]>
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> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
5517d86b | 66 | #include <linux/reciprocal_div.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
434d53b0 | 71 | #include <linux/bootmem.h> |
f00b45c1 PZ |
72 | #include <linux/debugfs.h> |
73 | #include <linux/ctype.h> | |
6cd8a4bb | 74 | #include <linux/ftrace.h> |
0a16b607 | 75 | #include <trace/sched.h> |
1da177e4 | 76 | |
5517d86b | 77 | #include <asm/tlb.h> |
838225b4 | 78 | #include <asm/irq_regs.h> |
1da177e4 | 79 | |
6e0534f2 GH |
80 | #include "sched_cpupri.h" |
81 | ||
1da177e4 LT |
82 | /* |
83 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
84 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
85 | * and back. | |
86 | */ | |
87 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
88 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
89 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
90 | ||
91 | /* | |
92 | * 'User priority' is the nice value converted to something we | |
93 | * can work with better when scaling various scheduler parameters, | |
94 | * it's a [ 0 ... 39 ] range. | |
95 | */ | |
96 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
97 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
98 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
99 | ||
100 | /* | |
d7876a08 | 101 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 102 | */ |
d6322faf | 103 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 104 | |
6aa645ea IM |
105 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
106 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
107 | ||
1da177e4 LT |
108 | /* |
109 | * These are the 'tuning knobs' of the scheduler: | |
110 | * | |
a4ec24b4 | 111 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
112 | * Timeslices get refilled after they expire. |
113 | */ | |
1da177e4 | 114 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 115 | |
d0b27fa7 PZ |
116 | /* |
117 | * single value that denotes runtime == period, ie unlimited time. | |
118 | */ | |
119 | #define RUNTIME_INF ((u64)~0ULL) | |
120 | ||
7e066fb8 MD |
121 | DEFINE_TRACE(sched_wait_task); |
122 | DEFINE_TRACE(sched_wakeup); | |
123 | DEFINE_TRACE(sched_wakeup_new); | |
124 | DEFINE_TRACE(sched_switch); | |
125 | DEFINE_TRACE(sched_migrate_task); | |
126 | ||
5517d86b ED |
127 | #ifdef CONFIG_SMP |
128 | /* | |
129 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
130 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
131 | */ | |
132 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
133 | { | |
134 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
135 | } | |
136 | ||
137 | /* | |
138 | * Each time a sched group cpu_power is changed, | |
139 | * we must compute its reciprocal value | |
140 | */ | |
141 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
142 | { | |
143 | sg->__cpu_power += val; | |
144 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
145 | } | |
146 | #endif | |
147 | ||
e05606d3 IM |
148 | static inline int rt_policy(int policy) |
149 | { | |
3f33a7ce | 150 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
151 | return 1; |
152 | return 0; | |
153 | } | |
154 | ||
155 | static inline int task_has_rt_policy(struct task_struct *p) | |
156 | { | |
157 | return rt_policy(p->policy); | |
158 | } | |
159 | ||
1da177e4 | 160 | /* |
6aa645ea | 161 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 162 | */ |
6aa645ea IM |
163 | struct rt_prio_array { |
164 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
165 | struct list_head queue[MAX_RT_PRIO]; | |
166 | }; | |
167 | ||
d0b27fa7 | 168 | struct rt_bandwidth { |
ea736ed5 IM |
169 | /* nests inside the rq lock: */ |
170 | spinlock_t rt_runtime_lock; | |
171 | ktime_t rt_period; | |
172 | u64 rt_runtime; | |
173 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
174 | }; |
175 | ||
176 | static struct rt_bandwidth def_rt_bandwidth; | |
177 | ||
178 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
179 | ||
180 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
181 | { | |
182 | struct rt_bandwidth *rt_b = | |
183 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
184 | ktime_t now; | |
185 | int overrun; | |
186 | int idle = 0; | |
187 | ||
188 | for (;;) { | |
189 | now = hrtimer_cb_get_time(timer); | |
190 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
191 | ||
192 | if (!overrun) | |
193 | break; | |
194 | ||
195 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
196 | } | |
197 | ||
198 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
199 | } | |
200 | ||
201 | static | |
202 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
203 | { | |
204 | rt_b->rt_period = ns_to_ktime(period); | |
205 | rt_b->rt_runtime = runtime; | |
206 | ||
ac086bc2 PZ |
207 | spin_lock_init(&rt_b->rt_runtime_lock); |
208 | ||
d0b27fa7 PZ |
209 | hrtimer_init(&rt_b->rt_period_timer, |
210 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
211 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
ccc7dadf | 212 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_UNLOCKED; |
d0b27fa7 PZ |
213 | } |
214 | ||
c8bfff6d KH |
215 | static inline int rt_bandwidth_enabled(void) |
216 | { | |
217 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
218 | } |
219 | ||
220 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
221 | { | |
222 | ktime_t now; | |
223 | ||
0b148fa0 | 224 | if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
225 | return; |
226 | ||
227 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
228 | return; | |
229 | ||
230 | spin_lock(&rt_b->rt_runtime_lock); | |
231 | for (;;) { | |
232 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
233 | break; | |
234 | ||
235 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
236 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
cc584b21 AV |
237 | hrtimer_start_expires(&rt_b->rt_period_timer, |
238 | HRTIMER_MODE_ABS); | |
d0b27fa7 PZ |
239 | } |
240 | spin_unlock(&rt_b->rt_runtime_lock); | |
241 | } | |
242 | ||
243 | #ifdef CONFIG_RT_GROUP_SCHED | |
244 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
245 | { | |
246 | hrtimer_cancel(&rt_b->rt_period_timer); | |
247 | } | |
248 | #endif | |
249 | ||
712555ee HC |
250 | /* |
251 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
252 | * detach_destroy_domains and partition_sched_domains. | |
253 | */ | |
254 | static DEFINE_MUTEX(sched_domains_mutex); | |
255 | ||
052f1dc7 | 256 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 257 | |
68318b8e SV |
258 | #include <linux/cgroup.h> |
259 | ||
29f59db3 SV |
260 | struct cfs_rq; |
261 | ||
6f505b16 PZ |
262 | static LIST_HEAD(task_groups); |
263 | ||
29f59db3 | 264 | /* task group related information */ |
4cf86d77 | 265 | struct task_group { |
052f1dc7 | 266 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
267 | struct cgroup_subsys_state css; |
268 | #endif | |
052f1dc7 | 269 | |
6c415b92 AB |
270 | #ifdef CONFIG_USER_SCHED |
271 | uid_t uid; | |
272 | #endif | |
273 | ||
052f1dc7 | 274 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
275 | /* schedulable entities of this group on each cpu */ |
276 | struct sched_entity **se; | |
277 | /* runqueue "owned" by this group on each cpu */ | |
278 | struct cfs_rq **cfs_rq; | |
279 | unsigned long shares; | |
052f1dc7 PZ |
280 | #endif |
281 | ||
282 | #ifdef CONFIG_RT_GROUP_SCHED | |
283 | struct sched_rt_entity **rt_se; | |
284 | struct rt_rq **rt_rq; | |
285 | ||
d0b27fa7 | 286 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 287 | #endif |
6b2d7700 | 288 | |
ae8393e5 | 289 | struct rcu_head rcu; |
6f505b16 | 290 | struct list_head list; |
f473aa5e PZ |
291 | |
292 | struct task_group *parent; | |
293 | struct list_head siblings; | |
294 | struct list_head children; | |
29f59db3 SV |
295 | }; |
296 | ||
354d60c2 | 297 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 298 | |
6c415b92 AB |
299 | /* Helper function to pass uid information to create_sched_user() */ |
300 | void set_tg_uid(struct user_struct *user) | |
301 | { | |
302 | user->tg->uid = user->uid; | |
303 | } | |
304 | ||
eff766a6 PZ |
305 | /* |
306 | * Root task group. | |
307 | * Every UID task group (including init_task_group aka UID-0) will | |
308 | * be a child to this group. | |
309 | */ | |
310 | struct task_group root_task_group; | |
311 | ||
052f1dc7 | 312 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
313 | /* Default task group's sched entity on each cpu */ |
314 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
315 | /* Default task group's cfs_rq on each cpu */ | |
316 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 317 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
318 | |
319 | #ifdef CONFIG_RT_GROUP_SCHED | |
320 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
321 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 322 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 323 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 324 | #define root_task_group init_task_group |
9a7e0b18 | 325 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 326 | |
8ed36996 | 327 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
328 | * a task group's cpu shares. |
329 | */ | |
8ed36996 | 330 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 331 | |
052f1dc7 | 332 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
333 | #ifdef CONFIG_USER_SCHED |
334 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 335 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 336 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 337 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 338 | |
cb4ad1ff | 339 | /* |
2e084786 LJ |
340 | * A weight of 0 or 1 can cause arithmetics problems. |
341 | * A weight of a cfs_rq is the sum of weights of which entities | |
342 | * are queued on this cfs_rq, so a weight of a entity should not be | |
343 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
344 | * (The default weight is 1024 - so there's no practical |
345 | * limitation from this.) | |
346 | */ | |
18d95a28 | 347 | #define MIN_SHARES 2 |
2e084786 | 348 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 349 | |
052f1dc7 PZ |
350 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
351 | #endif | |
352 | ||
29f59db3 | 353 | /* Default task group. |
3a252015 | 354 | * Every task in system belong to this group at bootup. |
29f59db3 | 355 | */ |
434d53b0 | 356 | struct task_group init_task_group; |
29f59db3 SV |
357 | |
358 | /* return group to which a task belongs */ | |
4cf86d77 | 359 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 360 | { |
4cf86d77 | 361 | struct task_group *tg; |
9b5b7751 | 362 | |
052f1dc7 | 363 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 364 | tg = p->user->tg; |
052f1dc7 | 365 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
366 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
367 | struct task_group, css); | |
24e377a8 | 368 | #else |
41a2d6cf | 369 | tg = &init_task_group; |
24e377a8 | 370 | #endif |
9b5b7751 | 371 | return tg; |
29f59db3 SV |
372 | } |
373 | ||
374 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 375 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 376 | { |
052f1dc7 | 377 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
378 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
379 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 380 | #endif |
6f505b16 | 381 | |
052f1dc7 | 382 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
383 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
384 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 385 | #endif |
29f59db3 SV |
386 | } |
387 | ||
388 | #else | |
389 | ||
6f505b16 | 390 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
391 | static inline struct task_group *task_group(struct task_struct *p) |
392 | { | |
393 | return NULL; | |
394 | } | |
29f59db3 | 395 | |
052f1dc7 | 396 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 397 | |
6aa645ea IM |
398 | /* CFS-related fields in a runqueue */ |
399 | struct cfs_rq { | |
400 | struct load_weight load; | |
401 | unsigned long nr_running; | |
402 | ||
6aa645ea | 403 | u64 exec_clock; |
e9acbff6 | 404 | u64 min_vruntime; |
6aa645ea IM |
405 | |
406 | struct rb_root tasks_timeline; | |
407 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
408 | |
409 | struct list_head tasks; | |
410 | struct list_head *balance_iterator; | |
411 | ||
412 | /* | |
413 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
414 | * It is set to NULL otherwise (i.e when none are currently running). |
415 | */ | |
4793241b | 416 | struct sched_entity *curr, *next, *last; |
ddc97297 | 417 | |
5ac5c4d6 | 418 | unsigned int nr_spread_over; |
ddc97297 | 419 | |
62160e3f | 420 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
421 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
422 | ||
41a2d6cf IM |
423 | /* |
424 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
425 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
426 | * (like users, containers etc.) | |
427 | * | |
428 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
429 | * list is used during load balance. | |
430 | */ | |
41a2d6cf IM |
431 | struct list_head leaf_cfs_rq_list; |
432 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
433 | |
434 | #ifdef CONFIG_SMP | |
c09595f6 | 435 | /* |
c8cba857 | 436 | * the part of load.weight contributed by tasks |
c09595f6 | 437 | */ |
c8cba857 | 438 | unsigned long task_weight; |
c09595f6 | 439 | |
c8cba857 PZ |
440 | /* |
441 | * h_load = weight * f(tg) | |
442 | * | |
443 | * Where f(tg) is the recursive weight fraction assigned to | |
444 | * this group. | |
445 | */ | |
446 | unsigned long h_load; | |
c09595f6 | 447 | |
c8cba857 PZ |
448 | /* |
449 | * this cpu's part of tg->shares | |
450 | */ | |
451 | unsigned long shares; | |
f1d239f7 PZ |
452 | |
453 | /* | |
454 | * load.weight at the time we set shares | |
455 | */ | |
456 | unsigned long rq_weight; | |
c09595f6 | 457 | #endif |
6aa645ea IM |
458 | #endif |
459 | }; | |
1da177e4 | 460 | |
6aa645ea IM |
461 | /* Real-Time classes' related field in a runqueue: */ |
462 | struct rt_rq { | |
463 | struct rt_prio_array active; | |
63489e45 | 464 | unsigned long rt_nr_running; |
052f1dc7 | 465 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
466 | struct { |
467 | int curr; /* highest queued rt task prio */ | |
468 | int next; /* next highest */ | |
469 | } highest_prio; | |
6f505b16 | 470 | #endif |
fa85ae24 | 471 | #ifdef CONFIG_SMP |
73fe6aae | 472 | unsigned long rt_nr_migratory; |
a22d7fc1 | 473 | int overloaded; |
fa85ae24 | 474 | #endif |
6f505b16 | 475 | int rt_throttled; |
fa85ae24 | 476 | u64 rt_time; |
ac086bc2 | 477 | u64 rt_runtime; |
ea736ed5 | 478 | /* Nests inside the rq lock: */ |
ac086bc2 | 479 | spinlock_t rt_runtime_lock; |
6f505b16 | 480 | |
052f1dc7 | 481 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
482 | unsigned long rt_nr_boosted; |
483 | ||
6f505b16 PZ |
484 | struct rq *rq; |
485 | struct list_head leaf_rt_rq_list; | |
486 | struct task_group *tg; | |
487 | struct sched_rt_entity *rt_se; | |
488 | #endif | |
6aa645ea IM |
489 | }; |
490 | ||
57d885fe GH |
491 | #ifdef CONFIG_SMP |
492 | ||
493 | /* | |
494 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
495 | * variables. Each exclusive cpuset essentially defines an island domain by |
496 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
497 | * exclusive cpuset is created, we also create and attach a new root-domain |
498 | * object. | |
499 | * | |
57d885fe GH |
500 | */ |
501 | struct root_domain { | |
502 | atomic_t refcount; | |
c6c4927b RR |
503 | cpumask_var_t span; |
504 | cpumask_var_t online; | |
637f5085 | 505 | |
0eab9146 | 506 | /* |
637f5085 GH |
507 | * The "RT overload" flag: it gets set if a CPU has more than |
508 | * one runnable RT task. | |
509 | */ | |
c6c4927b | 510 | cpumask_var_t rto_mask; |
0eab9146 | 511 | atomic_t rto_count; |
6e0534f2 GH |
512 | #ifdef CONFIG_SMP |
513 | struct cpupri cpupri; | |
514 | #endif | |
7a09b1a2 VS |
515 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
516 | /* | |
517 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
518 | * used when most cpus are idle in the system indicating overall very | |
519 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
520 | */ | |
521 | unsigned int sched_mc_preferred_wakeup_cpu; | |
522 | #endif | |
57d885fe GH |
523 | }; |
524 | ||
dc938520 GH |
525 | /* |
526 | * By default the system creates a single root-domain with all cpus as | |
527 | * members (mimicking the global state we have today). | |
528 | */ | |
57d885fe GH |
529 | static struct root_domain def_root_domain; |
530 | ||
531 | #endif | |
532 | ||
1da177e4 LT |
533 | /* |
534 | * This is the main, per-CPU runqueue data structure. | |
535 | * | |
536 | * Locking rule: those places that want to lock multiple runqueues | |
537 | * (such as the load balancing or the thread migration code), lock | |
538 | * acquire operations must be ordered by ascending &runqueue. | |
539 | */ | |
70b97a7f | 540 | struct rq { |
d8016491 IM |
541 | /* runqueue lock: */ |
542 | spinlock_t lock; | |
1da177e4 LT |
543 | |
544 | /* | |
545 | * nr_running and cpu_load should be in the same cacheline because | |
546 | * remote CPUs use both these fields when doing load calculation. | |
547 | */ | |
548 | unsigned long nr_running; | |
6aa645ea IM |
549 | #define CPU_LOAD_IDX_MAX 5 |
550 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 551 | unsigned char idle_at_tick; |
46cb4b7c | 552 | #ifdef CONFIG_NO_HZ |
15934a37 | 553 | unsigned long last_tick_seen; |
46cb4b7c SS |
554 | unsigned char in_nohz_recently; |
555 | #endif | |
d8016491 IM |
556 | /* capture load from *all* tasks on this cpu: */ |
557 | struct load_weight load; | |
6aa645ea IM |
558 | unsigned long nr_load_updates; |
559 | u64 nr_switches; | |
560 | ||
561 | struct cfs_rq cfs; | |
6f505b16 | 562 | struct rt_rq rt; |
6f505b16 | 563 | |
6aa645ea | 564 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
565 | /* list of leaf cfs_rq on this cpu: */ |
566 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
567 | #endif |
568 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 569 | struct list_head leaf_rt_rq_list; |
1da177e4 | 570 | #endif |
1da177e4 LT |
571 | |
572 | /* | |
573 | * This is part of a global counter where only the total sum | |
574 | * over all CPUs matters. A task can increase this counter on | |
575 | * one CPU and if it got migrated afterwards it may decrease | |
576 | * it on another CPU. Always updated under the runqueue lock: | |
577 | */ | |
578 | unsigned long nr_uninterruptible; | |
579 | ||
36c8b586 | 580 | struct task_struct *curr, *idle; |
c9819f45 | 581 | unsigned long next_balance; |
1da177e4 | 582 | struct mm_struct *prev_mm; |
6aa645ea | 583 | |
3e51f33f | 584 | u64 clock; |
6aa645ea | 585 | |
1da177e4 LT |
586 | atomic_t nr_iowait; |
587 | ||
588 | #ifdef CONFIG_SMP | |
0eab9146 | 589 | struct root_domain *rd; |
1da177e4 LT |
590 | struct sched_domain *sd; |
591 | ||
592 | /* For active balancing */ | |
593 | int active_balance; | |
594 | int push_cpu; | |
d8016491 IM |
595 | /* cpu of this runqueue: */ |
596 | int cpu; | |
1f11eb6a | 597 | int online; |
1da177e4 | 598 | |
a8a51d5e | 599 | unsigned long avg_load_per_task; |
1da177e4 | 600 | |
36c8b586 | 601 | struct task_struct *migration_thread; |
1da177e4 LT |
602 | struct list_head migration_queue; |
603 | #endif | |
604 | ||
8f4d37ec | 605 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
606 | #ifdef CONFIG_SMP |
607 | int hrtick_csd_pending; | |
608 | struct call_single_data hrtick_csd; | |
609 | #endif | |
8f4d37ec PZ |
610 | struct hrtimer hrtick_timer; |
611 | #endif | |
612 | ||
1da177e4 LT |
613 | #ifdef CONFIG_SCHEDSTATS |
614 | /* latency stats */ | |
615 | struct sched_info rq_sched_info; | |
616 | ||
617 | /* sys_sched_yield() stats */ | |
480b9434 KC |
618 | unsigned int yld_exp_empty; |
619 | unsigned int yld_act_empty; | |
620 | unsigned int yld_both_empty; | |
621 | unsigned int yld_count; | |
1da177e4 LT |
622 | |
623 | /* schedule() stats */ | |
480b9434 KC |
624 | unsigned int sched_switch; |
625 | unsigned int sched_count; | |
626 | unsigned int sched_goidle; | |
1da177e4 LT |
627 | |
628 | /* try_to_wake_up() stats */ | |
480b9434 KC |
629 | unsigned int ttwu_count; |
630 | unsigned int ttwu_local; | |
b8efb561 IM |
631 | |
632 | /* BKL stats */ | |
480b9434 | 633 | unsigned int bkl_count; |
1da177e4 LT |
634 | #endif |
635 | }; | |
636 | ||
f34e3b61 | 637 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 638 | |
15afe09b | 639 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 640 | { |
15afe09b | 641 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
642 | } |
643 | ||
0a2966b4 CL |
644 | static inline int cpu_of(struct rq *rq) |
645 | { | |
646 | #ifdef CONFIG_SMP | |
647 | return rq->cpu; | |
648 | #else | |
649 | return 0; | |
650 | #endif | |
651 | } | |
652 | ||
674311d5 NP |
653 | /* |
654 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 655 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
656 | * |
657 | * The domain tree of any CPU may only be accessed from within | |
658 | * preempt-disabled sections. | |
659 | */ | |
48f24c4d IM |
660 | #define for_each_domain(cpu, __sd) \ |
661 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
662 | |
663 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
664 | #define this_rq() (&__get_cpu_var(runqueues)) | |
665 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
666 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
667 | ||
3e51f33f PZ |
668 | static inline void update_rq_clock(struct rq *rq) |
669 | { | |
670 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
671 | } | |
672 | ||
bf5c91ba IM |
673 | /* |
674 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
675 | */ | |
676 | #ifdef CONFIG_SCHED_DEBUG | |
677 | # define const_debug __read_mostly | |
678 | #else | |
679 | # define const_debug static const | |
680 | #endif | |
681 | ||
017730c1 IM |
682 | /** |
683 | * runqueue_is_locked | |
684 | * | |
685 | * Returns true if the current cpu runqueue is locked. | |
686 | * This interface allows printk to be called with the runqueue lock | |
687 | * held and know whether or not it is OK to wake up the klogd. | |
688 | */ | |
689 | int runqueue_is_locked(void) | |
690 | { | |
691 | int cpu = get_cpu(); | |
692 | struct rq *rq = cpu_rq(cpu); | |
693 | int ret; | |
694 | ||
695 | ret = spin_is_locked(&rq->lock); | |
696 | put_cpu(); | |
697 | return ret; | |
698 | } | |
699 | ||
bf5c91ba IM |
700 | /* |
701 | * Debugging: various feature bits | |
702 | */ | |
f00b45c1 PZ |
703 | |
704 | #define SCHED_FEAT(name, enabled) \ | |
705 | __SCHED_FEAT_##name , | |
706 | ||
bf5c91ba | 707 | enum { |
f00b45c1 | 708 | #include "sched_features.h" |
bf5c91ba IM |
709 | }; |
710 | ||
f00b45c1 PZ |
711 | #undef SCHED_FEAT |
712 | ||
713 | #define SCHED_FEAT(name, enabled) \ | |
714 | (1UL << __SCHED_FEAT_##name) * enabled | | |
715 | ||
bf5c91ba | 716 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
717 | #include "sched_features.h" |
718 | 0; | |
719 | ||
720 | #undef SCHED_FEAT | |
721 | ||
722 | #ifdef CONFIG_SCHED_DEBUG | |
723 | #define SCHED_FEAT(name, enabled) \ | |
724 | #name , | |
725 | ||
983ed7a6 | 726 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
727 | #include "sched_features.h" |
728 | NULL | |
729 | }; | |
730 | ||
731 | #undef SCHED_FEAT | |
732 | ||
34f3a814 | 733 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 734 | { |
f00b45c1 PZ |
735 | int i; |
736 | ||
737 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
738 | if (!(sysctl_sched_features & (1UL << i))) |
739 | seq_puts(m, "NO_"); | |
740 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 741 | } |
34f3a814 | 742 | seq_puts(m, "\n"); |
f00b45c1 | 743 | |
34f3a814 | 744 | return 0; |
f00b45c1 PZ |
745 | } |
746 | ||
747 | static ssize_t | |
748 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
749 | size_t cnt, loff_t *ppos) | |
750 | { | |
751 | char buf[64]; | |
752 | char *cmp = buf; | |
753 | int neg = 0; | |
754 | int i; | |
755 | ||
756 | if (cnt > 63) | |
757 | cnt = 63; | |
758 | ||
759 | if (copy_from_user(&buf, ubuf, cnt)) | |
760 | return -EFAULT; | |
761 | ||
762 | buf[cnt] = 0; | |
763 | ||
c24b7c52 | 764 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
765 | neg = 1; |
766 | cmp += 3; | |
767 | } | |
768 | ||
769 | for (i = 0; sched_feat_names[i]; i++) { | |
770 | int len = strlen(sched_feat_names[i]); | |
771 | ||
772 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
773 | if (neg) | |
774 | sysctl_sched_features &= ~(1UL << i); | |
775 | else | |
776 | sysctl_sched_features |= (1UL << i); | |
777 | break; | |
778 | } | |
779 | } | |
780 | ||
781 | if (!sched_feat_names[i]) | |
782 | return -EINVAL; | |
783 | ||
784 | filp->f_pos += cnt; | |
785 | ||
786 | return cnt; | |
787 | } | |
788 | ||
34f3a814 LZ |
789 | static int sched_feat_open(struct inode *inode, struct file *filp) |
790 | { | |
791 | return single_open(filp, sched_feat_show, NULL); | |
792 | } | |
793 | ||
f00b45c1 | 794 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
795 | .open = sched_feat_open, |
796 | .write = sched_feat_write, | |
797 | .read = seq_read, | |
798 | .llseek = seq_lseek, | |
799 | .release = single_release, | |
f00b45c1 PZ |
800 | }; |
801 | ||
802 | static __init int sched_init_debug(void) | |
803 | { | |
f00b45c1 PZ |
804 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
805 | &sched_feat_fops); | |
806 | ||
807 | return 0; | |
808 | } | |
809 | late_initcall(sched_init_debug); | |
810 | ||
811 | #endif | |
812 | ||
813 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 814 | |
b82d9fdd PZ |
815 | /* |
816 | * Number of tasks to iterate in a single balance run. | |
817 | * Limited because this is done with IRQs disabled. | |
818 | */ | |
819 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
820 | ||
2398f2c6 PZ |
821 | /* |
822 | * ratelimit for updating the group shares. | |
55cd5340 | 823 | * default: 0.25ms |
2398f2c6 | 824 | */ |
55cd5340 | 825 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 826 | |
ffda12a1 PZ |
827 | /* |
828 | * Inject some fuzzyness into changing the per-cpu group shares | |
829 | * this avoids remote rq-locks at the expense of fairness. | |
830 | * default: 4 | |
831 | */ | |
832 | unsigned int sysctl_sched_shares_thresh = 4; | |
833 | ||
fa85ae24 | 834 | /* |
9f0c1e56 | 835 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
836 | * default: 1s |
837 | */ | |
9f0c1e56 | 838 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 839 | |
6892b75e IM |
840 | static __read_mostly int scheduler_running; |
841 | ||
9f0c1e56 PZ |
842 | /* |
843 | * part of the period that we allow rt tasks to run in us. | |
844 | * default: 0.95s | |
845 | */ | |
846 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 847 | |
d0b27fa7 PZ |
848 | static inline u64 global_rt_period(void) |
849 | { | |
850 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
851 | } | |
852 | ||
853 | static inline u64 global_rt_runtime(void) | |
854 | { | |
e26873bb | 855 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
856 | return RUNTIME_INF; |
857 | ||
858 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
859 | } | |
fa85ae24 | 860 | |
1da177e4 | 861 | #ifndef prepare_arch_switch |
4866cde0 NP |
862 | # define prepare_arch_switch(next) do { } while (0) |
863 | #endif | |
864 | #ifndef finish_arch_switch | |
865 | # define finish_arch_switch(prev) do { } while (0) | |
866 | #endif | |
867 | ||
051a1d1a DA |
868 | static inline int task_current(struct rq *rq, struct task_struct *p) |
869 | { | |
870 | return rq->curr == p; | |
871 | } | |
872 | ||
4866cde0 | 873 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 874 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 875 | { |
051a1d1a | 876 | return task_current(rq, p); |
4866cde0 NP |
877 | } |
878 | ||
70b97a7f | 879 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
880 | { |
881 | } | |
882 | ||
70b97a7f | 883 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 884 | { |
da04c035 IM |
885 | #ifdef CONFIG_DEBUG_SPINLOCK |
886 | /* this is a valid case when another task releases the spinlock */ | |
887 | rq->lock.owner = current; | |
888 | #endif | |
8a25d5de IM |
889 | /* |
890 | * If we are tracking spinlock dependencies then we have to | |
891 | * fix up the runqueue lock - which gets 'carried over' from | |
892 | * prev into current: | |
893 | */ | |
894 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
895 | ||
4866cde0 NP |
896 | spin_unlock_irq(&rq->lock); |
897 | } | |
898 | ||
899 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 900 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
901 | { |
902 | #ifdef CONFIG_SMP | |
903 | return p->oncpu; | |
904 | #else | |
051a1d1a | 905 | return task_current(rq, p); |
4866cde0 NP |
906 | #endif |
907 | } | |
908 | ||
70b97a7f | 909 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
910 | { |
911 | #ifdef CONFIG_SMP | |
912 | /* | |
913 | * We can optimise this out completely for !SMP, because the | |
914 | * SMP rebalancing from interrupt is the only thing that cares | |
915 | * here. | |
916 | */ | |
917 | next->oncpu = 1; | |
918 | #endif | |
919 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
920 | spin_unlock_irq(&rq->lock); | |
921 | #else | |
922 | spin_unlock(&rq->lock); | |
923 | #endif | |
924 | } | |
925 | ||
70b97a7f | 926 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
927 | { |
928 | #ifdef CONFIG_SMP | |
929 | /* | |
930 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
931 | * We must ensure this doesn't happen until the switch is completely | |
932 | * finished. | |
933 | */ | |
934 | smp_wmb(); | |
935 | prev->oncpu = 0; | |
936 | #endif | |
937 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
938 | local_irq_enable(); | |
1da177e4 | 939 | #endif |
4866cde0 NP |
940 | } |
941 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 942 | |
b29739f9 IM |
943 | /* |
944 | * __task_rq_lock - lock the runqueue a given task resides on. | |
945 | * Must be called interrupts disabled. | |
946 | */ | |
70b97a7f | 947 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
948 | __acquires(rq->lock) |
949 | { | |
3a5c359a AK |
950 | for (;;) { |
951 | struct rq *rq = task_rq(p); | |
952 | spin_lock(&rq->lock); | |
953 | if (likely(rq == task_rq(p))) | |
954 | return rq; | |
b29739f9 | 955 | spin_unlock(&rq->lock); |
b29739f9 | 956 | } |
b29739f9 IM |
957 | } |
958 | ||
1da177e4 LT |
959 | /* |
960 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 961 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
962 | * explicitly disabling preemption. |
963 | */ | |
70b97a7f | 964 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
965 | __acquires(rq->lock) |
966 | { | |
70b97a7f | 967 | struct rq *rq; |
1da177e4 | 968 | |
3a5c359a AK |
969 | for (;;) { |
970 | local_irq_save(*flags); | |
971 | rq = task_rq(p); | |
972 | spin_lock(&rq->lock); | |
973 | if (likely(rq == task_rq(p))) | |
974 | return rq; | |
1da177e4 | 975 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 976 | } |
1da177e4 LT |
977 | } |
978 | ||
ad474cac ON |
979 | void task_rq_unlock_wait(struct task_struct *p) |
980 | { | |
981 | struct rq *rq = task_rq(p); | |
982 | ||
983 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
984 | spin_unlock_wait(&rq->lock); | |
985 | } | |
986 | ||
a9957449 | 987 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
988 | __releases(rq->lock) |
989 | { | |
990 | spin_unlock(&rq->lock); | |
991 | } | |
992 | ||
70b97a7f | 993 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
994 | __releases(rq->lock) |
995 | { | |
996 | spin_unlock_irqrestore(&rq->lock, *flags); | |
997 | } | |
998 | ||
1da177e4 | 999 | /* |
cc2a73b5 | 1000 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1001 | */ |
a9957449 | 1002 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1003 | __acquires(rq->lock) |
1004 | { | |
70b97a7f | 1005 | struct rq *rq; |
1da177e4 LT |
1006 | |
1007 | local_irq_disable(); | |
1008 | rq = this_rq(); | |
1009 | spin_lock(&rq->lock); | |
1010 | ||
1011 | return rq; | |
1012 | } | |
1013 | ||
8f4d37ec PZ |
1014 | #ifdef CONFIG_SCHED_HRTICK |
1015 | /* | |
1016 | * Use HR-timers to deliver accurate preemption points. | |
1017 | * | |
1018 | * Its all a bit involved since we cannot program an hrt while holding the | |
1019 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1020 | * reschedule event. | |
1021 | * | |
1022 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1023 | * rq->lock. | |
1024 | */ | |
8f4d37ec PZ |
1025 | |
1026 | /* | |
1027 | * Use hrtick when: | |
1028 | * - enabled by features | |
1029 | * - hrtimer is actually high res | |
1030 | */ | |
1031 | static inline int hrtick_enabled(struct rq *rq) | |
1032 | { | |
1033 | if (!sched_feat(HRTICK)) | |
1034 | return 0; | |
ba42059f | 1035 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1036 | return 0; |
8f4d37ec PZ |
1037 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1038 | } | |
1039 | ||
8f4d37ec PZ |
1040 | static void hrtick_clear(struct rq *rq) |
1041 | { | |
1042 | if (hrtimer_active(&rq->hrtick_timer)) | |
1043 | hrtimer_cancel(&rq->hrtick_timer); | |
1044 | } | |
1045 | ||
8f4d37ec PZ |
1046 | /* |
1047 | * High-resolution timer tick. | |
1048 | * Runs from hardirq context with interrupts disabled. | |
1049 | */ | |
1050 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1051 | { | |
1052 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1053 | ||
1054 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1055 | ||
1056 | spin_lock(&rq->lock); | |
3e51f33f | 1057 | update_rq_clock(rq); |
8f4d37ec PZ |
1058 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1059 | spin_unlock(&rq->lock); | |
1060 | ||
1061 | return HRTIMER_NORESTART; | |
1062 | } | |
1063 | ||
95e904c7 | 1064 | #ifdef CONFIG_SMP |
31656519 PZ |
1065 | /* |
1066 | * called from hardirq (IPI) context | |
1067 | */ | |
1068 | static void __hrtick_start(void *arg) | |
b328ca18 | 1069 | { |
31656519 | 1070 | struct rq *rq = arg; |
b328ca18 | 1071 | |
31656519 PZ |
1072 | spin_lock(&rq->lock); |
1073 | hrtimer_restart(&rq->hrtick_timer); | |
1074 | rq->hrtick_csd_pending = 0; | |
1075 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1076 | } |
1077 | ||
31656519 PZ |
1078 | /* |
1079 | * Called to set the hrtick timer state. | |
1080 | * | |
1081 | * called with rq->lock held and irqs disabled | |
1082 | */ | |
1083 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1084 | { |
31656519 PZ |
1085 | struct hrtimer *timer = &rq->hrtick_timer; |
1086 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1087 | |
cc584b21 | 1088 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1089 | |
1090 | if (rq == this_rq()) { | |
1091 | hrtimer_restart(timer); | |
1092 | } else if (!rq->hrtick_csd_pending) { | |
1093 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1094 | rq->hrtick_csd_pending = 1; | |
1095 | } | |
b328ca18 PZ |
1096 | } |
1097 | ||
1098 | static int | |
1099 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1100 | { | |
1101 | int cpu = (int)(long)hcpu; | |
1102 | ||
1103 | switch (action) { | |
1104 | case CPU_UP_CANCELED: | |
1105 | case CPU_UP_CANCELED_FROZEN: | |
1106 | case CPU_DOWN_PREPARE: | |
1107 | case CPU_DOWN_PREPARE_FROZEN: | |
1108 | case CPU_DEAD: | |
1109 | case CPU_DEAD_FROZEN: | |
31656519 | 1110 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1111 | return NOTIFY_OK; |
1112 | } | |
1113 | ||
1114 | return NOTIFY_DONE; | |
1115 | } | |
1116 | ||
fa748203 | 1117 | static __init void init_hrtick(void) |
b328ca18 PZ |
1118 | { |
1119 | hotcpu_notifier(hotplug_hrtick, 0); | |
1120 | } | |
31656519 PZ |
1121 | #else |
1122 | /* | |
1123 | * Called to set the hrtick timer state. | |
1124 | * | |
1125 | * called with rq->lock held and irqs disabled | |
1126 | */ | |
1127 | static void hrtick_start(struct rq *rq, u64 delay) | |
1128 | { | |
1129 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1130 | } | |
b328ca18 | 1131 | |
006c75f1 | 1132 | static inline void init_hrtick(void) |
8f4d37ec | 1133 | { |
8f4d37ec | 1134 | } |
31656519 | 1135 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1136 | |
31656519 | 1137 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1138 | { |
31656519 PZ |
1139 | #ifdef CONFIG_SMP |
1140 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1141 | |
31656519 PZ |
1142 | rq->hrtick_csd.flags = 0; |
1143 | rq->hrtick_csd.func = __hrtick_start; | |
1144 | rq->hrtick_csd.info = rq; | |
1145 | #endif | |
8f4d37ec | 1146 | |
31656519 PZ |
1147 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1148 | rq->hrtick_timer.function = hrtick; | |
ccc7dadf | 1149 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_PERCPU; |
8f4d37ec | 1150 | } |
006c75f1 | 1151 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1152 | static inline void hrtick_clear(struct rq *rq) |
1153 | { | |
1154 | } | |
1155 | ||
8f4d37ec PZ |
1156 | static inline void init_rq_hrtick(struct rq *rq) |
1157 | { | |
1158 | } | |
1159 | ||
b328ca18 PZ |
1160 | static inline void init_hrtick(void) |
1161 | { | |
1162 | } | |
006c75f1 | 1163 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1164 | |
c24d20db IM |
1165 | /* |
1166 | * resched_task - mark a task 'to be rescheduled now'. | |
1167 | * | |
1168 | * On UP this means the setting of the need_resched flag, on SMP it | |
1169 | * might also involve a cross-CPU call to trigger the scheduler on | |
1170 | * the target CPU. | |
1171 | */ | |
1172 | #ifdef CONFIG_SMP | |
1173 | ||
1174 | #ifndef tsk_is_polling | |
1175 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1176 | #endif | |
1177 | ||
31656519 | 1178 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1179 | { |
1180 | int cpu; | |
1181 | ||
1182 | assert_spin_locked(&task_rq(p)->lock); | |
1183 | ||
31656519 | 1184 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1185 | return; |
1186 | ||
31656519 | 1187 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1188 | |
1189 | cpu = task_cpu(p); | |
1190 | if (cpu == smp_processor_id()) | |
1191 | return; | |
1192 | ||
1193 | /* NEED_RESCHED must be visible before we test polling */ | |
1194 | smp_mb(); | |
1195 | if (!tsk_is_polling(p)) | |
1196 | smp_send_reschedule(cpu); | |
1197 | } | |
1198 | ||
1199 | static void resched_cpu(int cpu) | |
1200 | { | |
1201 | struct rq *rq = cpu_rq(cpu); | |
1202 | unsigned long flags; | |
1203 | ||
1204 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1205 | return; | |
1206 | resched_task(cpu_curr(cpu)); | |
1207 | spin_unlock_irqrestore(&rq->lock, flags); | |
1208 | } | |
06d8308c TG |
1209 | |
1210 | #ifdef CONFIG_NO_HZ | |
1211 | /* | |
1212 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1213 | * idle CPU then this timer might expire before the next timer event | |
1214 | * which is scheduled to wake up that CPU. In case of a completely | |
1215 | * idle system the next event might even be infinite time into the | |
1216 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1217 | * leaves the inner idle loop so the newly added timer is taken into | |
1218 | * account when the CPU goes back to idle and evaluates the timer | |
1219 | * wheel for the next timer event. | |
1220 | */ | |
1221 | void wake_up_idle_cpu(int cpu) | |
1222 | { | |
1223 | struct rq *rq = cpu_rq(cpu); | |
1224 | ||
1225 | if (cpu == smp_processor_id()) | |
1226 | return; | |
1227 | ||
1228 | /* | |
1229 | * This is safe, as this function is called with the timer | |
1230 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1231 | * to idle and has not yet set rq->curr to idle then it will | |
1232 | * be serialized on the timer wheel base lock and take the new | |
1233 | * timer into account automatically. | |
1234 | */ | |
1235 | if (rq->curr != rq->idle) | |
1236 | return; | |
1237 | ||
1238 | /* | |
1239 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1240 | * lockless. The worst case is that the other CPU runs the | |
1241 | * idle task through an additional NOOP schedule() | |
1242 | */ | |
1243 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1244 | ||
1245 | /* NEED_RESCHED must be visible before we test polling */ | |
1246 | smp_mb(); | |
1247 | if (!tsk_is_polling(rq->idle)) | |
1248 | smp_send_reschedule(cpu); | |
1249 | } | |
6d6bc0ad | 1250 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1251 | |
6d6bc0ad | 1252 | #else /* !CONFIG_SMP */ |
31656519 | 1253 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1254 | { |
1255 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1256 | set_tsk_need_resched(p); |
c24d20db | 1257 | } |
6d6bc0ad | 1258 | #endif /* CONFIG_SMP */ |
c24d20db | 1259 | |
45bf76df IM |
1260 | #if BITS_PER_LONG == 32 |
1261 | # define WMULT_CONST (~0UL) | |
1262 | #else | |
1263 | # define WMULT_CONST (1UL << 32) | |
1264 | #endif | |
1265 | ||
1266 | #define WMULT_SHIFT 32 | |
1267 | ||
194081eb IM |
1268 | /* |
1269 | * Shift right and round: | |
1270 | */ | |
cf2ab469 | 1271 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1272 | |
a7be37ac PZ |
1273 | /* |
1274 | * delta *= weight / lw | |
1275 | */ | |
cb1c4fc9 | 1276 | static unsigned long |
45bf76df IM |
1277 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1278 | struct load_weight *lw) | |
1279 | { | |
1280 | u64 tmp; | |
1281 | ||
7a232e03 LJ |
1282 | if (!lw->inv_weight) { |
1283 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1284 | lw->inv_weight = 1; | |
1285 | else | |
1286 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1287 | / (lw->weight+1); | |
1288 | } | |
45bf76df IM |
1289 | |
1290 | tmp = (u64)delta_exec * weight; | |
1291 | /* | |
1292 | * Check whether we'd overflow the 64-bit multiplication: | |
1293 | */ | |
194081eb | 1294 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1295 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1296 | WMULT_SHIFT/2); |
1297 | else | |
cf2ab469 | 1298 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1299 | |
ecf691da | 1300 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1301 | } |
1302 | ||
1091985b | 1303 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1304 | { |
1305 | lw->weight += inc; | |
e89996ae | 1306 | lw->inv_weight = 0; |
45bf76df IM |
1307 | } |
1308 | ||
1091985b | 1309 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1310 | { |
1311 | lw->weight -= dec; | |
e89996ae | 1312 | lw->inv_weight = 0; |
45bf76df IM |
1313 | } |
1314 | ||
2dd73a4f PW |
1315 | /* |
1316 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1317 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1318 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1319 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1320 | * scaled version of the new time slice allocation that they receive on time |
1321 | * slice expiry etc. | |
1322 | */ | |
1323 | ||
dd41f596 IM |
1324 | #define WEIGHT_IDLEPRIO 2 |
1325 | #define WMULT_IDLEPRIO (1 << 31) | |
1326 | ||
1327 | /* | |
1328 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1329 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1330 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1331 | * that remained on nice 0. | |
1332 | * | |
1333 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1334 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1335 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1336 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1337 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1338 | */ |
1339 | static const int prio_to_weight[40] = { | |
254753dc IM |
1340 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1341 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1342 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1343 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1344 | /* 0 */ 1024, 820, 655, 526, 423, | |
1345 | /* 5 */ 335, 272, 215, 172, 137, | |
1346 | /* 10 */ 110, 87, 70, 56, 45, | |
1347 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1348 | }; |
1349 | ||
5714d2de IM |
1350 | /* |
1351 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1352 | * | |
1353 | * In cases where the weight does not change often, we can use the | |
1354 | * precalculated inverse to speed up arithmetics by turning divisions | |
1355 | * into multiplications: | |
1356 | */ | |
dd41f596 | 1357 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1358 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1359 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1360 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1361 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1362 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1363 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1364 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1365 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1366 | }; |
2dd73a4f | 1367 | |
dd41f596 IM |
1368 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1369 | ||
1370 | /* | |
1371 | * runqueue iterator, to support SMP load-balancing between different | |
1372 | * scheduling classes, without having to expose their internal data | |
1373 | * structures to the load-balancing proper: | |
1374 | */ | |
1375 | struct rq_iterator { | |
1376 | void *arg; | |
1377 | struct task_struct *(*start)(void *); | |
1378 | struct task_struct *(*next)(void *); | |
1379 | }; | |
1380 | ||
e1d1484f PW |
1381 | #ifdef CONFIG_SMP |
1382 | static unsigned long | |
1383 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1384 | unsigned long max_load_move, struct sched_domain *sd, | |
1385 | enum cpu_idle_type idle, int *all_pinned, | |
1386 | int *this_best_prio, struct rq_iterator *iterator); | |
1387 | ||
1388 | static int | |
1389 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1390 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1391 | struct rq_iterator *iterator); | |
e1d1484f | 1392 | #endif |
dd41f596 | 1393 | |
d842de87 SV |
1394 | #ifdef CONFIG_CGROUP_CPUACCT |
1395 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1396 | #else | |
1397 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1398 | #endif | |
1399 | ||
18d95a28 PZ |
1400 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1401 | { | |
1402 | update_load_add(&rq->load, load); | |
1403 | } | |
1404 | ||
1405 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1406 | { | |
1407 | update_load_sub(&rq->load, load); | |
1408 | } | |
1409 | ||
7940ca36 | 1410 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1411 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1412 | |
1413 | /* | |
1414 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1415 | * leaving it for the final time. | |
1416 | */ | |
eb755805 | 1417 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1418 | { |
1419 | struct task_group *parent, *child; | |
eb755805 | 1420 | int ret; |
c09595f6 PZ |
1421 | |
1422 | rcu_read_lock(); | |
1423 | parent = &root_task_group; | |
1424 | down: | |
eb755805 PZ |
1425 | ret = (*down)(parent, data); |
1426 | if (ret) | |
1427 | goto out_unlock; | |
c09595f6 PZ |
1428 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1429 | parent = child; | |
1430 | goto down; | |
1431 | ||
1432 | up: | |
1433 | continue; | |
1434 | } | |
eb755805 PZ |
1435 | ret = (*up)(parent, data); |
1436 | if (ret) | |
1437 | goto out_unlock; | |
c09595f6 PZ |
1438 | |
1439 | child = parent; | |
1440 | parent = parent->parent; | |
1441 | if (parent) | |
1442 | goto up; | |
eb755805 | 1443 | out_unlock: |
c09595f6 | 1444 | rcu_read_unlock(); |
eb755805 PZ |
1445 | |
1446 | return ret; | |
c09595f6 PZ |
1447 | } |
1448 | ||
eb755805 PZ |
1449 | static int tg_nop(struct task_group *tg, void *data) |
1450 | { | |
1451 | return 0; | |
c09595f6 | 1452 | } |
eb755805 PZ |
1453 | #endif |
1454 | ||
1455 | #ifdef CONFIG_SMP | |
1456 | static unsigned long source_load(int cpu, int type); | |
1457 | static unsigned long target_load(int cpu, int type); | |
1458 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1459 | ||
1460 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1461 | { | |
1462 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1463 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1464 | |
4cd42620 SR |
1465 | if (nr_running) |
1466 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1467 | else |
1468 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1469 | |
1470 | return rq->avg_load_per_task; | |
1471 | } | |
1472 | ||
1473 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1474 | |
c09595f6 PZ |
1475 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1476 | ||
1477 | /* | |
1478 | * Calculate and set the cpu's group shares. | |
1479 | */ | |
1480 | static void | |
ffda12a1 PZ |
1481 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1482 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1483 | { |
c09595f6 PZ |
1484 | unsigned long shares; |
1485 | unsigned long rq_weight; | |
1486 | ||
c8cba857 | 1487 | if (!tg->se[cpu]) |
c09595f6 PZ |
1488 | return; |
1489 | ||
ec4e0e2f | 1490 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1491 | |
c09595f6 PZ |
1492 | /* |
1493 | * \Sum shares * rq_weight | |
1494 | * shares = ----------------------- | |
1495 | * \Sum rq_weight | |
1496 | * | |
1497 | */ | |
ec4e0e2f | 1498 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1499 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1500 | |
ffda12a1 PZ |
1501 | if (abs(shares - tg->se[cpu]->load.weight) > |
1502 | sysctl_sched_shares_thresh) { | |
1503 | struct rq *rq = cpu_rq(cpu); | |
1504 | unsigned long flags; | |
c09595f6 | 1505 | |
ffda12a1 | 1506 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1507 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1508 | |
ffda12a1 PZ |
1509 | __set_se_shares(tg->se[cpu], shares); |
1510 | spin_unlock_irqrestore(&rq->lock, flags); | |
1511 | } | |
18d95a28 | 1512 | } |
c09595f6 PZ |
1513 | |
1514 | /* | |
c8cba857 PZ |
1515 | * Re-compute the task group their per cpu shares over the given domain. |
1516 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1517 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1518 | */ |
eb755805 | 1519 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1520 | { |
ec4e0e2f | 1521 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1522 | unsigned long shares = 0; |
eb755805 | 1523 | struct sched_domain *sd = data; |
c8cba857 | 1524 | int i; |
c09595f6 | 1525 | |
758b2cdc | 1526 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1527 | /* |
1528 | * If there are currently no tasks on the cpu pretend there | |
1529 | * is one of average load so that when a new task gets to | |
1530 | * run here it will not get delayed by group starvation. | |
1531 | */ | |
1532 | weight = tg->cfs_rq[i]->load.weight; | |
1533 | if (!weight) | |
1534 | weight = NICE_0_LOAD; | |
1535 | ||
1536 | tg->cfs_rq[i]->rq_weight = weight; | |
1537 | rq_weight += weight; | |
c8cba857 | 1538 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1539 | } |
c09595f6 | 1540 | |
c8cba857 PZ |
1541 | if ((!shares && rq_weight) || shares > tg->shares) |
1542 | shares = tg->shares; | |
1543 | ||
1544 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1545 | shares = tg->shares; | |
c09595f6 | 1546 | |
758b2cdc | 1547 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1548 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1549 | |
1550 | return 0; | |
c09595f6 PZ |
1551 | } |
1552 | ||
1553 | /* | |
c8cba857 PZ |
1554 | * Compute the cpu's hierarchical load factor for each task group. |
1555 | * This needs to be done in a top-down fashion because the load of a child | |
1556 | * group is a fraction of its parents load. | |
c09595f6 | 1557 | */ |
eb755805 | 1558 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1559 | { |
c8cba857 | 1560 | unsigned long load; |
eb755805 | 1561 | long cpu = (long)data; |
c09595f6 | 1562 | |
c8cba857 PZ |
1563 | if (!tg->parent) { |
1564 | load = cpu_rq(cpu)->load.weight; | |
1565 | } else { | |
1566 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1567 | load *= tg->cfs_rq[cpu]->shares; | |
1568 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1569 | } | |
c09595f6 | 1570 | |
c8cba857 | 1571 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1572 | |
eb755805 | 1573 | return 0; |
c09595f6 PZ |
1574 | } |
1575 | ||
c8cba857 | 1576 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1577 | { |
2398f2c6 PZ |
1578 | u64 now = cpu_clock(raw_smp_processor_id()); |
1579 | s64 elapsed = now - sd->last_update; | |
1580 | ||
1581 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1582 | sd->last_update = now; | |
eb755805 | 1583 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1584 | } |
4d8d595d PZ |
1585 | } |
1586 | ||
3e5459b4 PZ |
1587 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1588 | { | |
1589 | spin_unlock(&rq->lock); | |
1590 | update_shares(sd); | |
1591 | spin_lock(&rq->lock); | |
1592 | } | |
1593 | ||
eb755805 | 1594 | static void update_h_load(long cpu) |
c09595f6 | 1595 | { |
eb755805 | 1596 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1597 | } |
1598 | ||
c09595f6 PZ |
1599 | #else |
1600 | ||
c8cba857 | 1601 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1602 | { |
1603 | } | |
1604 | ||
3e5459b4 PZ |
1605 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1606 | { | |
1607 | } | |
1608 | ||
18d95a28 PZ |
1609 | #endif |
1610 | ||
70574a99 AD |
1611 | /* |
1612 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1613 | */ | |
1614 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1615 | __releases(this_rq->lock) | |
1616 | __acquires(busiest->lock) | |
1617 | __acquires(this_rq->lock) | |
1618 | { | |
1619 | int ret = 0; | |
1620 | ||
1621 | if (unlikely(!irqs_disabled())) { | |
1622 | /* printk() doesn't work good under rq->lock */ | |
1623 | spin_unlock(&this_rq->lock); | |
1624 | BUG_ON(1); | |
1625 | } | |
1626 | if (unlikely(!spin_trylock(&busiest->lock))) { | |
1627 | if (busiest < this_rq) { | |
1628 | spin_unlock(&this_rq->lock); | |
1629 | spin_lock(&busiest->lock); | |
1630 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1631 | ret = 1; | |
1632 | } else | |
1633 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1634 | } | |
1635 | return ret; | |
1636 | } | |
1637 | ||
1638 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1639 | __releases(busiest->lock) | |
1640 | { | |
1641 | spin_unlock(&busiest->lock); | |
1642 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1643 | } | |
18d95a28 PZ |
1644 | #endif |
1645 | ||
30432094 | 1646 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1647 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1648 | { | |
30432094 | 1649 | #ifdef CONFIG_SMP |
34e83e85 IM |
1650 | cfs_rq->shares = shares; |
1651 | #endif | |
1652 | } | |
30432094 | 1653 | #endif |
e7693a36 | 1654 | |
dd41f596 | 1655 | #include "sched_stats.h" |
dd41f596 | 1656 | #include "sched_idletask.c" |
5522d5d5 IM |
1657 | #include "sched_fair.c" |
1658 | #include "sched_rt.c" | |
dd41f596 IM |
1659 | #ifdef CONFIG_SCHED_DEBUG |
1660 | # include "sched_debug.c" | |
1661 | #endif | |
1662 | ||
1663 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1664 | #define for_each_class(class) \ |
1665 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1666 | |
c09595f6 | 1667 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1668 | { |
1669 | rq->nr_running++; | |
9c217245 IM |
1670 | } |
1671 | ||
c09595f6 | 1672 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1673 | { |
1674 | rq->nr_running--; | |
9c217245 IM |
1675 | } |
1676 | ||
45bf76df IM |
1677 | static void set_load_weight(struct task_struct *p) |
1678 | { | |
1679 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1680 | p->se.load.weight = prio_to_weight[0] * 2; |
1681 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1682 | return; | |
1683 | } | |
45bf76df | 1684 | |
dd41f596 IM |
1685 | /* |
1686 | * SCHED_IDLE tasks get minimal weight: | |
1687 | */ | |
1688 | if (p->policy == SCHED_IDLE) { | |
1689 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1690 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1691 | return; | |
1692 | } | |
71f8bd46 | 1693 | |
dd41f596 IM |
1694 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1695 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1696 | } |
1697 | ||
2087a1ad GH |
1698 | static void update_avg(u64 *avg, u64 sample) |
1699 | { | |
1700 | s64 diff = sample - *avg; | |
1701 | *avg += diff >> 3; | |
1702 | } | |
1703 | ||
8159f87e | 1704 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1705 | { |
dd41f596 | 1706 | sched_info_queued(p); |
fd390f6a | 1707 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1708 | p->se.on_rq = 1; |
71f8bd46 IM |
1709 | } |
1710 | ||
69be72c1 | 1711 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1712 | { |
2087a1ad GH |
1713 | if (sleep && p->se.last_wakeup) { |
1714 | update_avg(&p->se.avg_overlap, | |
1715 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1716 | p->se.last_wakeup = 0; | |
1717 | } | |
1718 | ||
46ac22ba | 1719 | sched_info_dequeued(p); |
f02231e5 | 1720 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1721 | p->se.on_rq = 0; |
71f8bd46 IM |
1722 | } |
1723 | ||
14531189 | 1724 | /* |
dd41f596 | 1725 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1726 | */ |
14531189 IM |
1727 | static inline int __normal_prio(struct task_struct *p) |
1728 | { | |
dd41f596 | 1729 | return p->static_prio; |
14531189 IM |
1730 | } |
1731 | ||
b29739f9 IM |
1732 | /* |
1733 | * Calculate the expected normal priority: i.e. priority | |
1734 | * without taking RT-inheritance into account. Might be | |
1735 | * boosted by interactivity modifiers. Changes upon fork, | |
1736 | * setprio syscalls, and whenever the interactivity | |
1737 | * estimator recalculates. | |
1738 | */ | |
36c8b586 | 1739 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1740 | { |
1741 | int prio; | |
1742 | ||
e05606d3 | 1743 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1744 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1745 | else | |
1746 | prio = __normal_prio(p); | |
1747 | return prio; | |
1748 | } | |
1749 | ||
1750 | /* | |
1751 | * Calculate the current priority, i.e. the priority | |
1752 | * taken into account by the scheduler. This value might | |
1753 | * be boosted by RT tasks, or might be boosted by | |
1754 | * interactivity modifiers. Will be RT if the task got | |
1755 | * RT-boosted. If not then it returns p->normal_prio. | |
1756 | */ | |
36c8b586 | 1757 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1758 | { |
1759 | p->normal_prio = normal_prio(p); | |
1760 | /* | |
1761 | * If we are RT tasks or we were boosted to RT priority, | |
1762 | * keep the priority unchanged. Otherwise, update priority | |
1763 | * to the normal priority: | |
1764 | */ | |
1765 | if (!rt_prio(p->prio)) | |
1766 | return p->normal_prio; | |
1767 | return p->prio; | |
1768 | } | |
1769 | ||
1da177e4 | 1770 | /* |
dd41f596 | 1771 | * activate_task - move a task to the runqueue. |
1da177e4 | 1772 | */ |
dd41f596 | 1773 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1774 | { |
d9514f6c | 1775 | if (task_contributes_to_load(p)) |
dd41f596 | 1776 | rq->nr_uninterruptible--; |
1da177e4 | 1777 | |
8159f87e | 1778 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1779 | inc_nr_running(rq); |
1da177e4 LT |
1780 | } |
1781 | ||
1da177e4 LT |
1782 | /* |
1783 | * deactivate_task - remove a task from the runqueue. | |
1784 | */ | |
2e1cb74a | 1785 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1786 | { |
d9514f6c | 1787 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1788 | rq->nr_uninterruptible++; |
1789 | ||
69be72c1 | 1790 | dequeue_task(rq, p, sleep); |
c09595f6 | 1791 | dec_nr_running(rq); |
1da177e4 LT |
1792 | } |
1793 | ||
1da177e4 LT |
1794 | /** |
1795 | * task_curr - is this task currently executing on a CPU? | |
1796 | * @p: the task in question. | |
1797 | */ | |
36c8b586 | 1798 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1799 | { |
1800 | return cpu_curr(task_cpu(p)) == p; | |
1801 | } | |
1802 | ||
dd41f596 IM |
1803 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1804 | { | |
6f505b16 | 1805 | set_task_rq(p, cpu); |
dd41f596 | 1806 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1807 | /* |
1808 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1809 | * successfuly executed on another CPU. We must ensure that updates of | |
1810 | * per-task data have been completed by this moment. | |
1811 | */ | |
1812 | smp_wmb(); | |
dd41f596 | 1813 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1814 | #endif |
2dd73a4f PW |
1815 | } |
1816 | ||
cb469845 SR |
1817 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1818 | const struct sched_class *prev_class, | |
1819 | int oldprio, int running) | |
1820 | { | |
1821 | if (prev_class != p->sched_class) { | |
1822 | if (prev_class->switched_from) | |
1823 | prev_class->switched_from(rq, p, running); | |
1824 | p->sched_class->switched_to(rq, p, running); | |
1825 | } else | |
1826 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1827 | } | |
1828 | ||
1da177e4 | 1829 | #ifdef CONFIG_SMP |
c65cc870 | 1830 | |
e958b360 TG |
1831 | /* Used instead of source_load when we know the type == 0 */ |
1832 | static unsigned long weighted_cpuload(const int cpu) | |
1833 | { | |
1834 | return cpu_rq(cpu)->load.weight; | |
1835 | } | |
1836 | ||
cc367732 IM |
1837 | /* |
1838 | * Is this task likely cache-hot: | |
1839 | */ | |
e7693a36 | 1840 | static int |
cc367732 IM |
1841 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1842 | { | |
1843 | s64 delta; | |
1844 | ||
f540a608 IM |
1845 | /* |
1846 | * Buddy candidates are cache hot: | |
1847 | */ | |
4793241b PZ |
1848 | if (sched_feat(CACHE_HOT_BUDDY) && |
1849 | (&p->se == cfs_rq_of(&p->se)->next || | |
1850 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1851 | return 1; |
1852 | ||
cc367732 IM |
1853 | if (p->sched_class != &fair_sched_class) |
1854 | return 0; | |
1855 | ||
6bc1665b IM |
1856 | if (sysctl_sched_migration_cost == -1) |
1857 | return 1; | |
1858 | if (sysctl_sched_migration_cost == 0) | |
1859 | return 0; | |
1860 | ||
cc367732 IM |
1861 | delta = now - p->se.exec_start; |
1862 | ||
1863 | return delta < (s64)sysctl_sched_migration_cost; | |
1864 | } | |
1865 | ||
1866 | ||
dd41f596 | 1867 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1868 | { |
dd41f596 IM |
1869 | int old_cpu = task_cpu(p); |
1870 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1871 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1872 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1873 | u64 clock_offset; |
dd41f596 IM |
1874 | |
1875 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1876 | |
1877 | #ifdef CONFIG_SCHEDSTATS | |
1878 | if (p->se.wait_start) | |
1879 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1880 | if (p->se.sleep_start) |
1881 | p->se.sleep_start -= clock_offset; | |
1882 | if (p->se.block_start) | |
1883 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1884 | if (old_cpu != new_cpu) { |
1885 | schedstat_inc(p, se.nr_migrations); | |
1886 | if (task_hot(p, old_rq->clock, NULL)) | |
1887 | schedstat_inc(p, se.nr_forced2_migrations); | |
1888 | } | |
6cfb0d5d | 1889 | #endif |
2830cf8c SV |
1890 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1891 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1892 | |
1893 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1894 | } |
1895 | ||
70b97a7f | 1896 | struct migration_req { |
1da177e4 | 1897 | struct list_head list; |
1da177e4 | 1898 | |
36c8b586 | 1899 | struct task_struct *task; |
1da177e4 LT |
1900 | int dest_cpu; |
1901 | ||
1da177e4 | 1902 | struct completion done; |
70b97a7f | 1903 | }; |
1da177e4 LT |
1904 | |
1905 | /* | |
1906 | * The task's runqueue lock must be held. | |
1907 | * Returns true if you have to wait for migration thread. | |
1908 | */ | |
36c8b586 | 1909 | static int |
70b97a7f | 1910 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1911 | { |
70b97a7f | 1912 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1913 | |
1914 | /* | |
1915 | * If the task is not on a runqueue (and not running), then | |
1916 | * it is sufficient to simply update the task's cpu field. | |
1917 | */ | |
dd41f596 | 1918 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1919 | set_task_cpu(p, dest_cpu); |
1920 | return 0; | |
1921 | } | |
1922 | ||
1923 | init_completion(&req->done); | |
1da177e4 LT |
1924 | req->task = p; |
1925 | req->dest_cpu = dest_cpu; | |
1926 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1927 | |
1da177e4 LT |
1928 | return 1; |
1929 | } | |
1930 | ||
1931 | /* | |
1932 | * wait_task_inactive - wait for a thread to unschedule. | |
1933 | * | |
85ba2d86 RM |
1934 | * If @match_state is nonzero, it's the @p->state value just checked and |
1935 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1936 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1937 | * we return a positive number (its total switch count). If a second call | |
1938 | * a short while later returns the same number, the caller can be sure that | |
1939 | * @p has remained unscheduled the whole time. | |
1940 | * | |
1da177e4 LT |
1941 | * The caller must ensure that the task *will* unschedule sometime soon, |
1942 | * else this function might spin for a *long* time. This function can't | |
1943 | * be called with interrupts off, or it may introduce deadlock with | |
1944 | * smp_call_function() if an IPI is sent by the same process we are | |
1945 | * waiting to become inactive. | |
1946 | */ | |
85ba2d86 | 1947 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1948 | { |
1949 | unsigned long flags; | |
dd41f596 | 1950 | int running, on_rq; |
85ba2d86 | 1951 | unsigned long ncsw; |
70b97a7f | 1952 | struct rq *rq; |
1da177e4 | 1953 | |
3a5c359a AK |
1954 | for (;;) { |
1955 | /* | |
1956 | * We do the initial early heuristics without holding | |
1957 | * any task-queue locks at all. We'll only try to get | |
1958 | * the runqueue lock when things look like they will | |
1959 | * work out! | |
1960 | */ | |
1961 | rq = task_rq(p); | |
fa490cfd | 1962 | |
3a5c359a AK |
1963 | /* |
1964 | * If the task is actively running on another CPU | |
1965 | * still, just relax and busy-wait without holding | |
1966 | * any locks. | |
1967 | * | |
1968 | * NOTE! Since we don't hold any locks, it's not | |
1969 | * even sure that "rq" stays as the right runqueue! | |
1970 | * But we don't care, since "task_running()" will | |
1971 | * return false if the runqueue has changed and p | |
1972 | * is actually now running somewhere else! | |
1973 | */ | |
85ba2d86 RM |
1974 | while (task_running(rq, p)) { |
1975 | if (match_state && unlikely(p->state != match_state)) | |
1976 | return 0; | |
3a5c359a | 1977 | cpu_relax(); |
85ba2d86 | 1978 | } |
fa490cfd | 1979 | |
3a5c359a AK |
1980 | /* |
1981 | * Ok, time to look more closely! We need the rq | |
1982 | * lock now, to be *sure*. If we're wrong, we'll | |
1983 | * just go back and repeat. | |
1984 | */ | |
1985 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 1986 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
1987 | running = task_running(rq, p); |
1988 | on_rq = p->se.on_rq; | |
85ba2d86 | 1989 | ncsw = 0; |
f31e11d8 | 1990 | if (!match_state || p->state == match_state) |
93dcf55f | 1991 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 1992 | task_rq_unlock(rq, &flags); |
fa490cfd | 1993 | |
85ba2d86 RM |
1994 | /* |
1995 | * If it changed from the expected state, bail out now. | |
1996 | */ | |
1997 | if (unlikely(!ncsw)) | |
1998 | break; | |
1999 | ||
3a5c359a AK |
2000 | /* |
2001 | * Was it really running after all now that we | |
2002 | * checked with the proper locks actually held? | |
2003 | * | |
2004 | * Oops. Go back and try again.. | |
2005 | */ | |
2006 | if (unlikely(running)) { | |
2007 | cpu_relax(); | |
2008 | continue; | |
2009 | } | |
fa490cfd | 2010 | |
3a5c359a AK |
2011 | /* |
2012 | * It's not enough that it's not actively running, | |
2013 | * it must be off the runqueue _entirely_, and not | |
2014 | * preempted! | |
2015 | * | |
2016 | * So if it wa still runnable (but just not actively | |
2017 | * running right now), it's preempted, and we should | |
2018 | * yield - it could be a while. | |
2019 | */ | |
2020 | if (unlikely(on_rq)) { | |
2021 | schedule_timeout_uninterruptible(1); | |
2022 | continue; | |
2023 | } | |
fa490cfd | 2024 | |
3a5c359a AK |
2025 | /* |
2026 | * Ahh, all good. It wasn't running, and it wasn't | |
2027 | * runnable, which means that it will never become | |
2028 | * running in the future either. We're all done! | |
2029 | */ | |
2030 | break; | |
2031 | } | |
85ba2d86 RM |
2032 | |
2033 | return ncsw; | |
1da177e4 LT |
2034 | } |
2035 | ||
2036 | /*** | |
2037 | * kick_process - kick a running thread to enter/exit the kernel | |
2038 | * @p: the to-be-kicked thread | |
2039 | * | |
2040 | * Cause a process which is running on another CPU to enter | |
2041 | * kernel-mode, without any delay. (to get signals handled.) | |
2042 | * | |
2043 | * NOTE: this function doesnt have to take the runqueue lock, | |
2044 | * because all it wants to ensure is that the remote task enters | |
2045 | * the kernel. If the IPI races and the task has been migrated | |
2046 | * to another CPU then no harm is done and the purpose has been | |
2047 | * achieved as well. | |
2048 | */ | |
36c8b586 | 2049 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2050 | { |
2051 | int cpu; | |
2052 | ||
2053 | preempt_disable(); | |
2054 | cpu = task_cpu(p); | |
2055 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2056 | smp_send_reschedule(cpu); | |
2057 | preempt_enable(); | |
2058 | } | |
2059 | ||
2060 | /* | |
2dd73a4f PW |
2061 | * Return a low guess at the load of a migration-source cpu weighted |
2062 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2063 | * |
2064 | * We want to under-estimate the load of migration sources, to | |
2065 | * balance conservatively. | |
2066 | */ | |
a9957449 | 2067 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2068 | { |
70b97a7f | 2069 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2070 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2071 | |
93b75217 | 2072 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2073 | return total; |
b910472d | 2074 | |
dd41f596 | 2075 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2076 | } |
2077 | ||
2078 | /* | |
2dd73a4f PW |
2079 | * Return a high guess at the load of a migration-target cpu weighted |
2080 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2081 | */ |
a9957449 | 2082 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2083 | { |
70b97a7f | 2084 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2085 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2086 | |
93b75217 | 2087 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2088 | return total; |
3b0bd9bc | 2089 | |
dd41f596 | 2090 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2091 | } |
2092 | ||
147cbb4b NP |
2093 | /* |
2094 | * find_idlest_group finds and returns the least busy CPU group within the | |
2095 | * domain. | |
2096 | */ | |
2097 | static struct sched_group * | |
2098 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2099 | { | |
2100 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2101 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2102 | int load_idx = sd->forkexec_idx; | |
2103 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2104 | ||
2105 | do { | |
2106 | unsigned long load, avg_load; | |
2107 | int local_group; | |
2108 | int i; | |
2109 | ||
da5a5522 | 2110 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2111 | if (!cpumask_intersects(sched_group_cpus(group), |
2112 | &p->cpus_allowed)) | |
3a5c359a | 2113 | continue; |
da5a5522 | 2114 | |
758b2cdc RR |
2115 | local_group = cpumask_test_cpu(this_cpu, |
2116 | sched_group_cpus(group)); | |
147cbb4b NP |
2117 | |
2118 | /* Tally up the load of all CPUs in the group */ | |
2119 | avg_load = 0; | |
2120 | ||
758b2cdc | 2121 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2122 | /* Bias balancing toward cpus of our domain */ |
2123 | if (local_group) | |
2124 | load = source_load(i, load_idx); | |
2125 | else | |
2126 | load = target_load(i, load_idx); | |
2127 | ||
2128 | avg_load += load; | |
2129 | } | |
2130 | ||
2131 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2132 | avg_load = sg_div_cpu_power(group, |
2133 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2134 | |
2135 | if (local_group) { | |
2136 | this_load = avg_load; | |
2137 | this = group; | |
2138 | } else if (avg_load < min_load) { | |
2139 | min_load = avg_load; | |
2140 | idlest = group; | |
2141 | } | |
3a5c359a | 2142 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2143 | |
2144 | if (!idlest || 100*this_load < imbalance*min_load) | |
2145 | return NULL; | |
2146 | return idlest; | |
2147 | } | |
2148 | ||
2149 | /* | |
0feaece9 | 2150 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2151 | */ |
95cdf3b7 | 2152 | static int |
758b2cdc | 2153 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2154 | { |
2155 | unsigned long load, min_load = ULONG_MAX; | |
2156 | int idlest = -1; | |
2157 | int i; | |
2158 | ||
da5a5522 | 2159 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2160 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2161 | load = weighted_cpuload(i); |
147cbb4b NP |
2162 | |
2163 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2164 | min_load = load; | |
2165 | idlest = i; | |
2166 | } | |
2167 | } | |
2168 | ||
2169 | return idlest; | |
2170 | } | |
2171 | ||
476d139c NP |
2172 | /* |
2173 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2174 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2175 | * SD_BALANCE_EXEC. | |
2176 | * | |
2177 | * Balance, ie. select the least loaded group. | |
2178 | * | |
2179 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2180 | * | |
2181 | * preempt must be disabled. | |
2182 | */ | |
2183 | static int sched_balance_self(int cpu, int flag) | |
2184 | { | |
2185 | struct task_struct *t = current; | |
2186 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2187 | |
c96d145e | 2188 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2189 | /* |
2190 | * If power savings logic is enabled for a domain, stop there. | |
2191 | */ | |
5c45bf27 SS |
2192 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2193 | break; | |
476d139c NP |
2194 | if (tmp->flags & flag) |
2195 | sd = tmp; | |
c96d145e | 2196 | } |
476d139c | 2197 | |
039a1c41 PZ |
2198 | if (sd) |
2199 | update_shares(sd); | |
2200 | ||
476d139c | 2201 | while (sd) { |
476d139c | 2202 | struct sched_group *group; |
1a848870 SS |
2203 | int new_cpu, weight; |
2204 | ||
2205 | if (!(sd->flags & flag)) { | |
2206 | sd = sd->child; | |
2207 | continue; | |
2208 | } | |
476d139c | 2209 | |
476d139c | 2210 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2211 | if (!group) { |
2212 | sd = sd->child; | |
2213 | continue; | |
2214 | } | |
476d139c | 2215 | |
758b2cdc | 2216 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2217 | if (new_cpu == -1 || new_cpu == cpu) { |
2218 | /* Now try balancing at a lower domain level of cpu */ | |
2219 | sd = sd->child; | |
2220 | continue; | |
2221 | } | |
476d139c | 2222 | |
1a848870 | 2223 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2224 | cpu = new_cpu; |
758b2cdc | 2225 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2226 | sd = NULL; |
476d139c | 2227 | for_each_domain(cpu, tmp) { |
758b2cdc | 2228 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2229 | break; |
2230 | if (tmp->flags & flag) | |
2231 | sd = tmp; | |
2232 | } | |
2233 | /* while loop will break here if sd == NULL */ | |
2234 | } | |
2235 | ||
2236 | return cpu; | |
2237 | } | |
2238 | ||
2239 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2240 | |
1da177e4 LT |
2241 | /*** |
2242 | * try_to_wake_up - wake up a thread | |
2243 | * @p: the to-be-woken-up thread | |
2244 | * @state: the mask of task states that can be woken | |
2245 | * @sync: do a synchronous wakeup? | |
2246 | * | |
2247 | * Put it on the run-queue if it's not already there. The "current" | |
2248 | * thread is always on the run-queue (except when the actual | |
2249 | * re-schedule is in progress), and as such you're allowed to do | |
2250 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2251 | * runnable without the overhead of this. | |
2252 | * | |
2253 | * returns failure only if the task is already active. | |
2254 | */ | |
36c8b586 | 2255 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2256 | { |
cc367732 | 2257 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2258 | unsigned long flags; |
2259 | long old_state; | |
70b97a7f | 2260 | struct rq *rq; |
1da177e4 | 2261 | |
b85d0667 IM |
2262 | if (!sched_feat(SYNC_WAKEUPS)) |
2263 | sync = 0; | |
2264 | ||
2398f2c6 PZ |
2265 | #ifdef CONFIG_SMP |
2266 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2267 | struct sched_domain *sd; | |
2268 | ||
2269 | this_cpu = raw_smp_processor_id(); | |
2270 | cpu = task_cpu(p); | |
2271 | ||
2272 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2273 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2274 | update_shares(sd); |
2275 | break; | |
2276 | } | |
2277 | } | |
2278 | } | |
2279 | #endif | |
2280 | ||
04e2f174 | 2281 | smp_wmb(); |
1da177e4 LT |
2282 | rq = task_rq_lock(p, &flags); |
2283 | old_state = p->state; | |
2284 | if (!(old_state & state)) | |
2285 | goto out; | |
2286 | ||
dd41f596 | 2287 | if (p->se.on_rq) |
1da177e4 LT |
2288 | goto out_running; |
2289 | ||
2290 | cpu = task_cpu(p); | |
cc367732 | 2291 | orig_cpu = cpu; |
1da177e4 LT |
2292 | this_cpu = smp_processor_id(); |
2293 | ||
2294 | #ifdef CONFIG_SMP | |
2295 | if (unlikely(task_running(rq, p))) | |
2296 | goto out_activate; | |
2297 | ||
5d2f5a61 DA |
2298 | cpu = p->sched_class->select_task_rq(p, sync); |
2299 | if (cpu != orig_cpu) { | |
2300 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2301 | task_rq_unlock(rq, &flags); |
2302 | /* might preempt at this point */ | |
2303 | rq = task_rq_lock(p, &flags); | |
2304 | old_state = p->state; | |
2305 | if (!(old_state & state)) | |
2306 | goto out; | |
dd41f596 | 2307 | if (p->se.on_rq) |
1da177e4 LT |
2308 | goto out_running; |
2309 | ||
2310 | this_cpu = smp_processor_id(); | |
2311 | cpu = task_cpu(p); | |
2312 | } | |
2313 | ||
e7693a36 GH |
2314 | #ifdef CONFIG_SCHEDSTATS |
2315 | schedstat_inc(rq, ttwu_count); | |
2316 | if (cpu == this_cpu) | |
2317 | schedstat_inc(rq, ttwu_local); | |
2318 | else { | |
2319 | struct sched_domain *sd; | |
2320 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2321 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2322 | schedstat_inc(sd, ttwu_wake_remote); |
2323 | break; | |
2324 | } | |
2325 | } | |
2326 | } | |
6d6bc0ad | 2327 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2328 | |
1da177e4 LT |
2329 | out_activate: |
2330 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2331 | schedstat_inc(p, se.nr_wakeups); |
2332 | if (sync) | |
2333 | schedstat_inc(p, se.nr_wakeups_sync); | |
2334 | if (orig_cpu != cpu) | |
2335 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2336 | if (cpu == this_cpu) | |
2337 | schedstat_inc(p, se.nr_wakeups_local); | |
2338 | else | |
2339 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2340 | update_rq_clock(rq); |
dd41f596 | 2341 | activate_task(rq, p, 1); |
1da177e4 LT |
2342 | success = 1; |
2343 | ||
2344 | out_running: | |
0a16b607 | 2345 | trace_sched_wakeup(rq, p); |
15afe09b | 2346 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2347 | |
1da177e4 | 2348 | p->state = TASK_RUNNING; |
9a897c5a SR |
2349 | #ifdef CONFIG_SMP |
2350 | if (p->sched_class->task_wake_up) | |
2351 | p->sched_class->task_wake_up(rq, p); | |
2352 | #endif | |
1da177e4 | 2353 | out: |
2087a1ad GH |
2354 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2355 | ||
1da177e4 LT |
2356 | task_rq_unlock(rq, &flags); |
2357 | ||
2358 | return success; | |
2359 | } | |
2360 | ||
7ad5b3a5 | 2361 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2362 | { |
d9514f6c | 2363 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2364 | } |
1da177e4 LT |
2365 | EXPORT_SYMBOL(wake_up_process); |
2366 | ||
7ad5b3a5 | 2367 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2368 | { |
2369 | return try_to_wake_up(p, state, 0); | |
2370 | } | |
2371 | ||
1da177e4 LT |
2372 | /* |
2373 | * Perform scheduler related setup for a newly forked process p. | |
2374 | * p is forked by current. | |
dd41f596 IM |
2375 | * |
2376 | * __sched_fork() is basic setup used by init_idle() too: | |
2377 | */ | |
2378 | static void __sched_fork(struct task_struct *p) | |
2379 | { | |
dd41f596 IM |
2380 | p->se.exec_start = 0; |
2381 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2382 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2383 | p->se.last_wakeup = 0; |
2384 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2385 | |
2386 | #ifdef CONFIG_SCHEDSTATS | |
2387 | p->se.wait_start = 0; | |
dd41f596 IM |
2388 | p->se.sum_sleep_runtime = 0; |
2389 | p->se.sleep_start = 0; | |
dd41f596 IM |
2390 | p->se.block_start = 0; |
2391 | p->se.sleep_max = 0; | |
2392 | p->se.block_max = 0; | |
2393 | p->se.exec_max = 0; | |
eba1ed4b | 2394 | p->se.slice_max = 0; |
dd41f596 | 2395 | p->se.wait_max = 0; |
6cfb0d5d | 2396 | #endif |
476d139c | 2397 | |
fa717060 | 2398 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2399 | p->se.on_rq = 0; |
4a55bd5e | 2400 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2401 | |
e107be36 AK |
2402 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2403 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2404 | #endif | |
2405 | ||
1da177e4 LT |
2406 | /* |
2407 | * We mark the process as running here, but have not actually | |
2408 | * inserted it onto the runqueue yet. This guarantees that | |
2409 | * nobody will actually run it, and a signal or other external | |
2410 | * event cannot wake it up and insert it on the runqueue either. | |
2411 | */ | |
2412 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2413 | } |
2414 | ||
2415 | /* | |
2416 | * fork()/clone()-time setup: | |
2417 | */ | |
2418 | void sched_fork(struct task_struct *p, int clone_flags) | |
2419 | { | |
2420 | int cpu = get_cpu(); | |
2421 | ||
2422 | __sched_fork(p); | |
2423 | ||
2424 | #ifdef CONFIG_SMP | |
2425 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2426 | #endif | |
02e4bac2 | 2427 | set_task_cpu(p, cpu); |
b29739f9 IM |
2428 | |
2429 | /* | |
2430 | * Make sure we do not leak PI boosting priority to the child: | |
2431 | */ | |
2432 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2433 | if (!rt_prio(p->prio)) |
2434 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2435 | |
52f17b6c | 2436 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2437 | if (likely(sched_info_on())) |
52f17b6c | 2438 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2439 | #endif |
d6077cb8 | 2440 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2441 | p->oncpu = 0; |
2442 | #endif | |
1da177e4 | 2443 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2444 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2445 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2446 | #endif |
476d139c | 2447 | put_cpu(); |
1da177e4 LT |
2448 | } |
2449 | ||
2450 | /* | |
2451 | * wake_up_new_task - wake up a newly created task for the first time. | |
2452 | * | |
2453 | * This function will do some initial scheduler statistics housekeeping | |
2454 | * that must be done for every newly created context, then puts the task | |
2455 | * on the runqueue and wakes it. | |
2456 | */ | |
7ad5b3a5 | 2457 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2458 | { |
2459 | unsigned long flags; | |
dd41f596 | 2460 | struct rq *rq; |
1da177e4 LT |
2461 | |
2462 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2463 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2464 | update_rq_clock(rq); |
1da177e4 LT |
2465 | |
2466 | p->prio = effective_prio(p); | |
2467 | ||
b9dca1e0 | 2468 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2469 | activate_task(rq, p, 0); |
1da177e4 | 2470 | } else { |
1da177e4 | 2471 | /* |
dd41f596 IM |
2472 | * Let the scheduling class do new task startup |
2473 | * management (if any): | |
1da177e4 | 2474 | */ |
ee0827d8 | 2475 | p->sched_class->task_new(rq, p); |
c09595f6 | 2476 | inc_nr_running(rq); |
1da177e4 | 2477 | } |
0a16b607 | 2478 | trace_sched_wakeup_new(rq, p); |
15afe09b | 2479 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2480 | #ifdef CONFIG_SMP |
2481 | if (p->sched_class->task_wake_up) | |
2482 | p->sched_class->task_wake_up(rq, p); | |
2483 | #endif | |
dd41f596 | 2484 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2485 | } |
2486 | ||
e107be36 AK |
2487 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2488 | ||
2489 | /** | |
421cee29 RD |
2490 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2491 | * @notifier: notifier struct to register | |
e107be36 AK |
2492 | */ |
2493 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2494 | { | |
2495 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2496 | } | |
2497 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2498 | ||
2499 | /** | |
2500 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2501 | * @notifier: notifier struct to unregister |
e107be36 AK |
2502 | * |
2503 | * This is safe to call from within a preemption notifier. | |
2504 | */ | |
2505 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2506 | { | |
2507 | hlist_del(¬ifier->link); | |
2508 | } | |
2509 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2510 | ||
2511 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2512 | { | |
2513 | struct preempt_notifier *notifier; | |
2514 | struct hlist_node *node; | |
2515 | ||
2516 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2517 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2518 | } | |
2519 | ||
2520 | static void | |
2521 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2522 | struct task_struct *next) | |
2523 | { | |
2524 | struct preempt_notifier *notifier; | |
2525 | struct hlist_node *node; | |
2526 | ||
2527 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2528 | notifier->ops->sched_out(notifier, next); | |
2529 | } | |
2530 | ||
6d6bc0ad | 2531 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2532 | |
2533 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2534 | { | |
2535 | } | |
2536 | ||
2537 | static void | |
2538 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2539 | struct task_struct *next) | |
2540 | { | |
2541 | } | |
2542 | ||
6d6bc0ad | 2543 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2544 | |
4866cde0 NP |
2545 | /** |
2546 | * prepare_task_switch - prepare to switch tasks | |
2547 | * @rq: the runqueue preparing to switch | |
421cee29 | 2548 | * @prev: the current task that is being switched out |
4866cde0 NP |
2549 | * @next: the task we are going to switch to. |
2550 | * | |
2551 | * This is called with the rq lock held and interrupts off. It must | |
2552 | * be paired with a subsequent finish_task_switch after the context | |
2553 | * switch. | |
2554 | * | |
2555 | * prepare_task_switch sets up locking and calls architecture specific | |
2556 | * hooks. | |
2557 | */ | |
e107be36 AK |
2558 | static inline void |
2559 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2560 | struct task_struct *next) | |
4866cde0 | 2561 | { |
e107be36 | 2562 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2563 | prepare_lock_switch(rq, next); |
2564 | prepare_arch_switch(next); | |
2565 | } | |
2566 | ||
1da177e4 LT |
2567 | /** |
2568 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2569 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2570 | * @prev: the thread we just switched away from. |
2571 | * | |
4866cde0 NP |
2572 | * finish_task_switch must be called after the context switch, paired |
2573 | * with a prepare_task_switch call before the context switch. | |
2574 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2575 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2576 | * |
2577 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2578 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2579 | * with the lock held can cause deadlocks; see schedule() for |
2580 | * details.) | |
2581 | */ | |
a9957449 | 2582 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2583 | __releases(rq->lock) |
2584 | { | |
1da177e4 | 2585 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2586 | long prev_state; |
1da177e4 LT |
2587 | |
2588 | rq->prev_mm = NULL; | |
2589 | ||
2590 | /* | |
2591 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2592 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2593 | * schedule one last time. The schedule call will never return, and |
2594 | * the scheduled task must drop that reference. | |
c394cc9f | 2595 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2596 | * still held, otherwise prev could be scheduled on another cpu, die |
2597 | * there before we look at prev->state, and then the reference would | |
2598 | * be dropped twice. | |
2599 | * Manfred Spraul <manfred@colorfullife.com> | |
2600 | */ | |
55a101f8 | 2601 | prev_state = prev->state; |
4866cde0 NP |
2602 | finish_arch_switch(prev); |
2603 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2604 | #ifdef CONFIG_SMP |
2605 | if (current->sched_class->post_schedule) | |
2606 | current->sched_class->post_schedule(rq); | |
2607 | #endif | |
e8fa1362 | 2608 | |
e107be36 | 2609 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2610 | if (mm) |
2611 | mmdrop(mm); | |
c394cc9f | 2612 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2613 | /* |
2614 | * Remove function-return probe instances associated with this | |
2615 | * task and put them back on the free list. | |
9761eea8 | 2616 | */ |
c6fd91f0 | 2617 | kprobe_flush_task(prev); |
1da177e4 | 2618 | put_task_struct(prev); |
c6fd91f0 | 2619 | } |
1da177e4 LT |
2620 | } |
2621 | ||
2622 | /** | |
2623 | * schedule_tail - first thing a freshly forked thread must call. | |
2624 | * @prev: the thread we just switched away from. | |
2625 | */ | |
36c8b586 | 2626 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2627 | __releases(rq->lock) |
2628 | { | |
70b97a7f IM |
2629 | struct rq *rq = this_rq(); |
2630 | ||
4866cde0 NP |
2631 | finish_task_switch(rq, prev); |
2632 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2633 | /* In this case, finish_task_switch does not reenable preemption */ | |
2634 | preempt_enable(); | |
2635 | #endif | |
1da177e4 | 2636 | if (current->set_child_tid) |
b488893a | 2637 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2638 | } |
2639 | ||
2640 | /* | |
2641 | * context_switch - switch to the new MM and the new | |
2642 | * thread's register state. | |
2643 | */ | |
dd41f596 | 2644 | static inline void |
70b97a7f | 2645 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2646 | struct task_struct *next) |
1da177e4 | 2647 | { |
dd41f596 | 2648 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2649 | |
e107be36 | 2650 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2651 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2652 | mm = next->mm; |
2653 | oldmm = prev->active_mm; | |
9226d125 ZA |
2654 | /* |
2655 | * For paravirt, this is coupled with an exit in switch_to to | |
2656 | * combine the page table reload and the switch backend into | |
2657 | * one hypercall. | |
2658 | */ | |
2659 | arch_enter_lazy_cpu_mode(); | |
2660 | ||
dd41f596 | 2661 | if (unlikely(!mm)) { |
1da177e4 LT |
2662 | next->active_mm = oldmm; |
2663 | atomic_inc(&oldmm->mm_count); | |
2664 | enter_lazy_tlb(oldmm, next); | |
2665 | } else | |
2666 | switch_mm(oldmm, mm, next); | |
2667 | ||
dd41f596 | 2668 | if (unlikely(!prev->mm)) { |
1da177e4 | 2669 | prev->active_mm = NULL; |
1da177e4 LT |
2670 | rq->prev_mm = oldmm; |
2671 | } | |
3a5f5e48 IM |
2672 | /* |
2673 | * Since the runqueue lock will be released by the next | |
2674 | * task (which is an invalid locking op but in the case | |
2675 | * of the scheduler it's an obvious special-case), so we | |
2676 | * do an early lockdep release here: | |
2677 | */ | |
2678 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2679 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2680 | #endif |
1da177e4 LT |
2681 | |
2682 | /* Here we just switch the register state and the stack. */ | |
2683 | switch_to(prev, next, prev); | |
2684 | ||
dd41f596 IM |
2685 | barrier(); |
2686 | /* | |
2687 | * this_rq must be evaluated again because prev may have moved | |
2688 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2689 | * frame will be invalid. | |
2690 | */ | |
2691 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2692 | } |
2693 | ||
2694 | /* | |
2695 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2696 | * | |
2697 | * externally visible scheduler statistics: current number of runnable | |
2698 | * threads, current number of uninterruptible-sleeping threads, total | |
2699 | * number of context switches performed since bootup. | |
2700 | */ | |
2701 | unsigned long nr_running(void) | |
2702 | { | |
2703 | unsigned long i, sum = 0; | |
2704 | ||
2705 | for_each_online_cpu(i) | |
2706 | sum += cpu_rq(i)->nr_running; | |
2707 | ||
2708 | return sum; | |
2709 | } | |
2710 | ||
2711 | unsigned long nr_uninterruptible(void) | |
2712 | { | |
2713 | unsigned long i, sum = 0; | |
2714 | ||
0a945022 | 2715 | for_each_possible_cpu(i) |
1da177e4 LT |
2716 | sum += cpu_rq(i)->nr_uninterruptible; |
2717 | ||
2718 | /* | |
2719 | * Since we read the counters lockless, it might be slightly | |
2720 | * inaccurate. Do not allow it to go below zero though: | |
2721 | */ | |
2722 | if (unlikely((long)sum < 0)) | |
2723 | sum = 0; | |
2724 | ||
2725 | return sum; | |
2726 | } | |
2727 | ||
2728 | unsigned long long nr_context_switches(void) | |
2729 | { | |
cc94abfc SR |
2730 | int i; |
2731 | unsigned long long sum = 0; | |
1da177e4 | 2732 | |
0a945022 | 2733 | for_each_possible_cpu(i) |
1da177e4 LT |
2734 | sum += cpu_rq(i)->nr_switches; |
2735 | ||
2736 | return sum; | |
2737 | } | |
2738 | ||
2739 | unsigned long nr_iowait(void) | |
2740 | { | |
2741 | unsigned long i, sum = 0; | |
2742 | ||
0a945022 | 2743 | for_each_possible_cpu(i) |
1da177e4 LT |
2744 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2745 | ||
2746 | return sum; | |
2747 | } | |
2748 | ||
db1b1fef JS |
2749 | unsigned long nr_active(void) |
2750 | { | |
2751 | unsigned long i, running = 0, uninterruptible = 0; | |
2752 | ||
2753 | for_each_online_cpu(i) { | |
2754 | running += cpu_rq(i)->nr_running; | |
2755 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2756 | } | |
2757 | ||
2758 | if (unlikely((long)uninterruptible < 0)) | |
2759 | uninterruptible = 0; | |
2760 | ||
2761 | return running + uninterruptible; | |
2762 | } | |
2763 | ||
48f24c4d | 2764 | /* |
dd41f596 IM |
2765 | * Update rq->cpu_load[] statistics. This function is usually called every |
2766 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2767 | */ |
dd41f596 | 2768 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2769 | { |
495eca49 | 2770 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2771 | int i, scale; |
2772 | ||
2773 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2774 | |
2775 | /* Update our load: */ | |
2776 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2777 | unsigned long old_load, new_load; | |
2778 | ||
2779 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2780 | ||
2781 | old_load = this_rq->cpu_load[i]; | |
2782 | new_load = this_load; | |
a25707f3 IM |
2783 | /* |
2784 | * Round up the averaging division if load is increasing. This | |
2785 | * prevents us from getting stuck on 9 if the load is 10, for | |
2786 | * example. | |
2787 | */ | |
2788 | if (new_load > old_load) | |
2789 | new_load += scale-1; | |
dd41f596 IM |
2790 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2791 | } | |
48f24c4d IM |
2792 | } |
2793 | ||
dd41f596 IM |
2794 | #ifdef CONFIG_SMP |
2795 | ||
1da177e4 LT |
2796 | /* |
2797 | * double_rq_lock - safely lock two runqueues | |
2798 | * | |
2799 | * Note this does not disable interrupts like task_rq_lock, | |
2800 | * you need to do so manually before calling. | |
2801 | */ | |
70b97a7f | 2802 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2803 | __acquires(rq1->lock) |
2804 | __acquires(rq2->lock) | |
2805 | { | |
054b9108 | 2806 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2807 | if (rq1 == rq2) { |
2808 | spin_lock(&rq1->lock); | |
2809 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2810 | } else { | |
c96d145e | 2811 | if (rq1 < rq2) { |
1da177e4 | 2812 | spin_lock(&rq1->lock); |
5e710e37 | 2813 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2814 | } else { |
2815 | spin_lock(&rq2->lock); | |
5e710e37 | 2816 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2817 | } |
2818 | } | |
6e82a3be IM |
2819 | update_rq_clock(rq1); |
2820 | update_rq_clock(rq2); | |
1da177e4 LT |
2821 | } |
2822 | ||
2823 | /* | |
2824 | * double_rq_unlock - safely unlock two runqueues | |
2825 | * | |
2826 | * Note this does not restore interrupts like task_rq_unlock, | |
2827 | * you need to do so manually after calling. | |
2828 | */ | |
70b97a7f | 2829 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2830 | __releases(rq1->lock) |
2831 | __releases(rq2->lock) | |
2832 | { | |
2833 | spin_unlock(&rq1->lock); | |
2834 | if (rq1 != rq2) | |
2835 | spin_unlock(&rq2->lock); | |
2836 | else | |
2837 | __release(rq2->lock); | |
2838 | } | |
2839 | ||
1da177e4 LT |
2840 | /* |
2841 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2842 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2843 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2844 | * the cpu_allowed mask is restored. |
2845 | */ | |
36c8b586 | 2846 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2847 | { |
70b97a7f | 2848 | struct migration_req req; |
1da177e4 | 2849 | unsigned long flags; |
70b97a7f | 2850 | struct rq *rq; |
1da177e4 LT |
2851 | |
2852 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 2853 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 2854 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2855 | goto out; |
2856 | ||
0a16b607 | 2857 | trace_sched_migrate_task(rq, p, dest_cpu); |
1da177e4 LT |
2858 | /* force the process onto the specified CPU */ |
2859 | if (migrate_task(p, dest_cpu, &req)) { | |
2860 | /* Need to wait for migration thread (might exit: take ref). */ | |
2861 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2862 | |
1da177e4 LT |
2863 | get_task_struct(mt); |
2864 | task_rq_unlock(rq, &flags); | |
2865 | wake_up_process(mt); | |
2866 | put_task_struct(mt); | |
2867 | wait_for_completion(&req.done); | |
36c8b586 | 2868 | |
1da177e4 LT |
2869 | return; |
2870 | } | |
2871 | out: | |
2872 | task_rq_unlock(rq, &flags); | |
2873 | } | |
2874 | ||
2875 | /* | |
476d139c NP |
2876 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2877 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2878 | */ |
2879 | void sched_exec(void) | |
2880 | { | |
1da177e4 | 2881 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2882 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2883 | put_cpu(); |
476d139c NP |
2884 | if (new_cpu != this_cpu) |
2885 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2886 | } |
2887 | ||
2888 | /* | |
2889 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2890 | * Both runqueues must be locked. | |
2891 | */ | |
dd41f596 IM |
2892 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2893 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2894 | { |
2e1cb74a | 2895 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2896 | set_task_cpu(p, this_cpu); |
dd41f596 | 2897 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2898 | /* |
2899 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2900 | * to be always true for them. | |
2901 | */ | |
15afe09b | 2902 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2903 | } |
2904 | ||
2905 | /* | |
2906 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2907 | */ | |
858119e1 | 2908 | static |
70b97a7f | 2909 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2910 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2911 | int *all_pinned) |
1da177e4 LT |
2912 | { |
2913 | /* | |
2914 | * We do not migrate tasks that are: | |
2915 | * 1) running (obviously), or | |
2916 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2917 | * 3) are cache-hot on their current CPU. | |
2918 | */ | |
96f874e2 | 2919 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 2920 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 2921 | return 0; |
cc367732 | 2922 | } |
81026794 NP |
2923 | *all_pinned = 0; |
2924 | ||
cc367732 IM |
2925 | if (task_running(rq, p)) { |
2926 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2927 | return 0; |
cc367732 | 2928 | } |
1da177e4 | 2929 | |
da84d961 IM |
2930 | /* |
2931 | * Aggressive migration if: | |
2932 | * 1) task is cache cold, or | |
2933 | * 2) too many balance attempts have failed. | |
2934 | */ | |
2935 | ||
6bc1665b IM |
2936 | if (!task_hot(p, rq->clock, sd) || |
2937 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2938 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2939 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2940 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2941 | schedstat_inc(p, se.nr_forced_migrations); |
2942 | } | |
da84d961 IM |
2943 | #endif |
2944 | return 1; | |
2945 | } | |
2946 | ||
cc367732 IM |
2947 | if (task_hot(p, rq->clock, sd)) { |
2948 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2949 | return 0; |
cc367732 | 2950 | } |
1da177e4 LT |
2951 | return 1; |
2952 | } | |
2953 | ||
e1d1484f PW |
2954 | static unsigned long |
2955 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2956 | unsigned long max_load_move, struct sched_domain *sd, | |
2957 | enum cpu_idle_type idle, int *all_pinned, | |
2958 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2959 | { |
051c6764 | 2960 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2961 | struct task_struct *p; |
2962 | long rem_load_move = max_load_move; | |
1da177e4 | 2963 | |
e1d1484f | 2964 | if (max_load_move == 0) |
1da177e4 LT |
2965 | goto out; |
2966 | ||
81026794 NP |
2967 | pinned = 1; |
2968 | ||
1da177e4 | 2969 | /* |
dd41f596 | 2970 | * Start the load-balancing iterator: |
1da177e4 | 2971 | */ |
dd41f596 IM |
2972 | p = iterator->start(iterator->arg); |
2973 | next: | |
b82d9fdd | 2974 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2975 | goto out; |
051c6764 PZ |
2976 | |
2977 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2978 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2979 | p = iterator->next(iterator->arg); |
2980 | goto next; | |
1da177e4 LT |
2981 | } |
2982 | ||
dd41f596 | 2983 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2984 | pulled++; |
dd41f596 | 2985 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2986 | |
2dd73a4f | 2987 | /* |
b82d9fdd | 2988 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2989 | */ |
e1d1484f | 2990 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2991 | if (p->prio < *this_best_prio) |
2992 | *this_best_prio = p->prio; | |
dd41f596 IM |
2993 | p = iterator->next(iterator->arg); |
2994 | goto next; | |
1da177e4 LT |
2995 | } |
2996 | out: | |
2997 | /* | |
e1d1484f | 2998 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2999 | * so we can safely collect pull_task() stats here rather than |
3000 | * inside pull_task(). | |
3001 | */ | |
3002 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3003 | |
3004 | if (all_pinned) | |
3005 | *all_pinned = pinned; | |
e1d1484f PW |
3006 | |
3007 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3008 | } |
3009 | ||
dd41f596 | 3010 | /* |
43010659 PW |
3011 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3012 | * this_rq, as part of a balancing operation within domain "sd". | |
3013 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3014 | * |
3015 | * Called with both runqueues locked. | |
3016 | */ | |
3017 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3018 | unsigned long max_load_move, |
dd41f596 IM |
3019 | struct sched_domain *sd, enum cpu_idle_type idle, |
3020 | int *all_pinned) | |
3021 | { | |
5522d5d5 | 3022 | const struct sched_class *class = sched_class_highest; |
43010659 | 3023 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3024 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3025 | |
3026 | do { | |
43010659 PW |
3027 | total_load_moved += |
3028 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3029 | max_load_move - total_load_moved, |
a4ac01c3 | 3030 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3031 | class = class->next; |
c4acb2c0 GH |
3032 | |
3033 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3034 | break; | |
3035 | ||
43010659 | 3036 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3037 | |
43010659 PW |
3038 | return total_load_moved > 0; |
3039 | } | |
3040 | ||
e1d1484f PW |
3041 | static int |
3042 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3043 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3044 | struct rq_iterator *iterator) | |
3045 | { | |
3046 | struct task_struct *p = iterator->start(iterator->arg); | |
3047 | int pinned = 0; | |
3048 | ||
3049 | while (p) { | |
3050 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3051 | pull_task(busiest, p, this_rq, this_cpu); | |
3052 | /* | |
3053 | * Right now, this is only the second place pull_task() | |
3054 | * is called, so we can safely collect pull_task() | |
3055 | * stats here rather than inside pull_task(). | |
3056 | */ | |
3057 | schedstat_inc(sd, lb_gained[idle]); | |
3058 | ||
3059 | return 1; | |
3060 | } | |
3061 | p = iterator->next(iterator->arg); | |
3062 | } | |
3063 | ||
3064 | return 0; | |
3065 | } | |
3066 | ||
43010659 PW |
3067 | /* |
3068 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3069 | * part of active balancing operations within "domain". | |
3070 | * Returns 1 if successful and 0 otherwise. | |
3071 | * | |
3072 | * Called with both runqueues locked. | |
3073 | */ | |
3074 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3075 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3076 | { | |
5522d5d5 | 3077 | const struct sched_class *class; |
43010659 PW |
3078 | |
3079 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3080 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3081 | return 1; |
3082 | ||
3083 | return 0; | |
dd41f596 IM |
3084 | } |
3085 | ||
1da177e4 LT |
3086 | /* |
3087 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3088 | * domain. It calculates and returns the amount of weighted load which |
3089 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3090 | */ |
3091 | static struct sched_group * | |
3092 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3093 | unsigned long *imbalance, enum cpu_idle_type idle, |
96f874e2 | 3094 | int *sd_idle, const struct cpumask *cpus, int *balance) |
1da177e4 LT |
3095 | { |
3096 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3097 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3098 | unsigned long max_pull; |
2dd73a4f PW |
3099 | unsigned long busiest_load_per_task, busiest_nr_running; |
3100 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3101 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3102 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3103 | int power_savings_balance = 1; | |
3104 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3105 | unsigned long min_nr_running = ULONG_MAX; | |
3106 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3107 | #endif | |
1da177e4 LT |
3108 | |
3109 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3110 | busiest_load_per_task = busiest_nr_running = 0; |
3111 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3112 | |
d15bcfdb | 3113 | if (idle == CPU_NOT_IDLE) |
7897986b | 3114 | load_idx = sd->busy_idx; |
d15bcfdb | 3115 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3116 | load_idx = sd->newidle_idx; |
3117 | else | |
3118 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3119 | |
3120 | do { | |
908a7c1b | 3121 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3122 | int local_group; |
3123 | int i; | |
908a7c1b | 3124 | int __group_imb = 0; |
783609c6 | 3125 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3126 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3127 | unsigned long sum_avg_load_per_task; |
3128 | unsigned long avg_load_per_task; | |
1da177e4 | 3129 | |
758b2cdc RR |
3130 | local_group = cpumask_test_cpu(this_cpu, |
3131 | sched_group_cpus(group)); | |
1da177e4 | 3132 | |
783609c6 | 3133 | if (local_group) |
758b2cdc | 3134 | balance_cpu = cpumask_first(sched_group_cpus(group)); |
783609c6 | 3135 | |
1da177e4 | 3136 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3137 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3138 | sum_avg_load_per_task = avg_load_per_task = 0; |
3139 | ||
908a7c1b KC |
3140 | max_cpu_load = 0; |
3141 | min_cpu_load = ~0UL; | |
1da177e4 | 3142 | |
758b2cdc RR |
3143 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3144 | struct rq *rq = cpu_rq(i); | |
2dd73a4f | 3145 | |
9439aab8 | 3146 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3147 | *sd_idle = 0; |
3148 | ||
1da177e4 | 3149 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3150 | if (local_group) { |
3151 | if (idle_cpu(i) && !first_idle_cpu) { | |
3152 | first_idle_cpu = 1; | |
3153 | balance_cpu = i; | |
3154 | } | |
3155 | ||
a2000572 | 3156 | load = target_load(i, load_idx); |
908a7c1b | 3157 | } else { |
a2000572 | 3158 | load = source_load(i, load_idx); |
908a7c1b KC |
3159 | if (load > max_cpu_load) |
3160 | max_cpu_load = load; | |
3161 | if (min_cpu_load > load) | |
3162 | min_cpu_load = load; | |
3163 | } | |
1da177e4 LT |
3164 | |
3165 | avg_load += load; | |
2dd73a4f | 3166 | sum_nr_running += rq->nr_running; |
dd41f596 | 3167 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3168 | |
3169 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3170 | } |
3171 | ||
783609c6 SS |
3172 | /* |
3173 | * First idle cpu or the first cpu(busiest) in this sched group | |
3174 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3175 | * domains. In the newly idle case, we will allow all the cpu's |
3176 | * to do the newly idle load balance. | |
783609c6 | 3177 | */ |
9439aab8 SS |
3178 | if (idle != CPU_NEWLY_IDLE && local_group && |
3179 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3180 | *balance = 0; |
3181 | goto ret; | |
3182 | } | |
3183 | ||
1da177e4 | 3184 | total_load += avg_load; |
5517d86b | 3185 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3186 | |
3187 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3188 | avg_load = sg_div_cpu_power(group, |
3189 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3190 | |
408ed066 PZ |
3191 | |
3192 | /* | |
3193 | * Consider the group unbalanced when the imbalance is larger | |
3194 | * than the average weight of two tasks. | |
3195 | * | |
3196 | * APZ: with cgroup the avg task weight can vary wildly and | |
3197 | * might not be a suitable number - should we keep a | |
3198 | * normalized nr_running number somewhere that negates | |
3199 | * the hierarchy? | |
3200 | */ | |
3201 | avg_load_per_task = sg_div_cpu_power(group, | |
3202 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3203 | ||
3204 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3205 | __group_imb = 1; |
3206 | ||
5517d86b | 3207 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3208 | |
1da177e4 LT |
3209 | if (local_group) { |
3210 | this_load = avg_load; | |
3211 | this = group; | |
2dd73a4f PW |
3212 | this_nr_running = sum_nr_running; |
3213 | this_load_per_task = sum_weighted_load; | |
3214 | } else if (avg_load > max_load && | |
908a7c1b | 3215 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3216 | max_load = avg_load; |
3217 | busiest = group; | |
2dd73a4f PW |
3218 | busiest_nr_running = sum_nr_running; |
3219 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3220 | group_imb = __group_imb; |
1da177e4 | 3221 | } |
5c45bf27 SS |
3222 | |
3223 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3224 | /* | |
3225 | * Busy processors will not participate in power savings | |
3226 | * balance. | |
3227 | */ | |
dd41f596 IM |
3228 | if (idle == CPU_NOT_IDLE || |
3229 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3230 | goto group_next; | |
5c45bf27 SS |
3231 | |
3232 | /* | |
3233 | * If the local group is idle or completely loaded | |
3234 | * no need to do power savings balance at this domain | |
3235 | */ | |
3236 | if (local_group && (this_nr_running >= group_capacity || | |
3237 | !this_nr_running)) | |
3238 | power_savings_balance = 0; | |
3239 | ||
dd41f596 | 3240 | /* |
5c45bf27 SS |
3241 | * If a group is already running at full capacity or idle, |
3242 | * don't include that group in power savings calculations | |
dd41f596 IM |
3243 | */ |
3244 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3245 | || !sum_nr_running) |
dd41f596 | 3246 | goto group_next; |
5c45bf27 | 3247 | |
dd41f596 | 3248 | /* |
5c45bf27 | 3249 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3250 | * This is the group from where we need to pick up the load |
3251 | * for saving power | |
3252 | */ | |
3253 | if ((sum_nr_running < min_nr_running) || | |
3254 | (sum_nr_running == min_nr_running && | |
d5679bd1 | 3255 | cpumask_first(sched_group_cpus(group)) > |
758b2cdc | 3256 | cpumask_first(sched_group_cpus(group_min)))) { |
dd41f596 IM |
3257 | group_min = group; |
3258 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3259 | min_load_per_task = sum_weighted_load / |
3260 | sum_nr_running; | |
dd41f596 | 3261 | } |
5c45bf27 | 3262 | |
dd41f596 | 3263 | /* |
5c45bf27 | 3264 | * Calculate the group which is almost near its |
dd41f596 IM |
3265 | * capacity but still has some space to pick up some load |
3266 | * from other group and save more power | |
3267 | */ | |
3268 | if (sum_nr_running <= group_capacity - 1) { | |
3269 | if (sum_nr_running > leader_nr_running || | |
3270 | (sum_nr_running == leader_nr_running && | |
d5679bd1 | 3271 | cpumask_first(sched_group_cpus(group)) < |
758b2cdc | 3272 | cpumask_first(sched_group_cpus(group_leader)))) { |
dd41f596 IM |
3273 | group_leader = group; |
3274 | leader_nr_running = sum_nr_running; | |
3275 | } | |
48f24c4d | 3276 | } |
5c45bf27 SS |
3277 | group_next: |
3278 | #endif | |
1da177e4 LT |
3279 | group = group->next; |
3280 | } while (group != sd->groups); | |
3281 | ||
2dd73a4f | 3282 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3283 | goto out_balanced; |
3284 | ||
3285 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3286 | ||
3287 | if (this_load >= avg_load || | |
3288 | 100*max_load <= sd->imbalance_pct*this_load) | |
3289 | goto out_balanced; | |
3290 | ||
2dd73a4f | 3291 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3292 | if (group_imb) |
3293 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3294 | ||
1da177e4 LT |
3295 | /* |
3296 | * We're trying to get all the cpus to the average_load, so we don't | |
3297 | * want to push ourselves above the average load, nor do we wish to | |
3298 | * reduce the max loaded cpu below the average load, as either of these | |
3299 | * actions would just result in more rebalancing later, and ping-pong | |
3300 | * tasks around. Thus we look for the minimum possible imbalance. | |
3301 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3302 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3303 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3304 | * appear as very large values with unsigned longs. |
3305 | */ | |
2dd73a4f PW |
3306 | if (max_load <= busiest_load_per_task) |
3307 | goto out_balanced; | |
3308 | ||
3309 | /* | |
3310 | * In the presence of smp nice balancing, certain scenarios can have | |
3311 | * max load less than avg load(as we skip the groups at or below | |
3312 | * its cpu_power, while calculating max_load..) | |
3313 | */ | |
3314 | if (max_load < avg_load) { | |
3315 | *imbalance = 0; | |
3316 | goto small_imbalance; | |
3317 | } | |
0c117f1b SS |
3318 | |
3319 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3320 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3321 | |
1da177e4 | 3322 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3323 | *imbalance = min(max_pull * busiest->__cpu_power, |
3324 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3325 | / SCHED_LOAD_SCALE; |
3326 | ||
2dd73a4f PW |
3327 | /* |
3328 | * if *imbalance is less than the average load per runnable task | |
3329 | * there is no gaurantee that any tasks will be moved so we'll have | |
3330 | * a think about bumping its value to force at least one task to be | |
3331 | * moved | |
3332 | */ | |
7fd0d2dd | 3333 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3334 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3335 | unsigned int imbn; |
3336 | ||
3337 | small_imbalance: | |
3338 | pwr_move = pwr_now = 0; | |
3339 | imbn = 2; | |
3340 | if (this_nr_running) { | |
3341 | this_load_per_task /= this_nr_running; | |
3342 | if (busiest_load_per_task > this_load_per_task) | |
3343 | imbn = 1; | |
3344 | } else | |
408ed066 | 3345 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3346 | |
01c8c57d | 3347 | if (max_load - this_load + busiest_load_per_task >= |
dd41f596 | 3348 | busiest_load_per_task * imbn) { |
2dd73a4f | 3349 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3350 | return busiest; |
3351 | } | |
3352 | ||
3353 | /* | |
3354 | * OK, we don't have enough imbalance to justify moving tasks, | |
3355 | * however we may be able to increase total CPU power used by | |
3356 | * moving them. | |
3357 | */ | |
3358 | ||
5517d86b ED |
3359 | pwr_now += busiest->__cpu_power * |
3360 | min(busiest_load_per_task, max_load); | |
3361 | pwr_now += this->__cpu_power * | |
3362 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3363 | pwr_now /= SCHED_LOAD_SCALE; |
3364 | ||
3365 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3366 | tmp = sg_div_cpu_power(busiest, |
3367 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3368 | if (max_load > tmp) |
5517d86b | 3369 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3370 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3371 | |
3372 | /* Amount of load we'd add */ | |
5517d86b | 3373 | if (max_load * busiest->__cpu_power < |
33859f7f | 3374 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3375 | tmp = sg_div_cpu_power(this, |
3376 | max_load * busiest->__cpu_power); | |
1da177e4 | 3377 | else |
5517d86b ED |
3378 | tmp = sg_div_cpu_power(this, |
3379 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3380 | pwr_move += this->__cpu_power * | |
3381 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3382 | pwr_move /= SCHED_LOAD_SCALE; |
3383 | ||
3384 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3385 | if (pwr_move > pwr_now) |
3386 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3387 | } |
3388 | ||
1da177e4 LT |
3389 | return busiest; |
3390 | ||
3391 | out_balanced: | |
5c45bf27 | 3392 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3393 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3394 | goto ret; |
1da177e4 | 3395 | |
5c45bf27 SS |
3396 | if (this == group_leader && group_leader != group_min) { |
3397 | *imbalance = min_load_per_task; | |
7a09b1a2 VS |
3398 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3399 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
9924da43 | 3400 | cpumask_first(sched_group_cpus(group_leader)); |
7a09b1a2 | 3401 | } |
5c45bf27 SS |
3402 | return group_min; |
3403 | } | |
5c45bf27 | 3404 | #endif |
783609c6 | 3405 | ret: |
1da177e4 LT |
3406 | *imbalance = 0; |
3407 | return NULL; | |
3408 | } | |
3409 | ||
3410 | /* | |
3411 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3412 | */ | |
70b97a7f | 3413 | static struct rq * |
d15bcfdb | 3414 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3415 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3416 | { |
70b97a7f | 3417 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3418 | unsigned long max_load = 0; |
1da177e4 LT |
3419 | int i; |
3420 | ||
758b2cdc | 3421 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3422 | unsigned long wl; |
0a2966b4 | 3423 | |
96f874e2 | 3424 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3425 | continue; |
3426 | ||
48f24c4d | 3427 | rq = cpu_rq(i); |
dd41f596 | 3428 | wl = weighted_cpuload(i); |
2dd73a4f | 3429 | |
dd41f596 | 3430 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3431 | continue; |
1da177e4 | 3432 | |
dd41f596 IM |
3433 | if (wl > max_load) { |
3434 | max_load = wl; | |
48f24c4d | 3435 | busiest = rq; |
1da177e4 LT |
3436 | } |
3437 | } | |
3438 | ||
3439 | return busiest; | |
3440 | } | |
3441 | ||
77391d71 NP |
3442 | /* |
3443 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3444 | * so long as it is large enough. | |
3445 | */ | |
3446 | #define MAX_PINNED_INTERVAL 512 | |
3447 | ||
1da177e4 LT |
3448 | /* |
3449 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3450 | * tasks if there is an imbalance. | |
1da177e4 | 3451 | */ |
70b97a7f | 3452 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3453 | struct sched_domain *sd, enum cpu_idle_type idle, |
96f874e2 | 3454 | int *balance, struct cpumask *cpus) |
1da177e4 | 3455 | { |
43010659 | 3456 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3457 | struct sched_group *group; |
1da177e4 | 3458 | unsigned long imbalance; |
70b97a7f | 3459 | struct rq *busiest; |
fe2eea3f | 3460 | unsigned long flags; |
5969fe06 | 3461 | |
96f874e2 | 3462 | cpumask_setall(cpus); |
7c16ec58 | 3463 | |
89c4710e SS |
3464 | /* |
3465 | * When power savings policy is enabled for the parent domain, idle | |
3466 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3467 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3468 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3469 | */ |
d15bcfdb | 3470 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3471 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3472 | sd_idle = 1; |
1da177e4 | 3473 | |
2d72376b | 3474 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3475 | |
0a2966b4 | 3476 | redo: |
c8cba857 | 3477 | update_shares(sd); |
0a2966b4 | 3478 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3479 | cpus, balance); |
783609c6 | 3480 | |
06066714 | 3481 | if (*balance == 0) |
783609c6 | 3482 | goto out_balanced; |
783609c6 | 3483 | |
1da177e4 LT |
3484 | if (!group) { |
3485 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3486 | goto out_balanced; | |
3487 | } | |
3488 | ||
7c16ec58 | 3489 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3490 | if (!busiest) { |
3491 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3492 | goto out_balanced; | |
3493 | } | |
3494 | ||
db935dbd | 3495 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3496 | |
3497 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3498 | ||
43010659 | 3499 | ld_moved = 0; |
1da177e4 LT |
3500 | if (busiest->nr_running > 1) { |
3501 | /* | |
3502 | * Attempt to move tasks. If find_busiest_group has found | |
3503 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3504 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3505 | * correctly treated as an imbalance. |
3506 | */ | |
fe2eea3f | 3507 | local_irq_save(flags); |
e17224bf | 3508 | double_rq_lock(this_rq, busiest); |
43010659 | 3509 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3510 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3511 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3512 | local_irq_restore(flags); |
81026794 | 3513 | |
46cb4b7c SS |
3514 | /* |
3515 | * some other cpu did the load balance for us. | |
3516 | */ | |
43010659 | 3517 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3518 | resched_cpu(this_cpu); |
3519 | ||
81026794 | 3520 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3521 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3522 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3523 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 3524 | goto redo; |
81026794 | 3525 | goto out_balanced; |
0a2966b4 | 3526 | } |
1da177e4 | 3527 | } |
81026794 | 3528 | |
43010659 | 3529 | if (!ld_moved) { |
1da177e4 LT |
3530 | schedstat_inc(sd, lb_failed[idle]); |
3531 | sd->nr_balance_failed++; | |
3532 | ||
3533 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3534 | |
fe2eea3f | 3535 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3536 | |
3537 | /* don't kick the migration_thread, if the curr | |
3538 | * task on busiest cpu can't be moved to this_cpu | |
3539 | */ | |
96f874e2 RR |
3540 | if (!cpumask_test_cpu(this_cpu, |
3541 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3542 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3543 | all_pinned = 1; |
3544 | goto out_one_pinned; | |
3545 | } | |
3546 | ||
1da177e4 LT |
3547 | if (!busiest->active_balance) { |
3548 | busiest->active_balance = 1; | |
3549 | busiest->push_cpu = this_cpu; | |
81026794 | 3550 | active_balance = 1; |
1da177e4 | 3551 | } |
fe2eea3f | 3552 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3553 | if (active_balance) |
1da177e4 LT |
3554 | wake_up_process(busiest->migration_thread); |
3555 | ||
3556 | /* | |
3557 | * We've kicked active balancing, reset the failure | |
3558 | * counter. | |
3559 | */ | |
39507451 | 3560 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3561 | } |
81026794 | 3562 | } else |
1da177e4 LT |
3563 | sd->nr_balance_failed = 0; |
3564 | ||
81026794 | 3565 | if (likely(!active_balance)) { |
1da177e4 LT |
3566 | /* We were unbalanced, so reset the balancing interval */ |
3567 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3568 | } else { |
3569 | /* | |
3570 | * If we've begun active balancing, start to back off. This | |
3571 | * case may not be covered by the all_pinned logic if there | |
3572 | * is only 1 task on the busy runqueue (because we don't call | |
3573 | * move_tasks). | |
3574 | */ | |
3575 | if (sd->balance_interval < sd->max_interval) | |
3576 | sd->balance_interval *= 2; | |
1da177e4 LT |
3577 | } |
3578 | ||
43010659 | 3579 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3580 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3581 | ld_moved = -1; |
3582 | ||
3583 | goto out; | |
1da177e4 LT |
3584 | |
3585 | out_balanced: | |
1da177e4 LT |
3586 | schedstat_inc(sd, lb_balanced[idle]); |
3587 | ||
16cfb1c0 | 3588 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3589 | |
3590 | out_one_pinned: | |
1da177e4 | 3591 | /* tune up the balancing interval */ |
77391d71 NP |
3592 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3593 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3594 | sd->balance_interval *= 2; |
3595 | ||
48f24c4d | 3596 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3597 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3598 | ld_moved = -1; |
3599 | else | |
3600 | ld_moved = 0; | |
3601 | out: | |
c8cba857 PZ |
3602 | if (ld_moved) |
3603 | update_shares(sd); | |
c09595f6 | 3604 | return ld_moved; |
1da177e4 LT |
3605 | } |
3606 | ||
3607 | /* | |
3608 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3609 | * tasks if there is an imbalance. | |
3610 | * | |
d15bcfdb | 3611 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3612 | * this_rq is locked. |
3613 | */ | |
48f24c4d | 3614 | static int |
7c16ec58 | 3615 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
96f874e2 | 3616 | struct cpumask *cpus) |
1da177e4 LT |
3617 | { |
3618 | struct sched_group *group; | |
70b97a7f | 3619 | struct rq *busiest = NULL; |
1da177e4 | 3620 | unsigned long imbalance; |
43010659 | 3621 | int ld_moved = 0; |
5969fe06 | 3622 | int sd_idle = 0; |
969bb4e4 | 3623 | int all_pinned = 0; |
7c16ec58 | 3624 | |
96f874e2 | 3625 | cpumask_setall(cpus); |
5969fe06 | 3626 | |
89c4710e SS |
3627 | /* |
3628 | * When power savings policy is enabled for the parent domain, idle | |
3629 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3630 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3631 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3632 | */ |
3633 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3634 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3635 | sd_idle = 1; |
1da177e4 | 3636 | |
2d72376b | 3637 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3638 | redo: |
3e5459b4 | 3639 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3640 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3641 | &sd_idle, cpus, NULL); |
1da177e4 | 3642 | if (!group) { |
d15bcfdb | 3643 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3644 | goto out_balanced; |
1da177e4 LT |
3645 | } |
3646 | ||
7c16ec58 | 3647 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3648 | if (!busiest) { |
d15bcfdb | 3649 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3650 | goto out_balanced; |
1da177e4 LT |
3651 | } |
3652 | ||
db935dbd NP |
3653 | BUG_ON(busiest == this_rq); |
3654 | ||
d15bcfdb | 3655 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3656 | |
43010659 | 3657 | ld_moved = 0; |
d6d5cfaf NP |
3658 | if (busiest->nr_running > 1) { |
3659 | /* Attempt to move tasks */ | |
3660 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3661 | /* this_rq->clock is already updated */ |
3662 | update_rq_clock(busiest); | |
43010659 | 3663 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3664 | imbalance, sd, CPU_NEWLY_IDLE, |
3665 | &all_pinned); | |
1b12bbc7 | 3666 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3667 | |
969bb4e4 | 3668 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3669 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3670 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
3671 | goto redo; |
3672 | } | |
d6d5cfaf NP |
3673 | } |
3674 | ||
43010659 | 3675 | if (!ld_moved) { |
36dffab6 | 3676 | int active_balance = 0; |
ad273b32 | 3677 | |
d15bcfdb | 3678 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3679 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3680 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3681 | return -1; |
ad273b32 VS |
3682 | |
3683 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
3684 | return -1; | |
3685 | ||
3686 | if (sd->nr_balance_failed++ < 2) | |
3687 | return -1; | |
3688 | ||
3689 | /* | |
3690 | * The only task running in a non-idle cpu can be moved to this | |
3691 | * cpu in an attempt to completely freeup the other CPU | |
3692 | * package. The same method used to move task in load_balance() | |
3693 | * have been extended for load_balance_newidle() to speedup | |
3694 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
3695 | * | |
3696 | * The package power saving logic comes from | |
3697 | * find_busiest_group(). If there are no imbalance, then | |
3698 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
3699 | * f_b_g() will select a group from which a running task may be | |
3700 | * pulled to this cpu in order to make the other package idle. | |
3701 | * If there is no opportunity to make a package idle and if | |
3702 | * there are no imbalance, then f_b_g() will return NULL and no | |
3703 | * action will be taken in load_balance_newidle(). | |
3704 | * | |
3705 | * Under normal task pull operation due to imbalance, there | |
3706 | * will be more than one task in the source run queue and | |
3707 | * move_tasks() will succeed. ld_moved will be true and this | |
3708 | * active balance code will not be triggered. | |
3709 | */ | |
3710 | ||
3711 | /* Lock busiest in correct order while this_rq is held */ | |
3712 | double_lock_balance(this_rq, busiest); | |
3713 | ||
3714 | /* | |
3715 | * don't kick the migration_thread, if the curr | |
3716 | * task on busiest cpu can't be moved to this_cpu | |
3717 | */ | |
3718 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
3719 | double_unlock_balance(this_rq, busiest); | |
3720 | all_pinned = 1; | |
3721 | return ld_moved; | |
3722 | } | |
3723 | ||
3724 | if (!busiest->active_balance) { | |
3725 | busiest->active_balance = 1; | |
3726 | busiest->push_cpu = this_cpu; | |
3727 | active_balance = 1; | |
3728 | } | |
3729 | ||
3730 | double_unlock_balance(this_rq, busiest); | |
3731 | if (active_balance) | |
3732 | wake_up_process(busiest->migration_thread); | |
3733 | ||
5969fe06 | 3734 | } else |
16cfb1c0 | 3735 | sd->nr_balance_failed = 0; |
1da177e4 | 3736 | |
3e5459b4 | 3737 | update_shares_locked(this_rq, sd); |
43010659 | 3738 | return ld_moved; |
16cfb1c0 NP |
3739 | |
3740 | out_balanced: | |
d15bcfdb | 3741 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3742 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3743 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3744 | return -1; |
16cfb1c0 | 3745 | sd->nr_balance_failed = 0; |
48f24c4d | 3746 | |
16cfb1c0 | 3747 | return 0; |
1da177e4 LT |
3748 | } |
3749 | ||
3750 | /* | |
3751 | * idle_balance is called by schedule() if this_cpu is about to become | |
3752 | * idle. Attempts to pull tasks from other CPUs. | |
3753 | */ | |
70b97a7f | 3754 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3755 | { |
3756 | struct sched_domain *sd; | |
efbe027e | 3757 | int pulled_task = 0; |
dd41f596 | 3758 | unsigned long next_balance = jiffies + HZ; |
4d2732c6 RR |
3759 | cpumask_var_t tmpmask; |
3760 | ||
3761 | if (!alloc_cpumask_var(&tmpmask, GFP_ATOMIC)) | |
3762 | return; | |
1da177e4 LT |
3763 | |
3764 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3765 | unsigned long interval; |
3766 | ||
3767 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3768 | continue; | |
3769 | ||
3770 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3771 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 3772 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
4d2732c6 | 3773 | sd, tmpmask); |
92c4ca5c CL |
3774 | |
3775 | interval = msecs_to_jiffies(sd->balance_interval); | |
3776 | if (time_after(next_balance, sd->last_balance + interval)) | |
3777 | next_balance = sd->last_balance + interval; | |
3778 | if (pulled_task) | |
3779 | break; | |
1da177e4 | 3780 | } |
dd41f596 | 3781 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3782 | /* |
3783 | * We are going idle. next_balance may be set based on | |
3784 | * a busy processor. So reset next_balance. | |
3785 | */ | |
3786 | this_rq->next_balance = next_balance; | |
dd41f596 | 3787 | } |
4d2732c6 | 3788 | free_cpumask_var(tmpmask); |
1da177e4 LT |
3789 | } |
3790 | ||
3791 | /* | |
3792 | * active_load_balance is run by migration threads. It pushes running tasks | |
3793 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3794 | * running on each physical CPU where possible, and avoids physical / | |
3795 | * logical imbalances. | |
3796 | * | |
3797 | * Called with busiest_rq locked. | |
3798 | */ | |
70b97a7f | 3799 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3800 | { |
39507451 | 3801 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3802 | struct sched_domain *sd; |
3803 | struct rq *target_rq; | |
39507451 | 3804 | |
48f24c4d | 3805 | /* Is there any task to move? */ |
39507451 | 3806 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3807 | return; |
3808 | ||
3809 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3810 | |
3811 | /* | |
39507451 | 3812 | * This condition is "impossible", if it occurs |
41a2d6cf | 3813 | * we need to fix it. Originally reported by |
39507451 | 3814 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3815 | */ |
39507451 | 3816 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3817 | |
39507451 NP |
3818 | /* move a task from busiest_rq to target_rq */ |
3819 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3820 | update_rq_clock(busiest_rq); |
3821 | update_rq_clock(target_rq); | |
39507451 NP |
3822 | |
3823 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3824 | for_each_domain(target_cpu, sd) { |
39507451 | 3825 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 3826 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 3827 | break; |
c96d145e | 3828 | } |
39507451 | 3829 | |
48f24c4d | 3830 | if (likely(sd)) { |
2d72376b | 3831 | schedstat_inc(sd, alb_count); |
39507451 | 3832 | |
43010659 PW |
3833 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3834 | sd, CPU_IDLE)) | |
48f24c4d IM |
3835 | schedstat_inc(sd, alb_pushed); |
3836 | else | |
3837 | schedstat_inc(sd, alb_failed); | |
3838 | } | |
1b12bbc7 | 3839 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3840 | } |
3841 | ||
46cb4b7c SS |
3842 | #ifdef CONFIG_NO_HZ |
3843 | static struct { | |
3844 | atomic_t load_balancer; | |
7d1e6a9b | 3845 | cpumask_var_t cpu_mask; |
46cb4b7c SS |
3846 | } nohz ____cacheline_aligned = { |
3847 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
3848 | }; |
3849 | ||
7835b98b | 3850 | /* |
46cb4b7c SS |
3851 | * This routine will try to nominate the ilb (idle load balancing) |
3852 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3853 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3854 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3855 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3856 | * arrives... | |
3857 | * | |
3858 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3859 | * for idle load balancing. ilb owner will still be part of | |
3860 | * nohz.cpu_mask.. | |
7835b98b | 3861 | * |
46cb4b7c SS |
3862 | * While stopping the tick, this cpu will become the ilb owner if there |
3863 | * is no other owner. And will be the owner till that cpu becomes busy | |
3864 | * or if all cpus in the system stop their ticks at which point | |
3865 | * there is no need for ilb owner. | |
3866 | * | |
3867 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3868 | * next busy scheduler_tick() | |
3869 | */ | |
3870 | int select_nohz_load_balancer(int stop_tick) | |
3871 | { | |
3872 | int cpu = smp_processor_id(); | |
3873 | ||
3874 | if (stop_tick) { | |
7d1e6a9b | 3875 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
3876 | cpu_rq(cpu)->in_nohz_recently = 1; |
3877 | ||
3878 | /* | |
3879 | * If we are going offline and still the leader, give up! | |
3880 | */ | |
e761b772 | 3881 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3882 | atomic_read(&nohz.load_balancer) == cpu) { |
3883 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3884 | BUG(); | |
3885 | return 0; | |
3886 | } | |
3887 | ||
3888 | /* time for ilb owner also to sleep */ | |
7d1e6a9b | 3889 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
3890 | if (atomic_read(&nohz.load_balancer) == cpu) |
3891 | atomic_set(&nohz.load_balancer, -1); | |
3892 | return 0; | |
3893 | } | |
3894 | ||
3895 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3896 | /* make me the ilb owner */ | |
3897 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3898 | return 1; | |
3899 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3900 | return 1; | |
3901 | } else { | |
7d1e6a9b | 3902 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
3903 | return 0; |
3904 | ||
7d1e6a9b | 3905 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
3906 | |
3907 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3908 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3909 | BUG(); | |
3910 | } | |
3911 | return 0; | |
3912 | } | |
3913 | #endif | |
3914 | ||
3915 | static DEFINE_SPINLOCK(balancing); | |
3916 | ||
3917 | /* | |
7835b98b CL |
3918 | * It checks each scheduling domain to see if it is due to be balanced, |
3919 | * and initiates a balancing operation if so. | |
3920 | * | |
3921 | * Balancing parameters are set up in arch_init_sched_domains. | |
3922 | */ | |
a9957449 | 3923 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3924 | { |
46cb4b7c SS |
3925 | int balance = 1; |
3926 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3927 | unsigned long interval; |
3928 | struct sched_domain *sd; | |
46cb4b7c | 3929 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3930 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3931 | int update_next_balance = 0; |
d07355f5 | 3932 | int need_serialize; |
a0e90245 RR |
3933 | cpumask_var_t tmp; |
3934 | ||
3935 | /* Fails alloc? Rebalancing probably not a priority right now. */ | |
3936 | if (!alloc_cpumask_var(&tmp, GFP_ATOMIC)) | |
3937 | return; | |
1da177e4 | 3938 | |
46cb4b7c | 3939 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3940 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3941 | continue; | |
3942 | ||
3943 | interval = sd->balance_interval; | |
d15bcfdb | 3944 | if (idle != CPU_IDLE) |
1da177e4 LT |
3945 | interval *= sd->busy_factor; |
3946 | ||
3947 | /* scale ms to jiffies */ | |
3948 | interval = msecs_to_jiffies(interval); | |
3949 | if (unlikely(!interval)) | |
3950 | interval = 1; | |
dd41f596 IM |
3951 | if (interval > HZ*NR_CPUS/10) |
3952 | interval = HZ*NR_CPUS/10; | |
3953 | ||
d07355f5 | 3954 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3955 | |
d07355f5 | 3956 | if (need_serialize) { |
08c183f3 CL |
3957 | if (!spin_trylock(&balancing)) |
3958 | goto out; | |
3959 | } | |
3960 | ||
c9819f45 | 3961 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
a0e90245 | 3962 | if (load_balance(cpu, rq, sd, idle, &balance, tmp)) { |
fa3b6ddc SS |
3963 | /* |
3964 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3965 | * longer idle, or one of our SMT siblings is |
3966 | * not idle. | |
3967 | */ | |
d15bcfdb | 3968 | idle = CPU_NOT_IDLE; |
1da177e4 | 3969 | } |
1bd77f2d | 3970 | sd->last_balance = jiffies; |
1da177e4 | 3971 | } |
d07355f5 | 3972 | if (need_serialize) |
08c183f3 CL |
3973 | spin_unlock(&balancing); |
3974 | out: | |
f549da84 | 3975 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3976 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3977 | update_next_balance = 1; |
3978 | } | |
783609c6 SS |
3979 | |
3980 | /* | |
3981 | * Stop the load balance at this level. There is another | |
3982 | * CPU in our sched group which is doing load balancing more | |
3983 | * actively. | |
3984 | */ | |
3985 | if (!balance) | |
3986 | break; | |
1da177e4 | 3987 | } |
f549da84 SS |
3988 | |
3989 | /* | |
3990 | * next_balance will be updated only when there is a need. | |
3991 | * When the cpu is attached to null domain for ex, it will not be | |
3992 | * updated. | |
3993 | */ | |
3994 | if (likely(update_next_balance)) | |
3995 | rq->next_balance = next_balance; | |
a0e90245 RR |
3996 | |
3997 | free_cpumask_var(tmp); | |
46cb4b7c SS |
3998 | } |
3999 | ||
4000 | /* | |
4001 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4002 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4003 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4004 | */ | |
4005 | static void run_rebalance_domains(struct softirq_action *h) | |
4006 | { | |
dd41f596 IM |
4007 | int this_cpu = smp_processor_id(); |
4008 | struct rq *this_rq = cpu_rq(this_cpu); | |
4009 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4010 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4011 | |
dd41f596 | 4012 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4013 | |
4014 | #ifdef CONFIG_NO_HZ | |
4015 | /* | |
4016 | * If this cpu is the owner for idle load balancing, then do the | |
4017 | * balancing on behalf of the other idle cpus whose ticks are | |
4018 | * stopped. | |
4019 | */ | |
dd41f596 IM |
4020 | if (this_rq->idle_at_tick && |
4021 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4022 | struct rq *rq; |
4023 | int balance_cpu; | |
4024 | ||
7d1e6a9b RR |
4025 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4026 | if (balance_cpu == this_cpu) | |
4027 | continue; | |
4028 | ||
46cb4b7c SS |
4029 | /* |
4030 | * If this cpu gets work to do, stop the load balancing | |
4031 | * work being done for other cpus. Next load | |
4032 | * balancing owner will pick it up. | |
4033 | */ | |
4034 | if (need_resched()) | |
4035 | break; | |
4036 | ||
de0cf899 | 4037 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4038 | |
4039 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4040 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4041 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4042 | } |
4043 | } | |
4044 | #endif | |
4045 | } | |
4046 | ||
4047 | /* | |
4048 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4049 | * | |
4050 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4051 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4052 | * if the whole system is idle. | |
4053 | */ | |
dd41f596 | 4054 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4055 | { |
46cb4b7c SS |
4056 | #ifdef CONFIG_NO_HZ |
4057 | /* | |
4058 | * If we were in the nohz mode recently and busy at the current | |
4059 | * scheduler tick, then check if we need to nominate new idle | |
4060 | * load balancer. | |
4061 | */ | |
4062 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4063 | rq->in_nohz_recently = 0; | |
4064 | ||
4065 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4066 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4067 | atomic_set(&nohz.load_balancer, -1); |
4068 | } | |
4069 | ||
4070 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4071 | /* | |
4072 | * simple selection for now: Nominate the | |
4073 | * first cpu in the nohz list to be the next | |
4074 | * ilb owner. | |
4075 | * | |
4076 | * TBD: Traverse the sched domains and nominate | |
4077 | * the nearest cpu in the nohz.cpu_mask. | |
4078 | */ | |
7d1e6a9b | 4079 | int ilb = cpumask_first(nohz.cpu_mask); |
46cb4b7c | 4080 | |
434d53b0 | 4081 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4082 | resched_cpu(ilb); |
4083 | } | |
4084 | } | |
4085 | ||
4086 | /* | |
4087 | * If this cpu is idle and doing idle load balancing for all the | |
4088 | * cpus with ticks stopped, is it time for that to stop? | |
4089 | */ | |
4090 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4091 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4092 | resched_cpu(cpu); |
4093 | return; | |
4094 | } | |
4095 | ||
4096 | /* | |
4097 | * If this cpu is idle and the idle load balancing is done by | |
4098 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4099 | */ | |
4100 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4101 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4102 | return; |
4103 | #endif | |
4104 | if (time_after_eq(jiffies, rq->next_balance)) | |
4105 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4106 | } |
dd41f596 IM |
4107 | |
4108 | #else /* CONFIG_SMP */ | |
4109 | ||
1da177e4 LT |
4110 | /* |
4111 | * on UP we do not need to balance between CPUs: | |
4112 | */ | |
70b97a7f | 4113 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4114 | { |
4115 | } | |
dd41f596 | 4116 | |
1da177e4 LT |
4117 | #endif |
4118 | ||
1da177e4 LT |
4119 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4120 | ||
4121 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4122 | ||
4123 | /* | |
f06febc9 FM |
4124 | * Return any ns on the sched_clock that have not yet been banked in |
4125 | * @p in case that task is currently running. | |
1da177e4 | 4126 | */ |
bb34d92f | 4127 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4128 | { |
1da177e4 | 4129 | unsigned long flags; |
41b86e9c | 4130 | struct rq *rq; |
bb34d92f | 4131 | u64 ns = 0; |
48f24c4d | 4132 | |
41b86e9c | 4133 | rq = task_rq_lock(p, &flags); |
1508487e | 4134 | |
051a1d1a | 4135 | if (task_current(rq, p)) { |
f06febc9 FM |
4136 | u64 delta_exec; |
4137 | ||
a8e504d2 IM |
4138 | update_rq_clock(rq); |
4139 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c | 4140 | if ((s64)delta_exec > 0) |
bb34d92f | 4141 | ns = delta_exec; |
41b86e9c | 4142 | } |
48f24c4d | 4143 | |
41b86e9c | 4144 | task_rq_unlock(rq, &flags); |
48f24c4d | 4145 | |
1da177e4 LT |
4146 | return ns; |
4147 | } | |
4148 | ||
1da177e4 LT |
4149 | /* |
4150 | * Account user cpu time to a process. | |
4151 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
4152 | * @cputime: the cpu time spent in user space since the last update |
4153 | */ | |
4154 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
4155 | { | |
4156 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4157 | cputime64_t tmp; | |
4158 | ||
4159 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4160 | account_group_user_time(p, cputime); |
1da177e4 LT |
4161 | |
4162 | /* Add user time to cpustat. */ | |
4163 | tmp = cputime_to_cputime64(cputime); | |
4164 | if (TASK_NICE(p) > 0) | |
4165 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4166 | else | |
4167 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4168 | /* Account for user time used */ |
4169 | acct_update_integrals(p); | |
1da177e4 LT |
4170 | } |
4171 | ||
94886b84 LV |
4172 | /* |
4173 | * Account guest cpu time to a process. | |
4174 | * @p: the process that the cpu time gets accounted to | |
4175 | * @cputime: the cpu time spent in virtual machine since the last update | |
4176 | */ | |
f7402e03 | 4177 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
4178 | { |
4179 | cputime64_t tmp; | |
4180 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4181 | ||
4182 | tmp = cputime_to_cputime64(cputime); | |
4183 | ||
4184 | p->utime = cputime_add(p->utime, cputime); | |
f06febc9 | 4185 | account_group_user_time(p, cputime); |
94886b84 LV |
4186 | p->gtime = cputime_add(p->gtime, cputime); |
4187 | ||
4188 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
4189 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4190 | } | |
4191 | ||
c66f08be MN |
4192 | /* |
4193 | * Account scaled user cpu time to a process. | |
4194 | * @p: the process that the cpu time gets accounted to | |
4195 | * @cputime: the cpu time spent in user space since the last update | |
4196 | */ | |
4197 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
4198 | { | |
4199 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
4200 | } | |
4201 | ||
1da177e4 LT |
4202 | /* |
4203 | * Account system cpu time to a process. | |
4204 | * @p: the process that the cpu time gets accounted to | |
4205 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4206 | * @cputime: the cpu time spent in kernel space since the last update | |
4207 | */ | |
4208 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
4209 | cputime_t cputime) | |
4210 | { | |
4211 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 4212 | struct rq *rq = this_rq(); |
1da177e4 LT |
4213 | cputime64_t tmp; |
4214 | ||
983ed7a6 HH |
4215 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
4216 | account_guest_time(p, cputime); | |
4217 | return; | |
4218 | } | |
94886b84 | 4219 | |
1da177e4 | 4220 | p->stime = cputime_add(p->stime, cputime); |
f06febc9 | 4221 | account_group_system_time(p, cputime); |
1da177e4 LT |
4222 | |
4223 | /* Add system time to cpustat. */ | |
4224 | tmp = cputime_to_cputime64(cputime); | |
4225 | if (hardirq_count() - hardirq_offset) | |
4226 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4227 | else if (softirq_count()) | |
4228 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 4229 | else if (p != rq->idle) |
1da177e4 | 4230 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 4231 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
4232 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
4233 | else | |
4234 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
4235 | /* Account for system time used */ | |
4236 | acct_update_integrals(p); | |
1da177e4 LT |
4237 | } |
4238 | ||
c66f08be MN |
4239 | /* |
4240 | * Account scaled system cpu time to a process. | |
4241 | * @p: the process that the cpu time gets accounted to | |
4242 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4243 | * @cputime: the cpu time spent in kernel space since the last update | |
4244 | */ | |
4245 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4246 | { | |
4247 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4248 | } | |
4249 | ||
1da177e4 LT |
4250 | /* |
4251 | * Account for involuntary wait time. | |
4252 | * @p: the process from which the cpu time has been stolen | |
4253 | * @steal: the cpu time spent in involuntary wait | |
4254 | */ | |
4255 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4256 | { | |
4257 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4258 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4259 | struct rq *rq = this_rq(); |
1da177e4 LT |
4260 | |
4261 | if (p == rq->idle) { | |
4262 | p->stime = cputime_add(p->stime, steal); | |
4263 | if (atomic_read(&rq->nr_iowait) > 0) | |
4264 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4265 | else | |
4266 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4267 | } else |
1da177e4 LT |
4268 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4269 | } | |
4270 | ||
49048622 BS |
4271 | /* |
4272 | * Use precise platform statistics if available: | |
4273 | */ | |
4274 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4275 | cputime_t task_utime(struct task_struct *p) | |
4276 | { | |
4277 | return p->utime; | |
4278 | } | |
4279 | ||
4280 | cputime_t task_stime(struct task_struct *p) | |
4281 | { | |
4282 | return p->stime; | |
4283 | } | |
4284 | #else | |
4285 | cputime_t task_utime(struct task_struct *p) | |
4286 | { | |
4287 | clock_t utime = cputime_to_clock_t(p->utime), | |
4288 | total = utime + cputime_to_clock_t(p->stime); | |
4289 | u64 temp; | |
4290 | ||
4291 | /* | |
4292 | * Use CFS's precise accounting: | |
4293 | */ | |
4294 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4295 | ||
4296 | if (total) { | |
4297 | temp *= utime; | |
4298 | do_div(temp, total); | |
4299 | } | |
4300 | utime = (clock_t)temp; | |
4301 | ||
4302 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4303 | return p->prev_utime; | |
4304 | } | |
4305 | ||
4306 | cputime_t task_stime(struct task_struct *p) | |
4307 | { | |
4308 | clock_t stime; | |
4309 | ||
4310 | /* | |
4311 | * Use CFS's precise accounting. (we subtract utime from | |
4312 | * the total, to make sure the total observed by userspace | |
4313 | * grows monotonically - apps rely on that): | |
4314 | */ | |
4315 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4316 | cputime_to_clock_t(task_utime(p)); | |
4317 | ||
4318 | if (stime >= 0) | |
4319 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4320 | ||
4321 | return p->prev_stime; | |
4322 | } | |
4323 | #endif | |
4324 | ||
4325 | inline cputime_t task_gtime(struct task_struct *p) | |
4326 | { | |
4327 | return p->gtime; | |
4328 | } | |
4329 | ||
7835b98b CL |
4330 | /* |
4331 | * This function gets called by the timer code, with HZ frequency. | |
4332 | * We call it with interrupts disabled. | |
4333 | * | |
4334 | * It also gets called by the fork code, when changing the parent's | |
4335 | * timeslices. | |
4336 | */ | |
4337 | void scheduler_tick(void) | |
4338 | { | |
7835b98b CL |
4339 | int cpu = smp_processor_id(); |
4340 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4341 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4342 | |
4343 | sched_clock_tick(); | |
dd41f596 IM |
4344 | |
4345 | spin_lock(&rq->lock); | |
3e51f33f | 4346 | update_rq_clock(rq); |
f1a438d8 | 4347 | update_cpu_load(rq); |
fa85ae24 | 4348 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4349 | spin_unlock(&rq->lock); |
7835b98b | 4350 | |
e418e1c2 | 4351 | #ifdef CONFIG_SMP |
dd41f596 IM |
4352 | rq->idle_at_tick = idle_cpu(cpu); |
4353 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4354 | #endif |
1da177e4 LT |
4355 | } |
4356 | ||
6cd8a4bb SR |
4357 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4358 | defined(CONFIG_PREEMPT_TRACER)) | |
4359 | ||
4360 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4361 | { | |
4362 | if (in_lock_functions(addr)) { | |
4363 | addr = CALLER_ADDR2; | |
4364 | if (in_lock_functions(addr)) | |
4365 | addr = CALLER_ADDR3; | |
4366 | } | |
4367 | return addr; | |
4368 | } | |
1da177e4 | 4369 | |
43627582 | 4370 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4371 | { |
6cd8a4bb | 4372 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4373 | /* |
4374 | * Underflow? | |
4375 | */ | |
9a11b49a IM |
4376 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4377 | return; | |
6cd8a4bb | 4378 | #endif |
1da177e4 | 4379 | preempt_count() += val; |
6cd8a4bb | 4380 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4381 | /* |
4382 | * Spinlock count overflowing soon? | |
4383 | */ | |
33859f7f MOS |
4384 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4385 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4386 | #endif |
4387 | if (preempt_count() == val) | |
4388 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4389 | } |
4390 | EXPORT_SYMBOL(add_preempt_count); | |
4391 | ||
43627582 | 4392 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4393 | { |
6cd8a4bb | 4394 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4395 | /* |
4396 | * Underflow? | |
4397 | */ | |
7317d7b8 | 4398 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count() - (!!kernel_locked()))) |
9a11b49a | 4399 | return; |
1da177e4 LT |
4400 | /* |
4401 | * Is the spinlock portion underflowing? | |
4402 | */ | |
9a11b49a IM |
4403 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4404 | !(preempt_count() & PREEMPT_MASK))) | |
4405 | return; | |
6cd8a4bb | 4406 | #endif |
9a11b49a | 4407 | |
6cd8a4bb SR |
4408 | if (preempt_count() == val) |
4409 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4410 | preempt_count() -= val; |
4411 | } | |
4412 | EXPORT_SYMBOL(sub_preempt_count); | |
4413 | ||
4414 | #endif | |
4415 | ||
4416 | /* | |
dd41f596 | 4417 | * Print scheduling while atomic bug: |
1da177e4 | 4418 | */ |
dd41f596 | 4419 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4420 | { |
838225b4 SS |
4421 | struct pt_regs *regs = get_irq_regs(); |
4422 | ||
4423 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4424 | prev->comm, prev->pid, preempt_count()); | |
4425 | ||
dd41f596 | 4426 | debug_show_held_locks(prev); |
e21f5b15 | 4427 | print_modules(); |
dd41f596 IM |
4428 | if (irqs_disabled()) |
4429 | print_irqtrace_events(prev); | |
838225b4 SS |
4430 | |
4431 | if (regs) | |
4432 | show_regs(regs); | |
4433 | else | |
4434 | dump_stack(); | |
dd41f596 | 4435 | } |
1da177e4 | 4436 | |
dd41f596 IM |
4437 | /* |
4438 | * Various schedule()-time debugging checks and statistics: | |
4439 | */ | |
4440 | static inline void schedule_debug(struct task_struct *prev) | |
4441 | { | |
1da177e4 | 4442 | /* |
41a2d6cf | 4443 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4444 | * schedule() atomically, we ignore that path for now. |
4445 | * Otherwise, whine if we are scheduling when we should not be. | |
4446 | */ | |
3f33a7ce | 4447 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4448 | __schedule_bug(prev); |
4449 | ||
1da177e4 LT |
4450 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4451 | ||
2d72376b | 4452 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4453 | #ifdef CONFIG_SCHEDSTATS |
4454 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4455 | schedstat_inc(this_rq(), bkl_count); |
4456 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4457 | } |
4458 | #endif | |
dd41f596 IM |
4459 | } |
4460 | ||
4461 | /* | |
4462 | * Pick up the highest-prio task: | |
4463 | */ | |
4464 | static inline struct task_struct * | |
ff95f3df | 4465 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4466 | { |
5522d5d5 | 4467 | const struct sched_class *class; |
dd41f596 | 4468 | struct task_struct *p; |
1da177e4 LT |
4469 | |
4470 | /* | |
dd41f596 IM |
4471 | * Optimization: we know that if all tasks are in |
4472 | * the fair class we can call that function directly: | |
1da177e4 | 4473 | */ |
dd41f596 | 4474 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4475 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4476 | if (likely(p)) |
4477 | return p; | |
1da177e4 LT |
4478 | } |
4479 | ||
dd41f596 IM |
4480 | class = sched_class_highest; |
4481 | for ( ; ; ) { | |
fb8d4724 | 4482 | p = class->pick_next_task(rq); |
dd41f596 IM |
4483 | if (p) |
4484 | return p; | |
4485 | /* | |
4486 | * Will never be NULL as the idle class always | |
4487 | * returns a non-NULL p: | |
4488 | */ | |
4489 | class = class->next; | |
4490 | } | |
4491 | } | |
1da177e4 | 4492 | |
dd41f596 IM |
4493 | /* |
4494 | * schedule() is the main scheduler function. | |
4495 | */ | |
4496 | asmlinkage void __sched schedule(void) | |
4497 | { | |
4498 | struct task_struct *prev, *next; | |
67ca7bde | 4499 | unsigned long *switch_count; |
dd41f596 | 4500 | struct rq *rq; |
31656519 | 4501 | int cpu; |
dd41f596 IM |
4502 | |
4503 | need_resched: | |
4504 | preempt_disable(); | |
4505 | cpu = smp_processor_id(); | |
4506 | rq = cpu_rq(cpu); | |
4507 | rcu_qsctr_inc(cpu); | |
4508 | prev = rq->curr; | |
4509 | switch_count = &prev->nivcsw; | |
4510 | ||
4511 | release_kernel_lock(prev); | |
4512 | need_resched_nonpreemptible: | |
4513 | ||
4514 | schedule_debug(prev); | |
1da177e4 | 4515 | |
31656519 | 4516 | if (sched_feat(HRTICK)) |
f333fdc9 | 4517 | hrtick_clear(rq); |
8f4d37ec | 4518 | |
8cd162ce | 4519 | spin_lock_irq(&rq->lock); |
3e51f33f | 4520 | update_rq_clock(rq); |
1e819950 | 4521 | clear_tsk_need_resched(prev); |
1da177e4 | 4522 | |
1da177e4 | 4523 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4524 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4525 | prev->state = TASK_RUNNING; |
16882c1e | 4526 | else |
2e1cb74a | 4527 | deactivate_task(rq, prev, 1); |
dd41f596 | 4528 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4529 | } |
4530 | ||
9a897c5a SR |
4531 | #ifdef CONFIG_SMP |
4532 | if (prev->sched_class->pre_schedule) | |
4533 | prev->sched_class->pre_schedule(rq, prev); | |
4534 | #endif | |
f65eda4f | 4535 | |
dd41f596 | 4536 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4537 | idle_balance(cpu, rq); |
1da177e4 | 4538 | |
31ee529c | 4539 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4540 | next = pick_next_task(rq, prev); |
1da177e4 | 4541 | |
1da177e4 | 4542 | if (likely(prev != next)) { |
673a90a1 DS |
4543 | sched_info_switch(prev, next); |
4544 | ||
1da177e4 LT |
4545 | rq->nr_switches++; |
4546 | rq->curr = next; | |
4547 | ++*switch_count; | |
4548 | ||
dd41f596 | 4549 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4550 | /* |
4551 | * the context switch might have flipped the stack from under | |
4552 | * us, hence refresh the local variables. | |
4553 | */ | |
4554 | cpu = smp_processor_id(); | |
4555 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4556 | } else |
4557 | spin_unlock_irq(&rq->lock); | |
4558 | ||
8f4d37ec | 4559 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4560 | goto need_resched_nonpreemptible; |
8f4d37ec | 4561 | |
1da177e4 LT |
4562 | preempt_enable_no_resched(); |
4563 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4564 | goto need_resched; | |
4565 | } | |
1da177e4 LT |
4566 | EXPORT_SYMBOL(schedule); |
4567 | ||
4568 | #ifdef CONFIG_PREEMPT | |
4569 | /* | |
2ed6e34f | 4570 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4571 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4572 | * occur there and call schedule directly. |
4573 | */ | |
4574 | asmlinkage void __sched preempt_schedule(void) | |
4575 | { | |
4576 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4577 | |
1da177e4 LT |
4578 | /* |
4579 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4580 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4581 | */ |
beed33a8 | 4582 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4583 | return; |
4584 | ||
3a5c359a AK |
4585 | do { |
4586 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4587 | schedule(); |
3a5c359a | 4588 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4589 | |
3a5c359a AK |
4590 | /* |
4591 | * Check again in case we missed a preemption opportunity | |
4592 | * between schedule and now. | |
4593 | */ | |
4594 | barrier(); | |
4595 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4596 | } |
1da177e4 LT |
4597 | EXPORT_SYMBOL(preempt_schedule); |
4598 | ||
4599 | /* | |
2ed6e34f | 4600 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4601 | * off of irq context. |
4602 | * Note, that this is called and return with irqs disabled. This will | |
4603 | * protect us against recursive calling from irq. | |
4604 | */ | |
4605 | asmlinkage void __sched preempt_schedule_irq(void) | |
4606 | { | |
4607 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4608 | |
2ed6e34f | 4609 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4610 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4611 | ||
3a5c359a AK |
4612 | do { |
4613 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4614 | local_irq_enable(); |
4615 | schedule(); | |
4616 | local_irq_disable(); | |
3a5c359a | 4617 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4618 | |
3a5c359a AK |
4619 | /* |
4620 | * Check again in case we missed a preemption opportunity | |
4621 | * between schedule and now. | |
4622 | */ | |
4623 | barrier(); | |
4624 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4625 | } |
4626 | ||
4627 | #endif /* CONFIG_PREEMPT */ | |
4628 | ||
95cdf3b7 IM |
4629 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4630 | void *key) | |
1da177e4 | 4631 | { |
48f24c4d | 4632 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4633 | } |
1da177e4 LT |
4634 | EXPORT_SYMBOL(default_wake_function); |
4635 | ||
4636 | /* | |
41a2d6cf IM |
4637 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4638 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4639 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4640 | * | |
4641 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4642 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4643 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4644 | */ | |
4645 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4646 | int nr_exclusive, int sync, void *key) | |
4647 | { | |
2e45874c | 4648 | wait_queue_t *curr, *next; |
1da177e4 | 4649 | |
2e45874c | 4650 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4651 | unsigned flags = curr->flags; |
4652 | ||
1da177e4 | 4653 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4654 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4655 | break; |
4656 | } | |
4657 | } | |
4658 | ||
4659 | /** | |
4660 | * __wake_up - wake up threads blocked on a waitqueue. | |
4661 | * @q: the waitqueue | |
4662 | * @mode: which threads | |
4663 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4664 | * @key: is directly passed to the wakeup function |
1da177e4 | 4665 | */ |
7ad5b3a5 | 4666 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4667 | int nr_exclusive, void *key) |
1da177e4 LT |
4668 | { |
4669 | unsigned long flags; | |
4670 | ||
4671 | spin_lock_irqsave(&q->lock, flags); | |
4672 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4673 | spin_unlock_irqrestore(&q->lock, flags); | |
4674 | } | |
1da177e4 LT |
4675 | EXPORT_SYMBOL(__wake_up); |
4676 | ||
4677 | /* | |
4678 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4679 | */ | |
7ad5b3a5 | 4680 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4681 | { |
4682 | __wake_up_common(q, mode, 1, 0, NULL); | |
4683 | } | |
4684 | ||
4685 | /** | |
67be2dd1 | 4686 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4687 | * @q: the waitqueue |
4688 | * @mode: which threads | |
4689 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4690 | * | |
4691 | * The sync wakeup differs that the waker knows that it will schedule | |
4692 | * away soon, so while the target thread will be woken up, it will not | |
4693 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4694 | * with each other. This can prevent needless bouncing between CPUs. | |
4695 | * | |
4696 | * On UP it can prevent extra preemption. | |
4697 | */ | |
7ad5b3a5 | 4698 | void |
95cdf3b7 | 4699 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4700 | { |
4701 | unsigned long flags; | |
4702 | int sync = 1; | |
4703 | ||
4704 | if (unlikely(!q)) | |
4705 | return; | |
4706 | ||
4707 | if (unlikely(!nr_exclusive)) | |
4708 | sync = 0; | |
4709 | ||
4710 | spin_lock_irqsave(&q->lock, flags); | |
4711 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4712 | spin_unlock_irqrestore(&q->lock, flags); | |
4713 | } | |
4714 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4715 | ||
65eb3dc6 KD |
4716 | /** |
4717 | * complete: - signals a single thread waiting on this completion | |
4718 | * @x: holds the state of this particular completion | |
4719 | * | |
4720 | * This will wake up a single thread waiting on this completion. Threads will be | |
4721 | * awakened in the same order in which they were queued. | |
4722 | * | |
4723 | * See also complete_all(), wait_for_completion() and related routines. | |
4724 | */ | |
b15136e9 | 4725 | void complete(struct completion *x) |
1da177e4 LT |
4726 | { |
4727 | unsigned long flags; | |
4728 | ||
4729 | spin_lock_irqsave(&x->wait.lock, flags); | |
4730 | x->done++; | |
d9514f6c | 4731 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4732 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4733 | } | |
4734 | EXPORT_SYMBOL(complete); | |
4735 | ||
65eb3dc6 KD |
4736 | /** |
4737 | * complete_all: - signals all threads waiting on this completion | |
4738 | * @x: holds the state of this particular completion | |
4739 | * | |
4740 | * This will wake up all threads waiting on this particular completion event. | |
4741 | */ | |
b15136e9 | 4742 | void complete_all(struct completion *x) |
1da177e4 LT |
4743 | { |
4744 | unsigned long flags; | |
4745 | ||
4746 | spin_lock_irqsave(&x->wait.lock, flags); | |
4747 | x->done += UINT_MAX/2; | |
d9514f6c | 4748 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4749 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4750 | } | |
4751 | EXPORT_SYMBOL(complete_all); | |
4752 | ||
8cbbe86d AK |
4753 | static inline long __sched |
4754 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4755 | { |
1da177e4 LT |
4756 | if (!x->done) { |
4757 | DECLARE_WAITQUEUE(wait, current); | |
4758 | ||
4759 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4760 | __add_wait_queue_tail(&x->wait, &wait); | |
4761 | do { | |
94d3d824 | 4762 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4763 | timeout = -ERESTARTSYS; |
4764 | break; | |
8cbbe86d AK |
4765 | } |
4766 | __set_current_state(state); | |
1da177e4 LT |
4767 | spin_unlock_irq(&x->wait.lock); |
4768 | timeout = schedule_timeout(timeout); | |
4769 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4770 | } while (!x->done && timeout); |
1da177e4 | 4771 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4772 | if (!x->done) |
4773 | return timeout; | |
1da177e4 LT |
4774 | } |
4775 | x->done--; | |
ea71a546 | 4776 | return timeout ?: 1; |
1da177e4 | 4777 | } |
1da177e4 | 4778 | |
8cbbe86d AK |
4779 | static long __sched |
4780 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4781 | { |
1da177e4 LT |
4782 | might_sleep(); |
4783 | ||
4784 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4785 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4786 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4787 | return timeout; |
4788 | } | |
1da177e4 | 4789 | |
65eb3dc6 KD |
4790 | /** |
4791 | * wait_for_completion: - waits for completion of a task | |
4792 | * @x: holds the state of this particular completion | |
4793 | * | |
4794 | * This waits to be signaled for completion of a specific task. It is NOT | |
4795 | * interruptible and there is no timeout. | |
4796 | * | |
4797 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4798 | * and interrupt capability. Also see complete(). | |
4799 | */ | |
b15136e9 | 4800 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4801 | { |
4802 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4803 | } |
8cbbe86d | 4804 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4805 | |
65eb3dc6 KD |
4806 | /** |
4807 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4808 | * @x: holds the state of this particular completion | |
4809 | * @timeout: timeout value in jiffies | |
4810 | * | |
4811 | * This waits for either a completion of a specific task to be signaled or for a | |
4812 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4813 | * interruptible. | |
4814 | */ | |
b15136e9 | 4815 | unsigned long __sched |
8cbbe86d | 4816 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4817 | { |
8cbbe86d | 4818 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4819 | } |
8cbbe86d | 4820 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4821 | |
65eb3dc6 KD |
4822 | /** |
4823 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4824 | * @x: holds the state of this particular completion | |
4825 | * | |
4826 | * This waits for completion of a specific task to be signaled. It is | |
4827 | * interruptible. | |
4828 | */ | |
8cbbe86d | 4829 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4830 | { |
51e97990 AK |
4831 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4832 | if (t == -ERESTARTSYS) | |
4833 | return t; | |
4834 | return 0; | |
0fec171c | 4835 | } |
8cbbe86d | 4836 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4837 | |
65eb3dc6 KD |
4838 | /** |
4839 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4840 | * @x: holds the state of this particular completion | |
4841 | * @timeout: timeout value in jiffies | |
4842 | * | |
4843 | * This waits for either a completion of a specific task to be signaled or for a | |
4844 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4845 | */ | |
b15136e9 | 4846 | unsigned long __sched |
8cbbe86d AK |
4847 | wait_for_completion_interruptible_timeout(struct completion *x, |
4848 | unsigned long timeout) | |
0fec171c | 4849 | { |
8cbbe86d | 4850 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4851 | } |
8cbbe86d | 4852 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4853 | |
65eb3dc6 KD |
4854 | /** |
4855 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4856 | * @x: holds the state of this particular completion | |
4857 | * | |
4858 | * This waits to be signaled for completion of a specific task. It can be | |
4859 | * interrupted by a kill signal. | |
4860 | */ | |
009e577e MW |
4861 | int __sched wait_for_completion_killable(struct completion *x) |
4862 | { | |
4863 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4864 | if (t == -ERESTARTSYS) | |
4865 | return t; | |
4866 | return 0; | |
4867 | } | |
4868 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4869 | ||
be4de352 DC |
4870 | /** |
4871 | * try_wait_for_completion - try to decrement a completion without blocking | |
4872 | * @x: completion structure | |
4873 | * | |
4874 | * Returns: 0 if a decrement cannot be done without blocking | |
4875 | * 1 if a decrement succeeded. | |
4876 | * | |
4877 | * If a completion is being used as a counting completion, | |
4878 | * attempt to decrement the counter without blocking. This | |
4879 | * enables us to avoid waiting if the resource the completion | |
4880 | * is protecting is not available. | |
4881 | */ | |
4882 | bool try_wait_for_completion(struct completion *x) | |
4883 | { | |
4884 | int ret = 1; | |
4885 | ||
4886 | spin_lock_irq(&x->wait.lock); | |
4887 | if (!x->done) | |
4888 | ret = 0; | |
4889 | else | |
4890 | x->done--; | |
4891 | spin_unlock_irq(&x->wait.lock); | |
4892 | return ret; | |
4893 | } | |
4894 | EXPORT_SYMBOL(try_wait_for_completion); | |
4895 | ||
4896 | /** | |
4897 | * completion_done - Test to see if a completion has any waiters | |
4898 | * @x: completion structure | |
4899 | * | |
4900 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4901 | * 1 if there are no waiters. | |
4902 | * | |
4903 | */ | |
4904 | bool completion_done(struct completion *x) | |
4905 | { | |
4906 | int ret = 1; | |
4907 | ||
4908 | spin_lock_irq(&x->wait.lock); | |
4909 | if (!x->done) | |
4910 | ret = 0; | |
4911 | spin_unlock_irq(&x->wait.lock); | |
4912 | return ret; | |
4913 | } | |
4914 | EXPORT_SYMBOL(completion_done); | |
4915 | ||
8cbbe86d AK |
4916 | static long __sched |
4917 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4918 | { |
0fec171c IM |
4919 | unsigned long flags; |
4920 | wait_queue_t wait; | |
4921 | ||
4922 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4923 | |
8cbbe86d | 4924 | __set_current_state(state); |
1da177e4 | 4925 | |
8cbbe86d AK |
4926 | spin_lock_irqsave(&q->lock, flags); |
4927 | __add_wait_queue(q, &wait); | |
4928 | spin_unlock(&q->lock); | |
4929 | timeout = schedule_timeout(timeout); | |
4930 | spin_lock_irq(&q->lock); | |
4931 | __remove_wait_queue(q, &wait); | |
4932 | spin_unlock_irqrestore(&q->lock, flags); | |
4933 | ||
4934 | return timeout; | |
4935 | } | |
4936 | ||
4937 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4938 | { | |
4939 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4940 | } |
1da177e4 LT |
4941 | EXPORT_SYMBOL(interruptible_sleep_on); |
4942 | ||
0fec171c | 4943 | long __sched |
95cdf3b7 | 4944 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4945 | { |
8cbbe86d | 4946 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4947 | } |
1da177e4 LT |
4948 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4949 | ||
0fec171c | 4950 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4951 | { |
8cbbe86d | 4952 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4953 | } |
1da177e4 LT |
4954 | EXPORT_SYMBOL(sleep_on); |
4955 | ||
0fec171c | 4956 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4957 | { |
8cbbe86d | 4958 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4959 | } |
1da177e4 LT |
4960 | EXPORT_SYMBOL(sleep_on_timeout); |
4961 | ||
b29739f9 IM |
4962 | #ifdef CONFIG_RT_MUTEXES |
4963 | ||
4964 | /* | |
4965 | * rt_mutex_setprio - set the current priority of a task | |
4966 | * @p: task | |
4967 | * @prio: prio value (kernel-internal form) | |
4968 | * | |
4969 | * This function changes the 'effective' priority of a task. It does | |
4970 | * not touch ->normal_prio like __setscheduler(). | |
4971 | * | |
4972 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4973 | */ | |
36c8b586 | 4974 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4975 | { |
4976 | unsigned long flags; | |
83b699ed | 4977 | int oldprio, on_rq, running; |
70b97a7f | 4978 | struct rq *rq; |
cb469845 | 4979 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4980 | |
4981 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4982 | ||
4983 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4984 | update_rq_clock(rq); |
b29739f9 | 4985 | |
d5f9f942 | 4986 | oldprio = p->prio; |
dd41f596 | 4987 | on_rq = p->se.on_rq; |
051a1d1a | 4988 | running = task_current(rq, p); |
0e1f3483 | 4989 | if (on_rq) |
69be72c1 | 4990 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4991 | if (running) |
4992 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4993 | |
4994 | if (rt_prio(prio)) | |
4995 | p->sched_class = &rt_sched_class; | |
4996 | else | |
4997 | p->sched_class = &fair_sched_class; | |
4998 | ||
b29739f9 IM |
4999 | p->prio = prio; |
5000 | ||
0e1f3483 HS |
5001 | if (running) |
5002 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5003 | if (on_rq) { |
8159f87e | 5004 | enqueue_task(rq, p, 0); |
cb469845 SR |
5005 | |
5006 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5007 | } |
5008 | task_rq_unlock(rq, &flags); | |
5009 | } | |
5010 | ||
5011 | #endif | |
5012 | ||
36c8b586 | 5013 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5014 | { |
dd41f596 | 5015 | int old_prio, delta, on_rq; |
1da177e4 | 5016 | unsigned long flags; |
70b97a7f | 5017 | struct rq *rq; |
1da177e4 LT |
5018 | |
5019 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5020 | return; | |
5021 | /* | |
5022 | * We have to be careful, if called from sys_setpriority(), | |
5023 | * the task might be in the middle of scheduling on another CPU. | |
5024 | */ | |
5025 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5026 | update_rq_clock(rq); |
1da177e4 LT |
5027 | /* |
5028 | * The RT priorities are set via sched_setscheduler(), but we still | |
5029 | * allow the 'normal' nice value to be set - but as expected | |
5030 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5031 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5032 | */ |
e05606d3 | 5033 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5034 | p->static_prio = NICE_TO_PRIO(nice); |
5035 | goto out_unlock; | |
5036 | } | |
dd41f596 | 5037 | on_rq = p->se.on_rq; |
c09595f6 | 5038 | if (on_rq) |
69be72c1 | 5039 | dequeue_task(rq, p, 0); |
1da177e4 | 5040 | |
1da177e4 | 5041 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5042 | set_load_weight(p); |
b29739f9 IM |
5043 | old_prio = p->prio; |
5044 | p->prio = effective_prio(p); | |
5045 | delta = p->prio - old_prio; | |
1da177e4 | 5046 | |
dd41f596 | 5047 | if (on_rq) { |
8159f87e | 5048 | enqueue_task(rq, p, 0); |
1da177e4 | 5049 | /* |
d5f9f942 AM |
5050 | * If the task increased its priority or is running and |
5051 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5052 | */ |
d5f9f942 | 5053 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5054 | resched_task(rq->curr); |
5055 | } | |
5056 | out_unlock: | |
5057 | task_rq_unlock(rq, &flags); | |
5058 | } | |
1da177e4 LT |
5059 | EXPORT_SYMBOL(set_user_nice); |
5060 | ||
e43379f1 MM |
5061 | /* |
5062 | * can_nice - check if a task can reduce its nice value | |
5063 | * @p: task | |
5064 | * @nice: nice value | |
5065 | */ | |
36c8b586 | 5066 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5067 | { |
024f4747 MM |
5068 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5069 | int nice_rlim = 20 - nice; | |
48f24c4d | 5070 | |
e43379f1 MM |
5071 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5072 | capable(CAP_SYS_NICE)); | |
5073 | } | |
5074 | ||
1da177e4 LT |
5075 | #ifdef __ARCH_WANT_SYS_NICE |
5076 | ||
5077 | /* | |
5078 | * sys_nice - change the priority of the current process. | |
5079 | * @increment: priority increment | |
5080 | * | |
5081 | * sys_setpriority is a more generic, but much slower function that | |
5082 | * does similar things. | |
5083 | */ | |
5084 | asmlinkage long sys_nice(int increment) | |
5085 | { | |
48f24c4d | 5086 | long nice, retval; |
1da177e4 LT |
5087 | |
5088 | /* | |
5089 | * Setpriority might change our priority at the same moment. | |
5090 | * We don't have to worry. Conceptually one call occurs first | |
5091 | * and we have a single winner. | |
5092 | */ | |
e43379f1 MM |
5093 | if (increment < -40) |
5094 | increment = -40; | |
1da177e4 LT |
5095 | if (increment > 40) |
5096 | increment = 40; | |
5097 | ||
5098 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
5099 | if (nice < -20) | |
5100 | nice = -20; | |
5101 | if (nice > 19) | |
5102 | nice = 19; | |
5103 | ||
e43379f1 MM |
5104 | if (increment < 0 && !can_nice(current, nice)) |
5105 | return -EPERM; | |
5106 | ||
1da177e4 LT |
5107 | retval = security_task_setnice(current, nice); |
5108 | if (retval) | |
5109 | return retval; | |
5110 | ||
5111 | set_user_nice(current, nice); | |
5112 | return 0; | |
5113 | } | |
5114 | ||
5115 | #endif | |
5116 | ||
5117 | /** | |
5118 | * task_prio - return the priority value of a given task. | |
5119 | * @p: the task in question. | |
5120 | * | |
5121 | * This is the priority value as seen by users in /proc. | |
5122 | * RT tasks are offset by -200. Normal tasks are centered | |
5123 | * around 0, value goes from -16 to +15. | |
5124 | */ | |
36c8b586 | 5125 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5126 | { |
5127 | return p->prio - MAX_RT_PRIO; | |
5128 | } | |
5129 | ||
5130 | /** | |
5131 | * task_nice - return the nice value of a given task. | |
5132 | * @p: the task in question. | |
5133 | */ | |
36c8b586 | 5134 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5135 | { |
5136 | return TASK_NICE(p); | |
5137 | } | |
150d8bed | 5138 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5139 | |
5140 | /** | |
5141 | * idle_cpu - is a given cpu idle currently? | |
5142 | * @cpu: the processor in question. | |
5143 | */ | |
5144 | int idle_cpu(int cpu) | |
5145 | { | |
5146 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5147 | } | |
5148 | ||
1da177e4 LT |
5149 | /** |
5150 | * idle_task - return the idle task for a given cpu. | |
5151 | * @cpu: the processor in question. | |
5152 | */ | |
36c8b586 | 5153 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5154 | { |
5155 | return cpu_rq(cpu)->idle; | |
5156 | } | |
5157 | ||
5158 | /** | |
5159 | * find_process_by_pid - find a process with a matching PID value. | |
5160 | * @pid: the pid in question. | |
5161 | */ | |
a9957449 | 5162 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5163 | { |
228ebcbe | 5164 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5165 | } |
5166 | ||
5167 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5168 | static void |
5169 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5170 | { |
dd41f596 | 5171 | BUG_ON(p->se.on_rq); |
48f24c4d | 5172 | |
1da177e4 | 5173 | p->policy = policy; |
dd41f596 IM |
5174 | switch (p->policy) { |
5175 | case SCHED_NORMAL: | |
5176 | case SCHED_BATCH: | |
5177 | case SCHED_IDLE: | |
5178 | p->sched_class = &fair_sched_class; | |
5179 | break; | |
5180 | case SCHED_FIFO: | |
5181 | case SCHED_RR: | |
5182 | p->sched_class = &rt_sched_class; | |
5183 | break; | |
5184 | } | |
5185 | ||
1da177e4 | 5186 | p->rt_priority = prio; |
b29739f9 IM |
5187 | p->normal_prio = normal_prio(p); |
5188 | /* we are holding p->pi_lock already */ | |
5189 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5190 | set_load_weight(p); |
1da177e4 LT |
5191 | } |
5192 | ||
961ccddd RR |
5193 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5194 | struct sched_param *param, bool user) | |
1da177e4 | 5195 | { |
83b699ed | 5196 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5197 | unsigned long flags; |
cb469845 | 5198 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5199 | struct rq *rq; |
1da177e4 | 5200 | |
66e5393a SR |
5201 | /* may grab non-irq protected spin_locks */ |
5202 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5203 | recheck: |
5204 | /* double check policy once rq lock held */ | |
5205 | if (policy < 0) | |
5206 | policy = oldpolicy = p->policy; | |
5207 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5208 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5209 | policy != SCHED_IDLE) | |
b0a9499c | 5210 | return -EINVAL; |
1da177e4 LT |
5211 | /* |
5212 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5213 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5214 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5215 | */ |
5216 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5217 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5218 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5219 | return -EINVAL; |
e05606d3 | 5220 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5221 | return -EINVAL; |
5222 | ||
37e4ab3f OC |
5223 | /* |
5224 | * Allow unprivileged RT tasks to decrease priority: | |
5225 | */ | |
961ccddd | 5226 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5227 | if (rt_policy(policy)) { |
8dc3e909 | 5228 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5229 | |
5230 | if (!lock_task_sighand(p, &flags)) | |
5231 | return -ESRCH; | |
5232 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5233 | unlock_task_sighand(p, &flags); | |
5234 | ||
5235 | /* can't set/change the rt policy */ | |
5236 | if (policy != p->policy && !rlim_rtprio) | |
5237 | return -EPERM; | |
5238 | ||
5239 | /* can't increase priority */ | |
5240 | if (param->sched_priority > p->rt_priority && | |
5241 | param->sched_priority > rlim_rtprio) | |
5242 | return -EPERM; | |
5243 | } | |
dd41f596 IM |
5244 | /* |
5245 | * Like positive nice levels, dont allow tasks to | |
5246 | * move out of SCHED_IDLE either: | |
5247 | */ | |
5248 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5249 | return -EPERM; | |
5fe1d75f | 5250 | |
37e4ab3f OC |
5251 | /* can't change other user's priorities */ |
5252 | if ((current->euid != p->euid) && | |
5253 | (current->euid != p->uid)) | |
5254 | return -EPERM; | |
5255 | } | |
1da177e4 | 5256 | |
725aad24 | 5257 | if (user) { |
b68aa230 | 5258 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5259 | /* |
5260 | * Do not allow realtime tasks into groups that have no runtime | |
5261 | * assigned. | |
5262 | */ | |
9a7e0b18 PZ |
5263 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5264 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5265 | return -EPERM; |
b68aa230 PZ |
5266 | #endif |
5267 | ||
725aad24 JF |
5268 | retval = security_task_setscheduler(p, policy, param); |
5269 | if (retval) | |
5270 | return retval; | |
5271 | } | |
5272 | ||
b29739f9 IM |
5273 | /* |
5274 | * make sure no PI-waiters arrive (or leave) while we are | |
5275 | * changing the priority of the task: | |
5276 | */ | |
5277 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5278 | /* |
5279 | * To be able to change p->policy safely, the apropriate | |
5280 | * runqueue lock must be held. | |
5281 | */ | |
b29739f9 | 5282 | rq = __task_rq_lock(p); |
1da177e4 LT |
5283 | /* recheck policy now with rq lock held */ |
5284 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5285 | policy = oldpolicy = -1; | |
b29739f9 IM |
5286 | __task_rq_unlock(rq); |
5287 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5288 | goto recheck; |
5289 | } | |
2daa3577 | 5290 | update_rq_clock(rq); |
dd41f596 | 5291 | on_rq = p->se.on_rq; |
051a1d1a | 5292 | running = task_current(rq, p); |
0e1f3483 | 5293 | if (on_rq) |
2e1cb74a | 5294 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5295 | if (running) |
5296 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5297 | |
1da177e4 | 5298 | oldprio = p->prio; |
dd41f596 | 5299 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5300 | |
0e1f3483 HS |
5301 | if (running) |
5302 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5303 | if (on_rq) { |
5304 | activate_task(rq, p, 0); | |
cb469845 SR |
5305 | |
5306 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5307 | } |
b29739f9 IM |
5308 | __task_rq_unlock(rq); |
5309 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5310 | ||
95e02ca9 TG |
5311 | rt_mutex_adjust_pi(p); |
5312 | ||
1da177e4 LT |
5313 | return 0; |
5314 | } | |
961ccddd RR |
5315 | |
5316 | /** | |
5317 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5318 | * @p: the task in question. | |
5319 | * @policy: new policy. | |
5320 | * @param: structure containing the new RT priority. | |
5321 | * | |
5322 | * NOTE that the task may be already dead. | |
5323 | */ | |
5324 | int sched_setscheduler(struct task_struct *p, int policy, | |
5325 | struct sched_param *param) | |
5326 | { | |
5327 | return __sched_setscheduler(p, policy, param, true); | |
5328 | } | |
1da177e4 LT |
5329 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5330 | ||
961ccddd RR |
5331 | /** |
5332 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5333 | * @p: the task in question. | |
5334 | * @policy: new policy. | |
5335 | * @param: structure containing the new RT priority. | |
5336 | * | |
5337 | * Just like sched_setscheduler, only don't bother checking if the | |
5338 | * current context has permission. For example, this is needed in | |
5339 | * stop_machine(): we create temporary high priority worker threads, | |
5340 | * but our caller might not have that capability. | |
5341 | */ | |
5342 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5343 | struct sched_param *param) | |
5344 | { | |
5345 | return __sched_setscheduler(p, policy, param, false); | |
5346 | } | |
5347 | ||
95cdf3b7 IM |
5348 | static int |
5349 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5350 | { |
1da177e4 LT |
5351 | struct sched_param lparam; |
5352 | struct task_struct *p; | |
36c8b586 | 5353 | int retval; |
1da177e4 LT |
5354 | |
5355 | if (!param || pid < 0) | |
5356 | return -EINVAL; | |
5357 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5358 | return -EFAULT; | |
5fe1d75f ON |
5359 | |
5360 | rcu_read_lock(); | |
5361 | retval = -ESRCH; | |
1da177e4 | 5362 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5363 | if (p != NULL) |
5364 | retval = sched_setscheduler(p, policy, &lparam); | |
5365 | rcu_read_unlock(); | |
36c8b586 | 5366 | |
1da177e4 LT |
5367 | return retval; |
5368 | } | |
5369 | ||
5370 | /** | |
5371 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5372 | * @pid: the pid in question. | |
5373 | * @policy: new policy. | |
5374 | * @param: structure containing the new RT priority. | |
5375 | */ | |
41a2d6cf IM |
5376 | asmlinkage long |
5377 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5378 | { |
c21761f1 JB |
5379 | /* negative values for policy are not valid */ |
5380 | if (policy < 0) | |
5381 | return -EINVAL; | |
5382 | ||
1da177e4 LT |
5383 | return do_sched_setscheduler(pid, policy, param); |
5384 | } | |
5385 | ||
5386 | /** | |
5387 | * sys_sched_setparam - set/change the RT priority of a thread | |
5388 | * @pid: the pid in question. | |
5389 | * @param: structure containing the new RT priority. | |
5390 | */ | |
5391 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
5392 | { | |
5393 | return do_sched_setscheduler(pid, -1, param); | |
5394 | } | |
5395 | ||
5396 | /** | |
5397 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5398 | * @pid: the pid in question. | |
5399 | */ | |
5400 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
5401 | { | |
36c8b586 | 5402 | struct task_struct *p; |
3a5c359a | 5403 | int retval; |
1da177e4 LT |
5404 | |
5405 | if (pid < 0) | |
3a5c359a | 5406 | return -EINVAL; |
1da177e4 LT |
5407 | |
5408 | retval = -ESRCH; | |
5409 | read_lock(&tasklist_lock); | |
5410 | p = find_process_by_pid(pid); | |
5411 | if (p) { | |
5412 | retval = security_task_getscheduler(p); | |
5413 | if (!retval) | |
5414 | retval = p->policy; | |
5415 | } | |
5416 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5417 | return retval; |
5418 | } | |
5419 | ||
5420 | /** | |
5421 | * sys_sched_getscheduler - get the RT priority of a thread | |
5422 | * @pid: the pid in question. | |
5423 | * @param: structure containing the RT priority. | |
5424 | */ | |
5425 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5426 | { | |
5427 | struct sched_param lp; | |
36c8b586 | 5428 | struct task_struct *p; |
3a5c359a | 5429 | int retval; |
1da177e4 LT |
5430 | |
5431 | if (!param || pid < 0) | |
3a5c359a | 5432 | return -EINVAL; |
1da177e4 LT |
5433 | |
5434 | read_lock(&tasklist_lock); | |
5435 | p = find_process_by_pid(pid); | |
5436 | retval = -ESRCH; | |
5437 | if (!p) | |
5438 | goto out_unlock; | |
5439 | ||
5440 | retval = security_task_getscheduler(p); | |
5441 | if (retval) | |
5442 | goto out_unlock; | |
5443 | ||
5444 | lp.sched_priority = p->rt_priority; | |
5445 | read_unlock(&tasklist_lock); | |
5446 | ||
5447 | /* | |
5448 | * This one might sleep, we cannot do it with a spinlock held ... | |
5449 | */ | |
5450 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5451 | ||
1da177e4 LT |
5452 | return retval; |
5453 | ||
5454 | out_unlock: | |
5455 | read_unlock(&tasklist_lock); | |
5456 | return retval; | |
5457 | } | |
5458 | ||
96f874e2 | 5459 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5460 | { |
5a16f3d3 | 5461 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5462 | struct task_struct *p; |
5463 | int retval; | |
1da177e4 | 5464 | |
95402b38 | 5465 | get_online_cpus(); |
1da177e4 LT |
5466 | read_lock(&tasklist_lock); |
5467 | ||
5468 | p = find_process_by_pid(pid); | |
5469 | if (!p) { | |
5470 | read_unlock(&tasklist_lock); | |
95402b38 | 5471 | put_online_cpus(); |
1da177e4 LT |
5472 | return -ESRCH; |
5473 | } | |
5474 | ||
5475 | /* | |
5476 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5477 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5478 | * usage count and then drop tasklist_lock. |
5479 | */ | |
5480 | get_task_struct(p); | |
5481 | read_unlock(&tasklist_lock); | |
5482 | ||
5a16f3d3 RR |
5483 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5484 | retval = -ENOMEM; | |
5485 | goto out_put_task; | |
5486 | } | |
5487 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5488 | retval = -ENOMEM; | |
5489 | goto out_free_cpus_allowed; | |
5490 | } | |
1da177e4 LT |
5491 | retval = -EPERM; |
5492 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5493 | !capable(CAP_SYS_NICE)) | |
5494 | goto out_unlock; | |
5495 | ||
e7834f8f DQ |
5496 | retval = security_task_setscheduler(p, 0, NULL); |
5497 | if (retval) | |
5498 | goto out_unlock; | |
5499 | ||
5a16f3d3 RR |
5500 | cpuset_cpus_allowed(p, cpus_allowed); |
5501 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 5502 | again: |
5a16f3d3 | 5503 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5504 | |
8707d8b8 | 5505 | if (!retval) { |
5a16f3d3 RR |
5506 | cpuset_cpus_allowed(p, cpus_allowed); |
5507 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5508 | /* |
5509 | * We must have raced with a concurrent cpuset | |
5510 | * update. Just reset the cpus_allowed to the | |
5511 | * cpuset's cpus_allowed | |
5512 | */ | |
5a16f3d3 | 5513 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5514 | goto again; |
5515 | } | |
5516 | } | |
1da177e4 | 5517 | out_unlock: |
5a16f3d3 RR |
5518 | free_cpumask_var(new_mask); |
5519 | out_free_cpus_allowed: | |
5520 | free_cpumask_var(cpus_allowed); | |
5521 | out_put_task: | |
1da177e4 | 5522 | put_task_struct(p); |
95402b38 | 5523 | put_online_cpus(); |
1da177e4 LT |
5524 | return retval; |
5525 | } | |
5526 | ||
5527 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5528 | struct cpumask *new_mask) |
1da177e4 | 5529 | { |
96f874e2 RR |
5530 | if (len < cpumask_size()) |
5531 | cpumask_clear(new_mask); | |
5532 | else if (len > cpumask_size()) | |
5533 | len = cpumask_size(); | |
5534 | ||
1da177e4 LT |
5535 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5536 | } | |
5537 | ||
5538 | /** | |
5539 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5540 | * @pid: pid of the process | |
5541 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5542 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5543 | */ | |
5544 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5545 | unsigned long __user *user_mask_ptr) | |
5546 | { | |
5a16f3d3 | 5547 | cpumask_var_t new_mask; |
1da177e4 LT |
5548 | int retval; |
5549 | ||
5a16f3d3 RR |
5550 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5551 | return -ENOMEM; | |
1da177e4 | 5552 | |
5a16f3d3 RR |
5553 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5554 | if (retval == 0) | |
5555 | retval = sched_setaffinity(pid, new_mask); | |
5556 | free_cpumask_var(new_mask); | |
5557 | return retval; | |
1da177e4 LT |
5558 | } |
5559 | ||
96f874e2 | 5560 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5561 | { |
36c8b586 | 5562 | struct task_struct *p; |
1da177e4 | 5563 | int retval; |
1da177e4 | 5564 | |
95402b38 | 5565 | get_online_cpus(); |
1da177e4 LT |
5566 | read_lock(&tasklist_lock); |
5567 | ||
5568 | retval = -ESRCH; | |
5569 | p = find_process_by_pid(pid); | |
5570 | if (!p) | |
5571 | goto out_unlock; | |
5572 | ||
e7834f8f DQ |
5573 | retval = security_task_getscheduler(p); |
5574 | if (retval) | |
5575 | goto out_unlock; | |
5576 | ||
96f874e2 | 5577 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
5578 | |
5579 | out_unlock: | |
5580 | read_unlock(&tasklist_lock); | |
95402b38 | 5581 | put_online_cpus(); |
1da177e4 | 5582 | |
9531b62f | 5583 | return retval; |
1da177e4 LT |
5584 | } |
5585 | ||
5586 | /** | |
5587 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5588 | * @pid: pid of the process | |
5589 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5590 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5591 | */ | |
5592 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5593 | unsigned long __user *user_mask_ptr) | |
5594 | { | |
5595 | int ret; | |
f17c8607 | 5596 | cpumask_var_t mask; |
1da177e4 | 5597 | |
f17c8607 | 5598 | if (len < cpumask_size()) |
1da177e4 LT |
5599 | return -EINVAL; |
5600 | ||
f17c8607 RR |
5601 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5602 | return -ENOMEM; | |
1da177e4 | 5603 | |
f17c8607 RR |
5604 | ret = sched_getaffinity(pid, mask); |
5605 | if (ret == 0) { | |
5606 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
5607 | ret = -EFAULT; | |
5608 | else | |
5609 | ret = cpumask_size(); | |
5610 | } | |
5611 | free_cpumask_var(mask); | |
1da177e4 | 5612 | |
f17c8607 | 5613 | return ret; |
1da177e4 LT |
5614 | } |
5615 | ||
5616 | /** | |
5617 | * sys_sched_yield - yield the current processor to other threads. | |
5618 | * | |
dd41f596 IM |
5619 | * This function yields the current CPU to other tasks. If there are no |
5620 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5621 | */ |
5622 | asmlinkage long sys_sched_yield(void) | |
5623 | { | |
70b97a7f | 5624 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5625 | |
2d72376b | 5626 | schedstat_inc(rq, yld_count); |
4530d7ab | 5627 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5628 | |
5629 | /* | |
5630 | * Since we are going to call schedule() anyway, there's | |
5631 | * no need to preempt or enable interrupts: | |
5632 | */ | |
5633 | __release(rq->lock); | |
8a25d5de | 5634 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5635 | _raw_spin_unlock(&rq->lock); |
5636 | preempt_enable_no_resched(); | |
5637 | ||
5638 | schedule(); | |
5639 | ||
5640 | return 0; | |
5641 | } | |
5642 | ||
e7b38404 | 5643 | static void __cond_resched(void) |
1da177e4 | 5644 | { |
8e0a43d8 IM |
5645 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5646 | __might_sleep(__FILE__, __LINE__); | |
5647 | #endif | |
5bbcfd90 IM |
5648 | /* |
5649 | * The BKS might be reacquired before we have dropped | |
5650 | * PREEMPT_ACTIVE, which could trigger a second | |
5651 | * cond_resched() call. | |
5652 | */ | |
1da177e4 LT |
5653 | do { |
5654 | add_preempt_count(PREEMPT_ACTIVE); | |
5655 | schedule(); | |
5656 | sub_preempt_count(PREEMPT_ACTIVE); | |
5657 | } while (need_resched()); | |
5658 | } | |
5659 | ||
02b67cc3 | 5660 | int __sched _cond_resched(void) |
1da177e4 | 5661 | { |
9414232f IM |
5662 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5663 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5664 | __cond_resched(); |
5665 | return 1; | |
5666 | } | |
5667 | return 0; | |
5668 | } | |
02b67cc3 | 5669 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5670 | |
5671 | /* | |
5672 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5673 | * call schedule, and on return reacquire the lock. | |
5674 | * | |
41a2d6cf | 5675 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5676 | * operations here to prevent schedule() from being called twice (once via |
5677 | * spin_unlock(), once by hand). | |
5678 | */ | |
95cdf3b7 | 5679 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5680 | { |
95c354fe | 5681 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5682 | int ret = 0; |
5683 | ||
95c354fe | 5684 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5685 | spin_unlock(lock); |
95c354fe NP |
5686 | if (resched && need_resched()) |
5687 | __cond_resched(); | |
5688 | else | |
5689 | cpu_relax(); | |
6df3cecb | 5690 | ret = 1; |
1da177e4 | 5691 | spin_lock(lock); |
1da177e4 | 5692 | } |
6df3cecb | 5693 | return ret; |
1da177e4 | 5694 | } |
1da177e4 LT |
5695 | EXPORT_SYMBOL(cond_resched_lock); |
5696 | ||
5697 | int __sched cond_resched_softirq(void) | |
5698 | { | |
5699 | BUG_ON(!in_softirq()); | |
5700 | ||
9414232f | 5701 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5702 | local_bh_enable(); |
1da177e4 LT |
5703 | __cond_resched(); |
5704 | local_bh_disable(); | |
5705 | return 1; | |
5706 | } | |
5707 | return 0; | |
5708 | } | |
1da177e4 LT |
5709 | EXPORT_SYMBOL(cond_resched_softirq); |
5710 | ||
1da177e4 LT |
5711 | /** |
5712 | * yield - yield the current processor to other threads. | |
5713 | * | |
72fd4a35 | 5714 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5715 | * thread runnable and calls sys_sched_yield(). |
5716 | */ | |
5717 | void __sched yield(void) | |
5718 | { | |
5719 | set_current_state(TASK_RUNNING); | |
5720 | sys_sched_yield(); | |
5721 | } | |
1da177e4 LT |
5722 | EXPORT_SYMBOL(yield); |
5723 | ||
5724 | /* | |
41a2d6cf | 5725 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5726 | * that process accounting knows that this is a task in IO wait state. |
5727 | * | |
5728 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5729 | * has set its backing_dev_info: the queue against which it should throttle) | |
5730 | */ | |
5731 | void __sched io_schedule(void) | |
5732 | { | |
70b97a7f | 5733 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5734 | |
0ff92245 | 5735 | delayacct_blkio_start(); |
1da177e4 LT |
5736 | atomic_inc(&rq->nr_iowait); |
5737 | schedule(); | |
5738 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5739 | delayacct_blkio_end(); |
1da177e4 | 5740 | } |
1da177e4 LT |
5741 | EXPORT_SYMBOL(io_schedule); |
5742 | ||
5743 | long __sched io_schedule_timeout(long timeout) | |
5744 | { | |
70b97a7f | 5745 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5746 | long ret; |
5747 | ||
0ff92245 | 5748 | delayacct_blkio_start(); |
1da177e4 LT |
5749 | atomic_inc(&rq->nr_iowait); |
5750 | ret = schedule_timeout(timeout); | |
5751 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5752 | delayacct_blkio_end(); |
1da177e4 LT |
5753 | return ret; |
5754 | } | |
5755 | ||
5756 | /** | |
5757 | * sys_sched_get_priority_max - return maximum RT priority. | |
5758 | * @policy: scheduling class. | |
5759 | * | |
5760 | * this syscall returns the maximum rt_priority that can be used | |
5761 | * by a given scheduling class. | |
5762 | */ | |
5763 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5764 | { | |
5765 | int ret = -EINVAL; | |
5766 | ||
5767 | switch (policy) { | |
5768 | case SCHED_FIFO: | |
5769 | case SCHED_RR: | |
5770 | ret = MAX_USER_RT_PRIO-1; | |
5771 | break; | |
5772 | case SCHED_NORMAL: | |
b0a9499c | 5773 | case SCHED_BATCH: |
dd41f596 | 5774 | case SCHED_IDLE: |
1da177e4 LT |
5775 | ret = 0; |
5776 | break; | |
5777 | } | |
5778 | return ret; | |
5779 | } | |
5780 | ||
5781 | /** | |
5782 | * sys_sched_get_priority_min - return minimum RT priority. | |
5783 | * @policy: scheduling class. | |
5784 | * | |
5785 | * this syscall returns the minimum rt_priority that can be used | |
5786 | * by a given scheduling class. | |
5787 | */ | |
5788 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5789 | { | |
5790 | int ret = -EINVAL; | |
5791 | ||
5792 | switch (policy) { | |
5793 | case SCHED_FIFO: | |
5794 | case SCHED_RR: | |
5795 | ret = 1; | |
5796 | break; | |
5797 | case SCHED_NORMAL: | |
b0a9499c | 5798 | case SCHED_BATCH: |
dd41f596 | 5799 | case SCHED_IDLE: |
1da177e4 LT |
5800 | ret = 0; |
5801 | } | |
5802 | return ret; | |
5803 | } | |
5804 | ||
5805 | /** | |
5806 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5807 | * @pid: pid of the process. | |
5808 | * @interval: userspace pointer to the timeslice value. | |
5809 | * | |
5810 | * this syscall writes the default timeslice value of a given process | |
5811 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5812 | */ | |
5813 | asmlinkage | |
5814 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5815 | { | |
36c8b586 | 5816 | struct task_struct *p; |
a4ec24b4 | 5817 | unsigned int time_slice; |
3a5c359a | 5818 | int retval; |
1da177e4 | 5819 | struct timespec t; |
1da177e4 LT |
5820 | |
5821 | if (pid < 0) | |
3a5c359a | 5822 | return -EINVAL; |
1da177e4 LT |
5823 | |
5824 | retval = -ESRCH; | |
5825 | read_lock(&tasklist_lock); | |
5826 | p = find_process_by_pid(pid); | |
5827 | if (!p) | |
5828 | goto out_unlock; | |
5829 | ||
5830 | retval = security_task_getscheduler(p); | |
5831 | if (retval) | |
5832 | goto out_unlock; | |
5833 | ||
77034937 IM |
5834 | /* |
5835 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5836 | * tasks that are on an otherwise idle runqueue: | |
5837 | */ | |
5838 | time_slice = 0; | |
5839 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5840 | time_slice = DEF_TIMESLICE; |
1868f958 | 5841 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5842 | struct sched_entity *se = &p->se; |
5843 | unsigned long flags; | |
5844 | struct rq *rq; | |
5845 | ||
5846 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5847 | if (rq->cfs.load.weight) |
5848 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5849 | task_rq_unlock(rq, &flags); |
5850 | } | |
1da177e4 | 5851 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5852 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5853 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5854 | return retval; |
3a5c359a | 5855 | |
1da177e4 LT |
5856 | out_unlock: |
5857 | read_unlock(&tasklist_lock); | |
5858 | return retval; | |
5859 | } | |
5860 | ||
7c731e0a | 5861 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5862 | |
82a1fcb9 | 5863 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5864 | { |
1da177e4 | 5865 | unsigned long free = 0; |
36c8b586 | 5866 | unsigned state; |
1da177e4 | 5867 | |
1da177e4 | 5868 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5869 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5870 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5871 | #if BITS_PER_LONG == 32 |
1da177e4 | 5872 | if (state == TASK_RUNNING) |
cc4ea795 | 5873 | printk(KERN_CONT " running "); |
1da177e4 | 5874 | else |
cc4ea795 | 5875 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5876 | #else |
5877 | if (state == TASK_RUNNING) | |
cc4ea795 | 5878 | printk(KERN_CONT " running task "); |
1da177e4 | 5879 | else |
cc4ea795 | 5880 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5881 | #endif |
5882 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5883 | { | |
10ebffde | 5884 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5885 | while (!*n) |
5886 | n++; | |
10ebffde | 5887 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5888 | } |
5889 | #endif | |
ba25f9dc | 5890 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5891 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5892 | |
5fb5e6de | 5893 | show_stack(p, NULL); |
1da177e4 LT |
5894 | } |
5895 | ||
e59e2ae2 | 5896 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5897 | { |
36c8b586 | 5898 | struct task_struct *g, *p; |
1da177e4 | 5899 | |
4bd77321 IM |
5900 | #if BITS_PER_LONG == 32 |
5901 | printk(KERN_INFO | |
5902 | " task PC stack pid father\n"); | |
1da177e4 | 5903 | #else |
4bd77321 IM |
5904 | printk(KERN_INFO |
5905 | " task PC stack pid father\n"); | |
1da177e4 LT |
5906 | #endif |
5907 | read_lock(&tasklist_lock); | |
5908 | do_each_thread(g, p) { | |
5909 | /* | |
5910 | * reset the NMI-timeout, listing all files on a slow | |
5911 | * console might take alot of time: | |
5912 | */ | |
5913 | touch_nmi_watchdog(); | |
39bc89fd | 5914 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5915 | sched_show_task(p); |
1da177e4 LT |
5916 | } while_each_thread(g, p); |
5917 | ||
04c9167f JF |
5918 | touch_all_softlockup_watchdogs(); |
5919 | ||
dd41f596 IM |
5920 | #ifdef CONFIG_SCHED_DEBUG |
5921 | sysrq_sched_debug_show(); | |
5922 | #endif | |
1da177e4 | 5923 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5924 | /* |
5925 | * Only show locks if all tasks are dumped: | |
5926 | */ | |
5927 | if (state_filter == -1) | |
5928 | debug_show_all_locks(); | |
1da177e4 LT |
5929 | } |
5930 | ||
1df21055 IM |
5931 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5932 | { | |
dd41f596 | 5933 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5934 | } |
5935 | ||
f340c0d1 IM |
5936 | /** |
5937 | * init_idle - set up an idle thread for a given CPU | |
5938 | * @idle: task in question | |
5939 | * @cpu: cpu the idle task belongs to | |
5940 | * | |
5941 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5942 | * flag, to make booting more robust. | |
5943 | */ | |
5c1e1767 | 5944 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5945 | { |
70b97a7f | 5946 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5947 | unsigned long flags; |
5948 | ||
5cbd54ef IM |
5949 | spin_lock_irqsave(&rq->lock, flags); |
5950 | ||
dd41f596 IM |
5951 | __sched_fork(idle); |
5952 | idle->se.exec_start = sched_clock(); | |
5953 | ||
b29739f9 | 5954 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 5955 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5956 | __set_task_cpu(idle, cpu); |
1da177e4 | 5957 | |
1da177e4 | 5958 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5959 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5960 | idle->oncpu = 1; | |
5961 | #endif | |
1da177e4 LT |
5962 | spin_unlock_irqrestore(&rq->lock, flags); |
5963 | ||
5964 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5965 | #if defined(CONFIG_PREEMPT) |
5966 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5967 | #else | |
a1261f54 | 5968 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5969 | #endif |
dd41f596 IM |
5970 | /* |
5971 | * The idle tasks have their own, simple scheduling class: | |
5972 | */ | |
5973 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5974 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5975 | } |
5976 | ||
5977 | /* | |
5978 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5979 | * indicates which cpus entered this state. This is used | |
5980 | * in the rcu update to wait only for active cpus. For system | |
5981 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5982 | * always be CPU_BITS_NONE. |
1da177e4 | 5983 | */ |
6a7b3dc3 | 5984 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5985 | |
19978ca6 IM |
5986 | /* |
5987 | * Increase the granularity value when there are more CPUs, | |
5988 | * because with more CPUs the 'effective latency' as visible | |
5989 | * to users decreases. But the relationship is not linear, | |
5990 | * so pick a second-best guess by going with the log2 of the | |
5991 | * number of CPUs. | |
5992 | * | |
5993 | * This idea comes from the SD scheduler of Con Kolivas: | |
5994 | */ | |
5995 | static inline void sched_init_granularity(void) | |
5996 | { | |
5997 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5998 | const unsigned long limit = 200000000; | |
5999 | ||
6000 | sysctl_sched_min_granularity *= factor; | |
6001 | if (sysctl_sched_min_granularity > limit) | |
6002 | sysctl_sched_min_granularity = limit; | |
6003 | ||
6004 | sysctl_sched_latency *= factor; | |
6005 | if (sysctl_sched_latency > limit) | |
6006 | sysctl_sched_latency = limit; | |
6007 | ||
6008 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6009 | |
6010 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6011 | } |
6012 | ||
1da177e4 LT |
6013 | #ifdef CONFIG_SMP |
6014 | /* | |
6015 | * This is how migration works: | |
6016 | * | |
70b97a7f | 6017 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6018 | * runqueue and wake up that CPU's migration thread. |
6019 | * 2) we down() the locked semaphore => thread blocks. | |
6020 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6021 | * thread off the CPU) | |
6022 | * 4) it gets the migration request and checks whether the migrated | |
6023 | * task is still in the wrong runqueue. | |
6024 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6025 | * it and puts it into the right queue. | |
6026 | * 6) migration thread up()s the semaphore. | |
6027 | * 7) we wake up and the migration is done. | |
6028 | */ | |
6029 | ||
6030 | /* | |
6031 | * Change a given task's CPU affinity. Migrate the thread to a | |
6032 | * proper CPU and schedule it away if the CPU it's executing on | |
6033 | * is removed from the allowed bitmask. | |
6034 | * | |
6035 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6036 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6037 | * call is not atomic; no spinlocks may be held. |
6038 | */ | |
96f874e2 | 6039 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6040 | { |
70b97a7f | 6041 | struct migration_req req; |
1da177e4 | 6042 | unsigned long flags; |
70b97a7f | 6043 | struct rq *rq; |
48f24c4d | 6044 | int ret = 0; |
1da177e4 LT |
6045 | |
6046 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6047 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6048 | ret = -EINVAL; |
6049 | goto out; | |
6050 | } | |
6051 | ||
9985b0ba | 6052 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6053 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6054 | ret = -EINVAL; |
6055 | goto out; | |
6056 | } | |
6057 | ||
73fe6aae | 6058 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6059 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6060 | else { |
96f874e2 RR |
6061 | cpumask_copy(&p->cpus_allowed, new_mask); |
6062 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6063 | } |
6064 | ||
1da177e4 | 6065 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6066 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6067 | goto out; |
6068 | ||
1e5ce4f4 | 6069 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6070 | /* Need help from migration thread: drop lock and wait. */ |
6071 | task_rq_unlock(rq, &flags); | |
6072 | wake_up_process(rq->migration_thread); | |
6073 | wait_for_completion(&req.done); | |
6074 | tlb_migrate_finish(p->mm); | |
6075 | return 0; | |
6076 | } | |
6077 | out: | |
6078 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6079 | |
1da177e4 LT |
6080 | return ret; |
6081 | } | |
cd8ba7cd | 6082 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6083 | |
6084 | /* | |
41a2d6cf | 6085 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6086 | * this because either it can't run here any more (set_cpus_allowed() |
6087 | * away from this CPU, or CPU going down), or because we're | |
6088 | * attempting to rebalance this task on exec (sched_exec). | |
6089 | * | |
6090 | * So we race with normal scheduler movements, but that's OK, as long | |
6091 | * as the task is no longer on this CPU. | |
efc30814 KK |
6092 | * |
6093 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6094 | */ |
efc30814 | 6095 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6096 | { |
70b97a7f | 6097 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6098 | int ret = 0, on_rq; |
1da177e4 | 6099 | |
e761b772 | 6100 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6101 | return ret; |
1da177e4 LT |
6102 | |
6103 | rq_src = cpu_rq(src_cpu); | |
6104 | rq_dest = cpu_rq(dest_cpu); | |
6105 | ||
6106 | double_rq_lock(rq_src, rq_dest); | |
6107 | /* Already moved. */ | |
6108 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6109 | goto done; |
1da177e4 | 6110 | /* Affinity changed (again). */ |
96f874e2 | 6111 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6112 | goto fail; |
1da177e4 | 6113 | |
dd41f596 | 6114 | on_rq = p->se.on_rq; |
6e82a3be | 6115 | if (on_rq) |
2e1cb74a | 6116 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6117 | |
1da177e4 | 6118 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6119 | if (on_rq) { |
6120 | activate_task(rq_dest, p, 0); | |
15afe09b | 6121 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6122 | } |
b1e38734 | 6123 | done: |
efc30814 | 6124 | ret = 1; |
b1e38734 | 6125 | fail: |
1da177e4 | 6126 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6127 | return ret; |
1da177e4 LT |
6128 | } |
6129 | ||
6130 | /* | |
6131 | * migration_thread - this is a highprio system thread that performs | |
6132 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6133 | * another runqueue. | |
6134 | */ | |
95cdf3b7 | 6135 | static int migration_thread(void *data) |
1da177e4 | 6136 | { |
1da177e4 | 6137 | int cpu = (long)data; |
70b97a7f | 6138 | struct rq *rq; |
1da177e4 LT |
6139 | |
6140 | rq = cpu_rq(cpu); | |
6141 | BUG_ON(rq->migration_thread != current); | |
6142 | ||
6143 | set_current_state(TASK_INTERRUPTIBLE); | |
6144 | while (!kthread_should_stop()) { | |
70b97a7f | 6145 | struct migration_req *req; |
1da177e4 | 6146 | struct list_head *head; |
1da177e4 | 6147 | |
1da177e4 LT |
6148 | spin_lock_irq(&rq->lock); |
6149 | ||
6150 | if (cpu_is_offline(cpu)) { | |
6151 | spin_unlock_irq(&rq->lock); | |
6152 | goto wait_to_die; | |
6153 | } | |
6154 | ||
6155 | if (rq->active_balance) { | |
6156 | active_load_balance(rq, cpu); | |
6157 | rq->active_balance = 0; | |
6158 | } | |
6159 | ||
6160 | head = &rq->migration_queue; | |
6161 | ||
6162 | if (list_empty(head)) { | |
6163 | spin_unlock_irq(&rq->lock); | |
6164 | schedule(); | |
6165 | set_current_state(TASK_INTERRUPTIBLE); | |
6166 | continue; | |
6167 | } | |
70b97a7f | 6168 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6169 | list_del_init(head->next); |
6170 | ||
674311d5 NP |
6171 | spin_unlock(&rq->lock); |
6172 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6173 | local_irq_enable(); | |
1da177e4 LT |
6174 | |
6175 | complete(&req->done); | |
6176 | } | |
6177 | __set_current_state(TASK_RUNNING); | |
6178 | return 0; | |
6179 | ||
6180 | wait_to_die: | |
6181 | /* Wait for kthread_stop */ | |
6182 | set_current_state(TASK_INTERRUPTIBLE); | |
6183 | while (!kthread_should_stop()) { | |
6184 | schedule(); | |
6185 | set_current_state(TASK_INTERRUPTIBLE); | |
6186 | } | |
6187 | __set_current_state(TASK_RUNNING); | |
6188 | return 0; | |
6189 | } | |
6190 | ||
6191 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6192 | |
6193 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6194 | { | |
6195 | int ret; | |
6196 | ||
6197 | local_irq_disable(); | |
6198 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6199 | local_irq_enable(); | |
6200 | return ret; | |
6201 | } | |
6202 | ||
054b9108 | 6203 | /* |
3a4fa0a2 | 6204 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 6205 | */ |
48f24c4d | 6206 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6207 | { |
70b97a7f | 6208 | int dest_cpu; |
e76bd8d9 RR |
6209 | /* FIXME: Use cpumask_of_node here. */ |
6210 | cpumask_t _nodemask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
6211 | const struct cpumask *nodemask = &_nodemask; | |
6212 | ||
6213 | again: | |
6214 | /* Look for allowed, online CPU in same node. */ | |
6215 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
6216 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
6217 | goto move; | |
6218 | ||
6219 | /* Any allowed, online CPU? */ | |
6220 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
6221 | if (dest_cpu < nr_cpu_ids) | |
6222 | goto move; | |
6223 | ||
6224 | /* No more Mr. Nice Guy. */ | |
6225 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
6226 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
6227 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 6228 | |
e76bd8d9 RR |
6229 | /* |
6230 | * Don't tell them about moving exiting tasks or | |
6231 | * kernel threads (both mm NULL), since they never | |
6232 | * leave kernel. | |
6233 | */ | |
6234 | if (p->mm && printk_ratelimit()) { | |
6235 | printk(KERN_INFO "process %d (%s) no " | |
6236 | "longer affine to cpu%d\n", | |
6237 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 6238 | } |
e76bd8d9 RR |
6239 | } |
6240 | ||
6241 | move: | |
6242 | /* It can have affinity changed while we were choosing. */ | |
6243 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
6244 | goto again; | |
1da177e4 LT |
6245 | } |
6246 | ||
6247 | /* | |
6248 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6249 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6250 | * for performance reasons the counter is not stricly tracking tasks to | |
6251 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6252 | * to keep the global sum constant after CPU-down: | |
6253 | */ | |
70b97a7f | 6254 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6255 | { |
1e5ce4f4 | 6256 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6257 | unsigned long flags; |
6258 | ||
6259 | local_irq_save(flags); | |
6260 | double_rq_lock(rq_src, rq_dest); | |
6261 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6262 | rq_src->nr_uninterruptible = 0; | |
6263 | double_rq_unlock(rq_src, rq_dest); | |
6264 | local_irq_restore(flags); | |
6265 | } | |
6266 | ||
6267 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6268 | static void migrate_live_tasks(int src_cpu) | |
6269 | { | |
48f24c4d | 6270 | struct task_struct *p, *t; |
1da177e4 | 6271 | |
f7b4cddc | 6272 | read_lock(&tasklist_lock); |
1da177e4 | 6273 | |
48f24c4d IM |
6274 | do_each_thread(t, p) { |
6275 | if (p == current) | |
1da177e4 LT |
6276 | continue; |
6277 | ||
48f24c4d IM |
6278 | if (task_cpu(p) == src_cpu) |
6279 | move_task_off_dead_cpu(src_cpu, p); | |
6280 | } while_each_thread(t, p); | |
1da177e4 | 6281 | |
f7b4cddc | 6282 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6283 | } |
6284 | ||
dd41f596 IM |
6285 | /* |
6286 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6287 | * It does so by boosting its priority to highest possible. |
6288 | * Used by CPU offline code. | |
1da177e4 LT |
6289 | */ |
6290 | void sched_idle_next(void) | |
6291 | { | |
48f24c4d | 6292 | int this_cpu = smp_processor_id(); |
70b97a7f | 6293 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6294 | struct task_struct *p = rq->idle; |
6295 | unsigned long flags; | |
6296 | ||
6297 | /* cpu has to be offline */ | |
48f24c4d | 6298 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6299 | |
48f24c4d IM |
6300 | /* |
6301 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6302 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6303 | */ |
6304 | spin_lock_irqsave(&rq->lock, flags); | |
6305 | ||
dd41f596 | 6306 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6307 | |
94bc9a7b DA |
6308 | update_rq_clock(rq); |
6309 | activate_task(rq, p, 0); | |
1da177e4 LT |
6310 | |
6311 | spin_unlock_irqrestore(&rq->lock, flags); | |
6312 | } | |
6313 | ||
48f24c4d IM |
6314 | /* |
6315 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6316 | * offline. |
6317 | */ | |
6318 | void idle_task_exit(void) | |
6319 | { | |
6320 | struct mm_struct *mm = current->active_mm; | |
6321 | ||
6322 | BUG_ON(cpu_online(smp_processor_id())); | |
6323 | ||
6324 | if (mm != &init_mm) | |
6325 | switch_mm(mm, &init_mm, current); | |
6326 | mmdrop(mm); | |
6327 | } | |
6328 | ||
054b9108 | 6329 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6330 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6331 | { |
70b97a7f | 6332 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6333 | |
6334 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6335 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6336 | |
6337 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6338 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6339 | |
48f24c4d | 6340 | get_task_struct(p); |
1da177e4 LT |
6341 | |
6342 | /* | |
6343 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6344 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6345 | * fine. |
6346 | */ | |
f7b4cddc | 6347 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6348 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6349 | spin_lock_irq(&rq->lock); |
1da177e4 | 6350 | |
48f24c4d | 6351 | put_task_struct(p); |
1da177e4 LT |
6352 | } |
6353 | ||
6354 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6355 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6356 | { | |
70b97a7f | 6357 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6358 | struct task_struct *next; |
48f24c4d | 6359 | |
dd41f596 IM |
6360 | for ( ; ; ) { |
6361 | if (!rq->nr_running) | |
6362 | break; | |
a8e504d2 | 6363 | update_rq_clock(rq); |
ff95f3df | 6364 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6365 | if (!next) |
6366 | break; | |
79c53799 | 6367 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6368 | migrate_dead(dead_cpu, next); |
e692ab53 | 6369 | |
1da177e4 LT |
6370 | } |
6371 | } | |
6372 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6373 | ||
e692ab53 NP |
6374 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6375 | ||
6376 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6377 | { |
6378 | .procname = "sched_domain", | |
c57baf1e | 6379 | .mode = 0555, |
e0361851 | 6380 | }, |
38605cae | 6381 | {0, }, |
e692ab53 NP |
6382 | }; |
6383 | ||
6384 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6385 | { |
c57baf1e | 6386 | .ctl_name = CTL_KERN, |
e0361851 | 6387 | .procname = "kernel", |
c57baf1e | 6388 | .mode = 0555, |
e0361851 AD |
6389 | .child = sd_ctl_dir, |
6390 | }, | |
38605cae | 6391 | {0, }, |
e692ab53 NP |
6392 | }; |
6393 | ||
6394 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6395 | { | |
6396 | struct ctl_table *entry = | |
5cf9f062 | 6397 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6398 | |
e692ab53 NP |
6399 | return entry; |
6400 | } | |
6401 | ||
6382bc90 MM |
6402 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6403 | { | |
cd790076 | 6404 | struct ctl_table *entry; |
6382bc90 | 6405 | |
cd790076 MM |
6406 | /* |
6407 | * In the intermediate directories, both the child directory and | |
6408 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6409 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6410 | * static strings and all have proc handlers. |
6411 | */ | |
6412 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6413 | if (entry->child) |
6414 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6415 | if (entry->proc_handler == NULL) |
6416 | kfree(entry->procname); | |
6417 | } | |
6382bc90 MM |
6418 | |
6419 | kfree(*tablep); | |
6420 | *tablep = NULL; | |
6421 | } | |
6422 | ||
e692ab53 | 6423 | static void |
e0361851 | 6424 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6425 | const char *procname, void *data, int maxlen, |
6426 | mode_t mode, proc_handler *proc_handler) | |
6427 | { | |
e692ab53 NP |
6428 | entry->procname = procname; |
6429 | entry->data = data; | |
6430 | entry->maxlen = maxlen; | |
6431 | entry->mode = mode; | |
6432 | entry->proc_handler = proc_handler; | |
6433 | } | |
6434 | ||
6435 | static struct ctl_table * | |
6436 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6437 | { | |
a5d8c348 | 6438 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6439 | |
ad1cdc1d MM |
6440 | if (table == NULL) |
6441 | return NULL; | |
6442 | ||
e0361851 | 6443 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6444 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6445 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6446 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6447 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6448 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6449 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6450 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6451 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6452 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6453 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6454 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6455 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6456 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6457 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6458 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6459 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6460 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6461 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6462 | &sd->cache_nice_tries, |
6463 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6464 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6465 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6466 | set_table_entry(&table[11], "name", sd->name, |
6467 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6468 | /* &table[12] is terminator */ | |
e692ab53 NP |
6469 | |
6470 | return table; | |
6471 | } | |
6472 | ||
9a4e7159 | 6473 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6474 | { |
6475 | struct ctl_table *entry, *table; | |
6476 | struct sched_domain *sd; | |
6477 | int domain_num = 0, i; | |
6478 | char buf[32]; | |
6479 | ||
6480 | for_each_domain(cpu, sd) | |
6481 | domain_num++; | |
6482 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6483 | if (table == NULL) |
6484 | return NULL; | |
e692ab53 NP |
6485 | |
6486 | i = 0; | |
6487 | for_each_domain(cpu, sd) { | |
6488 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6489 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6490 | entry->mode = 0555; |
e692ab53 NP |
6491 | entry->child = sd_alloc_ctl_domain_table(sd); |
6492 | entry++; | |
6493 | i++; | |
6494 | } | |
6495 | return table; | |
6496 | } | |
6497 | ||
6498 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6499 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6500 | { |
6501 | int i, cpu_num = num_online_cpus(); | |
6502 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6503 | char buf[32]; | |
6504 | ||
7378547f MM |
6505 | WARN_ON(sd_ctl_dir[0].child); |
6506 | sd_ctl_dir[0].child = entry; | |
6507 | ||
ad1cdc1d MM |
6508 | if (entry == NULL) |
6509 | return; | |
6510 | ||
97b6ea7b | 6511 | for_each_online_cpu(i) { |
e692ab53 | 6512 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6513 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6514 | entry->mode = 0555; |
e692ab53 | 6515 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6516 | entry++; |
e692ab53 | 6517 | } |
7378547f MM |
6518 | |
6519 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6520 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6521 | } | |
6382bc90 | 6522 | |
7378547f | 6523 | /* may be called multiple times per register */ |
6382bc90 MM |
6524 | static void unregister_sched_domain_sysctl(void) |
6525 | { | |
7378547f MM |
6526 | if (sd_sysctl_header) |
6527 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6528 | sd_sysctl_header = NULL; |
7378547f MM |
6529 | if (sd_ctl_dir[0].child) |
6530 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6531 | } |
e692ab53 | 6532 | #else |
6382bc90 MM |
6533 | static void register_sched_domain_sysctl(void) |
6534 | { | |
6535 | } | |
6536 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6537 | { |
6538 | } | |
6539 | #endif | |
6540 | ||
1f11eb6a GH |
6541 | static void set_rq_online(struct rq *rq) |
6542 | { | |
6543 | if (!rq->online) { | |
6544 | const struct sched_class *class; | |
6545 | ||
c6c4927b | 6546 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6547 | rq->online = 1; |
6548 | ||
6549 | for_each_class(class) { | |
6550 | if (class->rq_online) | |
6551 | class->rq_online(rq); | |
6552 | } | |
6553 | } | |
6554 | } | |
6555 | ||
6556 | static void set_rq_offline(struct rq *rq) | |
6557 | { | |
6558 | if (rq->online) { | |
6559 | const struct sched_class *class; | |
6560 | ||
6561 | for_each_class(class) { | |
6562 | if (class->rq_offline) | |
6563 | class->rq_offline(rq); | |
6564 | } | |
6565 | ||
c6c4927b | 6566 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6567 | rq->online = 0; |
6568 | } | |
6569 | } | |
6570 | ||
1da177e4 LT |
6571 | /* |
6572 | * migration_call - callback that gets triggered when a CPU is added. | |
6573 | * Here we can start up the necessary migration thread for the new CPU. | |
6574 | */ | |
48f24c4d IM |
6575 | static int __cpuinit |
6576 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6577 | { |
1da177e4 | 6578 | struct task_struct *p; |
48f24c4d | 6579 | int cpu = (long)hcpu; |
1da177e4 | 6580 | unsigned long flags; |
70b97a7f | 6581 | struct rq *rq; |
1da177e4 LT |
6582 | |
6583 | switch (action) { | |
5be9361c | 6584 | |
1da177e4 | 6585 | case CPU_UP_PREPARE: |
8bb78442 | 6586 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6587 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6588 | if (IS_ERR(p)) |
6589 | return NOTIFY_BAD; | |
1da177e4 LT |
6590 | kthread_bind(p, cpu); |
6591 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6592 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6593 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6594 | task_rq_unlock(rq, &flags); |
6595 | cpu_rq(cpu)->migration_thread = p; | |
6596 | break; | |
48f24c4d | 6597 | |
1da177e4 | 6598 | case CPU_ONLINE: |
8bb78442 | 6599 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6600 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6601 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6602 | |
6603 | /* Update our root-domain */ | |
6604 | rq = cpu_rq(cpu); | |
6605 | spin_lock_irqsave(&rq->lock, flags); | |
6606 | if (rq->rd) { | |
c6c4927b | 6607 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6608 | |
6609 | set_rq_online(rq); | |
1f94ef59 GH |
6610 | } |
6611 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6612 | break; |
48f24c4d | 6613 | |
1da177e4 LT |
6614 | #ifdef CONFIG_HOTPLUG_CPU |
6615 | case CPU_UP_CANCELED: | |
8bb78442 | 6616 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6617 | if (!cpu_rq(cpu)->migration_thread) |
6618 | break; | |
41a2d6cf | 6619 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 6620 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 6621 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6622 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6623 | cpu_rq(cpu)->migration_thread = NULL; | |
6624 | break; | |
48f24c4d | 6625 | |
1da177e4 | 6626 | case CPU_DEAD: |
8bb78442 | 6627 | case CPU_DEAD_FROZEN: |
470fd646 | 6628 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6629 | migrate_live_tasks(cpu); |
6630 | rq = cpu_rq(cpu); | |
6631 | kthread_stop(rq->migration_thread); | |
6632 | rq->migration_thread = NULL; | |
6633 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6634 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6635 | update_rq_clock(rq); |
2e1cb74a | 6636 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6637 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6638 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6639 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6640 | migrate_dead_tasks(cpu); |
d2da272a | 6641 | spin_unlock_irq(&rq->lock); |
470fd646 | 6642 | cpuset_unlock(); |
1da177e4 LT |
6643 | migrate_nr_uninterruptible(rq); |
6644 | BUG_ON(rq->nr_running != 0); | |
6645 | ||
41a2d6cf IM |
6646 | /* |
6647 | * No need to migrate the tasks: it was best-effort if | |
6648 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6649 | * the requestors. | |
6650 | */ | |
1da177e4 LT |
6651 | spin_lock_irq(&rq->lock); |
6652 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6653 | struct migration_req *req; |
6654 | ||
1da177e4 | 6655 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6656 | struct migration_req, list); |
1da177e4 | 6657 | list_del_init(&req->list); |
9a2bd244 | 6658 | spin_unlock_irq(&rq->lock); |
1da177e4 | 6659 | complete(&req->done); |
9a2bd244 | 6660 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
6661 | } |
6662 | spin_unlock_irq(&rq->lock); | |
6663 | break; | |
57d885fe | 6664 | |
08f503b0 GH |
6665 | case CPU_DYING: |
6666 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6667 | /* Update our root-domain */ |
6668 | rq = cpu_rq(cpu); | |
6669 | spin_lock_irqsave(&rq->lock, flags); | |
6670 | if (rq->rd) { | |
c6c4927b | 6671 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6672 | set_rq_offline(rq); |
57d885fe GH |
6673 | } |
6674 | spin_unlock_irqrestore(&rq->lock, flags); | |
6675 | break; | |
1da177e4 LT |
6676 | #endif |
6677 | } | |
6678 | return NOTIFY_OK; | |
6679 | } | |
6680 | ||
6681 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6682 | * happens before everything else. | |
6683 | */ | |
26c2143b | 6684 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6685 | .notifier_call = migration_call, |
6686 | .priority = 10 | |
6687 | }; | |
6688 | ||
7babe8db | 6689 | static int __init migration_init(void) |
1da177e4 LT |
6690 | { |
6691 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6692 | int err; |
48f24c4d IM |
6693 | |
6694 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6695 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6696 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6697 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6698 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6699 | |
6700 | return err; | |
1da177e4 | 6701 | } |
7babe8db | 6702 | early_initcall(migration_init); |
1da177e4 LT |
6703 | #endif |
6704 | ||
6705 | #ifdef CONFIG_SMP | |
476f3534 | 6706 | |
3e9830dc | 6707 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6708 | |
7c16ec58 | 6709 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6710 | struct cpumask *groupmask) |
1da177e4 | 6711 | { |
4dcf6aff | 6712 | struct sched_group *group = sd->groups; |
434d53b0 | 6713 | char str[256]; |
1da177e4 | 6714 | |
968ea6d8 | 6715 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6716 | cpumask_clear(groupmask); |
4dcf6aff IM |
6717 | |
6718 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6719 | ||
6720 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6721 | printk("does not load-balance\n"); | |
6722 | if (sd->parent) | |
6723 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6724 | " has parent"); | |
6725 | return -1; | |
41c7ce9a NP |
6726 | } |
6727 | ||
eefd796a | 6728 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6729 | |
758b2cdc | 6730 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
6731 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6732 | "CPU%d\n", cpu); | |
6733 | } | |
758b2cdc | 6734 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
6735 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6736 | " CPU%d\n", cpu); | |
6737 | } | |
1da177e4 | 6738 | |
4dcf6aff | 6739 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6740 | do { |
4dcf6aff IM |
6741 | if (!group) { |
6742 | printk("\n"); | |
6743 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6744 | break; |
6745 | } | |
6746 | ||
4dcf6aff IM |
6747 | if (!group->__cpu_power) { |
6748 | printk(KERN_CONT "\n"); | |
6749 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6750 | "set\n"); | |
6751 | break; | |
6752 | } | |
1da177e4 | 6753 | |
758b2cdc | 6754 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
6755 | printk(KERN_CONT "\n"); |
6756 | printk(KERN_ERR "ERROR: empty group\n"); | |
6757 | break; | |
6758 | } | |
1da177e4 | 6759 | |
758b2cdc | 6760 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
6761 | printk(KERN_CONT "\n"); |
6762 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6763 | break; | |
6764 | } | |
1da177e4 | 6765 | |
758b2cdc | 6766 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6767 | |
968ea6d8 | 6768 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
4dcf6aff | 6769 | printk(KERN_CONT " %s", str); |
1da177e4 | 6770 | |
4dcf6aff IM |
6771 | group = group->next; |
6772 | } while (group != sd->groups); | |
6773 | printk(KERN_CONT "\n"); | |
1da177e4 | 6774 | |
758b2cdc | 6775 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 6776 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6777 | |
758b2cdc RR |
6778 | if (sd->parent && |
6779 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
6780 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6781 | "of domain->span\n"); | |
6782 | return 0; | |
6783 | } | |
1da177e4 | 6784 | |
4dcf6aff IM |
6785 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6786 | { | |
d5dd3db1 | 6787 | cpumask_var_t groupmask; |
4dcf6aff | 6788 | int level = 0; |
1da177e4 | 6789 | |
4dcf6aff IM |
6790 | if (!sd) { |
6791 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6792 | return; | |
6793 | } | |
1da177e4 | 6794 | |
4dcf6aff IM |
6795 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6796 | ||
d5dd3db1 | 6797 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6798 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6799 | return; | |
6800 | } | |
6801 | ||
4dcf6aff | 6802 | for (;;) { |
7c16ec58 | 6803 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6804 | break; |
1da177e4 LT |
6805 | level++; |
6806 | sd = sd->parent; | |
33859f7f | 6807 | if (!sd) |
4dcf6aff IM |
6808 | break; |
6809 | } | |
d5dd3db1 | 6810 | free_cpumask_var(groupmask); |
1da177e4 | 6811 | } |
6d6bc0ad | 6812 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6813 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6814 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6815 | |
1a20ff27 | 6816 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6817 | { |
758b2cdc | 6818 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6819 | return 1; |
6820 | ||
6821 | /* Following flags need at least 2 groups */ | |
6822 | if (sd->flags & (SD_LOAD_BALANCE | | |
6823 | SD_BALANCE_NEWIDLE | | |
6824 | SD_BALANCE_FORK | | |
89c4710e SS |
6825 | SD_BALANCE_EXEC | |
6826 | SD_SHARE_CPUPOWER | | |
6827 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6828 | if (sd->groups != sd->groups->next) |
6829 | return 0; | |
6830 | } | |
6831 | ||
6832 | /* Following flags don't use groups */ | |
6833 | if (sd->flags & (SD_WAKE_IDLE | | |
6834 | SD_WAKE_AFFINE | | |
6835 | SD_WAKE_BALANCE)) | |
6836 | return 0; | |
6837 | ||
6838 | return 1; | |
6839 | } | |
6840 | ||
48f24c4d IM |
6841 | static int |
6842 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6843 | { |
6844 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6845 | ||
6846 | if (sd_degenerate(parent)) | |
6847 | return 1; | |
6848 | ||
758b2cdc | 6849 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6850 | return 0; |
6851 | ||
6852 | /* Does parent contain flags not in child? */ | |
6853 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6854 | if (cflags & SD_WAKE_AFFINE) | |
6855 | pflags &= ~SD_WAKE_BALANCE; | |
6856 | /* Flags needing groups don't count if only 1 group in parent */ | |
6857 | if (parent->groups == parent->groups->next) { | |
6858 | pflags &= ~(SD_LOAD_BALANCE | | |
6859 | SD_BALANCE_NEWIDLE | | |
6860 | SD_BALANCE_FORK | | |
89c4710e SS |
6861 | SD_BALANCE_EXEC | |
6862 | SD_SHARE_CPUPOWER | | |
6863 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6864 | if (nr_node_ids == 1) |
6865 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6866 | } |
6867 | if (~cflags & pflags) | |
6868 | return 0; | |
6869 | ||
6870 | return 1; | |
6871 | } | |
6872 | ||
c6c4927b RR |
6873 | static void free_rootdomain(struct root_domain *rd) |
6874 | { | |
68e74568 RR |
6875 | cpupri_cleanup(&rd->cpupri); |
6876 | ||
c6c4927b RR |
6877 | free_cpumask_var(rd->rto_mask); |
6878 | free_cpumask_var(rd->online); | |
6879 | free_cpumask_var(rd->span); | |
6880 | kfree(rd); | |
6881 | } | |
6882 | ||
57d885fe GH |
6883 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6884 | { | |
6885 | unsigned long flags; | |
57d885fe GH |
6886 | |
6887 | spin_lock_irqsave(&rq->lock, flags); | |
6888 | ||
6889 | if (rq->rd) { | |
6890 | struct root_domain *old_rd = rq->rd; | |
6891 | ||
c6c4927b | 6892 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6893 | set_rq_offline(rq); |
57d885fe | 6894 | |
c6c4927b | 6895 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6896 | |
57d885fe | 6897 | if (atomic_dec_and_test(&old_rd->refcount)) |
c6c4927b | 6898 | free_rootdomain(old_rd); |
57d885fe GH |
6899 | } |
6900 | ||
6901 | atomic_inc(&rd->refcount); | |
6902 | rq->rd = rd; | |
6903 | ||
c6c4927b RR |
6904 | cpumask_set_cpu(rq->cpu, rd->span); |
6905 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 6906 | set_rq_online(rq); |
57d885fe GH |
6907 | |
6908 | spin_unlock_irqrestore(&rq->lock, flags); | |
6909 | } | |
6910 | ||
c6c4927b | 6911 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe GH |
6912 | { |
6913 | memset(rd, 0, sizeof(*rd)); | |
6914 | ||
c6c4927b RR |
6915 | if (bootmem) { |
6916 | alloc_bootmem_cpumask_var(&def_root_domain.span); | |
6917 | alloc_bootmem_cpumask_var(&def_root_domain.online); | |
6918 | alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); | |
68e74568 | 6919 | cpupri_init(&rd->cpupri, true); |
c6c4927b RR |
6920 | return 0; |
6921 | } | |
6922 | ||
6923 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | |
6924 | goto free_rd; | |
6925 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) | |
6926 | goto free_span; | |
6927 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | |
6928 | goto free_online; | |
6e0534f2 | 6929 | |
68e74568 RR |
6930 | if (cpupri_init(&rd->cpupri, false) != 0) |
6931 | goto free_rto_mask; | |
c6c4927b | 6932 | return 0; |
6e0534f2 | 6933 | |
68e74568 RR |
6934 | free_rto_mask: |
6935 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6936 | free_online: |
6937 | free_cpumask_var(rd->online); | |
6938 | free_span: | |
6939 | free_cpumask_var(rd->span); | |
6940 | free_rd: | |
6941 | kfree(rd); | |
6942 | return -ENOMEM; | |
57d885fe GH |
6943 | } |
6944 | ||
6945 | static void init_defrootdomain(void) | |
6946 | { | |
c6c4927b RR |
6947 | init_rootdomain(&def_root_domain, true); |
6948 | ||
57d885fe GH |
6949 | atomic_set(&def_root_domain.refcount, 1); |
6950 | } | |
6951 | ||
dc938520 | 6952 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6953 | { |
6954 | struct root_domain *rd; | |
6955 | ||
6956 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6957 | if (!rd) | |
6958 | return NULL; | |
6959 | ||
c6c4927b RR |
6960 | if (init_rootdomain(rd, false) != 0) { |
6961 | kfree(rd); | |
6962 | return NULL; | |
6963 | } | |
57d885fe GH |
6964 | |
6965 | return rd; | |
6966 | } | |
6967 | ||
1da177e4 | 6968 | /* |
0eab9146 | 6969 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6970 | * hold the hotplug lock. |
6971 | */ | |
0eab9146 IM |
6972 | static void |
6973 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6974 | { |
70b97a7f | 6975 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6976 | struct sched_domain *tmp; |
6977 | ||
6978 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6979 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6980 | struct sched_domain *parent = tmp->parent; |
6981 | if (!parent) | |
6982 | break; | |
f29c9b1c | 6983 | |
1a848870 | 6984 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6985 | tmp->parent = parent->parent; |
1a848870 SS |
6986 | if (parent->parent) |
6987 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6988 | } else |
6989 | tmp = tmp->parent; | |
245af2c7 SS |
6990 | } |
6991 | ||
1a848870 | 6992 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6993 | sd = sd->parent; |
1a848870 SS |
6994 | if (sd) |
6995 | sd->child = NULL; | |
6996 | } | |
1da177e4 LT |
6997 | |
6998 | sched_domain_debug(sd, cpu); | |
6999 | ||
57d885fe | 7000 | rq_attach_root(rq, rd); |
674311d5 | 7001 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7002 | } |
7003 | ||
7004 | /* cpus with isolated domains */ | |
dcc30a35 | 7005 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7006 | |
7007 | /* Setup the mask of cpus configured for isolated domains */ | |
7008 | static int __init isolated_cpu_setup(char *str) | |
7009 | { | |
968ea6d8 | 7010 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7011 | return 1; |
7012 | } | |
7013 | ||
8927f494 | 7014 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7015 | |
7016 | /* | |
6711cab4 SS |
7017 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7018 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7019 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7020 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7021 | * |
7022 | * init_sched_build_groups will build a circular linked list of the groups | |
7023 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7024 | * and ->cpu_power to 0. | |
7025 | */ | |
a616058b | 7026 | static void |
96f874e2 RR |
7027 | init_sched_build_groups(const struct cpumask *span, |
7028 | const struct cpumask *cpu_map, | |
7029 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7030 | struct sched_group **sg, |
96f874e2 RR |
7031 | struct cpumask *tmpmask), |
7032 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7033 | { |
7034 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7035 | int i; |
7036 | ||
96f874e2 | 7037 | cpumask_clear(covered); |
7c16ec58 | 7038 | |
abcd083a | 7039 | for_each_cpu(i, span) { |
6711cab4 | 7040 | struct sched_group *sg; |
7c16ec58 | 7041 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7042 | int j; |
7043 | ||
758b2cdc | 7044 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7045 | continue; |
7046 | ||
758b2cdc | 7047 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7048 | sg->__cpu_power = 0; |
1da177e4 | 7049 | |
abcd083a | 7050 | for_each_cpu(j, span) { |
7c16ec58 | 7051 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7052 | continue; |
7053 | ||
96f874e2 | 7054 | cpumask_set_cpu(j, covered); |
758b2cdc | 7055 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7056 | } |
7057 | if (!first) | |
7058 | first = sg; | |
7059 | if (last) | |
7060 | last->next = sg; | |
7061 | last = sg; | |
7062 | } | |
7063 | last->next = first; | |
7064 | } | |
7065 | ||
9c1cfda2 | 7066 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7067 | |
9c1cfda2 | 7068 | #ifdef CONFIG_NUMA |
198e2f18 | 7069 | |
9c1cfda2 JH |
7070 | /** |
7071 | * find_next_best_node - find the next node to include in a sched_domain | |
7072 | * @node: node whose sched_domain we're building | |
7073 | * @used_nodes: nodes already in the sched_domain | |
7074 | * | |
41a2d6cf | 7075 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7076 | * finds the closest node not already in the @used_nodes map. |
7077 | * | |
7078 | * Should use nodemask_t. | |
7079 | */ | |
c5f59f08 | 7080 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7081 | { |
7082 | int i, n, val, min_val, best_node = 0; | |
7083 | ||
7084 | min_val = INT_MAX; | |
7085 | ||
076ac2af | 7086 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7087 | /* Start at @node */ |
076ac2af | 7088 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7089 | |
7090 | if (!nr_cpus_node(n)) | |
7091 | continue; | |
7092 | ||
7093 | /* Skip already used nodes */ | |
c5f59f08 | 7094 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7095 | continue; |
7096 | ||
7097 | /* Simple min distance search */ | |
7098 | val = node_distance(node, n); | |
7099 | ||
7100 | if (val < min_val) { | |
7101 | min_val = val; | |
7102 | best_node = n; | |
7103 | } | |
7104 | } | |
7105 | ||
c5f59f08 | 7106 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7107 | return best_node; |
7108 | } | |
7109 | ||
7110 | /** | |
7111 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7112 | * @node: node whose cpumask we're constructing | |
73486722 | 7113 | * @span: resulting cpumask |
9c1cfda2 | 7114 | * |
41a2d6cf | 7115 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7116 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7117 | * out optimally. | |
7118 | */ | |
96f874e2 | 7119 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7120 | { |
c5f59f08 | 7121 | nodemask_t used_nodes; |
96f874e2 | 7122 | /* FIXME: use cpumask_of_node() */ |
c5f59f08 | 7123 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 7124 | int i; |
9c1cfda2 | 7125 | |
4bdbaad3 | 7126 | cpus_clear(*span); |
c5f59f08 | 7127 | nodes_clear(used_nodes); |
9c1cfda2 | 7128 | |
4bdbaad3 | 7129 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 7130 | node_set(node, used_nodes); |
9c1cfda2 JH |
7131 | |
7132 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7133 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7134 | |
c5f59f08 | 7135 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 7136 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 7137 | } |
9c1cfda2 | 7138 | } |
6d6bc0ad | 7139 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7140 | |
5c45bf27 | 7141 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7142 | |
6c99e9ad RR |
7143 | /* |
7144 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
7145 | * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space | |
7146 | * for nr_cpu_ids < CONFIG_NR_CPUS. | |
7147 | */ | |
7148 | struct static_sched_group { | |
7149 | struct sched_group sg; | |
7150 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
7151 | }; | |
7152 | ||
7153 | struct static_sched_domain { | |
7154 | struct sched_domain sd; | |
7155 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
7156 | }; | |
7157 | ||
9c1cfda2 | 7158 | /* |
48f24c4d | 7159 | * SMT sched-domains: |
9c1cfda2 | 7160 | */ |
1da177e4 | 7161 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
7162 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
7163 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 7164 | |
41a2d6cf | 7165 | static int |
96f874e2 RR |
7166 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
7167 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 7168 | { |
6711cab4 | 7169 | if (sg) |
6c99e9ad | 7170 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
7171 | return cpu; |
7172 | } | |
6d6bc0ad | 7173 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7174 | |
48f24c4d IM |
7175 | /* |
7176 | * multi-core sched-domains: | |
7177 | */ | |
1e9f28fa | 7178 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
7179 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
7180 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 7181 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7182 | |
7183 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7184 | static int |
96f874e2 RR |
7185 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7186 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 7187 | { |
6711cab4 | 7188 | int group; |
7c16ec58 | 7189 | |
96f874e2 RR |
7190 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7191 | group = cpumask_first(mask); | |
6711cab4 | 7192 | if (sg) |
6c99e9ad | 7193 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 7194 | return group; |
1e9f28fa SS |
7195 | } |
7196 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7197 | static int |
96f874e2 RR |
7198 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7199 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 7200 | { |
6711cab4 | 7201 | if (sg) |
6c99e9ad | 7202 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
7203 | return cpu; |
7204 | } | |
7205 | #endif | |
7206 | ||
6c99e9ad RR |
7207 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
7208 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 7209 | |
41a2d6cf | 7210 | static int |
96f874e2 RR |
7211 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
7212 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 7213 | { |
6711cab4 | 7214 | int group; |
48f24c4d | 7215 | #ifdef CONFIG_SCHED_MC |
96f874e2 | 7216 | /* FIXME: Use cpu_coregroup_mask. */ |
7c16ec58 MT |
7217 | *mask = cpu_coregroup_map(cpu); |
7218 | cpus_and(*mask, *mask, *cpu_map); | |
96f874e2 | 7219 | group = cpumask_first(mask); |
1e9f28fa | 7220 | #elif defined(CONFIG_SCHED_SMT) |
96f874e2 RR |
7221 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7222 | group = cpumask_first(mask); | |
1da177e4 | 7223 | #else |
6711cab4 | 7224 | group = cpu; |
1da177e4 | 7225 | #endif |
6711cab4 | 7226 | if (sg) |
6c99e9ad | 7227 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 7228 | return group; |
1da177e4 LT |
7229 | } |
7230 | ||
7231 | #ifdef CONFIG_NUMA | |
1da177e4 | 7232 | /* |
9c1cfda2 JH |
7233 | * The init_sched_build_groups can't handle what we want to do with node |
7234 | * groups, so roll our own. Now each node has its own list of groups which | |
7235 | * gets dynamically allocated. | |
1da177e4 | 7236 | */ |
9c1cfda2 | 7237 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 7238 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7239 | |
9c1cfda2 | 7240 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6c99e9ad | 7241 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 7242 | |
96f874e2 RR |
7243 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
7244 | struct sched_group **sg, | |
7245 | struct cpumask *nodemask) | |
9c1cfda2 | 7246 | { |
6711cab4 | 7247 | int group; |
96f874e2 | 7248 | /* FIXME: use cpumask_of_node */ |
ea6f18ed | 7249 | node_to_cpumask_ptr(pnodemask, cpu_to_node(cpu)); |
6711cab4 | 7250 | |
96f874e2 RR |
7251 | cpumask_and(nodemask, pnodemask, cpu_map); |
7252 | group = cpumask_first(nodemask); | |
6711cab4 SS |
7253 | |
7254 | if (sg) | |
6c99e9ad | 7255 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7256 | return group; |
1da177e4 | 7257 | } |
6711cab4 | 7258 | |
08069033 SS |
7259 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7260 | { | |
7261 | struct sched_group *sg = group_head; | |
7262 | int j; | |
7263 | ||
7264 | if (!sg) | |
7265 | return; | |
3a5c359a | 7266 | do { |
758b2cdc | 7267 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7268 | struct sched_domain *sd; |
08069033 | 7269 | |
6c99e9ad | 7270 | sd = &per_cpu(phys_domains, j).sd; |
758b2cdc | 7271 | if (j != cpumask_first(sched_group_cpus(sd->groups))) { |
3a5c359a AK |
7272 | /* |
7273 | * Only add "power" once for each | |
7274 | * physical package. | |
7275 | */ | |
7276 | continue; | |
7277 | } | |
08069033 | 7278 | |
3a5c359a AK |
7279 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7280 | } | |
7281 | sg = sg->next; | |
7282 | } while (sg != group_head); | |
08069033 | 7283 | } |
6d6bc0ad | 7284 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7285 | |
a616058b | 7286 | #ifdef CONFIG_NUMA |
51888ca2 | 7287 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7288 | static void free_sched_groups(const struct cpumask *cpu_map, |
7289 | struct cpumask *nodemask) | |
51888ca2 | 7290 | { |
a616058b | 7291 | int cpu, i; |
51888ca2 | 7292 | |
abcd083a | 7293 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7294 | struct sched_group **sched_group_nodes |
7295 | = sched_group_nodes_bycpu[cpu]; | |
7296 | ||
51888ca2 SV |
7297 | if (!sched_group_nodes) |
7298 | continue; | |
7299 | ||
076ac2af | 7300 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 | 7301 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
96f874e2 | 7302 | /* FIXME: Use cpumask_of_node */ |
ea6f18ed | 7303 | node_to_cpumask_ptr(pnodemask, i); |
51888ca2 | 7304 | |
ea6f18ed | 7305 | cpus_and(*nodemask, *pnodemask, *cpu_map); |
96f874e2 | 7306 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7307 | continue; |
7308 | ||
7309 | if (sg == NULL) | |
7310 | continue; | |
7311 | sg = sg->next; | |
7312 | next_sg: | |
7313 | oldsg = sg; | |
7314 | sg = sg->next; | |
7315 | kfree(oldsg); | |
7316 | if (oldsg != sched_group_nodes[i]) | |
7317 | goto next_sg; | |
7318 | } | |
7319 | kfree(sched_group_nodes); | |
7320 | sched_group_nodes_bycpu[cpu] = NULL; | |
7321 | } | |
51888ca2 | 7322 | } |
6d6bc0ad | 7323 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7324 | static void free_sched_groups(const struct cpumask *cpu_map, |
7325 | struct cpumask *nodemask) | |
a616058b SS |
7326 | { |
7327 | } | |
6d6bc0ad | 7328 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7329 | |
89c4710e SS |
7330 | /* |
7331 | * Initialize sched groups cpu_power. | |
7332 | * | |
7333 | * cpu_power indicates the capacity of sched group, which is used while | |
7334 | * distributing the load between different sched groups in a sched domain. | |
7335 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7336 | * there are asymmetries in the topology. If there are asymmetries, group | |
7337 | * having more cpu_power will pickup more load compared to the group having | |
7338 | * less cpu_power. | |
7339 | * | |
7340 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7341 | * the maximum number of tasks a group can handle in the presence of other idle | |
7342 | * or lightly loaded groups in the same sched domain. | |
7343 | */ | |
7344 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7345 | { | |
7346 | struct sched_domain *child; | |
7347 | struct sched_group *group; | |
7348 | ||
7349 | WARN_ON(!sd || !sd->groups); | |
7350 | ||
758b2cdc | 7351 | if (cpu != cpumask_first(sched_group_cpus(sd->groups))) |
89c4710e SS |
7352 | return; |
7353 | ||
7354 | child = sd->child; | |
7355 | ||
5517d86b ED |
7356 | sd->groups->__cpu_power = 0; |
7357 | ||
89c4710e SS |
7358 | /* |
7359 | * For perf policy, if the groups in child domain share resources | |
7360 | * (for example cores sharing some portions of the cache hierarchy | |
7361 | * or SMT), then set this domain groups cpu_power such that each group | |
7362 | * can handle only one task, when there are other idle groups in the | |
7363 | * same sched domain. | |
7364 | */ | |
7365 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7366 | (child->flags & | |
7367 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7368 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7369 | return; |
7370 | } | |
7371 | ||
89c4710e SS |
7372 | /* |
7373 | * add cpu_power of each child group to this groups cpu_power | |
7374 | */ | |
7375 | group = child->groups; | |
7376 | do { | |
5517d86b | 7377 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7378 | group = group->next; |
7379 | } while (group != child->groups); | |
7380 | } | |
7381 | ||
7c16ec58 MT |
7382 | /* |
7383 | * Initializers for schedule domains | |
7384 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7385 | */ | |
7386 | ||
a5d8c348 IM |
7387 | #ifdef CONFIG_SCHED_DEBUG |
7388 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7389 | #else | |
7390 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7391 | #endif | |
7392 | ||
7c16ec58 | 7393 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7394 | |
7c16ec58 MT |
7395 | #define SD_INIT_FUNC(type) \ |
7396 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7397 | { \ | |
7398 | memset(sd, 0, sizeof(*sd)); \ | |
7399 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7400 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7401 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7402 | } |
7403 | ||
7404 | SD_INIT_FUNC(CPU) | |
7405 | #ifdef CONFIG_NUMA | |
7406 | SD_INIT_FUNC(ALLNODES) | |
7407 | SD_INIT_FUNC(NODE) | |
7408 | #endif | |
7409 | #ifdef CONFIG_SCHED_SMT | |
7410 | SD_INIT_FUNC(SIBLING) | |
7411 | #endif | |
7412 | #ifdef CONFIG_SCHED_MC | |
7413 | SD_INIT_FUNC(MC) | |
7414 | #endif | |
7415 | ||
1d3504fc HS |
7416 | static int default_relax_domain_level = -1; |
7417 | ||
7418 | static int __init setup_relax_domain_level(char *str) | |
7419 | { | |
30e0e178 LZ |
7420 | unsigned long val; |
7421 | ||
7422 | val = simple_strtoul(str, NULL, 0); | |
7423 | if (val < SD_LV_MAX) | |
7424 | default_relax_domain_level = val; | |
7425 | ||
1d3504fc HS |
7426 | return 1; |
7427 | } | |
7428 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7429 | ||
7430 | static void set_domain_attribute(struct sched_domain *sd, | |
7431 | struct sched_domain_attr *attr) | |
7432 | { | |
7433 | int request; | |
7434 | ||
7435 | if (!attr || attr->relax_domain_level < 0) { | |
7436 | if (default_relax_domain_level < 0) | |
7437 | return; | |
7438 | else | |
7439 | request = default_relax_domain_level; | |
7440 | } else | |
7441 | request = attr->relax_domain_level; | |
7442 | if (request < sd->level) { | |
7443 | /* turn off idle balance on this domain */ | |
7444 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7445 | } else { | |
7446 | /* turn on idle balance on this domain */ | |
7447 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7448 | } | |
7449 | } | |
7450 | ||
1da177e4 | 7451 | /* |
1a20ff27 DG |
7452 | * Build sched domains for a given set of cpus and attach the sched domains |
7453 | * to the individual cpus | |
1da177e4 | 7454 | */ |
96f874e2 | 7455 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 7456 | struct sched_domain_attr *attr) |
1da177e4 | 7457 | { |
3404c8d9 | 7458 | int i, err = -ENOMEM; |
57d885fe | 7459 | struct root_domain *rd; |
3404c8d9 RR |
7460 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
7461 | tmpmask; | |
d1b55138 | 7462 | #ifdef CONFIG_NUMA |
3404c8d9 | 7463 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 7464 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 7465 | int sd_allnodes = 0; |
d1b55138 | 7466 | |
3404c8d9 RR |
7467 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
7468 | goto out; | |
7469 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
7470 | goto free_domainspan; | |
7471 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
7472 | goto free_covered; | |
7473 | #endif | |
7474 | ||
7475 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
7476 | goto free_notcovered; | |
7477 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
7478 | goto free_nodemask; | |
7479 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
7480 | goto free_this_sibling_map; | |
7481 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
7482 | goto free_this_core_map; | |
7483 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
7484 | goto free_send_covered; | |
7485 | ||
7486 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
7487 | /* |
7488 | * Allocate the per-node list of sched groups | |
7489 | */ | |
076ac2af | 7490 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7491 | GFP_KERNEL); |
d1b55138 JH |
7492 | if (!sched_group_nodes) { |
7493 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 7494 | goto free_tmpmask; |
d1b55138 | 7495 | } |
d1b55138 | 7496 | #endif |
1da177e4 | 7497 | |
dc938520 | 7498 | rd = alloc_rootdomain(); |
57d885fe GH |
7499 | if (!rd) { |
7500 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 7501 | goto free_sched_groups; |
57d885fe GH |
7502 | } |
7503 | ||
7c16ec58 | 7504 | #ifdef CONFIG_NUMA |
96f874e2 | 7505 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
7506 | #endif |
7507 | ||
1da177e4 | 7508 | /* |
1a20ff27 | 7509 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7510 | */ |
abcd083a | 7511 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7512 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 7513 | |
96f874e2 | 7514 | /* FIXME: use cpumask_of_node */ |
7c16ec58 MT |
7515 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7516 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7517 | |
7518 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
7519 | if (cpumask_weight(cpu_map) > |
7520 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
9c1cfda2 | 7521 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7522 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7523 | set_domain_attribute(sd, attr); |
758b2cdc | 7524 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 7525 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7526 | p = sd; |
6711cab4 | 7527 | sd_allnodes = 1; |
9c1cfda2 JH |
7528 | } else |
7529 | p = NULL; | |
7530 | ||
1da177e4 | 7531 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7532 | SD_INIT(sd, NODE); |
1d3504fc | 7533 | set_domain_attribute(sd, attr); |
758b2cdc | 7534 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 7535 | sd->parent = p; |
1a848870 SS |
7536 | if (p) |
7537 | p->child = sd; | |
758b2cdc RR |
7538 | cpumask_and(sched_domain_span(sd), |
7539 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
7540 | #endif |
7541 | ||
7542 | p = sd; | |
6c99e9ad | 7543 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 7544 | SD_INIT(sd, CPU); |
1d3504fc | 7545 | set_domain_attribute(sd, attr); |
758b2cdc | 7546 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 7547 | sd->parent = p; |
1a848870 SS |
7548 | if (p) |
7549 | p->child = sd; | |
7c16ec58 | 7550 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7551 | |
1e9f28fa SS |
7552 | #ifdef CONFIG_SCHED_MC |
7553 | p = sd; | |
6c99e9ad | 7554 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 7555 | SD_INIT(sd, MC); |
1d3504fc | 7556 | set_domain_attribute(sd, attr); |
758b2cdc RR |
7557 | *sched_domain_span(sd) = cpu_coregroup_map(i); |
7558 | cpumask_and(sched_domain_span(sd), | |
7559 | sched_domain_span(sd), cpu_map); | |
1e9f28fa | 7560 | sd->parent = p; |
1a848870 | 7561 | p->child = sd; |
7c16ec58 | 7562 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7563 | #endif |
7564 | ||
1da177e4 LT |
7565 | #ifdef CONFIG_SCHED_SMT |
7566 | p = sd; | |
6c99e9ad | 7567 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 7568 | SD_INIT(sd, SIBLING); |
1d3504fc | 7569 | set_domain_attribute(sd, attr); |
758b2cdc RR |
7570 | cpumask_and(sched_domain_span(sd), |
7571 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
1da177e4 | 7572 | sd->parent = p; |
1a848870 | 7573 | p->child = sd; |
7c16ec58 | 7574 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7575 | #endif |
7576 | } | |
7577 | ||
7578 | #ifdef CONFIG_SCHED_SMT | |
7579 | /* Set up CPU (sibling) groups */ | |
abcd083a | 7580 | for_each_cpu(i, cpu_map) { |
96f874e2 RR |
7581 | cpumask_and(this_sibling_map, |
7582 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
7583 | if (i != cpumask_first(this_sibling_map)) | |
1da177e4 LT |
7584 | continue; |
7585 | ||
dd41f596 | 7586 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7587 | &cpu_to_cpu_group, |
7588 | send_covered, tmpmask); | |
1da177e4 LT |
7589 | } |
7590 | #endif | |
7591 | ||
1e9f28fa SS |
7592 | #ifdef CONFIG_SCHED_MC |
7593 | /* Set up multi-core groups */ | |
abcd083a | 7594 | for_each_cpu(i, cpu_map) { |
96f874e2 | 7595 | /* FIXME: Use cpu_coregroup_mask */ |
7c16ec58 MT |
7596 | *this_core_map = cpu_coregroup_map(i); |
7597 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
96f874e2 | 7598 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 7599 | continue; |
7c16ec58 | 7600 | |
dd41f596 | 7601 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7602 | &cpu_to_core_group, |
7603 | send_covered, tmpmask); | |
1e9f28fa SS |
7604 | } |
7605 | #endif | |
7606 | ||
1da177e4 | 7607 | /* Set up physical groups */ |
076ac2af | 7608 | for (i = 0; i < nr_node_ids; i++) { |
96f874e2 | 7609 | /* FIXME: Use cpumask_of_node */ |
7c16ec58 MT |
7610 | *nodemask = node_to_cpumask(i); |
7611 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
96f874e2 | 7612 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
7613 | continue; |
7614 | ||
7c16ec58 MT |
7615 | init_sched_build_groups(nodemask, cpu_map, |
7616 | &cpu_to_phys_group, | |
7617 | send_covered, tmpmask); | |
1da177e4 LT |
7618 | } |
7619 | ||
7620 | #ifdef CONFIG_NUMA | |
7621 | /* Set up node groups */ | |
7c16ec58 | 7622 | if (sd_allnodes) { |
7c16ec58 MT |
7623 | init_sched_build_groups(cpu_map, cpu_map, |
7624 | &cpu_to_allnodes_group, | |
7625 | send_covered, tmpmask); | |
7626 | } | |
9c1cfda2 | 7627 | |
076ac2af | 7628 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7629 | /* Set up node groups */ |
7630 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
7631 | int j; |
7632 | ||
96f874e2 | 7633 | /* FIXME: Use cpumask_of_node */ |
7c16ec58 | 7634 | *nodemask = node_to_cpumask(i); |
96f874e2 | 7635 | cpumask_clear(covered); |
7c16ec58 MT |
7636 | |
7637 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
96f874e2 | 7638 | if (cpumask_empty(nodemask)) { |
d1b55138 | 7639 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7640 | continue; |
d1b55138 | 7641 | } |
9c1cfda2 | 7642 | |
4bdbaad3 | 7643 | sched_domain_node_span(i, domainspan); |
96f874e2 | 7644 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 7645 | |
6c99e9ad RR |
7646 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
7647 | GFP_KERNEL, i); | |
51888ca2 SV |
7648 | if (!sg) { |
7649 | printk(KERN_WARNING "Can not alloc domain group for " | |
7650 | "node %d\n", i); | |
7651 | goto error; | |
7652 | } | |
9c1cfda2 | 7653 | sched_group_nodes[i] = sg; |
abcd083a | 7654 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 7655 | struct sched_domain *sd; |
9761eea8 | 7656 | |
9c1cfda2 JH |
7657 | sd = &per_cpu(node_domains, j); |
7658 | sd->groups = sg; | |
9c1cfda2 | 7659 | } |
5517d86b | 7660 | sg->__cpu_power = 0; |
758b2cdc | 7661 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 7662 | sg->next = sg; |
96f874e2 | 7663 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
7664 | prev = sg; |
7665 | ||
076ac2af | 7666 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 7667 | int n = (i + j) % nr_node_ids; |
96f874e2 | 7668 | /* FIXME: Use cpumask_of_node */ |
c5f59f08 | 7669 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7670 | |
96f874e2 RR |
7671 | cpumask_complement(notcovered, covered); |
7672 | cpumask_and(tmpmask, notcovered, cpu_map); | |
7673 | cpumask_and(tmpmask, tmpmask, domainspan); | |
7674 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
7675 | break; |
7676 | ||
96f874e2 RR |
7677 | cpumask_and(tmpmask, tmpmask, pnodemask); |
7678 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
7679 | continue; |
7680 | ||
6c99e9ad RR |
7681 | sg = kmalloc_node(sizeof(struct sched_group) + |
7682 | cpumask_size(), | |
15f0b676 | 7683 | GFP_KERNEL, i); |
9c1cfda2 JH |
7684 | if (!sg) { |
7685 | printk(KERN_WARNING | |
7686 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7687 | goto error; |
9c1cfda2 | 7688 | } |
5517d86b | 7689 | sg->__cpu_power = 0; |
758b2cdc | 7690 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 7691 | sg->next = prev->next; |
96f874e2 | 7692 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
7693 | prev->next = sg; |
7694 | prev = sg; | |
7695 | } | |
9c1cfda2 | 7696 | } |
1da177e4 LT |
7697 | #endif |
7698 | ||
7699 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7700 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7701 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 7702 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 7703 | |
89c4710e | 7704 | init_sched_groups_power(i, sd); |
5c45bf27 | 7705 | } |
1da177e4 | 7706 | #endif |
1e9f28fa | 7707 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7708 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 7709 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 7710 | |
89c4710e | 7711 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7712 | } |
7713 | #endif | |
1e9f28fa | 7714 | |
abcd083a | 7715 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 7716 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 7717 | |
89c4710e | 7718 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7719 | } |
7720 | ||
9c1cfda2 | 7721 | #ifdef CONFIG_NUMA |
076ac2af | 7722 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7723 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7724 | |
6711cab4 SS |
7725 | if (sd_allnodes) { |
7726 | struct sched_group *sg; | |
f712c0c7 | 7727 | |
96f874e2 | 7728 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 7729 | tmpmask); |
f712c0c7 SS |
7730 | init_numa_sched_groups_power(sg); |
7731 | } | |
9c1cfda2 JH |
7732 | #endif |
7733 | ||
1da177e4 | 7734 | /* Attach the domains */ |
abcd083a | 7735 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
7736 | struct sched_domain *sd; |
7737 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 7738 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7739 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7740 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7741 | #else |
6c99e9ad | 7742 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7743 | #endif |
57d885fe | 7744 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7745 | } |
51888ca2 | 7746 | |
3404c8d9 RR |
7747 | err = 0; |
7748 | ||
7749 | free_tmpmask: | |
7750 | free_cpumask_var(tmpmask); | |
7751 | free_send_covered: | |
7752 | free_cpumask_var(send_covered); | |
7753 | free_this_core_map: | |
7754 | free_cpumask_var(this_core_map); | |
7755 | free_this_sibling_map: | |
7756 | free_cpumask_var(this_sibling_map); | |
7757 | free_nodemask: | |
7758 | free_cpumask_var(nodemask); | |
7759 | free_notcovered: | |
7760 | #ifdef CONFIG_NUMA | |
7761 | free_cpumask_var(notcovered); | |
7762 | free_covered: | |
7763 | free_cpumask_var(covered); | |
7764 | free_domainspan: | |
7765 | free_cpumask_var(domainspan); | |
7766 | out: | |
7767 | #endif | |
7768 | return err; | |
7769 | ||
7770 | free_sched_groups: | |
7771 | #ifdef CONFIG_NUMA | |
7772 | kfree(sched_group_nodes); | |
7773 | #endif | |
7774 | goto free_tmpmask; | |
51888ca2 | 7775 | |
a616058b | 7776 | #ifdef CONFIG_NUMA |
51888ca2 | 7777 | error: |
7c16ec58 | 7778 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 7779 | free_rootdomain(rd); |
3404c8d9 | 7780 | goto free_tmpmask; |
a616058b | 7781 | #endif |
1da177e4 | 7782 | } |
029190c5 | 7783 | |
96f874e2 | 7784 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7785 | { |
7786 | return __build_sched_domains(cpu_map, NULL); | |
7787 | } | |
7788 | ||
96f874e2 | 7789 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 7790 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7791 | static struct sched_domain_attr *dattr_cur; |
7792 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7793 | |
7794 | /* | |
7795 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7796 | * cpumask) fails, then fallback to a single sched domain, |
7797 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7798 | */ |
4212823f | 7799 | static cpumask_var_t fallback_doms; |
029190c5 | 7800 | |
ee79d1bd HC |
7801 | /* |
7802 | * arch_update_cpu_topology lets virtualized architectures update the | |
7803 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7804 | * or 0 if it stayed the same. | |
7805 | */ | |
7806 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7807 | { |
ee79d1bd | 7808 | return 0; |
22e52b07 HC |
7809 | } |
7810 | ||
1a20ff27 | 7811 | /* |
41a2d6cf | 7812 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7813 | * For now this just excludes isolated cpus, but could be used to |
7814 | * exclude other special cases in the future. | |
1a20ff27 | 7815 | */ |
96f874e2 | 7816 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7817 | { |
7378547f MM |
7818 | int err; |
7819 | ||
22e52b07 | 7820 | arch_update_cpu_topology(); |
029190c5 | 7821 | ndoms_cur = 1; |
96f874e2 | 7822 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 7823 | if (!doms_cur) |
4212823f | 7824 | doms_cur = fallback_doms; |
dcc30a35 | 7825 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 7826 | dattr_cur = NULL; |
7378547f | 7827 | err = build_sched_domains(doms_cur); |
6382bc90 | 7828 | register_sched_domain_sysctl(); |
7378547f MM |
7829 | |
7830 | return err; | |
1a20ff27 DG |
7831 | } |
7832 | ||
96f874e2 RR |
7833 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7834 | struct cpumask *tmpmask) | |
1da177e4 | 7835 | { |
7c16ec58 | 7836 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7837 | } |
1da177e4 | 7838 | |
1a20ff27 DG |
7839 | /* |
7840 | * Detach sched domains from a group of cpus specified in cpu_map | |
7841 | * These cpus will now be attached to the NULL domain | |
7842 | */ | |
96f874e2 | 7843 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7844 | { |
96f874e2 RR |
7845 | /* Save because hotplug lock held. */ |
7846 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7847 | int i; |
7848 | ||
abcd083a | 7849 | for_each_cpu(i, cpu_map) |
57d885fe | 7850 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7851 | synchronize_sched(); |
96f874e2 | 7852 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7853 | } |
7854 | ||
1d3504fc HS |
7855 | /* handle null as "default" */ |
7856 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7857 | struct sched_domain_attr *new, int idx_new) | |
7858 | { | |
7859 | struct sched_domain_attr tmp; | |
7860 | ||
7861 | /* fast path */ | |
7862 | if (!new && !cur) | |
7863 | return 1; | |
7864 | ||
7865 | tmp = SD_ATTR_INIT; | |
7866 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7867 | new ? (new + idx_new) : &tmp, | |
7868 | sizeof(struct sched_domain_attr)); | |
7869 | } | |
7870 | ||
029190c5 PJ |
7871 | /* |
7872 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7873 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7874 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7875 | * It destroys each deleted domain and builds each new domain. | |
7876 | * | |
96f874e2 | 7877 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
7878 | * The masks don't intersect (don't overlap.) We should setup one |
7879 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7880 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7881 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7882 | * it as it is. | |
7883 | * | |
41a2d6cf IM |
7884 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7885 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
7886 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
7887 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
7888 | * the single partition 'fallback_doms', it also forces the domains | |
7889 | * to be rebuilt. | |
029190c5 | 7890 | * |
96f874e2 | 7891 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7892 | * ndoms_new == 0 is a special case for destroying existing domains, |
7893 | * and it will not create the default domain. | |
dfb512ec | 7894 | * |
029190c5 PJ |
7895 | * Call with hotplug lock held |
7896 | */ | |
96f874e2 RR |
7897 | /* FIXME: Change to struct cpumask *doms_new[] */ |
7898 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 7899 | struct sched_domain_attr *dattr_new) |
029190c5 | 7900 | { |
dfb512ec | 7901 | int i, j, n; |
d65bd5ec | 7902 | int new_topology; |
029190c5 | 7903 | |
712555ee | 7904 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7905 | |
7378547f MM |
7906 | /* always unregister in case we don't destroy any domains */ |
7907 | unregister_sched_domain_sysctl(); | |
7908 | ||
d65bd5ec HC |
7909 | /* Let architecture update cpu core mappings. */ |
7910 | new_topology = arch_update_cpu_topology(); | |
7911 | ||
dfb512ec | 7912 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7913 | |
7914 | /* Destroy deleted domains */ | |
7915 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7916 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 7917 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 7918 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7919 | goto match1; |
7920 | } | |
7921 | /* no match - a current sched domain not in new doms_new[] */ | |
7922 | detach_destroy_domains(doms_cur + i); | |
7923 | match1: | |
7924 | ; | |
7925 | } | |
7926 | ||
e761b772 MK |
7927 | if (doms_new == NULL) { |
7928 | ndoms_cur = 0; | |
4212823f | 7929 | doms_new = fallback_doms; |
dcc30a35 | 7930 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 7931 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7932 | } |
7933 | ||
029190c5 PJ |
7934 | /* Build new domains */ |
7935 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7936 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 7937 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 7938 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7939 | goto match2; |
7940 | } | |
7941 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7942 | __build_sched_domains(doms_new + i, |
7943 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7944 | match2: |
7945 | ; | |
7946 | } | |
7947 | ||
7948 | /* Remember the new sched domains */ | |
4212823f | 7949 | if (doms_cur != fallback_doms) |
029190c5 | 7950 | kfree(doms_cur); |
1d3504fc | 7951 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7952 | doms_cur = doms_new; |
1d3504fc | 7953 | dattr_cur = dattr_new; |
029190c5 | 7954 | ndoms_cur = ndoms_new; |
7378547f MM |
7955 | |
7956 | register_sched_domain_sysctl(); | |
a1835615 | 7957 | |
712555ee | 7958 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7959 | } |
7960 | ||
5c45bf27 | 7961 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7962 | int arch_reinit_sched_domains(void) |
5c45bf27 | 7963 | { |
95402b38 | 7964 | get_online_cpus(); |
dfb512ec MK |
7965 | |
7966 | /* Destroy domains first to force the rebuild */ | |
7967 | partition_sched_domains(0, NULL, NULL); | |
7968 | ||
e761b772 | 7969 | rebuild_sched_domains(); |
95402b38 | 7970 | put_online_cpus(); |
dfb512ec | 7971 | |
e761b772 | 7972 | return 0; |
5c45bf27 SS |
7973 | } |
7974 | ||
7975 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7976 | { | |
7977 | int ret; | |
afb8a9b7 | 7978 | unsigned int level = 0; |
5c45bf27 | 7979 | |
afb8a9b7 GS |
7980 | if (sscanf(buf, "%u", &level) != 1) |
7981 | return -EINVAL; | |
7982 | ||
7983 | /* | |
7984 | * level is always be positive so don't check for | |
7985 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7986 | * What happens on 0 or 1 byte write, | |
7987 | * need to check for count as well? | |
7988 | */ | |
7989 | ||
7990 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7991 | return -EINVAL; |
7992 | ||
7993 | if (smt) | |
afb8a9b7 | 7994 | sched_smt_power_savings = level; |
5c45bf27 | 7995 | else |
afb8a9b7 | 7996 | sched_mc_power_savings = level; |
5c45bf27 SS |
7997 | |
7998 | ret = arch_reinit_sched_domains(); | |
7999 | ||
8000 | return ret ? ret : count; | |
8001 | } | |
8002 | ||
5c45bf27 | 8003 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8004 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8005 | char *page) | |
5c45bf27 SS |
8006 | { |
8007 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8008 | } | |
f718cd4a | 8009 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8010 | const char *buf, size_t count) |
5c45bf27 SS |
8011 | { |
8012 | return sched_power_savings_store(buf, count, 0); | |
8013 | } | |
f718cd4a AK |
8014 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8015 | sched_mc_power_savings_show, | |
8016 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8017 | #endif |
8018 | ||
8019 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8020 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8021 | char *page) | |
5c45bf27 SS |
8022 | { |
8023 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8024 | } | |
f718cd4a | 8025 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8026 | const char *buf, size_t count) |
5c45bf27 SS |
8027 | { |
8028 | return sched_power_savings_store(buf, count, 1); | |
8029 | } | |
f718cd4a AK |
8030 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8031 | sched_smt_power_savings_show, | |
6707de00 AB |
8032 | sched_smt_power_savings_store); |
8033 | #endif | |
8034 | ||
8035 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
8036 | { | |
8037 | int err = 0; | |
8038 | ||
8039 | #ifdef CONFIG_SCHED_SMT | |
8040 | if (smt_capable()) | |
8041 | err = sysfs_create_file(&cls->kset.kobj, | |
8042 | &attr_sched_smt_power_savings.attr); | |
8043 | #endif | |
8044 | #ifdef CONFIG_SCHED_MC | |
8045 | if (!err && mc_capable()) | |
8046 | err = sysfs_create_file(&cls->kset.kobj, | |
8047 | &attr_sched_mc_power_savings.attr); | |
8048 | #endif | |
8049 | return err; | |
8050 | } | |
6d6bc0ad | 8051 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8052 | |
e761b772 | 8053 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8054 | /* |
e761b772 MK |
8055 | * Add online and remove offline CPUs from the scheduler domains. |
8056 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8057 | */ |
8058 | static int update_sched_domains(struct notifier_block *nfb, | |
8059 | unsigned long action, void *hcpu) | |
e761b772 MK |
8060 | { |
8061 | switch (action) { | |
8062 | case CPU_ONLINE: | |
8063 | case CPU_ONLINE_FROZEN: | |
8064 | case CPU_DEAD: | |
8065 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8066 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8067 | return NOTIFY_OK; |
8068 | ||
8069 | default: | |
8070 | return NOTIFY_DONE; | |
8071 | } | |
8072 | } | |
8073 | #endif | |
8074 | ||
8075 | static int update_runtime(struct notifier_block *nfb, | |
8076 | unsigned long action, void *hcpu) | |
1da177e4 | 8077 | { |
7def2be1 PZ |
8078 | int cpu = (int)(long)hcpu; |
8079 | ||
1da177e4 | 8080 | switch (action) { |
1da177e4 | 8081 | case CPU_DOWN_PREPARE: |
8bb78442 | 8082 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8083 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8084 | return NOTIFY_OK; |
8085 | ||
1da177e4 | 8086 | case CPU_DOWN_FAILED: |
8bb78442 | 8087 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8088 | case CPU_ONLINE: |
8bb78442 | 8089 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8090 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8091 | return NOTIFY_OK; |
8092 | ||
1da177e4 LT |
8093 | default: |
8094 | return NOTIFY_DONE; | |
8095 | } | |
1da177e4 | 8096 | } |
1da177e4 LT |
8097 | |
8098 | void __init sched_init_smp(void) | |
8099 | { | |
dcc30a35 RR |
8100 | cpumask_var_t non_isolated_cpus; |
8101 | ||
8102 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8103 | |
434d53b0 MT |
8104 | #if defined(CONFIG_NUMA) |
8105 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8106 | GFP_KERNEL); | |
8107 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8108 | #endif | |
95402b38 | 8109 | get_online_cpus(); |
712555ee | 8110 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8111 | arch_init_sched_domains(cpu_online_mask); |
8112 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
8113 | if (cpumask_empty(non_isolated_cpus)) | |
8114 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8115 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8116 | put_online_cpus(); |
e761b772 MK |
8117 | |
8118 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8119 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8120 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8121 | #endif |
8122 | ||
8123 | /* RT runtime code needs to handle some hotplug events */ | |
8124 | hotcpu_notifier(update_runtime, 0); | |
8125 | ||
b328ca18 | 8126 | init_hrtick(); |
5c1e1767 NP |
8127 | |
8128 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8129 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8130 | BUG(); |
19978ca6 | 8131 | sched_init_granularity(); |
dcc30a35 | 8132 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
8133 | |
8134 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 8135 | init_sched_rt_class(); |
1da177e4 LT |
8136 | } |
8137 | #else | |
8138 | void __init sched_init_smp(void) | |
8139 | { | |
19978ca6 | 8140 | sched_init_granularity(); |
1da177e4 LT |
8141 | } |
8142 | #endif /* CONFIG_SMP */ | |
8143 | ||
8144 | int in_sched_functions(unsigned long addr) | |
8145 | { | |
1da177e4 LT |
8146 | return in_lock_functions(addr) || |
8147 | (addr >= (unsigned long)__sched_text_start | |
8148 | && addr < (unsigned long)__sched_text_end); | |
8149 | } | |
8150 | ||
a9957449 | 8151 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8152 | { |
8153 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8154 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8155 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8156 | cfs_rq->rq = rq; | |
8157 | #endif | |
67e9fb2a | 8158 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8159 | } |
8160 | ||
fa85ae24 PZ |
8161 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8162 | { | |
8163 | struct rt_prio_array *array; | |
8164 | int i; | |
8165 | ||
8166 | array = &rt_rq->active; | |
8167 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8168 | INIT_LIST_HEAD(array->queue + i); | |
8169 | __clear_bit(i, array->bitmap); | |
8170 | } | |
8171 | /* delimiter for bitsearch: */ | |
8172 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8173 | ||
052f1dc7 | 8174 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
8175 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
8176 | rt_rq->highest_prio.next = MAX_RT_PRIO; | |
48d5e258 | 8177 | #endif |
fa85ae24 PZ |
8178 | #ifdef CONFIG_SMP |
8179 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
8180 | rt_rq->overloaded = 0; |
8181 | #endif | |
8182 | ||
8183 | rt_rq->rt_time = 0; | |
8184 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8185 | rt_rq->rt_runtime = 0; |
8186 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8187 | |
052f1dc7 | 8188 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8189 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8190 | rt_rq->rq = rq; |
8191 | #endif | |
fa85ae24 PZ |
8192 | } |
8193 | ||
6f505b16 | 8194 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8195 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8196 | struct sched_entity *se, int cpu, int add, | |
8197 | struct sched_entity *parent) | |
6f505b16 | 8198 | { |
ec7dc8ac | 8199 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8200 | tg->cfs_rq[cpu] = cfs_rq; |
8201 | init_cfs_rq(cfs_rq, rq); | |
8202 | cfs_rq->tg = tg; | |
8203 | if (add) | |
8204 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8205 | ||
8206 | tg->se[cpu] = se; | |
354d60c2 DG |
8207 | /* se could be NULL for init_task_group */ |
8208 | if (!se) | |
8209 | return; | |
8210 | ||
ec7dc8ac DG |
8211 | if (!parent) |
8212 | se->cfs_rq = &rq->cfs; | |
8213 | else | |
8214 | se->cfs_rq = parent->my_q; | |
8215 | ||
6f505b16 PZ |
8216 | se->my_q = cfs_rq; |
8217 | se->load.weight = tg->shares; | |
e05510d0 | 8218 | se->load.inv_weight = 0; |
ec7dc8ac | 8219 | se->parent = parent; |
6f505b16 | 8220 | } |
052f1dc7 | 8221 | #endif |
6f505b16 | 8222 | |
052f1dc7 | 8223 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8224 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8225 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8226 | struct sched_rt_entity *parent) | |
6f505b16 | 8227 | { |
ec7dc8ac DG |
8228 | struct rq *rq = cpu_rq(cpu); |
8229 | ||
6f505b16 PZ |
8230 | tg->rt_rq[cpu] = rt_rq; |
8231 | init_rt_rq(rt_rq, rq); | |
8232 | rt_rq->tg = tg; | |
8233 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8234 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8235 | if (add) |
8236 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8237 | ||
8238 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8239 | if (!rt_se) |
8240 | return; | |
8241 | ||
ec7dc8ac DG |
8242 | if (!parent) |
8243 | rt_se->rt_rq = &rq->rt; | |
8244 | else | |
8245 | rt_se->rt_rq = parent->my_q; | |
8246 | ||
6f505b16 | 8247 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8248 | rt_se->parent = parent; |
6f505b16 PZ |
8249 | INIT_LIST_HEAD(&rt_se->run_list); |
8250 | } | |
8251 | #endif | |
8252 | ||
1da177e4 LT |
8253 | void __init sched_init(void) |
8254 | { | |
dd41f596 | 8255 | int i, j; |
434d53b0 MT |
8256 | unsigned long alloc_size = 0, ptr; |
8257 | ||
8258 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8259 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8260 | #endif | |
8261 | #ifdef CONFIG_RT_GROUP_SCHED | |
8262 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8263 | #endif |
8264 | #ifdef CONFIG_USER_SCHED | |
8265 | alloc_size *= 2; | |
434d53b0 MT |
8266 | #endif |
8267 | /* | |
8268 | * As sched_init() is called before page_alloc is setup, | |
8269 | * we use alloc_bootmem(). | |
8270 | */ | |
8271 | if (alloc_size) { | |
5a9d3225 | 8272 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8273 | |
8274 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8275 | init_task_group.se = (struct sched_entity **)ptr; | |
8276 | ptr += nr_cpu_ids * sizeof(void **); | |
8277 | ||
8278 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8279 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8280 | |
8281 | #ifdef CONFIG_USER_SCHED | |
8282 | root_task_group.se = (struct sched_entity **)ptr; | |
8283 | ptr += nr_cpu_ids * sizeof(void **); | |
8284 | ||
8285 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8286 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8287 | #endif /* CONFIG_USER_SCHED */ |
8288 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8289 | #ifdef CONFIG_RT_GROUP_SCHED |
8290 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8291 | ptr += nr_cpu_ids * sizeof(void **); | |
8292 | ||
8293 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8294 | ptr += nr_cpu_ids * sizeof(void **); |
8295 | ||
8296 | #ifdef CONFIG_USER_SCHED | |
8297 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8298 | ptr += nr_cpu_ids * sizeof(void **); | |
8299 | ||
8300 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8301 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8302 | #endif /* CONFIG_USER_SCHED */ |
8303 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8304 | } |
dd41f596 | 8305 | |
57d885fe GH |
8306 | #ifdef CONFIG_SMP |
8307 | init_defrootdomain(); | |
8308 | #endif | |
8309 | ||
d0b27fa7 PZ |
8310 | init_rt_bandwidth(&def_rt_bandwidth, |
8311 | global_rt_period(), global_rt_runtime()); | |
8312 | ||
8313 | #ifdef CONFIG_RT_GROUP_SCHED | |
8314 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8315 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8316 | #ifdef CONFIG_USER_SCHED |
8317 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8318 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8319 | #endif /* CONFIG_USER_SCHED */ |
8320 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8321 | |
052f1dc7 | 8322 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8323 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8324 | INIT_LIST_HEAD(&init_task_group.children); |
8325 | ||
8326 | #ifdef CONFIG_USER_SCHED | |
8327 | INIT_LIST_HEAD(&root_task_group.children); | |
8328 | init_task_group.parent = &root_task_group; | |
8329 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8330 | #endif /* CONFIG_USER_SCHED */ |
8331 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8332 | |
0a945022 | 8333 | for_each_possible_cpu(i) { |
70b97a7f | 8334 | struct rq *rq; |
1da177e4 LT |
8335 | |
8336 | rq = cpu_rq(i); | |
8337 | spin_lock_init(&rq->lock); | |
7897986b | 8338 | rq->nr_running = 0; |
dd41f596 | 8339 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8340 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8341 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8342 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8343 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8344 | #ifdef CONFIG_CGROUP_SCHED |
8345 | /* | |
8346 | * How much cpu bandwidth does init_task_group get? | |
8347 | * | |
8348 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8349 | * gets 100% of the cpu resources in the system. This overall | |
8350 | * system cpu resource is divided among the tasks of | |
8351 | * init_task_group and its child task-groups in a fair manner, | |
8352 | * based on each entity's (task or task-group's) weight | |
8353 | * (se->load.weight). | |
8354 | * | |
8355 | * In other words, if init_task_group has 10 tasks of weight | |
8356 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8357 | * then A0's share of the cpu resource is: | |
8358 | * | |
8359 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8360 | * | |
8361 | * We achieve this by letting init_task_group's tasks sit | |
8362 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8363 | */ | |
ec7dc8ac | 8364 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8365 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8366 | root_task_group.shares = NICE_0_LOAD; |
8367 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8368 | /* |
8369 | * In case of task-groups formed thr' the user id of tasks, | |
8370 | * init_task_group represents tasks belonging to root user. | |
8371 | * Hence it forms a sibling of all subsequent groups formed. | |
8372 | * In this case, init_task_group gets only a fraction of overall | |
8373 | * system cpu resource, based on the weight assigned to root | |
8374 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8375 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8376 | * (init_cfs_rq) and having one entity represent this group of | |
8377 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8378 | */ | |
ec7dc8ac | 8379 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8380 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8381 | &per_cpu(init_sched_entity, i), i, 1, |
8382 | root_task_group.se[i]); | |
6f505b16 | 8383 | |
052f1dc7 | 8384 | #endif |
354d60c2 DG |
8385 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8386 | ||
8387 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8388 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8389 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8390 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8391 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8392 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8393 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8394 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8395 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8396 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8397 | root_task_group.rt_se[i]); | |
354d60c2 | 8398 | #endif |
dd41f596 | 8399 | #endif |
1da177e4 | 8400 | |
dd41f596 IM |
8401 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8402 | rq->cpu_load[j] = 0; | |
1da177e4 | 8403 | #ifdef CONFIG_SMP |
41c7ce9a | 8404 | rq->sd = NULL; |
57d885fe | 8405 | rq->rd = NULL; |
1da177e4 | 8406 | rq->active_balance = 0; |
dd41f596 | 8407 | rq->next_balance = jiffies; |
1da177e4 | 8408 | rq->push_cpu = 0; |
0a2966b4 | 8409 | rq->cpu = i; |
1f11eb6a | 8410 | rq->online = 0; |
1da177e4 LT |
8411 | rq->migration_thread = NULL; |
8412 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8413 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8414 | #endif |
8f4d37ec | 8415 | init_rq_hrtick(rq); |
1da177e4 | 8416 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8417 | } |
8418 | ||
2dd73a4f | 8419 | set_load_weight(&init_task); |
b50f60ce | 8420 | |
e107be36 AK |
8421 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8422 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8423 | #endif | |
8424 | ||
c9819f45 | 8425 | #ifdef CONFIG_SMP |
962cf36c | 8426 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8427 | #endif |
8428 | ||
b50f60ce HC |
8429 | #ifdef CONFIG_RT_MUTEXES |
8430 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8431 | #endif | |
8432 | ||
1da177e4 LT |
8433 | /* |
8434 | * The boot idle thread does lazy MMU switching as well: | |
8435 | */ | |
8436 | atomic_inc(&init_mm.mm_count); | |
8437 | enter_lazy_tlb(&init_mm, current); | |
8438 | ||
8439 | /* | |
8440 | * Make us the idle thread. Technically, schedule() should not be | |
8441 | * called from this thread, however somewhere below it might be, | |
8442 | * but because we are the idle thread, we just pick up running again | |
8443 | * when this runqueue becomes "idle". | |
8444 | */ | |
8445 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8446 | /* |
8447 | * During early bootup we pretend to be a normal task: | |
8448 | */ | |
8449 | current->sched_class = &fair_sched_class; | |
6892b75e | 8450 | |
6a7b3dc3 RR |
8451 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
8452 | alloc_bootmem_cpumask_var(&nohz_cpu_mask); | |
bf4d83f6 | 8453 | #ifdef CONFIG_SMP |
7d1e6a9b RR |
8454 | #ifdef CONFIG_NO_HZ |
8455 | alloc_bootmem_cpumask_var(&nohz.cpu_mask); | |
8456 | #endif | |
dcc30a35 | 8457 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
bf4d83f6 | 8458 | #endif /* SMP */ |
6a7b3dc3 | 8459 | |
6892b75e | 8460 | scheduler_running = 1; |
1da177e4 LT |
8461 | } |
8462 | ||
8463 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8464 | void __might_sleep(char *file, int line) | |
8465 | { | |
48f24c4d | 8466 | #ifdef in_atomic |
1da177e4 LT |
8467 | static unsigned long prev_jiffy; /* ratelimiting */ |
8468 | ||
aef745fc IM |
8469 | if ((!in_atomic() && !irqs_disabled()) || |
8470 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8471 | return; | |
8472 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8473 | return; | |
8474 | prev_jiffy = jiffies; | |
8475 | ||
8476 | printk(KERN_ERR | |
8477 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8478 | file, line); | |
8479 | printk(KERN_ERR | |
8480 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8481 | in_atomic(), irqs_disabled(), | |
8482 | current->pid, current->comm); | |
8483 | ||
8484 | debug_show_held_locks(current); | |
8485 | if (irqs_disabled()) | |
8486 | print_irqtrace_events(current); | |
8487 | dump_stack(); | |
1da177e4 LT |
8488 | #endif |
8489 | } | |
8490 | EXPORT_SYMBOL(__might_sleep); | |
8491 | #endif | |
8492 | ||
8493 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8494 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8495 | { | |
8496 | int on_rq; | |
3e51f33f | 8497 | |
3a5e4dc1 AK |
8498 | update_rq_clock(rq); |
8499 | on_rq = p->se.on_rq; | |
8500 | if (on_rq) | |
8501 | deactivate_task(rq, p, 0); | |
8502 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8503 | if (on_rq) { | |
8504 | activate_task(rq, p, 0); | |
8505 | resched_task(rq->curr); | |
8506 | } | |
8507 | } | |
8508 | ||
1da177e4 LT |
8509 | void normalize_rt_tasks(void) |
8510 | { | |
a0f98a1c | 8511 | struct task_struct *g, *p; |
1da177e4 | 8512 | unsigned long flags; |
70b97a7f | 8513 | struct rq *rq; |
1da177e4 | 8514 | |
4cf5d77a | 8515 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8516 | do_each_thread(g, p) { |
178be793 IM |
8517 | /* |
8518 | * Only normalize user tasks: | |
8519 | */ | |
8520 | if (!p->mm) | |
8521 | continue; | |
8522 | ||
6cfb0d5d | 8523 | p->se.exec_start = 0; |
6cfb0d5d | 8524 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8525 | p->se.wait_start = 0; |
dd41f596 | 8526 | p->se.sleep_start = 0; |
dd41f596 | 8527 | p->se.block_start = 0; |
6cfb0d5d | 8528 | #endif |
dd41f596 IM |
8529 | |
8530 | if (!rt_task(p)) { | |
8531 | /* | |
8532 | * Renice negative nice level userspace | |
8533 | * tasks back to 0: | |
8534 | */ | |
8535 | if (TASK_NICE(p) < 0 && p->mm) | |
8536 | set_user_nice(p, 0); | |
1da177e4 | 8537 | continue; |
dd41f596 | 8538 | } |
1da177e4 | 8539 | |
4cf5d77a | 8540 | spin_lock(&p->pi_lock); |
b29739f9 | 8541 | rq = __task_rq_lock(p); |
1da177e4 | 8542 | |
178be793 | 8543 | normalize_task(rq, p); |
3a5e4dc1 | 8544 | |
b29739f9 | 8545 | __task_rq_unlock(rq); |
4cf5d77a | 8546 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8547 | } while_each_thread(g, p); |
8548 | ||
4cf5d77a | 8549 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8550 | } |
8551 | ||
8552 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8553 | |
8554 | #ifdef CONFIG_IA64 | |
8555 | /* | |
8556 | * These functions are only useful for the IA64 MCA handling. | |
8557 | * | |
8558 | * They can only be called when the whole system has been | |
8559 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8560 | * activity can take place. Using them for anything else would | |
8561 | * be a serious bug, and as a result, they aren't even visible | |
8562 | * under any other configuration. | |
8563 | */ | |
8564 | ||
8565 | /** | |
8566 | * curr_task - return the current task for a given cpu. | |
8567 | * @cpu: the processor in question. | |
8568 | * | |
8569 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8570 | */ | |
36c8b586 | 8571 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8572 | { |
8573 | return cpu_curr(cpu); | |
8574 | } | |
8575 | ||
8576 | /** | |
8577 | * set_curr_task - set the current task for a given cpu. | |
8578 | * @cpu: the processor in question. | |
8579 | * @p: the task pointer to set. | |
8580 | * | |
8581 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8582 | * are serviced on a separate stack. It allows the architecture to switch the |
8583 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8584 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8585 | * and caller must save the original value of the current task (see | |
8586 | * curr_task() above) and restore that value before reenabling interrupts and | |
8587 | * re-starting the system. | |
8588 | * | |
8589 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8590 | */ | |
36c8b586 | 8591 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8592 | { |
8593 | cpu_curr(cpu) = p; | |
8594 | } | |
8595 | ||
8596 | #endif | |
29f59db3 | 8597 | |
bccbe08a PZ |
8598 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8599 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8600 | { |
8601 | int i; | |
8602 | ||
8603 | for_each_possible_cpu(i) { | |
8604 | if (tg->cfs_rq) | |
8605 | kfree(tg->cfs_rq[i]); | |
8606 | if (tg->se) | |
8607 | kfree(tg->se[i]); | |
6f505b16 PZ |
8608 | } |
8609 | ||
8610 | kfree(tg->cfs_rq); | |
8611 | kfree(tg->se); | |
6f505b16 PZ |
8612 | } |
8613 | ||
ec7dc8ac DG |
8614 | static |
8615 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8616 | { |
29f59db3 | 8617 | struct cfs_rq *cfs_rq; |
eab17229 | 8618 | struct sched_entity *se; |
9b5b7751 | 8619 | struct rq *rq; |
29f59db3 SV |
8620 | int i; |
8621 | ||
434d53b0 | 8622 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8623 | if (!tg->cfs_rq) |
8624 | goto err; | |
434d53b0 | 8625 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8626 | if (!tg->se) |
8627 | goto err; | |
052f1dc7 PZ |
8628 | |
8629 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8630 | |
8631 | for_each_possible_cpu(i) { | |
9b5b7751 | 8632 | rq = cpu_rq(i); |
29f59db3 | 8633 | |
eab17229 LZ |
8634 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8635 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8636 | if (!cfs_rq) |
8637 | goto err; | |
8638 | ||
eab17229 LZ |
8639 | se = kzalloc_node(sizeof(struct sched_entity), |
8640 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8641 | if (!se) |
8642 | goto err; | |
8643 | ||
eab17229 | 8644 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
8645 | } |
8646 | ||
8647 | return 1; | |
8648 | ||
8649 | err: | |
8650 | return 0; | |
8651 | } | |
8652 | ||
8653 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8654 | { | |
8655 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8656 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8657 | } | |
8658 | ||
8659 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8660 | { | |
8661 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8662 | } | |
6d6bc0ad | 8663 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8664 | static inline void free_fair_sched_group(struct task_group *tg) |
8665 | { | |
8666 | } | |
8667 | ||
ec7dc8ac DG |
8668 | static inline |
8669 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8670 | { |
8671 | return 1; | |
8672 | } | |
8673 | ||
8674 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8675 | { | |
8676 | } | |
8677 | ||
8678 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8679 | { | |
8680 | } | |
6d6bc0ad | 8681 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8682 | |
8683 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8684 | static void free_rt_sched_group(struct task_group *tg) |
8685 | { | |
8686 | int i; | |
8687 | ||
d0b27fa7 PZ |
8688 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8689 | ||
bccbe08a PZ |
8690 | for_each_possible_cpu(i) { |
8691 | if (tg->rt_rq) | |
8692 | kfree(tg->rt_rq[i]); | |
8693 | if (tg->rt_se) | |
8694 | kfree(tg->rt_se[i]); | |
8695 | } | |
8696 | ||
8697 | kfree(tg->rt_rq); | |
8698 | kfree(tg->rt_se); | |
8699 | } | |
8700 | ||
ec7dc8ac DG |
8701 | static |
8702 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8703 | { |
8704 | struct rt_rq *rt_rq; | |
eab17229 | 8705 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8706 | struct rq *rq; |
8707 | int i; | |
8708 | ||
434d53b0 | 8709 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8710 | if (!tg->rt_rq) |
8711 | goto err; | |
434d53b0 | 8712 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8713 | if (!tg->rt_se) |
8714 | goto err; | |
8715 | ||
d0b27fa7 PZ |
8716 | init_rt_bandwidth(&tg->rt_bandwidth, |
8717 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8718 | |
8719 | for_each_possible_cpu(i) { | |
8720 | rq = cpu_rq(i); | |
8721 | ||
eab17229 LZ |
8722 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8723 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8724 | if (!rt_rq) |
8725 | goto err; | |
29f59db3 | 8726 | |
eab17229 LZ |
8727 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8728 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8729 | if (!rt_se) |
8730 | goto err; | |
29f59db3 | 8731 | |
eab17229 | 8732 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8733 | } |
8734 | ||
bccbe08a PZ |
8735 | return 1; |
8736 | ||
8737 | err: | |
8738 | return 0; | |
8739 | } | |
8740 | ||
8741 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8742 | { | |
8743 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8744 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8745 | } | |
8746 | ||
8747 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8748 | { | |
8749 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8750 | } | |
6d6bc0ad | 8751 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8752 | static inline void free_rt_sched_group(struct task_group *tg) |
8753 | { | |
8754 | } | |
8755 | ||
ec7dc8ac DG |
8756 | static inline |
8757 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8758 | { |
8759 | return 1; | |
8760 | } | |
8761 | ||
8762 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8763 | { | |
8764 | } | |
8765 | ||
8766 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8767 | { | |
8768 | } | |
6d6bc0ad | 8769 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8770 | |
d0b27fa7 | 8771 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8772 | static void free_sched_group(struct task_group *tg) |
8773 | { | |
8774 | free_fair_sched_group(tg); | |
8775 | free_rt_sched_group(tg); | |
8776 | kfree(tg); | |
8777 | } | |
8778 | ||
8779 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8780 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8781 | { |
8782 | struct task_group *tg; | |
8783 | unsigned long flags; | |
8784 | int i; | |
8785 | ||
8786 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8787 | if (!tg) | |
8788 | return ERR_PTR(-ENOMEM); | |
8789 | ||
ec7dc8ac | 8790 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8791 | goto err; |
8792 | ||
ec7dc8ac | 8793 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8794 | goto err; |
8795 | ||
8ed36996 | 8796 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8797 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8798 | register_fair_sched_group(tg, i); |
8799 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8800 | } |
6f505b16 | 8801 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8802 | |
8803 | WARN_ON(!parent); /* root should already exist */ | |
8804 | ||
8805 | tg->parent = parent; | |
f473aa5e | 8806 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8807 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8808 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8809 | |
9b5b7751 | 8810 | return tg; |
29f59db3 SV |
8811 | |
8812 | err: | |
6f505b16 | 8813 | free_sched_group(tg); |
29f59db3 SV |
8814 | return ERR_PTR(-ENOMEM); |
8815 | } | |
8816 | ||
9b5b7751 | 8817 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8818 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8819 | { |
29f59db3 | 8820 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8821 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8822 | } |
8823 | ||
9b5b7751 | 8824 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8825 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8826 | { |
8ed36996 | 8827 | unsigned long flags; |
9b5b7751 | 8828 | int i; |
29f59db3 | 8829 | |
8ed36996 | 8830 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8831 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8832 | unregister_fair_sched_group(tg, i); |
8833 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8834 | } |
6f505b16 | 8835 | list_del_rcu(&tg->list); |
f473aa5e | 8836 | list_del_rcu(&tg->siblings); |
8ed36996 | 8837 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8838 | |
9b5b7751 | 8839 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8840 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8841 | } |
8842 | ||
9b5b7751 | 8843 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8844 | * The caller of this function should have put the task in its new group |
8845 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8846 | * reflect its new group. | |
9b5b7751 SV |
8847 | */ |
8848 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8849 | { |
8850 | int on_rq, running; | |
8851 | unsigned long flags; | |
8852 | struct rq *rq; | |
8853 | ||
8854 | rq = task_rq_lock(tsk, &flags); | |
8855 | ||
29f59db3 SV |
8856 | update_rq_clock(rq); |
8857 | ||
051a1d1a | 8858 | running = task_current(rq, tsk); |
29f59db3 SV |
8859 | on_rq = tsk->se.on_rq; |
8860 | ||
0e1f3483 | 8861 | if (on_rq) |
29f59db3 | 8862 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8863 | if (unlikely(running)) |
8864 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8865 | |
6f505b16 | 8866 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8867 | |
810b3817 PZ |
8868 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8869 | if (tsk->sched_class->moved_group) | |
8870 | tsk->sched_class->moved_group(tsk); | |
8871 | #endif | |
8872 | ||
0e1f3483 HS |
8873 | if (unlikely(running)) |
8874 | tsk->sched_class->set_curr_task(rq); | |
8875 | if (on_rq) | |
7074badb | 8876 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8877 | |
29f59db3 SV |
8878 | task_rq_unlock(rq, &flags); |
8879 | } | |
6d6bc0ad | 8880 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8881 | |
052f1dc7 | 8882 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8883 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8884 | { |
8885 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8886 | int on_rq; |
8887 | ||
29f59db3 | 8888 | on_rq = se->on_rq; |
62fb1851 | 8889 | if (on_rq) |
29f59db3 SV |
8890 | dequeue_entity(cfs_rq, se, 0); |
8891 | ||
8892 | se->load.weight = shares; | |
e05510d0 | 8893 | se->load.inv_weight = 0; |
29f59db3 | 8894 | |
62fb1851 | 8895 | if (on_rq) |
29f59db3 | 8896 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8897 | } |
62fb1851 | 8898 | |
c09595f6 PZ |
8899 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8900 | { | |
8901 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8902 | struct rq *rq = cfs_rq->rq; | |
8903 | unsigned long flags; | |
8904 | ||
8905 | spin_lock_irqsave(&rq->lock, flags); | |
8906 | __set_se_shares(se, shares); | |
8907 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8908 | } |
8909 | ||
8ed36996 PZ |
8910 | static DEFINE_MUTEX(shares_mutex); |
8911 | ||
4cf86d77 | 8912 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8913 | { |
8914 | int i; | |
8ed36996 | 8915 | unsigned long flags; |
c61935fd | 8916 | |
ec7dc8ac DG |
8917 | /* |
8918 | * We can't change the weight of the root cgroup. | |
8919 | */ | |
8920 | if (!tg->se[0]) | |
8921 | return -EINVAL; | |
8922 | ||
18d95a28 PZ |
8923 | if (shares < MIN_SHARES) |
8924 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8925 | else if (shares > MAX_SHARES) |
8926 | shares = MAX_SHARES; | |
62fb1851 | 8927 | |
8ed36996 | 8928 | mutex_lock(&shares_mutex); |
9b5b7751 | 8929 | if (tg->shares == shares) |
5cb350ba | 8930 | goto done; |
29f59db3 | 8931 | |
8ed36996 | 8932 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8933 | for_each_possible_cpu(i) |
8934 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8935 | list_del_rcu(&tg->siblings); |
8ed36996 | 8936 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8937 | |
8938 | /* wait for any ongoing reference to this group to finish */ | |
8939 | synchronize_sched(); | |
8940 | ||
8941 | /* | |
8942 | * Now we are free to modify the group's share on each cpu | |
8943 | * w/o tripping rebalance_share or load_balance_fair. | |
8944 | */ | |
9b5b7751 | 8945 | tg->shares = shares; |
c09595f6 PZ |
8946 | for_each_possible_cpu(i) { |
8947 | /* | |
8948 | * force a rebalance | |
8949 | */ | |
8950 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8951 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8952 | } |
29f59db3 | 8953 | |
6b2d7700 SV |
8954 | /* |
8955 | * Enable load balance activity on this group, by inserting it back on | |
8956 | * each cpu's rq->leaf_cfs_rq_list. | |
8957 | */ | |
8ed36996 | 8958 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8959 | for_each_possible_cpu(i) |
8960 | register_fair_sched_group(tg, i); | |
f473aa5e | 8961 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8962 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8963 | done: |
8ed36996 | 8964 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8965 | return 0; |
29f59db3 SV |
8966 | } |
8967 | ||
5cb350ba DG |
8968 | unsigned long sched_group_shares(struct task_group *tg) |
8969 | { | |
8970 | return tg->shares; | |
8971 | } | |
052f1dc7 | 8972 | #endif |
5cb350ba | 8973 | |
052f1dc7 | 8974 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8975 | /* |
9f0c1e56 | 8976 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8977 | */ |
9f0c1e56 PZ |
8978 | static DEFINE_MUTEX(rt_constraints_mutex); |
8979 | ||
8980 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8981 | { | |
8982 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8983 | return 1ULL << 20; |
9f0c1e56 | 8984 | |
9a7e0b18 | 8985 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8986 | } |
8987 | ||
9a7e0b18 PZ |
8988 | /* Must be called with tasklist_lock held */ |
8989 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8990 | { |
9a7e0b18 | 8991 | struct task_struct *g, *p; |
b40b2e8e | 8992 | |
9a7e0b18 PZ |
8993 | do_each_thread(g, p) { |
8994 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8995 | return 1; | |
8996 | } while_each_thread(g, p); | |
b40b2e8e | 8997 | |
9a7e0b18 PZ |
8998 | return 0; |
8999 | } | |
b40b2e8e | 9000 | |
9a7e0b18 PZ |
9001 | struct rt_schedulable_data { |
9002 | struct task_group *tg; | |
9003 | u64 rt_period; | |
9004 | u64 rt_runtime; | |
9005 | }; | |
b40b2e8e | 9006 | |
9a7e0b18 PZ |
9007 | static int tg_schedulable(struct task_group *tg, void *data) |
9008 | { | |
9009 | struct rt_schedulable_data *d = data; | |
9010 | struct task_group *child; | |
9011 | unsigned long total, sum = 0; | |
9012 | u64 period, runtime; | |
b40b2e8e | 9013 | |
9a7e0b18 PZ |
9014 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9015 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9016 | |
9a7e0b18 PZ |
9017 | if (tg == d->tg) { |
9018 | period = d->rt_period; | |
9019 | runtime = d->rt_runtime; | |
b40b2e8e | 9020 | } |
b40b2e8e | 9021 | |
4653f803 PZ |
9022 | /* |
9023 | * Cannot have more runtime than the period. | |
9024 | */ | |
9025 | if (runtime > period && runtime != RUNTIME_INF) | |
9026 | return -EINVAL; | |
6f505b16 | 9027 | |
4653f803 PZ |
9028 | /* |
9029 | * Ensure we don't starve existing RT tasks. | |
9030 | */ | |
9a7e0b18 PZ |
9031 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9032 | return -EBUSY; | |
6f505b16 | 9033 | |
9a7e0b18 | 9034 | total = to_ratio(period, runtime); |
6f505b16 | 9035 | |
4653f803 PZ |
9036 | /* |
9037 | * Nobody can have more than the global setting allows. | |
9038 | */ | |
9039 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9040 | return -EINVAL; | |
6f505b16 | 9041 | |
4653f803 PZ |
9042 | /* |
9043 | * The sum of our children's runtime should not exceed our own. | |
9044 | */ | |
9a7e0b18 PZ |
9045 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9046 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9047 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9048 | |
9a7e0b18 PZ |
9049 | if (child == d->tg) { |
9050 | period = d->rt_period; | |
9051 | runtime = d->rt_runtime; | |
9052 | } | |
6f505b16 | 9053 | |
9a7e0b18 | 9054 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9055 | } |
6f505b16 | 9056 | |
9a7e0b18 PZ |
9057 | if (sum > total) |
9058 | return -EINVAL; | |
9059 | ||
9060 | return 0; | |
6f505b16 PZ |
9061 | } |
9062 | ||
9a7e0b18 | 9063 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9064 | { |
9a7e0b18 PZ |
9065 | struct rt_schedulable_data data = { |
9066 | .tg = tg, | |
9067 | .rt_period = period, | |
9068 | .rt_runtime = runtime, | |
9069 | }; | |
9070 | ||
9071 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9072 | } |
9073 | ||
d0b27fa7 PZ |
9074 | static int tg_set_bandwidth(struct task_group *tg, |
9075 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9076 | { |
ac086bc2 | 9077 | int i, err = 0; |
9f0c1e56 | 9078 | |
9f0c1e56 | 9079 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9080 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9081 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9082 | if (err) | |
9f0c1e56 | 9083 | goto unlock; |
ac086bc2 PZ |
9084 | |
9085 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
9086 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
9087 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
9088 | |
9089 | for_each_possible_cpu(i) { | |
9090 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
9091 | ||
9092 | spin_lock(&rt_rq->rt_runtime_lock); | |
9093 | rt_rq->rt_runtime = rt_runtime; | |
9094 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9095 | } | |
9096 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 9097 | unlock: |
521f1a24 | 9098 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
9099 | mutex_unlock(&rt_constraints_mutex); |
9100 | ||
9101 | return err; | |
6f505b16 PZ |
9102 | } |
9103 | ||
d0b27fa7 PZ |
9104 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
9105 | { | |
9106 | u64 rt_runtime, rt_period; | |
9107 | ||
9108 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9109 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
9110 | if (rt_runtime_us < 0) | |
9111 | rt_runtime = RUNTIME_INF; | |
9112 | ||
9113 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
9114 | } | |
9115 | ||
9f0c1e56 PZ |
9116 | long sched_group_rt_runtime(struct task_group *tg) |
9117 | { | |
9118 | u64 rt_runtime_us; | |
9119 | ||
d0b27fa7 | 9120 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9121 | return -1; |
9122 | ||
d0b27fa7 | 9123 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9124 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9125 | return rt_runtime_us; | |
9126 | } | |
d0b27fa7 PZ |
9127 | |
9128 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
9129 | { | |
9130 | u64 rt_runtime, rt_period; | |
9131 | ||
9132 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
9133 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
9134 | ||
619b0488 R |
9135 | if (rt_period == 0) |
9136 | return -EINVAL; | |
9137 | ||
d0b27fa7 PZ |
9138 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9139 | } | |
9140 | ||
9141 | long sched_group_rt_period(struct task_group *tg) | |
9142 | { | |
9143 | u64 rt_period_us; | |
9144 | ||
9145 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9146 | do_div(rt_period_us, NSEC_PER_USEC); | |
9147 | return rt_period_us; | |
9148 | } | |
9149 | ||
9150 | static int sched_rt_global_constraints(void) | |
9151 | { | |
4653f803 | 9152 | u64 runtime, period; |
d0b27fa7 PZ |
9153 | int ret = 0; |
9154 | ||
ec5d4989 HS |
9155 | if (sysctl_sched_rt_period <= 0) |
9156 | return -EINVAL; | |
9157 | ||
4653f803 PZ |
9158 | runtime = global_rt_runtime(); |
9159 | period = global_rt_period(); | |
9160 | ||
9161 | /* | |
9162 | * Sanity check on the sysctl variables. | |
9163 | */ | |
9164 | if (runtime > period && runtime != RUNTIME_INF) | |
9165 | return -EINVAL; | |
10b612f4 | 9166 | |
d0b27fa7 | 9167 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9168 | read_lock(&tasklist_lock); |
4653f803 | 9169 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9170 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9171 | mutex_unlock(&rt_constraints_mutex); |
9172 | ||
9173 | return ret; | |
9174 | } | |
6d6bc0ad | 9175 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9176 | static int sched_rt_global_constraints(void) |
9177 | { | |
ac086bc2 PZ |
9178 | unsigned long flags; |
9179 | int i; | |
9180 | ||
ec5d4989 HS |
9181 | if (sysctl_sched_rt_period <= 0) |
9182 | return -EINVAL; | |
9183 | ||
ac086bc2 PZ |
9184 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9185 | for_each_possible_cpu(i) { | |
9186 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9187 | ||
9188 | spin_lock(&rt_rq->rt_runtime_lock); | |
9189 | rt_rq->rt_runtime = global_rt_runtime(); | |
9190 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9191 | } | |
9192 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9193 | ||
d0b27fa7 PZ |
9194 | return 0; |
9195 | } | |
6d6bc0ad | 9196 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9197 | |
9198 | int sched_rt_handler(struct ctl_table *table, int write, | |
9199 | struct file *filp, void __user *buffer, size_t *lenp, | |
9200 | loff_t *ppos) | |
9201 | { | |
9202 | int ret; | |
9203 | int old_period, old_runtime; | |
9204 | static DEFINE_MUTEX(mutex); | |
9205 | ||
9206 | mutex_lock(&mutex); | |
9207 | old_period = sysctl_sched_rt_period; | |
9208 | old_runtime = sysctl_sched_rt_runtime; | |
9209 | ||
9210 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9211 | ||
9212 | if (!ret && write) { | |
9213 | ret = sched_rt_global_constraints(); | |
9214 | if (ret) { | |
9215 | sysctl_sched_rt_period = old_period; | |
9216 | sysctl_sched_rt_runtime = old_runtime; | |
9217 | } else { | |
9218 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9219 | def_rt_bandwidth.rt_period = | |
9220 | ns_to_ktime(global_rt_period()); | |
9221 | } | |
9222 | } | |
9223 | mutex_unlock(&mutex); | |
9224 | ||
9225 | return ret; | |
9226 | } | |
68318b8e | 9227 | |
052f1dc7 | 9228 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9229 | |
9230 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9231 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9232 | { |
2b01dfe3 PM |
9233 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9234 | struct task_group, css); | |
68318b8e SV |
9235 | } |
9236 | ||
9237 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9238 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9239 | { |
ec7dc8ac | 9240 | struct task_group *tg, *parent; |
68318b8e | 9241 | |
2b01dfe3 | 9242 | if (!cgrp->parent) { |
68318b8e | 9243 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9244 | return &init_task_group.css; |
9245 | } | |
9246 | ||
ec7dc8ac DG |
9247 | parent = cgroup_tg(cgrp->parent); |
9248 | tg = sched_create_group(parent); | |
68318b8e SV |
9249 | if (IS_ERR(tg)) |
9250 | return ERR_PTR(-ENOMEM); | |
9251 | ||
68318b8e SV |
9252 | return &tg->css; |
9253 | } | |
9254 | ||
41a2d6cf IM |
9255 | static void |
9256 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9257 | { |
2b01dfe3 | 9258 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9259 | |
9260 | sched_destroy_group(tg); | |
9261 | } | |
9262 | ||
41a2d6cf IM |
9263 | static int |
9264 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9265 | struct task_struct *tsk) | |
68318b8e | 9266 | { |
b68aa230 PZ |
9267 | #ifdef CONFIG_RT_GROUP_SCHED |
9268 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 9269 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
9270 | return -EINVAL; |
9271 | #else | |
68318b8e SV |
9272 | /* We don't support RT-tasks being in separate groups */ |
9273 | if (tsk->sched_class != &fair_sched_class) | |
9274 | return -EINVAL; | |
b68aa230 | 9275 | #endif |
68318b8e SV |
9276 | |
9277 | return 0; | |
9278 | } | |
9279 | ||
9280 | static void | |
2b01dfe3 | 9281 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9282 | struct cgroup *old_cont, struct task_struct *tsk) |
9283 | { | |
9284 | sched_move_task(tsk); | |
9285 | } | |
9286 | ||
052f1dc7 | 9287 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9288 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9289 | u64 shareval) |
68318b8e | 9290 | { |
2b01dfe3 | 9291 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9292 | } |
9293 | ||
f4c753b7 | 9294 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9295 | { |
2b01dfe3 | 9296 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9297 | |
9298 | return (u64) tg->shares; | |
9299 | } | |
6d6bc0ad | 9300 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9301 | |
052f1dc7 | 9302 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9303 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9304 | s64 val) |
6f505b16 | 9305 | { |
06ecb27c | 9306 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9307 | } |
9308 | ||
06ecb27c | 9309 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9310 | { |
06ecb27c | 9311 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9312 | } |
d0b27fa7 PZ |
9313 | |
9314 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9315 | u64 rt_period_us) | |
9316 | { | |
9317 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9318 | } | |
9319 | ||
9320 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9321 | { | |
9322 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9323 | } | |
6d6bc0ad | 9324 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9325 | |
fe5c7cc2 | 9326 | static struct cftype cpu_files[] = { |
052f1dc7 | 9327 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9328 | { |
9329 | .name = "shares", | |
f4c753b7 PM |
9330 | .read_u64 = cpu_shares_read_u64, |
9331 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9332 | }, |
052f1dc7 PZ |
9333 | #endif |
9334 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9335 | { |
9f0c1e56 | 9336 | .name = "rt_runtime_us", |
06ecb27c PM |
9337 | .read_s64 = cpu_rt_runtime_read, |
9338 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9339 | }, |
d0b27fa7 PZ |
9340 | { |
9341 | .name = "rt_period_us", | |
f4c753b7 PM |
9342 | .read_u64 = cpu_rt_period_read_uint, |
9343 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9344 | }, |
052f1dc7 | 9345 | #endif |
68318b8e SV |
9346 | }; |
9347 | ||
9348 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9349 | { | |
fe5c7cc2 | 9350 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9351 | } |
9352 | ||
9353 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9354 | .name = "cpu", |
9355 | .create = cpu_cgroup_create, | |
9356 | .destroy = cpu_cgroup_destroy, | |
9357 | .can_attach = cpu_cgroup_can_attach, | |
9358 | .attach = cpu_cgroup_attach, | |
9359 | .populate = cpu_cgroup_populate, | |
9360 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9361 | .early_init = 1, |
9362 | }; | |
9363 | ||
052f1dc7 | 9364 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9365 | |
9366 | #ifdef CONFIG_CGROUP_CPUACCT | |
9367 | ||
9368 | /* | |
9369 | * CPU accounting code for task groups. | |
9370 | * | |
9371 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9372 | * (balbir@in.ibm.com). | |
9373 | */ | |
9374 | ||
934352f2 | 9375 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9376 | struct cpuacct { |
9377 | struct cgroup_subsys_state css; | |
9378 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9379 | u64 *cpuusage; | |
934352f2 | 9380 | struct cpuacct *parent; |
d842de87 SV |
9381 | }; |
9382 | ||
9383 | struct cgroup_subsys cpuacct_subsys; | |
9384 | ||
9385 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9386 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9387 | { |
32cd756a | 9388 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9389 | struct cpuacct, css); |
9390 | } | |
9391 | ||
9392 | /* return cpu accounting group to which this task belongs */ | |
9393 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9394 | { | |
9395 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9396 | struct cpuacct, css); | |
9397 | } | |
9398 | ||
9399 | /* create a new cpu accounting group */ | |
9400 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9401 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9402 | { |
9403 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9404 | ||
9405 | if (!ca) | |
9406 | return ERR_PTR(-ENOMEM); | |
9407 | ||
9408 | ca->cpuusage = alloc_percpu(u64); | |
9409 | if (!ca->cpuusage) { | |
9410 | kfree(ca); | |
9411 | return ERR_PTR(-ENOMEM); | |
9412 | } | |
9413 | ||
934352f2 BR |
9414 | if (cgrp->parent) |
9415 | ca->parent = cgroup_ca(cgrp->parent); | |
9416 | ||
d842de87 SV |
9417 | return &ca->css; |
9418 | } | |
9419 | ||
9420 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9421 | static void |
32cd756a | 9422 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9423 | { |
32cd756a | 9424 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9425 | |
9426 | free_percpu(ca->cpuusage); | |
9427 | kfree(ca); | |
9428 | } | |
9429 | ||
9430 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 9431 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9432 | { |
32cd756a | 9433 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9434 | u64 totalcpuusage = 0; |
9435 | int i; | |
9436 | ||
9437 | for_each_possible_cpu(i) { | |
9438 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9439 | ||
9440 | /* | |
9441 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
9442 | * platforms. | |
9443 | */ | |
9444 | spin_lock_irq(&cpu_rq(i)->lock); | |
9445 | totalcpuusage += *cpuusage; | |
9446 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9447 | } | |
9448 | ||
9449 | return totalcpuusage; | |
9450 | } | |
9451 | ||
0297b803 DG |
9452 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9453 | u64 reset) | |
9454 | { | |
9455 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9456 | int err = 0; | |
9457 | int i; | |
9458 | ||
9459 | if (reset) { | |
9460 | err = -EINVAL; | |
9461 | goto out; | |
9462 | } | |
9463 | ||
9464 | for_each_possible_cpu(i) { | |
9465 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
9466 | ||
9467 | spin_lock_irq(&cpu_rq(i)->lock); | |
9468 | *cpuusage = 0; | |
9469 | spin_unlock_irq(&cpu_rq(i)->lock); | |
9470 | } | |
9471 | out: | |
9472 | return err; | |
9473 | } | |
9474 | ||
d842de87 SV |
9475 | static struct cftype files[] = { |
9476 | { | |
9477 | .name = "usage", | |
f4c753b7 PM |
9478 | .read_u64 = cpuusage_read, |
9479 | .write_u64 = cpuusage_write, | |
d842de87 SV |
9480 | }, |
9481 | }; | |
9482 | ||
32cd756a | 9483 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9484 | { |
32cd756a | 9485 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9486 | } |
9487 | ||
9488 | /* | |
9489 | * charge this task's execution time to its accounting group. | |
9490 | * | |
9491 | * called with rq->lock held. | |
9492 | */ | |
9493 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9494 | { | |
9495 | struct cpuacct *ca; | |
934352f2 | 9496 | int cpu; |
d842de87 SV |
9497 | |
9498 | if (!cpuacct_subsys.active) | |
9499 | return; | |
9500 | ||
934352f2 | 9501 | cpu = task_cpu(tsk); |
d842de87 | 9502 | ca = task_ca(tsk); |
d842de87 | 9503 | |
934352f2 BR |
9504 | for (; ca; ca = ca->parent) { |
9505 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
d842de87 SV |
9506 | *cpuusage += cputime; |
9507 | } | |
9508 | } | |
9509 | ||
9510 | struct cgroup_subsys cpuacct_subsys = { | |
9511 | .name = "cpuacct", | |
9512 | .create = cpuacct_create, | |
9513 | .destroy = cpuacct_destroy, | |
9514 | .populate = cpuacct_populate, | |
9515 | .subsys_id = cpuacct_subsys_id, | |
9516 | }; | |
9517 | #endif /* CONFIG_CGROUP_CPUACCT */ |