqueue and wake up that CPU's migration thread. |
5659 | * 2) we down() the locked semaphore => thread blocks. | |
5660 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5661 | * thread off the CPU) | |
5662 | * 4) it gets the migration request and checks whether the migrated | |
5663 | * task is still in the wrong runqueue. | |
5664 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5665 | * it and puts it into the right queue. | |
5666 | * 6) migration thread up()s the semaphore. | |
5667 | * 7) we wake up and the migration is done. | |
5668 | */ | |
5669 | ||
5670 | /* | |
5671 | * Change a given task's CPU affinity. Migrate the thread to a | |
5672 | * proper CPU and schedule it away if the CPU it's executing on | |
5673 | * is removed from the allowed bitmask. | |
5674 | * | |
5675 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5676 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5677 | * call is not atomic; no spinlocks may be held. |
5678 | */ | |
cd8ba7cd | 5679 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5680 | { |
70b97a7f | 5681 | struct migration_req req; |
1da177e4 | 5682 | unsigned long flags; |
70b97a7f | 5683 | struct rq *rq; |
48f24c4d | 5684 | int ret = 0; |
1da177e4 LT |
5685 | |
5686 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5687 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5688 | ret = -EINVAL; |
5689 | goto out; | |
5690 | } | |
5691 | ||
9985b0ba DR |
5692 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
5693 | !cpus_equal(p->cpus_allowed, *new_mask))) { | |
5694 | ret = -EINVAL; | |
5695 | goto out; | |
5696 | } | |
5697 | ||
73fe6aae | 5698 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5699 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5700 | else { |
cd8ba7cd MT |
5701 | p->cpus_allowed = *new_mask; |
5702 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5703 | } |
5704 | ||
1da177e4 | 5705 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5706 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5707 | goto out; |
5708 | ||
cd8ba7cd | 5709 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5710 | /* Need help from migration thread: drop lock and wait. */ |
5711 | task_rq_unlock(rq, &flags); | |
5712 | wake_up_process(rq->migration_thread); | |
5713 | wait_for_completion(&req.done); | |
5714 | tlb_migrate_finish(p->mm); | |
5715 | return 0; | |
5716 | } | |
5717 | out: | |
5718 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5719 | |
1da177e4 LT |
5720 | return ret; |
5721 | } | |
cd8ba7cd | 5722 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5723 | |
5724 | /* | |
41a2d6cf | 5725 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5726 | * this because either it can't run here any more (set_cpus_allowed() |
5727 | * away from this CPU, or CPU going down), or because we're | |
5728 | * attempting to rebalance this task on exec (sched_exec). | |
5729 | * | |
5730 | * So we race with normal scheduler movements, but that's OK, as long | |
5731 | * as the task is no longer on this CPU. | |
efc30814 KK |
5732 | * |
5733 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5734 | */ |
efc30814 | 5735 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5736 | { |
70b97a7f | 5737 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5738 | int ret = 0, on_rq; |
1da177e4 LT |
5739 | |
5740 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5741 | return ret; |
1da177e4 LT |
5742 | |
5743 | rq_src = cpu_rq(src_cpu); | |
5744 | rq_dest = cpu_rq(dest_cpu); | |
5745 | ||
5746 | double_rq_lock(rq_src, rq_dest); | |
5747 | /* Already moved. */ | |
5748 | if (task_cpu(p) != src_cpu) | |
5749 | goto out; | |
5750 | /* Affinity changed (again). */ | |
5751 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5752 | goto out; | |
5753 | ||
dd41f596 | 5754 | on_rq = p->se.on_rq; |
6e82a3be | 5755 | if (on_rq) |
2e1cb74a | 5756 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5757 | |
1da177e4 | 5758 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5759 | if (on_rq) { |
5760 | activate_task(rq_dest, p, 0); | |
5761 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5762 | } |
efc30814 | 5763 | ret = 1; |
1da177e4 LT |
5764 | out: |
5765 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5766 | return ret; |
1da177e4 LT |
5767 | } |
5768 | ||
5769 | /* | |
5770 | * migration_thread - this is a highprio system thread that performs | |
5771 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5772 | * another runqueue. | |
5773 | */ | |
95cdf3b7 | 5774 | static int migration_thread(void *data) |
1da177e4 | 5775 | { |
1da177e4 | 5776 | int cpu = (long)data; |
70b97a7f | 5777 | struct rq *rq; |
1da177e4 LT |
5778 | |
5779 | rq = cpu_rq(cpu); | |
5780 | BUG_ON(rq->migration_thread != current); | |
5781 | ||
5782 | set_current_state(TASK_INTERRUPTIBLE); | |
5783 | while (!kthread_should_stop()) { | |
70b97a7f | 5784 | struct migration_req *req; |
1da177e4 | 5785 | struct list_head *head; |
1da177e4 | 5786 | |
1da177e4 LT |
5787 | spin_lock_irq(&rq->lock); |
5788 | ||
5789 | if (cpu_is_offline(cpu)) { | |
5790 | spin_unlock_irq(&rq->lock); | |
5791 | goto wait_to_die; | |
5792 | } | |
5793 | ||
5794 | if (rq->active_balance) { | |
5795 | active_load_balance(rq, cpu); | |
5796 | rq->active_balance = 0; | |
5797 | } | |
5798 | ||
5799 | head = &rq->migration_queue; | |
5800 | ||
5801 | if (list_empty(head)) { | |
5802 | spin_unlock_irq(&rq->lock); | |
5803 | schedule(); | |
5804 | set_current_state(TASK_INTERRUPTIBLE); | |
5805 | continue; | |
5806 | } | |
70b97a7f | 5807 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5808 | list_del_init(head->next); |
5809 | ||
674311d5 NP |
5810 | spin_unlock(&rq->lock); |
5811 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5812 | local_irq_enable(); | |
1da177e4 LT |
5813 | |
5814 | complete(&req->done); | |
5815 | } | |
5816 | __set_current_state(TASK_RUNNING); | |
5817 | return 0; | |
5818 | ||
5819 | wait_to_die: | |
5820 | /* Wait for kthread_stop */ | |
5821 | set_current_state(TASK_INTERRUPTIBLE); | |
5822 | while (!kthread_should_stop()) { | |
5823 | schedule(); | |
5824 | set_current_state(TASK_INTERRUPTIBLE); | |
5825 | } | |
5826 | __set_current_state(TASK_RUNNING); | |
5827 | return 0; | |
5828 | } | |
5829 | ||
5830 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5831 | |
5832 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5833 | { | |
5834 | int ret; | |
5835 | ||
5836 | local_irq_disable(); | |
5837 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5838 | local_irq_enable(); | |
5839 | return ret; | |
5840 | } | |
5841 | ||
054b9108 | 5842 | /* |
3a4fa0a2 | 5843 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5844 | * NOTE: interrupts should be disabled by the caller |
5845 | */ | |
48f24c4d | 5846 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5847 | { |
efc30814 | 5848 | unsigned long flags; |
1da177e4 | 5849 | cpumask_t mask; |
70b97a7f IM |
5850 | struct rq *rq; |
5851 | int dest_cpu; | |
1da177e4 | 5852 | |
3a5c359a AK |
5853 | do { |
5854 | /* On same node? */ | |
5855 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5856 | cpus_and(mask, mask, p->cpus_allowed); | |
5857 | dest_cpu = any_online_cpu(mask); | |
5858 | ||
5859 | /* On any allowed CPU? */ | |
434d53b0 | 5860 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5861 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5862 | ||
5863 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5864 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5865 | cpumask_t cpus_allowed; |
5866 | ||
5867 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5868 | /* |
5869 | * Try to stay on the same cpuset, where the | |
5870 | * current cpuset may be a subset of all cpus. | |
5871 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5872 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5873 | * called within calls to cpuset_lock/cpuset_unlock. |
5874 | */ | |
3a5c359a | 5875 | rq = task_rq_lock(p, &flags); |
470fd646 | 5876 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5877 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5878 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5879 | |
3a5c359a AK |
5880 | /* |
5881 | * Don't tell them about moving exiting tasks or | |
5882 | * kernel threads (both mm NULL), since they never | |
5883 | * leave kernel. | |
5884 | */ | |
41a2d6cf | 5885 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5886 | printk(KERN_INFO "process %d (%s) no " |
5887 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5888 | task_pid_nr(p), p->comm, dead_cpu); |
5889 | } | |
3a5c359a | 5890 | } |
f7b4cddc | 5891 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5892 | } |
5893 | ||
5894 | /* | |
5895 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5896 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5897 | * for performance reasons the counter is not stricly tracking tasks to | |
5898 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5899 | * to keep the global sum constant after CPU-down: | |
5900 | */ | |
70b97a7f | 5901 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5902 | { |
7c16ec58 | 5903 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
5904 | unsigned long flags; |
5905 | ||
5906 | local_irq_save(flags); | |
5907 | double_rq_lock(rq_src, rq_dest); | |
5908 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5909 | rq_src->nr_uninterruptible = 0; | |
5910 | double_rq_unlock(rq_src, rq_dest); | |
5911 | local_irq_restore(flags); | |
5912 | } | |
5913 | ||
5914 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5915 | static void migrate_live_tasks(int src_cpu) | |
5916 | { | |
48f24c4d | 5917 | struct task_struct *p, *t; |
1da177e4 | 5918 | |
f7b4cddc | 5919 | read_lock(&tasklist_lock); |
1da177e4 | 5920 | |
48f24c4d IM |
5921 | do_each_thread(t, p) { |
5922 | if (p == current) | |
1da177e4 LT |
5923 | continue; |
5924 | ||
48f24c4d IM |
5925 | if (task_cpu(p) == src_cpu) |
5926 | move_task_off_dead_cpu(src_cpu, p); | |
5927 | } while_each_thread(t, p); | |
1da177e4 | 5928 | |
f7b4cddc | 5929 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5930 | } |
5931 | ||
dd41f596 IM |
5932 | /* |
5933 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5934 | * It does so by boosting its priority to highest possible. |
5935 | * Used by CPU offline code. | |
1da177e4 LT |
5936 | */ |
5937 | void sched_idle_next(void) | |
5938 | { | |
48f24c4d | 5939 | int this_cpu = smp_processor_id(); |
70b97a7f | 5940 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5941 | struct task_struct *p = rq->idle; |
5942 | unsigned long flags; | |
5943 | ||
5944 | /* cpu has to be offline */ | |
48f24c4d | 5945 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5946 | |
48f24c4d IM |
5947 | /* |
5948 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5949 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5950 | */ |
5951 | spin_lock_irqsave(&rq->lock, flags); | |
5952 | ||
dd41f596 | 5953 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5954 | |
94bc9a7b DA |
5955 | update_rq_clock(rq); |
5956 | activate_task(rq, p, 0); | |
1da177e4 LT |
5957 | |
5958 | spin_unlock_irqrestore(&rq->lock, flags); | |
5959 | } | |
5960 | ||
48f24c4d IM |
5961 | /* |
5962 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5963 | * offline. |
5964 | */ | |
5965 | void idle_task_exit(void) | |
5966 | { | |
5967 | struct mm_struct *mm = current->active_mm; | |
5968 | ||
5969 | BUG_ON(cpu_online(smp_processor_id())); | |
5970 | ||
5971 | if (mm != &init_mm) | |
5972 | switch_mm(mm, &init_mm, current); | |
5973 | mmdrop(mm); | |
5974 | } | |
5975 | ||
054b9108 | 5976 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5977 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5978 | { |
70b97a7f | 5979 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5980 | |
5981 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5982 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5983 | |
5984 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5985 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5986 | |
48f24c4d | 5987 | get_task_struct(p); |
1da177e4 LT |
5988 | |
5989 | /* | |
5990 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5991 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5992 | * fine. |
5993 | */ | |
f7b4cddc | 5994 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5995 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5996 | spin_lock_irq(&rq->lock); |
1da177e4 | 5997 | |
48f24c4d | 5998 | put_task_struct(p); |
1da177e4 LT |
5999 | } |
6000 | ||
6001 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6002 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6003 | { | |
70b97a7f | 6004 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6005 | struct task_struct *next; |
48f24c4d | 6006 | |
dd41f596 IM |
6007 | for ( ; ; ) { |
6008 | if (!rq->nr_running) | |
6009 | break; | |
a8e504d2 | 6010 | update_rq_clock(rq); |
ff95f3df | 6011 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6012 | if (!next) |
6013 | break; | |
6014 | migrate_dead(dead_cpu, next); | |
e692ab53 | 6015 | |
1da177e4 LT |
6016 | } |
6017 | } | |
6018 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6019 | ||
e692ab53 NP |
6020 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6021 | ||
6022 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6023 | { |
6024 | .procname = "sched_domain", | |
c57baf1e | 6025 | .mode = 0555, |
e0361851 | 6026 | }, |
38605cae | 6027 | {0, }, |
e692ab53 NP |
6028 | }; |
6029 | ||
6030 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6031 | { |
c57baf1e | 6032 | .ctl_name = CTL_KERN, |
e0361851 | 6033 | .procname = "kernel", |
c57baf1e | 6034 | .mode = 0555, |
e0361851 AD |
6035 | .child = sd_ctl_dir, |
6036 | }, | |
38605cae | 6037 | {0, }, |
e692ab53 NP |
6038 | }; |
6039 | ||
6040 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6041 | { | |
6042 | struct ctl_table *entry = | |
5cf9f062 | 6043 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6044 | |
e692ab53 NP |
6045 | return entry; |
6046 | } | |
6047 | ||
6382bc90 MM |
6048 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6049 | { | |
cd790076 | 6050 | struct ctl_table *entry; |
6382bc90 | 6051 | |
cd790076 MM |
6052 | /* |
6053 | * In the intermediate directories, both the child directory and | |
6054 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6055 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6056 | * static strings and all have proc handlers. |
6057 | */ | |
6058 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6059 | if (entry->child) |
6060 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6061 | if (entry->proc_handler == NULL) |
6062 | kfree(entry->procname); | |
6063 | } | |
6382bc90 MM |
6064 | |
6065 | kfree(*tablep); | |
6066 | *tablep = NULL; | |
6067 | } | |
6068 | ||
e692ab53 | 6069 | static void |
e0361851 | 6070 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6071 | const char *procname, void *data, int maxlen, |
6072 | mode_t mode, proc_handler *proc_handler) | |
6073 | { | |
e692ab53 NP |
6074 | entry->procname = procname; |
6075 | entry->data = data; | |
6076 | entry->maxlen = maxlen; | |
6077 | entry->mode = mode; | |
6078 | entry->proc_handler = proc_handler; | |
6079 | } | |
6080 | ||
6081 | static struct ctl_table * | |
6082 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6083 | { | |
ace8b3d6 | 6084 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 6085 | |
ad1cdc1d MM |
6086 | if (table == NULL) |
6087 | return NULL; | |
6088 | ||
e0361851 | 6089 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6090 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6091 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6092 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6093 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6094 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6095 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6096 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6097 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6098 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6099 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6100 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6101 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6102 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6103 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6104 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6105 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6106 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6107 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6108 | &sd->cache_nice_tries, |
6109 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6110 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6111 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 6112 | /* &table[11] is terminator */ |
e692ab53 NP |
6113 | |
6114 | return table; | |
6115 | } | |
6116 | ||
9a4e7159 | 6117 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6118 | { |
6119 | struct ctl_table *entry, *table; | |
6120 | struct sched_domain *sd; | |
6121 | int domain_num = 0, i; | |
6122 | char buf[32]; | |
6123 | ||
6124 | for_each_domain(cpu, sd) | |
6125 | domain_num++; | |
6126 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6127 | if (table == NULL) |
6128 | return NULL; | |
e692ab53 NP |
6129 | |
6130 | i = 0; | |
6131 | for_each_domain(cpu, sd) { | |
6132 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6133 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6134 | entry->mode = 0555; |
e692ab53 NP |
6135 | entry->child = sd_alloc_ctl_domain_table(sd); |
6136 | entry++; | |
6137 | i++; | |
6138 | } | |
6139 | return table; | |
6140 | } | |
6141 | ||
6142 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6143 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6144 | { |
6145 | int i, cpu_num = num_online_cpus(); | |
6146 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6147 | char buf[32]; | |
6148 | ||
7378547f MM |
6149 | WARN_ON(sd_ctl_dir[0].child); |
6150 | sd_ctl_dir[0].child = entry; | |
6151 | ||
ad1cdc1d MM |
6152 | if (entry == NULL) |
6153 | return; | |
6154 | ||
97b6ea7b | 6155 | for_each_online_cpu(i) { |
e692ab53 | 6156 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6157 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6158 | entry->mode = 0555; |
e692ab53 | 6159 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6160 | entry++; |
e692ab53 | 6161 | } |
7378547f MM |
6162 | |
6163 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6164 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6165 | } | |
6382bc90 | 6166 | |
7378547f | 6167 | /* may be called multiple times per register */ |
6382bc90 MM |
6168 | static void unregister_sched_domain_sysctl(void) |
6169 | { | |
7378547f MM |
6170 | if (sd_sysctl_header) |
6171 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6172 | sd_sysctl_header = NULL; |
7378547f MM |
6173 | if (sd_ctl_dir[0].child) |
6174 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6175 | } |
e692ab53 | 6176 | #else |
6382bc90 MM |
6177 | static void register_sched_domain_sysctl(void) |
6178 | { | |
6179 | } | |
6180 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6181 | { |
6182 | } | |
6183 | #endif | |
6184 | ||
1f11eb6a GH |
6185 | static void set_rq_online(struct rq *rq) |
6186 | { | |
6187 | if (!rq->online) { | |
6188 | const struct sched_class *class; | |
6189 | ||
6190 | cpu_set(rq->cpu, rq->rd->online); | |
6191 | rq->online = 1; | |
6192 | ||
6193 | for_each_class(class) { | |
6194 | if (class->rq_online) | |
6195 | class->rq_online(rq); | |
6196 | } | |
6197 | } | |
6198 | } | |
6199 | ||
6200 | static void set_rq_offline(struct rq *rq) | |
6201 | { | |
6202 | if (rq->online) { | |
6203 | const struct sched_class *class; | |
6204 | ||
6205 | for_each_class(class) { | |
6206 | if (class->rq_offline) | |
6207 | class->rq_offline(rq); | |
6208 | } | |
6209 | ||
6210 | cpu_clear(rq->cpu, rq->rd->online); | |
6211 | rq->online = 0; | |
6212 | } | |
6213 | } | |
6214 | ||
1da177e4 LT |
6215 | /* |
6216 | * migration_call - callback that gets triggered when a CPU is added. | |
6217 | * Here we can start up the necessary migration thread for the new CPU. | |
6218 | */ | |
48f24c4d IM |
6219 | static int __cpuinit |
6220 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6221 | { |
1da177e4 | 6222 | struct task_struct *p; |
48f24c4d | 6223 | int cpu = (long)hcpu; |
1da177e4 | 6224 | unsigned long flags; |
70b97a7f | 6225 | struct rq *rq; |
1da177e4 LT |
6226 | |
6227 | switch (action) { | |
5be9361c | 6228 | |
1da177e4 | 6229 | case CPU_UP_PREPARE: |
8bb78442 | 6230 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6231 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6232 | if (IS_ERR(p)) |
6233 | return NOTIFY_BAD; | |
1da177e4 LT |
6234 | kthread_bind(p, cpu); |
6235 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6236 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6237 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6238 | task_rq_unlock(rq, &flags); |
6239 | cpu_rq(cpu)->migration_thread = p; | |
6240 | break; | |
48f24c4d | 6241 | |
1da177e4 | 6242 | case CPU_ONLINE: |
8bb78442 | 6243 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6244 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6245 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6246 | |
6247 | /* Update our root-domain */ | |
6248 | rq = cpu_rq(cpu); | |
6249 | spin_lock_irqsave(&rq->lock, flags); | |
6250 | if (rq->rd) { | |
6251 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a GH |
6252 | |
6253 | set_rq_online(rq); | |
1f94ef59 GH |
6254 | } |
6255 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6256 | break; |
48f24c4d | 6257 | |
1da177e4 LT |
6258 | #ifdef CONFIG_HOTPLUG_CPU |
6259 | case CPU_UP_CANCELED: | |
8bb78442 | 6260 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6261 | if (!cpu_rq(cpu)->migration_thread) |
6262 | break; | |
41a2d6cf | 6263 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6264 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6265 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6266 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6267 | cpu_rq(cpu)->migration_thread = NULL; | |
6268 | break; | |
48f24c4d | 6269 | |
1da177e4 | 6270 | case CPU_DEAD: |
8bb78442 | 6271 | case CPU_DEAD_FROZEN: |
470fd646 | 6272 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6273 | migrate_live_tasks(cpu); |
6274 | rq = cpu_rq(cpu); | |
6275 | kthread_stop(rq->migration_thread); | |
6276 | rq->migration_thread = NULL; | |
6277 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6278 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6279 | update_rq_clock(rq); |
2e1cb74a | 6280 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6281 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6282 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6283 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6284 | migrate_dead_tasks(cpu); |
d2da272a | 6285 | spin_unlock_irq(&rq->lock); |
470fd646 | 6286 | cpuset_unlock(); |
1da177e4 LT |
6287 | migrate_nr_uninterruptible(rq); |
6288 | BUG_ON(rq->nr_running != 0); | |
6289 | ||
41a2d6cf IM |
6290 | /* |
6291 | * No need to migrate the tasks: it was best-effort if | |
6292 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6293 | * the requestors. | |
6294 | */ | |
1da177e4 LT |
6295 | spin_lock_irq(&rq->lock); |
6296 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6297 | struct migration_req *req; |
6298 | ||
1da177e4 | 6299 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6300 | struct migration_req, list); |
1da177e4 LT |
6301 | list_del_init(&req->list); |
6302 | complete(&req->done); | |
6303 | } | |
6304 | spin_unlock_irq(&rq->lock); | |
6305 | break; | |
57d885fe | 6306 | |
08f503b0 GH |
6307 | case CPU_DYING: |
6308 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6309 | /* Update our root-domain */ |
6310 | rq = cpu_rq(cpu); | |
6311 | spin_lock_irqsave(&rq->lock, flags); | |
6312 | if (rq->rd) { | |
6313 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
1f11eb6a | 6314 | set_rq_offline(rq); |
57d885fe GH |
6315 | } |
6316 | spin_unlock_irqrestore(&rq->lock, flags); | |
6317 | break; | |
1da177e4 LT |
6318 | #endif |
6319 | } | |
6320 | return NOTIFY_OK; | |
6321 | } | |
6322 | ||
6323 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6324 | * happens before everything else. | |
6325 | */ | |
26c2143b | 6326 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6327 | .notifier_call = migration_call, |
6328 | .priority = 10 | |
6329 | }; | |
6330 | ||
e6fe6649 | 6331 | void __init migration_init(void) |
1da177e4 LT |
6332 | { |
6333 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6334 | int err; |
48f24c4d IM |
6335 | |
6336 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6337 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6338 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6339 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6340 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6341 | } |
6342 | #endif | |
6343 | ||
6344 | #ifdef CONFIG_SMP | |
476f3534 | 6345 | |
3e9830dc | 6346 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6347 | |
099f98c8 GS |
6348 | static inline const char *sd_level_to_string(enum sched_domain_level lvl) |
6349 | { | |
6350 | switch (lvl) { | |
6351 | case SD_LV_NONE: | |
6352 | return "NONE"; | |
6353 | case SD_LV_SIBLING: | |
6354 | return "SIBLING"; | |
6355 | case SD_LV_MC: | |
6356 | return "MC"; | |
6357 | case SD_LV_CPU: | |
6358 | return "CPU"; | |
6359 | case SD_LV_NODE: | |
6360 | return "NODE"; | |
6361 | case SD_LV_ALLNODES: | |
6362 | return "ALLNODES"; | |
6363 | case SD_LV_MAX: | |
6364 | return "MAX"; | |
6365 | ||
6366 | } | |
6367 | return "MAX"; | |
6368 | } | |
6369 | ||
7c16ec58 MT |
6370 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6371 | cpumask_t *groupmask) | |
1da177e4 | 6372 | { |
4dcf6aff | 6373 | struct sched_group *group = sd->groups; |
434d53b0 | 6374 | char str[256]; |
1da177e4 | 6375 | |
434d53b0 | 6376 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6377 | cpus_clear(*groupmask); |
4dcf6aff IM |
6378 | |
6379 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6380 | ||
6381 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6382 | printk("does not load-balance\n"); | |
6383 | if (sd->parent) | |
6384 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6385 | " has parent"); | |
6386 | return -1; | |
41c7ce9a NP |
6387 | } |
6388 | ||
099f98c8 GS |
6389 | printk(KERN_CONT "span %s level %s\n", |
6390 | str, sd_level_to_string(sd->level)); | |
4dcf6aff IM |
6391 | |
6392 | if (!cpu_isset(cpu, sd->span)) { | |
6393 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6394 | "CPU%d\n", cpu); | |
6395 | } | |
6396 | if (!cpu_isset(cpu, group->cpumask)) { | |
6397 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6398 | " CPU%d\n", cpu); | |
6399 | } | |
1da177e4 | 6400 | |
4dcf6aff | 6401 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6402 | do { |
4dcf6aff IM |
6403 | if (!group) { |
6404 | printk("\n"); | |
6405 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6406 | break; |
6407 | } | |
6408 | ||
4dcf6aff IM |
6409 | if (!group->__cpu_power) { |
6410 | printk(KERN_CONT "\n"); | |
6411 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6412 | "set\n"); | |
6413 | break; | |
6414 | } | |
1da177e4 | 6415 | |
4dcf6aff IM |
6416 | if (!cpus_weight(group->cpumask)) { |
6417 | printk(KERN_CONT "\n"); | |
6418 | printk(KERN_ERR "ERROR: empty group\n"); | |
6419 | break; | |
6420 | } | |
1da177e4 | 6421 | |
7c16ec58 | 6422 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6423 | printk(KERN_CONT "\n"); |
6424 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6425 | break; | |
6426 | } | |
1da177e4 | 6427 | |
7c16ec58 | 6428 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6429 | |
434d53b0 | 6430 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6431 | printk(KERN_CONT " %s", str); |
1da177e4 | 6432 | |
4dcf6aff IM |
6433 | group = group->next; |
6434 | } while (group != sd->groups); | |
6435 | printk(KERN_CONT "\n"); | |
1da177e4 | 6436 | |
7c16ec58 | 6437 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6438 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6439 | |
7c16ec58 | 6440 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6441 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6442 | "of domain->span\n"); | |
6443 | return 0; | |
6444 | } | |
1da177e4 | 6445 | |
4dcf6aff IM |
6446 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6447 | { | |
7c16ec58 | 6448 | cpumask_t *groupmask; |
4dcf6aff | 6449 | int level = 0; |
1da177e4 | 6450 | |
4dcf6aff IM |
6451 | if (!sd) { |
6452 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6453 | return; | |
6454 | } | |
1da177e4 | 6455 | |
4dcf6aff IM |
6456 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6457 | ||
7c16ec58 MT |
6458 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6459 | if (!groupmask) { | |
6460 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6461 | return; | |
6462 | } | |
6463 | ||
4dcf6aff | 6464 | for (;;) { |
7c16ec58 | 6465 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6466 | break; |
1da177e4 LT |
6467 | level++; |
6468 | sd = sd->parent; | |
33859f7f | 6469 | if (!sd) |
4dcf6aff IM |
6470 | break; |
6471 | } | |
7c16ec58 | 6472 | kfree(groupmask); |
1da177e4 | 6473 | } |
6d6bc0ad | 6474 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6475 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6476 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6477 | |
1a20ff27 | 6478 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6479 | { |
6480 | if (cpus_weight(sd->span) == 1) | |
6481 | return 1; | |
6482 | ||
6483 | /* Following flags need at least 2 groups */ | |
6484 | if (sd->flags & (SD_LOAD_BALANCE | | |
6485 | SD_BALANCE_NEWIDLE | | |
6486 | SD_BALANCE_FORK | | |
89c4710e SS |
6487 | SD_BALANCE_EXEC | |
6488 | SD_SHARE_CPUPOWER | | |
6489 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6490 | if (sd->groups != sd->groups->next) |
6491 | return 0; | |
6492 | } | |
6493 | ||
6494 | /* Following flags don't use groups */ | |
6495 | if (sd->flags & (SD_WAKE_IDLE | | |
6496 | SD_WAKE_AFFINE | | |
6497 | SD_WAKE_BALANCE)) | |
6498 | return 0; | |
6499 | ||
6500 | return 1; | |
6501 | } | |
6502 | ||
48f24c4d IM |
6503 | static int |
6504 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6505 | { |
6506 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6507 | ||
6508 | if (sd_degenerate(parent)) | |
6509 | return 1; | |
6510 | ||
6511 | if (!cpus_equal(sd->span, parent->span)) | |
6512 | return 0; | |
6513 | ||
6514 | /* Does parent contain flags not in child? */ | |
6515 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6516 | if (cflags & SD_WAKE_AFFINE) | |
6517 | pflags &= ~SD_WAKE_BALANCE; | |
6518 | /* Flags needing groups don't count if only 1 group in parent */ | |
6519 | if (parent->groups == parent->groups->next) { | |
6520 | pflags &= ~(SD_LOAD_BALANCE | | |
6521 | SD_BALANCE_NEWIDLE | | |
6522 | SD_BALANCE_FORK | | |
89c4710e SS |
6523 | SD_BALANCE_EXEC | |
6524 | SD_SHARE_CPUPOWER | | |
6525 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6526 | } |
6527 | if (~cflags & pflags) | |
6528 | return 0; | |
6529 | ||
6530 | return 1; | |
6531 | } | |
6532 | ||
57d885fe GH |
6533 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6534 | { | |
6535 | unsigned long flags; | |
57d885fe GH |
6536 | |
6537 | spin_lock_irqsave(&rq->lock, flags); | |
6538 | ||
6539 | if (rq->rd) { | |
6540 | struct root_domain *old_rd = rq->rd; | |
6541 | ||
1f11eb6a GH |
6542 | if (cpu_isset(rq->cpu, old_rd->online)) |
6543 | set_rq_offline(rq); | |
57d885fe | 6544 | |
dc938520 | 6545 | cpu_clear(rq->cpu, old_rd->span); |
dc938520 | 6546 | |
57d885fe GH |
6547 | if (atomic_dec_and_test(&old_rd->refcount)) |
6548 | kfree(old_rd); | |
6549 | } | |
6550 | ||
6551 | atomic_inc(&rd->refcount); | |
6552 | rq->rd = rd; | |
6553 | ||
dc938520 | 6554 | cpu_set(rq->cpu, rd->span); |
1f94ef59 | 6555 | if (cpu_isset(rq->cpu, cpu_online_map)) |
1f11eb6a | 6556 | set_rq_online(rq); |
57d885fe GH |
6557 | |
6558 | spin_unlock_irqrestore(&rq->lock, flags); | |
6559 | } | |
6560 | ||
dc938520 | 6561 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6562 | { |
6563 | memset(rd, 0, sizeof(*rd)); | |
6564 | ||
dc938520 GH |
6565 | cpus_clear(rd->span); |
6566 | cpus_clear(rd->online); | |
6e0534f2 GH |
6567 | |
6568 | cpupri_init(&rd->cpupri); | |
57d885fe GH |
6569 | } |
6570 | ||
6571 | static void init_defrootdomain(void) | |
6572 | { | |
dc938520 | 6573 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6574 | atomic_set(&def_root_domain.refcount, 1); |
6575 | } | |
6576 | ||
dc938520 | 6577 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6578 | { |
6579 | struct root_domain *rd; | |
6580 | ||
6581 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6582 | if (!rd) | |
6583 | return NULL; | |
6584 | ||
dc938520 | 6585 | init_rootdomain(rd); |
57d885fe GH |
6586 | |
6587 | return rd; | |
6588 | } | |
6589 | ||
1da177e4 | 6590 | /* |
0eab9146 | 6591 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6592 | * hold the hotplug lock. |
6593 | */ | |
0eab9146 IM |
6594 | static void |
6595 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6596 | { |
70b97a7f | 6597 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6598 | struct sched_domain *tmp; |
6599 | ||
6600 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6601 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6602 | struct sched_domain *parent = tmp->parent; | |
6603 | if (!parent) | |
6604 | break; | |
1a848870 | 6605 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6606 | tmp->parent = parent->parent; |
1a848870 SS |
6607 | if (parent->parent) |
6608 | parent->parent->child = tmp; | |
6609 | } | |
245af2c7 SS |
6610 | } |
6611 | ||
1a848870 | 6612 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6613 | sd = sd->parent; |
1a848870 SS |
6614 | if (sd) |
6615 | sd->child = NULL; | |
6616 | } | |
1da177e4 LT |
6617 | |
6618 | sched_domain_debug(sd, cpu); | |
6619 | ||
57d885fe | 6620 | rq_attach_root(rq, rd); |
674311d5 | 6621 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6622 | } |
6623 | ||
6624 | /* cpus with isolated domains */ | |
67af63a6 | 6625 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6626 | |
6627 | /* Setup the mask of cpus configured for isolated domains */ | |
6628 | static int __init isolated_cpu_setup(char *str) | |
6629 | { | |
6630 | int ints[NR_CPUS], i; | |
6631 | ||
6632 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6633 | cpus_clear(cpu_isolated_map); | |
6634 | for (i = 1; i <= ints[0]; i++) | |
6635 | if (ints[i] < NR_CPUS) | |
6636 | cpu_set(ints[i], cpu_isolated_map); | |
6637 | return 1; | |
6638 | } | |
6639 | ||
8927f494 | 6640 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6641 | |
6642 | /* | |
6711cab4 SS |
6643 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6644 | * to a function which identifies what group(along with sched group) a CPU | |
6645 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6646 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6647 | * |
6648 | * init_sched_build_groups will build a circular linked list of the groups | |
6649 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6650 | * and ->cpu_power to 0. | |
6651 | */ | |
a616058b | 6652 | static void |
7c16ec58 | 6653 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6654 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6655 | struct sched_group **sg, |
6656 | cpumask_t *tmpmask), | |
6657 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6658 | { |
6659 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6660 | int i; |
6661 | ||
7c16ec58 MT |
6662 | cpus_clear(*covered); |
6663 | ||
6664 | for_each_cpu_mask(i, *span) { | |
6711cab4 | 6665 | struct sched_group *sg; |
7c16ec58 | 6666 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6667 | int j; |
6668 | ||
7c16ec58 | 6669 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6670 | continue; |
6671 | ||
7c16ec58 | 6672 | cpus_clear(sg->cpumask); |
5517d86b | 6673 | sg->__cpu_power = 0; |
1da177e4 | 6674 | |
7c16ec58 MT |
6675 | for_each_cpu_mask(j, *span) { |
6676 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) | |
1da177e4 LT |
6677 | continue; |
6678 | ||
7c16ec58 | 6679 | cpu_set(j, *covered); |
1da177e4 LT |
6680 | cpu_set(j, sg->cpumask); |
6681 | } | |
6682 | if (!first) | |
6683 | first = sg; | |
6684 | if (last) | |
6685 | last->next = sg; | |
6686 | last = sg; | |
6687 | } | |
6688 | last->next = first; | |
6689 | } | |
6690 | ||
9c1cfda2 | 6691 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6692 | |
9c1cfda2 | 6693 | #ifdef CONFIG_NUMA |
198e2f18 | 6694 | |
9c1cfda2 JH |
6695 | /** |
6696 | * find_next_best_node - find the next node to include in a sched_domain | |
6697 | * @node: node whose sched_domain we're building | |
6698 | * @used_nodes: nodes already in the sched_domain | |
6699 | * | |
41a2d6cf | 6700 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6701 | * finds the closest node not already in the @used_nodes map. |
6702 | * | |
6703 | * Should use nodemask_t. | |
6704 | */ | |
c5f59f08 | 6705 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6706 | { |
6707 | int i, n, val, min_val, best_node = 0; | |
6708 | ||
6709 | min_val = INT_MAX; | |
6710 | ||
6711 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6712 | /* Start at @node */ | |
6713 | n = (node + i) % MAX_NUMNODES; | |
6714 | ||
6715 | if (!nr_cpus_node(n)) | |
6716 | continue; | |
6717 | ||
6718 | /* Skip already used nodes */ | |
c5f59f08 | 6719 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6720 | continue; |
6721 | ||
6722 | /* Simple min distance search */ | |
6723 | val = node_distance(node, n); | |
6724 | ||
6725 | if (val < min_val) { | |
6726 | min_val = val; | |
6727 | best_node = n; | |
6728 | } | |
6729 | } | |
6730 | ||
c5f59f08 | 6731 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6732 | return best_node; |
6733 | } | |
6734 | ||
6735 | /** | |
6736 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6737 | * @node: node whose cpumask we're constructing | |
73486722 | 6738 | * @span: resulting cpumask |
9c1cfda2 | 6739 | * |
41a2d6cf | 6740 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6741 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6742 | * out optimally. | |
6743 | */ | |
4bdbaad3 | 6744 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6745 | { |
c5f59f08 | 6746 | nodemask_t used_nodes; |
c5f59f08 | 6747 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6748 | int i; |
9c1cfda2 | 6749 | |
4bdbaad3 | 6750 | cpus_clear(*span); |
c5f59f08 | 6751 | nodes_clear(used_nodes); |
9c1cfda2 | 6752 | |
4bdbaad3 | 6753 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6754 | node_set(node, used_nodes); |
9c1cfda2 JH |
6755 | |
6756 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6757 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6758 | |
c5f59f08 | 6759 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6760 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6761 | } |
9c1cfda2 | 6762 | } |
6d6bc0ad | 6763 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6764 | |
5c45bf27 | 6765 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6766 | |
9c1cfda2 | 6767 | /* |
48f24c4d | 6768 | * SMT sched-domains: |
9c1cfda2 | 6769 | */ |
1da177e4 LT |
6770 | #ifdef CONFIG_SCHED_SMT |
6771 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6772 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6773 | |
41a2d6cf | 6774 | static int |
7c16ec58 MT |
6775 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6776 | cpumask_t *unused) | |
1da177e4 | 6777 | { |
6711cab4 SS |
6778 | if (sg) |
6779 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6780 | return cpu; |
6781 | } | |
6d6bc0ad | 6782 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6783 | |
48f24c4d IM |
6784 | /* |
6785 | * multi-core sched-domains: | |
6786 | */ | |
1e9f28fa SS |
6787 | #ifdef CONFIG_SCHED_MC |
6788 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6789 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
6d6bc0ad | 6790 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6791 | |
6792 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6793 | static int |
7c16ec58 MT |
6794 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6795 | cpumask_t *mask) | |
1e9f28fa | 6796 | { |
6711cab4 | 6797 | int group; |
7c16ec58 MT |
6798 | |
6799 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6800 | cpus_and(*mask, *mask, *cpu_map); | |
6801 | group = first_cpu(*mask); | |
6711cab4 SS |
6802 | if (sg) |
6803 | *sg = &per_cpu(sched_group_core, group); | |
6804 | return group; | |
1e9f28fa SS |
6805 | } |
6806 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6807 | static int |
7c16ec58 MT |
6808 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6809 | cpumask_t *unused) | |
1e9f28fa | 6810 | { |
6711cab4 SS |
6811 | if (sg) |
6812 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6813 | return cpu; |
6814 | } | |
6815 | #endif | |
6816 | ||
1da177e4 | 6817 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6818 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6819 | |
41a2d6cf | 6820 | static int |
7c16ec58 MT |
6821 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6822 | cpumask_t *mask) | |
1da177e4 | 6823 | { |
6711cab4 | 6824 | int group; |
48f24c4d | 6825 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6826 | *mask = cpu_coregroup_map(cpu); |
6827 | cpus_and(*mask, *mask, *cpu_map); | |
6828 | group = first_cpu(*mask); | |
1e9f28fa | 6829 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6830 | *mask = per_cpu(cpu_sibling_map, cpu); |
6831 | cpus_and(*mask, *mask, *cpu_map); | |
6832 | group = first_cpu(*mask); | |
1da177e4 | 6833 | #else |
6711cab4 | 6834 | group = cpu; |
1da177e4 | 6835 | #endif |
6711cab4 SS |
6836 | if (sg) |
6837 | *sg = &per_cpu(sched_group_phys, group); | |
6838 | return group; | |
1da177e4 LT |
6839 | } |
6840 | ||
6841 | #ifdef CONFIG_NUMA | |
1da177e4 | 6842 | /* |
9c1cfda2 JH |
6843 | * The init_sched_build_groups can't handle what we want to do with node |
6844 | * groups, so roll our own. Now each node has its own list of groups which | |
6845 | * gets dynamically allocated. | |
1da177e4 | 6846 | */ |
9c1cfda2 | 6847 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6848 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6849 | |
9c1cfda2 | 6850 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6851 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6852 | |
6711cab4 | 6853 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 6854 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 6855 | { |
6711cab4 SS |
6856 | int group; |
6857 | ||
7c16ec58 MT |
6858 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6859 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6860 | group = first_cpu(*nodemask); | |
6711cab4 SS |
6861 | |
6862 | if (sg) | |
6863 | *sg = &per_cpu(sched_group_allnodes, group); | |
6864 | return group; | |
1da177e4 | 6865 | } |
6711cab4 | 6866 | |
08069033 SS |
6867 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6868 | { | |
6869 | struct sched_group *sg = group_head; | |
6870 | int j; | |
6871 | ||
6872 | if (!sg) | |
6873 | return; | |
3a5c359a AK |
6874 | do { |
6875 | for_each_cpu_mask(j, sg->cpumask) { | |
6876 | struct sched_domain *sd; | |
08069033 | 6877 | |
3a5c359a AK |
6878 | sd = &per_cpu(phys_domains, j); |
6879 | if (j != first_cpu(sd->groups->cpumask)) { | |
6880 | /* | |
6881 | * Only add "power" once for each | |
6882 | * physical package. | |
6883 | */ | |
6884 | continue; | |
6885 | } | |
08069033 | 6886 | |
3a5c359a AK |
6887 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6888 | } | |
6889 | sg = sg->next; | |
6890 | } while (sg != group_head); | |
08069033 | 6891 | } |
6d6bc0ad | 6892 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6893 | |
a616058b | 6894 | #ifdef CONFIG_NUMA |
51888ca2 | 6895 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 6896 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 6897 | { |
a616058b | 6898 | int cpu, i; |
51888ca2 SV |
6899 | |
6900 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6901 | struct sched_group **sched_group_nodes |
6902 | = sched_group_nodes_bycpu[cpu]; | |
6903 | ||
51888ca2 SV |
6904 | if (!sched_group_nodes) |
6905 | continue; | |
6906 | ||
6907 | for (i = 0; i < MAX_NUMNODES; i++) { | |
51888ca2 SV |
6908 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6909 | ||
7c16ec58 MT |
6910 | *nodemask = node_to_cpumask(i); |
6911 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6912 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
6913 | continue; |
6914 | ||
6915 | if (sg == NULL) | |
6916 | continue; | |
6917 | sg = sg->next; | |
6918 | next_sg: | |
6919 | oldsg = sg; | |
6920 | sg = sg->next; | |
6921 | kfree(oldsg); | |
6922 | if (oldsg != sched_group_nodes[i]) | |
6923 | goto next_sg; | |
6924 | } | |
6925 | kfree(sched_group_nodes); | |
6926 | sched_group_nodes_bycpu[cpu] = NULL; | |
6927 | } | |
51888ca2 | 6928 | } |
6d6bc0ad | 6929 | #else /* !CONFIG_NUMA */ |
7c16ec58 | 6930 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
6931 | { |
6932 | } | |
6d6bc0ad | 6933 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6934 | |
89c4710e SS |
6935 | /* |
6936 | * Initialize sched groups cpu_power. | |
6937 | * | |
6938 | * cpu_power indicates the capacity of sched group, which is used while | |
6939 | * distributing the load between different sched groups in a sched domain. | |
6940 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6941 | * there are asymmetries in the topology. If there are asymmetries, group | |
6942 | * having more cpu_power will pickup more load compared to the group having | |
6943 | * less cpu_power. | |
6944 | * | |
6945 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6946 | * the maximum number of tasks a group can handle in the presence of other idle | |
6947 | * or lightly loaded groups in the same sched domain. | |
6948 | */ | |
6949 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6950 | { | |
6951 | struct sched_domain *child; | |
6952 | struct sched_group *group; | |
6953 | ||
6954 | WARN_ON(!sd || !sd->groups); | |
6955 | ||
6956 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6957 | return; | |
6958 | ||
6959 | child = sd->child; | |
6960 | ||
5517d86b ED |
6961 | sd->groups->__cpu_power = 0; |
6962 | ||
89c4710e SS |
6963 | /* |
6964 | * For perf policy, if the groups in child domain share resources | |
6965 | * (for example cores sharing some portions of the cache hierarchy | |
6966 | * or SMT), then set this domain groups cpu_power such that each group | |
6967 | * can handle only one task, when there are other idle groups in the | |
6968 | * same sched domain. | |
6969 | */ | |
6970 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6971 | (child->flags & | |
6972 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6973 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6974 | return; |
6975 | } | |
6976 | ||
89c4710e SS |
6977 | /* |
6978 | * add cpu_power of each child group to this groups cpu_power | |
6979 | */ | |
6980 | group = child->groups; | |
6981 | do { | |
5517d86b | 6982 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6983 | group = group->next; |
6984 | } while (group != child->groups); | |
6985 | } | |
6986 | ||
7c16ec58 MT |
6987 | /* |
6988 | * Initializers for schedule domains | |
6989 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6990 | */ | |
6991 | ||
6992 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
6993 | #define SD_INIT_FUNC(type) \ | |
6994 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6995 | { \ | |
6996 | memset(sd, 0, sizeof(*sd)); \ | |
6997 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6998 | sd->level = SD_LV_##type; \ |
7c16ec58 MT |
6999 | } |
7000 | ||
7001 | SD_INIT_FUNC(CPU) | |
7002 | #ifdef CONFIG_NUMA | |
7003 | SD_INIT_FUNC(ALLNODES) | |
7004 | SD_INIT_FUNC(NODE) | |
7005 | #endif | |
7006 | #ifdef CONFIG_SCHED_SMT | |
7007 | SD_INIT_FUNC(SIBLING) | |
7008 | #endif | |
7009 | #ifdef CONFIG_SCHED_MC | |
7010 | SD_INIT_FUNC(MC) | |
7011 | #endif | |
7012 | ||
7013 | /* | |
7014 | * To minimize stack usage kmalloc room for cpumasks and share the | |
7015 | * space as the usage in build_sched_domains() dictates. Used only | |
7016 | * if the amount of space is significant. | |
7017 | */ | |
7018 | struct allmasks { | |
7019 | cpumask_t tmpmask; /* make this one first */ | |
7020 | union { | |
7021 | cpumask_t nodemask; | |
7022 | cpumask_t this_sibling_map; | |
7023 | cpumask_t this_core_map; | |
7024 | }; | |
7025 | cpumask_t send_covered; | |
7026 | ||
7027 | #ifdef CONFIG_NUMA | |
7028 | cpumask_t domainspan; | |
7029 | cpumask_t covered; | |
7030 | cpumask_t notcovered; | |
7031 | #endif | |
7032 | }; | |
7033 | ||
7034 | #if NR_CPUS > 128 | |
7035 | #define SCHED_CPUMASK_ALLOC 1 | |
7036 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
7037 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
7038 | #else | |
7039 | #define SCHED_CPUMASK_ALLOC 0 | |
7040 | #define SCHED_CPUMASK_FREE(v) | |
7041 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
7042 | #endif | |
7043 | ||
7044 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
7045 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
7046 | ||
1d3504fc HS |
7047 | static int default_relax_domain_level = -1; |
7048 | ||
7049 | static int __init setup_relax_domain_level(char *str) | |
7050 | { | |
30e0e178 LZ |
7051 | unsigned long val; |
7052 | ||
7053 | val = simple_strtoul(str, NULL, 0); | |
7054 | if (val < SD_LV_MAX) | |
7055 | default_relax_domain_level = val; | |
7056 | ||
1d3504fc HS |
7057 | return 1; |
7058 | } | |
7059 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7060 | ||
7061 | static void set_domain_attribute(struct sched_domain *sd, | |
7062 | struct sched_domain_attr *attr) | |
7063 | { | |
7064 | int request; | |
7065 | ||
7066 | if (!attr || attr->relax_domain_level < 0) { | |
7067 | if (default_relax_domain_level < 0) | |
7068 | return; | |
7069 | else | |
7070 | request = default_relax_domain_level; | |
7071 | } else | |
7072 | request = attr->relax_domain_level; | |
7073 | if (request < sd->level) { | |
7074 | /* turn off idle balance on this domain */ | |
7075 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7076 | } else { | |
7077 | /* turn on idle balance on this domain */ | |
7078 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7079 | } | |
7080 | } | |
7081 | ||
1da177e4 | 7082 | /* |
1a20ff27 DG |
7083 | * Build sched domains for a given set of cpus and attach the sched domains |
7084 | * to the individual cpus | |
1da177e4 | 7085 | */ |
1d3504fc HS |
7086 | static int __build_sched_domains(const cpumask_t *cpu_map, |
7087 | struct sched_domain_attr *attr) | |
1da177e4 LT |
7088 | { |
7089 | int i; | |
57d885fe | 7090 | struct root_domain *rd; |
7c16ec58 MT |
7091 | SCHED_CPUMASK_DECLARE(allmasks); |
7092 | cpumask_t *tmpmask; | |
d1b55138 JH |
7093 | #ifdef CONFIG_NUMA |
7094 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 7095 | int sd_allnodes = 0; |
d1b55138 JH |
7096 | |
7097 | /* | |
7098 | * Allocate the per-node list of sched groups | |
7099 | */ | |
5cf9f062 | 7100 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 7101 | GFP_KERNEL); |
d1b55138 JH |
7102 | if (!sched_group_nodes) { |
7103 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 7104 | return -ENOMEM; |
d1b55138 | 7105 | } |
d1b55138 | 7106 | #endif |
1da177e4 | 7107 | |
dc938520 | 7108 | rd = alloc_rootdomain(); |
57d885fe GH |
7109 | if (!rd) { |
7110 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
7111 | #ifdef CONFIG_NUMA |
7112 | kfree(sched_group_nodes); | |
7113 | #endif | |
57d885fe GH |
7114 | return -ENOMEM; |
7115 | } | |
7116 | ||
7c16ec58 MT |
7117 | #if SCHED_CPUMASK_ALLOC |
7118 | /* get space for all scratch cpumask variables */ | |
7119 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7120 | if (!allmasks) { | |
7121 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7122 | kfree(rd); | |
7123 | #ifdef CONFIG_NUMA | |
7124 | kfree(sched_group_nodes); | |
7125 | #endif | |
7126 | return -ENOMEM; | |
7127 | } | |
7128 | #endif | |
7129 | tmpmask = (cpumask_t *)allmasks; | |
7130 | ||
7131 | ||
7132 | #ifdef CONFIG_NUMA | |
7133 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7134 | #endif | |
7135 | ||
1da177e4 | 7136 | /* |
1a20ff27 | 7137 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7138 | */ |
1a20ff27 | 7139 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 7140 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7141 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7142 | |
7c16ec58 MT |
7143 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7144 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7145 | |
7146 | #ifdef CONFIG_NUMA | |
dd41f596 | 7147 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7148 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7149 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7150 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7151 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7152 | sd->span = *cpu_map; |
7c16ec58 | 7153 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7154 | p = sd; |
6711cab4 | 7155 | sd_allnodes = 1; |
9c1cfda2 JH |
7156 | } else |
7157 | p = NULL; | |
7158 | ||
1da177e4 | 7159 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7160 | SD_INIT(sd, NODE); |
1d3504fc | 7161 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7162 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7163 | sd->parent = p; |
1a848870 SS |
7164 | if (p) |
7165 | p->child = sd; | |
9c1cfda2 | 7166 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7167 | #endif |
7168 | ||
7169 | p = sd; | |
7170 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7171 | SD_INIT(sd, CPU); |
1d3504fc | 7172 | set_domain_attribute(sd, attr); |
7c16ec58 | 7173 | sd->span = *nodemask; |
1da177e4 | 7174 | sd->parent = p; |
1a848870 SS |
7175 | if (p) |
7176 | p->child = sd; | |
7c16ec58 | 7177 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7178 | |
1e9f28fa SS |
7179 | #ifdef CONFIG_SCHED_MC |
7180 | p = sd; | |
7181 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7182 | SD_INIT(sd, MC); |
1d3504fc | 7183 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7184 | sd->span = cpu_coregroup_map(i); |
7185 | cpus_and(sd->span, sd->span, *cpu_map); | |
7186 | sd->parent = p; | |
1a848870 | 7187 | p->child = sd; |
7c16ec58 | 7188 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7189 | #endif |
7190 | ||
1da177e4 LT |
7191 | #ifdef CONFIG_SCHED_SMT |
7192 | p = sd; | |
7193 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7194 | SD_INIT(sd, SIBLING); |
1d3504fc | 7195 | set_domain_attribute(sd, attr); |
d5a7430d | 7196 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7197 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7198 | sd->parent = p; |
1a848870 | 7199 | p->child = sd; |
7c16ec58 | 7200 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7201 | #endif |
7202 | } | |
7203 | ||
7204 | #ifdef CONFIG_SCHED_SMT | |
7205 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 7206 | for_each_cpu_mask(i, *cpu_map) { |
7c16ec58 MT |
7207 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7208 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7209 | ||
7210 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7211 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7212 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7213 | continue; |
7214 | ||
dd41f596 | 7215 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7216 | &cpu_to_cpu_group, |
7217 | send_covered, tmpmask); | |
1da177e4 LT |
7218 | } |
7219 | #endif | |
7220 | ||
1e9f28fa SS |
7221 | #ifdef CONFIG_SCHED_MC |
7222 | /* Set up multi-core groups */ | |
7223 | for_each_cpu_mask(i, *cpu_map) { | |
7c16ec58 MT |
7224 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7225 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7226 | ||
7227 | *this_core_map = cpu_coregroup_map(i); | |
7228 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7229 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7230 | continue; |
7c16ec58 | 7231 | |
dd41f596 | 7232 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7233 | &cpu_to_core_group, |
7234 | send_covered, tmpmask); | |
1e9f28fa SS |
7235 | } |
7236 | #endif | |
7237 | ||
1da177e4 LT |
7238 | /* Set up physical groups */ |
7239 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7c16ec58 MT |
7240 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7241 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7242 | |
7c16ec58 MT |
7243 | *nodemask = node_to_cpumask(i); |
7244 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7245 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7246 | continue; |
7247 | ||
7c16ec58 MT |
7248 | init_sched_build_groups(nodemask, cpu_map, |
7249 | &cpu_to_phys_group, | |
7250 | send_covered, tmpmask); | |
1da177e4 LT |
7251 | } |
7252 | ||
7253 | #ifdef CONFIG_NUMA | |
7254 | /* Set up node groups */ | |
7c16ec58 MT |
7255 | if (sd_allnodes) { |
7256 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7257 | ||
7258 | init_sched_build_groups(cpu_map, cpu_map, | |
7259 | &cpu_to_allnodes_group, | |
7260 | send_covered, tmpmask); | |
7261 | } | |
9c1cfda2 JH |
7262 | |
7263 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7264 | /* Set up node groups */ | |
7265 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7266 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7267 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7268 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7269 | int j; |
7270 | ||
7c16ec58 MT |
7271 | *nodemask = node_to_cpumask(i); |
7272 | cpus_clear(*covered); | |
7273 | ||
7274 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7275 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7276 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7277 | continue; |
d1b55138 | 7278 | } |
9c1cfda2 | 7279 | |
4bdbaad3 | 7280 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7281 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7282 | |
15f0b676 | 7283 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7284 | if (!sg) { |
7285 | printk(KERN_WARNING "Can not alloc domain group for " | |
7286 | "node %d\n", i); | |
7287 | goto error; | |
7288 | } | |
9c1cfda2 | 7289 | sched_group_nodes[i] = sg; |
7c16ec58 | 7290 | for_each_cpu_mask(j, *nodemask) { |
9c1cfda2 | 7291 | struct sched_domain *sd; |
9761eea8 | 7292 | |
9c1cfda2 JH |
7293 | sd = &per_cpu(node_domains, j); |
7294 | sd->groups = sg; | |
9c1cfda2 | 7295 | } |
5517d86b | 7296 | sg->__cpu_power = 0; |
7c16ec58 | 7297 | sg->cpumask = *nodemask; |
51888ca2 | 7298 | sg->next = sg; |
7c16ec58 | 7299 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7300 | prev = sg; |
7301 | ||
7302 | for (j = 0; j < MAX_NUMNODES; j++) { | |
7c16ec58 | 7303 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
9c1cfda2 | 7304 | int n = (i + j) % MAX_NUMNODES; |
c5f59f08 | 7305 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7306 | |
7c16ec58 MT |
7307 | cpus_complement(*notcovered, *covered); |
7308 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7309 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7310 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7311 | break; |
7312 | ||
7c16ec58 MT |
7313 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7314 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7315 | continue; |
7316 | ||
15f0b676 SV |
7317 | sg = kmalloc_node(sizeof(struct sched_group), |
7318 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7319 | if (!sg) { |
7320 | printk(KERN_WARNING | |
7321 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7322 | goto error; |
9c1cfda2 | 7323 | } |
5517d86b | 7324 | sg->__cpu_power = 0; |
7c16ec58 | 7325 | sg->cpumask = *tmpmask; |
51888ca2 | 7326 | sg->next = prev->next; |
7c16ec58 | 7327 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7328 | prev->next = sg; |
7329 | prev = sg; | |
7330 | } | |
9c1cfda2 | 7331 | } |
1da177e4 LT |
7332 | #endif |
7333 | ||
7334 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7335 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 7336 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7337 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7338 | ||
89c4710e | 7339 | init_sched_groups_power(i, sd); |
5c45bf27 | 7340 | } |
1da177e4 | 7341 | #endif |
1e9f28fa | 7342 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 7343 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7344 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7345 | ||
89c4710e | 7346 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7347 | } |
7348 | #endif | |
1e9f28fa | 7349 | |
5c45bf27 | 7350 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7351 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7352 | ||
89c4710e | 7353 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7354 | } |
7355 | ||
9c1cfda2 | 7356 | #ifdef CONFIG_NUMA |
08069033 SS |
7357 | for (i = 0; i < MAX_NUMNODES; i++) |
7358 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 7359 | |
6711cab4 SS |
7360 | if (sd_allnodes) { |
7361 | struct sched_group *sg; | |
f712c0c7 | 7362 | |
7c16ec58 MT |
7363 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7364 | tmpmask); | |
f712c0c7 SS |
7365 | init_numa_sched_groups_power(sg); |
7366 | } | |
9c1cfda2 JH |
7367 | #endif |
7368 | ||
1da177e4 | 7369 | /* Attach the domains */ |
1a20ff27 | 7370 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
7371 | struct sched_domain *sd; |
7372 | #ifdef CONFIG_SCHED_SMT | |
7373 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7374 | #elif defined(CONFIG_SCHED_MC) |
7375 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7376 | #else |
7377 | sd = &per_cpu(phys_domains, i); | |
7378 | #endif | |
57d885fe | 7379 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7380 | } |
51888ca2 | 7381 | |
7c16ec58 | 7382 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7383 | return 0; |
7384 | ||
a616058b | 7385 | #ifdef CONFIG_NUMA |
51888ca2 | 7386 | error: |
7c16ec58 MT |
7387 | free_sched_groups(cpu_map, tmpmask); |
7388 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7389 | return -ENOMEM; |
a616058b | 7390 | #endif |
1da177e4 | 7391 | } |
029190c5 | 7392 | |
1d3504fc HS |
7393 | static int build_sched_domains(const cpumask_t *cpu_map) |
7394 | { | |
7395 | return __build_sched_domains(cpu_map, NULL); | |
7396 | } | |
7397 | ||
029190c5 PJ |
7398 | static cpumask_t *doms_cur; /* current sched domains */ |
7399 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7400 | static struct sched_domain_attr *dattr_cur; |
7401 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7402 | |
7403 | /* | |
7404 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7405 | * cpumask_t) fails, then fallback to a single sched domain, | |
7406 | * as determined by the single cpumask_t fallback_doms. | |
7407 | */ | |
7408 | static cpumask_t fallback_doms; | |
7409 | ||
22e52b07 HC |
7410 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7411 | { | |
7412 | } | |
7413 | ||
5c8e1ed1 MK |
7414 | /* |
7415 | * Free current domain masks. | |
7416 | * Called after all cpus are attached to NULL domain. | |
7417 | */ | |
7418 | static void free_sched_domains(void) | |
7419 | { | |
7420 | ndoms_cur = 0; | |
7421 | if (doms_cur != &fallback_doms) | |
7422 | kfree(doms_cur); | |
7423 | doms_cur = &fallback_doms; | |
7424 | } | |
7425 | ||
1a20ff27 | 7426 | /* |
41a2d6cf | 7427 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7428 | * For now this just excludes isolated cpus, but could be used to |
7429 | * exclude other special cases in the future. | |
1a20ff27 | 7430 | */ |
51888ca2 | 7431 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7432 | { |
7378547f MM |
7433 | int err; |
7434 | ||
22e52b07 | 7435 | arch_update_cpu_topology(); |
029190c5 PJ |
7436 | ndoms_cur = 1; |
7437 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7438 | if (!doms_cur) | |
7439 | doms_cur = &fallback_doms; | |
7440 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7441 | dattr_cur = NULL; |
7378547f | 7442 | err = build_sched_domains(doms_cur); |
6382bc90 | 7443 | register_sched_domain_sysctl(); |
7378547f MM |
7444 | |
7445 | return err; | |
1a20ff27 DG |
7446 | } |
7447 | ||
7c16ec58 MT |
7448 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7449 | cpumask_t *tmpmask) | |
1da177e4 | 7450 | { |
7c16ec58 | 7451 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7452 | } |
1da177e4 | 7453 | |
1a20ff27 DG |
7454 | /* |
7455 | * Detach sched domains from a group of cpus specified in cpu_map | |
7456 | * These cpus will now be attached to the NULL domain | |
7457 | */ | |
858119e1 | 7458 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7459 | { |
7c16ec58 | 7460 | cpumask_t tmpmask; |
1a20ff27 DG |
7461 | int i; |
7462 | ||
6382bc90 MM |
7463 | unregister_sched_domain_sysctl(); |
7464 | ||
1a20ff27 | 7465 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 7466 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7467 | synchronize_sched(); |
7c16ec58 | 7468 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7469 | } |
7470 | ||
1d3504fc HS |
7471 | /* handle null as "default" */ |
7472 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7473 | struct sched_domain_attr *new, int idx_new) | |
7474 | { | |
7475 | struct sched_domain_attr tmp; | |
7476 | ||
7477 | /* fast path */ | |
7478 | if (!new && !cur) | |
7479 | return 1; | |
7480 | ||
7481 | tmp = SD_ATTR_INIT; | |
7482 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7483 | new ? (new + idx_new) : &tmp, | |
7484 | sizeof(struct sched_domain_attr)); | |
7485 | } | |
7486 | ||
029190c5 PJ |
7487 | /* |
7488 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7489 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7490 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7491 | * It destroys each deleted domain and builds each new domain. | |
7492 | * | |
7493 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7494 | * The masks don't intersect (don't overlap.) We should setup one |
7495 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7496 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7497 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7498 | * it as it is. | |
7499 | * | |
41a2d6cf IM |
7500 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7501 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7502 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7503 | * and partition_sched_domains() will fallback to the single partition | |
7504 | * 'fallback_doms'. | |
7505 | * | |
7506 | * Call with hotplug lock held | |
7507 | */ | |
1d3504fc HS |
7508 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7509 | struct sched_domain_attr *dattr_new) | |
029190c5 PJ |
7510 | { |
7511 | int i, j; | |
7512 | ||
712555ee | 7513 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7514 | |
7378547f MM |
7515 | /* always unregister in case we don't destroy any domains */ |
7516 | unregister_sched_domain_sysctl(); | |
7517 | ||
029190c5 PJ |
7518 | if (doms_new == NULL) { |
7519 | ndoms_new = 1; | |
7520 | doms_new = &fallback_doms; | |
7521 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
1d3504fc | 7522 | dattr_new = NULL; |
029190c5 PJ |
7523 | } |
7524 | ||
7525 | /* Destroy deleted domains */ | |
7526 | for (i = 0; i < ndoms_cur; i++) { | |
7527 | for (j = 0; j < ndoms_new; j++) { | |
1d3504fc HS |
7528 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7529 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7530 | goto match1; |
7531 | } | |
7532 | /* no match - a current sched domain not in new doms_new[] */ | |
7533 | detach_destroy_domains(doms_cur + i); | |
7534 | match1: | |
7535 | ; | |
7536 | } | |
7537 | ||
7538 | /* Build new domains */ | |
7539 | for (i = 0; i < ndoms_new; i++) { | |
7540 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7541 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7542 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7543 | goto match2; |
7544 | } | |
7545 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7546 | __build_sched_domains(doms_new + i, |
7547 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7548 | match2: |
7549 | ; | |
7550 | } | |
7551 | ||
7552 | /* Remember the new sched domains */ | |
7553 | if (doms_cur != &fallback_doms) | |
7554 | kfree(doms_cur); | |
1d3504fc | 7555 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7556 | doms_cur = doms_new; |
1d3504fc | 7557 | dattr_cur = dattr_new; |
029190c5 | 7558 | ndoms_cur = ndoms_new; |
7378547f MM |
7559 | |
7560 | register_sched_domain_sysctl(); | |
a1835615 | 7561 | |
712555ee | 7562 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7563 | } |
7564 | ||
5c45bf27 | 7565 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7566 | int arch_reinit_sched_domains(void) |
5c45bf27 SS |
7567 | { |
7568 | int err; | |
7569 | ||
95402b38 | 7570 | get_online_cpus(); |
712555ee | 7571 | mutex_lock(&sched_domains_mutex); |
5c45bf27 | 7572 | detach_destroy_domains(&cpu_online_map); |
5c8e1ed1 | 7573 | free_sched_domains(); |
5c45bf27 | 7574 | err = arch_init_sched_domains(&cpu_online_map); |
712555ee | 7575 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7576 | put_online_cpus(); |
5c45bf27 SS |
7577 | |
7578 | return err; | |
7579 | } | |
7580 | ||
7581 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7582 | { | |
7583 | int ret; | |
7584 | ||
7585 | if (buf[0] != '0' && buf[0] != '1') | |
7586 | return -EINVAL; | |
7587 | ||
7588 | if (smt) | |
7589 | sched_smt_power_savings = (buf[0] == '1'); | |
7590 | else | |
7591 | sched_mc_power_savings = (buf[0] == '1'); | |
7592 | ||
7593 | ret = arch_reinit_sched_domains(); | |
7594 | ||
7595 | return ret ? ret : count; | |
7596 | } | |
7597 | ||
5c45bf27 SS |
7598 | #ifdef CONFIG_SCHED_MC |
7599 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
7600 | { | |
7601 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7602 | } | |
48f24c4d IM |
7603 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
7604 | const char *buf, size_t count) | |
5c45bf27 SS |
7605 | { |
7606 | return sched_power_savings_store(buf, count, 0); | |
7607 | } | |
6707de00 AB |
7608 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7609 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7610 | #endif |
7611 | ||
7612 | #ifdef CONFIG_SCHED_SMT | |
7613 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
7614 | { | |
7615 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7616 | } | |
48f24c4d IM |
7617 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
7618 | const char *buf, size_t count) | |
5c45bf27 SS |
7619 | { |
7620 | return sched_power_savings_store(buf, count, 1); | |
7621 | } | |
6707de00 AB |
7622 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7623 | sched_smt_power_savings_store); | |
7624 | #endif | |
7625 | ||
7626 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7627 | { | |
7628 | int err = 0; | |
7629 | ||
7630 | #ifdef CONFIG_SCHED_SMT | |
7631 | if (smt_capable()) | |
7632 | err = sysfs_create_file(&cls->kset.kobj, | |
7633 | &attr_sched_smt_power_savings.attr); | |
7634 | #endif | |
7635 | #ifdef CONFIG_SCHED_MC | |
7636 | if (!err && mc_capable()) | |
7637 | err = sysfs_create_file(&cls->kset.kobj, | |
7638 | &attr_sched_mc_power_savings.attr); | |
7639 | #endif | |
7640 | return err; | |
7641 | } | |
6d6bc0ad | 7642 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7643 | |
1da177e4 | 7644 | /* |
41a2d6cf | 7645 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 7646 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 7647 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
7648 | * which will prevent rebalancing while the sched domains are recalculated. |
7649 | */ | |
7650 | static int update_sched_domains(struct notifier_block *nfb, | |
7651 | unsigned long action, void *hcpu) | |
7652 | { | |
7def2be1 PZ |
7653 | int cpu = (int)(long)hcpu; |
7654 | ||
1da177e4 | 7655 | switch (action) { |
1da177e4 | 7656 | case CPU_DOWN_PREPARE: |
8bb78442 | 7657 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 PZ |
7658 | disable_runtime(cpu_rq(cpu)); |
7659 | /* fall-through */ | |
7660 | case CPU_UP_PREPARE: | |
7661 | case CPU_UP_PREPARE_FROZEN: | |
1a20ff27 | 7662 | detach_destroy_domains(&cpu_online_map); |
5c8e1ed1 | 7663 | free_sched_domains(); |
1da177e4 LT |
7664 | return NOTIFY_OK; |
7665 | ||
7def2be1 | 7666 | |
1da177e4 | 7667 | case CPU_DOWN_FAILED: |
8bb78442 | 7668 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7669 | case CPU_ONLINE: |
8bb78442 | 7670 | case CPU_ONLINE_FROZEN: |
7def2be1 PZ |
7671 | enable_runtime(cpu_rq(cpu)); |
7672 | /* fall-through */ | |
7673 | case CPU_UP_CANCELED: | |
7674 | case CPU_UP_CANCELED_FROZEN: | |
1da177e4 | 7675 | case CPU_DEAD: |
8bb78442 | 7676 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
7677 | /* |
7678 | * Fall through and re-initialise the domains. | |
7679 | */ | |
7680 | break; | |
7681 | default: | |
7682 | return NOTIFY_DONE; | |
7683 | } | |
7684 | ||
5c8e1ed1 MK |
7685 | #ifndef CONFIG_CPUSETS |
7686 | /* | |
7687 | * Create default domain partitioning if cpusets are disabled. | |
7688 | * Otherwise we let cpusets rebuild the domains based on the | |
7689 | * current setup. | |
7690 | */ | |
7691 | ||
1da177e4 | 7692 | /* The hotplug lock is already held by cpu_up/cpu_down */ |
1a20ff27 | 7693 | arch_init_sched_domains(&cpu_online_map); |
5c8e1ed1 | 7694 | #endif |
1da177e4 LT |
7695 | |
7696 | return NOTIFY_OK; | |
7697 | } | |
1da177e4 LT |
7698 | |
7699 | void __init sched_init_smp(void) | |
7700 | { | |
5c1e1767 NP |
7701 | cpumask_t non_isolated_cpus; |
7702 | ||
434d53b0 MT |
7703 | #if defined(CONFIG_NUMA) |
7704 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7705 | GFP_KERNEL); | |
7706 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7707 | #endif | |
95402b38 | 7708 | get_online_cpus(); |
712555ee | 7709 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7710 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7711 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7712 | if (cpus_empty(non_isolated_cpus)) |
7713 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7714 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7715 | put_online_cpus(); |
1da177e4 LT |
7716 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7717 | hotcpu_notifier(update_sched_domains, 0); | |
b328ca18 | 7718 | init_hrtick(); |
5c1e1767 NP |
7719 | |
7720 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7721 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7722 | BUG(); |
19978ca6 | 7723 | sched_init_granularity(); |
1da177e4 LT |
7724 | } |
7725 | #else | |
7726 | void __init sched_init_smp(void) | |
7727 | { | |
19978ca6 | 7728 | sched_init_granularity(); |
1da177e4 LT |
7729 | } |
7730 | #endif /* CONFIG_SMP */ | |
7731 | ||
7732 | int in_sched_functions(unsigned long addr) | |
7733 | { | |
1da177e4 LT |
7734 | return in_lock_functions(addr) || |
7735 | (addr >= (unsigned long)__sched_text_start | |
7736 | && addr < (unsigned long)__sched_text_end); | |
7737 | } | |
7738 | ||
a9957449 | 7739 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7740 | { |
7741 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7742 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7743 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7744 | cfs_rq->rq = rq; | |
7745 | #endif | |
67e9fb2a | 7746 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7747 | } |
7748 | ||
fa85ae24 PZ |
7749 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7750 | { | |
7751 | struct rt_prio_array *array; | |
7752 | int i; | |
7753 | ||
7754 | array = &rt_rq->active; | |
7755 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
20b6331b | 7756 | INIT_LIST_HEAD(array->queue + i); |
fa85ae24 PZ |
7757 | __clear_bit(i, array->bitmap); |
7758 | } | |
7759 | /* delimiter for bitsearch: */ | |
7760 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7761 | ||
052f1dc7 | 7762 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7763 | rt_rq->highest_prio = MAX_RT_PRIO; |
7764 | #endif | |
fa85ae24 PZ |
7765 | #ifdef CONFIG_SMP |
7766 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7767 | rt_rq->overloaded = 0; |
7768 | #endif | |
7769 | ||
7770 | rt_rq->rt_time = 0; | |
7771 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7772 | rt_rq->rt_runtime = 0; |
7773 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7774 | |
052f1dc7 | 7775 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7776 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7777 | rt_rq->rq = rq; |
7778 | #endif | |
fa85ae24 PZ |
7779 | } |
7780 | ||
6f505b16 | 7781 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7782 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7783 | struct sched_entity *se, int cpu, int add, | |
7784 | struct sched_entity *parent) | |
6f505b16 | 7785 | { |
ec7dc8ac | 7786 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7787 | tg->cfs_rq[cpu] = cfs_rq; |
7788 | init_cfs_rq(cfs_rq, rq); | |
7789 | cfs_rq->tg = tg; | |
7790 | if (add) | |
7791 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7792 | ||
7793 | tg->se[cpu] = se; | |
354d60c2 DG |
7794 | /* se could be NULL for init_task_group */ |
7795 | if (!se) | |
7796 | return; | |
7797 | ||
ec7dc8ac DG |
7798 | if (!parent) |
7799 | se->cfs_rq = &rq->cfs; | |
7800 | else | |
7801 | se->cfs_rq = parent->my_q; | |
7802 | ||
6f505b16 PZ |
7803 | se->my_q = cfs_rq; |
7804 | se->load.weight = tg->shares; | |
e05510d0 | 7805 | se->load.inv_weight = 0; |
ec7dc8ac | 7806 | se->parent = parent; |
6f505b16 | 7807 | } |
052f1dc7 | 7808 | #endif |
6f505b16 | 7809 | |
052f1dc7 | 7810 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7811 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7812 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7813 | struct sched_rt_entity *parent) | |
6f505b16 | 7814 | { |
ec7dc8ac DG |
7815 | struct rq *rq = cpu_rq(cpu); |
7816 | ||
6f505b16 PZ |
7817 | tg->rt_rq[cpu] = rt_rq; |
7818 | init_rt_rq(rt_rq, rq); | |
7819 | rt_rq->tg = tg; | |
7820 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7821 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7822 | if (add) |
7823 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7824 | ||
7825 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7826 | if (!rt_se) |
7827 | return; | |
7828 | ||
ec7dc8ac DG |
7829 | if (!parent) |
7830 | rt_se->rt_rq = &rq->rt; | |
7831 | else | |
7832 | rt_se->rt_rq = parent->my_q; | |
7833 | ||
6f505b16 | 7834 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7835 | rt_se->parent = parent; |
6f505b16 PZ |
7836 | INIT_LIST_HEAD(&rt_se->run_list); |
7837 | } | |
7838 | #endif | |
7839 | ||
1da177e4 LT |
7840 | void __init sched_init(void) |
7841 | { | |
dd41f596 | 7842 | int i, j; |
434d53b0 MT |
7843 | unsigned long alloc_size = 0, ptr; |
7844 | ||
7845 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7846 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7847 | #endif | |
7848 | #ifdef CONFIG_RT_GROUP_SCHED | |
7849 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7850 | #endif |
7851 | #ifdef CONFIG_USER_SCHED | |
7852 | alloc_size *= 2; | |
434d53b0 MT |
7853 | #endif |
7854 | /* | |
7855 | * As sched_init() is called before page_alloc is setup, | |
7856 | * we use alloc_bootmem(). | |
7857 | */ | |
7858 | if (alloc_size) { | |
5a9d3225 | 7859 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
7860 | |
7861 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7862 | init_task_group.se = (struct sched_entity **)ptr; | |
7863 | ptr += nr_cpu_ids * sizeof(void **); | |
7864 | ||
7865 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7866 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7867 | |
7868 | #ifdef CONFIG_USER_SCHED | |
7869 | root_task_group.se = (struct sched_entity **)ptr; | |
7870 | ptr += nr_cpu_ids * sizeof(void **); | |
7871 | ||
7872 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7873 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7874 | #endif /* CONFIG_USER_SCHED */ |
7875 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
7876 | #ifdef CONFIG_RT_GROUP_SCHED |
7877 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7878 | ptr += nr_cpu_ids * sizeof(void **); | |
7879 | ||
7880 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7881 | ptr += nr_cpu_ids * sizeof(void **); |
7882 | ||
7883 | #ifdef CONFIG_USER_SCHED | |
7884 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7885 | ptr += nr_cpu_ids * sizeof(void **); | |
7886 | ||
7887 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
7888 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
7889 | #endif /* CONFIG_USER_SCHED */ |
7890 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 7891 | } |
dd41f596 | 7892 | |
57d885fe GH |
7893 | #ifdef CONFIG_SMP |
7894 | init_defrootdomain(); | |
7895 | #endif | |
7896 | ||
d0b27fa7 PZ |
7897 | init_rt_bandwidth(&def_rt_bandwidth, |
7898 | global_rt_period(), global_rt_runtime()); | |
7899 | ||
7900 | #ifdef CONFIG_RT_GROUP_SCHED | |
7901 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7902 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
7903 | #ifdef CONFIG_USER_SCHED |
7904 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
7905 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
7906 | #endif /* CONFIG_USER_SCHED */ |
7907 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 7908 | |
052f1dc7 | 7909 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 7910 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7911 | INIT_LIST_HEAD(&init_task_group.children); |
7912 | ||
7913 | #ifdef CONFIG_USER_SCHED | |
7914 | INIT_LIST_HEAD(&root_task_group.children); | |
7915 | init_task_group.parent = &root_task_group; | |
7916 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
7917 | #endif /* CONFIG_USER_SCHED */ |
7918 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 7919 | |
0a945022 | 7920 | for_each_possible_cpu(i) { |
70b97a7f | 7921 | struct rq *rq; |
1da177e4 LT |
7922 | |
7923 | rq = cpu_rq(i); | |
7924 | spin_lock_init(&rq->lock); | |
fcb99371 | 7925 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7926 | rq->nr_running = 0; |
dd41f596 | 7927 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7928 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7929 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7930 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7931 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7932 | #ifdef CONFIG_CGROUP_SCHED |
7933 | /* | |
7934 | * How much cpu bandwidth does init_task_group get? | |
7935 | * | |
7936 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7937 | * gets 100% of the cpu resources in the system. This overall | |
7938 | * system cpu resource is divided among the tasks of | |
7939 | * init_task_group and its child task-groups in a fair manner, | |
7940 | * based on each entity's (task or task-group's) weight | |
7941 | * (se->load.weight). | |
7942 | * | |
7943 | * In other words, if init_task_group has 10 tasks of weight | |
7944 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7945 | * then A0's share of the cpu resource is: | |
7946 | * | |
7947 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
7948 | * | |
7949 | * We achieve this by letting init_task_group's tasks sit | |
7950 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7951 | */ | |
ec7dc8ac | 7952 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 7953 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
7954 | root_task_group.shares = NICE_0_LOAD; |
7955 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
7956 | /* |
7957 | * In case of task-groups formed thr' the user id of tasks, | |
7958 | * init_task_group represents tasks belonging to root user. | |
7959 | * Hence it forms a sibling of all subsequent groups formed. | |
7960 | * In this case, init_task_group gets only a fraction of overall | |
7961 | * system cpu resource, based on the weight assigned to root | |
7962 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
7963 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
7964 | * (init_cfs_rq) and having one entity represent this group of | |
7965 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
7966 | */ | |
ec7dc8ac | 7967 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 7968 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
7969 | &per_cpu(init_sched_entity, i), i, 1, |
7970 | root_task_group.se[i]); | |
6f505b16 | 7971 | |
052f1dc7 | 7972 | #endif |
354d60c2 DG |
7973 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7974 | ||
7975 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7976 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7977 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7978 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7979 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7980 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 7981 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 7982 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 7983 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
7984 | &per_cpu(init_sched_rt_entity, i), i, 1, |
7985 | root_task_group.rt_se[i]); | |
354d60c2 | 7986 | #endif |
dd41f596 | 7987 | #endif |
1da177e4 | 7988 | |
dd41f596 IM |
7989 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7990 | rq->cpu_load[j] = 0; | |
1da177e4 | 7991 | #ifdef CONFIG_SMP |
41c7ce9a | 7992 | rq->sd = NULL; |
57d885fe | 7993 | rq->rd = NULL; |
1da177e4 | 7994 | rq->active_balance = 0; |
dd41f596 | 7995 | rq->next_balance = jiffies; |
1da177e4 | 7996 | rq->push_cpu = 0; |
0a2966b4 | 7997 | rq->cpu = i; |
1f11eb6a | 7998 | rq->online = 0; |
1da177e4 LT |
7999 | rq->migration_thread = NULL; |
8000 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8001 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8002 | #endif |
8f4d37ec | 8003 | init_rq_hrtick(rq); |
1da177e4 | 8004 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8005 | } |
8006 | ||
2dd73a4f | 8007 | set_load_weight(&init_task); |
b50f60ce | 8008 | |
e107be36 AK |
8009 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8010 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8011 | #endif | |
8012 | ||
c9819f45 CL |
8013 | #ifdef CONFIG_SMP |
8014 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); | |
8015 | #endif | |
8016 | ||
b50f60ce HC |
8017 | #ifdef CONFIG_RT_MUTEXES |
8018 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8019 | #endif | |
8020 | ||
1da177e4 LT |
8021 | /* |
8022 | * The boot idle thread does lazy MMU switching as well: | |
8023 | */ | |
8024 | atomic_inc(&init_mm.mm_count); | |
8025 | enter_lazy_tlb(&init_mm, current); | |
8026 | ||
8027 | /* | |
8028 | * Make us the idle thread. Technically, schedule() should not be | |
8029 | * called from this thread, however somewhere below it might be, | |
8030 | * but because we are the idle thread, we just pick up running again | |
8031 | * when this runqueue becomes "idle". | |
8032 | */ | |
8033 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8034 | /* |
8035 | * During early bootup we pretend to be a normal task: | |
8036 | */ | |
8037 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
8038 | |
8039 | scheduler_running = 1; | |
1da177e4 LT |
8040 | } |
8041 | ||
8042 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8043 | void __might_sleep(char *file, int line) | |
8044 | { | |
48f24c4d | 8045 | #ifdef in_atomic |
1da177e4 LT |
8046 | static unsigned long prev_jiffy; /* ratelimiting */ |
8047 | ||
8048 | if ((in_atomic() || irqs_disabled()) && | |
8049 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
8050 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8051 | return; | |
8052 | prev_jiffy = jiffies; | |
91368d73 | 8053 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
8054 | " context at %s:%d\n", file, line); |
8055 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
8056 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 8057 | debug_show_held_locks(current); |
3117df04 IM |
8058 | if (irqs_disabled()) |
8059 | print_irqtrace_events(current); | |
1da177e4 LT |
8060 | dump_stack(); |
8061 | } | |
8062 | #endif | |
8063 | } | |
8064 | EXPORT_SYMBOL(__might_sleep); | |
8065 | #endif | |
8066 | ||
8067 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8068 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8069 | { | |
8070 | int on_rq; | |
3e51f33f | 8071 | |
3a5e4dc1 AK |
8072 | update_rq_clock(rq); |
8073 | on_rq = p->se.on_rq; | |
8074 | if (on_rq) | |
8075 | deactivate_task(rq, p, 0); | |
8076 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8077 | if (on_rq) { | |
8078 | activate_task(rq, p, 0); | |
8079 | resched_task(rq->curr); | |
8080 | } | |
8081 | } | |
8082 | ||
1da177e4 LT |
8083 | void normalize_rt_tasks(void) |
8084 | { | |
a0f98a1c | 8085 | struct task_struct *g, *p; |
1da177e4 | 8086 | unsigned long flags; |
70b97a7f | 8087 | struct rq *rq; |
1da177e4 | 8088 | |
4cf5d77a | 8089 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8090 | do_each_thread(g, p) { |
178be793 IM |
8091 | /* |
8092 | * Only normalize user tasks: | |
8093 | */ | |
8094 | if (!p->mm) | |
8095 | continue; | |
8096 | ||
6cfb0d5d | 8097 | p->se.exec_start = 0; |
6cfb0d5d | 8098 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8099 | p->se.wait_start = 0; |
dd41f596 | 8100 | p->se.sleep_start = 0; |
dd41f596 | 8101 | p->se.block_start = 0; |
6cfb0d5d | 8102 | #endif |
dd41f596 IM |
8103 | |
8104 | if (!rt_task(p)) { | |
8105 | /* | |
8106 | * Renice negative nice level userspace | |
8107 | * tasks back to 0: | |
8108 | */ | |
8109 | if (TASK_NICE(p) < 0 && p->mm) | |
8110 | set_user_nice(p, 0); | |
1da177e4 | 8111 | continue; |
dd41f596 | 8112 | } |
1da177e4 | 8113 | |
4cf5d77a | 8114 | spin_lock(&p->pi_lock); |
b29739f9 | 8115 | rq = __task_rq_lock(p); |
1da177e4 | 8116 | |
178be793 | 8117 | normalize_task(rq, p); |
3a5e4dc1 | 8118 | |
b29739f9 | 8119 | __task_rq_unlock(rq); |
4cf5d77a | 8120 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8121 | } while_each_thread(g, p); |
8122 | ||
4cf5d77a | 8123 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8124 | } |
8125 | ||
8126 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8127 | |
8128 | #ifdef CONFIG_IA64 | |
8129 | /* | |
8130 | * These functions are only useful for the IA64 MCA handling. | |
8131 | * | |
8132 | * They can only be called when the whole system has been | |
8133 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8134 | * activity can take place. Using them for anything else would | |
8135 | * be a serious bug, and as a result, they aren't even visible | |
8136 | * under any other configuration. | |
8137 | */ | |
8138 | ||
8139 | /** | |
8140 | * curr_task - return the current task for a given cpu. | |
8141 | * @cpu: the processor in question. | |
8142 | * | |
8143 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8144 | */ | |
36c8b586 | 8145 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8146 | { |
8147 | return cpu_curr(cpu); | |
8148 | } | |
8149 | ||
8150 | /** | |
8151 | * set_curr_task - set the current task for a given cpu. | |
8152 | * @cpu: the processor in question. | |
8153 | * @p: the task pointer to set. | |
8154 | * | |
8155 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8156 | * are serviced on a separate stack. It allows the architecture to switch the |
8157 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8158 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8159 | * and caller must save the original value of the current task (see | |
8160 | * curr_task() above) and restore that value before reenabling interrupts and | |
8161 | * re-starting the system. | |
8162 | * | |
8163 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8164 | */ | |
36c8b586 | 8165 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8166 | { |
8167 | cpu_curr(cpu) = p; | |
8168 | } | |
8169 | ||
8170 | #endif | |
29f59db3 | 8171 | |
bccbe08a PZ |
8172 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8173 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8174 | { |
8175 | int i; | |
8176 | ||
8177 | for_each_possible_cpu(i) { | |
8178 | if (tg->cfs_rq) | |
8179 | kfree(tg->cfs_rq[i]); | |
8180 | if (tg->se) | |
8181 | kfree(tg->se[i]); | |
6f505b16 PZ |
8182 | } |
8183 | ||
8184 | kfree(tg->cfs_rq); | |
8185 | kfree(tg->se); | |
6f505b16 PZ |
8186 | } |
8187 | ||
ec7dc8ac DG |
8188 | static |
8189 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8190 | { |
29f59db3 | 8191 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8192 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8193 | struct rq *rq; |
29f59db3 SV |
8194 | int i; |
8195 | ||
434d53b0 | 8196 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8197 | if (!tg->cfs_rq) |
8198 | goto err; | |
434d53b0 | 8199 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8200 | if (!tg->se) |
8201 | goto err; | |
052f1dc7 PZ |
8202 | |
8203 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8204 | |
8205 | for_each_possible_cpu(i) { | |
9b5b7751 | 8206 | rq = cpu_rq(i); |
29f59db3 | 8207 | |
6f505b16 PZ |
8208 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8209 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8210 | if (!cfs_rq) |
8211 | goto err; | |
8212 | ||
6f505b16 PZ |
8213 | se = kmalloc_node(sizeof(struct sched_entity), |
8214 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8215 | if (!se) |
8216 | goto err; | |
8217 | ||
ec7dc8ac DG |
8218 | parent_se = parent ? parent->se[i] : NULL; |
8219 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8220 | } |
8221 | ||
8222 | return 1; | |
8223 | ||
8224 | err: | |
8225 | return 0; | |
8226 | } | |
8227 | ||
8228 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8229 | { | |
8230 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8231 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8232 | } | |
8233 | ||
8234 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8235 | { | |
8236 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8237 | } | |
6d6bc0ad | 8238 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8239 | static inline void free_fair_sched_group(struct task_group *tg) |
8240 | { | |
8241 | } | |
8242 | ||
ec7dc8ac DG |
8243 | static inline |
8244 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8245 | { |
8246 | return 1; | |
8247 | } | |
8248 | ||
8249 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8250 | { | |
8251 | } | |
8252 | ||
8253 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8254 | { | |
8255 | } | |
6d6bc0ad | 8256 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8257 | |
8258 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8259 | static void free_rt_sched_group(struct task_group *tg) |
8260 | { | |
8261 | int i; | |
8262 | ||
d0b27fa7 PZ |
8263 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8264 | ||
bccbe08a PZ |
8265 | for_each_possible_cpu(i) { |
8266 | if (tg->rt_rq) | |
8267 | kfree(tg->rt_rq[i]); | |
8268 | if (tg->rt_se) | |
8269 | kfree(tg->rt_se[i]); | |
8270 | } | |
8271 | ||
8272 | kfree(tg->rt_rq); | |
8273 | kfree(tg->rt_se); | |
8274 | } | |
8275 | ||
ec7dc8ac DG |
8276 | static |
8277 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8278 | { |
8279 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8280 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8281 | struct rq *rq; |
8282 | int i; | |
8283 | ||
434d53b0 | 8284 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8285 | if (!tg->rt_rq) |
8286 | goto err; | |
434d53b0 | 8287 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8288 | if (!tg->rt_se) |
8289 | goto err; | |
8290 | ||
d0b27fa7 PZ |
8291 | init_rt_bandwidth(&tg->rt_bandwidth, |
8292 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8293 | |
8294 | for_each_possible_cpu(i) { | |
8295 | rq = cpu_rq(i); | |
8296 | ||
6f505b16 PZ |
8297 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8298 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8299 | if (!rt_rq) | |
8300 | goto err; | |
29f59db3 | 8301 | |
6f505b16 PZ |
8302 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8303 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8304 | if (!rt_se) | |
8305 | goto err; | |
29f59db3 | 8306 | |
ec7dc8ac DG |
8307 | parent_se = parent ? parent->rt_se[i] : NULL; |
8308 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8309 | } |
8310 | ||
bccbe08a PZ |
8311 | return 1; |
8312 | ||
8313 | err: | |
8314 | return 0; | |
8315 | } | |
8316 | ||
8317 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8318 | { | |
8319 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8320 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8321 | } | |
8322 | ||
8323 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8324 | { | |
8325 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8326 | } | |
6d6bc0ad | 8327 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8328 | static inline void free_rt_sched_group(struct task_group *tg) |
8329 | { | |
8330 | } | |
8331 | ||
ec7dc8ac DG |
8332 | static inline |
8333 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8334 | { |
8335 | return 1; | |
8336 | } | |
8337 | ||
8338 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8339 | { | |
8340 | } | |
8341 | ||
8342 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8343 | { | |
8344 | } | |
6d6bc0ad | 8345 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8346 | |
d0b27fa7 | 8347 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8348 | static void free_sched_group(struct task_group *tg) |
8349 | { | |
8350 | free_fair_sched_group(tg); | |
8351 | free_rt_sched_group(tg); | |
8352 | kfree(tg); | |
8353 | } | |
8354 | ||
8355 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8356 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8357 | { |
8358 | struct task_group *tg; | |
8359 | unsigned long flags; | |
8360 | int i; | |
8361 | ||
8362 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8363 | if (!tg) | |
8364 | return ERR_PTR(-ENOMEM); | |
8365 | ||
ec7dc8ac | 8366 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8367 | goto err; |
8368 | ||
ec7dc8ac | 8369 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8370 | goto err; |
8371 | ||
8ed36996 | 8372 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8373 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8374 | register_fair_sched_group(tg, i); |
8375 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8376 | } |
6f505b16 | 8377 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8378 | |
8379 | WARN_ON(!parent); /* root should already exist */ | |
8380 | ||
8381 | tg->parent = parent; | |
8382 | list_add_rcu(&tg->siblings, &parent->children); | |
8383 | INIT_LIST_HEAD(&tg->children); | |
8ed36996 | 8384 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8385 | |
9b5b7751 | 8386 | return tg; |
29f59db3 SV |
8387 | |
8388 | err: | |
6f505b16 | 8389 | free_sched_group(tg); |
29f59db3 SV |
8390 | return ERR_PTR(-ENOMEM); |
8391 | } | |
8392 | ||
9b5b7751 | 8393 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8394 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8395 | { |
29f59db3 | 8396 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8397 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8398 | } |
8399 | ||
9b5b7751 | 8400 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8401 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8402 | { |
8ed36996 | 8403 | unsigned long flags; |
9b5b7751 | 8404 | int i; |
29f59db3 | 8405 | |
8ed36996 | 8406 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8407 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8408 | unregister_fair_sched_group(tg, i); |
8409 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8410 | } |
6f505b16 | 8411 | list_del_rcu(&tg->list); |
f473aa5e | 8412 | list_del_rcu(&tg->siblings); |
8ed36996 | 8413 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8414 | |
9b5b7751 | 8415 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8416 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8417 | } |
8418 | ||
9b5b7751 | 8419 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8420 | * The caller of this function should have put the task in its new group |
8421 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8422 | * reflect its new group. | |
9b5b7751 SV |
8423 | */ |
8424 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8425 | { |
8426 | int on_rq, running; | |
8427 | unsigned long flags; | |
8428 | struct rq *rq; | |
8429 | ||
8430 | rq = task_rq_lock(tsk, &flags); | |
8431 | ||
29f59db3 SV |
8432 | update_rq_clock(rq); |
8433 | ||
051a1d1a | 8434 | running = task_current(rq, tsk); |
29f59db3 SV |
8435 | on_rq = tsk->se.on_rq; |
8436 | ||
0e1f3483 | 8437 | if (on_rq) |
29f59db3 | 8438 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8439 | if (unlikely(running)) |
8440 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8441 | |
6f505b16 | 8442 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8443 | |
810b3817 PZ |
8444 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8445 | if (tsk->sched_class->moved_group) | |
8446 | tsk->sched_class->moved_group(tsk); | |
8447 | #endif | |
8448 | ||
0e1f3483 HS |
8449 | if (unlikely(running)) |
8450 | tsk->sched_class->set_curr_task(rq); | |
8451 | if (on_rq) | |
7074badb | 8452 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8453 | |
29f59db3 SV |
8454 | task_rq_unlock(rq, &flags); |
8455 | } | |
6d6bc0ad | 8456 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8457 | |
052f1dc7 | 8458 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8459 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8460 | { |
8461 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8462 | int on_rq; |
8463 | ||
29f59db3 | 8464 | on_rq = se->on_rq; |
62fb1851 | 8465 | if (on_rq) |
29f59db3 SV |
8466 | dequeue_entity(cfs_rq, se, 0); |
8467 | ||
8468 | se->load.weight = shares; | |
e05510d0 | 8469 | se->load.inv_weight = 0; |
29f59db3 | 8470 | |
62fb1851 | 8471 | if (on_rq) |
29f59db3 | 8472 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8473 | } |
62fb1851 | 8474 | |
c09595f6 PZ |
8475 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8476 | { | |
8477 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8478 | struct rq *rq = cfs_rq->rq; | |
8479 | unsigned long flags; | |
8480 | ||
8481 | spin_lock_irqsave(&rq->lock, flags); | |
8482 | __set_se_shares(se, shares); | |
8483 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8484 | } |
8485 | ||
8ed36996 PZ |
8486 | static DEFINE_MUTEX(shares_mutex); |
8487 | ||
4cf86d77 | 8488 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8489 | { |
8490 | int i; | |
8ed36996 | 8491 | unsigned long flags; |
c61935fd | 8492 | |
ec7dc8ac DG |
8493 | /* |
8494 | * We can't change the weight of the root cgroup. | |
8495 | */ | |
8496 | if (!tg->se[0]) | |
8497 | return -EINVAL; | |
8498 | ||
18d95a28 PZ |
8499 | if (shares < MIN_SHARES) |
8500 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8501 | else if (shares > MAX_SHARES) |
8502 | shares = MAX_SHARES; | |
62fb1851 | 8503 | |
8ed36996 | 8504 | mutex_lock(&shares_mutex); |
9b5b7751 | 8505 | if (tg->shares == shares) |
5cb350ba | 8506 | goto done; |
29f59db3 | 8507 | |
8ed36996 | 8508 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8509 | for_each_possible_cpu(i) |
8510 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8511 | list_del_rcu(&tg->siblings); |
8ed36996 | 8512 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8513 | |
8514 | /* wait for any ongoing reference to this group to finish */ | |
8515 | synchronize_sched(); | |
8516 | ||
8517 | /* | |
8518 | * Now we are free to modify the group's share on each cpu | |
8519 | * w/o tripping rebalance_share or load_balance_fair. | |
8520 | */ | |
9b5b7751 | 8521 | tg->shares = shares; |
c09595f6 PZ |
8522 | for_each_possible_cpu(i) { |
8523 | /* | |
8524 | * force a rebalance | |
8525 | */ | |
8526 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8527 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8528 | } |
29f59db3 | 8529 | |
6b2d7700 SV |
8530 | /* |
8531 | * Enable load balance activity on this group, by inserting it back on | |
8532 | * each cpu's rq->leaf_cfs_rq_list. | |
8533 | */ | |
8ed36996 | 8534 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8535 | for_each_possible_cpu(i) |
8536 | register_fair_sched_group(tg, i); | |
f473aa5e | 8537 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8538 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8539 | done: |
8ed36996 | 8540 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8541 | return 0; |
29f59db3 SV |
8542 | } |
8543 | ||
5cb350ba DG |
8544 | unsigned long sched_group_shares(struct task_group *tg) |
8545 | { | |
8546 | return tg->shares; | |
8547 | } | |
052f1dc7 | 8548 | #endif |
5cb350ba | 8549 | |
052f1dc7 | 8550 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8551 | /* |
9f0c1e56 | 8552 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8553 | */ |
9f0c1e56 PZ |
8554 | static DEFINE_MUTEX(rt_constraints_mutex); |
8555 | ||
8556 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8557 | { | |
8558 | if (runtime == RUNTIME_INF) | |
8559 | return 1ULL << 16; | |
8560 | ||
6f6d6a1a | 8561 | return div64_u64(runtime << 16, period); |
9f0c1e56 PZ |
8562 | } |
8563 | ||
b40b2e8e PZ |
8564 | #ifdef CONFIG_CGROUP_SCHED |
8565 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8566 | { | |
10b612f4 | 8567 | struct task_group *tgi, *parent = tg->parent; |
b40b2e8e PZ |
8568 | unsigned long total = 0; |
8569 | ||
8570 | if (!parent) { | |
8571 | if (global_rt_period() < period) | |
8572 | return 0; | |
8573 | ||
8574 | return to_ratio(period, runtime) < | |
8575 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8576 | } | |
8577 | ||
8578 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8579 | return 0; | |
8580 | ||
8581 | rcu_read_lock(); | |
8582 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8583 | if (tgi == tg) | |
8584 | continue; | |
8585 | ||
8586 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8587 | tgi->rt_bandwidth.rt_runtime); | |
8588 | } | |
8589 | rcu_read_unlock(); | |
8590 | ||
10b612f4 | 8591 | return total + to_ratio(period, runtime) <= |
b40b2e8e PZ |
8592 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), |
8593 | parent->rt_bandwidth.rt_runtime); | |
8594 | } | |
8595 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8596 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8597 | { |
8598 | struct task_group *tgi; | |
8599 | unsigned long total = 0; | |
9f0c1e56 | 8600 | unsigned long global_ratio = |
d0b27fa7 | 8601 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8602 | |
8603 | rcu_read_lock(); | |
9f0c1e56 PZ |
8604 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8605 | if (tgi == tg) | |
8606 | continue; | |
6f505b16 | 8607 | |
d0b27fa7 PZ |
8608 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8609 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8610 | } |
8611 | rcu_read_unlock(); | |
6f505b16 | 8612 | |
9f0c1e56 | 8613 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8614 | } |
b40b2e8e | 8615 | #endif |
6f505b16 | 8616 | |
521f1a24 DG |
8617 | /* Must be called with tasklist_lock held */ |
8618 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8619 | { | |
8620 | struct task_struct *g, *p; | |
8621 | do_each_thread(g, p) { | |
8622 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8623 | return 1; | |
8624 | } while_each_thread(g, p); | |
8625 | return 0; | |
8626 | } | |
8627 | ||
d0b27fa7 PZ |
8628 | static int tg_set_bandwidth(struct task_group *tg, |
8629 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8630 | { |
ac086bc2 | 8631 | int i, err = 0; |
9f0c1e56 | 8632 | |
9f0c1e56 | 8633 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8634 | read_lock(&tasklist_lock); |
ac086bc2 | 8635 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8636 | err = -EBUSY; |
8637 | goto unlock; | |
8638 | } | |
9f0c1e56 PZ |
8639 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8640 | err = -EINVAL; | |
8641 | goto unlock; | |
8642 | } | |
ac086bc2 PZ |
8643 | |
8644 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8645 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8646 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8647 | |
8648 | for_each_possible_cpu(i) { | |
8649 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8650 | ||
8651 | spin_lock(&rt_rq->rt_runtime_lock); | |
8652 | rt_rq->rt_runtime = rt_runtime; | |
8653 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8654 | } | |
8655 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8656 | unlock: |
521f1a24 | 8657 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8658 | mutex_unlock(&rt_constraints_mutex); |
8659 | ||
8660 | return err; | |
6f505b16 PZ |
8661 | } |
8662 | ||
d0b27fa7 PZ |
8663 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8664 | { | |
8665 | u64 rt_runtime, rt_period; | |
8666 | ||
8667 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8668 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8669 | if (rt_runtime_us < 0) | |
8670 | rt_runtime = RUNTIME_INF; | |
8671 | ||
8672 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8673 | } | |
8674 | ||
9f0c1e56 PZ |
8675 | long sched_group_rt_runtime(struct task_group *tg) |
8676 | { | |
8677 | u64 rt_runtime_us; | |
8678 | ||
d0b27fa7 | 8679 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8680 | return -1; |
8681 | ||
d0b27fa7 | 8682 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8683 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8684 | return rt_runtime_us; | |
8685 | } | |
d0b27fa7 PZ |
8686 | |
8687 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8688 | { | |
8689 | u64 rt_runtime, rt_period; | |
8690 | ||
8691 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8692 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8693 | ||
8694 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8695 | } | |
8696 | ||
8697 | long sched_group_rt_period(struct task_group *tg) | |
8698 | { | |
8699 | u64 rt_period_us; | |
8700 | ||
8701 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8702 | do_div(rt_period_us, NSEC_PER_USEC); | |
8703 | return rt_period_us; | |
8704 | } | |
8705 | ||
8706 | static int sched_rt_global_constraints(void) | |
8707 | { | |
10b612f4 PZ |
8708 | struct task_group *tg = &root_task_group; |
8709 | u64 rt_runtime, rt_period; | |
d0b27fa7 PZ |
8710 | int ret = 0; |
8711 | ||
10b612f4 PZ |
8712 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8713 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8714 | ||
d0b27fa7 | 8715 | mutex_lock(&rt_constraints_mutex); |
10b612f4 | 8716 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) |
d0b27fa7 PZ |
8717 | ret = -EINVAL; |
8718 | mutex_unlock(&rt_constraints_mutex); | |
8719 | ||
8720 | return ret; | |
8721 | } | |
6d6bc0ad | 8722 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8723 | static int sched_rt_global_constraints(void) |
8724 | { | |
ac086bc2 PZ |
8725 | unsigned long flags; |
8726 | int i; | |
8727 | ||
8728 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8729 | for_each_possible_cpu(i) { | |
8730 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8731 | ||
8732 | spin_lock(&rt_rq->rt_runtime_lock); | |
8733 | rt_rq->rt_runtime = global_rt_runtime(); | |
8734 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8735 | } | |
8736 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8737 | ||
d0b27fa7 PZ |
8738 | return 0; |
8739 | } | |
6d6bc0ad | 8740 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8741 | |
8742 | int sched_rt_handler(struct ctl_table *table, int write, | |
8743 | struct file *filp, void __user *buffer, size_t *lenp, | |
8744 | loff_t *ppos) | |
8745 | { | |
8746 | int ret; | |
8747 | int old_period, old_runtime; | |
8748 | static DEFINE_MUTEX(mutex); | |
8749 | ||
8750 | mutex_lock(&mutex); | |
8751 | old_period = sysctl_sched_rt_period; | |
8752 | old_runtime = sysctl_sched_rt_runtime; | |
8753 | ||
8754 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8755 | ||
8756 | if (!ret && write) { | |
8757 | ret = sched_rt_global_constraints(); | |
8758 | if (ret) { | |
8759 | sysctl_sched_rt_period = old_period; | |
8760 | sysctl_sched_rt_runtime = old_runtime; | |
8761 | } else { | |
8762 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8763 | def_rt_bandwidth.rt_period = | |
8764 | ns_to_ktime(global_rt_period()); | |
8765 | } | |
8766 | } | |
8767 | mutex_unlock(&mutex); | |
8768 | ||
8769 | return ret; | |
8770 | } | |
68318b8e | 8771 | |
052f1dc7 | 8772 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8773 | |
8774 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8775 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8776 | { |
2b01dfe3 PM |
8777 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8778 | struct task_group, css); | |
68318b8e SV |
8779 | } |
8780 | ||
8781 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8782 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8783 | { |
ec7dc8ac | 8784 | struct task_group *tg, *parent; |
68318b8e | 8785 | |
2b01dfe3 | 8786 | if (!cgrp->parent) { |
68318b8e | 8787 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8788 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8789 | return &init_task_group.css; |
8790 | } | |
8791 | ||
ec7dc8ac DG |
8792 | parent = cgroup_tg(cgrp->parent); |
8793 | tg = sched_create_group(parent); | |
68318b8e SV |
8794 | if (IS_ERR(tg)) |
8795 | return ERR_PTR(-ENOMEM); | |
8796 | ||
8797 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8798 | tg->css.cgroup = cgrp; |
68318b8e SV |
8799 | |
8800 | return &tg->css; | |
8801 | } | |
8802 | ||
41a2d6cf IM |
8803 | static void |
8804 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8805 | { |
2b01dfe3 | 8806 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8807 | |
8808 | sched_destroy_group(tg); | |
8809 | } | |
8810 | ||
41a2d6cf IM |
8811 | static int |
8812 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8813 | struct task_struct *tsk) | |
68318b8e | 8814 | { |
b68aa230 PZ |
8815 | #ifdef CONFIG_RT_GROUP_SCHED |
8816 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8817 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8818 | return -EINVAL; |
8819 | #else | |
68318b8e SV |
8820 | /* We don't support RT-tasks being in separate groups */ |
8821 | if (tsk->sched_class != &fair_sched_class) | |
8822 | return -EINVAL; | |
b68aa230 | 8823 | #endif |
68318b8e SV |
8824 | |
8825 | return 0; | |
8826 | } | |
8827 | ||
8828 | static void | |
2b01dfe3 | 8829 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8830 | struct cgroup *old_cont, struct task_struct *tsk) |
8831 | { | |
8832 | sched_move_task(tsk); | |
8833 | } | |
8834 | ||
052f1dc7 | 8835 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8836 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8837 | u64 shareval) |
68318b8e | 8838 | { |
2b01dfe3 | 8839 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8840 | } |
8841 | ||
f4c753b7 | 8842 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8843 | { |
2b01dfe3 | 8844 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8845 | |
8846 | return (u64) tg->shares; | |
8847 | } | |
6d6bc0ad | 8848 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8849 | |
052f1dc7 | 8850 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8851 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8852 | s64 val) |
6f505b16 | 8853 | { |
06ecb27c | 8854 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8855 | } |
8856 | ||
06ecb27c | 8857 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8858 | { |
06ecb27c | 8859 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8860 | } |
d0b27fa7 PZ |
8861 | |
8862 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8863 | u64 rt_period_us) | |
8864 | { | |
8865 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8866 | } | |
8867 | ||
8868 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8869 | { | |
8870 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8871 | } | |
6d6bc0ad | 8872 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8873 | |
fe5c7cc2 | 8874 | static struct cftype cpu_files[] = { |
052f1dc7 | 8875 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8876 | { |
8877 | .name = "shares", | |
f4c753b7 PM |
8878 | .read_u64 = cpu_shares_read_u64, |
8879 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8880 | }, |
052f1dc7 PZ |
8881 | #endif |
8882 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8883 | { |
9f0c1e56 | 8884 | .name = "rt_runtime_us", |
06ecb27c PM |
8885 | .read_s64 = cpu_rt_runtime_read, |
8886 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8887 | }, |
d0b27fa7 PZ |
8888 | { |
8889 | .name = "rt_period_us", | |
f4c753b7 PM |
8890 | .read_u64 = cpu_rt_period_read_uint, |
8891 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8892 | }, |
052f1dc7 | 8893 | #endif |
68318b8e SV |
8894 | }; |
8895 | ||
8896 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8897 | { | |
fe5c7cc2 | 8898 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8899 | } |
8900 | ||
8901 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8902 | .name = "cpu", |
8903 | .create = cpu_cgroup_create, | |
8904 | .destroy = cpu_cgroup_destroy, | |
8905 | .can_attach = cpu_cgroup_can_attach, | |
8906 | .attach = cpu_cgroup_attach, | |
8907 | .populate = cpu_cgroup_populate, | |
8908 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8909 | .early_init = 1, |
8910 | }; | |
8911 | ||
052f1dc7 | 8912 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8913 | |
8914 | #ifdef CONFIG_CGROUP_CPUACCT | |
8915 | ||
8916 | /* | |
8917 | * CPU accounting code for task groups. | |
8918 | * | |
8919 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8920 | * (balbir@in.ibm.com). | |
8921 | */ | |
8922 | ||
8923 | /* track cpu usage of a group of tasks */ | |
8924 | struct cpuacct { | |
8925 | struct cgroup_subsys_state css; | |
8926 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8927 | u64 *cpuusage; | |
8928 | }; | |
8929 | ||
8930 | struct cgroup_subsys cpuacct_subsys; | |
8931 | ||
8932 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8933 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8934 | { |
32cd756a | 8935 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8936 | struct cpuacct, css); |
8937 | } | |
8938 | ||
8939 | /* return cpu accounting group to which this task belongs */ | |
8940 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8941 | { | |
8942 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8943 | struct cpuacct, css); | |
8944 | } | |
8945 | ||
8946 | /* create a new cpu accounting group */ | |
8947 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8948 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8949 | { |
8950 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
8951 | ||
8952 | if (!ca) | |
8953 | return ERR_PTR(-ENOMEM); | |
8954 | ||
8955 | ca->cpuusage = alloc_percpu(u64); | |
8956 | if (!ca->cpuusage) { | |
8957 | kfree(ca); | |
8958 | return ERR_PTR(-ENOMEM); | |
8959 | } | |
8960 | ||
8961 | return &ca->css; | |
8962 | } | |
8963 | ||
8964 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8965 | static void |
32cd756a | 8966 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8967 | { |
32cd756a | 8968 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8969 | |
8970 | free_percpu(ca->cpuusage); | |
8971 | kfree(ca); | |
8972 | } | |
8973 | ||
8974 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 8975 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8976 | { |
32cd756a | 8977 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8978 | u64 totalcpuusage = 0; |
8979 | int i; | |
8980 | ||
8981 | for_each_possible_cpu(i) { | |
8982 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8983 | ||
8984 | /* | |
8985 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
8986 | * platforms. | |
8987 | */ | |
8988 | spin_lock_irq(&cpu_rq(i)->lock); | |
8989 | totalcpuusage += *cpuusage; | |
8990 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8991 | } | |
8992 | ||
8993 | return totalcpuusage; | |
8994 | } | |
8995 | ||
0297b803 DG |
8996 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8997 | u64 reset) | |
8998 | { | |
8999 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9000 | int err = 0; | |
9001 | int i; | |
9002 | ||
9003 | if (reset) { | |
9004 | err = -EINVAL; | |
9005 | goto out; | |
9006 | } | |
9007 | ||
9008 | for_each_possible_cpu(i) { | |
9009 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9010 | ||
9011 | spin_lock_irq(&cpu_rq(i)->lock); | |
9012 | *cpuusage = 0; | |
9013 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9014 | } | |
9015 | out: | |
9016 | return err; | |
9017 | } | |
9018 | ||
d842de87 SV |
9019 | static struct cftype files[] = { |
9020 | { | |
9021 | .name = "usage", | |
f4c753b7 PM |
9022 | .read_u64 = cpuusage_read, |
9023 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9024 | }, |
9025 | }; | |
9026 | ||
32cd756a | 9027 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9028 | { |
32cd756a | 9029 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9030 | } |
9031 | ||
9032 | /* | |
9033 | * charge this task's execution time to its accounting group. | |
9034 | * | |
9035 | * called with rq->lock held. | |
9036 | */ | |
9037 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9038 | { | |
9039 | struct cpuacct *ca; | |
9040 | ||
9041 | if (!cpuacct_subsys.active) | |
9042 | return; | |
9043 | ||
9044 | ca = task_ca(tsk); | |
9045 | if (ca) { | |
9046 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
9047 | ||
9048 | *cpuusage += cputime; | |
9049 | } | |
9050 | } | |
9051 | ||
9052 | struct cgroup_subsys cpuacct_subsys = { | |
9053 | .name = "cpuacct", | |
9054 | .create = cpuacct_create, | |
9055 | .destroy = cpuacct_destroy, | |
9056 | .populate = cpuacct_populate, | |
9057 | .subsys_id = cpuacct_subsys_id, | |
9058 | }; | |
9059 | #endif /* CONFIG_CGROUP_CPUACCT */ |