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 | 127 | #ifdef CONFIG_SMP |
fd2ab30b SN |
128 | |
129 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
130 | ||
5517d86b ED |
131 | /* |
132 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
133 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
134 | */ | |
135 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
136 | { | |
137 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
138 | } | |
139 | ||
140 | /* | |
141 | * Each time a sched group cpu_power is changed, | |
142 | * we must compute its reciprocal value | |
143 | */ | |
144 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
145 | { | |
146 | sg->__cpu_power += val; | |
147 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
148 | } | |
149 | #endif | |
150 | ||
e05606d3 IM |
151 | static inline int rt_policy(int policy) |
152 | { | |
3f33a7ce | 153 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
154 | return 1; |
155 | return 0; | |
156 | } | |
157 | ||
158 | static inline int task_has_rt_policy(struct task_struct *p) | |
159 | { | |
160 | return rt_policy(p->policy); | |
161 | } | |
162 | ||
1da177e4 | 163 | /* |
6aa645ea | 164 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 165 | */ |
6aa645ea IM |
166 | struct rt_prio_array { |
167 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
168 | struct list_head queue[MAX_RT_PRIO]; | |
169 | }; | |
170 | ||
d0b27fa7 | 171 | struct rt_bandwidth { |
ea736ed5 IM |
172 | /* nests inside the rq lock: */ |
173 | spinlock_t rt_runtime_lock; | |
174 | ktime_t rt_period; | |
175 | u64 rt_runtime; | |
176 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
177 | }; |
178 | ||
179 | static struct rt_bandwidth def_rt_bandwidth; | |
180 | ||
181 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
182 | ||
183 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
184 | { | |
185 | struct rt_bandwidth *rt_b = | |
186 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
187 | ktime_t now; | |
188 | int overrun; | |
189 | int idle = 0; | |
190 | ||
191 | for (;;) { | |
192 | now = hrtimer_cb_get_time(timer); | |
193 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
194 | ||
195 | if (!overrun) | |
196 | break; | |
197 | ||
198 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
199 | } | |
200 | ||
201 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
202 | } | |
203 | ||
204 | static | |
205 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
206 | { | |
207 | rt_b->rt_period = ns_to_ktime(period); | |
208 | rt_b->rt_runtime = runtime; | |
209 | ||
ac086bc2 PZ |
210 | spin_lock_init(&rt_b->rt_runtime_lock); |
211 | ||
d0b27fa7 PZ |
212 | hrtimer_init(&rt_b->rt_period_timer, |
213 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
214 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
215 | } |
216 | ||
c8bfff6d KH |
217 | static inline int rt_bandwidth_enabled(void) |
218 | { | |
219 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
220 | } |
221 | ||
222 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
223 | { | |
224 | ktime_t now; | |
225 | ||
0b148fa0 | 226 | if (rt_bandwidth_enabled() && rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
227 | return; |
228 | ||
229 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
230 | return; | |
231 | ||
232 | spin_lock(&rt_b->rt_runtime_lock); | |
233 | for (;;) { | |
234 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
235 | break; | |
236 | ||
237 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
238 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
cc584b21 AV |
239 | hrtimer_start_expires(&rt_b->rt_period_timer, |
240 | HRTIMER_MODE_ABS); | |
d0b27fa7 PZ |
241 | } |
242 | spin_unlock(&rt_b->rt_runtime_lock); | |
243 | } | |
244 | ||
245 | #ifdef CONFIG_RT_GROUP_SCHED | |
246 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
247 | { | |
248 | hrtimer_cancel(&rt_b->rt_period_timer); | |
249 | } | |
250 | #endif | |
251 | ||
712555ee HC |
252 | /* |
253 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
254 | * detach_destroy_domains and partition_sched_domains. | |
255 | */ | |
256 | static DEFINE_MUTEX(sched_domains_mutex); | |
257 | ||
052f1dc7 | 258 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 259 | |
68318b8e SV |
260 | #include <linux/cgroup.h> |
261 | ||
29f59db3 SV |
262 | struct cfs_rq; |
263 | ||
6f505b16 PZ |
264 | static LIST_HEAD(task_groups); |
265 | ||
29f59db3 | 266 | /* task group related information */ |
4cf86d77 | 267 | struct task_group { |
052f1dc7 | 268 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
269 | struct cgroup_subsys_state css; |
270 | #endif | |
052f1dc7 | 271 | |
6c415b92 AB |
272 | #ifdef CONFIG_USER_SCHED |
273 | uid_t uid; | |
274 | #endif | |
275 | ||
052f1dc7 | 276 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
277 | /* schedulable entities of this group on each cpu */ |
278 | struct sched_entity **se; | |
279 | /* runqueue "owned" by this group on each cpu */ | |
280 | struct cfs_rq **cfs_rq; | |
281 | unsigned long shares; | |
052f1dc7 PZ |
282 | #endif |
283 | ||
284 | #ifdef CONFIG_RT_GROUP_SCHED | |
285 | struct sched_rt_entity **rt_se; | |
286 | struct rt_rq **rt_rq; | |
287 | ||
d0b27fa7 | 288 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 289 | #endif |
6b2d7700 | 290 | |
ae8393e5 | 291 | struct rcu_head rcu; |
6f505b16 | 292 | struct list_head list; |
f473aa5e PZ |
293 | |
294 | struct task_group *parent; | |
295 | struct list_head siblings; | |
296 | struct list_head children; | |
29f59db3 SV |
297 | }; |
298 | ||
354d60c2 | 299 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 300 | |
6c415b92 AB |
301 | /* Helper function to pass uid information to create_sched_user() */ |
302 | void set_tg_uid(struct user_struct *user) | |
303 | { | |
304 | user->tg->uid = user->uid; | |
305 | } | |
306 | ||
eff766a6 PZ |
307 | /* |
308 | * Root task group. | |
309 | * Every UID task group (including init_task_group aka UID-0) will | |
310 | * be a child to this group. | |
311 | */ | |
312 | struct task_group root_task_group; | |
313 | ||
052f1dc7 | 314 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
315 | /* Default task group's sched entity on each cpu */ |
316 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
317 | /* Default task group's cfs_rq on each cpu */ | |
318 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 319 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
320 | |
321 | #ifdef CONFIG_RT_GROUP_SCHED | |
322 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
323 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 324 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 325 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 326 | #define root_task_group init_task_group |
9a7e0b18 | 327 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 328 | |
8ed36996 | 329 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
330 | * a task group's cpu shares. |
331 | */ | |
8ed36996 | 332 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 333 | |
052f1dc7 | 334 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
335 | #ifdef CONFIG_USER_SCHED |
336 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 337 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 338 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 339 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 340 | |
cb4ad1ff | 341 | /* |
2e084786 LJ |
342 | * A weight of 0 or 1 can cause arithmetics problems. |
343 | * A weight of a cfs_rq is the sum of weights of which entities | |
344 | * are queued on this cfs_rq, so a weight of a entity should not be | |
345 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
346 | * (The default weight is 1024 - so there's no practical |
347 | * limitation from this.) | |
348 | */ | |
18d95a28 | 349 | #define MIN_SHARES 2 |
2e084786 | 350 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 351 | |
052f1dc7 PZ |
352 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
353 | #endif | |
354 | ||
29f59db3 | 355 | /* Default task group. |
3a252015 | 356 | * Every task in system belong to this group at bootup. |
29f59db3 | 357 | */ |
434d53b0 | 358 | struct task_group init_task_group; |
29f59db3 SV |
359 | |
360 | /* return group to which a task belongs */ | |
4cf86d77 | 361 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 362 | { |
4cf86d77 | 363 | struct task_group *tg; |
9b5b7751 | 364 | |
052f1dc7 | 365 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
366 | rcu_read_lock(); |
367 | tg = __task_cred(p)->user->tg; | |
368 | rcu_read_unlock(); | |
052f1dc7 | 369 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
370 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
371 | struct task_group, css); | |
24e377a8 | 372 | #else |
41a2d6cf | 373 | tg = &init_task_group; |
24e377a8 | 374 | #endif |
9b5b7751 | 375 | return tg; |
29f59db3 SV |
376 | } |
377 | ||
378 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 379 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 380 | { |
052f1dc7 | 381 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
382 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
383 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 384 | #endif |
6f505b16 | 385 | |
052f1dc7 | 386 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
387 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
388 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 389 | #endif |
29f59db3 SV |
390 | } |
391 | ||
392 | #else | |
393 | ||
6f505b16 | 394 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
395 | static inline struct task_group *task_group(struct task_struct *p) |
396 | { | |
397 | return NULL; | |
398 | } | |
29f59db3 | 399 | |
052f1dc7 | 400 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 401 | |
6aa645ea IM |
402 | /* CFS-related fields in a runqueue */ |
403 | struct cfs_rq { | |
404 | struct load_weight load; | |
405 | unsigned long nr_running; | |
406 | ||
6aa645ea | 407 | u64 exec_clock; |
e9acbff6 | 408 | u64 min_vruntime; |
6aa645ea IM |
409 | |
410 | struct rb_root tasks_timeline; | |
411 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
412 | |
413 | struct list_head tasks; | |
414 | struct list_head *balance_iterator; | |
415 | ||
416 | /* | |
417 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
418 | * It is set to NULL otherwise (i.e when none are currently running). |
419 | */ | |
4793241b | 420 | struct sched_entity *curr, *next, *last; |
ddc97297 | 421 | |
5ac5c4d6 | 422 | unsigned int nr_spread_over; |
ddc97297 | 423 | |
62160e3f | 424 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
425 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
426 | ||
41a2d6cf IM |
427 | /* |
428 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
429 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
430 | * (like users, containers etc.) | |
431 | * | |
432 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
433 | * list is used during load balance. | |
434 | */ | |
41a2d6cf IM |
435 | struct list_head leaf_cfs_rq_list; |
436 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
437 | |
438 | #ifdef CONFIG_SMP | |
c09595f6 | 439 | /* |
c8cba857 | 440 | * the part of load.weight contributed by tasks |
c09595f6 | 441 | */ |
c8cba857 | 442 | unsigned long task_weight; |
c09595f6 | 443 | |
c8cba857 PZ |
444 | /* |
445 | * h_load = weight * f(tg) | |
446 | * | |
447 | * Where f(tg) is the recursive weight fraction assigned to | |
448 | * this group. | |
449 | */ | |
450 | unsigned long h_load; | |
c09595f6 | 451 | |
c8cba857 PZ |
452 | /* |
453 | * this cpu's part of tg->shares | |
454 | */ | |
455 | unsigned long shares; | |
f1d239f7 PZ |
456 | |
457 | /* | |
458 | * load.weight at the time we set shares | |
459 | */ | |
460 | unsigned long rq_weight; | |
c09595f6 | 461 | #endif |
6aa645ea IM |
462 | #endif |
463 | }; | |
1da177e4 | 464 | |
6aa645ea IM |
465 | /* Real-Time classes' related field in a runqueue: */ |
466 | struct rt_rq { | |
467 | struct rt_prio_array active; | |
63489e45 | 468 | unsigned long rt_nr_running; |
052f1dc7 | 469 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
470 | int highest_prio; /* highest queued rt task prio */ |
471 | #endif | |
fa85ae24 | 472 | #ifdef CONFIG_SMP |
73fe6aae | 473 | unsigned long rt_nr_migratory; |
a22d7fc1 | 474 | int overloaded; |
fa85ae24 | 475 | #endif |
6f505b16 | 476 | int rt_throttled; |
fa85ae24 | 477 | u64 rt_time; |
ac086bc2 | 478 | u64 rt_runtime; |
ea736ed5 | 479 | /* Nests inside the rq lock: */ |
ac086bc2 | 480 | spinlock_t rt_runtime_lock; |
6f505b16 | 481 | |
052f1dc7 | 482 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
483 | unsigned long rt_nr_boosted; |
484 | ||
6f505b16 PZ |
485 | struct rq *rq; |
486 | struct list_head leaf_rt_rq_list; | |
487 | struct task_group *tg; | |
488 | struct sched_rt_entity *rt_se; | |
489 | #endif | |
6aa645ea IM |
490 | }; |
491 | ||
57d885fe GH |
492 | #ifdef CONFIG_SMP |
493 | ||
494 | /* | |
495 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
496 | * variables. Each exclusive cpuset essentially defines an island domain by |
497 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
498 | * exclusive cpuset is created, we also create and attach a new root-domain |
499 | * object. | |
500 | * | |
57d885fe GH |
501 | */ |
502 | struct root_domain { | |
503 | atomic_t refcount; | |
c6c4927b RR |
504 | cpumask_var_t span; |
505 | cpumask_var_t online; | |
637f5085 | 506 | |
0eab9146 | 507 | /* |
637f5085 GH |
508 | * The "RT overload" flag: it gets set if a CPU has more than |
509 | * one runnable RT task. | |
510 | */ | |
c6c4927b | 511 | cpumask_var_t rto_mask; |
0eab9146 | 512 | atomic_t rto_count; |
6e0534f2 GH |
513 | #ifdef CONFIG_SMP |
514 | struct cpupri cpupri; | |
515 | #endif | |
7a09b1a2 VS |
516 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
517 | /* | |
518 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
519 | * used when most cpus are idle in the system indicating overall very | |
520 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
521 | */ | |
522 | unsigned int sched_mc_preferred_wakeup_cpu; | |
523 | #endif | |
57d885fe GH |
524 | }; |
525 | ||
dc938520 GH |
526 | /* |
527 | * By default the system creates a single root-domain with all cpus as | |
528 | * members (mimicking the global state we have today). | |
529 | */ | |
57d885fe GH |
530 | static struct root_domain def_root_domain; |
531 | ||
532 | #endif | |
533 | ||
1da177e4 LT |
534 | /* |
535 | * This is the main, per-CPU runqueue data structure. | |
536 | * | |
537 | * Locking rule: those places that want to lock multiple runqueues | |
538 | * (such as the load balancing or the thread migration code), lock | |
539 | * acquire operations must be ordered by ascending &runqueue. | |
540 | */ | |
70b97a7f | 541 | struct rq { |
d8016491 IM |
542 | /* runqueue lock: */ |
543 | spinlock_t lock; | |
1da177e4 LT |
544 | |
545 | /* | |
546 | * nr_running and cpu_load should be in the same cacheline because | |
547 | * remote CPUs use both these fields when doing load calculation. | |
548 | */ | |
549 | unsigned long nr_running; | |
6aa645ea IM |
550 | #define CPU_LOAD_IDX_MAX 5 |
551 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 552 | unsigned char idle_at_tick; |
46cb4b7c | 553 | #ifdef CONFIG_NO_HZ |
15934a37 | 554 | unsigned long last_tick_seen; |
46cb4b7c SS |
555 | unsigned char in_nohz_recently; |
556 | #endif | |
d8016491 IM |
557 | /* capture load from *all* tasks on this cpu: */ |
558 | struct load_weight load; | |
6aa645ea IM |
559 | unsigned long nr_load_updates; |
560 | u64 nr_switches; | |
561 | ||
562 | struct cfs_rq cfs; | |
6f505b16 | 563 | struct rt_rq rt; |
6f505b16 | 564 | |
6aa645ea | 565 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
566 | /* list of leaf cfs_rq on this cpu: */ |
567 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
568 | #endif |
569 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 570 | struct list_head leaf_rt_rq_list; |
1da177e4 | 571 | #endif |
1da177e4 LT |
572 | |
573 | /* | |
574 | * This is part of a global counter where only the total sum | |
575 | * over all CPUs matters. A task can increase this counter on | |
576 | * one CPU and if it got migrated afterwards it may decrease | |
577 | * it on another CPU. Always updated under the runqueue lock: | |
578 | */ | |
579 | unsigned long nr_uninterruptible; | |
580 | ||
36c8b586 | 581 | struct task_struct *curr, *idle; |
c9819f45 | 582 | unsigned long next_balance; |
1da177e4 | 583 | struct mm_struct *prev_mm; |
6aa645ea | 584 | |
3e51f33f | 585 | u64 clock; |
6aa645ea | 586 | |
1da177e4 LT |
587 | atomic_t nr_iowait; |
588 | ||
589 | #ifdef CONFIG_SMP | |
0eab9146 | 590 | struct root_domain *rd; |
1da177e4 LT |
591 | struct sched_domain *sd; |
592 | ||
593 | /* For active balancing */ | |
594 | int active_balance; | |
595 | int push_cpu; | |
d8016491 IM |
596 | /* cpu of this runqueue: */ |
597 | int cpu; | |
1f11eb6a | 598 | int online; |
1da177e4 | 599 | |
a8a51d5e | 600 | unsigned long avg_load_per_task; |
1da177e4 | 601 | |
36c8b586 | 602 | struct task_struct *migration_thread; |
1da177e4 LT |
603 | struct list_head migration_queue; |
604 | #endif | |
605 | ||
8f4d37ec | 606 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
607 | #ifdef CONFIG_SMP |
608 | int hrtick_csd_pending; | |
609 | struct call_single_data hrtick_csd; | |
610 | #endif | |
8f4d37ec PZ |
611 | struct hrtimer hrtick_timer; |
612 | #endif | |
613 | ||
1da177e4 LT |
614 | #ifdef CONFIG_SCHEDSTATS |
615 | /* latency stats */ | |
616 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
617 | unsigned long long rq_cpu_time; |
618 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
619 | |
620 | /* sys_sched_yield() stats */ | |
480b9434 KC |
621 | unsigned int yld_exp_empty; |
622 | unsigned int yld_act_empty; | |
623 | unsigned int yld_both_empty; | |
624 | unsigned int yld_count; | |
1da177e4 LT |
625 | |
626 | /* schedule() stats */ | |
480b9434 KC |
627 | unsigned int sched_switch; |
628 | unsigned int sched_count; | |
629 | unsigned int sched_goidle; | |
1da177e4 LT |
630 | |
631 | /* try_to_wake_up() stats */ | |
480b9434 KC |
632 | unsigned int ttwu_count; |
633 | unsigned int ttwu_local; | |
b8efb561 IM |
634 | |
635 | /* BKL stats */ | |
480b9434 | 636 | unsigned int bkl_count; |
1da177e4 LT |
637 | #endif |
638 | }; | |
639 | ||
f34e3b61 | 640 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 641 | |
15afe09b | 642 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 643 | { |
15afe09b | 644 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
645 | } |
646 | ||
0a2966b4 CL |
647 | static inline int cpu_of(struct rq *rq) |
648 | { | |
649 | #ifdef CONFIG_SMP | |
650 | return rq->cpu; | |
651 | #else | |
652 | return 0; | |
653 | #endif | |
654 | } | |
655 | ||
674311d5 NP |
656 | /* |
657 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 658 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
659 | * |
660 | * The domain tree of any CPU may only be accessed from within | |
661 | * preempt-disabled sections. | |
662 | */ | |
48f24c4d IM |
663 | #define for_each_domain(cpu, __sd) \ |
664 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
665 | |
666 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
667 | #define this_rq() (&__get_cpu_var(runqueues)) | |
668 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
669 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
670 | ||
3e51f33f PZ |
671 | static inline void update_rq_clock(struct rq *rq) |
672 | { | |
673 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
674 | } | |
675 | ||
bf5c91ba IM |
676 | /* |
677 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
678 | */ | |
679 | #ifdef CONFIG_SCHED_DEBUG | |
680 | # define const_debug __read_mostly | |
681 | #else | |
682 | # define const_debug static const | |
683 | #endif | |
684 | ||
017730c1 IM |
685 | /** |
686 | * runqueue_is_locked | |
687 | * | |
688 | * Returns true if the current cpu runqueue is locked. | |
689 | * This interface allows printk to be called with the runqueue lock | |
690 | * held and know whether or not it is OK to wake up the klogd. | |
691 | */ | |
692 | int runqueue_is_locked(void) | |
693 | { | |
694 | int cpu = get_cpu(); | |
695 | struct rq *rq = cpu_rq(cpu); | |
696 | int ret; | |
697 | ||
698 | ret = spin_is_locked(&rq->lock); | |
699 | put_cpu(); | |
700 | return ret; | |
701 | } | |
702 | ||
bf5c91ba IM |
703 | /* |
704 | * Debugging: various feature bits | |
705 | */ | |
f00b45c1 PZ |
706 | |
707 | #define SCHED_FEAT(name, enabled) \ | |
708 | __SCHED_FEAT_##name , | |
709 | ||
bf5c91ba | 710 | enum { |
f00b45c1 | 711 | #include "sched_features.h" |
bf5c91ba IM |
712 | }; |
713 | ||
f00b45c1 PZ |
714 | #undef SCHED_FEAT |
715 | ||
716 | #define SCHED_FEAT(name, enabled) \ | |
717 | (1UL << __SCHED_FEAT_##name) * enabled | | |
718 | ||
bf5c91ba | 719 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
720 | #include "sched_features.h" |
721 | 0; | |
722 | ||
723 | #undef SCHED_FEAT | |
724 | ||
725 | #ifdef CONFIG_SCHED_DEBUG | |
726 | #define SCHED_FEAT(name, enabled) \ | |
727 | #name , | |
728 | ||
983ed7a6 | 729 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
730 | #include "sched_features.h" |
731 | NULL | |
732 | }; | |
733 | ||
734 | #undef SCHED_FEAT | |
735 | ||
34f3a814 | 736 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 737 | { |
f00b45c1 PZ |
738 | int i; |
739 | ||
740 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
741 | if (!(sysctl_sched_features & (1UL << i))) |
742 | seq_puts(m, "NO_"); | |
743 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 744 | } |
34f3a814 | 745 | seq_puts(m, "\n"); |
f00b45c1 | 746 | |
34f3a814 | 747 | return 0; |
f00b45c1 PZ |
748 | } |
749 | ||
750 | static ssize_t | |
751 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
752 | size_t cnt, loff_t *ppos) | |
753 | { | |
754 | char buf[64]; | |
755 | char *cmp = buf; | |
756 | int neg = 0; | |
757 | int i; | |
758 | ||
759 | if (cnt > 63) | |
760 | cnt = 63; | |
761 | ||
762 | if (copy_from_user(&buf, ubuf, cnt)) | |
763 | return -EFAULT; | |
764 | ||
765 | buf[cnt] = 0; | |
766 | ||
c24b7c52 | 767 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
768 | neg = 1; |
769 | cmp += 3; | |
770 | } | |
771 | ||
772 | for (i = 0; sched_feat_names[i]; i++) { | |
773 | int len = strlen(sched_feat_names[i]); | |
774 | ||
775 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
776 | if (neg) | |
777 | sysctl_sched_features &= ~(1UL << i); | |
778 | else | |
779 | sysctl_sched_features |= (1UL << i); | |
780 | break; | |
781 | } | |
782 | } | |
783 | ||
784 | if (!sched_feat_names[i]) | |
785 | return -EINVAL; | |
786 | ||
787 | filp->f_pos += cnt; | |
788 | ||
789 | return cnt; | |
790 | } | |
791 | ||
34f3a814 LZ |
792 | static int sched_feat_open(struct inode *inode, struct file *filp) |
793 | { | |
794 | return single_open(filp, sched_feat_show, NULL); | |
795 | } | |
796 | ||
f00b45c1 | 797 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
798 | .open = sched_feat_open, |
799 | .write = sched_feat_write, | |
800 | .read = seq_read, | |
801 | .llseek = seq_lseek, | |
802 | .release = single_release, | |
f00b45c1 PZ |
803 | }; |
804 | ||
805 | static __init int sched_init_debug(void) | |
806 | { | |
f00b45c1 PZ |
807 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
808 | &sched_feat_fops); | |
809 | ||
810 | return 0; | |
811 | } | |
812 | late_initcall(sched_init_debug); | |
813 | ||
814 | #endif | |
815 | ||
816 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 817 | |
b82d9fdd PZ |
818 | /* |
819 | * Number of tasks to iterate in a single balance run. | |
820 | * Limited because this is done with IRQs disabled. | |
821 | */ | |
822 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
823 | ||
2398f2c6 PZ |
824 | /* |
825 | * ratelimit for updating the group shares. | |
55cd5340 | 826 | * default: 0.25ms |
2398f2c6 | 827 | */ |
55cd5340 | 828 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 829 | |
ffda12a1 PZ |
830 | /* |
831 | * Inject some fuzzyness into changing the per-cpu group shares | |
832 | * this avoids remote rq-locks at the expense of fairness. | |
833 | * default: 4 | |
834 | */ | |
835 | unsigned int sysctl_sched_shares_thresh = 4; | |
836 | ||
fa85ae24 | 837 | /* |
9f0c1e56 | 838 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
839 | * default: 1s |
840 | */ | |
9f0c1e56 | 841 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 842 | |
6892b75e IM |
843 | static __read_mostly int scheduler_running; |
844 | ||
9f0c1e56 PZ |
845 | /* |
846 | * part of the period that we allow rt tasks to run in us. | |
847 | * default: 0.95s | |
848 | */ | |
849 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 850 | |
d0b27fa7 PZ |
851 | static inline u64 global_rt_period(void) |
852 | { | |
853 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
854 | } | |
855 | ||
856 | static inline u64 global_rt_runtime(void) | |
857 | { | |
e26873bb | 858 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
859 | return RUNTIME_INF; |
860 | ||
861 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
862 | } | |
fa85ae24 | 863 | |
1da177e4 | 864 | #ifndef prepare_arch_switch |
4866cde0 NP |
865 | # define prepare_arch_switch(next) do { } while (0) |
866 | #endif | |
867 | #ifndef finish_arch_switch | |
868 | # define finish_arch_switch(prev) do { } while (0) | |
869 | #endif | |
870 | ||
051a1d1a DA |
871 | static inline int task_current(struct rq *rq, struct task_struct *p) |
872 | { | |
873 | return rq->curr == p; | |
874 | } | |
875 | ||
4866cde0 | 876 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 877 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 878 | { |
051a1d1a | 879 | return task_current(rq, p); |
4866cde0 NP |
880 | } |
881 | ||
70b97a7f | 882 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
883 | { |
884 | } | |
885 | ||
70b97a7f | 886 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 887 | { |
da04c035 IM |
888 | #ifdef CONFIG_DEBUG_SPINLOCK |
889 | /* this is a valid case when another task releases the spinlock */ | |
890 | rq->lock.owner = current; | |
891 | #endif | |
8a25d5de IM |
892 | /* |
893 | * If we are tracking spinlock dependencies then we have to | |
894 | * fix up the runqueue lock - which gets 'carried over' from | |
895 | * prev into current: | |
896 | */ | |
897 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
898 | ||
4866cde0 NP |
899 | spin_unlock_irq(&rq->lock); |
900 | } | |
901 | ||
902 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 903 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
904 | { |
905 | #ifdef CONFIG_SMP | |
906 | return p->oncpu; | |
907 | #else | |
051a1d1a | 908 | return task_current(rq, p); |
4866cde0 NP |
909 | #endif |
910 | } | |
911 | ||
70b97a7f | 912 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
913 | { |
914 | #ifdef CONFIG_SMP | |
915 | /* | |
916 | * We can optimise this out completely for !SMP, because the | |
917 | * SMP rebalancing from interrupt is the only thing that cares | |
918 | * here. | |
919 | */ | |
920 | next->oncpu = 1; | |
921 | #endif | |
922 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
923 | spin_unlock_irq(&rq->lock); | |
924 | #else | |
925 | spin_unlock(&rq->lock); | |
926 | #endif | |
927 | } | |
928 | ||
70b97a7f | 929 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
930 | { |
931 | #ifdef CONFIG_SMP | |
932 | /* | |
933 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
934 | * We must ensure this doesn't happen until the switch is completely | |
935 | * finished. | |
936 | */ | |
937 | smp_wmb(); | |
938 | prev->oncpu = 0; | |
939 | #endif | |
940 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
941 | local_irq_enable(); | |
1da177e4 | 942 | #endif |
4866cde0 NP |
943 | } |
944 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 945 | |
b29739f9 IM |
946 | /* |
947 | * __task_rq_lock - lock the runqueue a given task resides on. | |
948 | * Must be called interrupts disabled. | |
949 | */ | |
70b97a7f | 950 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
951 | __acquires(rq->lock) |
952 | { | |
3a5c359a AK |
953 | for (;;) { |
954 | struct rq *rq = task_rq(p); | |
955 | spin_lock(&rq->lock); | |
956 | if (likely(rq == task_rq(p))) | |
957 | return rq; | |
b29739f9 | 958 | spin_unlock(&rq->lock); |
b29739f9 | 959 | } |
b29739f9 IM |
960 | } |
961 | ||
1da177e4 LT |
962 | /* |
963 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 964 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
965 | * explicitly disabling preemption. |
966 | */ | |
70b97a7f | 967 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
968 | __acquires(rq->lock) |
969 | { | |
70b97a7f | 970 | struct rq *rq; |
1da177e4 | 971 | |
3a5c359a AK |
972 | for (;;) { |
973 | local_irq_save(*flags); | |
974 | rq = task_rq(p); | |
975 | spin_lock(&rq->lock); | |
976 | if (likely(rq == task_rq(p))) | |
977 | return rq; | |
1da177e4 | 978 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 979 | } |
1da177e4 LT |
980 | } |
981 | ||
ad474cac ON |
982 | void task_rq_unlock_wait(struct task_struct *p) |
983 | { | |
984 | struct rq *rq = task_rq(p); | |
985 | ||
986 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
987 | spin_unlock_wait(&rq->lock); | |
988 | } | |
989 | ||
a9957449 | 990 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
991 | __releases(rq->lock) |
992 | { | |
993 | spin_unlock(&rq->lock); | |
994 | } | |
995 | ||
70b97a7f | 996 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
997 | __releases(rq->lock) |
998 | { | |
999 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1000 | } | |
1001 | ||
1da177e4 | 1002 | /* |
cc2a73b5 | 1003 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1004 | */ |
a9957449 | 1005 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1006 | __acquires(rq->lock) |
1007 | { | |
70b97a7f | 1008 | struct rq *rq; |
1da177e4 LT |
1009 | |
1010 | local_irq_disable(); | |
1011 | rq = this_rq(); | |
1012 | spin_lock(&rq->lock); | |
1013 | ||
1014 | return rq; | |
1015 | } | |
1016 | ||
8f4d37ec PZ |
1017 | #ifdef CONFIG_SCHED_HRTICK |
1018 | /* | |
1019 | * Use HR-timers to deliver accurate preemption points. | |
1020 | * | |
1021 | * Its all a bit involved since we cannot program an hrt while holding the | |
1022 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1023 | * reschedule event. | |
1024 | * | |
1025 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1026 | * rq->lock. | |
1027 | */ | |
8f4d37ec PZ |
1028 | |
1029 | /* | |
1030 | * Use hrtick when: | |
1031 | * - enabled by features | |
1032 | * - hrtimer is actually high res | |
1033 | */ | |
1034 | static inline int hrtick_enabled(struct rq *rq) | |
1035 | { | |
1036 | if (!sched_feat(HRTICK)) | |
1037 | return 0; | |
ba42059f | 1038 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1039 | return 0; |
8f4d37ec PZ |
1040 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1041 | } | |
1042 | ||
8f4d37ec PZ |
1043 | static void hrtick_clear(struct rq *rq) |
1044 | { | |
1045 | if (hrtimer_active(&rq->hrtick_timer)) | |
1046 | hrtimer_cancel(&rq->hrtick_timer); | |
1047 | } | |
1048 | ||
8f4d37ec PZ |
1049 | /* |
1050 | * High-resolution timer tick. | |
1051 | * Runs from hardirq context with interrupts disabled. | |
1052 | */ | |
1053 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1054 | { | |
1055 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1056 | ||
1057 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1058 | ||
1059 | spin_lock(&rq->lock); | |
3e51f33f | 1060 | update_rq_clock(rq); |
8f4d37ec PZ |
1061 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1062 | spin_unlock(&rq->lock); | |
1063 | ||
1064 | return HRTIMER_NORESTART; | |
1065 | } | |
1066 | ||
95e904c7 | 1067 | #ifdef CONFIG_SMP |
31656519 PZ |
1068 | /* |
1069 | * called from hardirq (IPI) context | |
1070 | */ | |
1071 | static void __hrtick_start(void *arg) | |
b328ca18 | 1072 | { |
31656519 | 1073 | struct rq *rq = arg; |
b328ca18 | 1074 | |
31656519 PZ |
1075 | spin_lock(&rq->lock); |
1076 | hrtimer_restart(&rq->hrtick_timer); | |
1077 | rq->hrtick_csd_pending = 0; | |
1078 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1079 | } |
1080 | ||
31656519 PZ |
1081 | /* |
1082 | * Called to set the hrtick timer state. | |
1083 | * | |
1084 | * called with rq->lock held and irqs disabled | |
1085 | */ | |
1086 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1087 | { |
31656519 PZ |
1088 | struct hrtimer *timer = &rq->hrtick_timer; |
1089 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1090 | |
cc584b21 | 1091 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1092 | |
1093 | if (rq == this_rq()) { | |
1094 | hrtimer_restart(timer); | |
1095 | } else if (!rq->hrtick_csd_pending) { | |
1096 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); | |
1097 | rq->hrtick_csd_pending = 1; | |
1098 | } | |
b328ca18 PZ |
1099 | } |
1100 | ||
1101 | static int | |
1102 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1103 | { | |
1104 | int cpu = (int)(long)hcpu; | |
1105 | ||
1106 | switch (action) { | |
1107 | case CPU_UP_CANCELED: | |
1108 | case CPU_UP_CANCELED_FROZEN: | |
1109 | case CPU_DOWN_PREPARE: | |
1110 | case CPU_DOWN_PREPARE_FROZEN: | |
1111 | case CPU_DEAD: | |
1112 | case CPU_DEAD_FROZEN: | |
31656519 | 1113 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1114 | return NOTIFY_OK; |
1115 | } | |
1116 | ||
1117 | return NOTIFY_DONE; | |
1118 | } | |
1119 | ||
fa748203 | 1120 | static __init void init_hrtick(void) |
b328ca18 PZ |
1121 | { |
1122 | hotcpu_notifier(hotplug_hrtick, 0); | |
1123 | } | |
31656519 PZ |
1124 | #else |
1125 | /* | |
1126 | * Called to set the hrtick timer state. | |
1127 | * | |
1128 | * called with rq->lock held and irqs disabled | |
1129 | */ | |
1130 | static void hrtick_start(struct rq *rq, u64 delay) | |
1131 | { | |
1132 | hrtimer_start(&rq->hrtick_timer, ns_to_ktime(delay), HRTIMER_MODE_REL); | |
1133 | } | |
b328ca18 | 1134 | |
006c75f1 | 1135 | static inline void init_hrtick(void) |
8f4d37ec | 1136 | { |
8f4d37ec | 1137 | } |
31656519 | 1138 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1139 | |
31656519 | 1140 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1141 | { |
31656519 PZ |
1142 | #ifdef CONFIG_SMP |
1143 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1144 | |
31656519 PZ |
1145 | rq->hrtick_csd.flags = 0; |
1146 | rq->hrtick_csd.func = __hrtick_start; | |
1147 | rq->hrtick_csd.info = rq; | |
1148 | #endif | |
8f4d37ec | 1149 | |
31656519 PZ |
1150 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1151 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1152 | } |
006c75f1 | 1153 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1154 | static inline void hrtick_clear(struct rq *rq) |
1155 | { | |
1156 | } | |
1157 | ||
8f4d37ec PZ |
1158 | static inline void init_rq_hrtick(struct rq *rq) |
1159 | { | |
1160 | } | |
1161 | ||
b328ca18 PZ |
1162 | static inline void init_hrtick(void) |
1163 | { | |
1164 | } | |
006c75f1 | 1165 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1166 | |
c24d20db IM |
1167 | /* |
1168 | * resched_task - mark a task 'to be rescheduled now'. | |
1169 | * | |
1170 | * On UP this means the setting of the need_resched flag, on SMP it | |
1171 | * might also involve a cross-CPU call to trigger the scheduler on | |
1172 | * the target CPU. | |
1173 | */ | |
1174 | #ifdef CONFIG_SMP | |
1175 | ||
1176 | #ifndef tsk_is_polling | |
1177 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1178 | #endif | |
1179 | ||
31656519 | 1180 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1181 | { |
1182 | int cpu; | |
1183 | ||
1184 | assert_spin_locked(&task_rq(p)->lock); | |
1185 | ||
31656519 | 1186 | if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) |
c24d20db IM |
1187 | return; |
1188 | ||
31656519 | 1189 | set_tsk_thread_flag(p, TIF_NEED_RESCHED); |
c24d20db IM |
1190 | |
1191 | cpu = task_cpu(p); | |
1192 | if (cpu == smp_processor_id()) | |
1193 | return; | |
1194 | ||
1195 | /* NEED_RESCHED must be visible before we test polling */ | |
1196 | smp_mb(); | |
1197 | if (!tsk_is_polling(p)) | |
1198 | smp_send_reschedule(cpu); | |
1199 | } | |
1200 | ||
1201 | static void resched_cpu(int cpu) | |
1202 | { | |
1203 | struct rq *rq = cpu_rq(cpu); | |
1204 | unsigned long flags; | |
1205 | ||
1206 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1207 | return; | |
1208 | resched_task(cpu_curr(cpu)); | |
1209 | spin_unlock_irqrestore(&rq->lock, flags); | |
1210 | } | |
06d8308c TG |
1211 | |
1212 | #ifdef CONFIG_NO_HZ | |
1213 | /* | |
1214 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1215 | * idle CPU then this timer might expire before the next timer event | |
1216 | * which is scheduled to wake up that CPU. In case of a completely | |
1217 | * idle system the next event might even be infinite time into the | |
1218 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1219 | * leaves the inner idle loop so the newly added timer is taken into | |
1220 | * account when the CPU goes back to idle and evaluates the timer | |
1221 | * wheel for the next timer event. | |
1222 | */ | |
1223 | void wake_up_idle_cpu(int cpu) | |
1224 | { | |
1225 | struct rq *rq = cpu_rq(cpu); | |
1226 | ||
1227 | if (cpu == smp_processor_id()) | |
1228 | return; | |
1229 | ||
1230 | /* | |
1231 | * This is safe, as this function is called with the timer | |
1232 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1233 | * to idle and has not yet set rq->curr to idle then it will | |
1234 | * be serialized on the timer wheel base lock and take the new | |
1235 | * timer into account automatically. | |
1236 | */ | |
1237 | if (rq->curr != rq->idle) | |
1238 | return; | |
1239 | ||
1240 | /* | |
1241 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1242 | * lockless. The worst case is that the other CPU runs the | |
1243 | * idle task through an additional NOOP schedule() | |
1244 | */ | |
1245 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1246 | ||
1247 | /* NEED_RESCHED must be visible before we test polling */ | |
1248 | smp_mb(); | |
1249 | if (!tsk_is_polling(rq->idle)) | |
1250 | smp_send_reschedule(cpu); | |
1251 | } | |
6d6bc0ad | 1252 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1253 | |
6d6bc0ad | 1254 | #else /* !CONFIG_SMP */ |
31656519 | 1255 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1256 | { |
1257 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1258 | set_tsk_need_resched(p); |
c24d20db | 1259 | } |
6d6bc0ad | 1260 | #endif /* CONFIG_SMP */ |
c24d20db | 1261 | |
45bf76df IM |
1262 | #if BITS_PER_LONG == 32 |
1263 | # define WMULT_CONST (~0UL) | |
1264 | #else | |
1265 | # define WMULT_CONST (1UL << 32) | |
1266 | #endif | |
1267 | ||
1268 | #define WMULT_SHIFT 32 | |
1269 | ||
194081eb IM |
1270 | /* |
1271 | * Shift right and round: | |
1272 | */ | |
cf2ab469 | 1273 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1274 | |
a7be37ac PZ |
1275 | /* |
1276 | * delta *= weight / lw | |
1277 | */ | |
cb1c4fc9 | 1278 | static unsigned long |
45bf76df IM |
1279 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1280 | struct load_weight *lw) | |
1281 | { | |
1282 | u64 tmp; | |
1283 | ||
7a232e03 LJ |
1284 | if (!lw->inv_weight) { |
1285 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1286 | lw->inv_weight = 1; | |
1287 | else | |
1288 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1289 | / (lw->weight+1); | |
1290 | } | |
45bf76df IM |
1291 | |
1292 | tmp = (u64)delta_exec * weight; | |
1293 | /* | |
1294 | * Check whether we'd overflow the 64-bit multiplication: | |
1295 | */ | |
194081eb | 1296 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1297 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1298 | WMULT_SHIFT/2); |
1299 | else | |
cf2ab469 | 1300 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1301 | |
ecf691da | 1302 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1303 | } |
1304 | ||
1091985b | 1305 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1306 | { |
1307 | lw->weight += inc; | |
e89996ae | 1308 | lw->inv_weight = 0; |
45bf76df IM |
1309 | } |
1310 | ||
1091985b | 1311 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1312 | { |
1313 | lw->weight -= dec; | |
e89996ae | 1314 | lw->inv_weight = 0; |
45bf76df IM |
1315 | } |
1316 | ||
2dd73a4f PW |
1317 | /* |
1318 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1319 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1320 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1321 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1322 | * scaled version of the new time slice allocation that they receive on time |
1323 | * slice expiry etc. | |
1324 | */ | |
1325 | ||
cce7ade8 PZ |
1326 | #define WEIGHT_IDLEPRIO 3 |
1327 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1328 | |
1329 | /* | |
1330 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1331 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1332 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1333 | * that remained on nice 0. | |
1334 | * | |
1335 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1336 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1337 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1338 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1339 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1340 | */ |
1341 | static const int prio_to_weight[40] = { | |
254753dc IM |
1342 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1343 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1344 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1345 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1346 | /* 0 */ 1024, 820, 655, 526, 423, | |
1347 | /* 5 */ 335, 272, 215, 172, 137, | |
1348 | /* 10 */ 110, 87, 70, 56, 45, | |
1349 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1350 | }; |
1351 | ||
5714d2de IM |
1352 | /* |
1353 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1354 | * | |
1355 | * In cases where the weight does not change often, we can use the | |
1356 | * precalculated inverse to speed up arithmetics by turning divisions | |
1357 | * into multiplications: | |
1358 | */ | |
dd41f596 | 1359 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1360 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1361 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1362 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1363 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1364 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1365 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1366 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1367 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1368 | }; |
2dd73a4f | 1369 | |
dd41f596 IM |
1370 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1371 | ||
1372 | /* | |
1373 | * runqueue iterator, to support SMP load-balancing between different | |
1374 | * scheduling classes, without having to expose their internal data | |
1375 | * structures to the load-balancing proper: | |
1376 | */ | |
1377 | struct rq_iterator { | |
1378 | void *arg; | |
1379 | struct task_struct *(*start)(void *); | |
1380 | struct task_struct *(*next)(void *); | |
1381 | }; | |
1382 | ||
e1d1484f PW |
1383 | #ifdef CONFIG_SMP |
1384 | static unsigned long | |
1385 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1386 | unsigned long max_load_move, struct sched_domain *sd, | |
1387 | enum cpu_idle_type idle, int *all_pinned, | |
1388 | int *this_best_prio, struct rq_iterator *iterator); | |
1389 | ||
1390 | static int | |
1391 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1392 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1393 | struct rq_iterator *iterator); | |
e1d1484f | 1394 | #endif |
dd41f596 | 1395 | |
d842de87 SV |
1396 | #ifdef CONFIG_CGROUP_CPUACCT |
1397 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1398 | #else | |
1399 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1400 | #endif | |
1401 | ||
18d95a28 PZ |
1402 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1403 | { | |
1404 | update_load_add(&rq->load, load); | |
1405 | } | |
1406 | ||
1407 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1408 | { | |
1409 | update_load_sub(&rq->load, load); | |
1410 | } | |
1411 | ||
7940ca36 | 1412 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1413 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1414 | |
1415 | /* | |
1416 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1417 | * leaving it for the final time. | |
1418 | */ | |
eb755805 | 1419 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1420 | { |
1421 | struct task_group *parent, *child; | |
eb755805 | 1422 | int ret; |
c09595f6 PZ |
1423 | |
1424 | rcu_read_lock(); | |
1425 | parent = &root_task_group; | |
1426 | down: | |
eb755805 PZ |
1427 | ret = (*down)(parent, data); |
1428 | if (ret) | |
1429 | goto out_unlock; | |
c09595f6 PZ |
1430 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1431 | parent = child; | |
1432 | goto down; | |
1433 | ||
1434 | up: | |
1435 | continue; | |
1436 | } | |
eb755805 PZ |
1437 | ret = (*up)(parent, data); |
1438 | if (ret) | |
1439 | goto out_unlock; | |
c09595f6 PZ |
1440 | |
1441 | child = parent; | |
1442 | parent = parent->parent; | |
1443 | if (parent) | |
1444 | goto up; | |
eb755805 | 1445 | out_unlock: |
c09595f6 | 1446 | rcu_read_unlock(); |
eb755805 PZ |
1447 | |
1448 | return ret; | |
c09595f6 PZ |
1449 | } |
1450 | ||
eb755805 PZ |
1451 | static int tg_nop(struct task_group *tg, void *data) |
1452 | { | |
1453 | return 0; | |
c09595f6 | 1454 | } |
eb755805 PZ |
1455 | #endif |
1456 | ||
1457 | #ifdef CONFIG_SMP | |
1458 | static unsigned long source_load(int cpu, int type); | |
1459 | static unsigned long target_load(int cpu, int type); | |
1460 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1461 | ||
1462 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1463 | { | |
1464 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1465 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1466 | |
4cd42620 SR |
1467 | if (nr_running) |
1468 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1469 | else |
1470 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1471 | |
1472 | return rq->avg_load_per_task; | |
1473 | } | |
1474 | ||
1475 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1476 | |
c09595f6 PZ |
1477 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1478 | ||
1479 | /* | |
1480 | * Calculate and set the cpu's group shares. | |
1481 | */ | |
1482 | static void | |
ffda12a1 PZ |
1483 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1484 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1485 | { |
c09595f6 PZ |
1486 | unsigned long shares; |
1487 | unsigned long rq_weight; | |
1488 | ||
c8cba857 | 1489 | if (!tg->se[cpu]) |
c09595f6 PZ |
1490 | return; |
1491 | ||
ec4e0e2f | 1492 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1493 | |
c09595f6 PZ |
1494 | /* |
1495 | * \Sum shares * rq_weight | |
1496 | * shares = ----------------------- | |
1497 | * \Sum rq_weight | |
1498 | * | |
1499 | */ | |
ec4e0e2f | 1500 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1501 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1502 | |
ffda12a1 PZ |
1503 | if (abs(shares - tg->se[cpu]->load.weight) > |
1504 | sysctl_sched_shares_thresh) { | |
1505 | struct rq *rq = cpu_rq(cpu); | |
1506 | unsigned long flags; | |
c09595f6 | 1507 | |
ffda12a1 | 1508 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1509 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1510 | |
ffda12a1 PZ |
1511 | __set_se_shares(tg->se[cpu], shares); |
1512 | spin_unlock_irqrestore(&rq->lock, flags); | |
1513 | } | |
18d95a28 | 1514 | } |
c09595f6 PZ |
1515 | |
1516 | /* | |
c8cba857 PZ |
1517 | * Re-compute the task group their per cpu shares over the given domain. |
1518 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1519 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1520 | */ |
eb755805 | 1521 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1522 | { |
ec4e0e2f | 1523 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1524 | unsigned long shares = 0; |
eb755805 | 1525 | struct sched_domain *sd = data; |
c8cba857 | 1526 | int i; |
c09595f6 | 1527 | |
758b2cdc | 1528 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1529 | /* |
1530 | * If there are currently no tasks on the cpu pretend there | |
1531 | * is one of average load so that when a new task gets to | |
1532 | * run here it will not get delayed by group starvation. | |
1533 | */ | |
1534 | weight = tg->cfs_rq[i]->load.weight; | |
1535 | if (!weight) | |
1536 | weight = NICE_0_LOAD; | |
1537 | ||
1538 | tg->cfs_rq[i]->rq_weight = weight; | |
1539 | rq_weight += weight; | |
c8cba857 | 1540 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1541 | } |
c09595f6 | 1542 | |
c8cba857 PZ |
1543 | if ((!shares && rq_weight) || shares > tg->shares) |
1544 | shares = tg->shares; | |
1545 | ||
1546 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1547 | shares = tg->shares; | |
c09595f6 | 1548 | |
758b2cdc | 1549 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1550 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1551 | |
1552 | return 0; | |
c09595f6 PZ |
1553 | } |
1554 | ||
1555 | /* | |
c8cba857 PZ |
1556 | * Compute the cpu's hierarchical load factor for each task group. |
1557 | * This needs to be done in a top-down fashion because the load of a child | |
1558 | * group is a fraction of its parents load. | |
c09595f6 | 1559 | */ |
eb755805 | 1560 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1561 | { |
c8cba857 | 1562 | unsigned long load; |
eb755805 | 1563 | long cpu = (long)data; |
c09595f6 | 1564 | |
c8cba857 PZ |
1565 | if (!tg->parent) { |
1566 | load = cpu_rq(cpu)->load.weight; | |
1567 | } else { | |
1568 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1569 | load *= tg->cfs_rq[cpu]->shares; | |
1570 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1571 | } | |
c09595f6 | 1572 | |
c8cba857 | 1573 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1574 | |
eb755805 | 1575 | return 0; |
c09595f6 PZ |
1576 | } |
1577 | ||
c8cba857 | 1578 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1579 | { |
2398f2c6 PZ |
1580 | u64 now = cpu_clock(raw_smp_processor_id()); |
1581 | s64 elapsed = now - sd->last_update; | |
1582 | ||
1583 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1584 | sd->last_update = now; | |
eb755805 | 1585 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1586 | } |
4d8d595d PZ |
1587 | } |
1588 | ||
3e5459b4 PZ |
1589 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1590 | { | |
1591 | spin_unlock(&rq->lock); | |
1592 | update_shares(sd); | |
1593 | spin_lock(&rq->lock); | |
1594 | } | |
1595 | ||
eb755805 | 1596 | static void update_h_load(long cpu) |
c09595f6 | 1597 | { |
eb755805 | 1598 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1599 | } |
1600 | ||
c09595f6 PZ |
1601 | #else |
1602 | ||
c8cba857 | 1603 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1604 | { |
1605 | } | |
1606 | ||
3e5459b4 PZ |
1607 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1608 | { | |
1609 | } | |
1610 | ||
18d95a28 PZ |
1611 | #endif |
1612 | ||
70574a99 AD |
1613 | /* |
1614 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1615 | */ | |
1616 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1617 | __releases(this_rq->lock) | |
1618 | __acquires(busiest->lock) | |
1619 | __acquires(this_rq->lock) | |
1620 | { | |
1621 | int ret = 0; | |
1622 | ||
1623 | if (unlikely(!irqs_disabled())) { | |
1624 | /* printk() doesn't work good under rq->lock */ | |
1625 | spin_unlock(&this_rq->lock); | |
1626 | BUG_ON(1); | |
1627 | } | |
1628 | if (unlikely(!spin_trylock(&busiest->lock))) { | |
1629 | if (busiest < this_rq) { | |
1630 | spin_unlock(&this_rq->lock); | |
1631 | spin_lock(&busiest->lock); | |
1632 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1633 | ret = 1; | |
1634 | } else | |
1635 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1636 | } | |
1637 | return ret; | |
1638 | } | |
1639 | ||
1640 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) | |
1641 | __releases(busiest->lock) | |
1642 | { | |
1643 | spin_unlock(&busiest->lock); | |
1644 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1645 | } | |
18d95a28 PZ |
1646 | #endif |
1647 | ||
30432094 | 1648 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1649 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1650 | { | |
30432094 | 1651 | #ifdef CONFIG_SMP |
34e83e85 IM |
1652 | cfs_rq->shares = shares; |
1653 | #endif | |
1654 | } | |
30432094 | 1655 | #endif |
e7693a36 | 1656 | |
dd41f596 | 1657 | #include "sched_stats.h" |
dd41f596 | 1658 | #include "sched_idletask.c" |
5522d5d5 IM |
1659 | #include "sched_fair.c" |
1660 | #include "sched_rt.c" | |
dd41f596 IM |
1661 | #ifdef CONFIG_SCHED_DEBUG |
1662 | # include "sched_debug.c" | |
1663 | #endif | |
1664 | ||
1665 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1666 | #define for_each_class(class) \ |
1667 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1668 | |
c09595f6 | 1669 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1670 | { |
1671 | rq->nr_running++; | |
9c217245 IM |
1672 | } |
1673 | ||
c09595f6 | 1674 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1675 | { |
1676 | rq->nr_running--; | |
9c217245 IM |
1677 | } |
1678 | ||
45bf76df IM |
1679 | static void set_load_weight(struct task_struct *p) |
1680 | { | |
1681 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1682 | p->se.load.weight = prio_to_weight[0] * 2; |
1683 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1684 | return; | |
1685 | } | |
45bf76df | 1686 | |
dd41f596 IM |
1687 | /* |
1688 | * SCHED_IDLE tasks get minimal weight: | |
1689 | */ | |
1690 | if (p->policy == SCHED_IDLE) { | |
1691 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1692 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1693 | return; | |
1694 | } | |
71f8bd46 | 1695 | |
dd41f596 IM |
1696 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1697 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1698 | } |
1699 | ||
2087a1ad GH |
1700 | static void update_avg(u64 *avg, u64 sample) |
1701 | { | |
1702 | s64 diff = sample - *avg; | |
1703 | *avg += diff >> 3; | |
1704 | } | |
1705 | ||
8159f87e | 1706 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1707 | { |
dd41f596 | 1708 | sched_info_queued(p); |
fd390f6a | 1709 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1710 | p->se.on_rq = 1; |
71f8bd46 IM |
1711 | } |
1712 | ||
69be72c1 | 1713 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1714 | { |
2087a1ad GH |
1715 | if (sleep && p->se.last_wakeup) { |
1716 | update_avg(&p->se.avg_overlap, | |
1717 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1718 | p->se.last_wakeup = 0; | |
1719 | } | |
1720 | ||
46ac22ba | 1721 | sched_info_dequeued(p); |
f02231e5 | 1722 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1723 | p->se.on_rq = 0; |
71f8bd46 IM |
1724 | } |
1725 | ||
14531189 | 1726 | /* |
dd41f596 | 1727 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1728 | */ |
14531189 IM |
1729 | static inline int __normal_prio(struct task_struct *p) |
1730 | { | |
dd41f596 | 1731 | return p->static_prio; |
14531189 IM |
1732 | } |
1733 | ||
b29739f9 IM |
1734 | /* |
1735 | * Calculate the expected normal priority: i.e. priority | |
1736 | * without taking RT-inheritance into account. Might be | |
1737 | * boosted by interactivity modifiers. Changes upon fork, | |
1738 | * setprio syscalls, and whenever the interactivity | |
1739 | * estimator recalculates. | |
1740 | */ | |
36c8b586 | 1741 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1742 | { |
1743 | int prio; | |
1744 | ||
e05606d3 | 1745 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1746 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1747 | else | |
1748 | prio = __normal_prio(p); | |
1749 | return prio; | |
1750 | } | |
1751 | ||
1752 | /* | |
1753 | * Calculate the current priority, i.e. the priority | |
1754 | * taken into account by the scheduler. This value might | |
1755 | * be boosted by RT tasks, or might be boosted by | |
1756 | * interactivity modifiers. Will be RT if the task got | |
1757 | * RT-boosted. If not then it returns p->normal_prio. | |
1758 | */ | |
36c8b586 | 1759 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1760 | { |
1761 | p->normal_prio = normal_prio(p); | |
1762 | /* | |
1763 | * If we are RT tasks or we were boosted to RT priority, | |
1764 | * keep the priority unchanged. Otherwise, update priority | |
1765 | * to the normal priority: | |
1766 | */ | |
1767 | if (!rt_prio(p->prio)) | |
1768 | return p->normal_prio; | |
1769 | return p->prio; | |
1770 | } | |
1771 | ||
1da177e4 | 1772 | /* |
dd41f596 | 1773 | * activate_task - move a task to the runqueue. |
1da177e4 | 1774 | */ |
dd41f596 | 1775 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1776 | { |
d9514f6c | 1777 | if (task_contributes_to_load(p)) |
dd41f596 | 1778 | rq->nr_uninterruptible--; |
1da177e4 | 1779 | |
8159f87e | 1780 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1781 | inc_nr_running(rq); |
1da177e4 LT |
1782 | } |
1783 | ||
1da177e4 LT |
1784 | /* |
1785 | * deactivate_task - remove a task from the runqueue. | |
1786 | */ | |
2e1cb74a | 1787 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1788 | { |
d9514f6c | 1789 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1790 | rq->nr_uninterruptible++; |
1791 | ||
69be72c1 | 1792 | dequeue_task(rq, p, sleep); |
c09595f6 | 1793 | dec_nr_running(rq); |
1da177e4 LT |
1794 | } |
1795 | ||
1da177e4 LT |
1796 | /** |
1797 | * task_curr - is this task currently executing on a CPU? | |
1798 | * @p: the task in question. | |
1799 | */ | |
36c8b586 | 1800 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1801 | { |
1802 | return cpu_curr(task_cpu(p)) == p; | |
1803 | } | |
1804 | ||
dd41f596 IM |
1805 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1806 | { | |
6f505b16 | 1807 | set_task_rq(p, cpu); |
dd41f596 | 1808 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1809 | /* |
1810 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1811 | * successfuly executed on another CPU. We must ensure that updates of | |
1812 | * per-task data have been completed by this moment. | |
1813 | */ | |
1814 | smp_wmb(); | |
dd41f596 | 1815 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1816 | #endif |
2dd73a4f PW |
1817 | } |
1818 | ||
cb469845 SR |
1819 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1820 | const struct sched_class *prev_class, | |
1821 | int oldprio, int running) | |
1822 | { | |
1823 | if (prev_class != p->sched_class) { | |
1824 | if (prev_class->switched_from) | |
1825 | prev_class->switched_from(rq, p, running); | |
1826 | p->sched_class->switched_to(rq, p, running); | |
1827 | } else | |
1828 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1829 | } | |
1830 | ||
1da177e4 | 1831 | #ifdef CONFIG_SMP |
c65cc870 | 1832 | |
e958b360 TG |
1833 | /* Used instead of source_load when we know the type == 0 */ |
1834 | static unsigned long weighted_cpuload(const int cpu) | |
1835 | { | |
1836 | return cpu_rq(cpu)->load.weight; | |
1837 | } | |
1838 | ||
cc367732 IM |
1839 | /* |
1840 | * Is this task likely cache-hot: | |
1841 | */ | |
e7693a36 | 1842 | static int |
cc367732 IM |
1843 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1844 | { | |
1845 | s64 delta; | |
1846 | ||
f540a608 IM |
1847 | /* |
1848 | * Buddy candidates are cache hot: | |
1849 | */ | |
4793241b PZ |
1850 | if (sched_feat(CACHE_HOT_BUDDY) && |
1851 | (&p->se == cfs_rq_of(&p->se)->next || | |
1852 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1853 | return 1; |
1854 | ||
cc367732 IM |
1855 | if (p->sched_class != &fair_sched_class) |
1856 | return 0; | |
1857 | ||
6bc1665b IM |
1858 | if (sysctl_sched_migration_cost == -1) |
1859 | return 1; | |
1860 | if (sysctl_sched_migration_cost == 0) | |
1861 | return 0; | |
1862 | ||
cc367732 IM |
1863 | delta = now - p->se.exec_start; |
1864 | ||
1865 | return delta < (s64)sysctl_sched_migration_cost; | |
1866 | } | |
1867 | ||
1868 | ||
dd41f596 | 1869 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1870 | { |
dd41f596 IM |
1871 | int old_cpu = task_cpu(p); |
1872 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1873 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1874 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1875 | u64 clock_offset; |
dd41f596 IM |
1876 | |
1877 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1878 | |
cbc34ed1 PZ |
1879 | trace_sched_migrate_task(p, task_cpu(p), new_cpu); |
1880 | ||
6cfb0d5d IM |
1881 | #ifdef CONFIG_SCHEDSTATS |
1882 | if (p->se.wait_start) | |
1883 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1884 | if (p->se.sleep_start) |
1885 | p->se.sleep_start -= clock_offset; | |
1886 | if (p->se.block_start) | |
1887 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1888 | if (old_cpu != new_cpu) { |
1889 | schedstat_inc(p, se.nr_migrations); | |
1890 | if (task_hot(p, old_rq->clock, NULL)) | |
1891 | schedstat_inc(p, se.nr_forced2_migrations); | |
1892 | } | |
6cfb0d5d | 1893 | #endif |
2830cf8c SV |
1894 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1895 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1896 | |
1897 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1898 | } |
1899 | ||
70b97a7f | 1900 | struct migration_req { |
1da177e4 | 1901 | struct list_head list; |
1da177e4 | 1902 | |
36c8b586 | 1903 | struct task_struct *task; |
1da177e4 LT |
1904 | int dest_cpu; |
1905 | ||
1da177e4 | 1906 | struct completion done; |
70b97a7f | 1907 | }; |
1da177e4 LT |
1908 | |
1909 | /* | |
1910 | * The task's runqueue lock must be held. | |
1911 | * Returns true if you have to wait for migration thread. | |
1912 | */ | |
36c8b586 | 1913 | static int |
70b97a7f | 1914 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1915 | { |
70b97a7f | 1916 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1917 | |
1918 | /* | |
1919 | * If the task is not on a runqueue (and not running), then | |
1920 | * it is sufficient to simply update the task's cpu field. | |
1921 | */ | |
dd41f596 | 1922 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1923 | set_task_cpu(p, dest_cpu); |
1924 | return 0; | |
1925 | } | |
1926 | ||
1927 | init_completion(&req->done); | |
1da177e4 LT |
1928 | req->task = p; |
1929 | req->dest_cpu = dest_cpu; | |
1930 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1931 | |
1da177e4 LT |
1932 | return 1; |
1933 | } | |
1934 | ||
1935 | /* | |
1936 | * wait_task_inactive - wait for a thread to unschedule. | |
1937 | * | |
85ba2d86 RM |
1938 | * If @match_state is nonzero, it's the @p->state value just checked and |
1939 | * not expected to change. If it changes, i.e. @p might have woken up, | |
1940 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
1941 | * we return a positive number (its total switch count). If a second call | |
1942 | * a short while later returns the same number, the caller can be sure that | |
1943 | * @p has remained unscheduled the whole time. | |
1944 | * | |
1da177e4 LT |
1945 | * The caller must ensure that the task *will* unschedule sometime soon, |
1946 | * else this function might spin for a *long* time. This function can't | |
1947 | * be called with interrupts off, or it may introduce deadlock with | |
1948 | * smp_call_function() if an IPI is sent by the same process we are | |
1949 | * waiting to become inactive. | |
1950 | */ | |
85ba2d86 | 1951 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
1952 | { |
1953 | unsigned long flags; | |
dd41f596 | 1954 | int running, on_rq; |
85ba2d86 | 1955 | unsigned long ncsw; |
70b97a7f | 1956 | struct rq *rq; |
1da177e4 | 1957 | |
3a5c359a AK |
1958 | for (;;) { |
1959 | /* | |
1960 | * We do the initial early heuristics without holding | |
1961 | * any task-queue locks at all. We'll only try to get | |
1962 | * the runqueue lock when things look like they will | |
1963 | * work out! | |
1964 | */ | |
1965 | rq = task_rq(p); | |
fa490cfd | 1966 | |
3a5c359a AK |
1967 | /* |
1968 | * If the task is actively running on another CPU | |
1969 | * still, just relax and busy-wait without holding | |
1970 | * any locks. | |
1971 | * | |
1972 | * NOTE! Since we don't hold any locks, it's not | |
1973 | * even sure that "rq" stays as the right runqueue! | |
1974 | * But we don't care, since "task_running()" will | |
1975 | * return false if the runqueue has changed and p | |
1976 | * is actually now running somewhere else! | |
1977 | */ | |
85ba2d86 RM |
1978 | while (task_running(rq, p)) { |
1979 | if (match_state && unlikely(p->state != match_state)) | |
1980 | return 0; | |
3a5c359a | 1981 | cpu_relax(); |
85ba2d86 | 1982 | } |
fa490cfd | 1983 | |
3a5c359a AK |
1984 | /* |
1985 | * Ok, time to look more closely! We need the rq | |
1986 | * lock now, to be *sure*. If we're wrong, we'll | |
1987 | * just go back and repeat. | |
1988 | */ | |
1989 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 1990 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
1991 | running = task_running(rq, p); |
1992 | on_rq = p->se.on_rq; | |
85ba2d86 | 1993 | ncsw = 0; |
f31e11d8 | 1994 | if (!match_state || p->state == match_state) |
93dcf55f | 1995 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 1996 | task_rq_unlock(rq, &flags); |
fa490cfd | 1997 | |
85ba2d86 RM |
1998 | /* |
1999 | * If it changed from the expected state, bail out now. | |
2000 | */ | |
2001 | if (unlikely(!ncsw)) | |
2002 | break; | |
2003 | ||
3a5c359a AK |
2004 | /* |
2005 | * Was it really running after all now that we | |
2006 | * checked with the proper locks actually held? | |
2007 | * | |
2008 | * Oops. Go back and try again.. | |
2009 | */ | |
2010 | if (unlikely(running)) { | |
2011 | cpu_relax(); | |
2012 | continue; | |
2013 | } | |
fa490cfd | 2014 | |
3a5c359a AK |
2015 | /* |
2016 | * It's not enough that it's not actively running, | |
2017 | * it must be off the runqueue _entirely_, and not | |
2018 | * preempted! | |
2019 | * | |
2020 | * So if it wa still runnable (but just not actively | |
2021 | * running right now), it's preempted, and we should | |
2022 | * yield - it could be a while. | |
2023 | */ | |
2024 | if (unlikely(on_rq)) { | |
2025 | schedule_timeout_uninterruptible(1); | |
2026 | continue; | |
2027 | } | |
fa490cfd | 2028 | |
3a5c359a AK |
2029 | /* |
2030 | * Ahh, all good. It wasn't running, and it wasn't | |
2031 | * runnable, which means that it will never become | |
2032 | * running in the future either. We're all done! | |
2033 | */ | |
2034 | break; | |
2035 | } | |
85ba2d86 RM |
2036 | |
2037 | return ncsw; | |
1da177e4 LT |
2038 | } |
2039 | ||
2040 | /*** | |
2041 | * kick_process - kick a running thread to enter/exit the kernel | |
2042 | * @p: the to-be-kicked thread | |
2043 | * | |
2044 | * Cause a process which is running on another CPU to enter | |
2045 | * kernel-mode, without any delay. (to get signals handled.) | |
2046 | * | |
2047 | * NOTE: this function doesnt have to take the runqueue lock, | |
2048 | * because all it wants to ensure is that the remote task enters | |
2049 | * the kernel. If the IPI races and the task has been migrated | |
2050 | * to another CPU then no harm is done and the purpose has been | |
2051 | * achieved as well. | |
2052 | */ | |
36c8b586 | 2053 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2054 | { |
2055 | int cpu; | |
2056 | ||
2057 | preempt_disable(); | |
2058 | cpu = task_cpu(p); | |
2059 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2060 | smp_send_reschedule(cpu); | |
2061 | preempt_enable(); | |
2062 | } | |
2063 | ||
2064 | /* | |
2dd73a4f PW |
2065 | * Return a low guess at the load of a migration-source cpu weighted |
2066 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2067 | * |
2068 | * We want to under-estimate the load of migration sources, to | |
2069 | * balance conservatively. | |
2070 | */ | |
a9957449 | 2071 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2072 | { |
70b97a7f | 2073 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2074 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2075 | |
93b75217 | 2076 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2077 | return total; |
b910472d | 2078 | |
dd41f596 | 2079 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2080 | } |
2081 | ||
2082 | /* | |
2dd73a4f PW |
2083 | * Return a high guess at the load of a migration-target cpu weighted |
2084 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2085 | */ |
a9957449 | 2086 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2087 | { |
70b97a7f | 2088 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2089 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2090 | |
93b75217 | 2091 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2092 | return total; |
3b0bd9bc | 2093 | |
dd41f596 | 2094 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2095 | } |
2096 | ||
147cbb4b NP |
2097 | /* |
2098 | * find_idlest_group finds and returns the least busy CPU group within the | |
2099 | * domain. | |
2100 | */ | |
2101 | static struct sched_group * | |
2102 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2103 | { | |
2104 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2105 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2106 | int load_idx = sd->forkexec_idx; | |
2107 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2108 | ||
2109 | do { | |
2110 | unsigned long load, avg_load; | |
2111 | int local_group; | |
2112 | int i; | |
2113 | ||
da5a5522 | 2114 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2115 | if (!cpumask_intersects(sched_group_cpus(group), |
2116 | &p->cpus_allowed)) | |
3a5c359a | 2117 | continue; |
da5a5522 | 2118 | |
758b2cdc RR |
2119 | local_group = cpumask_test_cpu(this_cpu, |
2120 | sched_group_cpus(group)); | |
147cbb4b NP |
2121 | |
2122 | /* Tally up the load of all CPUs in the group */ | |
2123 | avg_load = 0; | |
2124 | ||
758b2cdc | 2125 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2126 | /* Bias balancing toward cpus of our domain */ |
2127 | if (local_group) | |
2128 | load = source_load(i, load_idx); | |
2129 | else | |
2130 | load = target_load(i, load_idx); | |
2131 | ||
2132 | avg_load += load; | |
2133 | } | |
2134 | ||
2135 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2136 | avg_load = sg_div_cpu_power(group, |
2137 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2138 | |
2139 | if (local_group) { | |
2140 | this_load = avg_load; | |
2141 | this = group; | |
2142 | } else if (avg_load < min_load) { | |
2143 | min_load = avg_load; | |
2144 | idlest = group; | |
2145 | } | |
3a5c359a | 2146 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2147 | |
2148 | if (!idlest || 100*this_load < imbalance*min_load) | |
2149 | return NULL; | |
2150 | return idlest; | |
2151 | } | |
2152 | ||
2153 | /* | |
0feaece9 | 2154 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2155 | */ |
95cdf3b7 | 2156 | static int |
758b2cdc | 2157 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2158 | { |
2159 | unsigned long load, min_load = ULONG_MAX; | |
2160 | int idlest = -1; | |
2161 | int i; | |
2162 | ||
da5a5522 | 2163 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2164 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2165 | load = weighted_cpuload(i); |
147cbb4b NP |
2166 | |
2167 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2168 | min_load = load; | |
2169 | idlest = i; | |
2170 | } | |
2171 | } | |
2172 | ||
2173 | return idlest; | |
2174 | } | |
2175 | ||
476d139c NP |
2176 | /* |
2177 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2178 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2179 | * SD_BALANCE_EXEC. | |
2180 | * | |
2181 | * Balance, ie. select the least loaded group. | |
2182 | * | |
2183 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2184 | * | |
2185 | * preempt must be disabled. | |
2186 | */ | |
2187 | static int sched_balance_self(int cpu, int flag) | |
2188 | { | |
2189 | struct task_struct *t = current; | |
2190 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2191 | |
c96d145e | 2192 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2193 | /* |
2194 | * If power savings logic is enabled for a domain, stop there. | |
2195 | */ | |
5c45bf27 SS |
2196 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2197 | break; | |
476d139c NP |
2198 | if (tmp->flags & flag) |
2199 | sd = tmp; | |
c96d145e | 2200 | } |
476d139c | 2201 | |
039a1c41 PZ |
2202 | if (sd) |
2203 | update_shares(sd); | |
2204 | ||
476d139c | 2205 | while (sd) { |
476d139c | 2206 | struct sched_group *group; |
1a848870 SS |
2207 | int new_cpu, weight; |
2208 | ||
2209 | if (!(sd->flags & flag)) { | |
2210 | sd = sd->child; | |
2211 | continue; | |
2212 | } | |
476d139c | 2213 | |
476d139c | 2214 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2215 | if (!group) { |
2216 | sd = sd->child; | |
2217 | continue; | |
2218 | } | |
476d139c | 2219 | |
758b2cdc | 2220 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2221 | if (new_cpu == -1 || new_cpu == cpu) { |
2222 | /* Now try balancing at a lower domain level of cpu */ | |
2223 | sd = sd->child; | |
2224 | continue; | |
2225 | } | |
476d139c | 2226 | |
1a848870 | 2227 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2228 | cpu = new_cpu; |
758b2cdc | 2229 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2230 | sd = NULL; |
476d139c | 2231 | for_each_domain(cpu, tmp) { |
758b2cdc | 2232 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2233 | break; |
2234 | if (tmp->flags & flag) | |
2235 | sd = tmp; | |
2236 | } | |
2237 | /* while loop will break here if sd == NULL */ | |
2238 | } | |
2239 | ||
2240 | return cpu; | |
2241 | } | |
2242 | ||
2243 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2244 | |
1da177e4 LT |
2245 | /*** |
2246 | * try_to_wake_up - wake up a thread | |
2247 | * @p: the to-be-woken-up thread | |
2248 | * @state: the mask of task states that can be woken | |
2249 | * @sync: do a synchronous wakeup? | |
2250 | * | |
2251 | * Put it on the run-queue if it's not already there. The "current" | |
2252 | * thread is always on the run-queue (except when the actual | |
2253 | * re-schedule is in progress), and as such you're allowed to do | |
2254 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2255 | * runnable without the overhead of this. | |
2256 | * | |
2257 | * returns failure only if the task is already active. | |
2258 | */ | |
36c8b586 | 2259 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2260 | { |
cc367732 | 2261 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2262 | unsigned long flags; |
2263 | long old_state; | |
70b97a7f | 2264 | struct rq *rq; |
1da177e4 | 2265 | |
b85d0667 IM |
2266 | if (!sched_feat(SYNC_WAKEUPS)) |
2267 | sync = 0; | |
2268 | ||
d942fb6c PZ |
2269 | if (!sync && (current->se.avg_overlap < sysctl_sched_migration_cost && |
2270 | p->se.avg_overlap < sysctl_sched_migration_cost)) | |
2271 | sync = 1; | |
2272 | ||
2398f2c6 PZ |
2273 | #ifdef CONFIG_SMP |
2274 | if (sched_feat(LB_WAKEUP_UPDATE)) { | |
2275 | struct sched_domain *sd; | |
2276 | ||
2277 | this_cpu = raw_smp_processor_id(); | |
2278 | cpu = task_cpu(p); | |
2279 | ||
2280 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2281 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2282 | update_shares(sd); |
2283 | break; | |
2284 | } | |
2285 | } | |
2286 | } | |
2287 | #endif | |
2288 | ||
04e2f174 | 2289 | smp_wmb(); |
1da177e4 | 2290 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2291 | update_rq_clock(rq); |
1da177e4 LT |
2292 | old_state = p->state; |
2293 | if (!(old_state & state)) | |
2294 | goto out; | |
2295 | ||
dd41f596 | 2296 | if (p->se.on_rq) |
1da177e4 LT |
2297 | goto out_running; |
2298 | ||
2299 | cpu = task_cpu(p); | |
cc367732 | 2300 | orig_cpu = cpu; |
1da177e4 LT |
2301 | this_cpu = smp_processor_id(); |
2302 | ||
2303 | #ifdef CONFIG_SMP | |
2304 | if (unlikely(task_running(rq, p))) | |
2305 | goto out_activate; | |
2306 | ||
5d2f5a61 DA |
2307 | cpu = p->sched_class->select_task_rq(p, sync); |
2308 | if (cpu != orig_cpu) { | |
2309 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2310 | task_rq_unlock(rq, &flags); |
2311 | /* might preempt at this point */ | |
2312 | rq = task_rq_lock(p, &flags); | |
2313 | old_state = p->state; | |
2314 | if (!(old_state & state)) | |
2315 | goto out; | |
dd41f596 | 2316 | if (p->se.on_rq) |
1da177e4 LT |
2317 | goto out_running; |
2318 | ||
2319 | this_cpu = smp_processor_id(); | |
2320 | cpu = task_cpu(p); | |
2321 | } | |
2322 | ||
e7693a36 GH |
2323 | #ifdef CONFIG_SCHEDSTATS |
2324 | schedstat_inc(rq, ttwu_count); | |
2325 | if (cpu == this_cpu) | |
2326 | schedstat_inc(rq, ttwu_local); | |
2327 | else { | |
2328 | struct sched_domain *sd; | |
2329 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2330 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2331 | schedstat_inc(sd, ttwu_wake_remote); |
2332 | break; | |
2333 | } | |
2334 | } | |
2335 | } | |
6d6bc0ad | 2336 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2337 | |
1da177e4 LT |
2338 | out_activate: |
2339 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2340 | schedstat_inc(p, se.nr_wakeups); |
2341 | if (sync) | |
2342 | schedstat_inc(p, se.nr_wakeups_sync); | |
2343 | if (orig_cpu != cpu) | |
2344 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2345 | if (cpu == this_cpu) | |
2346 | schedstat_inc(p, se.nr_wakeups_local); | |
2347 | else | |
2348 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2349 | activate_task(rq, p, 1); |
1da177e4 LT |
2350 | success = 1; |
2351 | ||
2352 | out_running: | |
468a15bb | 2353 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2354 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2355 | |
1da177e4 | 2356 | p->state = TASK_RUNNING; |
9a897c5a SR |
2357 | #ifdef CONFIG_SMP |
2358 | if (p->sched_class->task_wake_up) | |
2359 | p->sched_class->task_wake_up(rq, p); | |
2360 | #endif | |
1da177e4 | 2361 | out: |
2087a1ad GH |
2362 | current->se.last_wakeup = current->se.sum_exec_runtime; |
2363 | ||
1da177e4 LT |
2364 | task_rq_unlock(rq, &flags); |
2365 | ||
2366 | return success; | |
2367 | } | |
2368 | ||
7ad5b3a5 | 2369 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2370 | { |
d9514f6c | 2371 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2372 | } |
1da177e4 LT |
2373 | EXPORT_SYMBOL(wake_up_process); |
2374 | ||
7ad5b3a5 | 2375 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2376 | { |
2377 | return try_to_wake_up(p, state, 0); | |
2378 | } | |
2379 | ||
1da177e4 LT |
2380 | /* |
2381 | * Perform scheduler related setup for a newly forked process p. | |
2382 | * p is forked by current. | |
dd41f596 IM |
2383 | * |
2384 | * __sched_fork() is basic setup used by init_idle() too: | |
2385 | */ | |
2386 | static void __sched_fork(struct task_struct *p) | |
2387 | { | |
dd41f596 IM |
2388 | p->se.exec_start = 0; |
2389 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2390 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2391 | p->se.last_wakeup = 0; |
2392 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2393 | |
2394 | #ifdef CONFIG_SCHEDSTATS | |
2395 | p->se.wait_start = 0; | |
dd41f596 IM |
2396 | p->se.sum_sleep_runtime = 0; |
2397 | p->se.sleep_start = 0; | |
dd41f596 IM |
2398 | p->se.block_start = 0; |
2399 | p->se.sleep_max = 0; | |
2400 | p->se.block_max = 0; | |
2401 | p->se.exec_max = 0; | |
eba1ed4b | 2402 | p->se.slice_max = 0; |
dd41f596 | 2403 | p->se.wait_max = 0; |
6cfb0d5d | 2404 | #endif |
476d139c | 2405 | |
fa717060 | 2406 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2407 | p->se.on_rq = 0; |
4a55bd5e | 2408 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2409 | |
e107be36 AK |
2410 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2411 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2412 | #endif | |
2413 | ||
1da177e4 LT |
2414 | /* |
2415 | * We mark the process as running here, but have not actually | |
2416 | * inserted it onto the runqueue yet. This guarantees that | |
2417 | * nobody will actually run it, and a signal or other external | |
2418 | * event cannot wake it up and insert it on the runqueue either. | |
2419 | */ | |
2420 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2421 | } |
2422 | ||
2423 | /* | |
2424 | * fork()/clone()-time setup: | |
2425 | */ | |
2426 | void sched_fork(struct task_struct *p, int clone_flags) | |
2427 | { | |
2428 | int cpu = get_cpu(); | |
2429 | ||
2430 | __sched_fork(p); | |
2431 | ||
2432 | #ifdef CONFIG_SMP | |
2433 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2434 | #endif | |
02e4bac2 | 2435 | set_task_cpu(p, cpu); |
b29739f9 IM |
2436 | |
2437 | /* | |
2438 | * Make sure we do not leak PI boosting priority to the child: | |
2439 | */ | |
2440 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2441 | if (!rt_prio(p->prio)) |
2442 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2443 | |
52f17b6c | 2444 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2445 | if (likely(sched_info_on())) |
52f17b6c | 2446 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2447 | #endif |
d6077cb8 | 2448 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2449 | p->oncpu = 0; |
2450 | #endif | |
1da177e4 | 2451 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2452 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2453 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2454 | #endif |
476d139c | 2455 | put_cpu(); |
1da177e4 LT |
2456 | } |
2457 | ||
2458 | /* | |
2459 | * wake_up_new_task - wake up a newly created task for the first time. | |
2460 | * | |
2461 | * This function will do some initial scheduler statistics housekeeping | |
2462 | * that must be done for every newly created context, then puts the task | |
2463 | * on the runqueue and wakes it. | |
2464 | */ | |
7ad5b3a5 | 2465 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2466 | { |
2467 | unsigned long flags; | |
dd41f596 | 2468 | struct rq *rq; |
1da177e4 LT |
2469 | |
2470 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2471 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2472 | update_rq_clock(rq); |
1da177e4 LT |
2473 | |
2474 | p->prio = effective_prio(p); | |
2475 | ||
b9dca1e0 | 2476 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2477 | activate_task(rq, p, 0); |
1da177e4 | 2478 | } else { |
1da177e4 | 2479 | /* |
dd41f596 IM |
2480 | * Let the scheduling class do new task startup |
2481 | * management (if any): | |
1da177e4 | 2482 | */ |
ee0827d8 | 2483 | p->sched_class->task_new(rq, p); |
c09595f6 | 2484 | inc_nr_running(rq); |
1da177e4 | 2485 | } |
c71dd42d | 2486 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2487 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2488 | #ifdef CONFIG_SMP |
2489 | if (p->sched_class->task_wake_up) | |
2490 | p->sched_class->task_wake_up(rq, p); | |
2491 | #endif | |
dd41f596 | 2492 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2493 | } |
2494 | ||
e107be36 AK |
2495 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2496 | ||
2497 | /** | |
421cee29 RD |
2498 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2499 | * @notifier: notifier struct to register | |
e107be36 AK |
2500 | */ |
2501 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2502 | { | |
2503 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2504 | } | |
2505 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2506 | ||
2507 | /** | |
2508 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2509 | * @notifier: notifier struct to unregister |
e107be36 AK |
2510 | * |
2511 | * This is safe to call from within a preemption notifier. | |
2512 | */ | |
2513 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2514 | { | |
2515 | hlist_del(¬ifier->link); | |
2516 | } | |
2517 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2518 | ||
2519 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2520 | { | |
2521 | struct preempt_notifier *notifier; | |
2522 | struct hlist_node *node; | |
2523 | ||
2524 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2525 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2526 | } | |
2527 | ||
2528 | static void | |
2529 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2530 | struct task_struct *next) | |
2531 | { | |
2532 | struct preempt_notifier *notifier; | |
2533 | struct hlist_node *node; | |
2534 | ||
2535 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2536 | notifier->ops->sched_out(notifier, next); | |
2537 | } | |
2538 | ||
6d6bc0ad | 2539 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2540 | |
2541 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2542 | { | |
2543 | } | |
2544 | ||
2545 | static void | |
2546 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2547 | struct task_struct *next) | |
2548 | { | |
2549 | } | |
2550 | ||
6d6bc0ad | 2551 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2552 | |
4866cde0 NP |
2553 | /** |
2554 | * prepare_task_switch - prepare to switch tasks | |
2555 | * @rq: the runqueue preparing to switch | |
421cee29 | 2556 | * @prev: the current task that is being switched out |
4866cde0 NP |
2557 | * @next: the task we are going to switch to. |
2558 | * | |
2559 | * This is called with the rq lock held and interrupts off. It must | |
2560 | * be paired with a subsequent finish_task_switch after the context | |
2561 | * switch. | |
2562 | * | |
2563 | * prepare_task_switch sets up locking and calls architecture specific | |
2564 | * hooks. | |
2565 | */ | |
e107be36 AK |
2566 | static inline void |
2567 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2568 | struct task_struct *next) | |
4866cde0 | 2569 | { |
e107be36 | 2570 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2571 | prepare_lock_switch(rq, next); |
2572 | prepare_arch_switch(next); | |
2573 | } | |
2574 | ||
1da177e4 LT |
2575 | /** |
2576 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2577 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2578 | * @prev: the thread we just switched away from. |
2579 | * | |
4866cde0 NP |
2580 | * finish_task_switch must be called after the context switch, paired |
2581 | * with a prepare_task_switch call before the context switch. | |
2582 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2583 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2584 | * |
2585 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2586 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2587 | * with the lock held can cause deadlocks; see schedule() for |
2588 | * details.) | |
2589 | */ | |
a9957449 | 2590 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2591 | __releases(rq->lock) |
2592 | { | |
1da177e4 | 2593 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2594 | long prev_state; |
1da177e4 LT |
2595 | |
2596 | rq->prev_mm = NULL; | |
2597 | ||
2598 | /* | |
2599 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2600 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2601 | * schedule one last time. The schedule call will never return, and |
2602 | * the scheduled task must drop that reference. | |
c394cc9f | 2603 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2604 | * still held, otherwise prev could be scheduled on another cpu, die |
2605 | * there before we look at prev->state, and then the reference would | |
2606 | * be dropped twice. | |
2607 | * Manfred Spraul <manfred@colorfullife.com> | |
2608 | */ | |
55a101f8 | 2609 | prev_state = prev->state; |
4866cde0 NP |
2610 | finish_arch_switch(prev); |
2611 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2612 | #ifdef CONFIG_SMP |
2613 | if (current->sched_class->post_schedule) | |
2614 | current->sched_class->post_schedule(rq); | |
2615 | #endif | |
e8fa1362 | 2616 | |
e107be36 | 2617 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2618 | if (mm) |
2619 | mmdrop(mm); | |
c394cc9f | 2620 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2621 | /* |
2622 | * Remove function-return probe instances associated with this | |
2623 | * task and put them back on the free list. | |
9761eea8 | 2624 | */ |
c6fd91f0 | 2625 | kprobe_flush_task(prev); |
1da177e4 | 2626 | put_task_struct(prev); |
c6fd91f0 | 2627 | } |
1da177e4 LT |
2628 | } |
2629 | ||
2630 | /** | |
2631 | * schedule_tail - first thing a freshly forked thread must call. | |
2632 | * @prev: the thread we just switched away from. | |
2633 | */ | |
36c8b586 | 2634 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2635 | __releases(rq->lock) |
2636 | { | |
70b97a7f IM |
2637 | struct rq *rq = this_rq(); |
2638 | ||
4866cde0 NP |
2639 | finish_task_switch(rq, prev); |
2640 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2641 | /* In this case, finish_task_switch does not reenable preemption */ | |
2642 | preempt_enable(); | |
2643 | #endif | |
1da177e4 | 2644 | if (current->set_child_tid) |
b488893a | 2645 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2646 | } |
2647 | ||
2648 | /* | |
2649 | * context_switch - switch to the new MM and the new | |
2650 | * thread's register state. | |
2651 | */ | |
dd41f596 | 2652 | static inline void |
70b97a7f | 2653 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2654 | struct task_struct *next) |
1da177e4 | 2655 | { |
dd41f596 | 2656 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2657 | |
e107be36 | 2658 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2659 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2660 | mm = next->mm; |
2661 | oldmm = prev->active_mm; | |
9226d125 ZA |
2662 | /* |
2663 | * For paravirt, this is coupled with an exit in switch_to to | |
2664 | * combine the page table reload and the switch backend into | |
2665 | * one hypercall. | |
2666 | */ | |
2667 | arch_enter_lazy_cpu_mode(); | |
2668 | ||
dd41f596 | 2669 | if (unlikely(!mm)) { |
1da177e4 LT |
2670 | next->active_mm = oldmm; |
2671 | atomic_inc(&oldmm->mm_count); | |
2672 | enter_lazy_tlb(oldmm, next); | |
2673 | } else | |
2674 | switch_mm(oldmm, mm, next); | |
2675 | ||
dd41f596 | 2676 | if (unlikely(!prev->mm)) { |
1da177e4 | 2677 | prev->active_mm = NULL; |
1da177e4 LT |
2678 | rq->prev_mm = oldmm; |
2679 | } | |
3a5f5e48 IM |
2680 | /* |
2681 | * Since the runqueue lock will be released by the next | |
2682 | * task (which is an invalid locking op but in the case | |
2683 | * of the scheduler it's an obvious special-case), so we | |
2684 | * do an early lockdep release here: | |
2685 | */ | |
2686 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2687 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2688 | #endif |
1da177e4 LT |
2689 | |
2690 | /* Here we just switch the register state and the stack. */ | |
2691 | switch_to(prev, next, prev); | |
2692 | ||
dd41f596 IM |
2693 | barrier(); |
2694 | /* | |
2695 | * this_rq must be evaluated again because prev may have moved | |
2696 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2697 | * frame will be invalid. | |
2698 | */ | |
2699 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2700 | } |
2701 | ||
2702 | /* | |
2703 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2704 | * | |
2705 | * externally visible scheduler statistics: current number of runnable | |
2706 | * threads, current number of uninterruptible-sleeping threads, total | |
2707 | * number of context switches performed since bootup. | |
2708 | */ | |
2709 | unsigned long nr_running(void) | |
2710 | { | |
2711 | unsigned long i, sum = 0; | |
2712 | ||
2713 | for_each_online_cpu(i) | |
2714 | sum += cpu_rq(i)->nr_running; | |
2715 | ||
2716 | return sum; | |
2717 | } | |
2718 | ||
2719 | unsigned long nr_uninterruptible(void) | |
2720 | { | |
2721 | unsigned long i, sum = 0; | |
2722 | ||
0a945022 | 2723 | for_each_possible_cpu(i) |
1da177e4 LT |
2724 | sum += cpu_rq(i)->nr_uninterruptible; |
2725 | ||
2726 | /* | |
2727 | * Since we read the counters lockless, it might be slightly | |
2728 | * inaccurate. Do not allow it to go below zero though: | |
2729 | */ | |
2730 | if (unlikely((long)sum < 0)) | |
2731 | sum = 0; | |
2732 | ||
2733 | return sum; | |
2734 | } | |
2735 | ||
2736 | unsigned long long nr_context_switches(void) | |
2737 | { | |
cc94abfc SR |
2738 | int i; |
2739 | unsigned long long sum = 0; | |
1da177e4 | 2740 | |
0a945022 | 2741 | for_each_possible_cpu(i) |
1da177e4 LT |
2742 | sum += cpu_rq(i)->nr_switches; |
2743 | ||
2744 | return sum; | |
2745 | } | |
2746 | ||
2747 | unsigned long nr_iowait(void) | |
2748 | { | |
2749 | unsigned long i, sum = 0; | |
2750 | ||
0a945022 | 2751 | for_each_possible_cpu(i) |
1da177e4 LT |
2752 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2753 | ||
2754 | return sum; | |
2755 | } | |
2756 | ||
db1b1fef JS |
2757 | unsigned long nr_active(void) |
2758 | { | |
2759 | unsigned long i, running = 0, uninterruptible = 0; | |
2760 | ||
2761 | for_each_online_cpu(i) { | |
2762 | running += cpu_rq(i)->nr_running; | |
2763 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2764 | } | |
2765 | ||
2766 | if (unlikely((long)uninterruptible < 0)) | |
2767 | uninterruptible = 0; | |
2768 | ||
2769 | return running + uninterruptible; | |
2770 | } | |
2771 | ||
48f24c4d | 2772 | /* |
dd41f596 IM |
2773 | * Update rq->cpu_load[] statistics. This function is usually called every |
2774 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2775 | */ |
dd41f596 | 2776 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2777 | { |
495eca49 | 2778 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2779 | int i, scale; |
2780 | ||
2781 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2782 | |
2783 | /* Update our load: */ | |
2784 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2785 | unsigned long old_load, new_load; | |
2786 | ||
2787 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2788 | ||
2789 | old_load = this_rq->cpu_load[i]; | |
2790 | new_load = this_load; | |
a25707f3 IM |
2791 | /* |
2792 | * Round up the averaging division if load is increasing. This | |
2793 | * prevents us from getting stuck on 9 if the load is 10, for | |
2794 | * example. | |
2795 | */ | |
2796 | if (new_load > old_load) | |
2797 | new_load += scale-1; | |
dd41f596 IM |
2798 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2799 | } | |
48f24c4d IM |
2800 | } |
2801 | ||
dd41f596 IM |
2802 | #ifdef CONFIG_SMP |
2803 | ||
1da177e4 LT |
2804 | /* |
2805 | * double_rq_lock - safely lock two runqueues | |
2806 | * | |
2807 | * Note this does not disable interrupts like task_rq_lock, | |
2808 | * you need to do so manually before calling. | |
2809 | */ | |
70b97a7f | 2810 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2811 | __acquires(rq1->lock) |
2812 | __acquires(rq2->lock) | |
2813 | { | |
054b9108 | 2814 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2815 | if (rq1 == rq2) { |
2816 | spin_lock(&rq1->lock); | |
2817 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2818 | } else { | |
c96d145e | 2819 | if (rq1 < rq2) { |
1da177e4 | 2820 | spin_lock(&rq1->lock); |
5e710e37 | 2821 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2822 | } else { |
2823 | spin_lock(&rq2->lock); | |
5e710e37 | 2824 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2825 | } |
2826 | } | |
6e82a3be IM |
2827 | update_rq_clock(rq1); |
2828 | update_rq_clock(rq2); | |
1da177e4 LT |
2829 | } |
2830 | ||
2831 | /* | |
2832 | * double_rq_unlock - safely unlock two runqueues | |
2833 | * | |
2834 | * Note this does not restore interrupts like task_rq_unlock, | |
2835 | * you need to do so manually after calling. | |
2836 | */ | |
70b97a7f | 2837 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2838 | __releases(rq1->lock) |
2839 | __releases(rq2->lock) | |
2840 | { | |
2841 | spin_unlock(&rq1->lock); | |
2842 | if (rq1 != rq2) | |
2843 | spin_unlock(&rq2->lock); | |
2844 | else | |
2845 | __release(rq2->lock); | |
2846 | } | |
2847 | ||
1da177e4 LT |
2848 | /* |
2849 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2850 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2851 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2852 | * the cpu_allowed mask is restored. |
2853 | */ | |
36c8b586 | 2854 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2855 | { |
70b97a7f | 2856 | struct migration_req req; |
1da177e4 | 2857 | unsigned long flags; |
70b97a7f | 2858 | struct rq *rq; |
1da177e4 LT |
2859 | |
2860 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 2861 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 2862 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
2863 | goto out; |
2864 | ||
2865 | /* force the process onto the specified CPU */ | |
2866 | if (migrate_task(p, dest_cpu, &req)) { | |
2867 | /* Need to wait for migration thread (might exit: take ref). */ | |
2868 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2869 | |
1da177e4 LT |
2870 | get_task_struct(mt); |
2871 | task_rq_unlock(rq, &flags); | |
2872 | wake_up_process(mt); | |
2873 | put_task_struct(mt); | |
2874 | wait_for_completion(&req.done); | |
36c8b586 | 2875 | |
1da177e4 LT |
2876 | return; |
2877 | } | |
2878 | out: | |
2879 | task_rq_unlock(rq, &flags); | |
2880 | } | |
2881 | ||
2882 | /* | |
476d139c NP |
2883 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2884 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2885 | */ |
2886 | void sched_exec(void) | |
2887 | { | |
1da177e4 | 2888 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2889 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2890 | put_cpu(); |
476d139c NP |
2891 | if (new_cpu != this_cpu) |
2892 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2893 | } |
2894 | ||
2895 | /* | |
2896 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2897 | * Both runqueues must be locked. | |
2898 | */ | |
dd41f596 IM |
2899 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2900 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2901 | { |
2e1cb74a | 2902 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2903 | set_task_cpu(p, this_cpu); |
dd41f596 | 2904 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2905 | /* |
2906 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2907 | * to be always true for them. | |
2908 | */ | |
15afe09b | 2909 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
2910 | } |
2911 | ||
2912 | /* | |
2913 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2914 | */ | |
858119e1 | 2915 | static |
70b97a7f | 2916 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2917 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2918 | int *all_pinned) |
1da177e4 LT |
2919 | { |
2920 | /* | |
2921 | * We do not migrate tasks that are: | |
2922 | * 1) running (obviously), or | |
2923 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2924 | * 3) are cache-hot on their current CPU. | |
2925 | */ | |
96f874e2 | 2926 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 2927 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 2928 | return 0; |
cc367732 | 2929 | } |
81026794 NP |
2930 | *all_pinned = 0; |
2931 | ||
cc367732 IM |
2932 | if (task_running(rq, p)) { |
2933 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2934 | return 0; |
cc367732 | 2935 | } |
1da177e4 | 2936 | |
da84d961 IM |
2937 | /* |
2938 | * Aggressive migration if: | |
2939 | * 1) task is cache cold, or | |
2940 | * 2) too many balance attempts have failed. | |
2941 | */ | |
2942 | ||
6bc1665b IM |
2943 | if (!task_hot(p, rq->clock, sd) || |
2944 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2945 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2946 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2947 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2948 | schedstat_inc(p, se.nr_forced_migrations); |
2949 | } | |
da84d961 IM |
2950 | #endif |
2951 | return 1; | |
2952 | } | |
2953 | ||
cc367732 IM |
2954 | if (task_hot(p, rq->clock, sd)) { |
2955 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2956 | return 0; |
cc367732 | 2957 | } |
1da177e4 LT |
2958 | return 1; |
2959 | } | |
2960 | ||
e1d1484f PW |
2961 | static unsigned long |
2962 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2963 | unsigned long max_load_move, struct sched_domain *sd, | |
2964 | enum cpu_idle_type idle, int *all_pinned, | |
2965 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2966 | { |
051c6764 | 2967 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
2968 | struct task_struct *p; |
2969 | long rem_load_move = max_load_move; | |
1da177e4 | 2970 | |
e1d1484f | 2971 | if (max_load_move == 0) |
1da177e4 LT |
2972 | goto out; |
2973 | ||
81026794 NP |
2974 | pinned = 1; |
2975 | ||
1da177e4 | 2976 | /* |
dd41f596 | 2977 | * Start the load-balancing iterator: |
1da177e4 | 2978 | */ |
dd41f596 IM |
2979 | p = iterator->start(iterator->arg); |
2980 | next: | |
b82d9fdd | 2981 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2982 | goto out; |
051c6764 PZ |
2983 | |
2984 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 2985 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2986 | p = iterator->next(iterator->arg); |
2987 | goto next; | |
1da177e4 LT |
2988 | } |
2989 | ||
dd41f596 | 2990 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2991 | pulled++; |
dd41f596 | 2992 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2993 | |
2dd73a4f | 2994 | /* |
b82d9fdd | 2995 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2996 | */ |
e1d1484f | 2997 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2998 | if (p->prio < *this_best_prio) |
2999 | *this_best_prio = p->prio; | |
dd41f596 IM |
3000 | p = iterator->next(iterator->arg); |
3001 | goto next; | |
1da177e4 LT |
3002 | } |
3003 | out: | |
3004 | /* | |
e1d1484f | 3005 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3006 | * so we can safely collect pull_task() stats here rather than |
3007 | * inside pull_task(). | |
3008 | */ | |
3009 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3010 | |
3011 | if (all_pinned) | |
3012 | *all_pinned = pinned; | |
e1d1484f PW |
3013 | |
3014 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3015 | } |
3016 | ||
dd41f596 | 3017 | /* |
43010659 PW |
3018 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3019 | * this_rq, as part of a balancing operation within domain "sd". | |
3020 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3021 | * |
3022 | * Called with both runqueues locked. | |
3023 | */ | |
3024 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3025 | unsigned long max_load_move, |
dd41f596 IM |
3026 | struct sched_domain *sd, enum cpu_idle_type idle, |
3027 | int *all_pinned) | |
3028 | { | |
5522d5d5 | 3029 | const struct sched_class *class = sched_class_highest; |
43010659 | 3030 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3031 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3032 | |
3033 | do { | |
43010659 PW |
3034 | total_load_moved += |
3035 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3036 | max_load_move - total_load_moved, |
a4ac01c3 | 3037 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3038 | class = class->next; |
c4acb2c0 GH |
3039 | |
3040 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) | |
3041 | break; | |
3042 | ||
43010659 | 3043 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3044 | |
43010659 PW |
3045 | return total_load_moved > 0; |
3046 | } | |
3047 | ||
e1d1484f PW |
3048 | static int |
3049 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3050 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3051 | struct rq_iterator *iterator) | |
3052 | { | |
3053 | struct task_struct *p = iterator->start(iterator->arg); | |
3054 | int pinned = 0; | |
3055 | ||
3056 | while (p) { | |
3057 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3058 | pull_task(busiest, p, this_rq, this_cpu); | |
3059 | /* | |
3060 | * Right now, this is only the second place pull_task() | |
3061 | * is called, so we can safely collect pull_task() | |
3062 | * stats here rather than inside pull_task(). | |
3063 | */ | |
3064 | schedstat_inc(sd, lb_gained[idle]); | |
3065 | ||
3066 | return 1; | |
3067 | } | |
3068 | p = iterator->next(iterator->arg); | |
3069 | } | |
3070 | ||
3071 | return 0; | |
3072 | } | |
3073 | ||
43010659 PW |
3074 | /* |
3075 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3076 | * part of active balancing operations within "domain". | |
3077 | * Returns 1 if successful and 0 otherwise. | |
3078 | * | |
3079 | * Called with both runqueues locked. | |
3080 | */ | |
3081 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3082 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3083 | { | |
5522d5d5 | 3084 | const struct sched_class *class; |
43010659 PW |
3085 | |
3086 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3087 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3088 | return 1; |
3089 | ||
3090 | return 0; | |
dd41f596 IM |
3091 | } |
3092 | ||
1da177e4 LT |
3093 | /* |
3094 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
3095 | * domain. It calculates and returns the amount of weighted load which |
3096 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
3097 | */ |
3098 | static struct sched_group * | |
3099 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 3100 | unsigned long *imbalance, enum cpu_idle_type idle, |
96f874e2 | 3101 | int *sd_idle, const struct cpumask *cpus, int *balance) |
1da177e4 LT |
3102 | { |
3103 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
3104 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 3105 | unsigned long max_pull; |
2dd73a4f PW |
3106 | unsigned long busiest_load_per_task, busiest_nr_running; |
3107 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 3108 | int load_idx, group_imb = 0; |
5c45bf27 SS |
3109 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3110 | int power_savings_balance = 1; | |
3111 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
3112 | unsigned long min_nr_running = ULONG_MAX; | |
3113 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
3114 | #endif | |
1da177e4 LT |
3115 | |
3116 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
3117 | busiest_load_per_task = busiest_nr_running = 0; |
3118 | this_load_per_task = this_nr_running = 0; | |
408ed066 | 3119 | |
d15bcfdb | 3120 | if (idle == CPU_NOT_IDLE) |
7897986b | 3121 | load_idx = sd->busy_idx; |
d15bcfdb | 3122 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
3123 | load_idx = sd->newidle_idx; |
3124 | else | |
3125 | load_idx = sd->idle_idx; | |
1da177e4 LT |
3126 | |
3127 | do { | |
908a7c1b | 3128 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
3129 | int local_group; |
3130 | int i; | |
908a7c1b | 3131 | int __group_imb = 0; |
783609c6 | 3132 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 3133 | unsigned long sum_nr_running, sum_weighted_load; |
408ed066 PZ |
3134 | unsigned long sum_avg_load_per_task; |
3135 | unsigned long avg_load_per_task; | |
1da177e4 | 3136 | |
758b2cdc RR |
3137 | local_group = cpumask_test_cpu(this_cpu, |
3138 | sched_group_cpus(group)); | |
1da177e4 | 3139 | |
783609c6 | 3140 | if (local_group) |
758b2cdc | 3141 | balance_cpu = cpumask_first(sched_group_cpus(group)); |
783609c6 | 3142 | |
1da177e4 | 3143 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 3144 | sum_weighted_load = sum_nr_running = avg_load = 0; |
408ed066 PZ |
3145 | sum_avg_load_per_task = avg_load_per_task = 0; |
3146 | ||
908a7c1b KC |
3147 | max_cpu_load = 0; |
3148 | min_cpu_load = ~0UL; | |
1da177e4 | 3149 | |
758b2cdc RR |
3150 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3151 | struct rq *rq = cpu_rq(i); | |
2dd73a4f | 3152 | |
9439aab8 | 3153 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3154 | *sd_idle = 0; |
3155 | ||
1da177e4 | 3156 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3157 | if (local_group) { |
3158 | if (idle_cpu(i) && !first_idle_cpu) { | |
3159 | first_idle_cpu = 1; | |
3160 | balance_cpu = i; | |
3161 | } | |
3162 | ||
a2000572 | 3163 | load = target_load(i, load_idx); |
908a7c1b | 3164 | } else { |
a2000572 | 3165 | load = source_load(i, load_idx); |
908a7c1b KC |
3166 | if (load > max_cpu_load) |
3167 | max_cpu_load = load; | |
3168 | if (min_cpu_load > load) | |
3169 | min_cpu_load = load; | |
3170 | } | |
1da177e4 LT |
3171 | |
3172 | avg_load += load; | |
2dd73a4f | 3173 | sum_nr_running += rq->nr_running; |
dd41f596 | 3174 | sum_weighted_load += weighted_cpuload(i); |
408ed066 PZ |
3175 | |
3176 | sum_avg_load_per_task += cpu_avg_load_per_task(i); | |
1da177e4 LT |
3177 | } |
3178 | ||
783609c6 SS |
3179 | /* |
3180 | * First idle cpu or the first cpu(busiest) in this sched group | |
3181 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3182 | * domains. In the newly idle case, we will allow all the cpu's |
3183 | * to do the newly idle load balance. | |
783609c6 | 3184 | */ |
9439aab8 SS |
3185 | if (idle != CPU_NEWLY_IDLE && local_group && |
3186 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3187 | *balance = 0; |
3188 | goto ret; | |
3189 | } | |
3190 | ||
1da177e4 | 3191 | total_load += avg_load; |
5517d86b | 3192 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3193 | |
3194 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3195 | avg_load = sg_div_cpu_power(group, |
3196 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3197 | |
408ed066 PZ |
3198 | |
3199 | /* | |
3200 | * Consider the group unbalanced when the imbalance is larger | |
3201 | * than the average weight of two tasks. | |
3202 | * | |
3203 | * APZ: with cgroup the avg task weight can vary wildly and | |
3204 | * might not be a suitable number - should we keep a | |
3205 | * normalized nr_running number somewhere that negates | |
3206 | * the hierarchy? | |
3207 | */ | |
3208 | avg_load_per_task = sg_div_cpu_power(group, | |
3209 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3210 | ||
3211 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
908a7c1b KC |
3212 | __group_imb = 1; |
3213 | ||
5517d86b | 3214 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3215 | |
1da177e4 LT |
3216 | if (local_group) { |
3217 | this_load = avg_load; | |
3218 | this = group; | |
2dd73a4f PW |
3219 | this_nr_running = sum_nr_running; |
3220 | this_load_per_task = sum_weighted_load; | |
3221 | } else if (avg_load > max_load && | |
908a7c1b | 3222 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3223 | max_load = avg_load; |
3224 | busiest = group; | |
2dd73a4f PW |
3225 | busiest_nr_running = sum_nr_running; |
3226 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3227 | group_imb = __group_imb; |
1da177e4 | 3228 | } |
5c45bf27 SS |
3229 | |
3230 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3231 | /* | |
3232 | * Busy processors will not participate in power savings | |
3233 | * balance. | |
3234 | */ | |
dd41f596 IM |
3235 | if (idle == CPU_NOT_IDLE || |
3236 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3237 | goto group_next; | |
5c45bf27 SS |
3238 | |
3239 | /* | |
3240 | * If the local group is idle or completely loaded | |
3241 | * no need to do power savings balance at this domain | |
3242 | */ | |
3243 | if (local_group && (this_nr_running >= group_capacity || | |
3244 | !this_nr_running)) | |
3245 | power_savings_balance = 0; | |
3246 | ||
dd41f596 | 3247 | /* |
5c45bf27 SS |
3248 | * If a group is already running at full capacity or idle, |
3249 | * don't include that group in power savings calculations | |
dd41f596 IM |
3250 | */ |
3251 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3252 | || !sum_nr_running) |
dd41f596 | 3253 | goto group_next; |
5c45bf27 | 3254 | |
dd41f596 | 3255 | /* |
5c45bf27 | 3256 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3257 | * This is the group from where we need to pick up the load |
3258 | * for saving power | |
3259 | */ | |
3260 | if ((sum_nr_running < min_nr_running) || | |
3261 | (sum_nr_running == min_nr_running && | |
d5679bd1 | 3262 | cpumask_first(sched_group_cpus(group)) > |
758b2cdc | 3263 | cpumask_first(sched_group_cpus(group_min)))) { |
dd41f596 IM |
3264 | group_min = group; |
3265 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3266 | min_load_per_task = sum_weighted_load / |
3267 | sum_nr_running; | |
dd41f596 | 3268 | } |
5c45bf27 | 3269 | |
dd41f596 | 3270 | /* |
5c45bf27 | 3271 | * Calculate the group which is almost near its |
dd41f596 IM |
3272 | * capacity but still has some space to pick up some load |
3273 | * from other group and save more power | |
3274 | */ | |
3275 | if (sum_nr_running <= group_capacity - 1) { | |
3276 | if (sum_nr_running > leader_nr_running || | |
3277 | (sum_nr_running == leader_nr_running && | |
d5679bd1 | 3278 | cpumask_first(sched_group_cpus(group)) < |
758b2cdc | 3279 | cpumask_first(sched_group_cpus(group_leader)))) { |
dd41f596 IM |
3280 | group_leader = group; |
3281 | leader_nr_running = sum_nr_running; | |
3282 | } | |
48f24c4d | 3283 | } |
5c45bf27 SS |
3284 | group_next: |
3285 | #endif | |
1da177e4 LT |
3286 | group = group->next; |
3287 | } while (group != sd->groups); | |
3288 | ||
2dd73a4f | 3289 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3290 | goto out_balanced; |
3291 | ||
3292 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3293 | ||
3294 | if (this_load >= avg_load || | |
3295 | 100*max_load <= sd->imbalance_pct*this_load) | |
3296 | goto out_balanced; | |
3297 | ||
2dd73a4f | 3298 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3299 | if (group_imb) |
3300 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3301 | ||
1da177e4 LT |
3302 | /* |
3303 | * We're trying to get all the cpus to the average_load, so we don't | |
3304 | * want to push ourselves above the average load, nor do we wish to | |
3305 | * reduce the max loaded cpu below the average load, as either of these | |
3306 | * actions would just result in more rebalancing later, and ping-pong | |
3307 | * tasks around. Thus we look for the minimum possible imbalance. | |
3308 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3309 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3310 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3311 | * appear as very large values with unsigned longs. |
3312 | */ | |
2dd73a4f PW |
3313 | if (max_load <= busiest_load_per_task) |
3314 | goto out_balanced; | |
3315 | ||
3316 | /* | |
3317 | * In the presence of smp nice balancing, certain scenarios can have | |
3318 | * max load less than avg load(as we skip the groups at or below | |
3319 | * its cpu_power, while calculating max_load..) | |
3320 | */ | |
3321 | if (max_load < avg_load) { | |
3322 | *imbalance = 0; | |
3323 | goto small_imbalance; | |
3324 | } | |
0c117f1b SS |
3325 | |
3326 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3327 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3328 | |
1da177e4 | 3329 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3330 | *imbalance = min(max_pull * busiest->__cpu_power, |
3331 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3332 | / SCHED_LOAD_SCALE; |
3333 | ||
2dd73a4f PW |
3334 | /* |
3335 | * if *imbalance is less than the average load per runnable task | |
3336 | * there is no gaurantee that any tasks will be moved so we'll have | |
3337 | * a think about bumping its value to force at least one task to be | |
3338 | * moved | |
3339 | */ | |
7fd0d2dd | 3340 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3341 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3342 | unsigned int imbn; |
3343 | ||
3344 | small_imbalance: | |
3345 | pwr_move = pwr_now = 0; | |
3346 | imbn = 2; | |
3347 | if (this_nr_running) { | |
3348 | this_load_per_task /= this_nr_running; | |
3349 | if (busiest_load_per_task > this_load_per_task) | |
3350 | imbn = 1; | |
3351 | } else | |
408ed066 | 3352 | this_load_per_task = cpu_avg_load_per_task(this_cpu); |
1da177e4 | 3353 | |
01c8c57d | 3354 | if (max_load - this_load + busiest_load_per_task >= |
dd41f596 | 3355 | busiest_load_per_task * imbn) { |
2dd73a4f | 3356 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3357 | return busiest; |
3358 | } | |
3359 | ||
3360 | /* | |
3361 | * OK, we don't have enough imbalance to justify moving tasks, | |
3362 | * however we may be able to increase total CPU power used by | |
3363 | * moving them. | |
3364 | */ | |
3365 | ||
5517d86b ED |
3366 | pwr_now += busiest->__cpu_power * |
3367 | min(busiest_load_per_task, max_load); | |
3368 | pwr_now += this->__cpu_power * | |
3369 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3370 | pwr_now /= SCHED_LOAD_SCALE; |
3371 | ||
3372 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3373 | tmp = sg_div_cpu_power(busiest, |
3374 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3375 | if (max_load > tmp) |
5517d86b | 3376 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3377 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3378 | |
3379 | /* Amount of load we'd add */ | |
5517d86b | 3380 | if (max_load * busiest->__cpu_power < |
33859f7f | 3381 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3382 | tmp = sg_div_cpu_power(this, |
3383 | max_load * busiest->__cpu_power); | |
1da177e4 | 3384 | else |
5517d86b ED |
3385 | tmp = sg_div_cpu_power(this, |
3386 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3387 | pwr_move += this->__cpu_power * | |
3388 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3389 | pwr_move /= SCHED_LOAD_SCALE; |
3390 | ||
3391 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3392 | if (pwr_move > pwr_now) |
3393 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3394 | } |
3395 | ||
1da177e4 LT |
3396 | return busiest; |
3397 | ||
3398 | out_balanced: | |
5c45bf27 | 3399 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3400 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3401 | goto ret; |
1da177e4 | 3402 | |
5c45bf27 SS |
3403 | if (this == group_leader && group_leader != group_min) { |
3404 | *imbalance = min_load_per_task; | |
7a09b1a2 VS |
3405 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3406 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
9924da43 | 3407 | cpumask_first(sched_group_cpus(group_leader)); |
7a09b1a2 | 3408 | } |
5c45bf27 SS |
3409 | return group_min; |
3410 | } | |
5c45bf27 | 3411 | #endif |
783609c6 | 3412 | ret: |
1da177e4 LT |
3413 | *imbalance = 0; |
3414 | return NULL; | |
3415 | } | |
3416 | ||
3417 | /* | |
3418 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3419 | */ | |
70b97a7f | 3420 | static struct rq * |
d15bcfdb | 3421 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3422 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3423 | { |
70b97a7f | 3424 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3425 | unsigned long max_load = 0; |
1da177e4 LT |
3426 | int i; |
3427 | ||
758b2cdc | 3428 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3429 | unsigned long wl; |
0a2966b4 | 3430 | |
96f874e2 | 3431 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3432 | continue; |
3433 | ||
48f24c4d | 3434 | rq = cpu_rq(i); |
dd41f596 | 3435 | wl = weighted_cpuload(i); |
2dd73a4f | 3436 | |
dd41f596 | 3437 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3438 | continue; |
1da177e4 | 3439 | |
dd41f596 IM |
3440 | if (wl > max_load) { |
3441 | max_load = wl; | |
48f24c4d | 3442 | busiest = rq; |
1da177e4 LT |
3443 | } |
3444 | } | |
3445 | ||
3446 | return busiest; | |
3447 | } | |
3448 | ||
77391d71 NP |
3449 | /* |
3450 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3451 | * so long as it is large enough. | |
3452 | */ | |
3453 | #define MAX_PINNED_INTERVAL 512 | |
3454 | ||
1da177e4 LT |
3455 | /* |
3456 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3457 | * tasks if there is an imbalance. | |
1da177e4 | 3458 | */ |
70b97a7f | 3459 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3460 | struct sched_domain *sd, enum cpu_idle_type idle, |
96f874e2 | 3461 | int *balance, struct cpumask *cpus) |
1da177e4 | 3462 | { |
43010659 | 3463 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3464 | struct sched_group *group; |
1da177e4 | 3465 | unsigned long imbalance; |
70b97a7f | 3466 | struct rq *busiest; |
fe2eea3f | 3467 | unsigned long flags; |
5969fe06 | 3468 | |
96f874e2 | 3469 | cpumask_setall(cpus); |
7c16ec58 | 3470 | |
89c4710e SS |
3471 | /* |
3472 | * When power savings policy is enabled for the parent domain, idle | |
3473 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3474 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3475 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3476 | */ |
d15bcfdb | 3477 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3478 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3479 | sd_idle = 1; |
1da177e4 | 3480 | |
2d72376b | 3481 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3482 | |
0a2966b4 | 3483 | redo: |
c8cba857 | 3484 | update_shares(sd); |
0a2966b4 | 3485 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3486 | cpus, balance); |
783609c6 | 3487 | |
06066714 | 3488 | if (*balance == 0) |
783609c6 | 3489 | goto out_balanced; |
783609c6 | 3490 | |
1da177e4 LT |
3491 | if (!group) { |
3492 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3493 | goto out_balanced; | |
3494 | } | |
3495 | ||
7c16ec58 | 3496 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3497 | if (!busiest) { |
3498 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3499 | goto out_balanced; | |
3500 | } | |
3501 | ||
db935dbd | 3502 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3503 | |
3504 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3505 | ||
43010659 | 3506 | ld_moved = 0; |
1da177e4 LT |
3507 | if (busiest->nr_running > 1) { |
3508 | /* | |
3509 | * Attempt to move tasks. If find_busiest_group has found | |
3510 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3511 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3512 | * correctly treated as an imbalance. |
3513 | */ | |
fe2eea3f | 3514 | local_irq_save(flags); |
e17224bf | 3515 | double_rq_lock(this_rq, busiest); |
43010659 | 3516 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3517 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3518 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3519 | local_irq_restore(flags); |
81026794 | 3520 | |
46cb4b7c SS |
3521 | /* |
3522 | * some other cpu did the load balance for us. | |
3523 | */ | |
43010659 | 3524 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3525 | resched_cpu(this_cpu); |
3526 | ||
81026794 | 3527 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3528 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3529 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3530 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 3531 | goto redo; |
81026794 | 3532 | goto out_balanced; |
0a2966b4 | 3533 | } |
1da177e4 | 3534 | } |
81026794 | 3535 | |
43010659 | 3536 | if (!ld_moved) { |
1da177e4 LT |
3537 | schedstat_inc(sd, lb_failed[idle]); |
3538 | sd->nr_balance_failed++; | |
3539 | ||
3540 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3541 | |
fe2eea3f | 3542 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3543 | |
3544 | /* don't kick the migration_thread, if the curr | |
3545 | * task on busiest cpu can't be moved to this_cpu | |
3546 | */ | |
96f874e2 RR |
3547 | if (!cpumask_test_cpu(this_cpu, |
3548 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3549 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3550 | all_pinned = 1; |
3551 | goto out_one_pinned; | |
3552 | } | |
3553 | ||
1da177e4 LT |
3554 | if (!busiest->active_balance) { |
3555 | busiest->active_balance = 1; | |
3556 | busiest->push_cpu = this_cpu; | |
81026794 | 3557 | active_balance = 1; |
1da177e4 | 3558 | } |
fe2eea3f | 3559 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3560 | if (active_balance) |
1da177e4 LT |
3561 | wake_up_process(busiest->migration_thread); |
3562 | ||
3563 | /* | |
3564 | * We've kicked active balancing, reset the failure | |
3565 | * counter. | |
3566 | */ | |
39507451 | 3567 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3568 | } |
81026794 | 3569 | } else |
1da177e4 LT |
3570 | sd->nr_balance_failed = 0; |
3571 | ||
81026794 | 3572 | if (likely(!active_balance)) { |
1da177e4 LT |
3573 | /* We were unbalanced, so reset the balancing interval */ |
3574 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3575 | } else { |
3576 | /* | |
3577 | * If we've begun active balancing, start to back off. This | |
3578 | * case may not be covered by the all_pinned logic if there | |
3579 | * is only 1 task on the busy runqueue (because we don't call | |
3580 | * move_tasks). | |
3581 | */ | |
3582 | if (sd->balance_interval < sd->max_interval) | |
3583 | sd->balance_interval *= 2; | |
1da177e4 LT |
3584 | } |
3585 | ||
43010659 | 3586 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3587 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3588 | ld_moved = -1; |
3589 | ||
3590 | goto out; | |
1da177e4 LT |
3591 | |
3592 | out_balanced: | |
1da177e4 LT |
3593 | schedstat_inc(sd, lb_balanced[idle]); |
3594 | ||
16cfb1c0 | 3595 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3596 | |
3597 | out_one_pinned: | |
1da177e4 | 3598 | /* tune up the balancing interval */ |
77391d71 NP |
3599 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3600 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3601 | sd->balance_interval *= 2; |
3602 | ||
48f24c4d | 3603 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3604 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
3605 | ld_moved = -1; |
3606 | else | |
3607 | ld_moved = 0; | |
3608 | out: | |
c8cba857 PZ |
3609 | if (ld_moved) |
3610 | update_shares(sd); | |
c09595f6 | 3611 | return ld_moved; |
1da177e4 LT |
3612 | } |
3613 | ||
3614 | /* | |
3615 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3616 | * tasks if there is an imbalance. | |
3617 | * | |
d15bcfdb | 3618 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3619 | * this_rq is locked. |
3620 | */ | |
48f24c4d | 3621 | static int |
7c16ec58 | 3622 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
96f874e2 | 3623 | struct cpumask *cpus) |
1da177e4 LT |
3624 | { |
3625 | struct sched_group *group; | |
70b97a7f | 3626 | struct rq *busiest = NULL; |
1da177e4 | 3627 | unsigned long imbalance; |
43010659 | 3628 | int ld_moved = 0; |
5969fe06 | 3629 | int sd_idle = 0; |
969bb4e4 | 3630 | int all_pinned = 0; |
7c16ec58 | 3631 | |
96f874e2 | 3632 | cpumask_setall(cpus); |
5969fe06 | 3633 | |
89c4710e SS |
3634 | /* |
3635 | * When power savings policy is enabled for the parent domain, idle | |
3636 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3637 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3638 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3639 | */ |
3640 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3641 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3642 | sd_idle = 1; |
1da177e4 | 3643 | |
2d72376b | 3644 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3645 | redo: |
3e5459b4 | 3646 | update_shares_locked(this_rq, sd); |
d15bcfdb | 3647 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3648 | &sd_idle, cpus, NULL); |
1da177e4 | 3649 | if (!group) { |
d15bcfdb | 3650 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3651 | goto out_balanced; |
1da177e4 LT |
3652 | } |
3653 | ||
7c16ec58 | 3654 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3655 | if (!busiest) { |
d15bcfdb | 3656 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3657 | goto out_balanced; |
1da177e4 LT |
3658 | } |
3659 | ||
db935dbd NP |
3660 | BUG_ON(busiest == this_rq); |
3661 | ||
d15bcfdb | 3662 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3663 | |
43010659 | 3664 | ld_moved = 0; |
d6d5cfaf NP |
3665 | if (busiest->nr_running > 1) { |
3666 | /* Attempt to move tasks */ | |
3667 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3668 | /* this_rq->clock is already updated */ |
3669 | update_rq_clock(busiest); | |
43010659 | 3670 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3671 | imbalance, sd, CPU_NEWLY_IDLE, |
3672 | &all_pinned); | |
1b12bbc7 | 3673 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 3674 | |
969bb4e4 | 3675 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3676 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3677 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
3678 | goto redo; |
3679 | } | |
d6d5cfaf NP |
3680 | } |
3681 | ||
43010659 | 3682 | if (!ld_moved) { |
36dffab6 | 3683 | int active_balance = 0; |
ad273b32 | 3684 | |
d15bcfdb | 3685 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3686 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3687 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3688 | return -1; |
ad273b32 VS |
3689 | |
3690 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
3691 | return -1; | |
3692 | ||
3693 | if (sd->nr_balance_failed++ < 2) | |
3694 | return -1; | |
3695 | ||
3696 | /* | |
3697 | * The only task running in a non-idle cpu can be moved to this | |
3698 | * cpu in an attempt to completely freeup the other CPU | |
3699 | * package. The same method used to move task in load_balance() | |
3700 | * have been extended for load_balance_newidle() to speedup | |
3701 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
3702 | * | |
3703 | * The package power saving logic comes from | |
3704 | * find_busiest_group(). If there are no imbalance, then | |
3705 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
3706 | * f_b_g() will select a group from which a running task may be | |
3707 | * pulled to this cpu in order to make the other package idle. | |
3708 | * If there is no opportunity to make a package idle and if | |
3709 | * there are no imbalance, then f_b_g() will return NULL and no | |
3710 | * action will be taken in load_balance_newidle(). | |
3711 | * | |
3712 | * Under normal task pull operation due to imbalance, there | |
3713 | * will be more than one task in the source run queue and | |
3714 | * move_tasks() will succeed. ld_moved will be true and this | |
3715 | * active balance code will not be triggered. | |
3716 | */ | |
3717 | ||
3718 | /* Lock busiest in correct order while this_rq is held */ | |
3719 | double_lock_balance(this_rq, busiest); | |
3720 | ||
3721 | /* | |
3722 | * don't kick the migration_thread, if the curr | |
3723 | * task on busiest cpu can't be moved to this_cpu | |
3724 | */ | |
6ca09dfc | 3725 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
3726 | double_unlock_balance(this_rq, busiest); |
3727 | all_pinned = 1; | |
3728 | return ld_moved; | |
3729 | } | |
3730 | ||
3731 | if (!busiest->active_balance) { | |
3732 | busiest->active_balance = 1; | |
3733 | busiest->push_cpu = this_cpu; | |
3734 | active_balance = 1; | |
3735 | } | |
3736 | ||
3737 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
3738 | /* |
3739 | * Should not call ttwu while holding a rq->lock | |
3740 | */ | |
3741 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
3742 | if (active_balance) |
3743 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 3744 | spin_lock(&this_rq->lock); |
ad273b32 | 3745 | |
5969fe06 | 3746 | } else |
16cfb1c0 | 3747 | sd->nr_balance_failed = 0; |
1da177e4 | 3748 | |
3e5459b4 | 3749 | update_shares_locked(this_rq, sd); |
43010659 | 3750 | return ld_moved; |
16cfb1c0 NP |
3751 | |
3752 | out_balanced: | |
d15bcfdb | 3753 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3754 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3755 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3756 | return -1; |
16cfb1c0 | 3757 | sd->nr_balance_failed = 0; |
48f24c4d | 3758 | |
16cfb1c0 | 3759 | return 0; |
1da177e4 LT |
3760 | } |
3761 | ||
3762 | /* | |
3763 | * idle_balance is called by schedule() if this_cpu is about to become | |
3764 | * idle. Attempts to pull tasks from other CPUs. | |
3765 | */ | |
70b97a7f | 3766 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3767 | { |
3768 | struct sched_domain *sd; | |
efbe027e | 3769 | int pulled_task = 0; |
dd41f596 | 3770 | unsigned long next_balance = jiffies + HZ; |
4d2732c6 RR |
3771 | cpumask_var_t tmpmask; |
3772 | ||
3773 | if (!alloc_cpumask_var(&tmpmask, GFP_ATOMIC)) | |
3774 | return; | |
1da177e4 LT |
3775 | |
3776 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3777 | unsigned long interval; |
3778 | ||
3779 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3780 | continue; | |
3781 | ||
3782 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3783 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 3784 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
4d2732c6 | 3785 | sd, tmpmask); |
92c4ca5c CL |
3786 | |
3787 | interval = msecs_to_jiffies(sd->balance_interval); | |
3788 | if (time_after(next_balance, sd->last_balance + interval)) | |
3789 | next_balance = sd->last_balance + interval; | |
3790 | if (pulled_task) | |
3791 | break; | |
1da177e4 | 3792 | } |
dd41f596 | 3793 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3794 | /* |
3795 | * We are going idle. next_balance may be set based on | |
3796 | * a busy processor. So reset next_balance. | |
3797 | */ | |
3798 | this_rq->next_balance = next_balance; | |
dd41f596 | 3799 | } |
4d2732c6 | 3800 | free_cpumask_var(tmpmask); |
1da177e4 LT |
3801 | } |
3802 | ||
3803 | /* | |
3804 | * active_load_balance is run by migration threads. It pushes running tasks | |
3805 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3806 | * running on each physical CPU where possible, and avoids physical / | |
3807 | * logical imbalances. | |
3808 | * | |
3809 | * Called with busiest_rq locked. | |
3810 | */ | |
70b97a7f | 3811 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3812 | { |
39507451 | 3813 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3814 | struct sched_domain *sd; |
3815 | struct rq *target_rq; | |
39507451 | 3816 | |
48f24c4d | 3817 | /* Is there any task to move? */ |
39507451 | 3818 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3819 | return; |
3820 | ||
3821 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3822 | |
3823 | /* | |
39507451 | 3824 | * This condition is "impossible", if it occurs |
41a2d6cf | 3825 | * we need to fix it. Originally reported by |
39507451 | 3826 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3827 | */ |
39507451 | 3828 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3829 | |
39507451 NP |
3830 | /* move a task from busiest_rq to target_rq */ |
3831 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3832 | update_rq_clock(busiest_rq); |
3833 | update_rq_clock(target_rq); | |
39507451 NP |
3834 | |
3835 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3836 | for_each_domain(target_cpu, sd) { |
39507451 | 3837 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 3838 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 3839 | break; |
c96d145e | 3840 | } |
39507451 | 3841 | |
48f24c4d | 3842 | if (likely(sd)) { |
2d72376b | 3843 | schedstat_inc(sd, alb_count); |
39507451 | 3844 | |
43010659 PW |
3845 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3846 | sd, CPU_IDLE)) | |
48f24c4d IM |
3847 | schedstat_inc(sd, alb_pushed); |
3848 | else | |
3849 | schedstat_inc(sd, alb_failed); | |
3850 | } | |
1b12bbc7 | 3851 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
3852 | } |
3853 | ||
46cb4b7c SS |
3854 | #ifdef CONFIG_NO_HZ |
3855 | static struct { | |
3856 | atomic_t load_balancer; | |
7d1e6a9b | 3857 | cpumask_var_t cpu_mask; |
46cb4b7c SS |
3858 | } nohz ____cacheline_aligned = { |
3859 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
3860 | }; |
3861 | ||
7835b98b | 3862 | /* |
46cb4b7c SS |
3863 | * This routine will try to nominate the ilb (idle load balancing) |
3864 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3865 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3866 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3867 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3868 | * arrives... | |
3869 | * | |
3870 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3871 | * for idle load balancing. ilb owner will still be part of | |
3872 | * nohz.cpu_mask.. | |
7835b98b | 3873 | * |
46cb4b7c SS |
3874 | * While stopping the tick, this cpu will become the ilb owner if there |
3875 | * is no other owner. And will be the owner till that cpu becomes busy | |
3876 | * or if all cpus in the system stop their ticks at which point | |
3877 | * there is no need for ilb owner. | |
3878 | * | |
3879 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3880 | * next busy scheduler_tick() | |
3881 | */ | |
3882 | int select_nohz_load_balancer(int stop_tick) | |
3883 | { | |
3884 | int cpu = smp_processor_id(); | |
3885 | ||
3886 | if (stop_tick) { | |
7d1e6a9b | 3887 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
3888 | cpu_rq(cpu)->in_nohz_recently = 1; |
3889 | ||
3890 | /* | |
3891 | * If we are going offline and still the leader, give up! | |
3892 | */ | |
e761b772 | 3893 | if (!cpu_active(cpu) && |
46cb4b7c SS |
3894 | atomic_read(&nohz.load_balancer) == cpu) { |
3895 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3896 | BUG(); | |
3897 | return 0; | |
3898 | } | |
3899 | ||
3900 | /* time for ilb owner also to sleep */ | |
7d1e6a9b | 3901 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
3902 | if (atomic_read(&nohz.load_balancer) == cpu) |
3903 | atomic_set(&nohz.load_balancer, -1); | |
3904 | return 0; | |
3905 | } | |
3906 | ||
3907 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3908 | /* make me the ilb owner */ | |
3909 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3910 | return 1; | |
3911 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3912 | return 1; | |
3913 | } else { | |
7d1e6a9b | 3914 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
3915 | return 0; |
3916 | ||
7d1e6a9b | 3917 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
3918 | |
3919 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3920 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3921 | BUG(); | |
3922 | } | |
3923 | return 0; | |
3924 | } | |
3925 | #endif | |
3926 | ||
3927 | static DEFINE_SPINLOCK(balancing); | |
3928 | ||
3929 | /* | |
7835b98b CL |
3930 | * It checks each scheduling domain to see if it is due to be balanced, |
3931 | * and initiates a balancing operation if so. | |
3932 | * | |
3933 | * Balancing parameters are set up in arch_init_sched_domains. | |
3934 | */ | |
a9957449 | 3935 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3936 | { |
46cb4b7c SS |
3937 | int balance = 1; |
3938 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3939 | unsigned long interval; |
3940 | struct sched_domain *sd; | |
46cb4b7c | 3941 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3942 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3943 | int update_next_balance = 0; |
d07355f5 | 3944 | int need_serialize; |
a0e90245 RR |
3945 | cpumask_var_t tmp; |
3946 | ||
3947 | /* Fails alloc? Rebalancing probably not a priority right now. */ | |
3948 | if (!alloc_cpumask_var(&tmp, GFP_ATOMIC)) | |
3949 | return; | |
1da177e4 | 3950 | |
46cb4b7c | 3951 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3952 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3953 | continue; | |
3954 | ||
3955 | interval = sd->balance_interval; | |
d15bcfdb | 3956 | if (idle != CPU_IDLE) |
1da177e4 LT |
3957 | interval *= sd->busy_factor; |
3958 | ||
3959 | /* scale ms to jiffies */ | |
3960 | interval = msecs_to_jiffies(interval); | |
3961 | if (unlikely(!interval)) | |
3962 | interval = 1; | |
dd41f596 IM |
3963 | if (interval > HZ*NR_CPUS/10) |
3964 | interval = HZ*NR_CPUS/10; | |
3965 | ||
d07355f5 | 3966 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 3967 | |
d07355f5 | 3968 | if (need_serialize) { |
08c183f3 CL |
3969 | if (!spin_trylock(&balancing)) |
3970 | goto out; | |
3971 | } | |
3972 | ||
c9819f45 | 3973 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
a0e90245 | 3974 | if (load_balance(cpu, rq, sd, idle, &balance, tmp)) { |
fa3b6ddc SS |
3975 | /* |
3976 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3977 | * longer idle, or one of our SMT siblings is |
3978 | * not idle. | |
3979 | */ | |
d15bcfdb | 3980 | idle = CPU_NOT_IDLE; |
1da177e4 | 3981 | } |
1bd77f2d | 3982 | sd->last_balance = jiffies; |
1da177e4 | 3983 | } |
d07355f5 | 3984 | if (need_serialize) |
08c183f3 CL |
3985 | spin_unlock(&balancing); |
3986 | out: | |
f549da84 | 3987 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3988 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3989 | update_next_balance = 1; |
3990 | } | |
783609c6 SS |
3991 | |
3992 | /* | |
3993 | * Stop the load balance at this level. There is another | |
3994 | * CPU in our sched group which is doing load balancing more | |
3995 | * actively. | |
3996 | */ | |
3997 | if (!balance) | |
3998 | break; | |
1da177e4 | 3999 | } |
f549da84 SS |
4000 | |
4001 | /* | |
4002 | * next_balance will be updated only when there is a need. | |
4003 | * When the cpu is attached to null domain for ex, it will not be | |
4004 | * updated. | |
4005 | */ | |
4006 | if (likely(update_next_balance)) | |
4007 | rq->next_balance = next_balance; | |
a0e90245 RR |
4008 | |
4009 | free_cpumask_var(tmp); | |
46cb4b7c SS |
4010 | } |
4011 | ||
4012 | /* | |
4013 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4014 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4015 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4016 | */ | |
4017 | static void run_rebalance_domains(struct softirq_action *h) | |
4018 | { | |
dd41f596 IM |
4019 | int this_cpu = smp_processor_id(); |
4020 | struct rq *this_rq = cpu_rq(this_cpu); | |
4021 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4022 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4023 | |
dd41f596 | 4024 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4025 | |
4026 | #ifdef CONFIG_NO_HZ | |
4027 | /* | |
4028 | * If this cpu is the owner for idle load balancing, then do the | |
4029 | * balancing on behalf of the other idle cpus whose ticks are | |
4030 | * stopped. | |
4031 | */ | |
dd41f596 IM |
4032 | if (this_rq->idle_at_tick && |
4033 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4034 | struct rq *rq; |
4035 | int balance_cpu; | |
4036 | ||
7d1e6a9b RR |
4037 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4038 | if (balance_cpu == this_cpu) | |
4039 | continue; | |
4040 | ||
46cb4b7c SS |
4041 | /* |
4042 | * If this cpu gets work to do, stop the load balancing | |
4043 | * work being done for other cpus. Next load | |
4044 | * balancing owner will pick it up. | |
4045 | */ | |
4046 | if (need_resched()) | |
4047 | break; | |
4048 | ||
de0cf899 | 4049 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4050 | |
4051 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4052 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4053 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4054 | } |
4055 | } | |
4056 | #endif | |
4057 | } | |
4058 | ||
4059 | /* | |
4060 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4061 | * | |
4062 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4063 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4064 | * if the whole system is idle. | |
4065 | */ | |
dd41f596 | 4066 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4067 | { |
46cb4b7c SS |
4068 | #ifdef CONFIG_NO_HZ |
4069 | /* | |
4070 | * If we were in the nohz mode recently and busy at the current | |
4071 | * scheduler tick, then check if we need to nominate new idle | |
4072 | * load balancer. | |
4073 | */ | |
4074 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4075 | rq->in_nohz_recently = 0; | |
4076 | ||
4077 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4078 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4079 | atomic_set(&nohz.load_balancer, -1); |
4080 | } | |
4081 | ||
4082 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4083 | /* | |
4084 | * simple selection for now: Nominate the | |
4085 | * first cpu in the nohz list to be the next | |
4086 | * ilb owner. | |
4087 | * | |
4088 | * TBD: Traverse the sched domains and nominate | |
4089 | * the nearest cpu in the nohz.cpu_mask. | |
4090 | */ | |
7d1e6a9b | 4091 | int ilb = cpumask_first(nohz.cpu_mask); |
46cb4b7c | 4092 | |
434d53b0 | 4093 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4094 | resched_cpu(ilb); |
4095 | } | |
4096 | } | |
4097 | ||
4098 | /* | |
4099 | * If this cpu is idle and doing idle load balancing for all the | |
4100 | * cpus with ticks stopped, is it time for that to stop? | |
4101 | */ | |
4102 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4103 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4104 | resched_cpu(cpu); |
4105 | return; | |
4106 | } | |
4107 | ||
4108 | /* | |
4109 | * If this cpu is idle and the idle load balancing is done by | |
4110 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4111 | */ | |
4112 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4113 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4114 | return; |
4115 | #endif | |
4116 | if (time_after_eq(jiffies, rq->next_balance)) | |
4117 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 4118 | } |
dd41f596 IM |
4119 | |
4120 | #else /* CONFIG_SMP */ | |
4121 | ||
1da177e4 LT |
4122 | /* |
4123 | * on UP we do not need to balance between CPUs: | |
4124 | */ | |
70b97a7f | 4125 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4126 | { |
4127 | } | |
dd41f596 | 4128 | |
1da177e4 LT |
4129 | #endif |
4130 | ||
1da177e4 LT |
4131 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4132 | ||
4133 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4134 | ||
4135 | /* | |
f06febc9 FM |
4136 | * Return any ns on the sched_clock that have not yet been banked in |
4137 | * @p in case that task is currently running. | |
1da177e4 | 4138 | */ |
bb34d92f | 4139 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4140 | { |
1da177e4 | 4141 | unsigned long flags; |
41b86e9c | 4142 | struct rq *rq; |
bb34d92f | 4143 | u64 ns = 0; |
48f24c4d | 4144 | |
41b86e9c | 4145 | rq = task_rq_lock(p, &flags); |
1508487e | 4146 | |
051a1d1a | 4147 | if (task_current(rq, p)) { |
f06febc9 FM |
4148 | u64 delta_exec; |
4149 | ||
a8e504d2 IM |
4150 | update_rq_clock(rq); |
4151 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c | 4152 | if ((s64)delta_exec > 0) |
bb34d92f | 4153 | ns = delta_exec; |
41b86e9c | 4154 | } |
48f24c4d | 4155 | |
41b86e9c | 4156 | task_rq_unlock(rq, &flags); |
48f24c4d | 4157 | |
1da177e4 LT |
4158 | return ns; |
4159 | } | |
4160 | ||
1da177e4 LT |
4161 | /* |
4162 | * Account user cpu time to a process. | |
4163 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4164 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4165 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4166 | */ |
457533a7 MS |
4167 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4168 | cputime_t cputime_scaled) | |
1da177e4 LT |
4169 | { |
4170 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4171 | cputime64_t tmp; | |
4172 | ||
457533a7 | 4173 | /* Add user time to process. */ |
1da177e4 | 4174 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4175 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4176 | account_group_user_time(p, cputime); |
1da177e4 LT |
4177 | |
4178 | /* Add user time to cpustat. */ | |
4179 | tmp = cputime_to_cputime64(cputime); | |
4180 | if (TASK_NICE(p) > 0) | |
4181 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4182 | else | |
4183 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
49b5cf34 JL |
4184 | /* Account for user time used */ |
4185 | acct_update_integrals(p); | |
1da177e4 LT |
4186 | } |
4187 | ||
94886b84 LV |
4188 | /* |
4189 | * Account guest cpu time to a process. | |
4190 | * @p: the process that the cpu time gets accounted to | |
4191 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4192 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4193 | */ |
457533a7 MS |
4194 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4195 | cputime_t cputime_scaled) | |
94886b84 LV |
4196 | { |
4197 | cputime64_t tmp; | |
4198 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4199 | ||
4200 | tmp = cputime_to_cputime64(cputime); | |
4201 | ||
457533a7 | 4202 | /* Add guest time to process. */ |
94886b84 | 4203 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4204 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4205 | account_group_user_time(p, cputime); |
94886b84 LV |
4206 | p->gtime = cputime_add(p->gtime, cputime); |
4207 | ||
457533a7 | 4208 | /* Add guest time to cpustat. */ |
94886b84 LV |
4209 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4210 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4211 | } | |
4212 | ||
1da177e4 LT |
4213 | /* |
4214 | * Account system cpu time to a process. | |
4215 | * @p: the process that the cpu time gets accounted to | |
4216 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4217 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4218 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4219 | */ |
4220 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4221 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4222 | { |
4223 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4224 | cputime64_t tmp; |
4225 | ||
983ed7a6 | 4226 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4227 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4228 | return; |
4229 | } | |
94886b84 | 4230 | |
457533a7 | 4231 | /* Add system time to process. */ |
1da177e4 | 4232 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4233 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4234 | account_group_system_time(p, cputime); |
1da177e4 LT |
4235 | |
4236 | /* Add system time to cpustat. */ | |
4237 | tmp = cputime_to_cputime64(cputime); | |
4238 | if (hardirq_count() - hardirq_offset) | |
4239 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4240 | else if (softirq_count()) | |
4241 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4242 | else |
79741dd3 MS |
4243 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4244 | ||
1da177e4 LT |
4245 | /* Account for system time used */ |
4246 | acct_update_integrals(p); | |
1da177e4 LT |
4247 | } |
4248 | ||
c66f08be | 4249 | /* |
1da177e4 | 4250 | * Account for involuntary wait time. |
1da177e4 | 4251 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4252 | */ |
79741dd3 | 4253 | void account_steal_time(cputime_t cputime) |
c66f08be | 4254 | { |
79741dd3 MS |
4255 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4256 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4257 | ||
4258 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4259 | } |
4260 | ||
1da177e4 | 4261 | /* |
79741dd3 MS |
4262 | * Account for idle time. |
4263 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4264 | */ |
79741dd3 | 4265 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4266 | { |
4267 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4268 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4269 | struct rq *rq = this_rq(); |
1da177e4 | 4270 | |
79741dd3 MS |
4271 | if (atomic_read(&rq->nr_iowait) > 0) |
4272 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4273 | else | |
4274 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4275 | } |
4276 | ||
79741dd3 MS |
4277 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4278 | ||
4279 | /* | |
4280 | * Account a single tick of cpu time. | |
4281 | * @p: the process that the cpu time gets accounted to | |
4282 | * @user_tick: indicates if the tick is a user or a system tick | |
4283 | */ | |
4284 | void account_process_tick(struct task_struct *p, int user_tick) | |
4285 | { | |
4286 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
4287 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
4288 | struct rq *rq = this_rq(); | |
4289 | ||
4290 | if (user_tick) | |
4291 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
4292 | else if (p != rq->idle) | |
4293 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, | |
4294 | one_jiffy_scaled); | |
4295 | else | |
4296 | account_idle_time(one_jiffy); | |
4297 | } | |
4298 | ||
4299 | /* | |
4300 | * Account multiple ticks of steal time. | |
4301 | * @p: the process from which the cpu time has been stolen | |
4302 | * @ticks: number of stolen ticks | |
4303 | */ | |
4304 | void account_steal_ticks(unsigned long ticks) | |
4305 | { | |
4306 | account_steal_time(jiffies_to_cputime(ticks)); | |
4307 | } | |
4308 | ||
4309 | /* | |
4310 | * Account multiple ticks of idle time. | |
4311 | * @ticks: number of stolen ticks | |
4312 | */ | |
4313 | void account_idle_ticks(unsigned long ticks) | |
4314 | { | |
4315 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
4316 | } |
4317 | ||
79741dd3 MS |
4318 | #endif |
4319 | ||
49048622 BS |
4320 | /* |
4321 | * Use precise platform statistics if available: | |
4322 | */ | |
4323 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4324 | cputime_t task_utime(struct task_struct *p) | |
4325 | { | |
4326 | return p->utime; | |
4327 | } | |
4328 | ||
4329 | cputime_t task_stime(struct task_struct *p) | |
4330 | { | |
4331 | return p->stime; | |
4332 | } | |
4333 | #else | |
4334 | cputime_t task_utime(struct task_struct *p) | |
4335 | { | |
4336 | clock_t utime = cputime_to_clock_t(p->utime), | |
4337 | total = utime + cputime_to_clock_t(p->stime); | |
4338 | u64 temp; | |
4339 | ||
4340 | /* | |
4341 | * Use CFS's precise accounting: | |
4342 | */ | |
4343 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4344 | ||
4345 | if (total) { | |
4346 | temp *= utime; | |
4347 | do_div(temp, total); | |
4348 | } | |
4349 | utime = (clock_t)temp; | |
4350 | ||
4351 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4352 | return p->prev_utime; | |
4353 | } | |
4354 | ||
4355 | cputime_t task_stime(struct task_struct *p) | |
4356 | { | |
4357 | clock_t stime; | |
4358 | ||
4359 | /* | |
4360 | * Use CFS's precise accounting. (we subtract utime from | |
4361 | * the total, to make sure the total observed by userspace | |
4362 | * grows monotonically - apps rely on that): | |
4363 | */ | |
4364 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4365 | cputime_to_clock_t(task_utime(p)); | |
4366 | ||
4367 | if (stime >= 0) | |
4368 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4369 | ||
4370 | return p->prev_stime; | |
4371 | } | |
4372 | #endif | |
4373 | ||
4374 | inline cputime_t task_gtime(struct task_struct *p) | |
4375 | { | |
4376 | return p->gtime; | |
4377 | } | |
4378 | ||
7835b98b CL |
4379 | /* |
4380 | * This function gets called by the timer code, with HZ frequency. | |
4381 | * We call it with interrupts disabled. | |
4382 | * | |
4383 | * It also gets called by the fork code, when changing the parent's | |
4384 | * timeslices. | |
4385 | */ | |
4386 | void scheduler_tick(void) | |
4387 | { | |
7835b98b CL |
4388 | int cpu = smp_processor_id(); |
4389 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4390 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4391 | |
4392 | sched_clock_tick(); | |
dd41f596 IM |
4393 | |
4394 | spin_lock(&rq->lock); | |
3e51f33f | 4395 | update_rq_clock(rq); |
f1a438d8 | 4396 | update_cpu_load(rq); |
fa85ae24 | 4397 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4398 | spin_unlock(&rq->lock); |
7835b98b | 4399 | |
e418e1c2 | 4400 | #ifdef CONFIG_SMP |
dd41f596 IM |
4401 | rq->idle_at_tick = idle_cpu(cpu); |
4402 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4403 | #endif |
1da177e4 LT |
4404 | } |
4405 | ||
6cd8a4bb SR |
4406 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4407 | defined(CONFIG_PREEMPT_TRACER)) | |
4408 | ||
4409 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4410 | { | |
4411 | if (in_lock_functions(addr)) { | |
4412 | addr = CALLER_ADDR2; | |
4413 | if (in_lock_functions(addr)) | |
4414 | addr = CALLER_ADDR3; | |
4415 | } | |
4416 | return addr; | |
4417 | } | |
1da177e4 | 4418 | |
43627582 | 4419 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4420 | { |
6cd8a4bb | 4421 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4422 | /* |
4423 | * Underflow? | |
4424 | */ | |
9a11b49a IM |
4425 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4426 | return; | |
6cd8a4bb | 4427 | #endif |
1da177e4 | 4428 | preempt_count() += val; |
6cd8a4bb | 4429 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4430 | /* |
4431 | * Spinlock count overflowing soon? | |
4432 | */ | |
33859f7f MOS |
4433 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4434 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4435 | #endif |
4436 | if (preempt_count() == val) | |
4437 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4438 | } |
4439 | EXPORT_SYMBOL(add_preempt_count); | |
4440 | ||
43627582 | 4441 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4442 | { |
6cd8a4bb | 4443 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4444 | /* |
4445 | * Underflow? | |
4446 | */ | |
01e3eb82 | 4447 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4448 | return; |
1da177e4 LT |
4449 | /* |
4450 | * Is the spinlock portion underflowing? | |
4451 | */ | |
9a11b49a IM |
4452 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4453 | !(preempt_count() & PREEMPT_MASK))) | |
4454 | return; | |
6cd8a4bb | 4455 | #endif |
9a11b49a | 4456 | |
6cd8a4bb SR |
4457 | if (preempt_count() == val) |
4458 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4459 | preempt_count() -= val; |
4460 | } | |
4461 | EXPORT_SYMBOL(sub_preempt_count); | |
4462 | ||
4463 | #endif | |
4464 | ||
4465 | /* | |
dd41f596 | 4466 | * Print scheduling while atomic bug: |
1da177e4 | 4467 | */ |
dd41f596 | 4468 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4469 | { |
838225b4 SS |
4470 | struct pt_regs *regs = get_irq_regs(); |
4471 | ||
4472 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4473 | prev->comm, prev->pid, preempt_count()); | |
4474 | ||
dd41f596 | 4475 | debug_show_held_locks(prev); |
e21f5b15 | 4476 | print_modules(); |
dd41f596 IM |
4477 | if (irqs_disabled()) |
4478 | print_irqtrace_events(prev); | |
838225b4 SS |
4479 | |
4480 | if (regs) | |
4481 | show_regs(regs); | |
4482 | else | |
4483 | dump_stack(); | |
dd41f596 | 4484 | } |
1da177e4 | 4485 | |
dd41f596 IM |
4486 | /* |
4487 | * Various schedule()-time debugging checks and statistics: | |
4488 | */ | |
4489 | static inline void schedule_debug(struct task_struct *prev) | |
4490 | { | |
1da177e4 | 4491 | /* |
41a2d6cf | 4492 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4493 | * schedule() atomically, we ignore that path for now. |
4494 | * Otherwise, whine if we are scheduling when we should not be. | |
4495 | */ | |
3f33a7ce | 4496 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4497 | __schedule_bug(prev); |
4498 | ||
1da177e4 LT |
4499 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4500 | ||
2d72376b | 4501 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4502 | #ifdef CONFIG_SCHEDSTATS |
4503 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4504 | schedstat_inc(this_rq(), bkl_count); |
4505 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4506 | } |
4507 | #endif | |
dd41f596 IM |
4508 | } |
4509 | ||
4510 | /* | |
4511 | * Pick up the highest-prio task: | |
4512 | */ | |
4513 | static inline struct task_struct * | |
ff95f3df | 4514 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4515 | { |
5522d5d5 | 4516 | const struct sched_class *class; |
dd41f596 | 4517 | struct task_struct *p; |
1da177e4 LT |
4518 | |
4519 | /* | |
dd41f596 IM |
4520 | * Optimization: we know that if all tasks are in |
4521 | * the fair class we can call that function directly: | |
1da177e4 | 4522 | */ |
dd41f596 | 4523 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4524 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4525 | if (likely(p)) |
4526 | return p; | |
1da177e4 LT |
4527 | } |
4528 | ||
dd41f596 IM |
4529 | class = sched_class_highest; |
4530 | for ( ; ; ) { | |
fb8d4724 | 4531 | p = class->pick_next_task(rq); |
dd41f596 IM |
4532 | if (p) |
4533 | return p; | |
4534 | /* | |
4535 | * Will never be NULL as the idle class always | |
4536 | * returns a non-NULL p: | |
4537 | */ | |
4538 | class = class->next; | |
4539 | } | |
4540 | } | |
1da177e4 | 4541 | |
dd41f596 IM |
4542 | /* |
4543 | * schedule() is the main scheduler function. | |
4544 | */ | |
4545 | asmlinkage void __sched schedule(void) | |
4546 | { | |
4547 | struct task_struct *prev, *next; | |
67ca7bde | 4548 | unsigned long *switch_count; |
dd41f596 | 4549 | struct rq *rq; |
31656519 | 4550 | int cpu; |
dd41f596 IM |
4551 | |
4552 | need_resched: | |
4553 | preempt_disable(); | |
4554 | cpu = smp_processor_id(); | |
4555 | rq = cpu_rq(cpu); | |
4556 | rcu_qsctr_inc(cpu); | |
4557 | prev = rq->curr; | |
4558 | switch_count = &prev->nivcsw; | |
4559 | ||
4560 | release_kernel_lock(prev); | |
4561 | need_resched_nonpreemptible: | |
4562 | ||
4563 | schedule_debug(prev); | |
1da177e4 | 4564 | |
31656519 | 4565 | if (sched_feat(HRTICK)) |
f333fdc9 | 4566 | hrtick_clear(rq); |
8f4d37ec | 4567 | |
8cd162ce | 4568 | spin_lock_irq(&rq->lock); |
3e51f33f | 4569 | update_rq_clock(rq); |
1e819950 | 4570 | clear_tsk_need_resched(prev); |
1da177e4 | 4571 | |
1da177e4 | 4572 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4573 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4574 | prev->state = TASK_RUNNING; |
16882c1e | 4575 | else |
2e1cb74a | 4576 | deactivate_task(rq, prev, 1); |
dd41f596 | 4577 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4578 | } |
4579 | ||
9a897c5a SR |
4580 | #ifdef CONFIG_SMP |
4581 | if (prev->sched_class->pre_schedule) | |
4582 | prev->sched_class->pre_schedule(rq, prev); | |
4583 | #endif | |
f65eda4f | 4584 | |
dd41f596 | 4585 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4586 | idle_balance(cpu, rq); |
1da177e4 | 4587 | |
31ee529c | 4588 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4589 | next = pick_next_task(rq, prev); |
1da177e4 | 4590 | |
1da177e4 | 4591 | if (likely(prev != next)) { |
673a90a1 DS |
4592 | sched_info_switch(prev, next); |
4593 | ||
1da177e4 LT |
4594 | rq->nr_switches++; |
4595 | rq->curr = next; | |
4596 | ++*switch_count; | |
4597 | ||
dd41f596 | 4598 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4599 | /* |
4600 | * the context switch might have flipped the stack from under | |
4601 | * us, hence refresh the local variables. | |
4602 | */ | |
4603 | cpu = smp_processor_id(); | |
4604 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4605 | } else |
4606 | spin_unlock_irq(&rq->lock); | |
4607 | ||
8f4d37ec | 4608 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 4609 | goto need_resched_nonpreemptible; |
8f4d37ec | 4610 | |
1da177e4 LT |
4611 | preempt_enable_no_resched(); |
4612 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4613 | goto need_resched; | |
4614 | } | |
1da177e4 LT |
4615 | EXPORT_SYMBOL(schedule); |
4616 | ||
4617 | #ifdef CONFIG_PREEMPT | |
4618 | /* | |
2ed6e34f | 4619 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4620 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4621 | * occur there and call schedule directly. |
4622 | */ | |
4623 | asmlinkage void __sched preempt_schedule(void) | |
4624 | { | |
4625 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4626 | |
1da177e4 LT |
4627 | /* |
4628 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4629 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4630 | */ |
beed33a8 | 4631 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4632 | return; |
4633 | ||
3a5c359a AK |
4634 | do { |
4635 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4636 | schedule(); |
3a5c359a | 4637 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4638 | |
3a5c359a AK |
4639 | /* |
4640 | * Check again in case we missed a preemption opportunity | |
4641 | * between schedule and now. | |
4642 | */ | |
4643 | barrier(); | |
4644 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4645 | } |
1da177e4 LT |
4646 | EXPORT_SYMBOL(preempt_schedule); |
4647 | ||
4648 | /* | |
2ed6e34f | 4649 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4650 | * off of irq context. |
4651 | * Note, that this is called and return with irqs disabled. This will | |
4652 | * protect us against recursive calling from irq. | |
4653 | */ | |
4654 | asmlinkage void __sched preempt_schedule_irq(void) | |
4655 | { | |
4656 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4657 | |
2ed6e34f | 4658 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4659 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4660 | ||
3a5c359a AK |
4661 | do { |
4662 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4663 | local_irq_enable(); |
4664 | schedule(); | |
4665 | local_irq_disable(); | |
3a5c359a | 4666 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4667 | |
3a5c359a AK |
4668 | /* |
4669 | * Check again in case we missed a preemption opportunity | |
4670 | * between schedule and now. | |
4671 | */ | |
4672 | barrier(); | |
4673 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4674 | } |
4675 | ||
4676 | #endif /* CONFIG_PREEMPT */ | |
4677 | ||
95cdf3b7 IM |
4678 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4679 | void *key) | |
1da177e4 | 4680 | { |
48f24c4d | 4681 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4682 | } |
1da177e4 LT |
4683 | EXPORT_SYMBOL(default_wake_function); |
4684 | ||
4685 | /* | |
41a2d6cf IM |
4686 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4687 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4688 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4689 | * | |
4690 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4691 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4692 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4693 | */ | |
4694 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4695 | int nr_exclusive, int sync, void *key) | |
4696 | { | |
2e45874c | 4697 | wait_queue_t *curr, *next; |
1da177e4 | 4698 | |
2e45874c | 4699 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4700 | unsigned flags = curr->flags; |
4701 | ||
1da177e4 | 4702 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4703 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4704 | break; |
4705 | } | |
4706 | } | |
4707 | ||
4708 | /** | |
4709 | * __wake_up - wake up threads blocked on a waitqueue. | |
4710 | * @q: the waitqueue | |
4711 | * @mode: which threads | |
4712 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4713 | * @key: is directly passed to the wakeup function |
1da177e4 | 4714 | */ |
7ad5b3a5 | 4715 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4716 | int nr_exclusive, void *key) |
1da177e4 LT |
4717 | { |
4718 | unsigned long flags; | |
4719 | ||
4720 | spin_lock_irqsave(&q->lock, flags); | |
4721 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4722 | spin_unlock_irqrestore(&q->lock, flags); | |
4723 | } | |
1da177e4 LT |
4724 | EXPORT_SYMBOL(__wake_up); |
4725 | ||
4726 | /* | |
4727 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4728 | */ | |
7ad5b3a5 | 4729 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4730 | { |
4731 | __wake_up_common(q, mode, 1, 0, NULL); | |
4732 | } | |
4733 | ||
4734 | /** | |
67be2dd1 | 4735 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4736 | * @q: the waitqueue |
4737 | * @mode: which threads | |
4738 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4739 | * | |
4740 | * The sync wakeup differs that the waker knows that it will schedule | |
4741 | * away soon, so while the target thread will be woken up, it will not | |
4742 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4743 | * with each other. This can prevent needless bouncing between CPUs. | |
4744 | * | |
4745 | * On UP it can prevent extra preemption. | |
4746 | */ | |
7ad5b3a5 | 4747 | void |
95cdf3b7 | 4748 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4749 | { |
4750 | unsigned long flags; | |
4751 | int sync = 1; | |
4752 | ||
4753 | if (unlikely(!q)) | |
4754 | return; | |
4755 | ||
4756 | if (unlikely(!nr_exclusive)) | |
4757 | sync = 0; | |
4758 | ||
4759 | spin_lock_irqsave(&q->lock, flags); | |
4760 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4761 | spin_unlock_irqrestore(&q->lock, flags); | |
4762 | } | |
4763 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4764 | ||
65eb3dc6 KD |
4765 | /** |
4766 | * complete: - signals a single thread waiting on this completion | |
4767 | * @x: holds the state of this particular completion | |
4768 | * | |
4769 | * This will wake up a single thread waiting on this completion. Threads will be | |
4770 | * awakened in the same order in which they were queued. | |
4771 | * | |
4772 | * See also complete_all(), wait_for_completion() and related routines. | |
4773 | */ | |
b15136e9 | 4774 | void complete(struct completion *x) |
1da177e4 LT |
4775 | { |
4776 | unsigned long flags; | |
4777 | ||
4778 | spin_lock_irqsave(&x->wait.lock, flags); | |
4779 | x->done++; | |
d9514f6c | 4780 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4781 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4782 | } | |
4783 | EXPORT_SYMBOL(complete); | |
4784 | ||
65eb3dc6 KD |
4785 | /** |
4786 | * complete_all: - signals all threads waiting on this completion | |
4787 | * @x: holds the state of this particular completion | |
4788 | * | |
4789 | * This will wake up all threads waiting on this particular completion event. | |
4790 | */ | |
b15136e9 | 4791 | void complete_all(struct completion *x) |
1da177e4 LT |
4792 | { |
4793 | unsigned long flags; | |
4794 | ||
4795 | spin_lock_irqsave(&x->wait.lock, flags); | |
4796 | x->done += UINT_MAX/2; | |
d9514f6c | 4797 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4798 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4799 | } | |
4800 | EXPORT_SYMBOL(complete_all); | |
4801 | ||
8cbbe86d AK |
4802 | static inline long __sched |
4803 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4804 | { |
1da177e4 LT |
4805 | if (!x->done) { |
4806 | DECLARE_WAITQUEUE(wait, current); | |
4807 | ||
4808 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4809 | __add_wait_queue_tail(&x->wait, &wait); | |
4810 | do { | |
94d3d824 | 4811 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4812 | timeout = -ERESTARTSYS; |
4813 | break; | |
8cbbe86d AK |
4814 | } |
4815 | __set_current_state(state); | |
1da177e4 LT |
4816 | spin_unlock_irq(&x->wait.lock); |
4817 | timeout = schedule_timeout(timeout); | |
4818 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4819 | } while (!x->done && timeout); |
1da177e4 | 4820 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4821 | if (!x->done) |
4822 | return timeout; | |
1da177e4 LT |
4823 | } |
4824 | x->done--; | |
ea71a546 | 4825 | return timeout ?: 1; |
1da177e4 | 4826 | } |
1da177e4 | 4827 | |
8cbbe86d AK |
4828 | static long __sched |
4829 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4830 | { |
1da177e4 LT |
4831 | might_sleep(); |
4832 | ||
4833 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4834 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4835 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4836 | return timeout; |
4837 | } | |
1da177e4 | 4838 | |
65eb3dc6 KD |
4839 | /** |
4840 | * wait_for_completion: - waits for completion of a task | |
4841 | * @x: holds the state of this particular completion | |
4842 | * | |
4843 | * This waits to be signaled for completion of a specific task. It is NOT | |
4844 | * interruptible and there is no timeout. | |
4845 | * | |
4846 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4847 | * and interrupt capability. Also see complete(). | |
4848 | */ | |
b15136e9 | 4849 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4850 | { |
4851 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4852 | } |
8cbbe86d | 4853 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4854 | |
65eb3dc6 KD |
4855 | /** |
4856 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4857 | * @x: holds the state of this particular completion | |
4858 | * @timeout: timeout value in jiffies | |
4859 | * | |
4860 | * This waits for either a completion of a specific task to be signaled or for a | |
4861 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4862 | * interruptible. | |
4863 | */ | |
b15136e9 | 4864 | unsigned long __sched |
8cbbe86d | 4865 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4866 | { |
8cbbe86d | 4867 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4868 | } |
8cbbe86d | 4869 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4870 | |
65eb3dc6 KD |
4871 | /** |
4872 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4873 | * @x: holds the state of this particular completion | |
4874 | * | |
4875 | * This waits for completion of a specific task to be signaled. It is | |
4876 | * interruptible. | |
4877 | */ | |
8cbbe86d | 4878 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4879 | { |
51e97990 AK |
4880 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4881 | if (t == -ERESTARTSYS) | |
4882 | return t; | |
4883 | return 0; | |
0fec171c | 4884 | } |
8cbbe86d | 4885 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4886 | |
65eb3dc6 KD |
4887 | /** |
4888 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4889 | * @x: holds the state of this particular completion | |
4890 | * @timeout: timeout value in jiffies | |
4891 | * | |
4892 | * This waits for either a completion of a specific task to be signaled or for a | |
4893 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4894 | */ | |
b15136e9 | 4895 | unsigned long __sched |
8cbbe86d AK |
4896 | wait_for_completion_interruptible_timeout(struct completion *x, |
4897 | unsigned long timeout) | |
0fec171c | 4898 | { |
8cbbe86d | 4899 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4900 | } |
8cbbe86d | 4901 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4902 | |
65eb3dc6 KD |
4903 | /** |
4904 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4905 | * @x: holds the state of this particular completion | |
4906 | * | |
4907 | * This waits to be signaled for completion of a specific task. It can be | |
4908 | * interrupted by a kill signal. | |
4909 | */ | |
009e577e MW |
4910 | int __sched wait_for_completion_killable(struct completion *x) |
4911 | { | |
4912 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4913 | if (t == -ERESTARTSYS) | |
4914 | return t; | |
4915 | return 0; | |
4916 | } | |
4917 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4918 | ||
be4de352 DC |
4919 | /** |
4920 | * try_wait_for_completion - try to decrement a completion without blocking | |
4921 | * @x: completion structure | |
4922 | * | |
4923 | * Returns: 0 if a decrement cannot be done without blocking | |
4924 | * 1 if a decrement succeeded. | |
4925 | * | |
4926 | * If a completion is being used as a counting completion, | |
4927 | * attempt to decrement the counter without blocking. This | |
4928 | * enables us to avoid waiting if the resource the completion | |
4929 | * is protecting is not available. | |
4930 | */ | |
4931 | bool try_wait_for_completion(struct completion *x) | |
4932 | { | |
4933 | int ret = 1; | |
4934 | ||
4935 | spin_lock_irq(&x->wait.lock); | |
4936 | if (!x->done) | |
4937 | ret = 0; | |
4938 | else | |
4939 | x->done--; | |
4940 | spin_unlock_irq(&x->wait.lock); | |
4941 | return ret; | |
4942 | } | |
4943 | EXPORT_SYMBOL(try_wait_for_completion); | |
4944 | ||
4945 | /** | |
4946 | * completion_done - Test to see if a completion has any waiters | |
4947 | * @x: completion structure | |
4948 | * | |
4949 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4950 | * 1 if there are no waiters. | |
4951 | * | |
4952 | */ | |
4953 | bool completion_done(struct completion *x) | |
4954 | { | |
4955 | int ret = 1; | |
4956 | ||
4957 | spin_lock_irq(&x->wait.lock); | |
4958 | if (!x->done) | |
4959 | ret = 0; | |
4960 | spin_unlock_irq(&x->wait.lock); | |
4961 | return ret; | |
4962 | } | |
4963 | EXPORT_SYMBOL(completion_done); | |
4964 | ||
8cbbe86d AK |
4965 | static long __sched |
4966 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4967 | { |
0fec171c IM |
4968 | unsigned long flags; |
4969 | wait_queue_t wait; | |
4970 | ||
4971 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4972 | |
8cbbe86d | 4973 | __set_current_state(state); |
1da177e4 | 4974 | |
8cbbe86d AK |
4975 | spin_lock_irqsave(&q->lock, flags); |
4976 | __add_wait_queue(q, &wait); | |
4977 | spin_unlock(&q->lock); | |
4978 | timeout = schedule_timeout(timeout); | |
4979 | spin_lock_irq(&q->lock); | |
4980 | __remove_wait_queue(q, &wait); | |
4981 | spin_unlock_irqrestore(&q->lock, flags); | |
4982 | ||
4983 | return timeout; | |
4984 | } | |
4985 | ||
4986 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4987 | { | |
4988 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4989 | } |
1da177e4 LT |
4990 | EXPORT_SYMBOL(interruptible_sleep_on); |
4991 | ||
0fec171c | 4992 | long __sched |
95cdf3b7 | 4993 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4994 | { |
8cbbe86d | 4995 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4996 | } |
1da177e4 LT |
4997 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4998 | ||
0fec171c | 4999 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5000 | { |
8cbbe86d | 5001 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5002 | } |
1da177e4 LT |
5003 | EXPORT_SYMBOL(sleep_on); |
5004 | ||
0fec171c | 5005 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5006 | { |
8cbbe86d | 5007 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5008 | } |
1da177e4 LT |
5009 | EXPORT_SYMBOL(sleep_on_timeout); |
5010 | ||
b29739f9 IM |
5011 | #ifdef CONFIG_RT_MUTEXES |
5012 | ||
5013 | /* | |
5014 | * rt_mutex_setprio - set the current priority of a task | |
5015 | * @p: task | |
5016 | * @prio: prio value (kernel-internal form) | |
5017 | * | |
5018 | * This function changes the 'effective' priority of a task. It does | |
5019 | * not touch ->normal_prio like __setscheduler(). | |
5020 | * | |
5021 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5022 | */ | |
36c8b586 | 5023 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5024 | { |
5025 | unsigned long flags; | |
83b699ed | 5026 | int oldprio, on_rq, running; |
70b97a7f | 5027 | struct rq *rq; |
cb469845 | 5028 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5029 | |
5030 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5031 | ||
5032 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5033 | update_rq_clock(rq); |
b29739f9 | 5034 | |
d5f9f942 | 5035 | oldprio = p->prio; |
dd41f596 | 5036 | on_rq = p->se.on_rq; |
051a1d1a | 5037 | running = task_current(rq, p); |
0e1f3483 | 5038 | if (on_rq) |
69be72c1 | 5039 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5040 | if (running) |
5041 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5042 | |
5043 | if (rt_prio(prio)) | |
5044 | p->sched_class = &rt_sched_class; | |
5045 | else | |
5046 | p->sched_class = &fair_sched_class; | |
5047 | ||
b29739f9 IM |
5048 | p->prio = prio; |
5049 | ||
0e1f3483 HS |
5050 | if (running) |
5051 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5052 | if (on_rq) { |
8159f87e | 5053 | enqueue_task(rq, p, 0); |
cb469845 SR |
5054 | |
5055 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5056 | } |
5057 | task_rq_unlock(rq, &flags); | |
5058 | } | |
5059 | ||
5060 | #endif | |
5061 | ||
36c8b586 | 5062 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5063 | { |
dd41f596 | 5064 | int old_prio, delta, on_rq; |
1da177e4 | 5065 | unsigned long flags; |
70b97a7f | 5066 | struct rq *rq; |
1da177e4 LT |
5067 | |
5068 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5069 | return; | |
5070 | /* | |
5071 | * We have to be careful, if called from sys_setpriority(), | |
5072 | * the task might be in the middle of scheduling on another CPU. | |
5073 | */ | |
5074 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5075 | update_rq_clock(rq); |
1da177e4 LT |
5076 | /* |
5077 | * The RT priorities are set via sched_setscheduler(), but we still | |
5078 | * allow the 'normal' nice value to be set - but as expected | |
5079 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5080 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5081 | */ |
e05606d3 | 5082 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5083 | p->static_prio = NICE_TO_PRIO(nice); |
5084 | goto out_unlock; | |
5085 | } | |
dd41f596 | 5086 | on_rq = p->se.on_rq; |
c09595f6 | 5087 | if (on_rq) |
69be72c1 | 5088 | dequeue_task(rq, p, 0); |
1da177e4 | 5089 | |
1da177e4 | 5090 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5091 | set_load_weight(p); |
b29739f9 IM |
5092 | old_prio = p->prio; |
5093 | p->prio = effective_prio(p); | |
5094 | delta = p->prio - old_prio; | |
1da177e4 | 5095 | |
dd41f596 | 5096 | if (on_rq) { |
8159f87e | 5097 | enqueue_task(rq, p, 0); |
1da177e4 | 5098 | /* |
d5f9f942 AM |
5099 | * If the task increased its priority or is running and |
5100 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5101 | */ |
d5f9f942 | 5102 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5103 | resched_task(rq->curr); |
5104 | } | |
5105 | out_unlock: | |
5106 | task_rq_unlock(rq, &flags); | |
5107 | } | |
1da177e4 LT |
5108 | EXPORT_SYMBOL(set_user_nice); |
5109 | ||
e43379f1 MM |
5110 | /* |
5111 | * can_nice - check if a task can reduce its nice value | |
5112 | * @p: task | |
5113 | * @nice: nice value | |
5114 | */ | |
36c8b586 | 5115 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5116 | { |
024f4747 MM |
5117 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5118 | int nice_rlim = 20 - nice; | |
48f24c4d | 5119 | |
e43379f1 MM |
5120 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5121 | capable(CAP_SYS_NICE)); | |
5122 | } | |
5123 | ||
1da177e4 LT |
5124 | #ifdef __ARCH_WANT_SYS_NICE |
5125 | ||
5126 | /* | |
5127 | * sys_nice - change the priority of the current process. | |
5128 | * @increment: priority increment | |
5129 | * | |
5130 | * sys_setpriority is a more generic, but much slower function that | |
5131 | * does similar things. | |
5132 | */ | |
5add95d4 | 5133 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5134 | { |
48f24c4d | 5135 | long nice, retval; |
1da177e4 LT |
5136 | |
5137 | /* | |
5138 | * Setpriority might change our priority at the same moment. | |
5139 | * We don't have to worry. Conceptually one call occurs first | |
5140 | * and we have a single winner. | |
5141 | */ | |
e43379f1 MM |
5142 | if (increment < -40) |
5143 | increment = -40; | |
1da177e4 LT |
5144 | if (increment > 40) |
5145 | increment = 40; | |
5146 | ||
5147 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
5148 | if (nice < -20) | |
5149 | nice = -20; | |
5150 | if (nice > 19) | |
5151 | nice = 19; | |
5152 | ||
e43379f1 MM |
5153 | if (increment < 0 && !can_nice(current, nice)) |
5154 | return -EPERM; | |
5155 | ||
1da177e4 LT |
5156 | retval = security_task_setnice(current, nice); |
5157 | if (retval) | |
5158 | return retval; | |
5159 | ||
5160 | set_user_nice(current, nice); | |
5161 | return 0; | |
5162 | } | |
5163 | ||
5164 | #endif | |
5165 | ||
5166 | /** | |
5167 | * task_prio - return the priority value of a given task. | |
5168 | * @p: the task in question. | |
5169 | * | |
5170 | * This is the priority value as seen by users in /proc. | |
5171 | * RT tasks are offset by -200. Normal tasks are centered | |
5172 | * around 0, value goes from -16 to +15. | |
5173 | */ | |
36c8b586 | 5174 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5175 | { |
5176 | return p->prio - MAX_RT_PRIO; | |
5177 | } | |
5178 | ||
5179 | /** | |
5180 | * task_nice - return the nice value of a given task. | |
5181 | * @p: the task in question. | |
5182 | */ | |
36c8b586 | 5183 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5184 | { |
5185 | return TASK_NICE(p); | |
5186 | } | |
150d8bed | 5187 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5188 | |
5189 | /** | |
5190 | * idle_cpu - is a given cpu idle currently? | |
5191 | * @cpu: the processor in question. | |
5192 | */ | |
5193 | int idle_cpu(int cpu) | |
5194 | { | |
5195 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5196 | } | |
5197 | ||
1da177e4 LT |
5198 | /** |
5199 | * idle_task - return the idle task for a given cpu. | |
5200 | * @cpu: the processor in question. | |
5201 | */ | |
36c8b586 | 5202 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5203 | { |
5204 | return cpu_rq(cpu)->idle; | |
5205 | } | |
5206 | ||
5207 | /** | |
5208 | * find_process_by_pid - find a process with a matching PID value. | |
5209 | * @pid: the pid in question. | |
5210 | */ | |
a9957449 | 5211 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5212 | { |
228ebcbe | 5213 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5214 | } |
5215 | ||
5216 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5217 | static void |
5218 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5219 | { |
dd41f596 | 5220 | BUG_ON(p->se.on_rq); |
48f24c4d | 5221 | |
1da177e4 | 5222 | p->policy = policy; |
dd41f596 IM |
5223 | switch (p->policy) { |
5224 | case SCHED_NORMAL: | |
5225 | case SCHED_BATCH: | |
5226 | case SCHED_IDLE: | |
5227 | p->sched_class = &fair_sched_class; | |
5228 | break; | |
5229 | case SCHED_FIFO: | |
5230 | case SCHED_RR: | |
5231 | p->sched_class = &rt_sched_class; | |
5232 | break; | |
5233 | } | |
5234 | ||
1da177e4 | 5235 | p->rt_priority = prio; |
b29739f9 IM |
5236 | p->normal_prio = normal_prio(p); |
5237 | /* we are holding p->pi_lock already */ | |
5238 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5239 | set_load_weight(p); |
1da177e4 LT |
5240 | } |
5241 | ||
c69e8d9c DH |
5242 | /* |
5243 | * check the target process has a UID that matches the current process's | |
5244 | */ | |
5245 | static bool check_same_owner(struct task_struct *p) | |
5246 | { | |
5247 | const struct cred *cred = current_cred(), *pcred; | |
5248 | bool match; | |
5249 | ||
5250 | rcu_read_lock(); | |
5251 | pcred = __task_cred(p); | |
5252 | match = (cred->euid == pcred->euid || | |
5253 | cred->euid == pcred->uid); | |
5254 | rcu_read_unlock(); | |
5255 | return match; | |
5256 | } | |
5257 | ||
961ccddd RR |
5258 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5259 | struct sched_param *param, bool user) | |
1da177e4 | 5260 | { |
83b699ed | 5261 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5262 | unsigned long flags; |
cb469845 | 5263 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5264 | struct rq *rq; |
1da177e4 | 5265 | |
66e5393a SR |
5266 | /* may grab non-irq protected spin_locks */ |
5267 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5268 | recheck: |
5269 | /* double check policy once rq lock held */ | |
5270 | if (policy < 0) | |
5271 | policy = oldpolicy = p->policy; | |
5272 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5273 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5274 | policy != SCHED_IDLE) | |
b0a9499c | 5275 | return -EINVAL; |
1da177e4 LT |
5276 | /* |
5277 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5278 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5279 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5280 | */ |
5281 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5282 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5283 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5284 | return -EINVAL; |
e05606d3 | 5285 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5286 | return -EINVAL; |
5287 | ||
37e4ab3f OC |
5288 | /* |
5289 | * Allow unprivileged RT tasks to decrease priority: | |
5290 | */ | |
961ccddd | 5291 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5292 | if (rt_policy(policy)) { |
8dc3e909 | 5293 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5294 | |
5295 | if (!lock_task_sighand(p, &flags)) | |
5296 | return -ESRCH; | |
5297 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5298 | unlock_task_sighand(p, &flags); | |
5299 | ||
5300 | /* can't set/change the rt policy */ | |
5301 | if (policy != p->policy && !rlim_rtprio) | |
5302 | return -EPERM; | |
5303 | ||
5304 | /* can't increase priority */ | |
5305 | if (param->sched_priority > p->rt_priority && | |
5306 | param->sched_priority > rlim_rtprio) | |
5307 | return -EPERM; | |
5308 | } | |
dd41f596 IM |
5309 | /* |
5310 | * Like positive nice levels, dont allow tasks to | |
5311 | * move out of SCHED_IDLE either: | |
5312 | */ | |
5313 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5314 | return -EPERM; | |
5fe1d75f | 5315 | |
37e4ab3f | 5316 | /* can't change other user's priorities */ |
c69e8d9c | 5317 | if (!check_same_owner(p)) |
37e4ab3f OC |
5318 | return -EPERM; |
5319 | } | |
1da177e4 | 5320 | |
725aad24 | 5321 | if (user) { |
b68aa230 | 5322 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5323 | /* |
5324 | * Do not allow realtime tasks into groups that have no runtime | |
5325 | * assigned. | |
5326 | */ | |
9a7e0b18 PZ |
5327 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5328 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5329 | return -EPERM; |
b68aa230 PZ |
5330 | #endif |
5331 | ||
725aad24 JF |
5332 | retval = security_task_setscheduler(p, policy, param); |
5333 | if (retval) | |
5334 | return retval; | |
5335 | } | |
5336 | ||
b29739f9 IM |
5337 | /* |
5338 | * make sure no PI-waiters arrive (or leave) while we are | |
5339 | * changing the priority of the task: | |
5340 | */ | |
5341 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5342 | /* |
5343 | * To be able to change p->policy safely, the apropriate | |
5344 | * runqueue lock must be held. | |
5345 | */ | |
b29739f9 | 5346 | rq = __task_rq_lock(p); |
1da177e4 LT |
5347 | /* recheck policy now with rq lock held */ |
5348 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5349 | policy = oldpolicy = -1; | |
b29739f9 IM |
5350 | __task_rq_unlock(rq); |
5351 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5352 | goto recheck; |
5353 | } | |
2daa3577 | 5354 | update_rq_clock(rq); |
dd41f596 | 5355 | on_rq = p->se.on_rq; |
051a1d1a | 5356 | running = task_current(rq, p); |
0e1f3483 | 5357 | if (on_rq) |
2e1cb74a | 5358 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5359 | if (running) |
5360 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5361 | |
1da177e4 | 5362 | oldprio = p->prio; |
dd41f596 | 5363 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5364 | |
0e1f3483 HS |
5365 | if (running) |
5366 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5367 | if (on_rq) { |
5368 | activate_task(rq, p, 0); | |
cb469845 SR |
5369 | |
5370 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5371 | } |
b29739f9 IM |
5372 | __task_rq_unlock(rq); |
5373 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5374 | ||
95e02ca9 TG |
5375 | rt_mutex_adjust_pi(p); |
5376 | ||
1da177e4 LT |
5377 | return 0; |
5378 | } | |
961ccddd RR |
5379 | |
5380 | /** | |
5381 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5382 | * @p: the task in question. | |
5383 | * @policy: new policy. | |
5384 | * @param: structure containing the new RT priority. | |
5385 | * | |
5386 | * NOTE that the task may be already dead. | |
5387 | */ | |
5388 | int sched_setscheduler(struct task_struct *p, int policy, | |
5389 | struct sched_param *param) | |
5390 | { | |
5391 | return __sched_setscheduler(p, policy, param, true); | |
5392 | } | |
1da177e4 LT |
5393 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5394 | ||
961ccddd RR |
5395 | /** |
5396 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5397 | * @p: the task in question. | |
5398 | * @policy: new policy. | |
5399 | * @param: structure containing the new RT priority. | |
5400 | * | |
5401 | * Just like sched_setscheduler, only don't bother checking if the | |
5402 | * current context has permission. For example, this is needed in | |
5403 | * stop_machine(): we create temporary high priority worker threads, | |
5404 | * but our caller might not have that capability. | |
5405 | */ | |
5406 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5407 | struct sched_param *param) | |
5408 | { | |
5409 | return __sched_setscheduler(p, policy, param, false); | |
5410 | } | |
5411 | ||
95cdf3b7 IM |
5412 | static int |
5413 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5414 | { |
1da177e4 LT |
5415 | struct sched_param lparam; |
5416 | struct task_struct *p; | |
36c8b586 | 5417 | int retval; |
1da177e4 LT |
5418 | |
5419 | if (!param || pid < 0) | |
5420 | return -EINVAL; | |
5421 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5422 | return -EFAULT; | |
5fe1d75f ON |
5423 | |
5424 | rcu_read_lock(); | |
5425 | retval = -ESRCH; | |
1da177e4 | 5426 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5427 | if (p != NULL) |
5428 | retval = sched_setscheduler(p, policy, &lparam); | |
5429 | rcu_read_unlock(); | |
36c8b586 | 5430 | |
1da177e4 LT |
5431 | return retval; |
5432 | } | |
5433 | ||
5434 | /** | |
5435 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5436 | * @pid: the pid in question. | |
5437 | * @policy: new policy. | |
5438 | * @param: structure containing the new RT priority. | |
5439 | */ | |
5add95d4 HC |
5440 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5441 | struct sched_param __user *, param) | |
1da177e4 | 5442 | { |
c21761f1 JB |
5443 | /* negative values for policy are not valid */ |
5444 | if (policy < 0) | |
5445 | return -EINVAL; | |
5446 | ||
1da177e4 LT |
5447 | return do_sched_setscheduler(pid, policy, param); |
5448 | } | |
5449 | ||
5450 | /** | |
5451 | * sys_sched_setparam - set/change the RT priority of a thread | |
5452 | * @pid: the pid in question. | |
5453 | * @param: structure containing the new RT priority. | |
5454 | */ | |
5add95d4 | 5455 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5456 | { |
5457 | return do_sched_setscheduler(pid, -1, param); | |
5458 | } | |
5459 | ||
5460 | /** | |
5461 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5462 | * @pid: the pid in question. | |
5463 | */ | |
5add95d4 | 5464 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5465 | { |
36c8b586 | 5466 | struct task_struct *p; |
3a5c359a | 5467 | int retval; |
1da177e4 LT |
5468 | |
5469 | if (pid < 0) | |
3a5c359a | 5470 | return -EINVAL; |
1da177e4 LT |
5471 | |
5472 | retval = -ESRCH; | |
5473 | read_lock(&tasklist_lock); | |
5474 | p = find_process_by_pid(pid); | |
5475 | if (p) { | |
5476 | retval = security_task_getscheduler(p); | |
5477 | if (!retval) | |
5478 | retval = p->policy; | |
5479 | } | |
5480 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
5481 | return retval; |
5482 | } | |
5483 | ||
5484 | /** | |
5485 | * sys_sched_getscheduler - get the RT priority of a thread | |
5486 | * @pid: the pid in question. | |
5487 | * @param: structure containing the RT priority. | |
5488 | */ | |
5add95d4 | 5489 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5490 | { |
5491 | struct sched_param lp; | |
36c8b586 | 5492 | struct task_struct *p; |
3a5c359a | 5493 | int retval; |
1da177e4 LT |
5494 | |
5495 | if (!param || pid < 0) | |
3a5c359a | 5496 | return -EINVAL; |
1da177e4 LT |
5497 | |
5498 | read_lock(&tasklist_lock); | |
5499 | p = find_process_by_pid(pid); | |
5500 | retval = -ESRCH; | |
5501 | if (!p) | |
5502 | goto out_unlock; | |
5503 | ||
5504 | retval = security_task_getscheduler(p); | |
5505 | if (retval) | |
5506 | goto out_unlock; | |
5507 | ||
5508 | lp.sched_priority = p->rt_priority; | |
5509 | read_unlock(&tasklist_lock); | |
5510 | ||
5511 | /* | |
5512 | * This one might sleep, we cannot do it with a spinlock held ... | |
5513 | */ | |
5514 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5515 | ||
1da177e4 LT |
5516 | return retval; |
5517 | ||
5518 | out_unlock: | |
5519 | read_unlock(&tasklist_lock); | |
5520 | return retval; | |
5521 | } | |
5522 | ||
96f874e2 | 5523 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5524 | { |
5a16f3d3 | 5525 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5526 | struct task_struct *p; |
5527 | int retval; | |
1da177e4 | 5528 | |
95402b38 | 5529 | get_online_cpus(); |
1da177e4 LT |
5530 | read_lock(&tasklist_lock); |
5531 | ||
5532 | p = find_process_by_pid(pid); | |
5533 | if (!p) { | |
5534 | read_unlock(&tasklist_lock); | |
95402b38 | 5535 | put_online_cpus(); |
1da177e4 LT |
5536 | return -ESRCH; |
5537 | } | |
5538 | ||
5539 | /* | |
5540 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5541 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5542 | * usage count and then drop tasklist_lock. |
5543 | */ | |
5544 | get_task_struct(p); | |
5545 | read_unlock(&tasklist_lock); | |
5546 | ||
5a16f3d3 RR |
5547 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5548 | retval = -ENOMEM; | |
5549 | goto out_put_task; | |
5550 | } | |
5551 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5552 | retval = -ENOMEM; | |
5553 | goto out_free_cpus_allowed; | |
5554 | } | |
1da177e4 | 5555 | retval = -EPERM; |
c69e8d9c | 5556 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
5557 | goto out_unlock; |
5558 | ||
e7834f8f DQ |
5559 | retval = security_task_setscheduler(p, 0, NULL); |
5560 | if (retval) | |
5561 | goto out_unlock; | |
5562 | ||
5a16f3d3 RR |
5563 | cpuset_cpus_allowed(p, cpus_allowed); |
5564 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 5565 | again: |
5a16f3d3 | 5566 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5567 | |
8707d8b8 | 5568 | if (!retval) { |
5a16f3d3 RR |
5569 | cpuset_cpus_allowed(p, cpus_allowed); |
5570 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5571 | /* |
5572 | * We must have raced with a concurrent cpuset | |
5573 | * update. Just reset the cpus_allowed to the | |
5574 | * cpuset's cpus_allowed | |
5575 | */ | |
5a16f3d3 | 5576 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5577 | goto again; |
5578 | } | |
5579 | } | |
1da177e4 | 5580 | out_unlock: |
5a16f3d3 RR |
5581 | free_cpumask_var(new_mask); |
5582 | out_free_cpus_allowed: | |
5583 | free_cpumask_var(cpus_allowed); | |
5584 | out_put_task: | |
1da177e4 | 5585 | put_task_struct(p); |
95402b38 | 5586 | put_online_cpus(); |
1da177e4 LT |
5587 | return retval; |
5588 | } | |
5589 | ||
5590 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5591 | struct cpumask *new_mask) |
1da177e4 | 5592 | { |
96f874e2 RR |
5593 | if (len < cpumask_size()) |
5594 | cpumask_clear(new_mask); | |
5595 | else if (len > cpumask_size()) | |
5596 | len = cpumask_size(); | |
5597 | ||
1da177e4 LT |
5598 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5599 | } | |
5600 | ||
5601 | /** | |
5602 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5603 | * @pid: pid of the process | |
5604 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5605 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5606 | */ | |
5add95d4 HC |
5607 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5608 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5609 | { |
5a16f3d3 | 5610 | cpumask_var_t new_mask; |
1da177e4 LT |
5611 | int retval; |
5612 | ||
5a16f3d3 RR |
5613 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5614 | return -ENOMEM; | |
1da177e4 | 5615 | |
5a16f3d3 RR |
5616 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5617 | if (retval == 0) | |
5618 | retval = sched_setaffinity(pid, new_mask); | |
5619 | free_cpumask_var(new_mask); | |
5620 | return retval; | |
1da177e4 LT |
5621 | } |
5622 | ||
96f874e2 | 5623 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5624 | { |
36c8b586 | 5625 | struct task_struct *p; |
1da177e4 | 5626 | int retval; |
1da177e4 | 5627 | |
95402b38 | 5628 | get_online_cpus(); |
1da177e4 LT |
5629 | read_lock(&tasklist_lock); |
5630 | ||
5631 | retval = -ESRCH; | |
5632 | p = find_process_by_pid(pid); | |
5633 | if (!p) | |
5634 | goto out_unlock; | |
5635 | ||
e7834f8f DQ |
5636 | retval = security_task_getscheduler(p); |
5637 | if (retval) | |
5638 | goto out_unlock; | |
5639 | ||
96f874e2 | 5640 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
5641 | |
5642 | out_unlock: | |
5643 | read_unlock(&tasklist_lock); | |
95402b38 | 5644 | put_online_cpus(); |
1da177e4 | 5645 | |
9531b62f | 5646 | return retval; |
1da177e4 LT |
5647 | } |
5648 | ||
5649 | /** | |
5650 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5651 | * @pid: pid of the process | |
5652 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5653 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5654 | */ | |
5add95d4 HC |
5655 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5656 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5657 | { |
5658 | int ret; | |
f17c8607 | 5659 | cpumask_var_t mask; |
1da177e4 | 5660 | |
f17c8607 | 5661 | if (len < cpumask_size()) |
1da177e4 LT |
5662 | return -EINVAL; |
5663 | ||
f17c8607 RR |
5664 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5665 | return -ENOMEM; | |
1da177e4 | 5666 | |
f17c8607 RR |
5667 | ret = sched_getaffinity(pid, mask); |
5668 | if (ret == 0) { | |
5669 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
5670 | ret = -EFAULT; | |
5671 | else | |
5672 | ret = cpumask_size(); | |
5673 | } | |
5674 | free_cpumask_var(mask); | |
1da177e4 | 5675 | |
f17c8607 | 5676 | return ret; |
1da177e4 LT |
5677 | } |
5678 | ||
5679 | /** | |
5680 | * sys_sched_yield - yield the current processor to other threads. | |
5681 | * | |
dd41f596 IM |
5682 | * This function yields the current CPU to other tasks. If there are no |
5683 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5684 | */ |
5add95d4 | 5685 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5686 | { |
70b97a7f | 5687 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5688 | |
2d72376b | 5689 | schedstat_inc(rq, yld_count); |
4530d7ab | 5690 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5691 | |
5692 | /* | |
5693 | * Since we are going to call schedule() anyway, there's | |
5694 | * no need to preempt or enable interrupts: | |
5695 | */ | |
5696 | __release(rq->lock); | |
8a25d5de | 5697 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5698 | _raw_spin_unlock(&rq->lock); |
5699 | preempt_enable_no_resched(); | |
5700 | ||
5701 | schedule(); | |
5702 | ||
5703 | return 0; | |
5704 | } | |
5705 | ||
e7b38404 | 5706 | static void __cond_resched(void) |
1da177e4 | 5707 | { |
8e0a43d8 IM |
5708 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5709 | __might_sleep(__FILE__, __LINE__); | |
5710 | #endif | |
5bbcfd90 IM |
5711 | /* |
5712 | * The BKS might be reacquired before we have dropped | |
5713 | * PREEMPT_ACTIVE, which could trigger a second | |
5714 | * cond_resched() call. | |
5715 | */ | |
1da177e4 LT |
5716 | do { |
5717 | add_preempt_count(PREEMPT_ACTIVE); | |
5718 | schedule(); | |
5719 | sub_preempt_count(PREEMPT_ACTIVE); | |
5720 | } while (need_resched()); | |
5721 | } | |
5722 | ||
02b67cc3 | 5723 | int __sched _cond_resched(void) |
1da177e4 | 5724 | { |
9414232f IM |
5725 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5726 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5727 | __cond_resched(); |
5728 | return 1; | |
5729 | } | |
5730 | return 0; | |
5731 | } | |
02b67cc3 | 5732 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5733 | |
5734 | /* | |
5735 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5736 | * call schedule, and on return reacquire the lock. | |
5737 | * | |
41a2d6cf | 5738 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5739 | * operations here to prevent schedule() from being called twice (once via |
5740 | * spin_unlock(), once by hand). | |
5741 | */ | |
95cdf3b7 | 5742 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5743 | { |
95c354fe | 5744 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5745 | int ret = 0; |
5746 | ||
95c354fe | 5747 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5748 | spin_unlock(lock); |
95c354fe NP |
5749 | if (resched && need_resched()) |
5750 | __cond_resched(); | |
5751 | else | |
5752 | cpu_relax(); | |
6df3cecb | 5753 | ret = 1; |
1da177e4 | 5754 | spin_lock(lock); |
1da177e4 | 5755 | } |
6df3cecb | 5756 | return ret; |
1da177e4 | 5757 | } |
1da177e4 LT |
5758 | EXPORT_SYMBOL(cond_resched_lock); |
5759 | ||
5760 | int __sched cond_resched_softirq(void) | |
5761 | { | |
5762 | BUG_ON(!in_softirq()); | |
5763 | ||
9414232f | 5764 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5765 | local_bh_enable(); |
1da177e4 LT |
5766 | __cond_resched(); |
5767 | local_bh_disable(); | |
5768 | return 1; | |
5769 | } | |
5770 | return 0; | |
5771 | } | |
1da177e4 LT |
5772 | EXPORT_SYMBOL(cond_resched_softirq); |
5773 | ||
1da177e4 LT |
5774 | /** |
5775 | * yield - yield the current processor to other threads. | |
5776 | * | |
72fd4a35 | 5777 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5778 | * thread runnable and calls sys_sched_yield(). |
5779 | */ | |
5780 | void __sched yield(void) | |
5781 | { | |
5782 | set_current_state(TASK_RUNNING); | |
5783 | sys_sched_yield(); | |
5784 | } | |
1da177e4 LT |
5785 | EXPORT_SYMBOL(yield); |
5786 | ||
5787 | /* | |
41a2d6cf | 5788 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5789 | * that process accounting knows that this is a task in IO wait state. |
5790 | * | |
5791 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5792 | * has set its backing_dev_info: the queue against which it should throttle) | |
5793 | */ | |
5794 | void __sched io_schedule(void) | |
5795 | { | |
70b97a7f | 5796 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5797 | |
0ff92245 | 5798 | delayacct_blkio_start(); |
1da177e4 LT |
5799 | atomic_inc(&rq->nr_iowait); |
5800 | schedule(); | |
5801 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5802 | delayacct_blkio_end(); |
1da177e4 | 5803 | } |
1da177e4 LT |
5804 | EXPORT_SYMBOL(io_schedule); |
5805 | ||
5806 | long __sched io_schedule_timeout(long timeout) | |
5807 | { | |
70b97a7f | 5808 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5809 | long ret; |
5810 | ||
0ff92245 | 5811 | delayacct_blkio_start(); |
1da177e4 LT |
5812 | atomic_inc(&rq->nr_iowait); |
5813 | ret = schedule_timeout(timeout); | |
5814 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5815 | delayacct_blkio_end(); |
1da177e4 LT |
5816 | return ret; |
5817 | } | |
5818 | ||
5819 | /** | |
5820 | * sys_sched_get_priority_max - return maximum RT priority. | |
5821 | * @policy: scheduling class. | |
5822 | * | |
5823 | * this syscall returns the maximum rt_priority that can be used | |
5824 | * by a given scheduling class. | |
5825 | */ | |
5add95d4 | 5826 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5827 | { |
5828 | int ret = -EINVAL; | |
5829 | ||
5830 | switch (policy) { | |
5831 | case SCHED_FIFO: | |
5832 | case SCHED_RR: | |
5833 | ret = MAX_USER_RT_PRIO-1; | |
5834 | break; | |
5835 | case SCHED_NORMAL: | |
b0a9499c | 5836 | case SCHED_BATCH: |
dd41f596 | 5837 | case SCHED_IDLE: |
1da177e4 LT |
5838 | ret = 0; |
5839 | break; | |
5840 | } | |
5841 | return ret; | |
5842 | } | |
5843 | ||
5844 | /** | |
5845 | * sys_sched_get_priority_min - return minimum RT priority. | |
5846 | * @policy: scheduling class. | |
5847 | * | |
5848 | * this syscall returns the minimum rt_priority that can be used | |
5849 | * by a given scheduling class. | |
5850 | */ | |
5add95d4 | 5851 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5852 | { |
5853 | int ret = -EINVAL; | |
5854 | ||
5855 | switch (policy) { | |
5856 | case SCHED_FIFO: | |
5857 | case SCHED_RR: | |
5858 | ret = 1; | |
5859 | break; | |
5860 | case SCHED_NORMAL: | |
b0a9499c | 5861 | case SCHED_BATCH: |
dd41f596 | 5862 | case SCHED_IDLE: |
1da177e4 LT |
5863 | ret = 0; |
5864 | } | |
5865 | return ret; | |
5866 | } | |
5867 | ||
5868 | /** | |
5869 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5870 | * @pid: pid of the process. | |
5871 | * @interval: userspace pointer to the timeslice value. | |
5872 | * | |
5873 | * this syscall writes the default timeslice value of a given process | |
5874 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5875 | */ | |
17da2bd9 | 5876 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5877 | struct timespec __user *, interval) |
1da177e4 | 5878 | { |
36c8b586 | 5879 | struct task_struct *p; |
a4ec24b4 | 5880 | unsigned int time_slice; |
3a5c359a | 5881 | int retval; |
1da177e4 | 5882 | struct timespec t; |
1da177e4 LT |
5883 | |
5884 | if (pid < 0) | |
3a5c359a | 5885 | return -EINVAL; |
1da177e4 LT |
5886 | |
5887 | retval = -ESRCH; | |
5888 | read_lock(&tasklist_lock); | |
5889 | p = find_process_by_pid(pid); | |
5890 | if (!p) | |
5891 | goto out_unlock; | |
5892 | ||
5893 | retval = security_task_getscheduler(p); | |
5894 | if (retval) | |
5895 | goto out_unlock; | |
5896 | ||
77034937 IM |
5897 | /* |
5898 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5899 | * tasks that are on an otherwise idle runqueue: | |
5900 | */ | |
5901 | time_slice = 0; | |
5902 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5903 | time_slice = DEF_TIMESLICE; |
1868f958 | 5904 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5905 | struct sched_entity *se = &p->se; |
5906 | unsigned long flags; | |
5907 | struct rq *rq; | |
5908 | ||
5909 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5910 | if (rq->cfs.load.weight) |
5911 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5912 | task_rq_unlock(rq, &flags); |
5913 | } | |
1da177e4 | 5914 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5915 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5916 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5917 | return retval; |
3a5c359a | 5918 | |
1da177e4 LT |
5919 | out_unlock: |
5920 | read_unlock(&tasklist_lock); | |
5921 | return retval; | |
5922 | } | |
5923 | ||
7c731e0a | 5924 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5925 | |
82a1fcb9 | 5926 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5927 | { |
1da177e4 | 5928 | unsigned long free = 0; |
36c8b586 | 5929 | unsigned state; |
1da177e4 | 5930 | |
1da177e4 | 5931 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5932 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5933 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5934 | #if BITS_PER_LONG == 32 |
1da177e4 | 5935 | if (state == TASK_RUNNING) |
cc4ea795 | 5936 | printk(KERN_CONT " running "); |
1da177e4 | 5937 | else |
cc4ea795 | 5938 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5939 | #else |
5940 | if (state == TASK_RUNNING) | |
cc4ea795 | 5941 | printk(KERN_CONT " running task "); |
1da177e4 | 5942 | else |
cc4ea795 | 5943 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5944 | #endif |
5945 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5946 | { | |
10ebffde | 5947 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5948 | while (!*n) |
5949 | n++; | |
10ebffde | 5950 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5951 | } |
5952 | #endif | |
ba25f9dc | 5953 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5954 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5955 | |
5fb5e6de | 5956 | show_stack(p, NULL); |
1da177e4 LT |
5957 | } |
5958 | ||
e59e2ae2 | 5959 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5960 | { |
36c8b586 | 5961 | struct task_struct *g, *p; |
1da177e4 | 5962 | |
4bd77321 IM |
5963 | #if BITS_PER_LONG == 32 |
5964 | printk(KERN_INFO | |
5965 | " task PC stack pid father\n"); | |
1da177e4 | 5966 | #else |
4bd77321 IM |
5967 | printk(KERN_INFO |
5968 | " task PC stack pid father\n"); | |
1da177e4 LT |
5969 | #endif |
5970 | read_lock(&tasklist_lock); | |
5971 | do_each_thread(g, p) { | |
5972 | /* | |
5973 | * reset the NMI-timeout, listing all files on a slow | |
5974 | * console might take alot of time: | |
5975 | */ | |
5976 | touch_nmi_watchdog(); | |
39bc89fd | 5977 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5978 | sched_show_task(p); |
1da177e4 LT |
5979 | } while_each_thread(g, p); |
5980 | ||
04c9167f JF |
5981 | touch_all_softlockup_watchdogs(); |
5982 | ||
dd41f596 IM |
5983 | #ifdef CONFIG_SCHED_DEBUG |
5984 | sysrq_sched_debug_show(); | |
5985 | #endif | |
1da177e4 | 5986 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5987 | /* |
5988 | * Only show locks if all tasks are dumped: | |
5989 | */ | |
5990 | if (state_filter == -1) | |
5991 | debug_show_all_locks(); | |
1da177e4 LT |
5992 | } |
5993 | ||
1df21055 IM |
5994 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5995 | { | |
dd41f596 | 5996 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5997 | } |
5998 | ||
f340c0d1 IM |
5999 | /** |
6000 | * init_idle - set up an idle thread for a given CPU | |
6001 | * @idle: task in question | |
6002 | * @cpu: cpu the idle task belongs to | |
6003 | * | |
6004 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6005 | * flag, to make booting more robust. | |
6006 | */ | |
5c1e1767 | 6007 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6008 | { |
70b97a7f | 6009 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6010 | unsigned long flags; |
6011 | ||
5cbd54ef IM |
6012 | spin_lock_irqsave(&rq->lock, flags); |
6013 | ||
dd41f596 IM |
6014 | __sched_fork(idle); |
6015 | idle->se.exec_start = sched_clock(); | |
6016 | ||
b29739f9 | 6017 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6018 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6019 | __set_task_cpu(idle, cpu); |
1da177e4 | 6020 | |
1da177e4 | 6021 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6022 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6023 | idle->oncpu = 1; | |
6024 | #endif | |
1da177e4 LT |
6025 | spin_unlock_irqrestore(&rq->lock, flags); |
6026 | ||
6027 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6028 | #if defined(CONFIG_PREEMPT) |
6029 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6030 | #else | |
a1261f54 | 6031 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6032 | #endif |
dd41f596 IM |
6033 | /* |
6034 | * The idle tasks have their own, simple scheduling class: | |
6035 | */ | |
6036 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6037 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6038 | } |
6039 | ||
6040 | /* | |
6041 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6042 | * indicates which cpus entered this state. This is used | |
6043 | * in the rcu update to wait only for active cpus. For system | |
6044 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6045 | * always be CPU_BITS_NONE. |
1da177e4 | 6046 | */ |
6a7b3dc3 | 6047 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6048 | |
19978ca6 IM |
6049 | /* |
6050 | * Increase the granularity value when there are more CPUs, | |
6051 | * because with more CPUs the 'effective latency' as visible | |
6052 | * to users decreases. But the relationship is not linear, | |
6053 | * so pick a second-best guess by going with the log2 of the | |
6054 | * number of CPUs. | |
6055 | * | |
6056 | * This idea comes from the SD scheduler of Con Kolivas: | |
6057 | */ | |
6058 | static inline void sched_init_granularity(void) | |
6059 | { | |
6060 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6061 | const unsigned long limit = 200000000; | |
6062 | ||
6063 | sysctl_sched_min_granularity *= factor; | |
6064 | if (sysctl_sched_min_granularity > limit) | |
6065 | sysctl_sched_min_granularity = limit; | |
6066 | ||
6067 | sysctl_sched_latency *= factor; | |
6068 | if (sysctl_sched_latency > limit) | |
6069 | sysctl_sched_latency = limit; | |
6070 | ||
6071 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6072 | |
6073 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6074 | } |
6075 | ||
1da177e4 LT |
6076 | #ifdef CONFIG_SMP |
6077 | /* | |
6078 | * This is how migration works: | |
6079 | * | |
70b97a7f | 6080 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6081 | * runqueue and wake up that CPU's migration thread. |
6082 | * 2) we down() the locked semaphore => thread blocks. | |
6083 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6084 | * thread off the CPU) | |
6085 | * 4) it gets the migration request and checks whether the migrated | |
6086 | * task is still in the wrong runqueue. | |
6087 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6088 | * it and puts it into the right queue. | |
6089 | * 6) migration thread up()s the semaphore. | |
6090 | * 7) we wake up and the migration is done. | |
6091 | */ | |
6092 | ||
6093 | /* | |
6094 | * Change a given task's CPU affinity. Migrate the thread to a | |
6095 | * proper CPU and schedule it away if the CPU it's executing on | |
6096 | * is removed from the allowed bitmask. | |
6097 | * | |
6098 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6099 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6100 | * call is not atomic; no spinlocks may be held. |
6101 | */ | |
96f874e2 | 6102 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6103 | { |
70b97a7f | 6104 | struct migration_req req; |
1da177e4 | 6105 | unsigned long flags; |
70b97a7f | 6106 | struct rq *rq; |
48f24c4d | 6107 | int ret = 0; |
1da177e4 LT |
6108 | |
6109 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6110 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6111 | ret = -EINVAL; |
6112 | goto out; | |
6113 | } | |
6114 | ||
9985b0ba | 6115 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6116 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6117 | ret = -EINVAL; |
6118 | goto out; | |
6119 | } | |
6120 | ||
73fe6aae | 6121 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6122 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6123 | else { |
96f874e2 RR |
6124 | cpumask_copy(&p->cpus_allowed, new_mask); |
6125 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6126 | } |
6127 | ||
1da177e4 | 6128 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6129 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6130 | goto out; |
6131 | ||
1e5ce4f4 | 6132 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6133 | /* Need help from migration thread: drop lock and wait. */ |
6134 | task_rq_unlock(rq, &flags); | |
6135 | wake_up_process(rq->migration_thread); | |
6136 | wait_for_completion(&req.done); | |
6137 | tlb_migrate_finish(p->mm); | |
6138 | return 0; | |
6139 | } | |
6140 | out: | |
6141 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6142 | |
1da177e4 LT |
6143 | return ret; |
6144 | } | |
cd8ba7cd | 6145 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6146 | |
6147 | /* | |
41a2d6cf | 6148 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6149 | * this because either it can't run here any more (set_cpus_allowed() |
6150 | * away from this CPU, or CPU going down), or because we're | |
6151 | * attempting to rebalance this task on exec (sched_exec). | |
6152 | * | |
6153 | * So we race with normal scheduler movements, but that's OK, as long | |
6154 | * as the task is no longer on this CPU. | |
efc30814 KK |
6155 | * |
6156 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6157 | */ |
efc30814 | 6158 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6159 | { |
70b97a7f | 6160 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6161 | int ret = 0, on_rq; |
1da177e4 | 6162 | |
e761b772 | 6163 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6164 | return ret; |
1da177e4 LT |
6165 | |
6166 | rq_src = cpu_rq(src_cpu); | |
6167 | rq_dest = cpu_rq(dest_cpu); | |
6168 | ||
6169 | double_rq_lock(rq_src, rq_dest); | |
6170 | /* Already moved. */ | |
6171 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6172 | goto done; |
1da177e4 | 6173 | /* Affinity changed (again). */ |
96f874e2 | 6174 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6175 | goto fail; |
1da177e4 | 6176 | |
dd41f596 | 6177 | on_rq = p->se.on_rq; |
6e82a3be | 6178 | if (on_rq) |
2e1cb74a | 6179 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6180 | |
1da177e4 | 6181 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6182 | if (on_rq) { |
6183 | activate_task(rq_dest, p, 0); | |
15afe09b | 6184 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6185 | } |
b1e38734 | 6186 | done: |
efc30814 | 6187 | ret = 1; |
b1e38734 | 6188 | fail: |
1da177e4 | 6189 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6190 | return ret; |
1da177e4 LT |
6191 | } |
6192 | ||
6193 | /* | |
6194 | * migration_thread - this is a highprio system thread that performs | |
6195 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6196 | * another runqueue. | |
6197 | */ | |
95cdf3b7 | 6198 | static int migration_thread(void *data) |
1da177e4 | 6199 | { |
1da177e4 | 6200 | int cpu = (long)data; |
70b97a7f | 6201 | struct rq *rq; |
1da177e4 LT |
6202 | |
6203 | rq = cpu_rq(cpu); | |
6204 | BUG_ON(rq->migration_thread != current); | |
6205 | ||
6206 | set_current_state(TASK_INTERRUPTIBLE); | |
6207 | while (!kthread_should_stop()) { | |
70b97a7f | 6208 | struct migration_req *req; |
1da177e4 | 6209 | struct list_head *head; |
1da177e4 | 6210 | |
1da177e4 LT |
6211 | spin_lock_irq(&rq->lock); |
6212 | ||
6213 | if (cpu_is_offline(cpu)) { | |
6214 | spin_unlock_irq(&rq->lock); | |
6215 | goto wait_to_die; | |
6216 | } | |
6217 | ||
6218 | if (rq->active_balance) { | |
6219 | active_load_balance(rq, cpu); | |
6220 | rq->active_balance = 0; | |
6221 | } | |
6222 | ||
6223 | head = &rq->migration_queue; | |
6224 | ||
6225 | if (list_empty(head)) { | |
6226 | spin_unlock_irq(&rq->lock); | |
6227 | schedule(); | |
6228 | set_current_state(TASK_INTERRUPTIBLE); | |
6229 | continue; | |
6230 | } | |
70b97a7f | 6231 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6232 | list_del_init(head->next); |
6233 | ||
674311d5 NP |
6234 | spin_unlock(&rq->lock); |
6235 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6236 | local_irq_enable(); | |
1da177e4 LT |
6237 | |
6238 | complete(&req->done); | |
6239 | } | |
6240 | __set_current_state(TASK_RUNNING); | |
6241 | return 0; | |
6242 | ||
6243 | wait_to_die: | |
6244 | /* Wait for kthread_stop */ | |
6245 | set_current_state(TASK_INTERRUPTIBLE); | |
6246 | while (!kthread_should_stop()) { | |
6247 | schedule(); | |
6248 | set_current_state(TASK_INTERRUPTIBLE); | |
6249 | } | |
6250 | __set_current_state(TASK_RUNNING); | |
6251 | return 0; | |
6252 | } | |
6253 | ||
6254 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6255 | |
6256 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6257 | { | |
6258 | int ret; | |
6259 | ||
6260 | local_irq_disable(); | |
6261 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6262 | local_irq_enable(); | |
6263 | return ret; | |
6264 | } | |
6265 | ||
054b9108 | 6266 | /* |
3a4fa0a2 | 6267 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 6268 | */ |
48f24c4d | 6269 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6270 | { |
70b97a7f | 6271 | int dest_cpu; |
6ca09dfc | 6272 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
6273 | |
6274 | again: | |
6275 | /* Look for allowed, online CPU in same node. */ | |
6276 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
6277 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
6278 | goto move; | |
6279 | ||
6280 | /* Any allowed, online CPU? */ | |
6281 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
6282 | if (dest_cpu < nr_cpu_ids) | |
6283 | goto move; | |
6284 | ||
6285 | /* No more Mr. Nice Guy. */ | |
6286 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
6287 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
6288 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 6289 | |
e76bd8d9 RR |
6290 | /* |
6291 | * Don't tell them about moving exiting tasks or | |
6292 | * kernel threads (both mm NULL), since they never | |
6293 | * leave kernel. | |
6294 | */ | |
6295 | if (p->mm && printk_ratelimit()) { | |
6296 | printk(KERN_INFO "process %d (%s) no " | |
6297 | "longer affine to cpu%d\n", | |
6298 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 6299 | } |
e76bd8d9 RR |
6300 | } |
6301 | ||
6302 | move: | |
6303 | /* It can have affinity changed while we were choosing. */ | |
6304 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
6305 | goto again; | |
1da177e4 LT |
6306 | } |
6307 | ||
6308 | /* | |
6309 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6310 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6311 | * for performance reasons the counter is not stricly tracking tasks to | |
6312 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6313 | * to keep the global sum constant after CPU-down: | |
6314 | */ | |
70b97a7f | 6315 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6316 | { |
1e5ce4f4 | 6317 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6318 | unsigned long flags; |
6319 | ||
6320 | local_irq_save(flags); | |
6321 | double_rq_lock(rq_src, rq_dest); | |
6322 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6323 | rq_src->nr_uninterruptible = 0; | |
6324 | double_rq_unlock(rq_src, rq_dest); | |
6325 | local_irq_restore(flags); | |
6326 | } | |
6327 | ||
6328 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6329 | static void migrate_live_tasks(int src_cpu) | |
6330 | { | |
48f24c4d | 6331 | struct task_struct *p, *t; |
1da177e4 | 6332 | |
f7b4cddc | 6333 | read_lock(&tasklist_lock); |
1da177e4 | 6334 | |
48f24c4d IM |
6335 | do_each_thread(t, p) { |
6336 | if (p == current) | |
1da177e4 LT |
6337 | continue; |
6338 | ||
48f24c4d IM |
6339 | if (task_cpu(p) == src_cpu) |
6340 | move_task_off_dead_cpu(src_cpu, p); | |
6341 | } while_each_thread(t, p); | |
1da177e4 | 6342 | |
f7b4cddc | 6343 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6344 | } |
6345 | ||
dd41f596 IM |
6346 | /* |
6347 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6348 | * It does so by boosting its priority to highest possible. |
6349 | * Used by CPU offline code. | |
1da177e4 LT |
6350 | */ |
6351 | void sched_idle_next(void) | |
6352 | { | |
48f24c4d | 6353 | int this_cpu = smp_processor_id(); |
70b97a7f | 6354 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6355 | struct task_struct *p = rq->idle; |
6356 | unsigned long flags; | |
6357 | ||
6358 | /* cpu has to be offline */ | |
48f24c4d | 6359 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6360 | |
48f24c4d IM |
6361 | /* |
6362 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6363 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6364 | */ |
6365 | spin_lock_irqsave(&rq->lock, flags); | |
6366 | ||
dd41f596 | 6367 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6368 | |
94bc9a7b DA |
6369 | update_rq_clock(rq); |
6370 | activate_task(rq, p, 0); | |
1da177e4 LT |
6371 | |
6372 | spin_unlock_irqrestore(&rq->lock, flags); | |
6373 | } | |
6374 | ||
48f24c4d IM |
6375 | /* |
6376 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6377 | * offline. |
6378 | */ | |
6379 | void idle_task_exit(void) | |
6380 | { | |
6381 | struct mm_struct *mm = current->active_mm; | |
6382 | ||
6383 | BUG_ON(cpu_online(smp_processor_id())); | |
6384 | ||
6385 | if (mm != &init_mm) | |
6386 | switch_mm(mm, &init_mm, current); | |
6387 | mmdrop(mm); | |
6388 | } | |
6389 | ||
054b9108 | 6390 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6391 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6392 | { |
70b97a7f | 6393 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6394 | |
6395 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6396 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6397 | |
6398 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6399 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6400 | |
48f24c4d | 6401 | get_task_struct(p); |
1da177e4 LT |
6402 | |
6403 | /* | |
6404 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6405 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6406 | * fine. |
6407 | */ | |
f7b4cddc | 6408 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6409 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6410 | spin_lock_irq(&rq->lock); |
1da177e4 | 6411 | |
48f24c4d | 6412 | put_task_struct(p); |
1da177e4 LT |
6413 | } |
6414 | ||
6415 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6416 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6417 | { | |
70b97a7f | 6418 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6419 | struct task_struct *next; |
48f24c4d | 6420 | |
dd41f596 IM |
6421 | for ( ; ; ) { |
6422 | if (!rq->nr_running) | |
6423 | break; | |
a8e504d2 | 6424 | update_rq_clock(rq); |
ff95f3df | 6425 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
6426 | if (!next) |
6427 | break; | |
79c53799 | 6428 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6429 | migrate_dead(dead_cpu, next); |
e692ab53 | 6430 | |
1da177e4 LT |
6431 | } |
6432 | } | |
6433 | #endif /* CONFIG_HOTPLUG_CPU */ | |
6434 | ||
e692ab53 NP |
6435 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6436 | ||
6437 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6438 | { |
6439 | .procname = "sched_domain", | |
c57baf1e | 6440 | .mode = 0555, |
e0361851 | 6441 | }, |
38605cae | 6442 | {0, }, |
e692ab53 NP |
6443 | }; |
6444 | ||
6445 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 6446 | { |
c57baf1e | 6447 | .ctl_name = CTL_KERN, |
e0361851 | 6448 | .procname = "kernel", |
c57baf1e | 6449 | .mode = 0555, |
e0361851 AD |
6450 | .child = sd_ctl_dir, |
6451 | }, | |
38605cae | 6452 | {0, }, |
e692ab53 NP |
6453 | }; |
6454 | ||
6455 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6456 | { | |
6457 | struct ctl_table *entry = | |
5cf9f062 | 6458 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6459 | |
e692ab53 NP |
6460 | return entry; |
6461 | } | |
6462 | ||
6382bc90 MM |
6463 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6464 | { | |
cd790076 | 6465 | struct ctl_table *entry; |
6382bc90 | 6466 | |
cd790076 MM |
6467 | /* |
6468 | * In the intermediate directories, both the child directory and | |
6469 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6470 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6471 | * static strings and all have proc handlers. |
6472 | */ | |
6473 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6474 | if (entry->child) |
6475 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6476 | if (entry->proc_handler == NULL) |
6477 | kfree(entry->procname); | |
6478 | } | |
6382bc90 MM |
6479 | |
6480 | kfree(*tablep); | |
6481 | *tablep = NULL; | |
6482 | } | |
6483 | ||
e692ab53 | 6484 | static void |
e0361851 | 6485 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6486 | const char *procname, void *data, int maxlen, |
6487 | mode_t mode, proc_handler *proc_handler) | |
6488 | { | |
e692ab53 NP |
6489 | entry->procname = procname; |
6490 | entry->data = data; | |
6491 | entry->maxlen = maxlen; | |
6492 | entry->mode = mode; | |
6493 | entry->proc_handler = proc_handler; | |
6494 | } | |
6495 | ||
6496 | static struct ctl_table * | |
6497 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6498 | { | |
a5d8c348 | 6499 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6500 | |
ad1cdc1d MM |
6501 | if (table == NULL) |
6502 | return NULL; | |
6503 | ||
e0361851 | 6504 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6505 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6506 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6507 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6508 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6509 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6510 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6511 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6512 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6513 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6514 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6515 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6516 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6517 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6518 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6519 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6520 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6521 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6522 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6523 | &sd->cache_nice_tries, |
6524 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6525 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6526 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6527 | set_table_entry(&table[11], "name", sd->name, |
6528 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6529 | /* &table[12] is terminator */ | |
e692ab53 NP |
6530 | |
6531 | return table; | |
6532 | } | |
6533 | ||
9a4e7159 | 6534 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6535 | { |
6536 | struct ctl_table *entry, *table; | |
6537 | struct sched_domain *sd; | |
6538 | int domain_num = 0, i; | |
6539 | char buf[32]; | |
6540 | ||
6541 | for_each_domain(cpu, sd) | |
6542 | domain_num++; | |
6543 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6544 | if (table == NULL) |
6545 | return NULL; | |
e692ab53 NP |
6546 | |
6547 | i = 0; | |
6548 | for_each_domain(cpu, sd) { | |
6549 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6550 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6551 | entry->mode = 0555; |
e692ab53 NP |
6552 | entry->child = sd_alloc_ctl_domain_table(sd); |
6553 | entry++; | |
6554 | i++; | |
6555 | } | |
6556 | return table; | |
6557 | } | |
6558 | ||
6559 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6560 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6561 | { |
6562 | int i, cpu_num = num_online_cpus(); | |
6563 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6564 | char buf[32]; | |
6565 | ||
7378547f MM |
6566 | WARN_ON(sd_ctl_dir[0].child); |
6567 | sd_ctl_dir[0].child = entry; | |
6568 | ||
ad1cdc1d MM |
6569 | if (entry == NULL) |
6570 | return; | |
6571 | ||
97b6ea7b | 6572 | for_each_online_cpu(i) { |
e692ab53 | 6573 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6574 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6575 | entry->mode = 0555; |
e692ab53 | 6576 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6577 | entry++; |
e692ab53 | 6578 | } |
7378547f MM |
6579 | |
6580 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6581 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6582 | } | |
6382bc90 | 6583 | |
7378547f | 6584 | /* may be called multiple times per register */ |
6382bc90 MM |
6585 | static void unregister_sched_domain_sysctl(void) |
6586 | { | |
7378547f MM |
6587 | if (sd_sysctl_header) |
6588 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6589 | sd_sysctl_header = NULL; |
7378547f MM |
6590 | if (sd_ctl_dir[0].child) |
6591 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6592 | } |
e692ab53 | 6593 | #else |
6382bc90 MM |
6594 | static void register_sched_domain_sysctl(void) |
6595 | { | |
6596 | } | |
6597 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6598 | { |
6599 | } | |
6600 | #endif | |
6601 | ||
1f11eb6a GH |
6602 | static void set_rq_online(struct rq *rq) |
6603 | { | |
6604 | if (!rq->online) { | |
6605 | const struct sched_class *class; | |
6606 | ||
c6c4927b | 6607 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6608 | rq->online = 1; |
6609 | ||
6610 | for_each_class(class) { | |
6611 | if (class->rq_online) | |
6612 | class->rq_online(rq); | |
6613 | } | |
6614 | } | |
6615 | } | |
6616 | ||
6617 | static void set_rq_offline(struct rq *rq) | |
6618 | { | |
6619 | if (rq->online) { | |
6620 | const struct sched_class *class; | |
6621 | ||
6622 | for_each_class(class) { | |
6623 | if (class->rq_offline) | |
6624 | class->rq_offline(rq); | |
6625 | } | |
6626 | ||
c6c4927b | 6627 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6628 | rq->online = 0; |
6629 | } | |
6630 | } | |
6631 | ||
1da177e4 LT |
6632 | /* |
6633 | * migration_call - callback that gets triggered when a CPU is added. | |
6634 | * Here we can start up the necessary migration thread for the new CPU. | |
6635 | */ | |
48f24c4d IM |
6636 | static int __cpuinit |
6637 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6638 | { |
1da177e4 | 6639 | struct task_struct *p; |
48f24c4d | 6640 | int cpu = (long)hcpu; |
1da177e4 | 6641 | unsigned long flags; |
70b97a7f | 6642 | struct rq *rq; |
1da177e4 LT |
6643 | |
6644 | switch (action) { | |
5be9361c | 6645 | |
1da177e4 | 6646 | case CPU_UP_PREPARE: |
8bb78442 | 6647 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6648 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6649 | if (IS_ERR(p)) |
6650 | return NOTIFY_BAD; | |
1da177e4 LT |
6651 | kthread_bind(p, cpu); |
6652 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6653 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6654 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6655 | task_rq_unlock(rq, &flags); |
6656 | cpu_rq(cpu)->migration_thread = p; | |
6657 | break; | |
48f24c4d | 6658 | |
1da177e4 | 6659 | case CPU_ONLINE: |
8bb78442 | 6660 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6661 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6662 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6663 | |
6664 | /* Update our root-domain */ | |
6665 | rq = cpu_rq(cpu); | |
6666 | spin_lock_irqsave(&rq->lock, flags); | |
6667 | if (rq->rd) { | |
c6c4927b | 6668 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6669 | |
6670 | set_rq_online(rq); | |
1f94ef59 GH |
6671 | } |
6672 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6673 | break; |
48f24c4d | 6674 | |
1da177e4 LT |
6675 | #ifdef CONFIG_HOTPLUG_CPU |
6676 | case CPU_UP_CANCELED: | |
8bb78442 | 6677 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6678 | if (!cpu_rq(cpu)->migration_thread) |
6679 | break; | |
41a2d6cf | 6680 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 6681 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 6682 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6683 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6684 | cpu_rq(cpu)->migration_thread = NULL; | |
6685 | break; | |
48f24c4d | 6686 | |
1da177e4 | 6687 | case CPU_DEAD: |
8bb78442 | 6688 | case CPU_DEAD_FROZEN: |
470fd646 | 6689 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6690 | migrate_live_tasks(cpu); |
6691 | rq = cpu_rq(cpu); | |
6692 | kthread_stop(rq->migration_thread); | |
6693 | rq->migration_thread = NULL; | |
6694 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6695 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6696 | update_rq_clock(rq); |
2e1cb74a | 6697 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6698 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6699 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6700 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6701 | migrate_dead_tasks(cpu); |
d2da272a | 6702 | spin_unlock_irq(&rq->lock); |
470fd646 | 6703 | cpuset_unlock(); |
1da177e4 LT |
6704 | migrate_nr_uninterruptible(rq); |
6705 | BUG_ON(rq->nr_running != 0); | |
6706 | ||
41a2d6cf IM |
6707 | /* |
6708 | * No need to migrate the tasks: it was best-effort if | |
6709 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6710 | * the requestors. | |
6711 | */ | |
1da177e4 LT |
6712 | spin_lock_irq(&rq->lock); |
6713 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6714 | struct migration_req *req; |
6715 | ||
1da177e4 | 6716 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6717 | struct migration_req, list); |
1da177e4 | 6718 | list_del_init(&req->list); |
9a2bd244 | 6719 | spin_unlock_irq(&rq->lock); |
1da177e4 | 6720 | complete(&req->done); |
9a2bd244 | 6721 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
6722 | } |
6723 | spin_unlock_irq(&rq->lock); | |
6724 | break; | |
57d885fe | 6725 | |
08f503b0 GH |
6726 | case CPU_DYING: |
6727 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6728 | /* Update our root-domain */ |
6729 | rq = cpu_rq(cpu); | |
6730 | spin_lock_irqsave(&rq->lock, flags); | |
6731 | if (rq->rd) { | |
c6c4927b | 6732 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6733 | set_rq_offline(rq); |
57d885fe GH |
6734 | } |
6735 | spin_unlock_irqrestore(&rq->lock, flags); | |
6736 | break; | |
1da177e4 LT |
6737 | #endif |
6738 | } | |
6739 | return NOTIFY_OK; | |
6740 | } | |
6741 | ||
6742 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6743 | * happens before everything else. | |
6744 | */ | |
26c2143b | 6745 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6746 | .notifier_call = migration_call, |
6747 | .priority = 10 | |
6748 | }; | |
6749 | ||
7babe8db | 6750 | static int __init migration_init(void) |
1da177e4 LT |
6751 | { |
6752 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6753 | int err; |
48f24c4d IM |
6754 | |
6755 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6756 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6757 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6758 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6759 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
6760 | |
6761 | return err; | |
1da177e4 | 6762 | } |
7babe8db | 6763 | early_initcall(migration_init); |
1da177e4 LT |
6764 | #endif |
6765 | ||
6766 | #ifdef CONFIG_SMP | |
476f3534 | 6767 | |
3e9830dc | 6768 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6769 | |
7c16ec58 | 6770 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6771 | struct cpumask *groupmask) |
1da177e4 | 6772 | { |
4dcf6aff | 6773 | struct sched_group *group = sd->groups; |
434d53b0 | 6774 | char str[256]; |
1da177e4 | 6775 | |
968ea6d8 | 6776 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6777 | cpumask_clear(groupmask); |
4dcf6aff IM |
6778 | |
6779 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6780 | ||
6781 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6782 | printk("does not load-balance\n"); | |
6783 | if (sd->parent) | |
6784 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6785 | " has parent"); | |
6786 | return -1; | |
41c7ce9a NP |
6787 | } |
6788 | ||
eefd796a | 6789 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6790 | |
758b2cdc | 6791 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
6792 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6793 | "CPU%d\n", cpu); | |
6794 | } | |
758b2cdc | 6795 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
6796 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6797 | " CPU%d\n", cpu); | |
6798 | } | |
1da177e4 | 6799 | |
4dcf6aff | 6800 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6801 | do { |
4dcf6aff IM |
6802 | if (!group) { |
6803 | printk("\n"); | |
6804 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6805 | break; |
6806 | } | |
6807 | ||
4dcf6aff IM |
6808 | if (!group->__cpu_power) { |
6809 | printk(KERN_CONT "\n"); | |
6810 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6811 | "set\n"); | |
6812 | break; | |
6813 | } | |
1da177e4 | 6814 | |
758b2cdc | 6815 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
6816 | printk(KERN_CONT "\n"); |
6817 | printk(KERN_ERR "ERROR: empty group\n"); | |
6818 | break; | |
6819 | } | |
1da177e4 | 6820 | |
758b2cdc | 6821 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
6822 | printk(KERN_CONT "\n"); |
6823 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6824 | break; | |
6825 | } | |
1da177e4 | 6826 | |
758b2cdc | 6827 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6828 | |
968ea6d8 | 6829 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
4dcf6aff | 6830 | printk(KERN_CONT " %s", str); |
1da177e4 | 6831 | |
4dcf6aff IM |
6832 | group = group->next; |
6833 | } while (group != sd->groups); | |
6834 | printk(KERN_CONT "\n"); | |
1da177e4 | 6835 | |
758b2cdc | 6836 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 6837 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6838 | |
758b2cdc RR |
6839 | if (sd->parent && |
6840 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
6841 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6842 | "of domain->span\n"); | |
6843 | return 0; | |
6844 | } | |
1da177e4 | 6845 | |
4dcf6aff IM |
6846 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6847 | { | |
d5dd3db1 | 6848 | cpumask_var_t groupmask; |
4dcf6aff | 6849 | int level = 0; |
1da177e4 | 6850 | |
4dcf6aff IM |
6851 | if (!sd) { |
6852 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6853 | return; | |
6854 | } | |
1da177e4 | 6855 | |
4dcf6aff IM |
6856 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6857 | ||
d5dd3db1 | 6858 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6859 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6860 | return; | |
6861 | } | |
6862 | ||
4dcf6aff | 6863 | for (;;) { |
7c16ec58 | 6864 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6865 | break; |
1da177e4 LT |
6866 | level++; |
6867 | sd = sd->parent; | |
33859f7f | 6868 | if (!sd) |
4dcf6aff IM |
6869 | break; |
6870 | } | |
d5dd3db1 | 6871 | free_cpumask_var(groupmask); |
1da177e4 | 6872 | } |
6d6bc0ad | 6873 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6874 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6875 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6876 | |
1a20ff27 | 6877 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6878 | { |
758b2cdc | 6879 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6880 | return 1; |
6881 | ||
6882 | /* Following flags need at least 2 groups */ | |
6883 | if (sd->flags & (SD_LOAD_BALANCE | | |
6884 | SD_BALANCE_NEWIDLE | | |
6885 | SD_BALANCE_FORK | | |
89c4710e SS |
6886 | SD_BALANCE_EXEC | |
6887 | SD_SHARE_CPUPOWER | | |
6888 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6889 | if (sd->groups != sd->groups->next) |
6890 | return 0; | |
6891 | } | |
6892 | ||
6893 | /* Following flags don't use groups */ | |
6894 | if (sd->flags & (SD_WAKE_IDLE | | |
6895 | SD_WAKE_AFFINE | | |
6896 | SD_WAKE_BALANCE)) | |
6897 | return 0; | |
6898 | ||
6899 | return 1; | |
6900 | } | |
6901 | ||
48f24c4d IM |
6902 | static int |
6903 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6904 | { |
6905 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6906 | ||
6907 | if (sd_degenerate(parent)) | |
6908 | return 1; | |
6909 | ||
758b2cdc | 6910 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6911 | return 0; |
6912 | ||
6913 | /* Does parent contain flags not in child? */ | |
6914 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6915 | if (cflags & SD_WAKE_AFFINE) | |
6916 | pflags &= ~SD_WAKE_BALANCE; | |
6917 | /* Flags needing groups don't count if only 1 group in parent */ | |
6918 | if (parent->groups == parent->groups->next) { | |
6919 | pflags &= ~(SD_LOAD_BALANCE | | |
6920 | SD_BALANCE_NEWIDLE | | |
6921 | SD_BALANCE_FORK | | |
89c4710e SS |
6922 | SD_BALANCE_EXEC | |
6923 | SD_SHARE_CPUPOWER | | |
6924 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6925 | if (nr_node_ids == 1) |
6926 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6927 | } |
6928 | if (~cflags & pflags) | |
6929 | return 0; | |
6930 | ||
6931 | return 1; | |
6932 | } | |
6933 | ||
c6c4927b RR |
6934 | static void free_rootdomain(struct root_domain *rd) |
6935 | { | |
68e74568 RR |
6936 | cpupri_cleanup(&rd->cpupri); |
6937 | ||
c6c4927b RR |
6938 | free_cpumask_var(rd->rto_mask); |
6939 | free_cpumask_var(rd->online); | |
6940 | free_cpumask_var(rd->span); | |
6941 | kfree(rd); | |
6942 | } | |
6943 | ||
57d885fe GH |
6944 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6945 | { | |
6946 | unsigned long flags; | |
57d885fe GH |
6947 | |
6948 | spin_lock_irqsave(&rq->lock, flags); | |
6949 | ||
6950 | if (rq->rd) { | |
6951 | struct root_domain *old_rd = rq->rd; | |
6952 | ||
c6c4927b | 6953 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6954 | set_rq_offline(rq); |
57d885fe | 6955 | |
c6c4927b | 6956 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6957 | |
57d885fe | 6958 | if (atomic_dec_and_test(&old_rd->refcount)) |
c6c4927b | 6959 | free_rootdomain(old_rd); |
57d885fe GH |
6960 | } |
6961 | ||
6962 | atomic_inc(&rd->refcount); | |
6963 | rq->rd = rd; | |
6964 | ||
c6c4927b RR |
6965 | cpumask_set_cpu(rq->cpu, rd->span); |
6966 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 6967 | set_rq_online(rq); |
57d885fe GH |
6968 | |
6969 | spin_unlock_irqrestore(&rq->lock, flags); | |
6970 | } | |
6971 | ||
db2f59c8 | 6972 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe GH |
6973 | { |
6974 | memset(rd, 0, sizeof(*rd)); | |
6975 | ||
c6c4927b RR |
6976 | if (bootmem) { |
6977 | alloc_bootmem_cpumask_var(&def_root_domain.span); | |
6978 | alloc_bootmem_cpumask_var(&def_root_domain.online); | |
6979 | alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); | |
68e74568 | 6980 | cpupri_init(&rd->cpupri, true); |
c6c4927b RR |
6981 | return 0; |
6982 | } | |
6983 | ||
6984 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | |
0c910d28 | 6985 | goto out; |
c6c4927b RR |
6986 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
6987 | goto free_span; | |
6988 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | |
6989 | goto free_online; | |
6e0534f2 | 6990 | |
68e74568 RR |
6991 | if (cpupri_init(&rd->cpupri, false) != 0) |
6992 | goto free_rto_mask; | |
c6c4927b | 6993 | return 0; |
6e0534f2 | 6994 | |
68e74568 RR |
6995 | free_rto_mask: |
6996 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6997 | free_online: |
6998 | free_cpumask_var(rd->online); | |
6999 | free_span: | |
7000 | free_cpumask_var(rd->span); | |
0c910d28 | 7001 | out: |
c6c4927b | 7002 | return -ENOMEM; |
57d885fe GH |
7003 | } |
7004 | ||
7005 | static void init_defrootdomain(void) | |
7006 | { | |
c6c4927b RR |
7007 | init_rootdomain(&def_root_domain, true); |
7008 | ||
57d885fe GH |
7009 | atomic_set(&def_root_domain.refcount, 1); |
7010 | } | |
7011 | ||
dc938520 | 7012 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7013 | { |
7014 | struct root_domain *rd; | |
7015 | ||
7016 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7017 | if (!rd) | |
7018 | return NULL; | |
7019 | ||
c6c4927b RR |
7020 | if (init_rootdomain(rd, false) != 0) { |
7021 | kfree(rd); | |
7022 | return NULL; | |
7023 | } | |
57d885fe GH |
7024 | |
7025 | return rd; | |
7026 | } | |
7027 | ||
1da177e4 | 7028 | /* |
0eab9146 | 7029 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7030 | * hold the hotplug lock. |
7031 | */ | |
0eab9146 IM |
7032 | static void |
7033 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7034 | { |
70b97a7f | 7035 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7036 | struct sched_domain *tmp; |
7037 | ||
7038 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7039 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7040 | struct sched_domain *parent = tmp->parent; |
7041 | if (!parent) | |
7042 | break; | |
f29c9b1c | 7043 | |
1a848870 | 7044 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7045 | tmp->parent = parent->parent; |
1a848870 SS |
7046 | if (parent->parent) |
7047 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7048 | } else |
7049 | tmp = tmp->parent; | |
245af2c7 SS |
7050 | } |
7051 | ||
1a848870 | 7052 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7053 | sd = sd->parent; |
1a848870 SS |
7054 | if (sd) |
7055 | sd->child = NULL; | |
7056 | } | |
1da177e4 LT |
7057 | |
7058 | sched_domain_debug(sd, cpu); | |
7059 | ||
57d885fe | 7060 | rq_attach_root(rq, rd); |
674311d5 | 7061 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7062 | } |
7063 | ||
7064 | /* cpus with isolated domains */ | |
dcc30a35 | 7065 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7066 | |
7067 | /* Setup the mask of cpus configured for isolated domains */ | |
7068 | static int __init isolated_cpu_setup(char *str) | |
7069 | { | |
968ea6d8 | 7070 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7071 | return 1; |
7072 | } | |
7073 | ||
8927f494 | 7074 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7075 | |
7076 | /* | |
6711cab4 SS |
7077 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7078 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7079 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7080 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7081 | * |
7082 | * init_sched_build_groups will build a circular linked list of the groups | |
7083 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7084 | * and ->cpu_power to 0. | |
7085 | */ | |
a616058b | 7086 | static void |
96f874e2 RR |
7087 | init_sched_build_groups(const struct cpumask *span, |
7088 | const struct cpumask *cpu_map, | |
7089 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7090 | struct sched_group **sg, |
96f874e2 RR |
7091 | struct cpumask *tmpmask), |
7092 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7093 | { |
7094 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7095 | int i; |
7096 | ||
96f874e2 | 7097 | cpumask_clear(covered); |
7c16ec58 | 7098 | |
abcd083a | 7099 | for_each_cpu(i, span) { |
6711cab4 | 7100 | struct sched_group *sg; |
7c16ec58 | 7101 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7102 | int j; |
7103 | ||
758b2cdc | 7104 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7105 | continue; |
7106 | ||
758b2cdc | 7107 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7108 | sg->__cpu_power = 0; |
1da177e4 | 7109 | |
abcd083a | 7110 | for_each_cpu(j, span) { |
7c16ec58 | 7111 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7112 | continue; |
7113 | ||
96f874e2 | 7114 | cpumask_set_cpu(j, covered); |
758b2cdc | 7115 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7116 | } |
7117 | if (!first) | |
7118 | first = sg; | |
7119 | if (last) | |
7120 | last->next = sg; | |
7121 | last = sg; | |
7122 | } | |
7123 | last->next = first; | |
7124 | } | |
7125 | ||
9c1cfda2 | 7126 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7127 | |
9c1cfda2 | 7128 | #ifdef CONFIG_NUMA |
198e2f18 | 7129 | |
9c1cfda2 JH |
7130 | /** |
7131 | * find_next_best_node - find the next node to include in a sched_domain | |
7132 | * @node: node whose sched_domain we're building | |
7133 | * @used_nodes: nodes already in the sched_domain | |
7134 | * | |
41a2d6cf | 7135 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7136 | * finds the closest node not already in the @used_nodes map. |
7137 | * | |
7138 | * Should use nodemask_t. | |
7139 | */ | |
c5f59f08 | 7140 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7141 | { |
7142 | int i, n, val, min_val, best_node = 0; | |
7143 | ||
7144 | min_val = INT_MAX; | |
7145 | ||
076ac2af | 7146 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7147 | /* Start at @node */ |
076ac2af | 7148 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7149 | |
7150 | if (!nr_cpus_node(n)) | |
7151 | continue; | |
7152 | ||
7153 | /* Skip already used nodes */ | |
c5f59f08 | 7154 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7155 | continue; |
7156 | ||
7157 | /* Simple min distance search */ | |
7158 | val = node_distance(node, n); | |
7159 | ||
7160 | if (val < min_val) { | |
7161 | min_val = val; | |
7162 | best_node = n; | |
7163 | } | |
7164 | } | |
7165 | ||
c5f59f08 | 7166 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7167 | return best_node; |
7168 | } | |
7169 | ||
7170 | /** | |
7171 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7172 | * @node: node whose cpumask we're constructing | |
73486722 | 7173 | * @span: resulting cpumask |
9c1cfda2 | 7174 | * |
41a2d6cf | 7175 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7176 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7177 | * out optimally. | |
7178 | */ | |
96f874e2 | 7179 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7180 | { |
c5f59f08 | 7181 | nodemask_t used_nodes; |
48f24c4d | 7182 | int i; |
9c1cfda2 | 7183 | |
6ca09dfc | 7184 | cpumask_clear(span); |
c5f59f08 | 7185 | nodes_clear(used_nodes); |
9c1cfda2 | 7186 | |
6ca09dfc | 7187 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7188 | node_set(node, used_nodes); |
9c1cfda2 JH |
7189 | |
7190 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7191 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7192 | |
6ca09dfc | 7193 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 7194 | } |
9c1cfda2 | 7195 | } |
6d6bc0ad | 7196 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7197 | |
5c45bf27 | 7198 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7199 | |
6c99e9ad RR |
7200 | /* |
7201 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
7202 | * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space | |
7203 | * for nr_cpu_ids < CONFIG_NR_CPUS. | |
7204 | */ | |
7205 | struct static_sched_group { | |
7206 | struct sched_group sg; | |
7207 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
7208 | }; | |
7209 | ||
7210 | struct static_sched_domain { | |
7211 | struct sched_domain sd; | |
7212 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
7213 | }; | |
7214 | ||
9c1cfda2 | 7215 | /* |
48f24c4d | 7216 | * SMT sched-domains: |
9c1cfda2 | 7217 | */ |
1da177e4 | 7218 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
7219 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
7220 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 7221 | |
41a2d6cf | 7222 | static int |
96f874e2 RR |
7223 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
7224 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 7225 | { |
6711cab4 | 7226 | if (sg) |
6c99e9ad | 7227 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
7228 | return cpu; |
7229 | } | |
6d6bc0ad | 7230 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7231 | |
48f24c4d IM |
7232 | /* |
7233 | * multi-core sched-domains: | |
7234 | */ | |
1e9f28fa | 7235 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
7236 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
7237 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 7238 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7239 | |
7240 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7241 | static int |
96f874e2 RR |
7242 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7243 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 7244 | { |
6711cab4 | 7245 | int group; |
7c16ec58 | 7246 | |
96f874e2 RR |
7247 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7248 | group = cpumask_first(mask); | |
6711cab4 | 7249 | if (sg) |
6c99e9ad | 7250 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 7251 | return group; |
1e9f28fa SS |
7252 | } |
7253 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7254 | static int |
96f874e2 RR |
7255 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7256 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 7257 | { |
6711cab4 | 7258 | if (sg) |
6c99e9ad | 7259 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
7260 | return cpu; |
7261 | } | |
7262 | #endif | |
7263 | ||
6c99e9ad RR |
7264 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
7265 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 7266 | |
41a2d6cf | 7267 | static int |
96f874e2 RR |
7268 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
7269 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 7270 | { |
6711cab4 | 7271 | int group; |
48f24c4d | 7272 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 7273 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 7274 | group = cpumask_first(mask); |
1e9f28fa | 7275 | #elif defined(CONFIG_SCHED_SMT) |
96f874e2 RR |
7276 | cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); |
7277 | group = cpumask_first(mask); | |
1da177e4 | 7278 | #else |
6711cab4 | 7279 | group = cpu; |
1da177e4 | 7280 | #endif |
6711cab4 | 7281 | if (sg) |
6c99e9ad | 7282 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 7283 | return group; |
1da177e4 LT |
7284 | } |
7285 | ||
7286 | #ifdef CONFIG_NUMA | |
1da177e4 | 7287 | /* |
9c1cfda2 JH |
7288 | * The init_sched_build_groups can't handle what we want to do with node |
7289 | * groups, so roll our own. Now each node has its own list of groups which | |
7290 | * gets dynamically allocated. | |
1da177e4 | 7291 | */ |
62ea9ceb | 7292 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 7293 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7294 | |
62ea9ceb | 7295 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 7296 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 7297 | |
96f874e2 RR |
7298 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
7299 | struct sched_group **sg, | |
7300 | struct cpumask *nodemask) | |
9c1cfda2 | 7301 | { |
6711cab4 SS |
7302 | int group; |
7303 | ||
6ca09dfc | 7304 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 7305 | group = cpumask_first(nodemask); |
6711cab4 SS |
7306 | |
7307 | if (sg) | |
6c99e9ad | 7308 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7309 | return group; |
1da177e4 | 7310 | } |
6711cab4 | 7311 | |
08069033 SS |
7312 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7313 | { | |
7314 | struct sched_group *sg = group_head; | |
7315 | int j; | |
7316 | ||
7317 | if (!sg) | |
7318 | return; | |
3a5c359a | 7319 | do { |
758b2cdc | 7320 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7321 | struct sched_domain *sd; |
08069033 | 7322 | |
6c99e9ad | 7323 | sd = &per_cpu(phys_domains, j).sd; |
758b2cdc | 7324 | if (j != cpumask_first(sched_group_cpus(sd->groups))) { |
3a5c359a AK |
7325 | /* |
7326 | * Only add "power" once for each | |
7327 | * physical package. | |
7328 | */ | |
7329 | continue; | |
7330 | } | |
08069033 | 7331 | |
3a5c359a AK |
7332 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7333 | } | |
7334 | sg = sg->next; | |
7335 | } while (sg != group_head); | |
08069033 | 7336 | } |
6d6bc0ad | 7337 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7338 | |
a616058b | 7339 | #ifdef CONFIG_NUMA |
51888ca2 | 7340 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7341 | static void free_sched_groups(const struct cpumask *cpu_map, |
7342 | struct cpumask *nodemask) | |
51888ca2 | 7343 | { |
a616058b | 7344 | int cpu, i; |
51888ca2 | 7345 | |
abcd083a | 7346 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7347 | struct sched_group **sched_group_nodes |
7348 | = sched_group_nodes_bycpu[cpu]; | |
7349 | ||
51888ca2 SV |
7350 | if (!sched_group_nodes) |
7351 | continue; | |
7352 | ||
076ac2af | 7353 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7354 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7355 | ||
6ca09dfc | 7356 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7357 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7358 | continue; |
7359 | ||
7360 | if (sg == NULL) | |
7361 | continue; | |
7362 | sg = sg->next; | |
7363 | next_sg: | |
7364 | oldsg = sg; | |
7365 | sg = sg->next; | |
7366 | kfree(oldsg); | |
7367 | if (oldsg != sched_group_nodes[i]) | |
7368 | goto next_sg; | |
7369 | } | |
7370 | kfree(sched_group_nodes); | |
7371 | sched_group_nodes_bycpu[cpu] = NULL; | |
7372 | } | |
51888ca2 | 7373 | } |
6d6bc0ad | 7374 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7375 | static void free_sched_groups(const struct cpumask *cpu_map, |
7376 | struct cpumask *nodemask) | |
a616058b SS |
7377 | { |
7378 | } | |
6d6bc0ad | 7379 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7380 | |
89c4710e SS |
7381 | /* |
7382 | * Initialize sched groups cpu_power. | |
7383 | * | |
7384 | * cpu_power indicates the capacity of sched group, which is used while | |
7385 | * distributing the load between different sched groups in a sched domain. | |
7386 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7387 | * there are asymmetries in the topology. If there are asymmetries, group | |
7388 | * having more cpu_power will pickup more load compared to the group having | |
7389 | * less cpu_power. | |
7390 | * | |
7391 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7392 | * the maximum number of tasks a group can handle in the presence of other idle | |
7393 | * or lightly loaded groups in the same sched domain. | |
7394 | */ | |
7395 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7396 | { | |
7397 | struct sched_domain *child; | |
7398 | struct sched_group *group; | |
7399 | ||
7400 | WARN_ON(!sd || !sd->groups); | |
7401 | ||
758b2cdc | 7402 | if (cpu != cpumask_first(sched_group_cpus(sd->groups))) |
89c4710e SS |
7403 | return; |
7404 | ||
7405 | child = sd->child; | |
7406 | ||
5517d86b ED |
7407 | sd->groups->__cpu_power = 0; |
7408 | ||
89c4710e SS |
7409 | /* |
7410 | * For perf policy, if the groups in child domain share resources | |
7411 | * (for example cores sharing some portions of the cache hierarchy | |
7412 | * or SMT), then set this domain groups cpu_power such that each group | |
7413 | * can handle only one task, when there are other idle groups in the | |
7414 | * same sched domain. | |
7415 | */ | |
7416 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
7417 | (child->flags & | |
7418 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 7419 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
7420 | return; |
7421 | } | |
7422 | ||
89c4710e SS |
7423 | /* |
7424 | * add cpu_power of each child group to this groups cpu_power | |
7425 | */ | |
7426 | group = child->groups; | |
7427 | do { | |
5517d86b | 7428 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
7429 | group = group->next; |
7430 | } while (group != child->groups); | |
7431 | } | |
7432 | ||
7c16ec58 MT |
7433 | /* |
7434 | * Initializers for schedule domains | |
7435 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7436 | */ | |
7437 | ||
a5d8c348 IM |
7438 | #ifdef CONFIG_SCHED_DEBUG |
7439 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7440 | #else | |
7441 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7442 | #endif | |
7443 | ||
7c16ec58 | 7444 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7445 | |
7c16ec58 MT |
7446 | #define SD_INIT_FUNC(type) \ |
7447 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7448 | { \ | |
7449 | memset(sd, 0, sizeof(*sd)); \ | |
7450 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7451 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7452 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7453 | } |
7454 | ||
7455 | SD_INIT_FUNC(CPU) | |
7456 | #ifdef CONFIG_NUMA | |
7457 | SD_INIT_FUNC(ALLNODES) | |
7458 | SD_INIT_FUNC(NODE) | |
7459 | #endif | |
7460 | #ifdef CONFIG_SCHED_SMT | |
7461 | SD_INIT_FUNC(SIBLING) | |
7462 | #endif | |
7463 | #ifdef CONFIG_SCHED_MC | |
7464 | SD_INIT_FUNC(MC) | |
7465 | #endif | |
7466 | ||
1d3504fc HS |
7467 | static int default_relax_domain_level = -1; |
7468 | ||
7469 | static int __init setup_relax_domain_level(char *str) | |
7470 | { | |
30e0e178 LZ |
7471 | unsigned long val; |
7472 | ||
7473 | val = simple_strtoul(str, NULL, 0); | |
7474 | if (val < SD_LV_MAX) | |
7475 | default_relax_domain_level = val; | |
7476 | ||
1d3504fc HS |
7477 | return 1; |
7478 | } | |
7479 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7480 | ||
7481 | static void set_domain_attribute(struct sched_domain *sd, | |
7482 | struct sched_domain_attr *attr) | |
7483 | { | |
7484 | int request; | |
7485 | ||
7486 | if (!attr || attr->relax_domain_level < 0) { | |
7487 | if (default_relax_domain_level < 0) | |
7488 | return; | |
7489 | else | |
7490 | request = default_relax_domain_level; | |
7491 | } else | |
7492 | request = attr->relax_domain_level; | |
7493 | if (request < sd->level) { | |
7494 | /* turn off idle balance on this domain */ | |
7495 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
7496 | } else { | |
7497 | /* turn on idle balance on this domain */ | |
7498 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
7499 | } | |
7500 | } | |
7501 | ||
1da177e4 | 7502 | /* |
1a20ff27 DG |
7503 | * Build sched domains for a given set of cpus and attach the sched domains |
7504 | * to the individual cpus | |
1da177e4 | 7505 | */ |
96f874e2 | 7506 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 7507 | struct sched_domain_attr *attr) |
1da177e4 | 7508 | { |
3404c8d9 | 7509 | int i, err = -ENOMEM; |
57d885fe | 7510 | struct root_domain *rd; |
3404c8d9 RR |
7511 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
7512 | tmpmask; | |
d1b55138 | 7513 | #ifdef CONFIG_NUMA |
3404c8d9 | 7514 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 7515 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 7516 | int sd_allnodes = 0; |
d1b55138 | 7517 | |
3404c8d9 RR |
7518 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
7519 | goto out; | |
7520 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
7521 | goto free_domainspan; | |
7522 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
7523 | goto free_covered; | |
7524 | #endif | |
7525 | ||
7526 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
7527 | goto free_notcovered; | |
7528 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
7529 | goto free_nodemask; | |
7530 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
7531 | goto free_this_sibling_map; | |
7532 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
7533 | goto free_this_core_map; | |
7534 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
7535 | goto free_send_covered; | |
7536 | ||
7537 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
7538 | /* |
7539 | * Allocate the per-node list of sched groups | |
7540 | */ | |
076ac2af | 7541 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 7542 | GFP_KERNEL); |
d1b55138 JH |
7543 | if (!sched_group_nodes) { |
7544 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 7545 | goto free_tmpmask; |
d1b55138 | 7546 | } |
d1b55138 | 7547 | #endif |
1da177e4 | 7548 | |
dc938520 | 7549 | rd = alloc_rootdomain(); |
57d885fe GH |
7550 | if (!rd) { |
7551 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 7552 | goto free_sched_groups; |
57d885fe GH |
7553 | } |
7554 | ||
7c16ec58 | 7555 | #ifdef CONFIG_NUMA |
96f874e2 | 7556 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
7557 | #endif |
7558 | ||
1da177e4 | 7559 | /* |
1a20ff27 | 7560 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7561 | */ |
abcd083a | 7562 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7563 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 7564 | |
6ca09dfc | 7565 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
7566 | |
7567 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
7568 | if (cpumask_weight(cpu_map) > |
7569 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 7570 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 7571 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7572 | set_domain_attribute(sd, attr); |
758b2cdc | 7573 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 7574 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7575 | p = sd; |
6711cab4 | 7576 | sd_allnodes = 1; |
9c1cfda2 JH |
7577 | } else |
7578 | p = NULL; | |
7579 | ||
62ea9ceb | 7580 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 7581 | SD_INIT(sd, NODE); |
1d3504fc | 7582 | set_domain_attribute(sd, attr); |
758b2cdc | 7583 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 7584 | sd->parent = p; |
1a848870 SS |
7585 | if (p) |
7586 | p->child = sd; | |
758b2cdc RR |
7587 | cpumask_and(sched_domain_span(sd), |
7588 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
7589 | #endif |
7590 | ||
7591 | p = sd; | |
6c99e9ad | 7592 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 7593 | SD_INIT(sd, CPU); |
1d3504fc | 7594 | set_domain_attribute(sd, attr); |
758b2cdc | 7595 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 7596 | sd->parent = p; |
1a848870 SS |
7597 | if (p) |
7598 | p->child = sd; | |
7c16ec58 | 7599 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7600 | |
1e9f28fa SS |
7601 | #ifdef CONFIG_SCHED_MC |
7602 | p = sd; | |
6c99e9ad | 7603 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 7604 | SD_INIT(sd, MC); |
1d3504fc | 7605 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
7606 | cpumask_and(sched_domain_span(sd), cpu_map, |
7607 | cpu_coregroup_mask(i)); | |
1e9f28fa | 7608 | sd->parent = p; |
1a848870 | 7609 | p->child = sd; |
7c16ec58 | 7610 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7611 | #endif |
7612 | ||
1da177e4 LT |
7613 | #ifdef CONFIG_SCHED_SMT |
7614 | p = sd; | |
6c99e9ad | 7615 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 7616 | SD_INIT(sd, SIBLING); |
1d3504fc | 7617 | set_domain_attribute(sd, attr); |
758b2cdc RR |
7618 | cpumask_and(sched_domain_span(sd), |
7619 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
1da177e4 | 7620 | sd->parent = p; |
1a848870 | 7621 | p->child = sd; |
7c16ec58 | 7622 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7623 | #endif |
7624 | } | |
7625 | ||
7626 | #ifdef CONFIG_SCHED_SMT | |
7627 | /* Set up CPU (sibling) groups */ | |
abcd083a | 7628 | for_each_cpu(i, cpu_map) { |
96f874e2 RR |
7629 | cpumask_and(this_sibling_map, |
7630 | &per_cpu(cpu_sibling_map, i), cpu_map); | |
7631 | if (i != cpumask_first(this_sibling_map)) | |
1da177e4 LT |
7632 | continue; |
7633 | ||
dd41f596 | 7634 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7635 | &cpu_to_cpu_group, |
7636 | send_covered, tmpmask); | |
1da177e4 LT |
7637 | } |
7638 | #endif | |
7639 | ||
1e9f28fa SS |
7640 | #ifdef CONFIG_SCHED_MC |
7641 | /* Set up multi-core groups */ | |
abcd083a | 7642 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 7643 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 7644 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 7645 | continue; |
7c16ec58 | 7646 | |
dd41f596 | 7647 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7648 | &cpu_to_core_group, |
7649 | send_covered, tmpmask); | |
1e9f28fa SS |
7650 | } |
7651 | #endif | |
7652 | ||
1da177e4 | 7653 | /* Set up physical groups */ |
076ac2af | 7654 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 7655 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7656 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
7657 | continue; |
7658 | ||
7c16ec58 MT |
7659 | init_sched_build_groups(nodemask, cpu_map, |
7660 | &cpu_to_phys_group, | |
7661 | send_covered, tmpmask); | |
1da177e4 LT |
7662 | } |
7663 | ||
7664 | #ifdef CONFIG_NUMA | |
7665 | /* Set up node groups */ | |
7c16ec58 | 7666 | if (sd_allnodes) { |
7c16ec58 MT |
7667 | init_sched_build_groups(cpu_map, cpu_map, |
7668 | &cpu_to_allnodes_group, | |
7669 | send_covered, tmpmask); | |
7670 | } | |
9c1cfda2 | 7671 | |
076ac2af | 7672 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
7673 | /* Set up node groups */ |
7674 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
7675 | int j; |
7676 | ||
96f874e2 | 7677 | cpumask_clear(covered); |
6ca09dfc | 7678 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7679 | if (cpumask_empty(nodemask)) { |
d1b55138 | 7680 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7681 | continue; |
d1b55138 | 7682 | } |
9c1cfda2 | 7683 | |
4bdbaad3 | 7684 | sched_domain_node_span(i, domainspan); |
96f874e2 | 7685 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 7686 | |
6c99e9ad RR |
7687 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
7688 | GFP_KERNEL, i); | |
51888ca2 SV |
7689 | if (!sg) { |
7690 | printk(KERN_WARNING "Can not alloc domain group for " | |
7691 | "node %d\n", i); | |
7692 | goto error; | |
7693 | } | |
9c1cfda2 | 7694 | sched_group_nodes[i] = sg; |
abcd083a | 7695 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 7696 | struct sched_domain *sd; |
9761eea8 | 7697 | |
62ea9ceb | 7698 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 7699 | sd->groups = sg; |
9c1cfda2 | 7700 | } |
5517d86b | 7701 | sg->__cpu_power = 0; |
758b2cdc | 7702 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 7703 | sg->next = sg; |
96f874e2 | 7704 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
7705 | prev = sg; |
7706 | ||
076ac2af | 7707 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 7708 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 7709 | |
96f874e2 RR |
7710 | cpumask_complement(notcovered, covered); |
7711 | cpumask_and(tmpmask, notcovered, cpu_map); | |
7712 | cpumask_and(tmpmask, tmpmask, domainspan); | |
7713 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
7714 | break; |
7715 | ||
6ca09dfc | 7716 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 7717 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
7718 | continue; |
7719 | ||
6c99e9ad RR |
7720 | sg = kmalloc_node(sizeof(struct sched_group) + |
7721 | cpumask_size(), | |
15f0b676 | 7722 | GFP_KERNEL, i); |
9c1cfda2 JH |
7723 | if (!sg) { |
7724 | printk(KERN_WARNING | |
7725 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7726 | goto error; |
9c1cfda2 | 7727 | } |
5517d86b | 7728 | sg->__cpu_power = 0; |
758b2cdc | 7729 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 7730 | sg->next = prev->next; |
96f874e2 | 7731 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
7732 | prev->next = sg; |
7733 | prev = sg; | |
7734 | } | |
9c1cfda2 | 7735 | } |
1da177e4 LT |
7736 | #endif |
7737 | ||
7738 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7739 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7740 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 7741 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 7742 | |
89c4710e | 7743 | init_sched_groups_power(i, sd); |
5c45bf27 | 7744 | } |
1da177e4 | 7745 | #endif |
1e9f28fa | 7746 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7747 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 7748 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 7749 | |
89c4710e | 7750 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7751 | } |
7752 | #endif | |
1e9f28fa | 7753 | |
abcd083a | 7754 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 7755 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 7756 | |
89c4710e | 7757 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7758 | } |
7759 | ||
9c1cfda2 | 7760 | #ifdef CONFIG_NUMA |
076ac2af | 7761 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 7762 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 7763 | |
6711cab4 SS |
7764 | if (sd_allnodes) { |
7765 | struct sched_group *sg; | |
f712c0c7 | 7766 | |
96f874e2 | 7767 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 7768 | tmpmask); |
f712c0c7 SS |
7769 | init_numa_sched_groups_power(sg); |
7770 | } | |
9c1cfda2 JH |
7771 | #endif |
7772 | ||
1da177e4 | 7773 | /* Attach the domains */ |
abcd083a | 7774 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
7775 | struct sched_domain *sd; |
7776 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 7777 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7778 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7779 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7780 | #else |
6c99e9ad | 7781 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7782 | #endif |
57d885fe | 7783 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7784 | } |
51888ca2 | 7785 | |
3404c8d9 RR |
7786 | err = 0; |
7787 | ||
7788 | free_tmpmask: | |
7789 | free_cpumask_var(tmpmask); | |
7790 | free_send_covered: | |
7791 | free_cpumask_var(send_covered); | |
7792 | free_this_core_map: | |
7793 | free_cpumask_var(this_core_map); | |
7794 | free_this_sibling_map: | |
7795 | free_cpumask_var(this_sibling_map); | |
7796 | free_nodemask: | |
7797 | free_cpumask_var(nodemask); | |
7798 | free_notcovered: | |
7799 | #ifdef CONFIG_NUMA | |
7800 | free_cpumask_var(notcovered); | |
7801 | free_covered: | |
7802 | free_cpumask_var(covered); | |
7803 | free_domainspan: | |
7804 | free_cpumask_var(domainspan); | |
7805 | out: | |
7806 | #endif | |
7807 | return err; | |
7808 | ||
7809 | free_sched_groups: | |
7810 | #ifdef CONFIG_NUMA | |
7811 | kfree(sched_group_nodes); | |
7812 | #endif | |
7813 | goto free_tmpmask; | |
51888ca2 | 7814 | |
a616058b | 7815 | #ifdef CONFIG_NUMA |
51888ca2 | 7816 | error: |
7c16ec58 | 7817 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 7818 | free_rootdomain(rd); |
3404c8d9 | 7819 | goto free_tmpmask; |
a616058b | 7820 | #endif |
1da177e4 | 7821 | } |
029190c5 | 7822 | |
96f874e2 | 7823 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7824 | { |
7825 | return __build_sched_domains(cpu_map, NULL); | |
7826 | } | |
7827 | ||
96f874e2 | 7828 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 7829 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7830 | static struct sched_domain_attr *dattr_cur; |
7831 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7832 | |
7833 | /* | |
7834 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7835 | * cpumask) fails, then fallback to a single sched domain, |
7836 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7837 | */ |
4212823f | 7838 | static cpumask_var_t fallback_doms; |
029190c5 | 7839 | |
ee79d1bd HC |
7840 | /* |
7841 | * arch_update_cpu_topology lets virtualized architectures update the | |
7842 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7843 | * or 0 if it stayed the same. | |
7844 | */ | |
7845 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7846 | { |
ee79d1bd | 7847 | return 0; |
22e52b07 HC |
7848 | } |
7849 | ||
1a20ff27 | 7850 | /* |
41a2d6cf | 7851 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7852 | * For now this just excludes isolated cpus, but could be used to |
7853 | * exclude other special cases in the future. | |
1a20ff27 | 7854 | */ |
96f874e2 | 7855 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7856 | { |
7378547f MM |
7857 | int err; |
7858 | ||
22e52b07 | 7859 | arch_update_cpu_topology(); |
029190c5 | 7860 | ndoms_cur = 1; |
96f874e2 | 7861 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 7862 | if (!doms_cur) |
4212823f | 7863 | doms_cur = fallback_doms; |
dcc30a35 | 7864 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 7865 | dattr_cur = NULL; |
7378547f | 7866 | err = build_sched_domains(doms_cur); |
6382bc90 | 7867 | register_sched_domain_sysctl(); |
7378547f MM |
7868 | |
7869 | return err; | |
1a20ff27 DG |
7870 | } |
7871 | ||
96f874e2 RR |
7872 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7873 | struct cpumask *tmpmask) | |
1da177e4 | 7874 | { |
7c16ec58 | 7875 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7876 | } |
1da177e4 | 7877 | |
1a20ff27 DG |
7878 | /* |
7879 | * Detach sched domains from a group of cpus specified in cpu_map | |
7880 | * These cpus will now be attached to the NULL domain | |
7881 | */ | |
96f874e2 | 7882 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7883 | { |
96f874e2 RR |
7884 | /* Save because hotplug lock held. */ |
7885 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7886 | int i; |
7887 | ||
abcd083a | 7888 | for_each_cpu(i, cpu_map) |
57d885fe | 7889 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7890 | synchronize_sched(); |
96f874e2 | 7891 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7892 | } |
7893 | ||
1d3504fc HS |
7894 | /* handle null as "default" */ |
7895 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7896 | struct sched_domain_attr *new, int idx_new) | |
7897 | { | |
7898 | struct sched_domain_attr tmp; | |
7899 | ||
7900 | /* fast path */ | |
7901 | if (!new && !cur) | |
7902 | return 1; | |
7903 | ||
7904 | tmp = SD_ATTR_INIT; | |
7905 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7906 | new ? (new + idx_new) : &tmp, | |
7907 | sizeof(struct sched_domain_attr)); | |
7908 | } | |
7909 | ||
029190c5 PJ |
7910 | /* |
7911 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7912 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7913 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7914 | * It destroys each deleted domain and builds each new domain. | |
7915 | * | |
96f874e2 | 7916 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
7917 | * The masks don't intersect (don't overlap.) We should setup one |
7918 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7919 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7920 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7921 | * it as it is. | |
7922 | * | |
41a2d6cf IM |
7923 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7924 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
7925 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
7926 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
7927 | * the single partition 'fallback_doms', it also forces the domains | |
7928 | * to be rebuilt. | |
029190c5 | 7929 | * |
96f874e2 | 7930 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7931 | * ndoms_new == 0 is a special case for destroying existing domains, |
7932 | * and it will not create the default domain. | |
dfb512ec | 7933 | * |
029190c5 PJ |
7934 | * Call with hotplug lock held |
7935 | */ | |
96f874e2 RR |
7936 | /* FIXME: Change to struct cpumask *doms_new[] */ |
7937 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 7938 | struct sched_domain_attr *dattr_new) |
029190c5 | 7939 | { |
dfb512ec | 7940 | int i, j, n; |
d65bd5ec | 7941 | int new_topology; |
029190c5 | 7942 | |
712555ee | 7943 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7944 | |
7378547f MM |
7945 | /* always unregister in case we don't destroy any domains */ |
7946 | unregister_sched_domain_sysctl(); | |
7947 | ||
d65bd5ec HC |
7948 | /* Let architecture update cpu core mappings. */ |
7949 | new_topology = arch_update_cpu_topology(); | |
7950 | ||
dfb512ec | 7951 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7952 | |
7953 | /* Destroy deleted domains */ | |
7954 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7955 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 7956 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 7957 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7958 | goto match1; |
7959 | } | |
7960 | /* no match - a current sched domain not in new doms_new[] */ | |
7961 | detach_destroy_domains(doms_cur + i); | |
7962 | match1: | |
7963 | ; | |
7964 | } | |
7965 | ||
e761b772 MK |
7966 | if (doms_new == NULL) { |
7967 | ndoms_cur = 0; | |
4212823f | 7968 | doms_new = fallback_doms; |
dcc30a35 | 7969 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 7970 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7971 | } |
7972 | ||
029190c5 PJ |
7973 | /* Build new domains */ |
7974 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7975 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 7976 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 7977 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7978 | goto match2; |
7979 | } | |
7980 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7981 | __build_sched_domains(doms_new + i, |
7982 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7983 | match2: |
7984 | ; | |
7985 | } | |
7986 | ||
7987 | /* Remember the new sched domains */ | |
4212823f | 7988 | if (doms_cur != fallback_doms) |
029190c5 | 7989 | kfree(doms_cur); |
1d3504fc | 7990 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7991 | doms_cur = doms_new; |
1d3504fc | 7992 | dattr_cur = dattr_new; |
029190c5 | 7993 | ndoms_cur = ndoms_new; |
7378547f MM |
7994 | |
7995 | register_sched_domain_sysctl(); | |
a1835615 | 7996 | |
712555ee | 7997 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7998 | } |
7999 | ||
5c45bf27 | 8000 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8001 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8002 | { |
95402b38 | 8003 | get_online_cpus(); |
dfb512ec MK |
8004 | |
8005 | /* Destroy domains first to force the rebuild */ | |
8006 | partition_sched_domains(0, NULL, NULL); | |
8007 | ||
e761b772 | 8008 | rebuild_sched_domains(); |
95402b38 | 8009 | put_online_cpus(); |
5c45bf27 SS |
8010 | } |
8011 | ||
8012 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8013 | { | |
afb8a9b7 | 8014 | unsigned int level = 0; |
5c45bf27 | 8015 | |
afb8a9b7 GS |
8016 | if (sscanf(buf, "%u", &level) != 1) |
8017 | return -EINVAL; | |
8018 | ||
8019 | /* | |
8020 | * level is always be positive so don't check for | |
8021 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8022 | * What happens on 0 or 1 byte write, | |
8023 | * need to check for count as well? | |
8024 | */ | |
8025 | ||
8026 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8027 | return -EINVAL; |
8028 | ||
8029 | if (smt) | |
afb8a9b7 | 8030 | sched_smt_power_savings = level; |
5c45bf27 | 8031 | else |
afb8a9b7 | 8032 | sched_mc_power_savings = level; |
5c45bf27 | 8033 | |
c70f22d2 | 8034 | arch_reinit_sched_domains(); |
5c45bf27 | 8035 | |
c70f22d2 | 8036 | return count; |
5c45bf27 SS |
8037 | } |
8038 | ||
5c45bf27 | 8039 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8040 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8041 | char *page) | |
5c45bf27 SS |
8042 | { |
8043 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8044 | } | |
f718cd4a | 8045 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8046 | const char *buf, size_t count) |
5c45bf27 SS |
8047 | { |
8048 | return sched_power_savings_store(buf, count, 0); | |
8049 | } | |
f718cd4a AK |
8050 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8051 | sched_mc_power_savings_show, | |
8052 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8053 | #endif |
8054 | ||
8055 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8056 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8057 | char *page) | |
5c45bf27 SS |
8058 | { |
8059 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8060 | } | |
f718cd4a | 8061 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8062 | const char *buf, size_t count) |
5c45bf27 SS |
8063 | { |
8064 | return sched_power_savings_store(buf, count, 1); | |
8065 | } | |
f718cd4a AK |
8066 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8067 | sched_smt_power_savings_show, | |
6707de00 AB |
8068 | sched_smt_power_savings_store); |
8069 | #endif | |
8070 | ||
39aac648 | 8071 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8072 | { |
8073 | int err = 0; | |
8074 | ||
8075 | #ifdef CONFIG_SCHED_SMT | |
8076 | if (smt_capable()) | |
8077 | err = sysfs_create_file(&cls->kset.kobj, | |
8078 | &attr_sched_smt_power_savings.attr); | |
8079 | #endif | |
8080 | #ifdef CONFIG_SCHED_MC | |
8081 | if (!err && mc_capable()) | |
8082 | err = sysfs_create_file(&cls->kset.kobj, | |
8083 | &attr_sched_mc_power_savings.attr); | |
8084 | #endif | |
8085 | return err; | |
8086 | } | |
6d6bc0ad | 8087 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8088 | |
e761b772 | 8089 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8090 | /* |
e761b772 MK |
8091 | * Add online and remove offline CPUs from the scheduler domains. |
8092 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8093 | */ |
8094 | static int update_sched_domains(struct notifier_block *nfb, | |
8095 | unsigned long action, void *hcpu) | |
e761b772 MK |
8096 | { |
8097 | switch (action) { | |
8098 | case CPU_ONLINE: | |
8099 | case CPU_ONLINE_FROZEN: | |
8100 | case CPU_DEAD: | |
8101 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8102 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8103 | return NOTIFY_OK; |
8104 | ||
8105 | default: | |
8106 | return NOTIFY_DONE; | |
8107 | } | |
8108 | } | |
8109 | #endif | |
8110 | ||
8111 | static int update_runtime(struct notifier_block *nfb, | |
8112 | unsigned long action, void *hcpu) | |
1da177e4 | 8113 | { |
7def2be1 PZ |
8114 | int cpu = (int)(long)hcpu; |
8115 | ||
1da177e4 | 8116 | switch (action) { |
1da177e4 | 8117 | case CPU_DOWN_PREPARE: |
8bb78442 | 8118 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8119 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8120 | return NOTIFY_OK; |
8121 | ||
1da177e4 | 8122 | case CPU_DOWN_FAILED: |
8bb78442 | 8123 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8124 | case CPU_ONLINE: |
8bb78442 | 8125 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8126 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8127 | return NOTIFY_OK; |
8128 | ||
1da177e4 LT |
8129 | default: |
8130 | return NOTIFY_DONE; | |
8131 | } | |
1da177e4 | 8132 | } |
1da177e4 LT |
8133 | |
8134 | void __init sched_init_smp(void) | |
8135 | { | |
dcc30a35 RR |
8136 | cpumask_var_t non_isolated_cpus; |
8137 | ||
8138 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8139 | |
434d53b0 MT |
8140 | #if defined(CONFIG_NUMA) |
8141 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8142 | GFP_KERNEL); | |
8143 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8144 | #endif | |
95402b38 | 8145 | get_online_cpus(); |
712555ee | 8146 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8147 | arch_init_sched_domains(cpu_online_mask); |
8148 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
8149 | if (cpumask_empty(non_isolated_cpus)) | |
8150 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8151 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8152 | put_online_cpus(); |
e761b772 MK |
8153 | |
8154 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8155 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8156 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8157 | #endif |
8158 | ||
8159 | /* RT runtime code needs to handle some hotplug events */ | |
8160 | hotcpu_notifier(update_runtime, 0); | |
8161 | ||
b328ca18 | 8162 | init_hrtick(); |
5c1e1767 NP |
8163 | |
8164 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8165 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8166 | BUG(); |
19978ca6 | 8167 | sched_init_granularity(); |
dcc30a35 | 8168 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
8169 | |
8170 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 8171 | init_sched_rt_class(); |
1da177e4 LT |
8172 | } |
8173 | #else | |
8174 | void __init sched_init_smp(void) | |
8175 | { | |
19978ca6 | 8176 | sched_init_granularity(); |
1da177e4 LT |
8177 | } |
8178 | #endif /* CONFIG_SMP */ | |
8179 | ||
8180 | int in_sched_functions(unsigned long addr) | |
8181 | { | |
1da177e4 LT |
8182 | return in_lock_functions(addr) || |
8183 | (addr >= (unsigned long)__sched_text_start | |
8184 | && addr < (unsigned long)__sched_text_end); | |
8185 | } | |
8186 | ||
a9957449 | 8187 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8188 | { |
8189 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8190 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8191 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8192 | cfs_rq->rq = rq; | |
8193 | #endif | |
67e9fb2a | 8194 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8195 | } |
8196 | ||
fa85ae24 PZ |
8197 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8198 | { | |
8199 | struct rt_prio_array *array; | |
8200 | int i; | |
8201 | ||
8202 | array = &rt_rq->active; | |
8203 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8204 | INIT_LIST_HEAD(array->queue + i); | |
8205 | __clear_bit(i, array->bitmap); | |
8206 | } | |
8207 | /* delimiter for bitsearch: */ | |
8208 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8209 | ||
052f1dc7 | 8210 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
8211 | rt_rq->highest_prio = MAX_RT_PRIO; |
8212 | #endif | |
fa85ae24 PZ |
8213 | #ifdef CONFIG_SMP |
8214 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
8215 | rt_rq->overloaded = 0; |
8216 | #endif | |
8217 | ||
8218 | rt_rq->rt_time = 0; | |
8219 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8220 | rt_rq->rt_runtime = 0; |
8221 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8222 | |
052f1dc7 | 8223 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8224 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8225 | rt_rq->rq = rq; |
8226 | #endif | |
fa85ae24 PZ |
8227 | } |
8228 | ||
6f505b16 | 8229 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8230 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8231 | struct sched_entity *se, int cpu, int add, | |
8232 | struct sched_entity *parent) | |
6f505b16 | 8233 | { |
ec7dc8ac | 8234 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8235 | tg->cfs_rq[cpu] = cfs_rq; |
8236 | init_cfs_rq(cfs_rq, rq); | |
8237 | cfs_rq->tg = tg; | |
8238 | if (add) | |
8239 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8240 | ||
8241 | tg->se[cpu] = se; | |
354d60c2 DG |
8242 | /* se could be NULL for init_task_group */ |
8243 | if (!se) | |
8244 | return; | |
8245 | ||
ec7dc8ac DG |
8246 | if (!parent) |
8247 | se->cfs_rq = &rq->cfs; | |
8248 | else | |
8249 | se->cfs_rq = parent->my_q; | |
8250 | ||
6f505b16 PZ |
8251 | se->my_q = cfs_rq; |
8252 | se->load.weight = tg->shares; | |
e05510d0 | 8253 | se->load.inv_weight = 0; |
ec7dc8ac | 8254 | se->parent = parent; |
6f505b16 | 8255 | } |
052f1dc7 | 8256 | #endif |
6f505b16 | 8257 | |
052f1dc7 | 8258 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8259 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8260 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8261 | struct sched_rt_entity *parent) | |
6f505b16 | 8262 | { |
ec7dc8ac DG |
8263 | struct rq *rq = cpu_rq(cpu); |
8264 | ||
6f505b16 PZ |
8265 | tg->rt_rq[cpu] = rt_rq; |
8266 | init_rt_rq(rt_rq, rq); | |
8267 | rt_rq->tg = tg; | |
8268 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8269 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8270 | if (add) |
8271 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8272 | ||
8273 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8274 | if (!rt_se) |
8275 | return; | |
8276 | ||
ec7dc8ac DG |
8277 | if (!parent) |
8278 | rt_se->rt_rq = &rq->rt; | |
8279 | else | |
8280 | rt_se->rt_rq = parent->my_q; | |
8281 | ||
6f505b16 | 8282 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8283 | rt_se->parent = parent; |
6f505b16 PZ |
8284 | INIT_LIST_HEAD(&rt_se->run_list); |
8285 | } | |
8286 | #endif | |
8287 | ||
1da177e4 LT |
8288 | void __init sched_init(void) |
8289 | { | |
dd41f596 | 8290 | int i, j; |
434d53b0 MT |
8291 | unsigned long alloc_size = 0, ptr; |
8292 | ||
8293 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8294 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8295 | #endif | |
8296 | #ifdef CONFIG_RT_GROUP_SCHED | |
8297 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8298 | #endif |
8299 | #ifdef CONFIG_USER_SCHED | |
8300 | alloc_size *= 2; | |
434d53b0 MT |
8301 | #endif |
8302 | /* | |
8303 | * As sched_init() is called before page_alloc is setup, | |
8304 | * we use alloc_bootmem(). | |
8305 | */ | |
8306 | if (alloc_size) { | |
5a9d3225 | 8307 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8308 | |
8309 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8310 | init_task_group.se = (struct sched_entity **)ptr; | |
8311 | ptr += nr_cpu_ids * sizeof(void **); | |
8312 | ||
8313 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8314 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8315 | |
8316 | #ifdef CONFIG_USER_SCHED | |
8317 | root_task_group.se = (struct sched_entity **)ptr; | |
8318 | ptr += nr_cpu_ids * sizeof(void **); | |
8319 | ||
8320 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8321 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8322 | #endif /* CONFIG_USER_SCHED */ |
8323 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8324 | #ifdef CONFIG_RT_GROUP_SCHED |
8325 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8326 | ptr += nr_cpu_ids * sizeof(void **); | |
8327 | ||
8328 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8329 | ptr += nr_cpu_ids * sizeof(void **); |
8330 | ||
8331 | #ifdef CONFIG_USER_SCHED | |
8332 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8333 | ptr += nr_cpu_ids * sizeof(void **); | |
8334 | ||
8335 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8336 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8337 | #endif /* CONFIG_USER_SCHED */ |
8338 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
434d53b0 | 8339 | } |
dd41f596 | 8340 | |
57d885fe GH |
8341 | #ifdef CONFIG_SMP |
8342 | init_defrootdomain(); | |
8343 | #endif | |
8344 | ||
d0b27fa7 PZ |
8345 | init_rt_bandwidth(&def_rt_bandwidth, |
8346 | global_rt_period(), global_rt_runtime()); | |
8347 | ||
8348 | #ifdef CONFIG_RT_GROUP_SCHED | |
8349 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8350 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8351 | #ifdef CONFIG_USER_SCHED |
8352 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8353 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8354 | #endif /* CONFIG_USER_SCHED */ |
8355 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8356 | |
052f1dc7 | 8357 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8358 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8359 | INIT_LIST_HEAD(&init_task_group.children); |
8360 | ||
8361 | #ifdef CONFIG_USER_SCHED | |
8362 | INIT_LIST_HEAD(&root_task_group.children); | |
8363 | init_task_group.parent = &root_task_group; | |
8364 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8365 | #endif /* CONFIG_USER_SCHED */ |
8366 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8367 | |
0a945022 | 8368 | for_each_possible_cpu(i) { |
70b97a7f | 8369 | struct rq *rq; |
1da177e4 LT |
8370 | |
8371 | rq = cpu_rq(i); | |
8372 | spin_lock_init(&rq->lock); | |
7897986b | 8373 | rq->nr_running = 0; |
dd41f596 | 8374 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8375 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8376 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8377 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8378 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8379 | #ifdef CONFIG_CGROUP_SCHED |
8380 | /* | |
8381 | * How much cpu bandwidth does init_task_group get? | |
8382 | * | |
8383 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8384 | * gets 100% of the cpu resources in the system. This overall | |
8385 | * system cpu resource is divided among the tasks of | |
8386 | * init_task_group and its child task-groups in a fair manner, | |
8387 | * based on each entity's (task or task-group's) weight | |
8388 | * (se->load.weight). | |
8389 | * | |
8390 | * In other words, if init_task_group has 10 tasks of weight | |
8391 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8392 | * then A0's share of the cpu resource is: | |
8393 | * | |
8394 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
8395 | * | |
8396 | * We achieve this by letting init_task_group's tasks sit | |
8397 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8398 | */ | |
ec7dc8ac | 8399 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 8400 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
8401 | root_task_group.shares = NICE_0_LOAD; |
8402 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
8403 | /* |
8404 | * In case of task-groups formed thr' the user id of tasks, | |
8405 | * init_task_group represents tasks belonging to root user. | |
8406 | * Hence it forms a sibling of all subsequent groups formed. | |
8407 | * In this case, init_task_group gets only a fraction of overall | |
8408 | * system cpu resource, based on the weight assigned to root | |
8409 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
8410 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
8411 | * (init_cfs_rq) and having one entity represent this group of | |
8412 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
8413 | */ | |
ec7dc8ac | 8414 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 8415 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
8416 | &per_cpu(init_sched_entity, i), i, 1, |
8417 | root_task_group.se[i]); | |
6f505b16 | 8418 | |
052f1dc7 | 8419 | #endif |
354d60c2 DG |
8420 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8421 | ||
8422 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8423 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8424 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8425 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8426 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8427 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 8428 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 8429 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 8430 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
8431 | &per_cpu(init_sched_rt_entity, i), i, 1, |
8432 | root_task_group.rt_se[i]); | |
354d60c2 | 8433 | #endif |
dd41f596 | 8434 | #endif |
1da177e4 | 8435 | |
dd41f596 IM |
8436 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8437 | rq->cpu_load[j] = 0; | |
1da177e4 | 8438 | #ifdef CONFIG_SMP |
41c7ce9a | 8439 | rq->sd = NULL; |
57d885fe | 8440 | rq->rd = NULL; |
1da177e4 | 8441 | rq->active_balance = 0; |
dd41f596 | 8442 | rq->next_balance = jiffies; |
1da177e4 | 8443 | rq->push_cpu = 0; |
0a2966b4 | 8444 | rq->cpu = i; |
1f11eb6a | 8445 | rq->online = 0; |
1da177e4 LT |
8446 | rq->migration_thread = NULL; |
8447 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 8448 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 8449 | #endif |
8f4d37ec | 8450 | init_rq_hrtick(rq); |
1da177e4 | 8451 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8452 | } |
8453 | ||
2dd73a4f | 8454 | set_load_weight(&init_task); |
b50f60ce | 8455 | |
e107be36 AK |
8456 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8457 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8458 | #endif | |
8459 | ||
c9819f45 | 8460 | #ifdef CONFIG_SMP |
962cf36c | 8461 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8462 | #endif |
8463 | ||
b50f60ce HC |
8464 | #ifdef CONFIG_RT_MUTEXES |
8465 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
8466 | #endif | |
8467 | ||
1da177e4 LT |
8468 | /* |
8469 | * The boot idle thread does lazy MMU switching as well: | |
8470 | */ | |
8471 | atomic_inc(&init_mm.mm_count); | |
8472 | enter_lazy_tlb(&init_mm, current); | |
8473 | ||
8474 | /* | |
8475 | * Make us the idle thread. Technically, schedule() should not be | |
8476 | * called from this thread, however somewhere below it might be, | |
8477 | * but because we are the idle thread, we just pick up running again | |
8478 | * when this runqueue becomes "idle". | |
8479 | */ | |
8480 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
8481 | /* |
8482 | * During early bootup we pretend to be a normal task: | |
8483 | */ | |
8484 | current->sched_class = &fair_sched_class; | |
6892b75e | 8485 | |
6a7b3dc3 RR |
8486 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
8487 | alloc_bootmem_cpumask_var(&nohz_cpu_mask); | |
bf4d83f6 | 8488 | #ifdef CONFIG_SMP |
7d1e6a9b RR |
8489 | #ifdef CONFIG_NO_HZ |
8490 | alloc_bootmem_cpumask_var(&nohz.cpu_mask); | |
8491 | #endif | |
dcc30a35 | 8492 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
bf4d83f6 | 8493 | #endif /* SMP */ |
6a7b3dc3 | 8494 | |
6892b75e | 8495 | scheduler_running = 1; |
1da177e4 LT |
8496 | } |
8497 | ||
8498 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
8499 | void __might_sleep(char *file, int line) | |
8500 | { | |
48f24c4d | 8501 | #ifdef in_atomic |
1da177e4 LT |
8502 | static unsigned long prev_jiffy; /* ratelimiting */ |
8503 | ||
aef745fc IM |
8504 | if ((!in_atomic() && !irqs_disabled()) || |
8505 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
8506 | return; | |
8507 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8508 | return; | |
8509 | prev_jiffy = jiffies; | |
8510 | ||
8511 | printk(KERN_ERR | |
8512 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8513 | file, line); | |
8514 | printk(KERN_ERR | |
8515 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8516 | in_atomic(), irqs_disabled(), | |
8517 | current->pid, current->comm); | |
8518 | ||
8519 | debug_show_held_locks(current); | |
8520 | if (irqs_disabled()) | |
8521 | print_irqtrace_events(current); | |
8522 | dump_stack(); | |
1da177e4 LT |
8523 | #endif |
8524 | } | |
8525 | EXPORT_SYMBOL(__might_sleep); | |
8526 | #endif | |
8527 | ||
8528 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8529 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8530 | { | |
8531 | int on_rq; | |
3e51f33f | 8532 | |
3a5e4dc1 AK |
8533 | update_rq_clock(rq); |
8534 | on_rq = p->se.on_rq; | |
8535 | if (on_rq) | |
8536 | deactivate_task(rq, p, 0); | |
8537 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8538 | if (on_rq) { | |
8539 | activate_task(rq, p, 0); | |
8540 | resched_task(rq->curr); | |
8541 | } | |
8542 | } | |
8543 | ||
1da177e4 LT |
8544 | void normalize_rt_tasks(void) |
8545 | { | |
a0f98a1c | 8546 | struct task_struct *g, *p; |
1da177e4 | 8547 | unsigned long flags; |
70b97a7f | 8548 | struct rq *rq; |
1da177e4 | 8549 | |
4cf5d77a | 8550 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8551 | do_each_thread(g, p) { |
178be793 IM |
8552 | /* |
8553 | * Only normalize user tasks: | |
8554 | */ | |
8555 | if (!p->mm) | |
8556 | continue; | |
8557 | ||
6cfb0d5d | 8558 | p->se.exec_start = 0; |
6cfb0d5d | 8559 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 8560 | p->se.wait_start = 0; |
dd41f596 | 8561 | p->se.sleep_start = 0; |
dd41f596 | 8562 | p->se.block_start = 0; |
6cfb0d5d | 8563 | #endif |
dd41f596 IM |
8564 | |
8565 | if (!rt_task(p)) { | |
8566 | /* | |
8567 | * Renice negative nice level userspace | |
8568 | * tasks back to 0: | |
8569 | */ | |
8570 | if (TASK_NICE(p) < 0 && p->mm) | |
8571 | set_user_nice(p, 0); | |
1da177e4 | 8572 | continue; |
dd41f596 | 8573 | } |
1da177e4 | 8574 | |
4cf5d77a | 8575 | spin_lock(&p->pi_lock); |
b29739f9 | 8576 | rq = __task_rq_lock(p); |
1da177e4 | 8577 | |
178be793 | 8578 | normalize_task(rq, p); |
3a5e4dc1 | 8579 | |
b29739f9 | 8580 | __task_rq_unlock(rq); |
4cf5d77a | 8581 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8582 | } while_each_thread(g, p); |
8583 | ||
4cf5d77a | 8584 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8585 | } |
8586 | ||
8587 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
8588 | |
8589 | #ifdef CONFIG_IA64 | |
8590 | /* | |
8591 | * These functions are only useful for the IA64 MCA handling. | |
8592 | * | |
8593 | * They can only be called when the whole system has been | |
8594 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8595 | * activity can take place. Using them for anything else would | |
8596 | * be a serious bug, and as a result, they aren't even visible | |
8597 | * under any other configuration. | |
8598 | */ | |
8599 | ||
8600 | /** | |
8601 | * curr_task - return the current task for a given cpu. | |
8602 | * @cpu: the processor in question. | |
8603 | * | |
8604 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8605 | */ | |
36c8b586 | 8606 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8607 | { |
8608 | return cpu_curr(cpu); | |
8609 | } | |
8610 | ||
8611 | /** | |
8612 | * set_curr_task - set the current task for a given cpu. | |
8613 | * @cpu: the processor in question. | |
8614 | * @p: the task pointer to set. | |
8615 | * | |
8616 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8617 | * are serviced on a separate stack. It allows the architecture to switch the |
8618 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8619 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8620 | * and caller must save the original value of the current task (see | |
8621 | * curr_task() above) and restore that value before reenabling interrupts and | |
8622 | * re-starting the system. | |
8623 | * | |
8624 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8625 | */ | |
36c8b586 | 8626 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8627 | { |
8628 | cpu_curr(cpu) = p; | |
8629 | } | |
8630 | ||
8631 | #endif | |
29f59db3 | 8632 | |
bccbe08a PZ |
8633 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8634 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8635 | { |
8636 | int i; | |
8637 | ||
8638 | for_each_possible_cpu(i) { | |
8639 | if (tg->cfs_rq) | |
8640 | kfree(tg->cfs_rq[i]); | |
8641 | if (tg->se) | |
8642 | kfree(tg->se[i]); | |
6f505b16 PZ |
8643 | } |
8644 | ||
8645 | kfree(tg->cfs_rq); | |
8646 | kfree(tg->se); | |
6f505b16 PZ |
8647 | } |
8648 | ||
ec7dc8ac DG |
8649 | static |
8650 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8651 | { |
29f59db3 | 8652 | struct cfs_rq *cfs_rq; |
eab17229 | 8653 | struct sched_entity *se; |
9b5b7751 | 8654 | struct rq *rq; |
29f59db3 SV |
8655 | int i; |
8656 | ||
434d53b0 | 8657 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8658 | if (!tg->cfs_rq) |
8659 | goto err; | |
434d53b0 | 8660 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8661 | if (!tg->se) |
8662 | goto err; | |
052f1dc7 PZ |
8663 | |
8664 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8665 | |
8666 | for_each_possible_cpu(i) { | |
9b5b7751 | 8667 | rq = cpu_rq(i); |
29f59db3 | 8668 | |
eab17229 LZ |
8669 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8670 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8671 | if (!cfs_rq) |
8672 | goto err; | |
8673 | ||
eab17229 LZ |
8674 | se = kzalloc_node(sizeof(struct sched_entity), |
8675 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8676 | if (!se) |
8677 | goto err; | |
8678 | ||
eab17229 | 8679 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
8680 | } |
8681 | ||
8682 | return 1; | |
8683 | ||
8684 | err: | |
8685 | return 0; | |
8686 | } | |
8687 | ||
8688 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8689 | { | |
8690 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8691 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8692 | } | |
8693 | ||
8694 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8695 | { | |
8696 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8697 | } | |
6d6bc0ad | 8698 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8699 | static inline void free_fair_sched_group(struct task_group *tg) |
8700 | { | |
8701 | } | |
8702 | ||
ec7dc8ac DG |
8703 | static inline |
8704 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8705 | { |
8706 | return 1; | |
8707 | } | |
8708 | ||
8709 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8710 | { | |
8711 | } | |
8712 | ||
8713 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8714 | { | |
8715 | } | |
6d6bc0ad | 8716 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8717 | |
8718 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8719 | static void free_rt_sched_group(struct task_group *tg) |
8720 | { | |
8721 | int i; | |
8722 | ||
d0b27fa7 PZ |
8723 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8724 | ||
bccbe08a PZ |
8725 | for_each_possible_cpu(i) { |
8726 | if (tg->rt_rq) | |
8727 | kfree(tg->rt_rq[i]); | |
8728 | if (tg->rt_se) | |
8729 | kfree(tg->rt_se[i]); | |
8730 | } | |
8731 | ||
8732 | kfree(tg->rt_rq); | |
8733 | kfree(tg->rt_se); | |
8734 | } | |
8735 | ||
ec7dc8ac DG |
8736 | static |
8737 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8738 | { |
8739 | struct rt_rq *rt_rq; | |
eab17229 | 8740 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8741 | struct rq *rq; |
8742 | int i; | |
8743 | ||
434d53b0 | 8744 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8745 | if (!tg->rt_rq) |
8746 | goto err; | |
434d53b0 | 8747 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8748 | if (!tg->rt_se) |
8749 | goto err; | |
8750 | ||
d0b27fa7 PZ |
8751 | init_rt_bandwidth(&tg->rt_bandwidth, |
8752 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8753 | |
8754 | for_each_possible_cpu(i) { | |
8755 | rq = cpu_rq(i); | |
8756 | ||
eab17229 LZ |
8757 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8758 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8759 | if (!rt_rq) |
8760 | goto err; | |
29f59db3 | 8761 | |
eab17229 LZ |
8762 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8763 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8764 | if (!rt_se) |
8765 | goto err; | |
29f59db3 | 8766 | |
eab17229 | 8767 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8768 | } |
8769 | ||
bccbe08a PZ |
8770 | return 1; |
8771 | ||
8772 | err: | |
8773 | return 0; | |
8774 | } | |
8775 | ||
8776 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8777 | { | |
8778 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8779 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8780 | } | |
8781 | ||
8782 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8783 | { | |
8784 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8785 | } | |
6d6bc0ad | 8786 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8787 | static inline void free_rt_sched_group(struct task_group *tg) |
8788 | { | |
8789 | } | |
8790 | ||
ec7dc8ac DG |
8791 | static inline |
8792 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8793 | { |
8794 | return 1; | |
8795 | } | |
8796 | ||
8797 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8798 | { | |
8799 | } | |
8800 | ||
8801 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8802 | { | |
8803 | } | |
6d6bc0ad | 8804 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8805 | |
d0b27fa7 | 8806 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8807 | static void free_sched_group(struct task_group *tg) |
8808 | { | |
8809 | free_fair_sched_group(tg); | |
8810 | free_rt_sched_group(tg); | |
8811 | kfree(tg); | |
8812 | } | |
8813 | ||
8814 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8815 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8816 | { |
8817 | struct task_group *tg; | |
8818 | unsigned long flags; | |
8819 | int i; | |
8820 | ||
8821 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8822 | if (!tg) | |
8823 | return ERR_PTR(-ENOMEM); | |
8824 | ||
ec7dc8ac | 8825 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8826 | goto err; |
8827 | ||
ec7dc8ac | 8828 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8829 | goto err; |
8830 | ||
8ed36996 | 8831 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8832 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8833 | register_fair_sched_group(tg, i); |
8834 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8835 | } |
6f505b16 | 8836 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8837 | |
8838 | WARN_ON(!parent); /* root should already exist */ | |
8839 | ||
8840 | tg->parent = parent; | |
f473aa5e | 8841 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8842 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8843 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8844 | |
9b5b7751 | 8845 | return tg; |
29f59db3 SV |
8846 | |
8847 | err: | |
6f505b16 | 8848 | free_sched_group(tg); |
29f59db3 SV |
8849 | return ERR_PTR(-ENOMEM); |
8850 | } | |
8851 | ||
9b5b7751 | 8852 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8853 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8854 | { |
29f59db3 | 8855 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8856 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8857 | } |
8858 | ||
9b5b7751 | 8859 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8860 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8861 | { |
8ed36996 | 8862 | unsigned long flags; |
9b5b7751 | 8863 | int i; |
29f59db3 | 8864 | |
8ed36996 | 8865 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8866 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8867 | unregister_fair_sched_group(tg, i); |
8868 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8869 | } |
6f505b16 | 8870 | list_del_rcu(&tg->list); |
f473aa5e | 8871 | list_del_rcu(&tg->siblings); |
8ed36996 | 8872 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8873 | |
9b5b7751 | 8874 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8875 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8876 | } |
8877 | ||
9b5b7751 | 8878 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8879 | * The caller of this function should have put the task in its new group |
8880 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8881 | * reflect its new group. | |
9b5b7751 SV |
8882 | */ |
8883 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8884 | { |
8885 | int on_rq, running; | |
8886 | unsigned long flags; | |
8887 | struct rq *rq; | |
8888 | ||
8889 | rq = task_rq_lock(tsk, &flags); | |
8890 | ||
29f59db3 SV |
8891 | update_rq_clock(rq); |
8892 | ||
051a1d1a | 8893 | running = task_current(rq, tsk); |
29f59db3 SV |
8894 | on_rq = tsk->se.on_rq; |
8895 | ||
0e1f3483 | 8896 | if (on_rq) |
29f59db3 | 8897 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8898 | if (unlikely(running)) |
8899 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8900 | |
6f505b16 | 8901 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8902 | |
810b3817 PZ |
8903 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8904 | if (tsk->sched_class->moved_group) | |
8905 | tsk->sched_class->moved_group(tsk); | |
8906 | #endif | |
8907 | ||
0e1f3483 HS |
8908 | if (unlikely(running)) |
8909 | tsk->sched_class->set_curr_task(rq); | |
8910 | if (on_rq) | |
7074badb | 8911 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8912 | |
29f59db3 SV |
8913 | task_rq_unlock(rq, &flags); |
8914 | } | |
6d6bc0ad | 8915 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 8916 | |
052f1dc7 | 8917 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8918 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8919 | { |
8920 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8921 | int on_rq; |
8922 | ||
29f59db3 | 8923 | on_rq = se->on_rq; |
62fb1851 | 8924 | if (on_rq) |
29f59db3 SV |
8925 | dequeue_entity(cfs_rq, se, 0); |
8926 | ||
8927 | se->load.weight = shares; | |
e05510d0 | 8928 | se->load.inv_weight = 0; |
29f59db3 | 8929 | |
62fb1851 | 8930 | if (on_rq) |
29f59db3 | 8931 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8932 | } |
62fb1851 | 8933 | |
c09595f6 PZ |
8934 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8935 | { | |
8936 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8937 | struct rq *rq = cfs_rq->rq; | |
8938 | unsigned long flags; | |
8939 | ||
8940 | spin_lock_irqsave(&rq->lock, flags); | |
8941 | __set_se_shares(se, shares); | |
8942 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
8943 | } |
8944 | ||
8ed36996 PZ |
8945 | static DEFINE_MUTEX(shares_mutex); |
8946 | ||
4cf86d77 | 8947 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8948 | { |
8949 | int i; | |
8ed36996 | 8950 | unsigned long flags; |
c61935fd | 8951 | |
ec7dc8ac DG |
8952 | /* |
8953 | * We can't change the weight of the root cgroup. | |
8954 | */ | |
8955 | if (!tg->se[0]) | |
8956 | return -EINVAL; | |
8957 | ||
18d95a28 PZ |
8958 | if (shares < MIN_SHARES) |
8959 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8960 | else if (shares > MAX_SHARES) |
8961 | shares = MAX_SHARES; | |
62fb1851 | 8962 | |
8ed36996 | 8963 | mutex_lock(&shares_mutex); |
9b5b7751 | 8964 | if (tg->shares == shares) |
5cb350ba | 8965 | goto done; |
29f59db3 | 8966 | |
8ed36996 | 8967 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8968 | for_each_possible_cpu(i) |
8969 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8970 | list_del_rcu(&tg->siblings); |
8ed36996 | 8971 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8972 | |
8973 | /* wait for any ongoing reference to this group to finish */ | |
8974 | synchronize_sched(); | |
8975 | ||
8976 | /* | |
8977 | * Now we are free to modify the group's share on each cpu | |
8978 | * w/o tripping rebalance_share or load_balance_fair. | |
8979 | */ | |
9b5b7751 | 8980 | tg->shares = shares; |
c09595f6 PZ |
8981 | for_each_possible_cpu(i) { |
8982 | /* | |
8983 | * force a rebalance | |
8984 | */ | |
8985 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8986 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8987 | } |
29f59db3 | 8988 | |
6b2d7700 SV |
8989 | /* |
8990 | * Enable load balance activity on this group, by inserting it back on | |
8991 | * each cpu's rq->leaf_cfs_rq_list. | |
8992 | */ | |
8ed36996 | 8993 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8994 | for_each_possible_cpu(i) |
8995 | register_fair_sched_group(tg, i); | |
f473aa5e | 8996 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8997 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8998 | done: |
8ed36996 | 8999 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9000 | return 0; |
29f59db3 SV |
9001 | } |
9002 | ||
5cb350ba DG |
9003 | unsigned long sched_group_shares(struct task_group *tg) |
9004 | { | |
9005 | return tg->shares; | |
9006 | } | |
052f1dc7 | 9007 | #endif |
5cb350ba | 9008 | |
052f1dc7 | 9009 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9010 | /* |
9f0c1e56 | 9011 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9012 | */ |
9f0c1e56 PZ |
9013 | static DEFINE_MUTEX(rt_constraints_mutex); |
9014 | ||
9015 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9016 | { | |
9017 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9018 | return 1ULL << 20; |
9f0c1e56 | 9019 | |
9a7e0b18 | 9020 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9021 | } |
9022 | ||
9a7e0b18 PZ |
9023 | /* Must be called with tasklist_lock held */ |
9024 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9025 | { |
9a7e0b18 | 9026 | struct task_struct *g, *p; |
b40b2e8e | 9027 | |
9a7e0b18 PZ |
9028 | do_each_thread(g, p) { |
9029 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9030 | return 1; | |
9031 | } while_each_thread(g, p); | |
b40b2e8e | 9032 | |
9a7e0b18 PZ |
9033 | return 0; |
9034 | } | |
b40b2e8e | 9035 | |
9a7e0b18 PZ |
9036 | struct rt_schedulable_data { |
9037 | struct task_group *tg; | |
9038 | u64 rt_period; | |
9039 | u64 rt_runtime; | |
9040 | }; | |
b40b2e8e | 9041 | |
9a7e0b18 PZ |
9042 | static int tg_schedulable(struct task_group *tg, void *data) |
9043 | { | |
9044 | struct rt_schedulable_data *d = data; | |
9045 | struct task_group *child; | |
9046 | unsigned long total, sum = 0; | |
9047 | u64 period, runtime; | |
b40b2e8e | 9048 | |
9a7e0b18 PZ |
9049 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9050 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9051 | |
9a7e0b18 PZ |
9052 | if (tg == d->tg) { |
9053 | period = d->rt_period; | |
9054 | runtime = d->rt_runtime; | |
b40b2e8e | 9055 | } |
b40b2e8e | 9056 | |
98a4826b PZ |
9057 | #ifdef CONFIG_USER_SCHED |
9058 | if (tg == &root_task_group) { | |
9059 | period = global_rt_period(); | |
9060 | runtime = global_rt_runtime(); | |
9061 | } | |
9062 | #endif | |
9063 | ||
4653f803 PZ |
9064 | /* |
9065 | * Cannot have more runtime than the period. | |
9066 | */ | |
9067 | if (runtime > period && runtime != RUNTIME_INF) | |
9068 | return -EINVAL; | |
6f505b16 | 9069 | |
4653f803 PZ |
9070 | /* |
9071 | * Ensure we don't starve existing RT tasks. | |
9072 | */ | |
9a7e0b18 PZ |
9073 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9074 | return -EBUSY; | |
6f505b16 | 9075 | |
9a7e0b18 | 9076 | total = to_ratio(period, runtime); |
6f505b16 | 9077 | |
4653f803 PZ |
9078 | /* |
9079 | * Nobody can have more than the global setting allows. | |
9080 | */ | |
9081 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9082 | return -EINVAL; | |
6f505b16 | 9083 | |
4653f803 PZ |
9084 | /* |
9085 | * The sum of our children's runtime should not exceed our own. | |
9086 | */ | |
9a7e0b18 PZ |
9087 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9088 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9089 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9090 | |
9a7e0b18 PZ |
9091 | if (child == d->tg) { |
9092 | period = d->rt_period; | |
9093 | runtime = d->rt_runtime; | |
9094 | } | |
6f505b16 | 9095 | |
9a7e0b18 | 9096 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9097 | } |
6f505b16 | 9098 | |
9a7e0b18 PZ |
9099 | if (sum > total) |
9100 | return -EINVAL; | |
9101 | ||
9102 | return 0; | |
6f505b16 PZ |
9103 | } |
9104 | ||
9a7e0b18 | 9105 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9106 | { |
9a7e0b18 PZ |
9107 | struct rt_schedulable_data data = { |
9108 | .tg = tg, | |
9109 | .rt_period = period, | |
9110 | .rt_runtime = runtime, | |
9111 | }; | |
9112 | ||
9113 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9114 | } |
9115 | ||
d0b27fa7 PZ |
9116 | static int tg_set_bandwidth(struct task_group *tg, |
9117 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9118 | { |
ac086bc2 | 9119 | int i, err = 0; |
9f0c1e56 | 9120 | |
9f0c1e56 | 9121 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9122 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9123 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9124 | if (err) | |
9f0c1e56 | 9125 | goto unlock; |
ac086bc2 PZ |
9126 | |
9127 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
9128 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
9129 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
9130 | |
9131 | for_each_possible_cpu(i) { | |
9132 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
9133 | ||
9134 | spin_lock(&rt_rq->rt_runtime_lock); | |
9135 | rt_rq->rt_runtime = rt_runtime; | |
9136 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9137 | } | |
9138 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 9139 | unlock: |
521f1a24 | 9140 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
9141 | mutex_unlock(&rt_constraints_mutex); |
9142 | ||
9143 | return err; | |
6f505b16 PZ |
9144 | } |
9145 | ||
d0b27fa7 PZ |
9146 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
9147 | { | |
9148 | u64 rt_runtime, rt_period; | |
9149 | ||
9150 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9151 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
9152 | if (rt_runtime_us < 0) | |
9153 | rt_runtime = RUNTIME_INF; | |
9154 | ||
9155 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
9156 | } | |
9157 | ||
9f0c1e56 PZ |
9158 | long sched_group_rt_runtime(struct task_group *tg) |
9159 | { | |
9160 | u64 rt_runtime_us; | |
9161 | ||
d0b27fa7 | 9162 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9163 | return -1; |
9164 | ||
d0b27fa7 | 9165 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9166 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9167 | return rt_runtime_us; | |
9168 | } | |
d0b27fa7 PZ |
9169 | |
9170 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
9171 | { | |
9172 | u64 rt_runtime, rt_period; | |
9173 | ||
9174 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
9175 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
9176 | ||
619b0488 R |
9177 | if (rt_period == 0) |
9178 | return -EINVAL; | |
9179 | ||
d0b27fa7 PZ |
9180 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9181 | } | |
9182 | ||
9183 | long sched_group_rt_period(struct task_group *tg) | |
9184 | { | |
9185 | u64 rt_period_us; | |
9186 | ||
9187 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9188 | do_div(rt_period_us, NSEC_PER_USEC); | |
9189 | return rt_period_us; | |
9190 | } | |
9191 | ||
9192 | static int sched_rt_global_constraints(void) | |
9193 | { | |
4653f803 | 9194 | u64 runtime, period; |
d0b27fa7 PZ |
9195 | int ret = 0; |
9196 | ||
ec5d4989 HS |
9197 | if (sysctl_sched_rt_period <= 0) |
9198 | return -EINVAL; | |
9199 | ||
4653f803 PZ |
9200 | runtime = global_rt_runtime(); |
9201 | period = global_rt_period(); | |
9202 | ||
9203 | /* | |
9204 | * Sanity check on the sysctl variables. | |
9205 | */ | |
9206 | if (runtime > period && runtime != RUNTIME_INF) | |
9207 | return -EINVAL; | |
10b612f4 | 9208 | |
d0b27fa7 | 9209 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9210 | read_lock(&tasklist_lock); |
4653f803 | 9211 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9212 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9213 | mutex_unlock(&rt_constraints_mutex); |
9214 | ||
9215 | return ret; | |
9216 | } | |
6d6bc0ad | 9217 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9218 | static int sched_rt_global_constraints(void) |
9219 | { | |
ac086bc2 PZ |
9220 | unsigned long flags; |
9221 | int i; | |
9222 | ||
ec5d4989 HS |
9223 | if (sysctl_sched_rt_period <= 0) |
9224 | return -EINVAL; | |
9225 | ||
ac086bc2 PZ |
9226 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9227 | for_each_possible_cpu(i) { | |
9228 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9229 | ||
9230 | spin_lock(&rt_rq->rt_runtime_lock); | |
9231 | rt_rq->rt_runtime = global_rt_runtime(); | |
9232 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9233 | } | |
9234 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9235 | ||
d0b27fa7 PZ |
9236 | return 0; |
9237 | } | |
6d6bc0ad | 9238 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9239 | |
9240 | int sched_rt_handler(struct ctl_table *table, int write, | |
9241 | struct file *filp, void __user *buffer, size_t *lenp, | |
9242 | loff_t *ppos) | |
9243 | { | |
9244 | int ret; | |
9245 | int old_period, old_runtime; | |
9246 | static DEFINE_MUTEX(mutex); | |
9247 | ||
9248 | mutex_lock(&mutex); | |
9249 | old_period = sysctl_sched_rt_period; | |
9250 | old_runtime = sysctl_sched_rt_runtime; | |
9251 | ||
9252 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9253 | ||
9254 | if (!ret && write) { | |
9255 | ret = sched_rt_global_constraints(); | |
9256 | if (ret) { | |
9257 | sysctl_sched_rt_period = old_period; | |
9258 | sysctl_sched_rt_runtime = old_runtime; | |
9259 | } else { | |
9260 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9261 | def_rt_bandwidth.rt_period = | |
9262 | ns_to_ktime(global_rt_period()); | |
9263 | } | |
9264 | } | |
9265 | mutex_unlock(&mutex); | |
9266 | ||
9267 | return ret; | |
9268 | } | |
68318b8e | 9269 | |
052f1dc7 | 9270 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9271 | |
9272 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9273 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9274 | { |
2b01dfe3 PM |
9275 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9276 | struct task_group, css); | |
68318b8e SV |
9277 | } |
9278 | ||
9279 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9280 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9281 | { |
ec7dc8ac | 9282 | struct task_group *tg, *parent; |
68318b8e | 9283 | |
2b01dfe3 | 9284 | if (!cgrp->parent) { |
68318b8e | 9285 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9286 | return &init_task_group.css; |
9287 | } | |
9288 | ||
ec7dc8ac DG |
9289 | parent = cgroup_tg(cgrp->parent); |
9290 | tg = sched_create_group(parent); | |
68318b8e SV |
9291 | if (IS_ERR(tg)) |
9292 | return ERR_PTR(-ENOMEM); | |
9293 | ||
68318b8e SV |
9294 | return &tg->css; |
9295 | } | |
9296 | ||
41a2d6cf IM |
9297 | static void |
9298 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9299 | { |
2b01dfe3 | 9300 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9301 | |
9302 | sched_destroy_group(tg); | |
9303 | } | |
9304 | ||
41a2d6cf IM |
9305 | static int |
9306 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9307 | struct task_struct *tsk) | |
68318b8e | 9308 | { |
b68aa230 PZ |
9309 | #ifdef CONFIG_RT_GROUP_SCHED |
9310 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 9311 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
9312 | return -EINVAL; |
9313 | #else | |
68318b8e SV |
9314 | /* We don't support RT-tasks being in separate groups */ |
9315 | if (tsk->sched_class != &fair_sched_class) | |
9316 | return -EINVAL; | |
b68aa230 | 9317 | #endif |
68318b8e SV |
9318 | |
9319 | return 0; | |
9320 | } | |
9321 | ||
9322 | static void | |
2b01dfe3 | 9323 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9324 | struct cgroup *old_cont, struct task_struct *tsk) |
9325 | { | |
9326 | sched_move_task(tsk); | |
9327 | } | |
9328 | ||
052f1dc7 | 9329 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9330 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9331 | u64 shareval) |
68318b8e | 9332 | { |
2b01dfe3 | 9333 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9334 | } |
9335 | ||
f4c753b7 | 9336 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9337 | { |
2b01dfe3 | 9338 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9339 | |
9340 | return (u64) tg->shares; | |
9341 | } | |
6d6bc0ad | 9342 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9343 | |
052f1dc7 | 9344 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9345 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9346 | s64 val) |
6f505b16 | 9347 | { |
06ecb27c | 9348 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9349 | } |
9350 | ||
06ecb27c | 9351 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9352 | { |
06ecb27c | 9353 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9354 | } |
d0b27fa7 PZ |
9355 | |
9356 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9357 | u64 rt_period_us) | |
9358 | { | |
9359 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9360 | } | |
9361 | ||
9362 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9363 | { | |
9364 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9365 | } | |
6d6bc0ad | 9366 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9367 | |
fe5c7cc2 | 9368 | static struct cftype cpu_files[] = { |
052f1dc7 | 9369 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9370 | { |
9371 | .name = "shares", | |
f4c753b7 PM |
9372 | .read_u64 = cpu_shares_read_u64, |
9373 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9374 | }, |
052f1dc7 PZ |
9375 | #endif |
9376 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9377 | { |
9f0c1e56 | 9378 | .name = "rt_runtime_us", |
06ecb27c PM |
9379 | .read_s64 = cpu_rt_runtime_read, |
9380 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9381 | }, |
d0b27fa7 PZ |
9382 | { |
9383 | .name = "rt_period_us", | |
f4c753b7 PM |
9384 | .read_u64 = cpu_rt_period_read_uint, |
9385 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9386 | }, |
052f1dc7 | 9387 | #endif |
68318b8e SV |
9388 | }; |
9389 | ||
9390 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9391 | { | |
fe5c7cc2 | 9392 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9393 | } |
9394 | ||
9395 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9396 | .name = "cpu", |
9397 | .create = cpu_cgroup_create, | |
9398 | .destroy = cpu_cgroup_destroy, | |
9399 | .can_attach = cpu_cgroup_can_attach, | |
9400 | .attach = cpu_cgroup_attach, | |
9401 | .populate = cpu_cgroup_populate, | |
9402 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9403 | .early_init = 1, |
9404 | }; | |
9405 | ||
052f1dc7 | 9406 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9407 | |
9408 | #ifdef CONFIG_CGROUP_CPUACCT | |
9409 | ||
9410 | /* | |
9411 | * CPU accounting code for task groups. | |
9412 | * | |
9413 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9414 | * (balbir@in.ibm.com). | |
9415 | */ | |
9416 | ||
934352f2 | 9417 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9418 | struct cpuacct { |
9419 | struct cgroup_subsys_state css; | |
9420 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
9421 | u64 *cpuusage; | |
934352f2 | 9422 | struct cpuacct *parent; |
d842de87 SV |
9423 | }; |
9424 | ||
9425 | struct cgroup_subsys cpuacct_subsys; | |
9426 | ||
9427 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9428 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9429 | { |
32cd756a | 9430 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9431 | struct cpuacct, css); |
9432 | } | |
9433 | ||
9434 | /* return cpu accounting group to which this task belongs */ | |
9435 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9436 | { | |
9437 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9438 | struct cpuacct, css); | |
9439 | } | |
9440 | ||
9441 | /* create a new cpu accounting group */ | |
9442 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9443 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9444 | { |
9445 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
9446 | ||
9447 | if (!ca) | |
9448 | return ERR_PTR(-ENOMEM); | |
9449 | ||
9450 | ca->cpuusage = alloc_percpu(u64); | |
9451 | if (!ca->cpuusage) { | |
9452 | kfree(ca); | |
9453 | return ERR_PTR(-ENOMEM); | |
9454 | } | |
9455 | ||
934352f2 BR |
9456 | if (cgrp->parent) |
9457 | ca->parent = cgroup_ca(cgrp->parent); | |
9458 | ||
d842de87 SV |
9459 | return &ca->css; |
9460 | } | |
9461 | ||
9462 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9463 | static void |
32cd756a | 9464 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9465 | { |
32cd756a | 9466 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9467 | |
9468 | free_percpu(ca->cpuusage); | |
9469 | kfree(ca); | |
9470 | } | |
9471 | ||
720f5498 KC |
9472 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9473 | { | |
9474 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
9475 | u64 data; | |
9476 | ||
9477 | #ifndef CONFIG_64BIT | |
9478 | /* | |
9479 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9480 | */ | |
9481 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
9482 | data = *cpuusage; | |
9483 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
9484 | #else | |
9485 | data = *cpuusage; | |
9486 | #endif | |
9487 | ||
9488 | return data; | |
9489 | } | |
9490 | ||
9491 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9492 | { | |
9493 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
9494 | ||
9495 | #ifndef CONFIG_64BIT | |
9496 | /* | |
9497 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9498 | */ | |
9499 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
9500 | *cpuusage = val; | |
9501 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
9502 | #else | |
9503 | *cpuusage = val; | |
9504 | #endif | |
9505 | } | |
9506 | ||
d842de87 | 9507 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9508 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9509 | { |
32cd756a | 9510 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9511 | u64 totalcpuusage = 0; |
9512 | int i; | |
9513 | ||
720f5498 KC |
9514 | for_each_present_cpu(i) |
9515 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9516 | |
9517 | return totalcpuusage; | |
9518 | } | |
9519 | ||
0297b803 DG |
9520 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9521 | u64 reset) | |
9522 | { | |
9523 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9524 | int err = 0; | |
9525 | int i; | |
9526 | ||
9527 | if (reset) { | |
9528 | err = -EINVAL; | |
9529 | goto out; | |
9530 | } | |
9531 | ||
720f5498 KC |
9532 | for_each_present_cpu(i) |
9533 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9534 | |
0297b803 DG |
9535 | out: |
9536 | return err; | |
9537 | } | |
9538 | ||
e9515c3c KC |
9539 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9540 | struct seq_file *m) | |
9541 | { | |
9542 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9543 | u64 percpu; | |
9544 | int i; | |
9545 | ||
9546 | for_each_present_cpu(i) { | |
9547 | percpu = cpuacct_cpuusage_read(ca, i); | |
9548 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9549 | } | |
9550 | seq_printf(m, "\n"); | |
9551 | return 0; | |
9552 | } | |
9553 | ||
d842de87 SV |
9554 | static struct cftype files[] = { |
9555 | { | |
9556 | .name = "usage", | |
f4c753b7 PM |
9557 | .read_u64 = cpuusage_read, |
9558 | .write_u64 = cpuusage_write, | |
d842de87 | 9559 | }, |
e9515c3c KC |
9560 | { |
9561 | .name = "usage_percpu", | |
9562 | .read_seq_string = cpuacct_percpu_seq_read, | |
9563 | }, | |
9564 | ||
d842de87 SV |
9565 | }; |
9566 | ||
32cd756a | 9567 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9568 | { |
32cd756a | 9569 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9570 | } |
9571 | ||
9572 | /* | |
9573 | * charge this task's execution time to its accounting group. | |
9574 | * | |
9575 | * called with rq->lock held. | |
9576 | */ | |
9577 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9578 | { | |
9579 | struct cpuacct *ca; | |
934352f2 | 9580 | int cpu; |
d842de87 SV |
9581 | |
9582 | if (!cpuacct_subsys.active) | |
9583 | return; | |
9584 | ||
934352f2 | 9585 | cpu = task_cpu(tsk); |
d842de87 | 9586 | ca = task_ca(tsk); |
d842de87 | 9587 | |
934352f2 BR |
9588 | for (; ca; ca = ca->parent) { |
9589 | u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); | |
d842de87 SV |
9590 | *cpuusage += cputime; |
9591 | } | |
9592 | } | |
9593 | ||
9594 | struct cgroup_subsys cpuacct_subsys = { | |
9595 | .name = "cpuacct", | |
9596 | .create = cpuacct_create, | |
9597 | .destroy = cpuacct_destroy, | |
9598 | .populate = cpuacct_populate, | |
9599 | .subsys_id = cpuacct_subsys_id, | |
9600 | }; | |
9601 | #endif /* CONFIG_CGROUP_CPUACCT */ |