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> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 GH |
79 | #include "sched_cpupri.h" |
80 | ||
a8d154b0 | 81 | #define CREATE_TRACE_POINTS |
ad8d75ff | 82 | #include <trace/events/sched.h> |
a8d154b0 | 83 | |
1da177e4 LT |
84 | /* |
85 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
86 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
87 | * and back. | |
88 | */ | |
89 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
90 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
91 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
92 | ||
93 | /* | |
94 | * 'User priority' is the nice value converted to something we | |
95 | * can work with better when scaling various scheduler parameters, | |
96 | * it's a [ 0 ... 39 ] range. | |
97 | */ | |
98 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
99 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
100 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
101 | ||
102 | /* | |
d7876a08 | 103 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 104 | */ |
d6322faf | 105 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 106 | |
6aa645ea IM |
107 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
108 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
109 | ||
1da177e4 LT |
110 | /* |
111 | * These are the 'tuning knobs' of the scheduler: | |
112 | * | |
a4ec24b4 | 113 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
114 | * Timeslices get refilled after they expire. |
115 | */ | |
1da177e4 | 116 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 117 | |
d0b27fa7 PZ |
118 | /* |
119 | * single value that denotes runtime == period, ie unlimited time. | |
120 | */ | |
121 | #define RUNTIME_INF ((u64)~0ULL) | |
122 | ||
5517d86b | 123 | #ifdef CONFIG_SMP |
fd2ab30b SN |
124 | |
125 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
126 | ||
5517d86b ED |
127 | /* |
128 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
129 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
130 | */ | |
131 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
132 | { | |
133 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
134 | } | |
135 | ||
136 | /* | |
137 | * Each time a sched group cpu_power is changed, | |
138 | * we must compute its reciprocal value | |
139 | */ | |
140 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
141 | { | |
142 | sg->__cpu_power += val; | |
143 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
144 | } | |
145 | #endif | |
146 | ||
e05606d3 IM |
147 | static inline int rt_policy(int policy) |
148 | { | |
3f33a7ce | 149 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
150 | return 1; |
151 | return 0; | |
152 | } | |
153 | ||
154 | static inline int task_has_rt_policy(struct task_struct *p) | |
155 | { | |
156 | return rt_policy(p->policy); | |
157 | } | |
158 | ||
1da177e4 | 159 | /* |
6aa645ea | 160 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 161 | */ |
6aa645ea IM |
162 | struct rt_prio_array { |
163 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
164 | struct list_head queue[MAX_RT_PRIO]; | |
165 | }; | |
166 | ||
d0b27fa7 | 167 | struct rt_bandwidth { |
ea736ed5 IM |
168 | /* nests inside the rq lock: */ |
169 | spinlock_t rt_runtime_lock; | |
170 | ktime_t rt_period; | |
171 | u64 rt_runtime; | |
172 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
173 | }; |
174 | ||
175 | static struct rt_bandwidth def_rt_bandwidth; | |
176 | ||
177 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
178 | ||
179 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
180 | { | |
181 | struct rt_bandwidth *rt_b = | |
182 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
183 | ktime_t now; | |
184 | int overrun; | |
185 | int idle = 0; | |
186 | ||
187 | for (;;) { | |
188 | now = hrtimer_cb_get_time(timer); | |
189 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
190 | ||
191 | if (!overrun) | |
192 | break; | |
193 | ||
194 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
195 | } | |
196 | ||
197 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
198 | } | |
199 | ||
200 | static | |
201 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
202 | { | |
203 | rt_b->rt_period = ns_to_ktime(period); | |
204 | rt_b->rt_runtime = runtime; | |
205 | ||
ac086bc2 PZ |
206 | spin_lock_init(&rt_b->rt_runtime_lock); |
207 | ||
d0b27fa7 PZ |
208 | hrtimer_init(&rt_b->rt_period_timer, |
209 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
210 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
211 | } |
212 | ||
c8bfff6d KH |
213 | static inline int rt_bandwidth_enabled(void) |
214 | { | |
215 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
216 | } |
217 | ||
218 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
219 | { | |
220 | ktime_t now; | |
221 | ||
cac64d00 | 222 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
223 | return; |
224 | ||
225 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
226 | return; | |
227 | ||
228 | spin_lock(&rt_b->rt_runtime_lock); | |
229 | for (;;) { | |
7f1e2ca9 PZ |
230 | unsigned long delta; |
231 | ktime_t soft, hard; | |
232 | ||
d0b27fa7 PZ |
233 | if (hrtimer_active(&rt_b->rt_period_timer)) |
234 | break; | |
235 | ||
236 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
237 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
238 | |
239 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
240 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
241 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
242 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
243 | HRTIMER_MODE_ABS, 0); | |
d0b27fa7 PZ |
244 | } |
245 | spin_unlock(&rt_b->rt_runtime_lock); | |
246 | } | |
247 | ||
248 | #ifdef CONFIG_RT_GROUP_SCHED | |
249 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
250 | { | |
251 | hrtimer_cancel(&rt_b->rt_period_timer); | |
252 | } | |
253 | #endif | |
254 | ||
712555ee HC |
255 | /* |
256 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
257 | * detach_destroy_domains and partition_sched_domains. | |
258 | */ | |
259 | static DEFINE_MUTEX(sched_domains_mutex); | |
260 | ||
052f1dc7 | 261 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 262 | |
68318b8e SV |
263 | #include <linux/cgroup.h> |
264 | ||
29f59db3 SV |
265 | struct cfs_rq; |
266 | ||
6f505b16 PZ |
267 | static LIST_HEAD(task_groups); |
268 | ||
29f59db3 | 269 | /* task group related information */ |
4cf86d77 | 270 | struct task_group { |
052f1dc7 | 271 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
272 | struct cgroup_subsys_state css; |
273 | #endif | |
052f1dc7 | 274 | |
6c415b92 AB |
275 | #ifdef CONFIG_USER_SCHED |
276 | uid_t uid; | |
277 | #endif | |
278 | ||
052f1dc7 | 279 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
280 | /* schedulable entities of this group on each cpu */ |
281 | struct sched_entity **se; | |
282 | /* runqueue "owned" by this group on each cpu */ | |
283 | struct cfs_rq **cfs_rq; | |
284 | unsigned long shares; | |
052f1dc7 PZ |
285 | #endif |
286 | ||
287 | #ifdef CONFIG_RT_GROUP_SCHED | |
288 | struct sched_rt_entity **rt_se; | |
289 | struct rt_rq **rt_rq; | |
290 | ||
d0b27fa7 | 291 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 292 | #endif |
6b2d7700 | 293 | |
ae8393e5 | 294 | struct rcu_head rcu; |
6f505b16 | 295 | struct list_head list; |
f473aa5e PZ |
296 | |
297 | struct task_group *parent; | |
298 | struct list_head siblings; | |
299 | struct list_head children; | |
29f59db3 SV |
300 | }; |
301 | ||
354d60c2 | 302 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 303 | |
6c415b92 AB |
304 | /* Helper function to pass uid information to create_sched_user() */ |
305 | void set_tg_uid(struct user_struct *user) | |
306 | { | |
307 | user->tg->uid = user->uid; | |
308 | } | |
309 | ||
eff766a6 PZ |
310 | /* |
311 | * Root task group. | |
312 | * Every UID task group (including init_task_group aka UID-0) will | |
313 | * be a child to this group. | |
314 | */ | |
315 | struct task_group root_task_group; | |
316 | ||
052f1dc7 | 317 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
318 | /* Default task group's sched entity on each cpu */ |
319 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
320 | /* Default task group's cfs_rq on each cpu */ | |
321 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 322 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
323 | |
324 | #ifdef CONFIG_RT_GROUP_SCHED | |
325 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
326 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 327 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 328 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 329 | #define root_task_group init_task_group |
9a7e0b18 | 330 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 331 | |
8ed36996 | 332 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
333 | * a task group's cpu shares. |
334 | */ | |
8ed36996 | 335 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 336 | |
57310a98 PZ |
337 | #ifdef CONFIG_SMP |
338 | static int root_task_group_empty(void) | |
339 | { | |
340 | return list_empty(&root_task_group.children); | |
341 | } | |
342 | #endif | |
343 | ||
052f1dc7 | 344 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
345 | #ifdef CONFIG_USER_SCHED |
346 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 347 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 348 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 349 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 350 | |
cb4ad1ff | 351 | /* |
2e084786 LJ |
352 | * A weight of 0 or 1 can cause arithmetics problems. |
353 | * A weight of a cfs_rq is the sum of weights of which entities | |
354 | * are queued on this cfs_rq, so a weight of a entity should not be | |
355 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
356 | * (The default weight is 1024 - so there's no practical |
357 | * limitation from this.) | |
358 | */ | |
18d95a28 | 359 | #define MIN_SHARES 2 |
2e084786 | 360 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 361 | |
052f1dc7 PZ |
362 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
363 | #endif | |
364 | ||
29f59db3 | 365 | /* Default task group. |
3a252015 | 366 | * Every task in system belong to this group at bootup. |
29f59db3 | 367 | */ |
434d53b0 | 368 | struct task_group init_task_group; |
29f59db3 SV |
369 | |
370 | /* return group to which a task belongs */ | |
4cf86d77 | 371 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 372 | { |
4cf86d77 | 373 | struct task_group *tg; |
9b5b7751 | 374 | |
052f1dc7 | 375 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
376 | rcu_read_lock(); |
377 | tg = __task_cred(p)->user->tg; | |
378 | rcu_read_unlock(); | |
052f1dc7 | 379 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
380 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
381 | struct task_group, css); | |
24e377a8 | 382 | #else |
41a2d6cf | 383 | tg = &init_task_group; |
24e377a8 | 384 | #endif |
9b5b7751 | 385 | return tg; |
29f59db3 SV |
386 | } |
387 | ||
388 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 389 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 390 | { |
052f1dc7 | 391 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
392 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
393 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 394 | #endif |
6f505b16 | 395 | |
052f1dc7 | 396 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
397 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
398 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 399 | #endif |
29f59db3 SV |
400 | } |
401 | ||
402 | #else | |
403 | ||
57310a98 PZ |
404 | #ifdef CONFIG_SMP |
405 | static int root_task_group_empty(void) | |
406 | { | |
407 | return 1; | |
408 | } | |
409 | #endif | |
410 | ||
6f505b16 | 411 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
412 | static inline struct task_group *task_group(struct task_struct *p) |
413 | { | |
414 | return NULL; | |
415 | } | |
29f59db3 | 416 | |
052f1dc7 | 417 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 418 | |
6aa645ea IM |
419 | /* CFS-related fields in a runqueue */ |
420 | struct cfs_rq { | |
421 | struct load_weight load; | |
422 | unsigned long nr_running; | |
423 | ||
6aa645ea | 424 | u64 exec_clock; |
e9acbff6 | 425 | u64 min_vruntime; |
6aa645ea IM |
426 | |
427 | struct rb_root tasks_timeline; | |
428 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
429 | |
430 | struct list_head tasks; | |
431 | struct list_head *balance_iterator; | |
432 | ||
433 | /* | |
434 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
435 | * It is set to NULL otherwise (i.e when none are currently running). |
436 | */ | |
4793241b | 437 | struct sched_entity *curr, *next, *last; |
ddc97297 | 438 | |
5ac5c4d6 | 439 | unsigned int nr_spread_over; |
ddc97297 | 440 | |
62160e3f | 441 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
442 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
443 | ||
41a2d6cf IM |
444 | /* |
445 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
446 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
447 | * (like users, containers etc.) | |
448 | * | |
449 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
450 | * list is used during load balance. | |
451 | */ | |
41a2d6cf IM |
452 | struct list_head leaf_cfs_rq_list; |
453 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
454 | |
455 | #ifdef CONFIG_SMP | |
c09595f6 | 456 | /* |
c8cba857 | 457 | * the part of load.weight contributed by tasks |
c09595f6 | 458 | */ |
c8cba857 | 459 | unsigned long task_weight; |
c09595f6 | 460 | |
c8cba857 PZ |
461 | /* |
462 | * h_load = weight * f(tg) | |
463 | * | |
464 | * Where f(tg) is the recursive weight fraction assigned to | |
465 | * this group. | |
466 | */ | |
467 | unsigned long h_load; | |
c09595f6 | 468 | |
c8cba857 PZ |
469 | /* |
470 | * this cpu's part of tg->shares | |
471 | */ | |
472 | unsigned long shares; | |
f1d239f7 PZ |
473 | |
474 | /* | |
475 | * load.weight at the time we set shares | |
476 | */ | |
477 | unsigned long rq_weight; | |
c09595f6 | 478 | #endif |
6aa645ea IM |
479 | #endif |
480 | }; | |
1da177e4 | 481 | |
6aa645ea IM |
482 | /* Real-Time classes' related field in a runqueue: */ |
483 | struct rt_rq { | |
484 | struct rt_prio_array active; | |
63489e45 | 485 | unsigned long rt_nr_running; |
052f1dc7 | 486 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
487 | struct { |
488 | int curr; /* highest queued rt task prio */ | |
398a153b | 489 | #ifdef CONFIG_SMP |
e864c499 | 490 | int next; /* next highest */ |
398a153b | 491 | #endif |
e864c499 | 492 | } highest_prio; |
6f505b16 | 493 | #endif |
fa85ae24 | 494 | #ifdef CONFIG_SMP |
73fe6aae | 495 | unsigned long rt_nr_migratory; |
a22d7fc1 | 496 | int overloaded; |
917b627d | 497 | struct plist_head pushable_tasks; |
fa85ae24 | 498 | #endif |
6f505b16 | 499 | int rt_throttled; |
fa85ae24 | 500 | u64 rt_time; |
ac086bc2 | 501 | u64 rt_runtime; |
ea736ed5 | 502 | /* Nests inside the rq lock: */ |
ac086bc2 | 503 | spinlock_t rt_runtime_lock; |
6f505b16 | 504 | |
052f1dc7 | 505 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
506 | unsigned long rt_nr_boosted; |
507 | ||
6f505b16 PZ |
508 | struct rq *rq; |
509 | struct list_head leaf_rt_rq_list; | |
510 | struct task_group *tg; | |
511 | struct sched_rt_entity *rt_se; | |
512 | #endif | |
6aa645ea IM |
513 | }; |
514 | ||
57d885fe GH |
515 | #ifdef CONFIG_SMP |
516 | ||
517 | /* | |
518 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
519 | * variables. Each exclusive cpuset essentially defines an island domain by |
520 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
521 | * exclusive cpuset is created, we also create and attach a new root-domain |
522 | * object. | |
523 | * | |
57d885fe GH |
524 | */ |
525 | struct root_domain { | |
526 | atomic_t refcount; | |
c6c4927b RR |
527 | cpumask_var_t span; |
528 | cpumask_var_t online; | |
637f5085 | 529 | |
0eab9146 | 530 | /* |
637f5085 GH |
531 | * The "RT overload" flag: it gets set if a CPU has more than |
532 | * one runnable RT task. | |
533 | */ | |
c6c4927b | 534 | cpumask_var_t rto_mask; |
0eab9146 | 535 | atomic_t rto_count; |
6e0534f2 GH |
536 | #ifdef CONFIG_SMP |
537 | struct cpupri cpupri; | |
538 | #endif | |
7a09b1a2 VS |
539 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
540 | /* | |
541 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
542 | * used when most cpus are idle in the system indicating overall very | |
543 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
544 | */ | |
545 | unsigned int sched_mc_preferred_wakeup_cpu; | |
546 | #endif | |
57d885fe GH |
547 | }; |
548 | ||
dc938520 GH |
549 | /* |
550 | * By default the system creates a single root-domain with all cpus as | |
551 | * members (mimicking the global state we have today). | |
552 | */ | |
57d885fe GH |
553 | static struct root_domain def_root_domain; |
554 | ||
555 | #endif | |
556 | ||
1da177e4 LT |
557 | /* |
558 | * This is the main, per-CPU runqueue data structure. | |
559 | * | |
560 | * Locking rule: those places that want to lock multiple runqueues | |
561 | * (such as the load balancing or the thread migration code), lock | |
562 | * acquire operations must be ordered by ascending &runqueue. | |
563 | */ | |
70b97a7f | 564 | struct rq { |
d8016491 IM |
565 | /* runqueue lock: */ |
566 | spinlock_t lock; | |
1da177e4 LT |
567 | |
568 | /* | |
569 | * nr_running and cpu_load should be in the same cacheline because | |
570 | * remote CPUs use both these fields when doing load calculation. | |
571 | */ | |
572 | unsigned long nr_running; | |
6aa645ea IM |
573 | #define CPU_LOAD_IDX_MAX 5 |
574 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 575 | #ifdef CONFIG_NO_HZ |
15934a37 | 576 | unsigned long last_tick_seen; |
46cb4b7c SS |
577 | unsigned char in_nohz_recently; |
578 | #endif | |
d8016491 IM |
579 | /* capture load from *all* tasks on this cpu: */ |
580 | struct load_weight load; | |
6aa645ea IM |
581 | unsigned long nr_load_updates; |
582 | u64 nr_switches; | |
583 | ||
584 | struct cfs_rq cfs; | |
6f505b16 | 585 | struct rt_rq rt; |
6f505b16 | 586 | |
6aa645ea | 587 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
588 | /* list of leaf cfs_rq on this cpu: */ |
589 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
590 | #endif |
591 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 592 | struct list_head leaf_rt_rq_list; |
1da177e4 | 593 | #endif |
1da177e4 LT |
594 | |
595 | /* | |
596 | * This is part of a global counter where only the total sum | |
597 | * over all CPUs matters. A task can increase this counter on | |
598 | * one CPU and if it got migrated afterwards it may decrease | |
599 | * it on another CPU. Always updated under the runqueue lock: | |
600 | */ | |
601 | unsigned long nr_uninterruptible; | |
602 | ||
36c8b586 | 603 | struct task_struct *curr, *idle; |
c9819f45 | 604 | unsigned long next_balance; |
1da177e4 | 605 | struct mm_struct *prev_mm; |
6aa645ea | 606 | |
3e51f33f | 607 | u64 clock; |
6aa645ea | 608 | |
1da177e4 LT |
609 | atomic_t nr_iowait; |
610 | ||
611 | #ifdef CONFIG_SMP | |
0eab9146 | 612 | struct root_domain *rd; |
1da177e4 LT |
613 | struct sched_domain *sd; |
614 | ||
a0a522ce | 615 | unsigned char idle_at_tick; |
1da177e4 LT |
616 | /* For active balancing */ |
617 | int active_balance; | |
618 | int push_cpu; | |
d8016491 IM |
619 | /* cpu of this runqueue: */ |
620 | int cpu; | |
1f11eb6a | 621 | int online; |
1da177e4 | 622 | |
a8a51d5e | 623 | unsigned long avg_load_per_task; |
1da177e4 | 624 | |
36c8b586 | 625 | struct task_struct *migration_thread; |
1da177e4 LT |
626 | struct list_head migration_queue; |
627 | #endif | |
628 | ||
dce48a84 TG |
629 | /* calc_load related fields */ |
630 | unsigned long calc_load_update; | |
631 | long calc_load_active; | |
632 | ||
8f4d37ec | 633 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
634 | #ifdef CONFIG_SMP |
635 | int hrtick_csd_pending; | |
636 | struct call_single_data hrtick_csd; | |
637 | #endif | |
8f4d37ec PZ |
638 | struct hrtimer hrtick_timer; |
639 | #endif | |
640 | ||
1da177e4 LT |
641 | #ifdef CONFIG_SCHEDSTATS |
642 | /* latency stats */ | |
643 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
644 | unsigned long long rq_cpu_time; |
645 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
646 | |
647 | /* sys_sched_yield() stats */ | |
480b9434 | 648 | unsigned int yld_count; |
1da177e4 LT |
649 | |
650 | /* schedule() stats */ | |
480b9434 KC |
651 | unsigned int sched_switch; |
652 | unsigned int sched_count; | |
653 | unsigned int sched_goidle; | |
1da177e4 LT |
654 | |
655 | /* try_to_wake_up() stats */ | |
480b9434 KC |
656 | unsigned int ttwu_count; |
657 | unsigned int ttwu_local; | |
b8efb561 IM |
658 | |
659 | /* BKL stats */ | |
480b9434 | 660 | unsigned int bkl_count; |
1da177e4 LT |
661 | #endif |
662 | }; | |
663 | ||
f34e3b61 | 664 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 665 | |
15afe09b | 666 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 667 | { |
15afe09b | 668 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
669 | } |
670 | ||
0a2966b4 CL |
671 | static inline int cpu_of(struct rq *rq) |
672 | { | |
673 | #ifdef CONFIG_SMP | |
674 | return rq->cpu; | |
675 | #else | |
676 | return 0; | |
677 | #endif | |
678 | } | |
679 | ||
674311d5 NP |
680 | /* |
681 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 682 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
683 | * |
684 | * The domain tree of any CPU may only be accessed from within | |
685 | * preempt-disabled sections. | |
686 | */ | |
48f24c4d IM |
687 | #define for_each_domain(cpu, __sd) \ |
688 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
689 | |
690 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
691 | #define this_rq() (&__get_cpu_var(runqueues)) | |
692 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
693 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
694 | ||
3e51f33f PZ |
695 | static inline void update_rq_clock(struct rq *rq) |
696 | { | |
697 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
698 | } | |
699 | ||
bf5c91ba IM |
700 | /* |
701 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
702 | */ | |
703 | #ifdef CONFIG_SCHED_DEBUG | |
704 | # define const_debug __read_mostly | |
705 | #else | |
706 | # define const_debug static const | |
707 | #endif | |
708 | ||
017730c1 IM |
709 | /** |
710 | * runqueue_is_locked | |
711 | * | |
712 | * Returns true if the current cpu runqueue is locked. | |
713 | * This interface allows printk to be called with the runqueue lock | |
714 | * held and know whether or not it is OK to wake up the klogd. | |
715 | */ | |
716 | int runqueue_is_locked(void) | |
717 | { | |
718 | int cpu = get_cpu(); | |
719 | struct rq *rq = cpu_rq(cpu); | |
720 | int ret; | |
721 | ||
722 | ret = spin_is_locked(&rq->lock); | |
723 | put_cpu(); | |
724 | return ret; | |
725 | } | |
726 | ||
bf5c91ba IM |
727 | /* |
728 | * Debugging: various feature bits | |
729 | */ | |
f00b45c1 PZ |
730 | |
731 | #define SCHED_FEAT(name, enabled) \ | |
732 | __SCHED_FEAT_##name , | |
733 | ||
bf5c91ba | 734 | enum { |
f00b45c1 | 735 | #include "sched_features.h" |
bf5c91ba IM |
736 | }; |
737 | ||
f00b45c1 PZ |
738 | #undef SCHED_FEAT |
739 | ||
740 | #define SCHED_FEAT(name, enabled) \ | |
741 | (1UL << __SCHED_FEAT_##name) * enabled | | |
742 | ||
bf5c91ba | 743 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
744 | #include "sched_features.h" |
745 | 0; | |
746 | ||
747 | #undef SCHED_FEAT | |
748 | ||
749 | #ifdef CONFIG_SCHED_DEBUG | |
750 | #define SCHED_FEAT(name, enabled) \ | |
751 | #name , | |
752 | ||
983ed7a6 | 753 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
754 | #include "sched_features.h" |
755 | NULL | |
756 | }; | |
757 | ||
758 | #undef SCHED_FEAT | |
759 | ||
34f3a814 | 760 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 761 | { |
f00b45c1 PZ |
762 | int i; |
763 | ||
764 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
765 | if (!(sysctl_sched_features & (1UL << i))) |
766 | seq_puts(m, "NO_"); | |
767 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 768 | } |
34f3a814 | 769 | seq_puts(m, "\n"); |
f00b45c1 | 770 | |
34f3a814 | 771 | return 0; |
f00b45c1 PZ |
772 | } |
773 | ||
774 | static ssize_t | |
775 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
776 | size_t cnt, loff_t *ppos) | |
777 | { | |
778 | char buf[64]; | |
779 | char *cmp = buf; | |
780 | int neg = 0; | |
781 | int i; | |
782 | ||
783 | if (cnt > 63) | |
784 | cnt = 63; | |
785 | ||
786 | if (copy_from_user(&buf, ubuf, cnt)) | |
787 | return -EFAULT; | |
788 | ||
789 | buf[cnt] = 0; | |
790 | ||
c24b7c52 | 791 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
792 | neg = 1; |
793 | cmp += 3; | |
794 | } | |
795 | ||
796 | for (i = 0; sched_feat_names[i]; i++) { | |
797 | int len = strlen(sched_feat_names[i]); | |
798 | ||
799 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
800 | if (neg) | |
801 | sysctl_sched_features &= ~(1UL << i); | |
802 | else | |
803 | sysctl_sched_features |= (1UL << i); | |
804 | break; | |
805 | } | |
806 | } | |
807 | ||
808 | if (!sched_feat_names[i]) | |
809 | return -EINVAL; | |
810 | ||
811 | filp->f_pos += cnt; | |
812 | ||
813 | return cnt; | |
814 | } | |
815 | ||
34f3a814 LZ |
816 | static int sched_feat_open(struct inode *inode, struct file *filp) |
817 | { | |
818 | return single_open(filp, sched_feat_show, NULL); | |
819 | } | |
820 | ||
f00b45c1 | 821 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
822 | .open = sched_feat_open, |
823 | .write = sched_feat_write, | |
824 | .read = seq_read, | |
825 | .llseek = seq_lseek, | |
826 | .release = single_release, | |
f00b45c1 PZ |
827 | }; |
828 | ||
829 | static __init int sched_init_debug(void) | |
830 | { | |
f00b45c1 PZ |
831 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
832 | &sched_feat_fops); | |
833 | ||
834 | return 0; | |
835 | } | |
836 | late_initcall(sched_init_debug); | |
837 | ||
838 | #endif | |
839 | ||
840 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 841 | |
b82d9fdd PZ |
842 | /* |
843 | * Number of tasks to iterate in a single balance run. | |
844 | * Limited because this is done with IRQs disabled. | |
845 | */ | |
846 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
847 | ||
2398f2c6 PZ |
848 | /* |
849 | * ratelimit for updating the group shares. | |
55cd5340 | 850 | * default: 0.25ms |
2398f2c6 | 851 | */ |
55cd5340 | 852 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 853 | |
ffda12a1 PZ |
854 | /* |
855 | * Inject some fuzzyness into changing the per-cpu group shares | |
856 | * this avoids remote rq-locks at the expense of fairness. | |
857 | * default: 4 | |
858 | */ | |
859 | unsigned int sysctl_sched_shares_thresh = 4; | |
860 | ||
fa85ae24 | 861 | /* |
9f0c1e56 | 862 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
863 | * default: 1s |
864 | */ | |
9f0c1e56 | 865 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 866 | |
6892b75e IM |
867 | static __read_mostly int scheduler_running; |
868 | ||
9f0c1e56 PZ |
869 | /* |
870 | * part of the period that we allow rt tasks to run in us. | |
871 | * default: 0.95s | |
872 | */ | |
873 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 874 | |
d0b27fa7 PZ |
875 | static inline u64 global_rt_period(void) |
876 | { | |
877 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
878 | } | |
879 | ||
880 | static inline u64 global_rt_runtime(void) | |
881 | { | |
e26873bb | 882 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
883 | return RUNTIME_INF; |
884 | ||
885 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
886 | } | |
fa85ae24 | 887 | |
1da177e4 | 888 | #ifndef prepare_arch_switch |
4866cde0 NP |
889 | # define prepare_arch_switch(next) do { } while (0) |
890 | #endif | |
891 | #ifndef finish_arch_switch | |
892 | # define finish_arch_switch(prev) do { } while (0) | |
893 | #endif | |
894 | ||
051a1d1a DA |
895 | static inline int task_current(struct rq *rq, struct task_struct *p) |
896 | { | |
897 | return rq->curr == p; | |
898 | } | |
899 | ||
4866cde0 | 900 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 901 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 902 | { |
051a1d1a | 903 | return task_current(rq, p); |
4866cde0 NP |
904 | } |
905 | ||
70b97a7f | 906 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
907 | { |
908 | } | |
909 | ||
70b97a7f | 910 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 911 | { |
da04c035 IM |
912 | #ifdef CONFIG_DEBUG_SPINLOCK |
913 | /* this is a valid case when another task releases the spinlock */ | |
914 | rq->lock.owner = current; | |
915 | #endif | |
8a25d5de IM |
916 | /* |
917 | * If we are tracking spinlock dependencies then we have to | |
918 | * fix up the runqueue lock - which gets 'carried over' from | |
919 | * prev into current: | |
920 | */ | |
921 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
922 | ||
4866cde0 NP |
923 | spin_unlock_irq(&rq->lock); |
924 | } | |
925 | ||
926 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 927 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
928 | { |
929 | #ifdef CONFIG_SMP | |
930 | return p->oncpu; | |
931 | #else | |
051a1d1a | 932 | return task_current(rq, p); |
4866cde0 NP |
933 | #endif |
934 | } | |
935 | ||
70b97a7f | 936 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
937 | { |
938 | #ifdef CONFIG_SMP | |
939 | /* | |
940 | * We can optimise this out completely for !SMP, because the | |
941 | * SMP rebalancing from interrupt is the only thing that cares | |
942 | * here. | |
943 | */ | |
944 | next->oncpu = 1; | |
945 | #endif | |
946 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
947 | spin_unlock_irq(&rq->lock); | |
948 | #else | |
949 | spin_unlock(&rq->lock); | |
950 | #endif | |
951 | } | |
952 | ||
70b97a7f | 953 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
954 | { |
955 | #ifdef CONFIG_SMP | |
956 | /* | |
957 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
958 | * We must ensure this doesn't happen until the switch is completely | |
959 | * finished. | |
960 | */ | |
961 | smp_wmb(); | |
962 | prev->oncpu = 0; | |
963 | #endif | |
964 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
965 | local_irq_enable(); | |
1da177e4 | 966 | #endif |
4866cde0 NP |
967 | } |
968 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 969 | |
b29739f9 IM |
970 | /* |
971 | * __task_rq_lock - lock the runqueue a given task resides on. | |
972 | * Must be called interrupts disabled. | |
973 | */ | |
70b97a7f | 974 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
975 | __acquires(rq->lock) |
976 | { | |
3a5c359a AK |
977 | for (;;) { |
978 | struct rq *rq = task_rq(p); | |
979 | spin_lock(&rq->lock); | |
980 | if (likely(rq == task_rq(p))) | |
981 | return rq; | |
b29739f9 | 982 | spin_unlock(&rq->lock); |
b29739f9 | 983 | } |
b29739f9 IM |
984 | } |
985 | ||
1da177e4 LT |
986 | /* |
987 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 988 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
989 | * explicitly disabling preemption. |
990 | */ | |
70b97a7f | 991 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
992 | __acquires(rq->lock) |
993 | { | |
70b97a7f | 994 | struct rq *rq; |
1da177e4 | 995 | |
3a5c359a AK |
996 | for (;;) { |
997 | local_irq_save(*flags); | |
998 | rq = task_rq(p); | |
999 | spin_lock(&rq->lock); | |
1000 | if (likely(rq == task_rq(p))) | |
1001 | return rq; | |
1da177e4 | 1002 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 1003 | } |
1da177e4 LT |
1004 | } |
1005 | ||
ad474cac ON |
1006 | void task_rq_unlock_wait(struct task_struct *p) |
1007 | { | |
1008 | struct rq *rq = task_rq(p); | |
1009 | ||
1010 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1011 | spin_unlock_wait(&rq->lock); | |
1012 | } | |
1013 | ||
a9957449 | 1014 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1015 | __releases(rq->lock) |
1016 | { | |
1017 | spin_unlock(&rq->lock); | |
1018 | } | |
1019 | ||
70b97a7f | 1020 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1021 | __releases(rq->lock) |
1022 | { | |
1023 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1024 | } | |
1025 | ||
1da177e4 | 1026 | /* |
cc2a73b5 | 1027 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1028 | */ |
a9957449 | 1029 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1030 | __acquires(rq->lock) |
1031 | { | |
70b97a7f | 1032 | struct rq *rq; |
1da177e4 LT |
1033 | |
1034 | local_irq_disable(); | |
1035 | rq = this_rq(); | |
1036 | spin_lock(&rq->lock); | |
1037 | ||
1038 | return rq; | |
1039 | } | |
1040 | ||
8f4d37ec PZ |
1041 | #ifdef CONFIG_SCHED_HRTICK |
1042 | /* | |
1043 | * Use HR-timers to deliver accurate preemption points. | |
1044 | * | |
1045 | * Its all a bit involved since we cannot program an hrt while holding the | |
1046 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1047 | * reschedule event. | |
1048 | * | |
1049 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1050 | * rq->lock. | |
1051 | */ | |
8f4d37ec PZ |
1052 | |
1053 | /* | |
1054 | * Use hrtick when: | |
1055 | * - enabled by features | |
1056 | * - hrtimer is actually high res | |
1057 | */ | |
1058 | static inline int hrtick_enabled(struct rq *rq) | |
1059 | { | |
1060 | if (!sched_feat(HRTICK)) | |
1061 | return 0; | |
ba42059f | 1062 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1063 | return 0; |
8f4d37ec PZ |
1064 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1065 | } | |
1066 | ||
8f4d37ec PZ |
1067 | static void hrtick_clear(struct rq *rq) |
1068 | { | |
1069 | if (hrtimer_active(&rq->hrtick_timer)) | |
1070 | hrtimer_cancel(&rq->hrtick_timer); | |
1071 | } | |
1072 | ||
8f4d37ec PZ |
1073 | /* |
1074 | * High-resolution timer tick. | |
1075 | * Runs from hardirq context with interrupts disabled. | |
1076 | */ | |
1077 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1078 | { | |
1079 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1080 | ||
1081 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1082 | ||
1083 | spin_lock(&rq->lock); | |
3e51f33f | 1084 | update_rq_clock(rq); |
8f4d37ec PZ |
1085 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1086 | spin_unlock(&rq->lock); | |
1087 | ||
1088 | return HRTIMER_NORESTART; | |
1089 | } | |
1090 | ||
95e904c7 | 1091 | #ifdef CONFIG_SMP |
31656519 PZ |
1092 | /* |
1093 | * called from hardirq (IPI) context | |
1094 | */ | |
1095 | static void __hrtick_start(void *arg) | |
b328ca18 | 1096 | { |
31656519 | 1097 | struct rq *rq = arg; |
b328ca18 | 1098 | |
31656519 PZ |
1099 | spin_lock(&rq->lock); |
1100 | hrtimer_restart(&rq->hrtick_timer); | |
1101 | rq->hrtick_csd_pending = 0; | |
1102 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1103 | } |
1104 | ||
31656519 PZ |
1105 | /* |
1106 | * Called to set the hrtick timer state. | |
1107 | * | |
1108 | * called with rq->lock held and irqs disabled | |
1109 | */ | |
1110 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1111 | { |
31656519 PZ |
1112 | struct hrtimer *timer = &rq->hrtick_timer; |
1113 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1114 | |
cc584b21 | 1115 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1116 | |
1117 | if (rq == this_rq()) { | |
1118 | hrtimer_restart(timer); | |
1119 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1120 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1121 | rq->hrtick_csd_pending = 1; |
1122 | } | |
b328ca18 PZ |
1123 | } |
1124 | ||
1125 | static int | |
1126 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1127 | { | |
1128 | int cpu = (int)(long)hcpu; | |
1129 | ||
1130 | switch (action) { | |
1131 | case CPU_UP_CANCELED: | |
1132 | case CPU_UP_CANCELED_FROZEN: | |
1133 | case CPU_DOWN_PREPARE: | |
1134 | case CPU_DOWN_PREPARE_FROZEN: | |
1135 | case CPU_DEAD: | |
1136 | case CPU_DEAD_FROZEN: | |
31656519 | 1137 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1138 | return NOTIFY_OK; |
1139 | } | |
1140 | ||
1141 | return NOTIFY_DONE; | |
1142 | } | |
1143 | ||
fa748203 | 1144 | static __init void init_hrtick(void) |
b328ca18 PZ |
1145 | { |
1146 | hotcpu_notifier(hotplug_hrtick, 0); | |
1147 | } | |
31656519 PZ |
1148 | #else |
1149 | /* | |
1150 | * Called to set the hrtick timer state. | |
1151 | * | |
1152 | * called with rq->lock held and irqs disabled | |
1153 | */ | |
1154 | static void hrtick_start(struct rq *rq, u64 delay) | |
1155 | { | |
7f1e2ca9 PZ |
1156 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
1157 | HRTIMER_MODE_REL, 0); | |
31656519 | 1158 | } |
b328ca18 | 1159 | |
006c75f1 | 1160 | static inline void init_hrtick(void) |
8f4d37ec | 1161 | { |
8f4d37ec | 1162 | } |
31656519 | 1163 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1164 | |
31656519 | 1165 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1166 | { |
31656519 PZ |
1167 | #ifdef CONFIG_SMP |
1168 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1169 | |
31656519 PZ |
1170 | rq->hrtick_csd.flags = 0; |
1171 | rq->hrtick_csd.func = __hrtick_start; | |
1172 | rq->hrtick_csd.info = rq; | |
1173 | #endif | |
8f4d37ec | 1174 | |
31656519 PZ |
1175 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1176 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1177 | } |
006c75f1 | 1178 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1179 | static inline void hrtick_clear(struct rq *rq) |
1180 | { | |
1181 | } | |
1182 | ||
8f4d37ec PZ |
1183 | static inline void init_rq_hrtick(struct rq *rq) |
1184 | { | |
1185 | } | |
1186 | ||
b328ca18 PZ |
1187 | static inline void init_hrtick(void) |
1188 | { | |
1189 | } | |
006c75f1 | 1190 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1191 | |
c24d20db IM |
1192 | /* |
1193 | * resched_task - mark a task 'to be rescheduled now'. | |
1194 | * | |
1195 | * On UP this means the setting of the need_resched flag, on SMP it | |
1196 | * might also involve a cross-CPU call to trigger the scheduler on | |
1197 | * the target CPU. | |
1198 | */ | |
1199 | #ifdef CONFIG_SMP | |
1200 | ||
1201 | #ifndef tsk_is_polling | |
1202 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1203 | #endif | |
1204 | ||
31656519 | 1205 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1206 | { |
1207 | int cpu; | |
1208 | ||
1209 | assert_spin_locked(&task_rq(p)->lock); | |
1210 | ||
5ed0cec0 | 1211 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1212 | return; |
1213 | ||
5ed0cec0 | 1214 | set_tsk_need_resched(p); |
c24d20db IM |
1215 | |
1216 | cpu = task_cpu(p); | |
1217 | if (cpu == smp_processor_id()) | |
1218 | return; | |
1219 | ||
1220 | /* NEED_RESCHED must be visible before we test polling */ | |
1221 | smp_mb(); | |
1222 | if (!tsk_is_polling(p)) | |
1223 | smp_send_reschedule(cpu); | |
1224 | } | |
1225 | ||
1226 | static void resched_cpu(int cpu) | |
1227 | { | |
1228 | struct rq *rq = cpu_rq(cpu); | |
1229 | unsigned long flags; | |
1230 | ||
1231 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1232 | return; | |
1233 | resched_task(cpu_curr(cpu)); | |
1234 | spin_unlock_irqrestore(&rq->lock, flags); | |
1235 | } | |
06d8308c TG |
1236 | |
1237 | #ifdef CONFIG_NO_HZ | |
1238 | /* | |
1239 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1240 | * idle CPU then this timer might expire before the next timer event | |
1241 | * which is scheduled to wake up that CPU. In case of a completely | |
1242 | * idle system the next event might even be infinite time into the | |
1243 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1244 | * leaves the inner idle loop so the newly added timer is taken into | |
1245 | * account when the CPU goes back to idle and evaluates the timer | |
1246 | * wheel for the next timer event. | |
1247 | */ | |
1248 | void wake_up_idle_cpu(int cpu) | |
1249 | { | |
1250 | struct rq *rq = cpu_rq(cpu); | |
1251 | ||
1252 | if (cpu == smp_processor_id()) | |
1253 | return; | |
1254 | ||
1255 | /* | |
1256 | * This is safe, as this function is called with the timer | |
1257 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1258 | * to idle and has not yet set rq->curr to idle then it will | |
1259 | * be serialized on the timer wheel base lock and take the new | |
1260 | * timer into account automatically. | |
1261 | */ | |
1262 | if (rq->curr != rq->idle) | |
1263 | return; | |
1264 | ||
1265 | /* | |
1266 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1267 | * lockless. The worst case is that the other CPU runs the | |
1268 | * idle task through an additional NOOP schedule() | |
1269 | */ | |
5ed0cec0 | 1270 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1271 | |
1272 | /* NEED_RESCHED must be visible before we test polling */ | |
1273 | smp_mb(); | |
1274 | if (!tsk_is_polling(rq->idle)) | |
1275 | smp_send_reschedule(cpu); | |
1276 | } | |
6d6bc0ad | 1277 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1278 | |
6d6bc0ad | 1279 | #else /* !CONFIG_SMP */ |
31656519 | 1280 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1281 | { |
1282 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1283 | set_tsk_need_resched(p); |
c24d20db | 1284 | } |
6d6bc0ad | 1285 | #endif /* CONFIG_SMP */ |
c24d20db | 1286 | |
45bf76df IM |
1287 | #if BITS_PER_LONG == 32 |
1288 | # define WMULT_CONST (~0UL) | |
1289 | #else | |
1290 | # define WMULT_CONST (1UL << 32) | |
1291 | #endif | |
1292 | ||
1293 | #define WMULT_SHIFT 32 | |
1294 | ||
194081eb IM |
1295 | /* |
1296 | * Shift right and round: | |
1297 | */ | |
cf2ab469 | 1298 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1299 | |
a7be37ac PZ |
1300 | /* |
1301 | * delta *= weight / lw | |
1302 | */ | |
cb1c4fc9 | 1303 | static unsigned long |
45bf76df IM |
1304 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1305 | struct load_weight *lw) | |
1306 | { | |
1307 | u64 tmp; | |
1308 | ||
7a232e03 LJ |
1309 | if (!lw->inv_weight) { |
1310 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1311 | lw->inv_weight = 1; | |
1312 | else | |
1313 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1314 | / (lw->weight+1); | |
1315 | } | |
45bf76df IM |
1316 | |
1317 | tmp = (u64)delta_exec * weight; | |
1318 | /* | |
1319 | * Check whether we'd overflow the 64-bit multiplication: | |
1320 | */ | |
194081eb | 1321 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1322 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1323 | WMULT_SHIFT/2); |
1324 | else | |
cf2ab469 | 1325 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1326 | |
ecf691da | 1327 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1328 | } |
1329 | ||
1091985b | 1330 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1331 | { |
1332 | lw->weight += inc; | |
e89996ae | 1333 | lw->inv_weight = 0; |
45bf76df IM |
1334 | } |
1335 | ||
1091985b | 1336 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1337 | { |
1338 | lw->weight -= dec; | |
e89996ae | 1339 | lw->inv_weight = 0; |
45bf76df IM |
1340 | } |
1341 | ||
2dd73a4f PW |
1342 | /* |
1343 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1344 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1345 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1346 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1347 | * scaled version of the new time slice allocation that they receive on time |
1348 | * slice expiry etc. | |
1349 | */ | |
1350 | ||
cce7ade8 PZ |
1351 | #define WEIGHT_IDLEPRIO 3 |
1352 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1353 | |
1354 | /* | |
1355 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1356 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1357 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1358 | * that remained on nice 0. | |
1359 | * | |
1360 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1361 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1362 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1363 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1364 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1365 | */ |
1366 | static const int prio_to_weight[40] = { | |
254753dc IM |
1367 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1368 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1369 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1370 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1371 | /* 0 */ 1024, 820, 655, 526, 423, | |
1372 | /* 5 */ 335, 272, 215, 172, 137, | |
1373 | /* 10 */ 110, 87, 70, 56, 45, | |
1374 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1375 | }; |
1376 | ||
5714d2de IM |
1377 | /* |
1378 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1379 | * | |
1380 | * In cases where the weight does not change often, we can use the | |
1381 | * precalculated inverse to speed up arithmetics by turning divisions | |
1382 | * into multiplications: | |
1383 | */ | |
dd41f596 | 1384 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1385 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1386 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1387 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1388 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1389 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1390 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1391 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1392 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1393 | }; |
2dd73a4f | 1394 | |
dd41f596 IM |
1395 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1396 | ||
1397 | /* | |
1398 | * runqueue iterator, to support SMP load-balancing between different | |
1399 | * scheduling classes, without having to expose their internal data | |
1400 | * structures to the load-balancing proper: | |
1401 | */ | |
1402 | struct rq_iterator { | |
1403 | void *arg; | |
1404 | struct task_struct *(*start)(void *); | |
1405 | struct task_struct *(*next)(void *); | |
1406 | }; | |
1407 | ||
e1d1484f PW |
1408 | #ifdef CONFIG_SMP |
1409 | static unsigned long | |
1410 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1411 | unsigned long max_load_move, struct sched_domain *sd, | |
1412 | enum cpu_idle_type idle, int *all_pinned, | |
1413 | int *this_best_prio, struct rq_iterator *iterator); | |
1414 | ||
1415 | static int | |
1416 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1417 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1418 | struct rq_iterator *iterator); | |
e1d1484f | 1419 | #endif |
dd41f596 | 1420 | |
ef12fefa BR |
1421 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1422 | enum cpuacct_stat_index { | |
1423 | CPUACCT_STAT_USER, /* ... user mode */ | |
1424 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1425 | ||
1426 | CPUACCT_STAT_NSTATS, | |
1427 | }; | |
1428 | ||
d842de87 SV |
1429 | #ifdef CONFIG_CGROUP_CPUACCT |
1430 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1431 | static void cpuacct_update_stats(struct task_struct *tsk, |
1432 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1433 | #else |
1434 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1435 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1436 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1437 | #endif |
1438 | ||
18d95a28 PZ |
1439 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1440 | { | |
1441 | update_load_add(&rq->load, load); | |
1442 | } | |
1443 | ||
1444 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1445 | { | |
1446 | update_load_sub(&rq->load, load); | |
1447 | } | |
1448 | ||
7940ca36 | 1449 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1450 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1451 | |
1452 | /* | |
1453 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1454 | * leaving it for the final time. | |
1455 | */ | |
eb755805 | 1456 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1457 | { |
1458 | struct task_group *parent, *child; | |
eb755805 | 1459 | int ret; |
c09595f6 PZ |
1460 | |
1461 | rcu_read_lock(); | |
1462 | parent = &root_task_group; | |
1463 | down: | |
eb755805 PZ |
1464 | ret = (*down)(parent, data); |
1465 | if (ret) | |
1466 | goto out_unlock; | |
c09595f6 PZ |
1467 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1468 | parent = child; | |
1469 | goto down; | |
1470 | ||
1471 | up: | |
1472 | continue; | |
1473 | } | |
eb755805 PZ |
1474 | ret = (*up)(parent, data); |
1475 | if (ret) | |
1476 | goto out_unlock; | |
c09595f6 PZ |
1477 | |
1478 | child = parent; | |
1479 | parent = parent->parent; | |
1480 | if (parent) | |
1481 | goto up; | |
eb755805 | 1482 | out_unlock: |
c09595f6 | 1483 | rcu_read_unlock(); |
eb755805 PZ |
1484 | |
1485 | return ret; | |
c09595f6 PZ |
1486 | } |
1487 | ||
eb755805 PZ |
1488 | static int tg_nop(struct task_group *tg, void *data) |
1489 | { | |
1490 | return 0; | |
c09595f6 | 1491 | } |
eb755805 PZ |
1492 | #endif |
1493 | ||
1494 | #ifdef CONFIG_SMP | |
1495 | static unsigned long source_load(int cpu, int type); | |
1496 | static unsigned long target_load(int cpu, int type); | |
1497 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1498 | ||
1499 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1500 | { | |
1501 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1502 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1503 | |
4cd42620 SR |
1504 | if (nr_running) |
1505 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1506 | else |
1507 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1508 | |
1509 | return rq->avg_load_per_task; | |
1510 | } | |
1511 | ||
1512 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1513 | |
c09595f6 PZ |
1514 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1515 | ||
1516 | /* | |
1517 | * Calculate and set the cpu's group shares. | |
1518 | */ | |
1519 | static void | |
ffda12a1 PZ |
1520 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1521 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1522 | { |
c09595f6 PZ |
1523 | unsigned long shares; |
1524 | unsigned long rq_weight; | |
1525 | ||
c8cba857 | 1526 | if (!tg->se[cpu]) |
c09595f6 PZ |
1527 | return; |
1528 | ||
ec4e0e2f | 1529 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1530 | |
c09595f6 PZ |
1531 | /* |
1532 | * \Sum shares * rq_weight | |
1533 | * shares = ----------------------- | |
1534 | * \Sum rq_weight | |
1535 | * | |
1536 | */ | |
ec4e0e2f | 1537 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1538 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1539 | |
ffda12a1 PZ |
1540 | if (abs(shares - tg->se[cpu]->load.weight) > |
1541 | sysctl_sched_shares_thresh) { | |
1542 | struct rq *rq = cpu_rq(cpu); | |
1543 | unsigned long flags; | |
c09595f6 | 1544 | |
ffda12a1 | 1545 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1546 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1547 | |
ffda12a1 PZ |
1548 | __set_se_shares(tg->se[cpu], shares); |
1549 | spin_unlock_irqrestore(&rq->lock, flags); | |
1550 | } | |
18d95a28 | 1551 | } |
c09595f6 PZ |
1552 | |
1553 | /* | |
c8cba857 PZ |
1554 | * Re-compute the task group their per cpu shares over the given domain. |
1555 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1556 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1557 | */ |
eb755805 | 1558 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1559 | { |
ec4e0e2f | 1560 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1561 | unsigned long shares = 0; |
eb755805 | 1562 | struct sched_domain *sd = data; |
c8cba857 | 1563 | int i; |
c09595f6 | 1564 | |
758b2cdc | 1565 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1566 | /* |
1567 | * If there are currently no tasks on the cpu pretend there | |
1568 | * is one of average load so that when a new task gets to | |
1569 | * run here it will not get delayed by group starvation. | |
1570 | */ | |
1571 | weight = tg->cfs_rq[i]->load.weight; | |
1572 | if (!weight) | |
1573 | weight = NICE_0_LOAD; | |
1574 | ||
1575 | tg->cfs_rq[i]->rq_weight = weight; | |
1576 | rq_weight += weight; | |
c8cba857 | 1577 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1578 | } |
c09595f6 | 1579 | |
c8cba857 PZ |
1580 | if ((!shares && rq_weight) || shares > tg->shares) |
1581 | shares = tg->shares; | |
1582 | ||
1583 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1584 | shares = tg->shares; | |
c09595f6 | 1585 | |
758b2cdc | 1586 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1587 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1588 | |
1589 | return 0; | |
c09595f6 PZ |
1590 | } |
1591 | ||
1592 | /* | |
c8cba857 PZ |
1593 | * Compute the cpu's hierarchical load factor for each task group. |
1594 | * This needs to be done in a top-down fashion because the load of a child | |
1595 | * group is a fraction of its parents load. | |
c09595f6 | 1596 | */ |
eb755805 | 1597 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1598 | { |
c8cba857 | 1599 | unsigned long load; |
eb755805 | 1600 | long cpu = (long)data; |
c09595f6 | 1601 | |
c8cba857 PZ |
1602 | if (!tg->parent) { |
1603 | load = cpu_rq(cpu)->load.weight; | |
1604 | } else { | |
1605 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1606 | load *= tg->cfs_rq[cpu]->shares; | |
1607 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1608 | } | |
c09595f6 | 1609 | |
c8cba857 | 1610 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1611 | |
eb755805 | 1612 | return 0; |
c09595f6 PZ |
1613 | } |
1614 | ||
c8cba857 | 1615 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1616 | { |
2398f2c6 PZ |
1617 | u64 now = cpu_clock(raw_smp_processor_id()); |
1618 | s64 elapsed = now - sd->last_update; | |
1619 | ||
1620 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1621 | sd->last_update = now; | |
eb755805 | 1622 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1623 | } |
4d8d595d PZ |
1624 | } |
1625 | ||
3e5459b4 PZ |
1626 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1627 | { | |
1628 | spin_unlock(&rq->lock); | |
1629 | update_shares(sd); | |
1630 | spin_lock(&rq->lock); | |
1631 | } | |
1632 | ||
eb755805 | 1633 | static void update_h_load(long cpu) |
c09595f6 | 1634 | { |
eb755805 | 1635 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1636 | } |
1637 | ||
c09595f6 PZ |
1638 | #else |
1639 | ||
c8cba857 | 1640 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1641 | { |
1642 | } | |
1643 | ||
3e5459b4 PZ |
1644 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1645 | { | |
1646 | } | |
1647 | ||
18d95a28 PZ |
1648 | #endif |
1649 | ||
8f45e2b5 GH |
1650 | #ifdef CONFIG_PREEMPT |
1651 | ||
70574a99 | 1652 | /* |
8f45e2b5 GH |
1653 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1654 | * way at the expense of forcing extra atomic operations in all | |
1655 | * invocations. This assures that the double_lock is acquired using the | |
1656 | * same underlying policy as the spinlock_t on this architecture, which | |
1657 | * reduces latency compared to the unfair variant below. However, it | |
1658 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1659 | */ |
8f45e2b5 GH |
1660 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1661 | __releases(this_rq->lock) | |
1662 | __acquires(busiest->lock) | |
1663 | __acquires(this_rq->lock) | |
1664 | { | |
1665 | spin_unlock(&this_rq->lock); | |
1666 | double_rq_lock(this_rq, busiest); | |
1667 | ||
1668 | return 1; | |
1669 | } | |
1670 | ||
1671 | #else | |
1672 | /* | |
1673 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1674 | * latency by eliminating extra atomic operations when the locks are | |
1675 | * already in proper order on entry. This favors lower cpu-ids and will | |
1676 | * grant the double lock to lower cpus over higher ids under contention, | |
1677 | * regardless of entry order into the function. | |
1678 | */ | |
1679 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1680 | __releases(this_rq->lock) |
1681 | __acquires(busiest->lock) | |
1682 | __acquires(this_rq->lock) | |
1683 | { | |
1684 | int ret = 0; | |
1685 | ||
70574a99 AD |
1686 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1687 | if (busiest < this_rq) { | |
1688 | spin_unlock(&this_rq->lock); | |
1689 | spin_lock(&busiest->lock); | |
1690 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1691 | ret = 1; | |
1692 | } else | |
1693 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1694 | } | |
1695 | return ret; | |
1696 | } | |
1697 | ||
8f45e2b5 GH |
1698 | #endif /* CONFIG_PREEMPT */ |
1699 | ||
1700 | /* | |
1701 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1702 | */ | |
1703 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1704 | { | |
1705 | if (unlikely(!irqs_disabled())) { | |
1706 | /* printk() doesn't work good under rq->lock */ | |
1707 | spin_unlock(&this_rq->lock); | |
1708 | BUG_ON(1); | |
1709 | } | |
1710 | ||
1711 | return _double_lock_balance(this_rq, busiest); | |
1712 | } | |
1713 | ||
70574a99 AD |
1714 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1715 | __releases(busiest->lock) | |
1716 | { | |
1717 | spin_unlock(&busiest->lock); | |
1718 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1719 | } | |
18d95a28 PZ |
1720 | #endif |
1721 | ||
30432094 | 1722 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1723 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1724 | { | |
30432094 | 1725 | #ifdef CONFIG_SMP |
34e83e85 IM |
1726 | cfs_rq->shares = shares; |
1727 | #endif | |
1728 | } | |
30432094 | 1729 | #endif |
e7693a36 | 1730 | |
dce48a84 TG |
1731 | static void calc_load_account_active(struct rq *this_rq); |
1732 | ||
dd41f596 | 1733 | #include "sched_stats.h" |
dd41f596 | 1734 | #include "sched_idletask.c" |
5522d5d5 IM |
1735 | #include "sched_fair.c" |
1736 | #include "sched_rt.c" | |
dd41f596 IM |
1737 | #ifdef CONFIG_SCHED_DEBUG |
1738 | # include "sched_debug.c" | |
1739 | #endif | |
1740 | ||
1741 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1742 | #define for_each_class(class) \ |
1743 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1744 | |
c09595f6 | 1745 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1746 | { |
1747 | rq->nr_running++; | |
9c217245 IM |
1748 | } |
1749 | ||
c09595f6 | 1750 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1751 | { |
1752 | rq->nr_running--; | |
9c217245 IM |
1753 | } |
1754 | ||
45bf76df IM |
1755 | static void set_load_weight(struct task_struct *p) |
1756 | { | |
1757 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1758 | p->se.load.weight = prio_to_weight[0] * 2; |
1759 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1760 | return; | |
1761 | } | |
45bf76df | 1762 | |
dd41f596 IM |
1763 | /* |
1764 | * SCHED_IDLE tasks get minimal weight: | |
1765 | */ | |
1766 | if (p->policy == SCHED_IDLE) { | |
1767 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1768 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1769 | return; | |
1770 | } | |
71f8bd46 | 1771 | |
dd41f596 IM |
1772 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1773 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1774 | } |
1775 | ||
2087a1ad GH |
1776 | static void update_avg(u64 *avg, u64 sample) |
1777 | { | |
1778 | s64 diff = sample - *avg; | |
1779 | *avg += diff >> 3; | |
1780 | } | |
1781 | ||
8159f87e | 1782 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1783 | { |
831451ac PZ |
1784 | if (wakeup) |
1785 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1786 | ||
dd41f596 | 1787 | sched_info_queued(p); |
fd390f6a | 1788 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1789 | p->se.on_rq = 1; |
71f8bd46 IM |
1790 | } |
1791 | ||
69be72c1 | 1792 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1793 | { |
831451ac PZ |
1794 | if (sleep) { |
1795 | if (p->se.last_wakeup) { | |
1796 | update_avg(&p->se.avg_overlap, | |
1797 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1798 | p->se.last_wakeup = 0; | |
1799 | } else { | |
1800 | update_avg(&p->se.avg_wakeup, | |
1801 | sysctl_sched_wakeup_granularity); | |
1802 | } | |
2087a1ad GH |
1803 | } |
1804 | ||
46ac22ba | 1805 | sched_info_dequeued(p); |
f02231e5 | 1806 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1807 | p->se.on_rq = 0; |
71f8bd46 IM |
1808 | } |
1809 | ||
14531189 | 1810 | /* |
dd41f596 | 1811 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1812 | */ |
14531189 IM |
1813 | static inline int __normal_prio(struct task_struct *p) |
1814 | { | |
dd41f596 | 1815 | return p->static_prio; |
14531189 IM |
1816 | } |
1817 | ||
b29739f9 IM |
1818 | /* |
1819 | * Calculate the expected normal priority: i.e. priority | |
1820 | * without taking RT-inheritance into account. Might be | |
1821 | * boosted by interactivity modifiers. Changes upon fork, | |
1822 | * setprio syscalls, and whenever the interactivity | |
1823 | * estimator recalculates. | |
1824 | */ | |
36c8b586 | 1825 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1826 | { |
1827 | int prio; | |
1828 | ||
e05606d3 | 1829 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1830 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1831 | else | |
1832 | prio = __normal_prio(p); | |
1833 | return prio; | |
1834 | } | |
1835 | ||
1836 | /* | |
1837 | * Calculate the current priority, i.e. the priority | |
1838 | * taken into account by the scheduler. This value might | |
1839 | * be boosted by RT tasks, or might be boosted by | |
1840 | * interactivity modifiers. Will be RT if the task got | |
1841 | * RT-boosted. If not then it returns p->normal_prio. | |
1842 | */ | |
36c8b586 | 1843 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1844 | { |
1845 | p->normal_prio = normal_prio(p); | |
1846 | /* | |
1847 | * If we are RT tasks or we were boosted to RT priority, | |
1848 | * keep the priority unchanged. Otherwise, update priority | |
1849 | * to the normal priority: | |
1850 | */ | |
1851 | if (!rt_prio(p->prio)) | |
1852 | return p->normal_prio; | |
1853 | return p->prio; | |
1854 | } | |
1855 | ||
1da177e4 | 1856 | /* |
dd41f596 | 1857 | * activate_task - move a task to the runqueue. |
1da177e4 | 1858 | */ |
dd41f596 | 1859 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1860 | { |
d9514f6c | 1861 | if (task_contributes_to_load(p)) |
dd41f596 | 1862 | rq->nr_uninterruptible--; |
1da177e4 | 1863 | |
8159f87e | 1864 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1865 | inc_nr_running(rq); |
1da177e4 LT |
1866 | } |
1867 | ||
1da177e4 LT |
1868 | /* |
1869 | * deactivate_task - remove a task from the runqueue. | |
1870 | */ | |
2e1cb74a | 1871 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1872 | { |
d9514f6c | 1873 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1874 | rq->nr_uninterruptible++; |
1875 | ||
69be72c1 | 1876 | dequeue_task(rq, p, sleep); |
c09595f6 | 1877 | dec_nr_running(rq); |
1da177e4 LT |
1878 | } |
1879 | ||
1da177e4 LT |
1880 | /** |
1881 | * task_curr - is this task currently executing on a CPU? | |
1882 | * @p: the task in question. | |
1883 | */ | |
36c8b586 | 1884 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1885 | { |
1886 | return cpu_curr(task_cpu(p)) == p; | |
1887 | } | |
1888 | ||
dd41f596 IM |
1889 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1890 | { | |
6f505b16 | 1891 | set_task_rq(p, cpu); |
dd41f596 | 1892 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1893 | /* |
1894 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1895 | * successfuly executed on another CPU. We must ensure that updates of | |
1896 | * per-task data have been completed by this moment. | |
1897 | */ | |
1898 | smp_wmb(); | |
dd41f596 | 1899 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1900 | #endif |
2dd73a4f PW |
1901 | } |
1902 | ||
cb469845 SR |
1903 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1904 | const struct sched_class *prev_class, | |
1905 | int oldprio, int running) | |
1906 | { | |
1907 | if (prev_class != p->sched_class) { | |
1908 | if (prev_class->switched_from) | |
1909 | prev_class->switched_from(rq, p, running); | |
1910 | p->sched_class->switched_to(rq, p, running); | |
1911 | } else | |
1912 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1913 | } | |
1914 | ||
1da177e4 | 1915 | #ifdef CONFIG_SMP |
c65cc870 | 1916 | |
e958b360 TG |
1917 | /* Used instead of source_load when we know the type == 0 */ |
1918 | static unsigned long weighted_cpuload(const int cpu) | |
1919 | { | |
1920 | return cpu_rq(cpu)->load.weight; | |
1921 | } | |
1922 | ||
cc367732 IM |
1923 | /* |
1924 | * Is this task likely cache-hot: | |
1925 | */ | |
e7693a36 | 1926 | static int |
cc367732 IM |
1927 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1928 | { | |
1929 | s64 delta; | |
1930 | ||
f540a608 IM |
1931 | /* |
1932 | * Buddy candidates are cache hot: | |
1933 | */ | |
4793241b PZ |
1934 | if (sched_feat(CACHE_HOT_BUDDY) && |
1935 | (&p->se == cfs_rq_of(&p->se)->next || | |
1936 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1937 | return 1; |
1938 | ||
cc367732 IM |
1939 | if (p->sched_class != &fair_sched_class) |
1940 | return 0; | |
1941 | ||
6bc1665b IM |
1942 | if (sysctl_sched_migration_cost == -1) |
1943 | return 1; | |
1944 | if (sysctl_sched_migration_cost == 0) | |
1945 | return 0; | |
1946 | ||
cc367732 IM |
1947 | delta = now - p->se.exec_start; |
1948 | ||
1949 | return delta < (s64)sysctl_sched_migration_cost; | |
1950 | } | |
1951 | ||
1952 | ||
dd41f596 | 1953 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1954 | { |
dd41f596 IM |
1955 | int old_cpu = task_cpu(p); |
1956 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1957 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1958 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1959 | u64 clock_offset; |
dd41f596 IM |
1960 | |
1961 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1962 | |
de1d7286 | 1963 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1964 | |
6cfb0d5d IM |
1965 | #ifdef CONFIG_SCHEDSTATS |
1966 | if (p->se.wait_start) | |
1967 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1968 | if (p->se.sleep_start) |
1969 | p->se.sleep_start -= clock_offset; | |
1970 | if (p->se.block_start) | |
1971 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1972 | if (old_cpu != new_cpu) { |
1973 | schedstat_inc(p, se.nr_migrations); | |
1974 | if (task_hot(p, old_rq->clock, NULL)) | |
1975 | schedstat_inc(p, se.nr_forced2_migrations); | |
1976 | } | |
6cfb0d5d | 1977 | #endif |
2830cf8c SV |
1978 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1979 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1980 | |
1981 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1982 | } |
1983 | ||
70b97a7f | 1984 | struct migration_req { |
1da177e4 | 1985 | struct list_head list; |
1da177e4 | 1986 | |
36c8b586 | 1987 | struct task_struct *task; |
1da177e4 LT |
1988 | int dest_cpu; |
1989 | ||
1da177e4 | 1990 | struct completion done; |
70b97a7f | 1991 | }; |
1da177e4 LT |
1992 | |
1993 | /* | |
1994 | * The task's runqueue lock must be held. | |
1995 | * Returns true if you have to wait for migration thread. | |
1996 | */ | |
36c8b586 | 1997 | static int |
70b97a7f | 1998 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1999 | { |
70b97a7f | 2000 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2001 | |
2002 | /* | |
2003 | * If the task is not on a runqueue (and not running), then | |
2004 | * it is sufficient to simply update the task's cpu field. | |
2005 | */ | |
dd41f596 | 2006 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2007 | set_task_cpu(p, dest_cpu); |
2008 | return 0; | |
2009 | } | |
2010 | ||
2011 | init_completion(&req->done); | |
1da177e4 LT |
2012 | req->task = p; |
2013 | req->dest_cpu = dest_cpu; | |
2014 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2015 | |
1da177e4 LT |
2016 | return 1; |
2017 | } | |
2018 | ||
a26b89f0 MM |
2019 | /* |
2020 | * wait_task_context_switch - wait for a thread to complete at least one | |
2021 | * context switch. | |
2022 | * | |
2023 | * @p must not be current. | |
2024 | */ | |
2025 | void wait_task_context_switch(struct task_struct *p) | |
2026 | { | |
2027 | unsigned long nvcsw, nivcsw, flags; | |
2028 | int running; | |
2029 | struct rq *rq; | |
2030 | ||
2031 | nvcsw = p->nvcsw; | |
2032 | nivcsw = p->nivcsw; | |
2033 | for (;;) { | |
2034 | /* | |
2035 | * The runqueue is assigned before the actual context | |
2036 | * switch. We need to take the runqueue lock. | |
2037 | * | |
2038 | * We could check initially without the lock but it is | |
2039 | * very likely that we need to take the lock in every | |
2040 | * iteration. | |
2041 | */ | |
2042 | rq = task_rq_lock(p, &flags); | |
2043 | running = task_running(rq, p); | |
2044 | task_rq_unlock(rq, &flags); | |
2045 | ||
2046 | if (likely(!running)) | |
2047 | break; | |
2048 | /* | |
2049 | * The switch count is incremented before the actual | |
2050 | * context switch. We thus wait for two switches to be | |
2051 | * sure at least one completed. | |
2052 | */ | |
2053 | if ((p->nvcsw - nvcsw) > 1) | |
2054 | break; | |
2055 | if ((p->nivcsw - nivcsw) > 1) | |
2056 | break; | |
2057 | ||
2058 | cpu_relax(); | |
2059 | } | |
2060 | } | |
2061 | ||
1da177e4 LT |
2062 | /* |
2063 | * wait_task_inactive - wait for a thread to unschedule. | |
2064 | * | |
85ba2d86 RM |
2065 | * If @match_state is nonzero, it's the @p->state value just checked and |
2066 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2067 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2068 | * we return a positive number (its total switch count). If a second call | |
2069 | * a short while later returns the same number, the caller can be sure that | |
2070 | * @p has remained unscheduled the whole time. | |
2071 | * | |
1da177e4 LT |
2072 | * The caller must ensure that the task *will* unschedule sometime soon, |
2073 | * else this function might spin for a *long* time. This function can't | |
2074 | * be called with interrupts off, or it may introduce deadlock with | |
2075 | * smp_call_function() if an IPI is sent by the same process we are | |
2076 | * waiting to become inactive. | |
2077 | */ | |
85ba2d86 | 2078 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2079 | { |
2080 | unsigned long flags; | |
dd41f596 | 2081 | int running, on_rq; |
85ba2d86 | 2082 | unsigned long ncsw; |
70b97a7f | 2083 | struct rq *rq; |
1da177e4 | 2084 | |
3a5c359a AK |
2085 | for (;;) { |
2086 | /* | |
2087 | * We do the initial early heuristics without holding | |
2088 | * any task-queue locks at all. We'll only try to get | |
2089 | * the runqueue lock when things look like they will | |
2090 | * work out! | |
2091 | */ | |
2092 | rq = task_rq(p); | |
fa490cfd | 2093 | |
3a5c359a AK |
2094 | /* |
2095 | * If the task is actively running on another CPU | |
2096 | * still, just relax and busy-wait without holding | |
2097 | * any locks. | |
2098 | * | |
2099 | * NOTE! Since we don't hold any locks, it's not | |
2100 | * even sure that "rq" stays as the right runqueue! | |
2101 | * But we don't care, since "task_running()" will | |
2102 | * return false if the runqueue has changed and p | |
2103 | * is actually now running somewhere else! | |
2104 | */ | |
85ba2d86 RM |
2105 | while (task_running(rq, p)) { |
2106 | if (match_state && unlikely(p->state != match_state)) | |
2107 | return 0; | |
3a5c359a | 2108 | cpu_relax(); |
85ba2d86 | 2109 | } |
fa490cfd | 2110 | |
3a5c359a AK |
2111 | /* |
2112 | * Ok, time to look more closely! We need the rq | |
2113 | * lock now, to be *sure*. If we're wrong, we'll | |
2114 | * just go back and repeat. | |
2115 | */ | |
2116 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2117 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2118 | running = task_running(rq, p); |
2119 | on_rq = p->se.on_rq; | |
85ba2d86 | 2120 | ncsw = 0; |
f31e11d8 | 2121 | if (!match_state || p->state == match_state) |
93dcf55f | 2122 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2123 | task_rq_unlock(rq, &flags); |
fa490cfd | 2124 | |
85ba2d86 RM |
2125 | /* |
2126 | * If it changed from the expected state, bail out now. | |
2127 | */ | |
2128 | if (unlikely(!ncsw)) | |
2129 | break; | |
2130 | ||
3a5c359a AK |
2131 | /* |
2132 | * Was it really running after all now that we | |
2133 | * checked with the proper locks actually held? | |
2134 | * | |
2135 | * Oops. Go back and try again.. | |
2136 | */ | |
2137 | if (unlikely(running)) { | |
2138 | cpu_relax(); | |
2139 | continue; | |
2140 | } | |
fa490cfd | 2141 | |
3a5c359a AK |
2142 | /* |
2143 | * It's not enough that it's not actively running, | |
2144 | * it must be off the runqueue _entirely_, and not | |
2145 | * preempted! | |
2146 | * | |
80dd99b3 | 2147 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2148 | * running right now), it's preempted, and we should |
2149 | * yield - it could be a while. | |
2150 | */ | |
2151 | if (unlikely(on_rq)) { | |
2152 | schedule_timeout_uninterruptible(1); | |
2153 | continue; | |
2154 | } | |
fa490cfd | 2155 | |
3a5c359a AK |
2156 | /* |
2157 | * Ahh, all good. It wasn't running, and it wasn't | |
2158 | * runnable, which means that it will never become | |
2159 | * running in the future either. We're all done! | |
2160 | */ | |
2161 | break; | |
2162 | } | |
85ba2d86 RM |
2163 | |
2164 | return ncsw; | |
1da177e4 LT |
2165 | } |
2166 | ||
2167 | /*** | |
2168 | * kick_process - kick a running thread to enter/exit the kernel | |
2169 | * @p: the to-be-kicked thread | |
2170 | * | |
2171 | * Cause a process which is running on another CPU to enter | |
2172 | * kernel-mode, without any delay. (to get signals handled.) | |
2173 | * | |
2174 | * NOTE: this function doesnt have to take the runqueue lock, | |
2175 | * because all it wants to ensure is that the remote task enters | |
2176 | * the kernel. If the IPI races and the task has been migrated | |
2177 | * to another CPU then no harm is done and the purpose has been | |
2178 | * achieved as well. | |
2179 | */ | |
36c8b586 | 2180 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2181 | { |
2182 | int cpu; | |
2183 | ||
2184 | preempt_disable(); | |
2185 | cpu = task_cpu(p); | |
2186 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2187 | smp_send_reschedule(cpu); | |
2188 | preempt_enable(); | |
2189 | } | |
2190 | ||
2191 | /* | |
2dd73a4f PW |
2192 | * Return a low guess at the load of a migration-source cpu weighted |
2193 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2194 | * |
2195 | * We want to under-estimate the load of migration sources, to | |
2196 | * balance conservatively. | |
2197 | */ | |
a9957449 | 2198 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2199 | { |
70b97a7f | 2200 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2201 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2202 | |
93b75217 | 2203 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2204 | return total; |
b910472d | 2205 | |
dd41f596 | 2206 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2207 | } |
2208 | ||
2209 | /* | |
2dd73a4f PW |
2210 | * Return a high guess at the load of a migration-target cpu weighted |
2211 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2212 | */ |
a9957449 | 2213 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2214 | { |
70b97a7f | 2215 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2216 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2217 | |
93b75217 | 2218 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2219 | return total; |
3b0bd9bc | 2220 | |
dd41f596 | 2221 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2222 | } |
2223 | ||
147cbb4b NP |
2224 | /* |
2225 | * find_idlest_group finds and returns the least busy CPU group within the | |
2226 | * domain. | |
2227 | */ | |
2228 | static struct sched_group * | |
2229 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2230 | { | |
2231 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2232 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2233 | int load_idx = sd->forkexec_idx; | |
2234 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2235 | ||
2236 | do { | |
2237 | unsigned long load, avg_load; | |
2238 | int local_group; | |
2239 | int i; | |
2240 | ||
da5a5522 | 2241 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2242 | if (!cpumask_intersects(sched_group_cpus(group), |
2243 | &p->cpus_allowed)) | |
3a5c359a | 2244 | continue; |
da5a5522 | 2245 | |
758b2cdc RR |
2246 | local_group = cpumask_test_cpu(this_cpu, |
2247 | sched_group_cpus(group)); | |
147cbb4b NP |
2248 | |
2249 | /* Tally up the load of all CPUs in the group */ | |
2250 | avg_load = 0; | |
2251 | ||
758b2cdc | 2252 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2253 | /* Bias balancing toward cpus of our domain */ |
2254 | if (local_group) | |
2255 | load = source_load(i, load_idx); | |
2256 | else | |
2257 | load = target_load(i, load_idx); | |
2258 | ||
2259 | avg_load += load; | |
2260 | } | |
2261 | ||
2262 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2263 | avg_load = sg_div_cpu_power(group, |
2264 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2265 | |
2266 | if (local_group) { | |
2267 | this_load = avg_load; | |
2268 | this = group; | |
2269 | } else if (avg_load < min_load) { | |
2270 | min_load = avg_load; | |
2271 | idlest = group; | |
2272 | } | |
3a5c359a | 2273 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2274 | |
2275 | if (!idlest || 100*this_load < imbalance*min_load) | |
2276 | return NULL; | |
2277 | return idlest; | |
2278 | } | |
2279 | ||
2280 | /* | |
0feaece9 | 2281 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2282 | */ |
95cdf3b7 | 2283 | static int |
758b2cdc | 2284 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2285 | { |
2286 | unsigned long load, min_load = ULONG_MAX; | |
2287 | int idlest = -1; | |
2288 | int i; | |
2289 | ||
da5a5522 | 2290 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2291 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2292 | load = weighted_cpuload(i); |
147cbb4b NP |
2293 | |
2294 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2295 | min_load = load; | |
2296 | idlest = i; | |
2297 | } | |
2298 | } | |
2299 | ||
2300 | return idlest; | |
2301 | } | |
2302 | ||
476d139c NP |
2303 | /* |
2304 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2305 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2306 | * SD_BALANCE_EXEC. | |
2307 | * | |
2308 | * Balance, ie. select the least loaded group. | |
2309 | * | |
2310 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2311 | * | |
2312 | * preempt must be disabled. | |
2313 | */ | |
2314 | static int sched_balance_self(int cpu, int flag) | |
2315 | { | |
2316 | struct task_struct *t = current; | |
2317 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2318 | |
c96d145e | 2319 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2320 | /* |
2321 | * If power savings logic is enabled for a domain, stop there. | |
2322 | */ | |
5c45bf27 SS |
2323 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2324 | break; | |
476d139c NP |
2325 | if (tmp->flags & flag) |
2326 | sd = tmp; | |
c96d145e | 2327 | } |
476d139c | 2328 | |
039a1c41 PZ |
2329 | if (sd) |
2330 | update_shares(sd); | |
2331 | ||
476d139c | 2332 | while (sd) { |
476d139c | 2333 | struct sched_group *group; |
1a848870 SS |
2334 | int new_cpu, weight; |
2335 | ||
2336 | if (!(sd->flags & flag)) { | |
2337 | sd = sd->child; | |
2338 | continue; | |
2339 | } | |
476d139c | 2340 | |
476d139c | 2341 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2342 | if (!group) { |
2343 | sd = sd->child; | |
2344 | continue; | |
2345 | } | |
476d139c | 2346 | |
758b2cdc | 2347 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2348 | if (new_cpu == -1 || new_cpu == cpu) { |
2349 | /* Now try balancing at a lower domain level of cpu */ | |
2350 | sd = sd->child; | |
2351 | continue; | |
2352 | } | |
476d139c | 2353 | |
1a848870 | 2354 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2355 | cpu = new_cpu; |
758b2cdc | 2356 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2357 | sd = NULL; |
476d139c | 2358 | for_each_domain(cpu, tmp) { |
758b2cdc | 2359 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2360 | break; |
2361 | if (tmp->flags & flag) | |
2362 | sd = tmp; | |
2363 | } | |
2364 | /* while loop will break here if sd == NULL */ | |
2365 | } | |
2366 | ||
2367 | return cpu; | |
2368 | } | |
2369 | ||
2370 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2371 | |
1da177e4 LT |
2372 | /*** |
2373 | * try_to_wake_up - wake up a thread | |
2374 | * @p: the to-be-woken-up thread | |
2375 | * @state: the mask of task states that can be woken | |
2376 | * @sync: do a synchronous wakeup? | |
2377 | * | |
2378 | * Put it on the run-queue if it's not already there. The "current" | |
2379 | * thread is always on the run-queue (except when the actual | |
2380 | * re-schedule is in progress), and as such you're allowed to do | |
2381 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2382 | * runnable without the overhead of this. | |
2383 | * | |
2384 | * returns failure only if the task is already active. | |
2385 | */ | |
36c8b586 | 2386 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2387 | { |
cc367732 | 2388 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2389 | unsigned long flags; |
2390 | long old_state; | |
70b97a7f | 2391 | struct rq *rq; |
1da177e4 | 2392 | |
b85d0667 IM |
2393 | if (!sched_feat(SYNC_WAKEUPS)) |
2394 | sync = 0; | |
2395 | ||
2398f2c6 | 2396 | #ifdef CONFIG_SMP |
57310a98 | 2397 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2398 | struct sched_domain *sd; |
2399 | ||
2400 | this_cpu = raw_smp_processor_id(); | |
2401 | cpu = task_cpu(p); | |
2402 | ||
2403 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2404 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2405 | update_shares(sd); |
2406 | break; | |
2407 | } | |
2408 | } | |
2409 | } | |
2410 | #endif | |
2411 | ||
04e2f174 | 2412 | smp_wmb(); |
1da177e4 | 2413 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2414 | update_rq_clock(rq); |
1da177e4 LT |
2415 | old_state = p->state; |
2416 | if (!(old_state & state)) | |
2417 | goto out; | |
2418 | ||
dd41f596 | 2419 | if (p->se.on_rq) |
1da177e4 LT |
2420 | goto out_running; |
2421 | ||
2422 | cpu = task_cpu(p); | |
cc367732 | 2423 | orig_cpu = cpu; |
1da177e4 LT |
2424 | this_cpu = smp_processor_id(); |
2425 | ||
2426 | #ifdef CONFIG_SMP | |
2427 | if (unlikely(task_running(rq, p))) | |
2428 | goto out_activate; | |
2429 | ||
5d2f5a61 DA |
2430 | cpu = p->sched_class->select_task_rq(p, sync); |
2431 | if (cpu != orig_cpu) { | |
2432 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2433 | task_rq_unlock(rq, &flags); |
2434 | /* might preempt at this point */ | |
2435 | rq = task_rq_lock(p, &flags); | |
2436 | old_state = p->state; | |
2437 | if (!(old_state & state)) | |
2438 | goto out; | |
dd41f596 | 2439 | if (p->se.on_rq) |
1da177e4 LT |
2440 | goto out_running; |
2441 | ||
2442 | this_cpu = smp_processor_id(); | |
2443 | cpu = task_cpu(p); | |
2444 | } | |
2445 | ||
e7693a36 GH |
2446 | #ifdef CONFIG_SCHEDSTATS |
2447 | schedstat_inc(rq, ttwu_count); | |
2448 | if (cpu == this_cpu) | |
2449 | schedstat_inc(rq, ttwu_local); | |
2450 | else { | |
2451 | struct sched_domain *sd; | |
2452 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2453 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2454 | schedstat_inc(sd, ttwu_wake_remote); |
2455 | break; | |
2456 | } | |
2457 | } | |
2458 | } | |
6d6bc0ad | 2459 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2460 | |
1da177e4 LT |
2461 | out_activate: |
2462 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2463 | schedstat_inc(p, se.nr_wakeups); |
2464 | if (sync) | |
2465 | schedstat_inc(p, se.nr_wakeups_sync); | |
2466 | if (orig_cpu != cpu) | |
2467 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2468 | if (cpu == this_cpu) | |
2469 | schedstat_inc(p, se.nr_wakeups_local); | |
2470 | else | |
2471 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2472 | activate_task(rq, p, 1); |
1da177e4 LT |
2473 | success = 1; |
2474 | ||
831451ac PZ |
2475 | /* |
2476 | * Only attribute actual wakeups done by this task. | |
2477 | */ | |
2478 | if (!in_interrupt()) { | |
2479 | struct sched_entity *se = ¤t->se; | |
2480 | u64 sample = se->sum_exec_runtime; | |
2481 | ||
2482 | if (se->last_wakeup) | |
2483 | sample -= se->last_wakeup; | |
2484 | else | |
2485 | sample -= se->start_runtime; | |
2486 | update_avg(&se->avg_wakeup, sample); | |
2487 | ||
2488 | se->last_wakeup = se->sum_exec_runtime; | |
2489 | } | |
2490 | ||
1da177e4 | 2491 | out_running: |
468a15bb | 2492 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2493 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2494 | |
1da177e4 | 2495 | p->state = TASK_RUNNING; |
9a897c5a SR |
2496 | #ifdef CONFIG_SMP |
2497 | if (p->sched_class->task_wake_up) | |
2498 | p->sched_class->task_wake_up(rq, p); | |
2499 | #endif | |
1da177e4 LT |
2500 | out: |
2501 | task_rq_unlock(rq, &flags); | |
2502 | ||
2503 | return success; | |
2504 | } | |
2505 | ||
50fa610a DH |
2506 | /** |
2507 | * wake_up_process - Wake up a specific process | |
2508 | * @p: The process to be woken up. | |
2509 | * | |
2510 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2511 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2512 | * running. | |
2513 | * | |
2514 | * It may be assumed that this function implies a write memory barrier before | |
2515 | * changing the task state if and only if any tasks are woken up. | |
2516 | */ | |
7ad5b3a5 | 2517 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2518 | { |
d9514f6c | 2519 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2520 | } |
1da177e4 LT |
2521 | EXPORT_SYMBOL(wake_up_process); |
2522 | ||
7ad5b3a5 | 2523 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2524 | { |
2525 | return try_to_wake_up(p, state, 0); | |
2526 | } | |
2527 | ||
1da177e4 LT |
2528 | /* |
2529 | * Perform scheduler related setup for a newly forked process p. | |
2530 | * p is forked by current. | |
dd41f596 IM |
2531 | * |
2532 | * __sched_fork() is basic setup used by init_idle() too: | |
2533 | */ | |
2534 | static void __sched_fork(struct task_struct *p) | |
2535 | { | |
dd41f596 IM |
2536 | p->se.exec_start = 0; |
2537 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2538 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2539 | p->se.last_wakeup = 0; |
2540 | p->se.avg_overlap = 0; | |
831451ac PZ |
2541 | p->se.start_runtime = 0; |
2542 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2543 | |
2544 | #ifdef CONFIG_SCHEDSTATS | |
2545 | p->se.wait_start = 0; | |
dd41f596 IM |
2546 | p->se.sum_sleep_runtime = 0; |
2547 | p->se.sleep_start = 0; | |
dd41f596 IM |
2548 | p->se.block_start = 0; |
2549 | p->se.sleep_max = 0; | |
2550 | p->se.block_max = 0; | |
2551 | p->se.exec_max = 0; | |
eba1ed4b | 2552 | p->se.slice_max = 0; |
dd41f596 | 2553 | p->se.wait_max = 0; |
6cfb0d5d | 2554 | #endif |
476d139c | 2555 | |
fa717060 | 2556 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2557 | p->se.on_rq = 0; |
4a55bd5e | 2558 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2559 | |
e107be36 AK |
2560 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2561 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2562 | #endif | |
2563 | ||
1da177e4 LT |
2564 | /* |
2565 | * We mark the process as running here, but have not actually | |
2566 | * inserted it onto the runqueue yet. This guarantees that | |
2567 | * nobody will actually run it, and a signal or other external | |
2568 | * event cannot wake it up and insert it on the runqueue either. | |
2569 | */ | |
2570 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2571 | } |
2572 | ||
2573 | /* | |
2574 | * fork()/clone()-time setup: | |
2575 | */ | |
2576 | void sched_fork(struct task_struct *p, int clone_flags) | |
2577 | { | |
2578 | int cpu = get_cpu(); | |
2579 | ||
2580 | __sched_fork(p); | |
2581 | ||
2582 | #ifdef CONFIG_SMP | |
2583 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2584 | #endif | |
02e4bac2 | 2585 | set_task_cpu(p, cpu); |
b29739f9 IM |
2586 | |
2587 | /* | |
2588 | * Make sure we do not leak PI boosting priority to the child: | |
2589 | */ | |
2590 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2591 | if (!rt_prio(p->prio)) |
2592 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2593 | |
52f17b6c | 2594 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2595 | if (likely(sched_info_on())) |
52f17b6c | 2596 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2597 | #endif |
d6077cb8 | 2598 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2599 | p->oncpu = 0; |
2600 | #endif | |
1da177e4 | 2601 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2602 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2603 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2604 | #endif |
917b627d GH |
2605 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2606 | ||
476d139c | 2607 | put_cpu(); |
1da177e4 LT |
2608 | } |
2609 | ||
2610 | /* | |
2611 | * wake_up_new_task - wake up a newly created task for the first time. | |
2612 | * | |
2613 | * This function will do some initial scheduler statistics housekeeping | |
2614 | * that must be done for every newly created context, then puts the task | |
2615 | * on the runqueue and wakes it. | |
2616 | */ | |
7ad5b3a5 | 2617 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2618 | { |
2619 | unsigned long flags; | |
dd41f596 | 2620 | struct rq *rq; |
1da177e4 LT |
2621 | |
2622 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2623 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2624 | update_rq_clock(rq); |
1da177e4 LT |
2625 | |
2626 | p->prio = effective_prio(p); | |
2627 | ||
b9dca1e0 | 2628 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2629 | activate_task(rq, p, 0); |
1da177e4 | 2630 | } else { |
1da177e4 | 2631 | /* |
dd41f596 IM |
2632 | * Let the scheduling class do new task startup |
2633 | * management (if any): | |
1da177e4 | 2634 | */ |
ee0827d8 | 2635 | p->sched_class->task_new(rq, p); |
c09595f6 | 2636 | inc_nr_running(rq); |
1da177e4 | 2637 | } |
c71dd42d | 2638 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2639 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2640 | #ifdef CONFIG_SMP |
2641 | if (p->sched_class->task_wake_up) | |
2642 | p->sched_class->task_wake_up(rq, p); | |
2643 | #endif | |
dd41f596 | 2644 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2645 | } |
2646 | ||
e107be36 AK |
2647 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2648 | ||
2649 | /** | |
80dd99b3 | 2650 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2651 | * @notifier: notifier struct to register |
e107be36 AK |
2652 | */ |
2653 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2654 | { | |
2655 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2656 | } | |
2657 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2658 | ||
2659 | /** | |
2660 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2661 | * @notifier: notifier struct to unregister |
e107be36 AK |
2662 | * |
2663 | * This is safe to call from within a preemption notifier. | |
2664 | */ | |
2665 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2666 | { | |
2667 | hlist_del(¬ifier->link); | |
2668 | } | |
2669 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2670 | ||
2671 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2672 | { | |
2673 | struct preempt_notifier *notifier; | |
2674 | struct hlist_node *node; | |
2675 | ||
2676 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2677 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2678 | } | |
2679 | ||
2680 | static void | |
2681 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2682 | struct task_struct *next) | |
2683 | { | |
2684 | struct preempt_notifier *notifier; | |
2685 | struct hlist_node *node; | |
2686 | ||
2687 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2688 | notifier->ops->sched_out(notifier, next); | |
2689 | } | |
2690 | ||
6d6bc0ad | 2691 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2692 | |
2693 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2694 | { | |
2695 | } | |
2696 | ||
2697 | static void | |
2698 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2699 | struct task_struct *next) | |
2700 | { | |
2701 | } | |
2702 | ||
6d6bc0ad | 2703 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2704 | |
4866cde0 NP |
2705 | /** |
2706 | * prepare_task_switch - prepare to switch tasks | |
2707 | * @rq: the runqueue preparing to switch | |
421cee29 | 2708 | * @prev: the current task that is being switched out |
4866cde0 NP |
2709 | * @next: the task we are going to switch to. |
2710 | * | |
2711 | * This is called with the rq lock held and interrupts off. It must | |
2712 | * be paired with a subsequent finish_task_switch after the context | |
2713 | * switch. | |
2714 | * | |
2715 | * prepare_task_switch sets up locking and calls architecture specific | |
2716 | * hooks. | |
2717 | */ | |
e107be36 AK |
2718 | static inline void |
2719 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2720 | struct task_struct *next) | |
4866cde0 | 2721 | { |
e107be36 | 2722 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2723 | prepare_lock_switch(rq, next); |
2724 | prepare_arch_switch(next); | |
2725 | } | |
2726 | ||
1da177e4 LT |
2727 | /** |
2728 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2729 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2730 | * @prev: the thread we just switched away from. |
2731 | * | |
4866cde0 NP |
2732 | * finish_task_switch must be called after the context switch, paired |
2733 | * with a prepare_task_switch call before the context switch. | |
2734 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2735 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2736 | * |
2737 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2738 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2739 | * with the lock held can cause deadlocks; see schedule() for |
2740 | * details.) | |
2741 | */ | |
a9957449 | 2742 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2743 | __releases(rq->lock) |
2744 | { | |
1da177e4 | 2745 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2746 | long prev_state; |
967fc046 GH |
2747 | #ifdef CONFIG_SMP |
2748 | int post_schedule = 0; | |
2749 | ||
2750 | if (current->sched_class->needs_post_schedule) | |
2751 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2752 | #endif | |
1da177e4 LT |
2753 | |
2754 | rq->prev_mm = NULL; | |
2755 | ||
2756 | /* | |
2757 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2758 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2759 | * schedule one last time. The schedule call will never return, and |
2760 | * the scheduled task must drop that reference. | |
c394cc9f | 2761 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2762 | * still held, otherwise prev could be scheduled on another cpu, die |
2763 | * there before we look at prev->state, and then the reference would | |
2764 | * be dropped twice. | |
2765 | * Manfred Spraul <manfred@colorfullife.com> | |
2766 | */ | |
55a101f8 | 2767 | prev_state = prev->state; |
4866cde0 NP |
2768 | finish_arch_switch(prev); |
2769 | finish_lock_switch(rq, prev); | |
9a897c5a | 2770 | #ifdef CONFIG_SMP |
967fc046 | 2771 | if (post_schedule) |
9a897c5a SR |
2772 | current->sched_class->post_schedule(rq); |
2773 | #endif | |
e8fa1362 | 2774 | |
e107be36 | 2775 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2776 | if (mm) |
2777 | mmdrop(mm); | |
c394cc9f | 2778 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2779 | /* |
2780 | * Remove function-return probe instances associated with this | |
2781 | * task and put them back on the free list. | |
9761eea8 | 2782 | */ |
c6fd91f0 | 2783 | kprobe_flush_task(prev); |
1da177e4 | 2784 | put_task_struct(prev); |
c6fd91f0 | 2785 | } |
1da177e4 LT |
2786 | } |
2787 | ||
2788 | /** | |
2789 | * schedule_tail - first thing a freshly forked thread must call. | |
2790 | * @prev: the thread we just switched away from. | |
2791 | */ | |
36c8b586 | 2792 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2793 | __releases(rq->lock) |
2794 | { | |
70b97a7f IM |
2795 | struct rq *rq = this_rq(); |
2796 | ||
4866cde0 NP |
2797 | finish_task_switch(rq, prev); |
2798 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2799 | /* In this case, finish_task_switch does not reenable preemption */ | |
2800 | preempt_enable(); | |
2801 | #endif | |
1da177e4 | 2802 | if (current->set_child_tid) |
b488893a | 2803 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2804 | } |
2805 | ||
2806 | /* | |
2807 | * context_switch - switch to the new MM and the new | |
2808 | * thread's register state. | |
2809 | */ | |
dd41f596 | 2810 | static inline void |
70b97a7f | 2811 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2812 | struct task_struct *next) |
1da177e4 | 2813 | { |
dd41f596 | 2814 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2815 | |
e107be36 | 2816 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2817 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2818 | mm = next->mm; |
2819 | oldmm = prev->active_mm; | |
9226d125 ZA |
2820 | /* |
2821 | * For paravirt, this is coupled with an exit in switch_to to | |
2822 | * combine the page table reload and the switch backend into | |
2823 | * one hypercall. | |
2824 | */ | |
224101ed | 2825 | arch_start_context_switch(prev); |
9226d125 | 2826 | |
dd41f596 | 2827 | if (unlikely(!mm)) { |
1da177e4 LT |
2828 | next->active_mm = oldmm; |
2829 | atomic_inc(&oldmm->mm_count); | |
2830 | enter_lazy_tlb(oldmm, next); | |
2831 | } else | |
2832 | switch_mm(oldmm, mm, next); | |
2833 | ||
dd41f596 | 2834 | if (unlikely(!prev->mm)) { |
1da177e4 | 2835 | prev->active_mm = NULL; |
1da177e4 LT |
2836 | rq->prev_mm = oldmm; |
2837 | } | |
3a5f5e48 IM |
2838 | /* |
2839 | * Since the runqueue lock will be released by the next | |
2840 | * task (which is an invalid locking op but in the case | |
2841 | * of the scheduler it's an obvious special-case), so we | |
2842 | * do an early lockdep release here: | |
2843 | */ | |
2844 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2845 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2846 | #endif |
1da177e4 LT |
2847 | |
2848 | /* Here we just switch the register state and the stack. */ | |
2849 | switch_to(prev, next, prev); | |
2850 | ||
dd41f596 IM |
2851 | barrier(); |
2852 | /* | |
2853 | * this_rq must be evaluated again because prev may have moved | |
2854 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2855 | * frame will be invalid. | |
2856 | */ | |
2857 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2858 | } |
2859 | ||
2860 | /* | |
2861 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2862 | * | |
2863 | * externally visible scheduler statistics: current number of runnable | |
2864 | * threads, current number of uninterruptible-sleeping threads, total | |
2865 | * number of context switches performed since bootup. | |
2866 | */ | |
2867 | unsigned long nr_running(void) | |
2868 | { | |
2869 | unsigned long i, sum = 0; | |
2870 | ||
2871 | for_each_online_cpu(i) | |
2872 | sum += cpu_rq(i)->nr_running; | |
2873 | ||
2874 | return sum; | |
2875 | } | |
2876 | ||
2877 | unsigned long nr_uninterruptible(void) | |
2878 | { | |
2879 | unsigned long i, sum = 0; | |
2880 | ||
0a945022 | 2881 | for_each_possible_cpu(i) |
1da177e4 LT |
2882 | sum += cpu_rq(i)->nr_uninterruptible; |
2883 | ||
2884 | /* | |
2885 | * Since we read the counters lockless, it might be slightly | |
2886 | * inaccurate. Do not allow it to go below zero though: | |
2887 | */ | |
2888 | if (unlikely((long)sum < 0)) | |
2889 | sum = 0; | |
2890 | ||
2891 | return sum; | |
2892 | } | |
2893 | ||
2894 | unsigned long long nr_context_switches(void) | |
2895 | { | |
cc94abfc SR |
2896 | int i; |
2897 | unsigned long long sum = 0; | |
1da177e4 | 2898 | |
0a945022 | 2899 | for_each_possible_cpu(i) |
1da177e4 LT |
2900 | sum += cpu_rq(i)->nr_switches; |
2901 | ||
2902 | return sum; | |
2903 | } | |
2904 | ||
2905 | unsigned long nr_iowait(void) | |
2906 | { | |
2907 | unsigned long i, sum = 0; | |
2908 | ||
0a945022 | 2909 | for_each_possible_cpu(i) |
1da177e4 LT |
2910 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2911 | ||
2912 | return sum; | |
2913 | } | |
2914 | ||
dce48a84 TG |
2915 | /* Variables and functions for calc_load */ |
2916 | static atomic_long_t calc_load_tasks; | |
2917 | static unsigned long calc_load_update; | |
2918 | unsigned long avenrun[3]; | |
2919 | EXPORT_SYMBOL(avenrun); | |
2920 | ||
2d02494f TG |
2921 | /** |
2922 | * get_avenrun - get the load average array | |
2923 | * @loads: pointer to dest load array | |
2924 | * @offset: offset to add | |
2925 | * @shift: shift count to shift the result left | |
2926 | * | |
2927 | * These values are estimates at best, so no need for locking. | |
2928 | */ | |
2929 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2930 | { | |
2931 | loads[0] = (avenrun[0] + offset) << shift; | |
2932 | loads[1] = (avenrun[1] + offset) << shift; | |
2933 | loads[2] = (avenrun[2] + offset) << shift; | |
2934 | } | |
2935 | ||
dce48a84 TG |
2936 | static unsigned long |
2937 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2938 | { |
dce48a84 TG |
2939 | load *= exp; |
2940 | load += active * (FIXED_1 - exp); | |
2941 | return load >> FSHIFT; | |
2942 | } | |
db1b1fef | 2943 | |
dce48a84 TG |
2944 | /* |
2945 | * calc_load - update the avenrun load estimates 10 ticks after the | |
2946 | * CPUs have updated calc_load_tasks. | |
2947 | */ | |
2948 | void calc_global_load(void) | |
2949 | { | |
2950 | unsigned long upd = calc_load_update + 10; | |
2951 | long active; | |
2952 | ||
2953 | if (time_before(jiffies, upd)) | |
2954 | return; | |
db1b1fef | 2955 | |
dce48a84 TG |
2956 | active = atomic_long_read(&calc_load_tasks); |
2957 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 2958 | |
dce48a84 TG |
2959 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2960 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2961 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
2962 | ||
2963 | calc_load_update += LOAD_FREQ; | |
2964 | } | |
2965 | ||
2966 | /* | |
2967 | * Either called from update_cpu_load() or from a cpu going idle | |
2968 | */ | |
2969 | static void calc_load_account_active(struct rq *this_rq) | |
2970 | { | |
2971 | long nr_active, delta; | |
2972 | ||
2973 | nr_active = this_rq->nr_running; | |
2974 | nr_active += (long) this_rq->nr_uninterruptible; | |
2975 | ||
2976 | if (nr_active != this_rq->calc_load_active) { | |
2977 | delta = nr_active - this_rq->calc_load_active; | |
2978 | this_rq->calc_load_active = nr_active; | |
2979 | atomic_long_add(delta, &calc_load_tasks); | |
2980 | } | |
db1b1fef JS |
2981 | } |
2982 | ||
48f24c4d | 2983 | /* |
dd41f596 IM |
2984 | * Update rq->cpu_load[] statistics. This function is usually called every |
2985 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2986 | */ |
dd41f596 | 2987 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2988 | { |
495eca49 | 2989 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2990 | int i, scale; |
2991 | ||
2992 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2993 | |
2994 | /* Update our load: */ | |
2995 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2996 | unsigned long old_load, new_load; | |
2997 | ||
2998 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2999 | ||
3000 | old_load = this_rq->cpu_load[i]; | |
3001 | new_load = this_load; | |
a25707f3 IM |
3002 | /* |
3003 | * Round up the averaging division if load is increasing. This | |
3004 | * prevents us from getting stuck on 9 if the load is 10, for | |
3005 | * example. | |
3006 | */ | |
3007 | if (new_load > old_load) | |
3008 | new_load += scale-1; | |
dd41f596 IM |
3009 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3010 | } | |
dce48a84 TG |
3011 | |
3012 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3013 | this_rq->calc_load_update += LOAD_FREQ; | |
3014 | calc_load_account_active(this_rq); | |
3015 | } | |
48f24c4d IM |
3016 | } |
3017 | ||
dd41f596 IM |
3018 | #ifdef CONFIG_SMP |
3019 | ||
1da177e4 LT |
3020 | /* |
3021 | * double_rq_lock - safely lock two runqueues | |
3022 | * | |
3023 | * Note this does not disable interrupts like task_rq_lock, | |
3024 | * you need to do so manually before calling. | |
3025 | */ | |
70b97a7f | 3026 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3027 | __acquires(rq1->lock) |
3028 | __acquires(rq2->lock) | |
3029 | { | |
054b9108 | 3030 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3031 | if (rq1 == rq2) { |
3032 | spin_lock(&rq1->lock); | |
3033 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3034 | } else { | |
c96d145e | 3035 | if (rq1 < rq2) { |
1da177e4 | 3036 | spin_lock(&rq1->lock); |
5e710e37 | 3037 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3038 | } else { |
3039 | spin_lock(&rq2->lock); | |
5e710e37 | 3040 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3041 | } |
3042 | } | |
6e82a3be IM |
3043 | update_rq_clock(rq1); |
3044 | update_rq_clock(rq2); | |
1da177e4 LT |
3045 | } |
3046 | ||
3047 | /* | |
3048 | * double_rq_unlock - safely unlock two runqueues | |
3049 | * | |
3050 | * Note this does not restore interrupts like task_rq_unlock, | |
3051 | * you need to do so manually after calling. | |
3052 | */ | |
70b97a7f | 3053 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3054 | __releases(rq1->lock) |
3055 | __releases(rq2->lock) | |
3056 | { | |
3057 | spin_unlock(&rq1->lock); | |
3058 | if (rq1 != rq2) | |
3059 | spin_unlock(&rq2->lock); | |
3060 | else | |
3061 | __release(rq2->lock); | |
3062 | } | |
3063 | ||
1da177e4 LT |
3064 | /* |
3065 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3066 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3067 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3068 | * the cpu_allowed mask is restored. |
3069 | */ | |
36c8b586 | 3070 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3071 | { |
70b97a7f | 3072 | struct migration_req req; |
1da177e4 | 3073 | unsigned long flags; |
70b97a7f | 3074 | struct rq *rq; |
1da177e4 LT |
3075 | |
3076 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3077 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3078 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3079 | goto out; |
3080 | ||
3081 | /* force the process onto the specified CPU */ | |
3082 | if (migrate_task(p, dest_cpu, &req)) { | |
3083 | /* Need to wait for migration thread (might exit: take ref). */ | |
3084 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3085 | |
1da177e4 LT |
3086 | get_task_struct(mt); |
3087 | task_rq_unlock(rq, &flags); | |
3088 | wake_up_process(mt); | |
3089 | put_task_struct(mt); | |
3090 | wait_for_completion(&req.done); | |
36c8b586 | 3091 | |
1da177e4 LT |
3092 | return; |
3093 | } | |
3094 | out: | |
3095 | task_rq_unlock(rq, &flags); | |
3096 | } | |
3097 | ||
3098 | /* | |
476d139c NP |
3099 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3100 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3101 | */ |
3102 | void sched_exec(void) | |
3103 | { | |
1da177e4 | 3104 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3105 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3106 | put_cpu(); |
476d139c NP |
3107 | if (new_cpu != this_cpu) |
3108 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3109 | } |
3110 | ||
3111 | /* | |
3112 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3113 | * Both runqueues must be locked. | |
3114 | */ | |
dd41f596 IM |
3115 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3116 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3117 | { |
2e1cb74a | 3118 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3119 | set_task_cpu(p, this_cpu); |
dd41f596 | 3120 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3121 | /* |
3122 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3123 | * to be always true for them. | |
3124 | */ | |
15afe09b | 3125 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3126 | } |
3127 | ||
3128 | /* | |
3129 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3130 | */ | |
858119e1 | 3131 | static |
70b97a7f | 3132 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3133 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3134 | int *all_pinned) |
1da177e4 | 3135 | { |
708dc512 | 3136 | int tsk_cache_hot = 0; |
1da177e4 LT |
3137 | /* |
3138 | * We do not migrate tasks that are: | |
3139 | * 1) running (obviously), or | |
3140 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3141 | * 3) are cache-hot on their current CPU. | |
3142 | */ | |
96f874e2 | 3143 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3144 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3145 | return 0; |
cc367732 | 3146 | } |
81026794 NP |
3147 | *all_pinned = 0; |
3148 | ||
cc367732 IM |
3149 | if (task_running(rq, p)) { |
3150 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3151 | return 0; |
cc367732 | 3152 | } |
1da177e4 | 3153 | |
da84d961 IM |
3154 | /* |
3155 | * Aggressive migration if: | |
3156 | * 1) task is cache cold, or | |
3157 | * 2) too many balance attempts have failed. | |
3158 | */ | |
3159 | ||
708dc512 LH |
3160 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3161 | if (!tsk_cache_hot || | |
3162 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3163 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3164 | if (tsk_cache_hot) { |
da84d961 | 3165 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3166 | schedstat_inc(p, se.nr_forced_migrations); |
3167 | } | |
da84d961 IM |
3168 | #endif |
3169 | return 1; | |
3170 | } | |
3171 | ||
708dc512 | 3172 | if (tsk_cache_hot) { |
cc367732 | 3173 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3174 | return 0; |
cc367732 | 3175 | } |
1da177e4 LT |
3176 | return 1; |
3177 | } | |
3178 | ||
e1d1484f PW |
3179 | static unsigned long |
3180 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3181 | unsigned long max_load_move, struct sched_domain *sd, | |
3182 | enum cpu_idle_type idle, int *all_pinned, | |
3183 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3184 | { |
051c6764 | 3185 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3186 | struct task_struct *p; |
3187 | long rem_load_move = max_load_move; | |
1da177e4 | 3188 | |
e1d1484f | 3189 | if (max_load_move == 0) |
1da177e4 LT |
3190 | goto out; |
3191 | ||
81026794 NP |
3192 | pinned = 1; |
3193 | ||
1da177e4 | 3194 | /* |
dd41f596 | 3195 | * Start the load-balancing iterator: |
1da177e4 | 3196 | */ |
dd41f596 IM |
3197 | p = iterator->start(iterator->arg); |
3198 | next: | |
b82d9fdd | 3199 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3200 | goto out; |
051c6764 PZ |
3201 | |
3202 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3203 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3204 | p = iterator->next(iterator->arg); |
3205 | goto next; | |
1da177e4 LT |
3206 | } |
3207 | ||
dd41f596 | 3208 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3209 | pulled++; |
dd41f596 | 3210 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3211 | |
7e96fa58 GH |
3212 | #ifdef CONFIG_PREEMPT |
3213 | /* | |
3214 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3215 | * will stop after the first task is pulled to minimize the critical | |
3216 | * section. | |
3217 | */ | |
3218 | if (idle == CPU_NEWLY_IDLE) | |
3219 | goto out; | |
3220 | #endif | |
3221 | ||
2dd73a4f | 3222 | /* |
b82d9fdd | 3223 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3224 | */ |
e1d1484f | 3225 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3226 | if (p->prio < *this_best_prio) |
3227 | *this_best_prio = p->prio; | |
dd41f596 IM |
3228 | p = iterator->next(iterator->arg); |
3229 | goto next; | |
1da177e4 LT |
3230 | } |
3231 | out: | |
3232 | /* | |
e1d1484f | 3233 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3234 | * so we can safely collect pull_task() stats here rather than |
3235 | * inside pull_task(). | |
3236 | */ | |
3237 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3238 | |
3239 | if (all_pinned) | |
3240 | *all_pinned = pinned; | |
e1d1484f PW |
3241 | |
3242 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3243 | } |
3244 | ||
dd41f596 | 3245 | /* |
43010659 PW |
3246 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3247 | * this_rq, as part of a balancing operation within domain "sd". | |
3248 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3249 | * |
3250 | * Called with both runqueues locked. | |
3251 | */ | |
3252 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3253 | unsigned long max_load_move, |
dd41f596 IM |
3254 | struct sched_domain *sd, enum cpu_idle_type idle, |
3255 | int *all_pinned) | |
3256 | { | |
5522d5d5 | 3257 | const struct sched_class *class = sched_class_highest; |
43010659 | 3258 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3259 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3260 | |
3261 | do { | |
43010659 PW |
3262 | total_load_moved += |
3263 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3264 | max_load_move - total_load_moved, |
a4ac01c3 | 3265 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3266 | class = class->next; |
c4acb2c0 | 3267 | |
7e96fa58 GH |
3268 | #ifdef CONFIG_PREEMPT |
3269 | /* | |
3270 | * NEWIDLE balancing is a source of latency, so preemptible | |
3271 | * kernels will stop after the first task is pulled to minimize | |
3272 | * the critical section. | |
3273 | */ | |
c4acb2c0 GH |
3274 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3275 | break; | |
7e96fa58 | 3276 | #endif |
43010659 | 3277 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3278 | |
43010659 PW |
3279 | return total_load_moved > 0; |
3280 | } | |
3281 | ||
e1d1484f PW |
3282 | static int |
3283 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3284 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3285 | struct rq_iterator *iterator) | |
3286 | { | |
3287 | struct task_struct *p = iterator->start(iterator->arg); | |
3288 | int pinned = 0; | |
3289 | ||
3290 | while (p) { | |
3291 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3292 | pull_task(busiest, p, this_rq, this_cpu); | |
3293 | /* | |
3294 | * Right now, this is only the second place pull_task() | |
3295 | * is called, so we can safely collect pull_task() | |
3296 | * stats here rather than inside pull_task(). | |
3297 | */ | |
3298 | schedstat_inc(sd, lb_gained[idle]); | |
3299 | ||
3300 | return 1; | |
3301 | } | |
3302 | p = iterator->next(iterator->arg); | |
3303 | } | |
3304 | ||
3305 | return 0; | |
3306 | } | |
3307 | ||
43010659 PW |
3308 | /* |
3309 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3310 | * part of active balancing operations within "domain". | |
3311 | * Returns 1 if successful and 0 otherwise. | |
3312 | * | |
3313 | * Called with both runqueues locked. | |
3314 | */ | |
3315 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3316 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3317 | { | |
5522d5d5 | 3318 | const struct sched_class *class; |
43010659 PW |
3319 | |
3320 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3321 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3322 | return 1; |
3323 | ||
3324 | return 0; | |
dd41f596 | 3325 | } |
67bb6c03 | 3326 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3327 | /* |
222d656d GS |
3328 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3329 | * during load balancing. | |
1da177e4 | 3330 | */ |
222d656d GS |
3331 | struct sd_lb_stats { |
3332 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3333 | struct sched_group *this; /* Local group in this sd */ | |
3334 | unsigned long total_load; /* Total load of all groups in sd */ | |
3335 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3336 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3337 | ||
3338 | /** Statistics of this group */ | |
3339 | unsigned long this_load; | |
3340 | unsigned long this_load_per_task; | |
3341 | unsigned long this_nr_running; | |
3342 | ||
3343 | /* Statistics of the busiest group */ | |
3344 | unsigned long max_load; | |
3345 | unsigned long busiest_load_per_task; | |
3346 | unsigned long busiest_nr_running; | |
3347 | ||
3348 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3349 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3350 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3351 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3352 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3353 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3354 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3355 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3356 | #endif |
222d656d | 3357 | }; |
1da177e4 | 3358 | |
d5ac537e | 3359 | /* |
381be78f GS |
3360 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3361 | */ | |
3362 | struct sg_lb_stats { | |
3363 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3364 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3365 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3366 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3367 | unsigned long group_capacity; | |
3368 | int group_imb; /* Is there an imbalance in the group ? */ | |
3369 | }; | |
408ed066 | 3370 | |
67bb6c03 GS |
3371 | /** |
3372 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3373 | * @group: The group whose first cpu is to be returned. | |
3374 | */ | |
3375 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3376 | { | |
3377 | return cpumask_first(sched_group_cpus(group)); | |
3378 | } | |
3379 | ||
3380 | /** | |
3381 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3382 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3383 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3384 | */ | |
3385 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3386 | enum cpu_idle_type idle) | |
3387 | { | |
3388 | int load_idx; | |
3389 | ||
3390 | switch (idle) { | |
3391 | case CPU_NOT_IDLE: | |
7897986b | 3392 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3393 | break; |
3394 | ||
3395 | case CPU_NEWLY_IDLE: | |
7897986b | 3396 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3397 | break; |
3398 | default: | |
7897986b | 3399 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3400 | break; |
3401 | } | |
1da177e4 | 3402 | |
67bb6c03 GS |
3403 | return load_idx; |
3404 | } | |
1da177e4 | 3405 | |
1da177e4 | 3406 | |
c071df18 GS |
3407 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3408 | /** | |
3409 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3410 | * the given sched_domain, during load balancing. | |
3411 | * | |
3412 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3413 | * @sds: Variable containing the statistics for sd. | |
3414 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3415 | */ | |
3416 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3417 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3418 | { | |
3419 | /* | |
3420 | * Busy processors will not participate in power savings | |
3421 | * balance. | |
3422 | */ | |
3423 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3424 | sds->power_savings_balance = 0; | |
3425 | else { | |
3426 | sds->power_savings_balance = 1; | |
3427 | sds->min_nr_running = ULONG_MAX; | |
3428 | sds->leader_nr_running = 0; | |
3429 | } | |
3430 | } | |
783609c6 | 3431 | |
c071df18 GS |
3432 | /** |
3433 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3434 | * sched_domain while performing load balancing. | |
3435 | * | |
3436 | * @group: sched_group belonging to the sched_domain under consideration. | |
3437 | * @sds: Variable containing the statistics of the sched_domain | |
3438 | * @local_group: Does group contain the CPU for which we're performing | |
3439 | * load balancing ? | |
3440 | * @sgs: Variable containing the statistics of the group. | |
3441 | */ | |
3442 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3443 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3444 | { | |
408ed066 | 3445 | |
c071df18 GS |
3446 | if (!sds->power_savings_balance) |
3447 | return; | |
1da177e4 | 3448 | |
c071df18 GS |
3449 | /* |
3450 | * If the local group is idle or completely loaded | |
3451 | * no need to do power savings balance at this domain | |
3452 | */ | |
3453 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3454 | !sds->this_nr_running)) | |
3455 | sds->power_savings_balance = 0; | |
2dd73a4f | 3456 | |
c071df18 GS |
3457 | /* |
3458 | * If a group is already running at full capacity or idle, | |
3459 | * don't include that group in power savings calculations | |
3460 | */ | |
3461 | if (!sds->power_savings_balance || | |
3462 | sgs->sum_nr_running >= sgs->group_capacity || | |
3463 | !sgs->sum_nr_running) | |
3464 | return; | |
5969fe06 | 3465 | |
c071df18 GS |
3466 | /* |
3467 | * Calculate the group which has the least non-idle load. | |
3468 | * This is the group from where we need to pick up the load | |
3469 | * for saving power | |
3470 | */ | |
3471 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3472 | (sgs->sum_nr_running == sds->min_nr_running && | |
3473 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3474 | sds->group_min = group; | |
3475 | sds->min_nr_running = sgs->sum_nr_running; | |
3476 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3477 | sgs->sum_nr_running; | |
3478 | } | |
783609c6 | 3479 | |
c071df18 GS |
3480 | /* |
3481 | * Calculate the group which is almost near its | |
3482 | * capacity but still has some space to pick up some load | |
3483 | * from other group and save more power | |
3484 | */ | |
3485 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3486 | return; | |
1da177e4 | 3487 | |
c071df18 GS |
3488 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3489 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3490 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3491 | sds->group_leader = group; | |
3492 | sds->leader_nr_running = sgs->sum_nr_running; | |
3493 | } | |
3494 | } | |
408ed066 | 3495 | |
c071df18 | 3496 | /** |
d5ac537e | 3497 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3498 | * @sds: Variable containing the statistics of the sched_domain |
3499 | * under consideration. | |
3500 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3501 | * @imbalance: Variable to store the imbalance. | |
3502 | * | |
d5ac537e RD |
3503 | * Description: |
3504 | * Check if we have potential to perform some power-savings balance. | |
3505 | * If yes, set the busiest group to be the least loaded group in the | |
3506 | * sched_domain, so that it's CPUs can be put to idle. | |
3507 | * | |
c071df18 GS |
3508 | * Returns 1 if there is potential to perform power-savings balance. |
3509 | * Else returns 0. | |
3510 | */ | |
3511 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3512 | int this_cpu, unsigned long *imbalance) | |
3513 | { | |
3514 | if (!sds->power_savings_balance) | |
3515 | return 0; | |
1da177e4 | 3516 | |
c071df18 GS |
3517 | if (sds->this != sds->group_leader || |
3518 | sds->group_leader == sds->group_min) | |
3519 | return 0; | |
783609c6 | 3520 | |
c071df18 GS |
3521 | *imbalance = sds->min_load_per_task; |
3522 | sds->busiest = sds->group_min; | |
1da177e4 | 3523 | |
c071df18 GS |
3524 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3525 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3526 | group_first_cpu(sds->group_leader); | |
3527 | } | |
3528 | ||
3529 | return 1; | |
1da177e4 | 3530 | |
c071df18 GS |
3531 | } |
3532 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3533 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3534 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3535 | { | |
3536 | return; | |
3537 | } | |
408ed066 | 3538 | |
c071df18 GS |
3539 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3540 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3541 | { | |
3542 | return; | |
3543 | } | |
3544 | ||
3545 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3546 | int this_cpu, unsigned long *imbalance) | |
3547 | { | |
3548 | return 0; | |
3549 | } | |
3550 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3551 | ||
3552 | ||
1f8c553d GS |
3553 | /** |
3554 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3555 | * @group: sched_group whose statistics are to be updated. | |
3556 | * @this_cpu: Cpu for which load balance is currently performed. | |
3557 | * @idle: Idle status of this_cpu | |
3558 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3559 | * @sd_idle: Idle status of the sched_domain containing group. | |
3560 | * @local_group: Does group contain this_cpu. | |
3561 | * @cpus: Set of cpus considered for load balancing. | |
3562 | * @balance: Should we balance. | |
3563 | * @sgs: variable to hold the statistics for this group. | |
3564 | */ | |
3565 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3566 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3567 | int local_group, const struct cpumask *cpus, | |
3568 | int *balance, struct sg_lb_stats *sgs) | |
3569 | { | |
3570 | unsigned long load, max_cpu_load, min_cpu_load; | |
3571 | int i; | |
3572 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3573 | unsigned long sum_avg_load_per_task; | |
3574 | unsigned long avg_load_per_task; | |
3575 | ||
3576 | if (local_group) | |
3577 | balance_cpu = group_first_cpu(group); | |
3578 | ||
3579 | /* Tally up the load of all CPUs in the group */ | |
3580 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3581 | max_cpu_load = 0; | |
3582 | min_cpu_load = ~0UL; | |
408ed066 | 3583 | |
1f8c553d GS |
3584 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3585 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3586 | |
1f8c553d GS |
3587 | if (*sd_idle && rq->nr_running) |
3588 | *sd_idle = 0; | |
5c45bf27 | 3589 | |
1f8c553d | 3590 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3591 | if (local_group) { |
1f8c553d GS |
3592 | if (idle_cpu(i) && !first_idle_cpu) { |
3593 | first_idle_cpu = 1; | |
3594 | balance_cpu = i; | |
3595 | } | |
3596 | ||
3597 | load = target_load(i, load_idx); | |
3598 | } else { | |
3599 | load = source_load(i, load_idx); | |
3600 | if (load > max_cpu_load) | |
3601 | max_cpu_load = load; | |
3602 | if (min_cpu_load > load) | |
3603 | min_cpu_load = load; | |
1da177e4 | 3604 | } |
5c45bf27 | 3605 | |
1f8c553d GS |
3606 | sgs->group_load += load; |
3607 | sgs->sum_nr_running += rq->nr_running; | |
3608 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3609 | |
1f8c553d GS |
3610 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3611 | } | |
5c45bf27 | 3612 | |
1f8c553d GS |
3613 | /* |
3614 | * First idle cpu or the first cpu(busiest) in this sched group | |
3615 | * is eligible for doing load balancing at this and above | |
3616 | * domains. In the newly idle case, we will allow all the cpu's | |
3617 | * to do the newly idle load balance. | |
3618 | */ | |
3619 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3620 | balance_cpu != this_cpu && balance) { | |
3621 | *balance = 0; | |
3622 | return; | |
3623 | } | |
5c45bf27 | 3624 | |
1f8c553d GS |
3625 | /* Adjust by relative CPU power of the group */ |
3626 | sgs->avg_load = sg_div_cpu_power(group, | |
3627 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3628 | |
1f8c553d GS |
3629 | |
3630 | /* | |
3631 | * Consider the group unbalanced when the imbalance is larger | |
3632 | * than the average weight of two tasks. | |
3633 | * | |
3634 | * APZ: with cgroup the avg task weight can vary wildly and | |
3635 | * might not be a suitable number - should we keep a | |
3636 | * normalized nr_running number somewhere that negates | |
3637 | * the hierarchy? | |
3638 | */ | |
3639 | avg_load_per_task = sg_div_cpu_power(group, | |
3640 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3641 | ||
3642 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3643 | sgs->group_imb = 1; | |
3644 | ||
3645 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3646 | ||
3647 | } | |
dd41f596 | 3648 | |
37abe198 GS |
3649 | /** |
3650 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3651 | * @sd: sched_domain whose statistics are to be updated. | |
3652 | * @this_cpu: Cpu for which load balance is currently performed. | |
3653 | * @idle: Idle status of this_cpu | |
3654 | * @sd_idle: Idle status of the sched_domain containing group. | |
3655 | * @cpus: Set of cpus considered for load balancing. | |
3656 | * @balance: Should we balance. | |
3657 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3658 | */ |
37abe198 GS |
3659 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3660 | enum cpu_idle_type idle, int *sd_idle, | |
3661 | const struct cpumask *cpus, int *balance, | |
3662 | struct sd_lb_stats *sds) | |
1da177e4 | 3663 | { |
222d656d | 3664 | struct sched_group *group = sd->groups; |
37abe198 | 3665 | struct sg_lb_stats sgs; |
222d656d GS |
3666 | int load_idx; |
3667 | ||
c071df18 | 3668 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3669 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3670 | |
3671 | do { | |
1da177e4 | 3672 | int local_group; |
1da177e4 | 3673 | |
758b2cdc RR |
3674 | local_group = cpumask_test_cpu(this_cpu, |
3675 | sched_group_cpus(group)); | |
381be78f | 3676 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3677 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3678 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3679 | |
37abe198 GS |
3680 | if (local_group && balance && !(*balance)) |
3681 | return; | |
783609c6 | 3682 | |
37abe198 GS |
3683 | sds->total_load += sgs.group_load; |
3684 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3685 | |
1da177e4 | 3686 | if (local_group) { |
37abe198 GS |
3687 | sds->this_load = sgs.avg_load; |
3688 | sds->this = group; | |
3689 | sds->this_nr_running = sgs.sum_nr_running; | |
3690 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3691 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3692 | (sgs.sum_nr_running > sgs.group_capacity || |
3693 | sgs.group_imb)) { | |
37abe198 GS |
3694 | sds->max_load = sgs.avg_load; |
3695 | sds->busiest = group; | |
3696 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3697 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3698 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3699 | } |
5c45bf27 | 3700 | |
c071df18 | 3701 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3702 | group = group->next; |
3703 | } while (group != sd->groups); | |
3704 | ||
37abe198 | 3705 | } |
1da177e4 | 3706 | |
2e6f44ae GS |
3707 | /** |
3708 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3709 | * amongst the groups of a sched_domain, during |
3710 | * load balancing. | |
2e6f44ae GS |
3711 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3712 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3713 | * @imbalance: Variable to store the imbalance. | |
3714 | */ | |
3715 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3716 | int this_cpu, unsigned long *imbalance) | |
3717 | { | |
3718 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3719 | unsigned int imbn = 2; | |
3720 | ||
3721 | if (sds->this_nr_running) { | |
3722 | sds->this_load_per_task /= sds->this_nr_running; | |
3723 | if (sds->busiest_load_per_task > | |
3724 | sds->this_load_per_task) | |
3725 | imbn = 1; | |
3726 | } else | |
3727 | sds->this_load_per_task = | |
3728 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3729 | |
2e6f44ae GS |
3730 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3731 | sds->busiest_load_per_task * imbn) { | |
3732 | *imbalance = sds->busiest_load_per_task; | |
3733 | return; | |
3734 | } | |
908a7c1b | 3735 | |
1da177e4 | 3736 | /* |
2e6f44ae GS |
3737 | * OK, we don't have enough imbalance to justify moving tasks, |
3738 | * however we may be able to increase total CPU power used by | |
3739 | * moving them. | |
1da177e4 | 3740 | */ |
2dd73a4f | 3741 | |
2e6f44ae GS |
3742 | pwr_now += sds->busiest->__cpu_power * |
3743 | min(sds->busiest_load_per_task, sds->max_load); | |
3744 | pwr_now += sds->this->__cpu_power * | |
3745 | min(sds->this_load_per_task, sds->this_load); | |
3746 | pwr_now /= SCHED_LOAD_SCALE; | |
3747 | ||
3748 | /* Amount of load we'd subtract */ | |
3749 | tmp = sg_div_cpu_power(sds->busiest, | |
3750 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3751 | if (sds->max_load > tmp) | |
3752 | pwr_move += sds->busiest->__cpu_power * | |
3753 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3754 | ||
3755 | /* Amount of load we'd add */ | |
3756 | if (sds->max_load * sds->busiest->__cpu_power < | |
3757 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3758 | tmp = sg_div_cpu_power(sds->this, | |
3759 | sds->max_load * sds->busiest->__cpu_power); | |
3760 | else | |
3761 | tmp = sg_div_cpu_power(sds->this, | |
3762 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3763 | pwr_move += sds->this->__cpu_power * | |
3764 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3765 | pwr_move /= SCHED_LOAD_SCALE; | |
3766 | ||
3767 | /* Move if we gain throughput */ | |
3768 | if (pwr_move > pwr_now) | |
3769 | *imbalance = sds->busiest_load_per_task; | |
3770 | } | |
dbc523a3 GS |
3771 | |
3772 | /** | |
3773 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3774 | * groups of a given sched_domain during load balance. | |
3775 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3776 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3777 | * @imbalance: The variable to store the imbalance. | |
3778 | */ | |
3779 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3780 | unsigned long *imbalance) | |
3781 | { | |
3782 | unsigned long max_pull; | |
2dd73a4f PW |
3783 | /* |
3784 | * In the presence of smp nice balancing, certain scenarios can have | |
3785 | * max load less than avg load(as we skip the groups at or below | |
3786 | * its cpu_power, while calculating max_load..) | |
3787 | */ | |
dbc523a3 | 3788 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3789 | *imbalance = 0; |
dbc523a3 | 3790 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3791 | } |
0c117f1b SS |
3792 | |
3793 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3794 | max_pull = min(sds->max_load - sds->avg_load, |
3795 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3796 | |
1da177e4 | 3797 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
3798 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
3799 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
3800 | / SCHED_LOAD_SCALE; |
3801 | ||
2dd73a4f PW |
3802 | /* |
3803 | * if *imbalance is less than the average load per runnable task | |
3804 | * there is no gaurantee that any tasks will be moved so we'll have | |
3805 | * a think about bumping its value to force at least one task to be | |
3806 | * moved | |
3807 | */ | |
dbc523a3 GS |
3808 | if (*imbalance < sds->busiest_load_per_task) |
3809 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3810 | |
dbc523a3 | 3811 | } |
37abe198 | 3812 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3813 | |
b7bb4c9b GS |
3814 | /** |
3815 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3816 | * if there is an imbalance. If there isn't an imbalance, and | |
3817 | * the user has opted for power-savings, it returns a group whose | |
3818 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3819 | * such a group exists. | |
3820 | * | |
3821 | * Also calculates the amount of weighted load which should be moved | |
3822 | * to restore balance. | |
3823 | * | |
3824 | * @sd: The sched_domain whose busiest group is to be returned. | |
3825 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3826 | * @imbalance: Variable which stores amount of weighted load which should | |
3827 | * be moved to restore balance/put a group to idle. | |
3828 | * @idle: The idle status of this_cpu. | |
3829 | * @sd_idle: The idleness of sd | |
3830 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3831 | * @balance: Pointer to a variable indicating if this_cpu | |
3832 | * is the appropriate cpu to perform load balancing at this_level. | |
3833 | * | |
3834 | * Returns: - the busiest group if imbalance exists. | |
3835 | * - If no imbalance and user has opted for power-savings balance, | |
3836 | * return the least loaded group whose CPUs can be | |
3837 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3838 | */ |
3839 | static struct sched_group * | |
3840 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3841 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3842 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3843 | { | |
3844 | struct sd_lb_stats sds; | |
1da177e4 | 3845 | |
37abe198 | 3846 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3847 | |
37abe198 GS |
3848 | /* |
3849 | * Compute the various statistics relavent for load balancing at | |
3850 | * this level. | |
3851 | */ | |
3852 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3853 | balance, &sds); | |
3854 | ||
b7bb4c9b GS |
3855 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3856 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3857 | * at this level. | |
3858 | * 2) There is no busy sibling group to pull from. | |
3859 | * 3) This group is the busiest group. | |
3860 | * 4) This group is more busy than the avg busieness at this | |
3861 | * sched_domain. | |
3862 | * 5) The imbalance is within the specified limit. | |
3863 | * 6) Any rebalance would lead to ping-pong | |
3864 | */ | |
37abe198 GS |
3865 | if (balance && !(*balance)) |
3866 | goto ret; | |
1da177e4 | 3867 | |
b7bb4c9b GS |
3868 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3869 | goto out_balanced; | |
1da177e4 | 3870 | |
b7bb4c9b | 3871 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3872 | goto out_balanced; |
1da177e4 | 3873 | |
222d656d | 3874 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3875 | |
b7bb4c9b GS |
3876 | if (sds.this_load >= sds.avg_load) |
3877 | goto out_balanced; | |
3878 | ||
3879 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3880 | goto out_balanced; |
3881 | ||
222d656d GS |
3882 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3883 | if (sds.group_imb) | |
3884 | sds.busiest_load_per_task = | |
3885 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3886 | |
1da177e4 LT |
3887 | /* |
3888 | * We're trying to get all the cpus to the average_load, so we don't | |
3889 | * want to push ourselves above the average load, nor do we wish to | |
3890 | * reduce the max loaded cpu below the average load, as either of these | |
3891 | * actions would just result in more rebalancing later, and ping-pong | |
3892 | * tasks around. Thus we look for the minimum possible imbalance. | |
3893 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3894 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3895 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3896 | * appear as very large values with unsigned longs. |
3897 | */ | |
222d656d | 3898 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3899 | goto out_balanced; |
3900 | ||
dbc523a3 GS |
3901 | /* Looks like there is an imbalance. Compute it */ |
3902 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3903 | return sds.busiest; |
1da177e4 LT |
3904 | |
3905 | out_balanced: | |
c071df18 GS |
3906 | /* |
3907 | * There is no obvious imbalance. But check if we can do some balancing | |
3908 | * to save power. | |
3909 | */ | |
3910 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3911 | return sds.busiest; | |
783609c6 | 3912 | ret: |
1da177e4 LT |
3913 | *imbalance = 0; |
3914 | return NULL; | |
3915 | } | |
3916 | ||
3917 | /* | |
3918 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3919 | */ | |
70b97a7f | 3920 | static struct rq * |
d15bcfdb | 3921 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3922 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3923 | { |
70b97a7f | 3924 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3925 | unsigned long max_load = 0; |
1da177e4 LT |
3926 | int i; |
3927 | ||
758b2cdc | 3928 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3929 | unsigned long wl; |
0a2966b4 | 3930 | |
96f874e2 | 3931 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3932 | continue; |
3933 | ||
48f24c4d | 3934 | rq = cpu_rq(i); |
dd41f596 | 3935 | wl = weighted_cpuload(i); |
2dd73a4f | 3936 | |
dd41f596 | 3937 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3938 | continue; |
1da177e4 | 3939 | |
dd41f596 IM |
3940 | if (wl > max_load) { |
3941 | max_load = wl; | |
48f24c4d | 3942 | busiest = rq; |
1da177e4 LT |
3943 | } |
3944 | } | |
3945 | ||
3946 | return busiest; | |
3947 | } | |
3948 | ||
77391d71 NP |
3949 | /* |
3950 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3951 | * so long as it is large enough. | |
3952 | */ | |
3953 | #define MAX_PINNED_INTERVAL 512 | |
3954 | ||
df7c8e84 RR |
3955 | /* Working cpumask for load_balance and load_balance_newidle. */ |
3956 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
3957 | ||
1da177e4 LT |
3958 | /* |
3959 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3960 | * tasks if there is an imbalance. | |
1da177e4 | 3961 | */ |
70b97a7f | 3962 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3963 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 3964 | int *balance) |
1da177e4 | 3965 | { |
43010659 | 3966 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3967 | struct sched_group *group; |
1da177e4 | 3968 | unsigned long imbalance; |
70b97a7f | 3969 | struct rq *busiest; |
fe2eea3f | 3970 | unsigned long flags; |
df7c8e84 | 3971 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 3972 | |
96f874e2 | 3973 | cpumask_setall(cpus); |
7c16ec58 | 3974 | |
89c4710e SS |
3975 | /* |
3976 | * When power savings policy is enabled for the parent domain, idle | |
3977 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3978 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3979 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3980 | */ |
d15bcfdb | 3981 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3982 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3983 | sd_idle = 1; |
1da177e4 | 3984 | |
2d72376b | 3985 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3986 | |
0a2966b4 | 3987 | redo: |
c8cba857 | 3988 | update_shares(sd); |
0a2966b4 | 3989 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3990 | cpus, balance); |
783609c6 | 3991 | |
06066714 | 3992 | if (*balance == 0) |
783609c6 | 3993 | goto out_balanced; |
783609c6 | 3994 | |
1da177e4 LT |
3995 | if (!group) { |
3996 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3997 | goto out_balanced; | |
3998 | } | |
3999 | ||
7c16ec58 | 4000 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4001 | if (!busiest) { |
4002 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4003 | goto out_balanced; | |
4004 | } | |
4005 | ||
db935dbd | 4006 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4007 | |
4008 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4009 | ||
43010659 | 4010 | ld_moved = 0; |
1da177e4 LT |
4011 | if (busiest->nr_running > 1) { |
4012 | /* | |
4013 | * Attempt to move tasks. If find_busiest_group has found | |
4014 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4015 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4016 | * correctly treated as an imbalance. |
4017 | */ | |
fe2eea3f | 4018 | local_irq_save(flags); |
e17224bf | 4019 | double_rq_lock(this_rq, busiest); |
43010659 | 4020 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4021 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4022 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4023 | local_irq_restore(flags); |
81026794 | 4024 | |
46cb4b7c SS |
4025 | /* |
4026 | * some other cpu did the load balance for us. | |
4027 | */ | |
43010659 | 4028 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4029 | resched_cpu(this_cpu); |
4030 | ||
81026794 | 4031 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4032 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4033 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4034 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4035 | goto redo; |
81026794 | 4036 | goto out_balanced; |
0a2966b4 | 4037 | } |
1da177e4 | 4038 | } |
81026794 | 4039 | |
43010659 | 4040 | if (!ld_moved) { |
1da177e4 LT |
4041 | schedstat_inc(sd, lb_failed[idle]); |
4042 | sd->nr_balance_failed++; | |
4043 | ||
4044 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4045 | |
fe2eea3f | 4046 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4047 | |
4048 | /* don't kick the migration_thread, if the curr | |
4049 | * task on busiest cpu can't be moved to this_cpu | |
4050 | */ | |
96f874e2 RR |
4051 | if (!cpumask_test_cpu(this_cpu, |
4052 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4053 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4054 | all_pinned = 1; |
4055 | goto out_one_pinned; | |
4056 | } | |
4057 | ||
1da177e4 LT |
4058 | if (!busiest->active_balance) { |
4059 | busiest->active_balance = 1; | |
4060 | busiest->push_cpu = this_cpu; | |
81026794 | 4061 | active_balance = 1; |
1da177e4 | 4062 | } |
fe2eea3f | 4063 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4064 | if (active_balance) |
1da177e4 LT |
4065 | wake_up_process(busiest->migration_thread); |
4066 | ||
4067 | /* | |
4068 | * We've kicked active balancing, reset the failure | |
4069 | * counter. | |
4070 | */ | |
39507451 | 4071 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4072 | } |
81026794 | 4073 | } else |
1da177e4 LT |
4074 | sd->nr_balance_failed = 0; |
4075 | ||
81026794 | 4076 | if (likely(!active_balance)) { |
1da177e4 LT |
4077 | /* We were unbalanced, so reset the balancing interval */ |
4078 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4079 | } else { |
4080 | /* | |
4081 | * If we've begun active balancing, start to back off. This | |
4082 | * case may not be covered by the all_pinned logic if there | |
4083 | * is only 1 task on the busy runqueue (because we don't call | |
4084 | * move_tasks). | |
4085 | */ | |
4086 | if (sd->balance_interval < sd->max_interval) | |
4087 | sd->balance_interval *= 2; | |
1da177e4 LT |
4088 | } |
4089 | ||
43010659 | 4090 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4091 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4092 | ld_moved = -1; |
4093 | ||
4094 | goto out; | |
1da177e4 LT |
4095 | |
4096 | out_balanced: | |
1da177e4 LT |
4097 | schedstat_inc(sd, lb_balanced[idle]); |
4098 | ||
16cfb1c0 | 4099 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4100 | |
4101 | out_one_pinned: | |
1da177e4 | 4102 | /* tune up the balancing interval */ |
77391d71 NP |
4103 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4104 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4105 | sd->balance_interval *= 2; |
4106 | ||
48f24c4d | 4107 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4108 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4109 | ld_moved = -1; |
4110 | else | |
4111 | ld_moved = 0; | |
4112 | out: | |
c8cba857 PZ |
4113 | if (ld_moved) |
4114 | update_shares(sd); | |
c09595f6 | 4115 | return ld_moved; |
1da177e4 LT |
4116 | } |
4117 | ||
4118 | /* | |
4119 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4120 | * tasks if there is an imbalance. | |
4121 | * | |
d15bcfdb | 4122 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4123 | * this_rq is locked. |
4124 | */ | |
48f24c4d | 4125 | static int |
df7c8e84 | 4126 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4127 | { |
4128 | struct sched_group *group; | |
70b97a7f | 4129 | struct rq *busiest = NULL; |
1da177e4 | 4130 | unsigned long imbalance; |
43010659 | 4131 | int ld_moved = 0; |
5969fe06 | 4132 | int sd_idle = 0; |
969bb4e4 | 4133 | int all_pinned = 0; |
df7c8e84 | 4134 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4135 | |
96f874e2 | 4136 | cpumask_setall(cpus); |
5969fe06 | 4137 | |
89c4710e SS |
4138 | /* |
4139 | * When power savings policy is enabled for the parent domain, idle | |
4140 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4141 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4142 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4143 | */ |
4144 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4145 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4146 | sd_idle = 1; |
1da177e4 | 4147 | |
2d72376b | 4148 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4149 | redo: |
3e5459b4 | 4150 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4151 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4152 | &sd_idle, cpus, NULL); |
1da177e4 | 4153 | if (!group) { |
d15bcfdb | 4154 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4155 | goto out_balanced; |
1da177e4 LT |
4156 | } |
4157 | ||
7c16ec58 | 4158 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4159 | if (!busiest) { |
d15bcfdb | 4160 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4161 | goto out_balanced; |
1da177e4 LT |
4162 | } |
4163 | ||
db935dbd NP |
4164 | BUG_ON(busiest == this_rq); |
4165 | ||
d15bcfdb | 4166 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4167 | |
43010659 | 4168 | ld_moved = 0; |
d6d5cfaf NP |
4169 | if (busiest->nr_running > 1) { |
4170 | /* Attempt to move tasks */ | |
4171 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4172 | /* this_rq->clock is already updated */ |
4173 | update_rq_clock(busiest); | |
43010659 | 4174 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4175 | imbalance, sd, CPU_NEWLY_IDLE, |
4176 | &all_pinned); | |
1b12bbc7 | 4177 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4178 | |
969bb4e4 | 4179 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4180 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4181 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4182 | goto redo; |
4183 | } | |
d6d5cfaf NP |
4184 | } |
4185 | ||
43010659 | 4186 | if (!ld_moved) { |
36dffab6 | 4187 | int active_balance = 0; |
ad273b32 | 4188 | |
d15bcfdb | 4189 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4190 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4191 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4192 | return -1; |
ad273b32 VS |
4193 | |
4194 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4195 | return -1; | |
4196 | ||
4197 | if (sd->nr_balance_failed++ < 2) | |
4198 | return -1; | |
4199 | ||
4200 | /* | |
4201 | * The only task running in a non-idle cpu can be moved to this | |
4202 | * cpu in an attempt to completely freeup the other CPU | |
4203 | * package. The same method used to move task in load_balance() | |
4204 | * have been extended for load_balance_newidle() to speedup | |
4205 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4206 | * | |
4207 | * The package power saving logic comes from | |
4208 | * find_busiest_group(). If there are no imbalance, then | |
4209 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4210 | * f_b_g() will select a group from which a running task may be | |
4211 | * pulled to this cpu in order to make the other package idle. | |
4212 | * If there is no opportunity to make a package idle and if | |
4213 | * there are no imbalance, then f_b_g() will return NULL and no | |
4214 | * action will be taken in load_balance_newidle(). | |
4215 | * | |
4216 | * Under normal task pull operation due to imbalance, there | |
4217 | * will be more than one task in the source run queue and | |
4218 | * move_tasks() will succeed. ld_moved will be true and this | |
4219 | * active balance code will not be triggered. | |
4220 | */ | |
4221 | ||
4222 | /* Lock busiest in correct order while this_rq is held */ | |
4223 | double_lock_balance(this_rq, busiest); | |
4224 | ||
4225 | /* | |
4226 | * don't kick the migration_thread, if the curr | |
4227 | * task on busiest cpu can't be moved to this_cpu | |
4228 | */ | |
6ca09dfc | 4229 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4230 | double_unlock_balance(this_rq, busiest); |
4231 | all_pinned = 1; | |
4232 | return ld_moved; | |
4233 | } | |
4234 | ||
4235 | if (!busiest->active_balance) { | |
4236 | busiest->active_balance = 1; | |
4237 | busiest->push_cpu = this_cpu; | |
4238 | active_balance = 1; | |
4239 | } | |
4240 | ||
4241 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4242 | /* |
4243 | * Should not call ttwu while holding a rq->lock | |
4244 | */ | |
4245 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4246 | if (active_balance) |
4247 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4248 | spin_lock(&this_rq->lock); |
ad273b32 | 4249 | |
5969fe06 | 4250 | } else |
16cfb1c0 | 4251 | sd->nr_balance_failed = 0; |
1da177e4 | 4252 | |
3e5459b4 | 4253 | update_shares_locked(this_rq, sd); |
43010659 | 4254 | return ld_moved; |
16cfb1c0 NP |
4255 | |
4256 | out_balanced: | |
d15bcfdb | 4257 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4258 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4259 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4260 | return -1; |
16cfb1c0 | 4261 | sd->nr_balance_failed = 0; |
48f24c4d | 4262 | |
16cfb1c0 | 4263 | return 0; |
1da177e4 LT |
4264 | } |
4265 | ||
4266 | /* | |
4267 | * idle_balance is called by schedule() if this_cpu is about to become | |
4268 | * idle. Attempts to pull tasks from other CPUs. | |
4269 | */ | |
70b97a7f | 4270 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4271 | { |
4272 | struct sched_domain *sd; | |
efbe027e | 4273 | int pulled_task = 0; |
dd41f596 | 4274 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4275 | |
4276 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4277 | unsigned long interval; |
4278 | ||
4279 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4280 | continue; | |
4281 | ||
4282 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4283 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4284 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4285 | sd); |
92c4ca5c CL |
4286 | |
4287 | interval = msecs_to_jiffies(sd->balance_interval); | |
4288 | if (time_after(next_balance, sd->last_balance + interval)) | |
4289 | next_balance = sd->last_balance + interval; | |
4290 | if (pulled_task) | |
4291 | break; | |
1da177e4 | 4292 | } |
dd41f596 | 4293 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4294 | /* |
4295 | * We are going idle. next_balance may be set based on | |
4296 | * a busy processor. So reset next_balance. | |
4297 | */ | |
4298 | this_rq->next_balance = next_balance; | |
dd41f596 | 4299 | } |
1da177e4 LT |
4300 | } |
4301 | ||
4302 | /* | |
4303 | * active_load_balance is run by migration threads. It pushes running tasks | |
4304 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4305 | * running on each physical CPU where possible, and avoids physical / | |
4306 | * logical imbalances. | |
4307 | * | |
4308 | * Called with busiest_rq locked. | |
4309 | */ | |
70b97a7f | 4310 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4311 | { |
39507451 | 4312 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4313 | struct sched_domain *sd; |
4314 | struct rq *target_rq; | |
39507451 | 4315 | |
48f24c4d | 4316 | /* Is there any task to move? */ |
39507451 | 4317 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4318 | return; |
4319 | ||
4320 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4321 | |
4322 | /* | |
39507451 | 4323 | * This condition is "impossible", if it occurs |
41a2d6cf | 4324 | * we need to fix it. Originally reported by |
39507451 | 4325 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4326 | */ |
39507451 | 4327 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4328 | |
39507451 NP |
4329 | /* move a task from busiest_rq to target_rq */ |
4330 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4331 | update_rq_clock(busiest_rq); |
4332 | update_rq_clock(target_rq); | |
39507451 NP |
4333 | |
4334 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4335 | for_each_domain(target_cpu, sd) { |
39507451 | 4336 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4337 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4338 | break; |
c96d145e | 4339 | } |
39507451 | 4340 | |
48f24c4d | 4341 | if (likely(sd)) { |
2d72376b | 4342 | schedstat_inc(sd, alb_count); |
39507451 | 4343 | |
43010659 PW |
4344 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4345 | sd, CPU_IDLE)) | |
48f24c4d IM |
4346 | schedstat_inc(sd, alb_pushed); |
4347 | else | |
4348 | schedstat_inc(sd, alb_failed); | |
4349 | } | |
1b12bbc7 | 4350 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4351 | } |
4352 | ||
46cb4b7c SS |
4353 | #ifdef CONFIG_NO_HZ |
4354 | static struct { | |
4355 | atomic_t load_balancer; | |
7d1e6a9b | 4356 | cpumask_var_t cpu_mask; |
f711f609 | 4357 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4358 | } nohz ____cacheline_aligned = { |
4359 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4360 | }; |
4361 | ||
f711f609 GS |
4362 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4363 | /** | |
4364 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4365 | * @cpu: The cpu whose lowest level of sched domain is to | |
4366 | * be returned. | |
4367 | * @flag: The flag to check for the lowest sched_domain | |
4368 | * for the given cpu. | |
4369 | * | |
4370 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4371 | */ | |
4372 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4373 | { | |
4374 | struct sched_domain *sd; | |
4375 | ||
4376 | for_each_domain(cpu, sd) | |
4377 | if (sd && (sd->flags & flag)) | |
4378 | break; | |
4379 | ||
4380 | return sd; | |
4381 | } | |
4382 | ||
4383 | /** | |
4384 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4385 | * @cpu: The cpu whose domains we're iterating over. | |
4386 | * @sd: variable holding the value of the power_savings_sd | |
4387 | * for cpu. | |
4388 | * @flag: The flag to filter the sched_domains to be iterated. | |
4389 | * | |
4390 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4391 | * set, starting from the lowest sched_domain to the highest. | |
4392 | */ | |
4393 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4394 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4395 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4396 | ||
4397 | /** | |
4398 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4399 | * @ilb_group: group to be checked for semi-idleness | |
4400 | * | |
4401 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4402 | * | |
4403 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4404 | * and atleast one non-idle CPU. This helper function checks if the given | |
4405 | * sched_group is semi-idle or not. | |
4406 | */ | |
4407 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4408 | { | |
4409 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4410 | sched_group_cpus(ilb_group)); | |
4411 | ||
4412 | /* | |
4413 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4414 | * and atleast one idle cpu. | |
4415 | */ | |
4416 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4417 | return 0; | |
4418 | ||
4419 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4420 | return 0; | |
4421 | ||
4422 | return 1; | |
4423 | } | |
4424 | /** | |
4425 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4426 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4427 | * | |
4428 | * Returns: Returns the id of the idle load balancer if it exists, | |
4429 | * Else, returns >= nr_cpu_ids. | |
4430 | * | |
4431 | * This algorithm picks the idle load balancer such that it belongs to a | |
4432 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4433 | * completely idle packages/cores just for the purpose of idle load balancing | |
4434 | * when there are other idle cpu's which are better suited for that job. | |
4435 | */ | |
4436 | static int find_new_ilb(int cpu) | |
4437 | { | |
4438 | struct sched_domain *sd; | |
4439 | struct sched_group *ilb_group; | |
4440 | ||
4441 | /* | |
4442 | * Have idle load balancer selection from semi-idle packages only | |
4443 | * when power-aware load balancing is enabled | |
4444 | */ | |
4445 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4446 | goto out_done; | |
4447 | ||
4448 | /* | |
4449 | * Optimize for the case when we have no idle CPUs or only one | |
4450 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4451 | */ | |
4452 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4453 | goto out_done; | |
4454 | ||
4455 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4456 | ilb_group = sd->groups; | |
4457 | ||
4458 | do { | |
4459 | if (is_semi_idle_group(ilb_group)) | |
4460 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4461 | ||
4462 | ilb_group = ilb_group->next; | |
4463 | ||
4464 | } while (ilb_group != sd->groups); | |
4465 | } | |
4466 | ||
4467 | out_done: | |
4468 | return cpumask_first(nohz.cpu_mask); | |
4469 | } | |
4470 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4471 | static inline int find_new_ilb(int call_cpu) | |
4472 | { | |
6e29ec57 | 4473 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4474 | } |
4475 | #endif | |
4476 | ||
7835b98b | 4477 | /* |
46cb4b7c SS |
4478 | * This routine will try to nominate the ilb (idle load balancing) |
4479 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4480 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4481 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4482 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4483 | * arrives... | |
4484 | * | |
4485 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4486 | * for idle load balancing. ilb owner will still be part of | |
4487 | * nohz.cpu_mask.. | |
7835b98b | 4488 | * |
46cb4b7c SS |
4489 | * While stopping the tick, this cpu will become the ilb owner if there |
4490 | * is no other owner. And will be the owner till that cpu becomes busy | |
4491 | * or if all cpus in the system stop their ticks at which point | |
4492 | * there is no need for ilb owner. | |
4493 | * | |
4494 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4495 | * next busy scheduler_tick() | |
4496 | */ | |
4497 | int select_nohz_load_balancer(int stop_tick) | |
4498 | { | |
4499 | int cpu = smp_processor_id(); | |
4500 | ||
4501 | if (stop_tick) { | |
46cb4b7c SS |
4502 | cpu_rq(cpu)->in_nohz_recently = 1; |
4503 | ||
483b4ee6 SS |
4504 | if (!cpu_active(cpu)) { |
4505 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4506 | return 0; | |
4507 | ||
4508 | /* | |
4509 | * If we are going offline and still the leader, | |
4510 | * give up! | |
4511 | */ | |
46cb4b7c SS |
4512 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4513 | BUG(); | |
483b4ee6 | 4514 | |
46cb4b7c SS |
4515 | return 0; |
4516 | } | |
4517 | ||
483b4ee6 SS |
4518 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4519 | ||
46cb4b7c | 4520 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4521 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4522 | if (atomic_read(&nohz.load_balancer) == cpu) |
4523 | atomic_set(&nohz.load_balancer, -1); | |
4524 | return 0; | |
4525 | } | |
4526 | ||
4527 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4528 | /* make me the ilb owner */ | |
4529 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4530 | return 1; | |
e790fb0b GS |
4531 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4532 | int new_ilb; | |
4533 | ||
4534 | if (!(sched_smt_power_savings || | |
4535 | sched_mc_power_savings)) | |
4536 | return 1; | |
4537 | /* | |
4538 | * Check to see if there is a more power-efficient | |
4539 | * ilb. | |
4540 | */ | |
4541 | new_ilb = find_new_ilb(cpu); | |
4542 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4543 | atomic_set(&nohz.load_balancer, -1); | |
4544 | resched_cpu(new_ilb); | |
4545 | return 0; | |
4546 | } | |
46cb4b7c | 4547 | return 1; |
e790fb0b | 4548 | } |
46cb4b7c | 4549 | } else { |
7d1e6a9b | 4550 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4551 | return 0; |
4552 | ||
7d1e6a9b | 4553 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4554 | |
4555 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4556 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4557 | BUG(); | |
4558 | } | |
4559 | return 0; | |
4560 | } | |
4561 | #endif | |
4562 | ||
4563 | static DEFINE_SPINLOCK(balancing); | |
4564 | ||
4565 | /* | |
7835b98b CL |
4566 | * It checks each scheduling domain to see if it is due to be balanced, |
4567 | * and initiates a balancing operation if so. | |
4568 | * | |
4569 | * Balancing parameters are set up in arch_init_sched_domains. | |
4570 | */ | |
a9957449 | 4571 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4572 | { |
46cb4b7c SS |
4573 | int balance = 1; |
4574 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4575 | unsigned long interval; |
4576 | struct sched_domain *sd; | |
46cb4b7c | 4577 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4578 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4579 | int update_next_balance = 0; |
d07355f5 | 4580 | int need_serialize; |
1da177e4 | 4581 | |
46cb4b7c | 4582 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4583 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4584 | continue; | |
4585 | ||
4586 | interval = sd->balance_interval; | |
d15bcfdb | 4587 | if (idle != CPU_IDLE) |
1da177e4 LT |
4588 | interval *= sd->busy_factor; |
4589 | ||
4590 | /* scale ms to jiffies */ | |
4591 | interval = msecs_to_jiffies(interval); | |
4592 | if (unlikely(!interval)) | |
4593 | interval = 1; | |
dd41f596 IM |
4594 | if (interval > HZ*NR_CPUS/10) |
4595 | interval = HZ*NR_CPUS/10; | |
4596 | ||
d07355f5 | 4597 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4598 | |
d07355f5 | 4599 | if (need_serialize) { |
08c183f3 CL |
4600 | if (!spin_trylock(&balancing)) |
4601 | goto out; | |
4602 | } | |
4603 | ||
c9819f45 | 4604 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4605 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4606 | /* |
4607 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4608 | * longer idle, or one of our SMT siblings is |
4609 | * not idle. | |
4610 | */ | |
d15bcfdb | 4611 | idle = CPU_NOT_IDLE; |
1da177e4 | 4612 | } |
1bd77f2d | 4613 | sd->last_balance = jiffies; |
1da177e4 | 4614 | } |
d07355f5 | 4615 | if (need_serialize) |
08c183f3 CL |
4616 | spin_unlock(&balancing); |
4617 | out: | |
f549da84 | 4618 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4619 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4620 | update_next_balance = 1; |
4621 | } | |
783609c6 SS |
4622 | |
4623 | /* | |
4624 | * Stop the load balance at this level. There is another | |
4625 | * CPU in our sched group which is doing load balancing more | |
4626 | * actively. | |
4627 | */ | |
4628 | if (!balance) | |
4629 | break; | |
1da177e4 | 4630 | } |
f549da84 SS |
4631 | |
4632 | /* | |
4633 | * next_balance will be updated only when there is a need. | |
4634 | * When the cpu is attached to null domain for ex, it will not be | |
4635 | * updated. | |
4636 | */ | |
4637 | if (likely(update_next_balance)) | |
4638 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4639 | } |
4640 | ||
4641 | /* | |
4642 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4643 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4644 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4645 | */ | |
4646 | static void run_rebalance_domains(struct softirq_action *h) | |
4647 | { | |
dd41f596 IM |
4648 | int this_cpu = smp_processor_id(); |
4649 | struct rq *this_rq = cpu_rq(this_cpu); | |
4650 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4651 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4652 | |
dd41f596 | 4653 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4654 | |
4655 | #ifdef CONFIG_NO_HZ | |
4656 | /* | |
4657 | * If this cpu is the owner for idle load balancing, then do the | |
4658 | * balancing on behalf of the other idle cpus whose ticks are | |
4659 | * stopped. | |
4660 | */ | |
dd41f596 IM |
4661 | if (this_rq->idle_at_tick && |
4662 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4663 | struct rq *rq; |
4664 | int balance_cpu; | |
4665 | ||
7d1e6a9b RR |
4666 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4667 | if (balance_cpu == this_cpu) | |
4668 | continue; | |
4669 | ||
46cb4b7c SS |
4670 | /* |
4671 | * If this cpu gets work to do, stop the load balancing | |
4672 | * work being done for other cpus. Next load | |
4673 | * balancing owner will pick it up. | |
4674 | */ | |
4675 | if (need_resched()) | |
4676 | break; | |
4677 | ||
de0cf899 | 4678 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4679 | |
4680 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4681 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4682 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4683 | } |
4684 | } | |
4685 | #endif | |
4686 | } | |
4687 | ||
8a0be9ef FW |
4688 | static inline int on_null_domain(int cpu) |
4689 | { | |
4690 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4691 | } | |
4692 | ||
46cb4b7c SS |
4693 | /* |
4694 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4695 | * | |
4696 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4697 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4698 | * if the whole system is idle. | |
4699 | */ | |
dd41f596 | 4700 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4701 | { |
46cb4b7c SS |
4702 | #ifdef CONFIG_NO_HZ |
4703 | /* | |
4704 | * If we were in the nohz mode recently and busy at the current | |
4705 | * scheduler tick, then check if we need to nominate new idle | |
4706 | * load balancer. | |
4707 | */ | |
4708 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4709 | rq->in_nohz_recently = 0; | |
4710 | ||
4711 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4712 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4713 | atomic_set(&nohz.load_balancer, -1); |
4714 | } | |
4715 | ||
4716 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4717 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4718 | |
434d53b0 | 4719 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4720 | resched_cpu(ilb); |
4721 | } | |
4722 | } | |
4723 | ||
4724 | /* | |
4725 | * If this cpu is idle and doing idle load balancing for all the | |
4726 | * cpus with ticks stopped, is it time for that to stop? | |
4727 | */ | |
4728 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4729 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4730 | resched_cpu(cpu); |
4731 | return; | |
4732 | } | |
4733 | ||
4734 | /* | |
4735 | * If this cpu is idle and the idle load balancing is done by | |
4736 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4737 | */ | |
4738 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4739 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4740 | return; |
4741 | #endif | |
8a0be9ef FW |
4742 | /* Don't need to rebalance while attached to NULL domain */ |
4743 | if (time_after_eq(jiffies, rq->next_balance) && | |
4744 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4745 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4746 | } |
dd41f596 IM |
4747 | |
4748 | #else /* CONFIG_SMP */ | |
4749 | ||
1da177e4 LT |
4750 | /* |
4751 | * on UP we do not need to balance between CPUs: | |
4752 | */ | |
70b97a7f | 4753 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4754 | { |
4755 | } | |
dd41f596 | 4756 | |
1da177e4 LT |
4757 | #endif |
4758 | ||
1da177e4 LT |
4759 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4760 | ||
4761 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4762 | ||
4763 | /* | |
c5f8d995 | 4764 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4765 | * @p in case that task is currently running. |
c5f8d995 HS |
4766 | * |
4767 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4768 | */ |
c5f8d995 HS |
4769 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4770 | { | |
4771 | u64 ns = 0; | |
4772 | ||
4773 | if (task_current(rq, p)) { | |
4774 | update_rq_clock(rq); | |
4775 | ns = rq->clock - p->se.exec_start; | |
4776 | if ((s64)ns < 0) | |
4777 | ns = 0; | |
4778 | } | |
4779 | ||
4780 | return ns; | |
4781 | } | |
4782 | ||
bb34d92f | 4783 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4784 | { |
1da177e4 | 4785 | unsigned long flags; |
41b86e9c | 4786 | struct rq *rq; |
bb34d92f | 4787 | u64 ns = 0; |
48f24c4d | 4788 | |
41b86e9c | 4789 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4790 | ns = do_task_delta_exec(p, rq); |
4791 | task_rq_unlock(rq, &flags); | |
1508487e | 4792 | |
c5f8d995 HS |
4793 | return ns; |
4794 | } | |
f06febc9 | 4795 | |
c5f8d995 HS |
4796 | /* |
4797 | * Return accounted runtime for the task. | |
4798 | * In case the task is currently running, return the runtime plus current's | |
4799 | * pending runtime that have not been accounted yet. | |
4800 | */ | |
4801 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4802 | { | |
4803 | unsigned long flags; | |
4804 | struct rq *rq; | |
4805 | u64 ns = 0; | |
4806 | ||
4807 | rq = task_rq_lock(p, &flags); | |
4808 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4809 | task_rq_unlock(rq, &flags); | |
4810 | ||
4811 | return ns; | |
4812 | } | |
48f24c4d | 4813 | |
c5f8d995 HS |
4814 | /* |
4815 | * Return sum_exec_runtime for the thread group. | |
4816 | * In case the task is currently running, return the sum plus current's | |
4817 | * pending runtime that have not been accounted yet. | |
4818 | * | |
4819 | * Note that the thread group might have other running tasks as well, | |
4820 | * so the return value not includes other pending runtime that other | |
4821 | * running tasks might have. | |
4822 | */ | |
4823 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4824 | { | |
4825 | struct task_cputime totals; | |
4826 | unsigned long flags; | |
4827 | struct rq *rq; | |
4828 | u64 ns; | |
4829 | ||
4830 | rq = task_rq_lock(p, &flags); | |
4831 | thread_group_cputime(p, &totals); | |
4832 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4833 | task_rq_unlock(rq, &flags); |
48f24c4d | 4834 | |
1da177e4 LT |
4835 | return ns; |
4836 | } | |
4837 | ||
1da177e4 LT |
4838 | /* |
4839 | * Account user cpu time to a process. | |
4840 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4841 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4842 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4843 | */ |
457533a7 MS |
4844 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4845 | cputime_t cputime_scaled) | |
1da177e4 LT |
4846 | { |
4847 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4848 | cputime64_t tmp; | |
4849 | ||
457533a7 | 4850 | /* Add user time to process. */ |
1da177e4 | 4851 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4852 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4853 | account_group_user_time(p, cputime); |
1da177e4 LT |
4854 | |
4855 | /* Add user time to cpustat. */ | |
4856 | tmp = cputime_to_cputime64(cputime); | |
4857 | if (TASK_NICE(p) > 0) | |
4858 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4859 | else | |
4860 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
4861 | |
4862 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
4863 | /* Account for user time used */ |
4864 | acct_update_integrals(p); | |
1da177e4 LT |
4865 | } |
4866 | ||
94886b84 LV |
4867 | /* |
4868 | * Account guest cpu time to a process. | |
4869 | * @p: the process that the cpu time gets accounted to | |
4870 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4871 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4872 | */ |
457533a7 MS |
4873 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4874 | cputime_t cputime_scaled) | |
94886b84 LV |
4875 | { |
4876 | cputime64_t tmp; | |
4877 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4878 | ||
4879 | tmp = cputime_to_cputime64(cputime); | |
4880 | ||
457533a7 | 4881 | /* Add guest time to process. */ |
94886b84 | 4882 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4883 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4884 | account_group_user_time(p, cputime); |
94886b84 LV |
4885 | p->gtime = cputime_add(p->gtime, cputime); |
4886 | ||
457533a7 | 4887 | /* Add guest time to cpustat. */ |
94886b84 LV |
4888 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4889 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4890 | } | |
4891 | ||
1da177e4 LT |
4892 | /* |
4893 | * Account system cpu time to a process. | |
4894 | * @p: the process that the cpu time gets accounted to | |
4895 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4896 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4897 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4898 | */ |
4899 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4900 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4901 | { |
4902 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4903 | cputime64_t tmp; |
4904 | ||
983ed7a6 | 4905 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4906 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4907 | return; |
4908 | } | |
94886b84 | 4909 | |
457533a7 | 4910 | /* Add system time to process. */ |
1da177e4 | 4911 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4912 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4913 | account_group_system_time(p, cputime); |
1da177e4 LT |
4914 | |
4915 | /* Add system time to cpustat. */ | |
4916 | tmp = cputime_to_cputime64(cputime); | |
4917 | if (hardirq_count() - hardirq_offset) | |
4918 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4919 | else if (softirq_count()) | |
4920 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4921 | else |
79741dd3 MS |
4922 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4923 | ||
ef12fefa BR |
4924 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
4925 | ||
1da177e4 LT |
4926 | /* Account for system time used */ |
4927 | acct_update_integrals(p); | |
1da177e4 LT |
4928 | } |
4929 | ||
c66f08be | 4930 | /* |
1da177e4 | 4931 | * Account for involuntary wait time. |
1da177e4 | 4932 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4933 | */ |
79741dd3 | 4934 | void account_steal_time(cputime_t cputime) |
c66f08be | 4935 | { |
79741dd3 MS |
4936 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4937 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4938 | ||
4939 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4940 | } |
4941 | ||
1da177e4 | 4942 | /* |
79741dd3 MS |
4943 | * Account for idle time. |
4944 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4945 | */ |
79741dd3 | 4946 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4947 | { |
4948 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4949 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4950 | struct rq *rq = this_rq(); |
1da177e4 | 4951 | |
79741dd3 MS |
4952 | if (atomic_read(&rq->nr_iowait) > 0) |
4953 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4954 | else | |
4955 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4956 | } |
4957 | ||
79741dd3 MS |
4958 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4959 | ||
4960 | /* | |
4961 | * Account a single tick of cpu time. | |
4962 | * @p: the process that the cpu time gets accounted to | |
4963 | * @user_tick: indicates if the tick is a user or a system tick | |
4964 | */ | |
4965 | void account_process_tick(struct task_struct *p, int user_tick) | |
4966 | { | |
4967 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
4968 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
4969 | struct rq *rq = this_rq(); | |
4970 | ||
4971 | if (user_tick) | |
4972 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 4973 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
4974 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
4975 | one_jiffy_scaled); | |
4976 | else | |
4977 | account_idle_time(one_jiffy); | |
4978 | } | |
4979 | ||
4980 | /* | |
4981 | * Account multiple ticks of steal time. | |
4982 | * @p: the process from which the cpu time has been stolen | |
4983 | * @ticks: number of stolen ticks | |
4984 | */ | |
4985 | void account_steal_ticks(unsigned long ticks) | |
4986 | { | |
4987 | account_steal_time(jiffies_to_cputime(ticks)); | |
4988 | } | |
4989 | ||
4990 | /* | |
4991 | * Account multiple ticks of idle time. | |
4992 | * @ticks: number of stolen ticks | |
4993 | */ | |
4994 | void account_idle_ticks(unsigned long ticks) | |
4995 | { | |
4996 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
4997 | } |
4998 | ||
79741dd3 MS |
4999 | #endif |
5000 | ||
49048622 BS |
5001 | /* |
5002 | * Use precise platform statistics if available: | |
5003 | */ | |
5004 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5005 | cputime_t task_utime(struct task_struct *p) | |
5006 | { | |
5007 | return p->utime; | |
5008 | } | |
5009 | ||
5010 | cputime_t task_stime(struct task_struct *p) | |
5011 | { | |
5012 | return p->stime; | |
5013 | } | |
5014 | #else | |
5015 | cputime_t task_utime(struct task_struct *p) | |
5016 | { | |
5017 | clock_t utime = cputime_to_clock_t(p->utime), | |
5018 | total = utime + cputime_to_clock_t(p->stime); | |
5019 | u64 temp; | |
5020 | ||
5021 | /* | |
5022 | * Use CFS's precise accounting: | |
5023 | */ | |
5024 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
5025 | ||
5026 | if (total) { | |
5027 | temp *= utime; | |
5028 | do_div(temp, total); | |
5029 | } | |
5030 | utime = (clock_t)temp; | |
5031 | ||
5032 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
5033 | return p->prev_utime; | |
5034 | } | |
5035 | ||
5036 | cputime_t task_stime(struct task_struct *p) | |
5037 | { | |
5038 | clock_t stime; | |
5039 | ||
5040 | /* | |
5041 | * Use CFS's precise accounting. (we subtract utime from | |
5042 | * the total, to make sure the total observed by userspace | |
5043 | * grows monotonically - apps rely on that): | |
5044 | */ | |
5045 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5046 | cputime_to_clock_t(task_utime(p)); | |
5047 | ||
5048 | if (stime >= 0) | |
5049 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5050 | ||
5051 | return p->prev_stime; | |
5052 | } | |
5053 | #endif | |
5054 | ||
5055 | inline cputime_t task_gtime(struct task_struct *p) | |
5056 | { | |
5057 | return p->gtime; | |
5058 | } | |
5059 | ||
7835b98b CL |
5060 | /* |
5061 | * This function gets called by the timer code, with HZ frequency. | |
5062 | * We call it with interrupts disabled. | |
5063 | * | |
5064 | * It also gets called by the fork code, when changing the parent's | |
5065 | * timeslices. | |
5066 | */ | |
5067 | void scheduler_tick(void) | |
5068 | { | |
7835b98b CL |
5069 | int cpu = smp_processor_id(); |
5070 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5071 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5072 | |
5073 | sched_clock_tick(); | |
dd41f596 IM |
5074 | |
5075 | spin_lock(&rq->lock); | |
3e51f33f | 5076 | update_rq_clock(rq); |
f1a438d8 | 5077 | update_cpu_load(rq); |
fa85ae24 | 5078 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5079 | spin_unlock(&rq->lock); |
7835b98b | 5080 | |
e418e1c2 | 5081 | #ifdef CONFIG_SMP |
dd41f596 IM |
5082 | rq->idle_at_tick = idle_cpu(cpu); |
5083 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5084 | #endif |
1da177e4 LT |
5085 | } |
5086 | ||
132380a0 | 5087 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5088 | { |
5089 | if (in_lock_functions(addr)) { | |
5090 | addr = CALLER_ADDR2; | |
5091 | if (in_lock_functions(addr)) | |
5092 | addr = CALLER_ADDR3; | |
5093 | } | |
5094 | return addr; | |
5095 | } | |
1da177e4 | 5096 | |
7e49fcce SR |
5097 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5098 | defined(CONFIG_PREEMPT_TRACER)) | |
5099 | ||
43627582 | 5100 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5101 | { |
6cd8a4bb | 5102 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5103 | /* |
5104 | * Underflow? | |
5105 | */ | |
9a11b49a IM |
5106 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5107 | return; | |
6cd8a4bb | 5108 | #endif |
1da177e4 | 5109 | preempt_count() += val; |
6cd8a4bb | 5110 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5111 | /* |
5112 | * Spinlock count overflowing soon? | |
5113 | */ | |
33859f7f MOS |
5114 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5115 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5116 | #endif |
5117 | if (preempt_count() == val) | |
5118 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5119 | } |
5120 | EXPORT_SYMBOL(add_preempt_count); | |
5121 | ||
43627582 | 5122 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5123 | { |
6cd8a4bb | 5124 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5125 | /* |
5126 | * Underflow? | |
5127 | */ | |
01e3eb82 | 5128 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5129 | return; |
1da177e4 LT |
5130 | /* |
5131 | * Is the spinlock portion underflowing? | |
5132 | */ | |
9a11b49a IM |
5133 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5134 | !(preempt_count() & PREEMPT_MASK))) | |
5135 | return; | |
6cd8a4bb | 5136 | #endif |
9a11b49a | 5137 | |
6cd8a4bb SR |
5138 | if (preempt_count() == val) |
5139 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5140 | preempt_count() -= val; |
5141 | } | |
5142 | EXPORT_SYMBOL(sub_preempt_count); | |
5143 | ||
5144 | #endif | |
5145 | ||
5146 | /* | |
dd41f596 | 5147 | * Print scheduling while atomic bug: |
1da177e4 | 5148 | */ |
dd41f596 | 5149 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5150 | { |
838225b4 SS |
5151 | struct pt_regs *regs = get_irq_regs(); |
5152 | ||
5153 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5154 | prev->comm, prev->pid, preempt_count()); | |
5155 | ||
dd41f596 | 5156 | debug_show_held_locks(prev); |
e21f5b15 | 5157 | print_modules(); |
dd41f596 IM |
5158 | if (irqs_disabled()) |
5159 | print_irqtrace_events(prev); | |
838225b4 SS |
5160 | |
5161 | if (regs) | |
5162 | show_regs(regs); | |
5163 | else | |
5164 | dump_stack(); | |
dd41f596 | 5165 | } |
1da177e4 | 5166 | |
dd41f596 IM |
5167 | /* |
5168 | * Various schedule()-time debugging checks and statistics: | |
5169 | */ | |
5170 | static inline void schedule_debug(struct task_struct *prev) | |
5171 | { | |
1da177e4 | 5172 | /* |
41a2d6cf | 5173 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5174 | * schedule() atomically, we ignore that path for now. |
5175 | * Otherwise, whine if we are scheduling when we should not be. | |
5176 | */ | |
3f33a7ce | 5177 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5178 | __schedule_bug(prev); |
5179 | ||
1da177e4 LT |
5180 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5181 | ||
2d72376b | 5182 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5183 | #ifdef CONFIG_SCHEDSTATS |
5184 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5185 | schedstat_inc(this_rq(), bkl_count); |
5186 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5187 | } |
5188 | #endif | |
dd41f596 IM |
5189 | } |
5190 | ||
df1c99d4 MG |
5191 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5192 | { | |
5193 | if (prev->state == TASK_RUNNING) { | |
5194 | u64 runtime = prev->se.sum_exec_runtime; | |
5195 | ||
5196 | runtime -= prev->se.prev_sum_exec_runtime; | |
5197 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5198 | ||
5199 | /* | |
5200 | * In order to avoid avg_overlap growing stale when we are | |
5201 | * indeed overlapping and hence not getting put to sleep, grow | |
5202 | * the avg_overlap on preemption. | |
5203 | * | |
5204 | * We use the average preemption runtime because that | |
5205 | * correlates to the amount of cache footprint a task can | |
5206 | * build up. | |
5207 | */ | |
5208 | update_avg(&prev->se.avg_overlap, runtime); | |
5209 | } | |
5210 | prev->sched_class->put_prev_task(rq, prev); | |
5211 | } | |
5212 | ||
dd41f596 IM |
5213 | /* |
5214 | * Pick up the highest-prio task: | |
5215 | */ | |
5216 | static inline struct task_struct * | |
b67802ea | 5217 | pick_next_task(struct rq *rq) |
dd41f596 | 5218 | { |
5522d5d5 | 5219 | const struct sched_class *class; |
dd41f596 | 5220 | struct task_struct *p; |
1da177e4 LT |
5221 | |
5222 | /* | |
dd41f596 IM |
5223 | * Optimization: we know that if all tasks are in |
5224 | * the fair class we can call that function directly: | |
1da177e4 | 5225 | */ |
dd41f596 | 5226 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5227 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5228 | if (likely(p)) |
5229 | return p; | |
1da177e4 LT |
5230 | } |
5231 | ||
dd41f596 IM |
5232 | class = sched_class_highest; |
5233 | for ( ; ; ) { | |
fb8d4724 | 5234 | p = class->pick_next_task(rq); |
dd41f596 IM |
5235 | if (p) |
5236 | return p; | |
5237 | /* | |
5238 | * Will never be NULL as the idle class always | |
5239 | * returns a non-NULL p: | |
5240 | */ | |
5241 | class = class->next; | |
5242 | } | |
5243 | } | |
1da177e4 | 5244 | |
dd41f596 IM |
5245 | /* |
5246 | * schedule() is the main scheduler function. | |
5247 | */ | |
ff743345 | 5248 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5249 | { |
5250 | struct task_struct *prev, *next; | |
67ca7bde | 5251 | unsigned long *switch_count; |
dd41f596 | 5252 | struct rq *rq; |
31656519 | 5253 | int cpu; |
dd41f596 | 5254 | |
ff743345 PZ |
5255 | need_resched: |
5256 | preempt_disable(); | |
dd41f596 IM |
5257 | cpu = smp_processor_id(); |
5258 | rq = cpu_rq(cpu); | |
5259 | rcu_qsctr_inc(cpu); | |
5260 | prev = rq->curr; | |
5261 | switch_count = &prev->nivcsw; | |
5262 | ||
5263 | release_kernel_lock(prev); | |
5264 | need_resched_nonpreemptible: | |
5265 | ||
5266 | schedule_debug(prev); | |
1da177e4 | 5267 | |
31656519 | 5268 | if (sched_feat(HRTICK)) |
f333fdc9 | 5269 | hrtick_clear(rq); |
8f4d37ec | 5270 | |
8cd162ce | 5271 | spin_lock_irq(&rq->lock); |
3e51f33f | 5272 | update_rq_clock(rq); |
1e819950 | 5273 | clear_tsk_need_resched(prev); |
1da177e4 | 5274 | |
1da177e4 | 5275 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5276 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5277 | prev->state = TASK_RUNNING; |
16882c1e | 5278 | else |
2e1cb74a | 5279 | deactivate_task(rq, prev, 1); |
dd41f596 | 5280 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5281 | } |
5282 | ||
9a897c5a SR |
5283 | #ifdef CONFIG_SMP |
5284 | if (prev->sched_class->pre_schedule) | |
5285 | prev->sched_class->pre_schedule(rq, prev); | |
5286 | #endif | |
f65eda4f | 5287 | |
dd41f596 | 5288 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5289 | idle_balance(cpu, rq); |
1da177e4 | 5290 | |
df1c99d4 | 5291 | put_prev_task(rq, prev); |
b67802ea | 5292 | next = pick_next_task(rq); |
1da177e4 | 5293 | |
1da177e4 | 5294 | if (likely(prev != next)) { |
673a90a1 DS |
5295 | sched_info_switch(prev, next); |
5296 | ||
1da177e4 LT |
5297 | rq->nr_switches++; |
5298 | rq->curr = next; | |
5299 | ++*switch_count; | |
5300 | ||
dd41f596 | 5301 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5302 | /* |
5303 | * the context switch might have flipped the stack from under | |
5304 | * us, hence refresh the local variables. | |
5305 | */ | |
5306 | cpu = smp_processor_id(); | |
5307 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5308 | } else |
5309 | spin_unlock_irq(&rq->lock); | |
5310 | ||
8f4d37ec | 5311 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5312 | goto need_resched_nonpreemptible; |
8f4d37ec | 5313 | |
1da177e4 | 5314 | preempt_enable_no_resched(); |
ff743345 | 5315 | if (need_resched()) |
1da177e4 LT |
5316 | goto need_resched; |
5317 | } | |
1da177e4 LT |
5318 | EXPORT_SYMBOL(schedule); |
5319 | ||
0d66bf6d PZ |
5320 | #ifdef CONFIG_SMP |
5321 | /* | |
5322 | * Look out! "owner" is an entirely speculative pointer | |
5323 | * access and not reliable. | |
5324 | */ | |
5325 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5326 | { | |
5327 | unsigned int cpu; | |
5328 | struct rq *rq; | |
5329 | ||
5330 | if (!sched_feat(OWNER_SPIN)) | |
5331 | return 0; | |
5332 | ||
5333 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5334 | /* | |
5335 | * Need to access the cpu field knowing that | |
5336 | * DEBUG_PAGEALLOC could have unmapped it if | |
5337 | * the mutex owner just released it and exited. | |
5338 | */ | |
5339 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5340 | goto out; | |
5341 | #else | |
5342 | cpu = owner->cpu; | |
5343 | #endif | |
5344 | ||
5345 | /* | |
5346 | * Even if the access succeeded (likely case), | |
5347 | * the cpu field may no longer be valid. | |
5348 | */ | |
5349 | if (cpu >= nr_cpumask_bits) | |
5350 | goto out; | |
5351 | ||
5352 | /* | |
5353 | * We need to validate that we can do a | |
5354 | * get_cpu() and that we have the percpu area. | |
5355 | */ | |
5356 | if (!cpu_online(cpu)) | |
5357 | goto out; | |
5358 | ||
5359 | rq = cpu_rq(cpu); | |
5360 | ||
5361 | for (;;) { | |
5362 | /* | |
5363 | * Owner changed, break to re-assess state. | |
5364 | */ | |
5365 | if (lock->owner != owner) | |
5366 | break; | |
5367 | ||
5368 | /* | |
5369 | * Is that owner really running on that cpu? | |
5370 | */ | |
5371 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5372 | return 0; | |
5373 | ||
5374 | cpu_relax(); | |
5375 | } | |
5376 | out: | |
5377 | return 1; | |
5378 | } | |
5379 | #endif | |
5380 | ||
1da177e4 LT |
5381 | #ifdef CONFIG_PREEMPT |
5382 | /* | |
2ed6e34f | 5383 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5384 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5385 | * occur there and call schedule directly. |
5386 | */ | |
5387 | asmlinkage void __sched preempt_schedule(void) | |
5388 | { | |
5389 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5390 | |
1da177e4 LT |
5391 | /* |
5392 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5393 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5394 | */ |
beed33a8 | 5395 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5396 | return; |
5397 | ||
3a5c359a AK |
5398 | do { |
5399 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5400 | schedule(); |
3a5c359a | 5401 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5402 | |
3a5c359a AK |
5403 | /* |
5404 | * Check again in case we missed a preemption opportunity | |
5405 | * between schedule and now. | |
5406 | */ | |
5407 | barrier(); | |
5ed0cec0 | 5408 | } while (need_resched()); |
1da177e4 | 5409 | } |
1da177e4 LT |
5410 | EXPORT_SYMBOL(preempt_schedule); |
5411 | ||
5412 | /* | |
2ed6e34f | 5413 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5414 | * off of irq context. |
5415 | * Note, that this is called and return with irqs disabled. This will | |
5416 | * protect us against recursive calling from irq. | |
5417 | */ | |
5418 | asmlinkage void __sched preempt_schedule_irq(void) | |
5419 | { | |
5420 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5421 | |
2ed6e34f | 5422 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5423 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5424 | ||
3a5c359a AK |
5425 | do { |
5426 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5427 | local_irq_enable(); |
5428 | schedule(); | |
5429 | local_irq_disable(); | |
3a5c359a | 5430 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5431 | |
3a5c359a AK |
5432 | /* |
5433 | * Check again in case we missed a preemption opportunity | |
5434 | * between schedule and now. | |
5435 | */ | |
5436 | barrier(); | |
5ed0cec0 | 5437 | } while (need_resched()); |
1da177e4 LT |
5438 | } |
5439 | ||
5440 | #endif /* CONFIG_PREEMPT */ | |
5441 | ||
95cdf3b7 IM |
5442 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5443 | void *key) | |
1da177e4 | 5444 | { |
48f24c4d | 5445 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5446 | } |
1da177e4 LT |
5447 | EXPORT_SYMBOL(default_wake_function); |
5448 | ||
5449 | /* | |
41a2d6cf IM |
5450 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5451 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5452 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5453 | * | |
5454 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5455 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5456 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5457 | */ | |
78ddb08f | 5458 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5459 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5460 | { |
2e45874c | 5461 | wait_queue_t *curr, *next; |
1da177e4 | 5462 | |
2e45874c | 5463 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5464 | unsigned flags = curr->flags; |
5465 | ||
1da177e4 | 5466 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5467 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5468 | break; |
5469 | } | |
5470 | } | |
5471 | ||
5472 | /** | |
5473 | * __wake_up - wake up threads blocked on a waitqueue. | |
5474 | * @q: the waitqueue | |
5475 | * @mode: which threads | |
5476 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5477 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5478 | * |
5479 | * It may be assumed that this function implies a write memory barrier before | |
5480 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5481 | */ |
7ad5b3a5 | 5482 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5483 | int nr_exclusive, void *key) |
1da177e4 LT |
5484 | { |
5485 | unsigned long flags; | |
5486 | ||
5487 | spin_lock_irqsave(&q->lock, flags); | |
5488 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5489 | spin_unlock_irqrestore(&q->lock, flags); | |
5490 | } | |
1da177e4 LT |
5491 | EXPORT_SYMBOL(__wake_up); |
5492 | ||
5493 | /* | |
5494 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5495 | */ | |
7ad5b3a5 | 5496 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5497 | { |
5498 | __wake_up_common(q, mode, 1, 0, NULL); | |
5499 | } | |
5500 | ||
4ede816a DL |
5501 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5502 | { | |
5503 | __wake_up_common(q, mode, 1, 0, key); | |
5504 | } | |
5505 | ||
1da177e4 | 5506 | /** |
4ede816a | 5507 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5508 | * @q: the waitqueue |
5509 | * @mode: which threads | |
5510 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5511 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5512 | * |
5513 | * The sync wakeup differs that the waker knows that it will schedule | |
5514 | * away soon, so while the target thread will be woken up, it will not | |
5515 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5516 | * with each other. This can prevent needless bouncing between CPUs. | |
5517 | * | |
5518 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5519 | * |
5520 | * It may be assumed that this function implies a write memory barrier before | |
5521 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5522 | */ |
4ede816a DL |
5523 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5524 | int nr_exclusive, void *key) | |
1da177e4 LT |
5525 | { |
5526 | unsigned long flags; | |
5527 | int sync = 1; | |
5528 | ||
5529 | if (unlikely(!q)) | |
5530 | return; | |
5531 | ||
5532 | if (unlikely(!nr_exclusive)) | |
5533 | sync = 0; | |
5534 | ||
5535 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5536 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5537 | spin_unlock_irqrestore(&q->lock, flags); |
5538 | } | |
4ede816a DL |
5539 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5540 | ||
5541 | /* | |
5542 | * __wake_up_sync - see __wake_up_sync_key() | |
5543 | */ | |
5544 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5545 | { | |
5546 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5547 | } | |
1da177e4 LT |
5548 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5549 | ||
65eb3dc6 KD |
5550 | /** |
5551 | * complete: - signals a single thread waiting on this completion | |
5552 | * @x: holds the state of this particular completion | |
5553 | * | |
5554 | * This will wake up a single thread waiting on this completion. Threads will be | |
5555 | * awakened in the same order in which they were queued. | |
5556 | * | |
5557 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5558 | * |
5559 | * It may be assumed that this function implies a write memory barrier before | |
5560 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5561 | */ |
b15136e9 | 5562 | void complete(struct completion *x) |
1da177e4 LT |
5563 | { |
5564 | unsigned long flags; | |
5565 | ||
5566 | spin_lock_irqsave(&x->wait.lock, flags); | |
5567 | x->done++; | |
d9514f6c | 5568 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5569 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5570 | } | |
5571 | EXPORT_SYMBOL(complete); | |
5572 | ||
65eb3dc6 KD |
5573 | /** |
5574 | * complete_all: - signals all threads waiting on this completion | |
5575 | * @x: holds the state of this particular completion | |
5576 | * | |
5577 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5578 | * |
5579 | * It may be assumed that this function implies a write memory barrier before | |
5580 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5581 | */ |
b15136e9 | 5582 | void complete_all(struct completion *x) |
1da177e4 LT |
5583 | { |
5584 | unsigned long flags; | |
5585 | ||
5586 | spin_lock_irqsave(&x->wait.lock, flags); | |
5587 | x->done += UINT_MAX/2; | |
d9514f6c | 5588 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5589 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5590 | } | |
5591 | EXPORT_SYMBOL(complete_all); | |
5592 | ||
8cbbe86d AK |
5593 | static inline long __sched |
5594 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5595 | { |
1da177e4 LT |
5596 | if (!x->done) { |
5597 | DECLARE_WAITQUEUE(wait, current); | |
5598 | ||
5599 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5600 | __add_wait_queue_tail(&x->wait, &wait); | |
5601 | do { | |
94d3d824 | 5602 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5603 | timeout = -ERESTARTSYS; |
5604 | break; | |
8cbbe86d AK |
5605 | } |
5606 | __set_current_state(state); | |
1da177e4 LT |
5607 | spin_unlock_irq(&x->wait.lock); |
5608 | timeout = schedule_timeout(timeout); | |
5609 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5610 | } while (!x->done && timeout); |
1da177e4 | 5611 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5612 | if (!x->done) |
5613 | return timeout; | |
1da177e4 LT |
5614 | } |
5615 | x->done--; | |
ea71a546 | 5616 | return timeout ?: 1; |
1da177e4 | 5617 | } |
1da177e4 | 5618 | |
8cbbe86d AK |
5619 | static long __sched |
5620 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5621 | { |
1da177e4 LT |
5622 | might_sleep(); |
5623 | ||
5624 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5625 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5626 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5627 | return timeout; |
5628 | } | |
1da177e4 | 5629 | |
65eb3dc6 KD |
5630 | /** |
5631 | * wait_for_completion: - waits for completion of a task | |
5632 | * @x: holds the state of this particular completion | |
5633 | * | |
5634 | * This waits to be signaled for completion of a specific task. It is NOT | |
5635 | * interruptible and there is no timeout. | |
5636 | * | |
5637 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5638 | * and interrupt capability. Also see complete(). | |
5639 | */ | |
b15136e9 | 5640 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5641 | { |
5642 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5643 | } |
8cbbe86d | 5644 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5645 | |
65eb3dc6 KD |
5646 | /** |
5647 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5648 | * @x: holds the state of this particular completion | |
5649 | * @timeout: timeout value in jiffies | |
5650 | * | |
5651 | * This waits for either a completion of a specific task to be signaled or for a | |
5652 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5653 | * interruptible. | |
5654 | */ | |
b15136e9 | 5655 | unsigned long __sched |
8cbbe86d | 5656 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5657 | { |
8cbbe86d | 5658 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5659 | } |
8cbbe86d | 5660 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5661 | |
65eb3dc6 KD |
5662 | /** |
5663 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5664 | * @x: holds the state of this particular completion | |
5665 | * | |
5666 | * This waits for completion of a specific task to be signaled. It is | |
5667 | * interruptible. | |
5668 | */ | |
8cbbe86d | 5669 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5670 | { |
51e97990 AK |
5671 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5672 | if (t == -ERESTARTSYS) | |
5673 | return t; | |
5674 | return 0; | |
0fec171c | 5675 | } |
8cbbe86d | 5676 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5677 | |
65eb3dc6 KD |
5678 | /** |
5679 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5680 | * @x: holds the state of this particular completion | |
5681 | * @timeout: timeout value in jiffies | |
5682 | * | |
5683 | * This waits for either a completion of a specific task to be signaled or for a | |
5684 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5685 | */ | |
b15136e9 | 5686 | unsigned long __sched |
8cbbe86d AK |
5687 | wait_for_completion_interruptible_timeout(struct completion *x, |
5688 | unsigned long timeout) | |
0fec171c | 5689 | { |
8cbbe86d | 5690 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5691 | } |
8cbbe86d | 5692 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5693 | |
65eb3dc6 KD |
5694 | /** |
5695 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5696 | * @x: holds the state of this particular completion | |
5697 | * | |
5698 | * This waits to be signaled for completion of a specific task. It can be | |
5699 | * interrupted by a kill signal. | |
5700 | */ | |
009e577e MW |
5701 | int __sched wait_for_completion_killable(struct completion *x) |
5702 | { | |
5703 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5704 | if (t == -ERESTARTSYS) | |
5705 | return t; | |
5706 | return 0; | |
5707 | } | |
5708 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5709 | ||
be4de352 DC |
5710 | /** |
5711 | * try_wait_for_completion - try to decrement a completion without blocking | |
5712 | * @x: completion structure | |
5713 | * | |
5714 | * Returns: 0 if a decrement cannot be done without blocking | |
5715 | * 1 if a decrement succeeded. | |
5716 | * | |
5717 | * If a completion is being used as a counting completion, | |
5718 | * attempt to decrement the counter without blocking. This | |
5719 | * enables us to avoid waiting if the resource the completion | |
5720 | * is protecting is not available. | |
5721 | */ | |
5722 | bool try_wait_for_completion(struct completion *x) | |
5723 | { | |
5724 | int ret = 1; | |
5725 | ||
5726 | spin_lock_irq(&x->wait.lock); | |
5727 | if (!x->done) | |
5728 | ret = 0; | |
5729 | else | |
5730 | x->done--; | |
5731 | spin_unlock_irq(&x->wait.lock); | |
5732 | return ret; | |
5733 | } | |
5734 | EXPORT_SYMBOL(try_wait_for_completion); | |
5735 | ||
5736 | /** | |
5737 | * completion_done - Test to see if a completion has any waiters | |
5738 | * @x: completion structure | |
5739 | * | |
5740 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5741 | * 1 if there are no waiters. | |
5742 | * | |
5743 | */ | |
5744 | bool completion_done(struct completion *x) | |
5745 | { | |
5746 | int ret = 1; | |
5747 | ||
5748 | spin_lock_irq(&x->wait.lock); | |
5749 | if (!x->done) | |
5750 | ret = 0; | |
5751 | spin_unlock_irq(&x->wait.lock); | |
5752 | return ret; | |
5753 | } | |
5754 | EXPORT_SYMBOL(completion_done); | |
5755 | ||
8cbbe86d AK |
5756 | static long __sched |
5757 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5758 | { |
0fec171c IM |
5759 | unsigned long flags; |
5760 | wait_queue_t wait; | |
5761 | ||
5762 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5763 | |
8cbbe86d | 5764 | __set_current_state(state); |
1da177e4 | 5765 | |
8cbbe86d AK |
5766 | spin_lock_irqsave(&q->lock, flags); |
5767 | __add_wait_queue(q, &wait); | |
5768 | spin_unlock(&q->lock); | |
5769 | timeout = schedule_timeout(timeout); | |
5770 | spin_lock_irq(&q->lock); | |
5771 | __remove_wait_queue(q, &wait); | |
5772 | spin_unlock_irqrestore(&q->lock, flags); | |
5773 | ||
5774 | return timeout; | |
5775 | } | |
5776 | ||
5777 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5778 | { | |
5779 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5780 | } |
1da177e4 LT |
5781 | EXPORT_SYMBOL(interruptible_sleep_on); |
5782 | ||
0fec171c | 5783 | long __sched |
95cdf3b7 | 5784 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5785 | { |
8cbbe86d | 5786 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5787 | } |
1da177e4 LT |
5788 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5789 | ||
0fec171c | 5790 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5791 | { |
8cbbe86d | 5792 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5793 | } |
1da177e4 LT |
5794 | EXPORT_SYMBOL(sleep_on); |
5795 | ||
0fec171c | 5796 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5797 | { |
8cbbe86d | 5798 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5799 | } |
1da177e4 LT |
5800 | EXPORT_SYMBOL(sleep_on_timeout); |
5801 | ||
b29739f9 IM |
5802 | #ifdef CONFIG_RT_MUTEXES |
5803 | ||
5804 | /* | |
5805 | * rt_mutex_setprio - set the current priority of a task | |
5806 | * @p: task | |
5807 | * @prio: prio value (kernel-internal form) | |
5808 | * | |
5809 | * This function changes the 'effective' priority of a task. It does | |
5810 | * not touch ->normal_prio like __setscheduler(). | |
5811 | * | |
5812 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5813 | */ | |
36c8b586 | 5814 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5815 | { |
5816 | unsigned long flags; | |
83b699ed | 5817 | int oldprio, on_rq, running; |
70b97a7f | 5818 | struct rq *rq; |
cb469845 | 5819 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5820 | |
5821 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5822 | ||
5823 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5824 | update_rq_clock(rq); |
b29739f9 | 5825 | |
d5f9f942 | 5826 | oldprio = p->prio; |
dd41f596 | 5827 | on_rq = p->se.on_rq; |
051a1d1a | 5828 | running = task_current(rq, p); |
0e1f3483 | 5829 | if (on_rq) |
69be72c1 | 5830 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5831 | if (running) |
5832 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5833 | |
5834 | if (rt_prio(prio)) | |
5835 | p->sched_class = &rt_sched_class; | |
5836 | else | |
5837 | p->sched_class = &fair_sched_class; | |
5838 | ||
b29739f9 IM |
5839 | p->prio = prio; |
5840 | ||
0e1f3483 HS |
5841 | if (running) |
5842 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5843 | if (on_rq) { |
8159f87e | 5844 | enqueue_task(rq, p, 0); |
cb469845 SR |
5845 | |
5846 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5847 | } |
5848 | task_rq_unlock(rq, &flags); | |
5849 | } | |
5850 | ||
5851 | #endif | |
5852 | ||
36c8b586 | 5853 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5854 | { |
dd41f596 | 5855 | int old_prio, delta, on_rq; |
1da177e4 | 5856 | unsigned long flags; |
70b97a7f | 5857 | struct rq *rq; |
1da177e4 LT |
5858 | |
5859 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5860 | return; | |
5861 | /* | |
5862 | * We have to be careful, if called from sys_setpriority(), | |
5863 | * the task might be in the middle of scheduling on another CPU. | |
5864 | */ | |
5865 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5866 | update_rq_clock(rq); |
1da177e4 LT |
5867 | /* |
5868 | * The RT priorities are set via sched_setscheduler(), but we still | |
5869 | * allow the 'normal' nice value to be set - but as expected | |
5870 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5871 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5872 | */ |
e05606d3 | 5873 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5874 | p->static_prio = NICE_TO_PRIO(nice); |
5875 | goto out_unlock; | |
5876 | } | |
dd41f596 | 5877 | on_rq = p->se.on_rq; |
c09595f6 | 5878 | if (on_rq) |
69be72c1 | 5879 | dequeue_task(rq, p, 0); |
1da177e4 | 5880 | |
1da177e4 | 5881 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5882 | set_load_weight(p); |
b29739f9 IM |
5883 | old_prio = p->prio; |
5884 | p->prio = effective_prio(p); | |
5885 | delta = p->prio - old_prio; | |
1da177e4 | 5886 | |
dd41f596 | 5887 | if (on_rq) { |
8159f87e | 5888 | enqueue_task(rq, p, 0); |
1da177e4 | 5889 | /* |
d5f9f942 AM |
5890 | * If the task increased its priority or is running and |
5891 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5892 | */ |
d5f9f942 | 5893 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5894 | resched_task(rq->curr); |
5895 | } | |
5896 | out_unlock: | |
5897 | task_rq_unlock(rq, &flags); | |
5898 | } | |
1da177e4 LT |
5899 | EXPORT_SYMBOL(set_user_nice); |
5900 | ||
e43379f1 MM |
5901 | /* |
5902 | * can_nice - check if a task can reduce its nice value | |
5903 | * @p: task | |
5904 | * @nice: nice value | |
5905 | */ | |
36c8b586 | 5906 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5907 | { |
024f4747 MM |
5908 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5909 | int nice_rlim = 20 - nice; | |
48f24c4d | 5910 | |
e43379f1 MM |
5911 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5912 | capable(CAP_SYS_NICE)); | |
5913 | } | |
5914 | ||
1da177e4 LT |
5915 | #ifdef __ARCH_WANT_SYS_NICE |
5916 | ||
5917 | /* | |
5918 | * sys_nice - change the priority of the current process. | |
5919 | * @increment: priority increment | |
5920 | * | |
5921 | * sys_setpriority is a more generic, but much slower function that | |
5922 | * does similar things. | |
5923 | */ | |
5add95d4 | 5924 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5925 | { |
48f24c4d | 5926 | long nice, retval; |
1da177e4 LT |
5927 | |
5928 | /* | |
5929 | * Setpriority might change our priority at the same moment. | |
5930 | * We don't have to worry. Conceptually one call occurs first | |
5931 | * and we have a single winner. | |
5932 | */ | |
e43379f1 MM |
5933 | if (increment < -40) |
5934 | increment = -40; | |
1da177e4 LT |
5935 | if (increment > 40) |
5936 | increment = 40; | |
5937 | ||
2b8f836f | 5938 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5939 | if (nice < -20) |
5940 | nice = -20; | |
5941 | if (nice > 19) | |
5942 | nice = 19; | |
5943 | ||
e43379f1 MM |
5944 | if (increment < 0 && !can_nice(current, nice)) |
5945 | return -EPERM; | |
5946 | ||
1da177e4 LT |
5947 | retval = security_task_setnice(current, nice); |
5948 | if (retval) | |
5949 | return retval; | |
5950 | ||
5951 | set_user_nice(current, nice); | |
5952 | return 0; | |
5953 | } | |
5954 | ||
5955 | #endif | |
5956 | ||
5957 | /** | |
5958 | * task_prio - return the priority value of a given task. | |
5959 | * @p: the task in question. | |
5960 | * | |
5961 | * This is the priority value as seen by users in /proc. | |
5962 | * RT tasks are offset by -200. Normal tasks are centered | |
5963 | * around 0, value goes from -16 to +15. | |
5964 | */ | |
36c8b586 | 5965 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5966 | { |
5967 | return p->prio - MAX_RT_PRIO; | |
5968 | } | |
5969 | ||
5970 | /** | |
5971 | * task_nice - return the nice value of a given task. | |
5972 | * @p: the task in question. | |
5973 | */ | |
36c8b586 | 5974 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5975 | { |
5976 | return TASK_NICE(p); | |
5977 | } | |
150d8bed | 5978 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5979 | |
5980 | /** | |
5981 | * idle_cpu - is a given cpu idle currently? | |
5982 | * @cpu: the processor in question. | |
5983 | */ | |
5984 | int idle_cpu(int cpu) | |
5985 | { | |
5986 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5987 | } | |
5988 | ||
1da177e4 LT |
5989 | /** |
5990 | * idle_task - return the idle task for a given cpu. | |
5991 | * @cpu: the processor in question. | |
5992 | */ | |
36c8b586 | 5993 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5994 | { |
5995 | return cpu_rq(cpu)->idle; | |
5996 | } | |
5997 | ||
5998 | /** | |
5999 | * find_process_by_pid - find a process with a matching PID value. | |
6000 | * @pid: the pid in question. | |
6001 | */ | |
a9957449 | 6002 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6003 | { |
228ebcbe | 6004 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6005 | } |
6006 | ||
6007 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6008 | static void |
6009 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6010 | { |
dd41f596 | 6011 | BUG_ON(p->se.on_rq); |
48f24c4d | 6012 | |
1da177e4 | 6013 | p->policy = policy; |
dd41f596 IM |
6014 | switch (p->policy) { |
6015 | case SCHED_NORMAL: | |
6016 | case SCHED_BATCH: | |
6017 | case SCHED_IDLE: | |
6018 | p->sched_class = &fair_sched_class; | |
6019 | break; | |
6020 | case SCHED_FIFO: | |
6021 | case SCHED_RR: | |
6022 | p->sched_class = &rt_sched_class; | |
6023 | break; | |
6024 | } | |
6025 | ||
1da177e4 | 6026 | p->rt_priority = prio; |
b29739f9 IM |
6027 | p->normal_prio = normal_prio(p); |
6028 | /* we are holding p->pi_lock already */ | |
6029 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 6030 | set_load_weight(p); |
1da177e4 LT |
6031 | } |
6032 | ||
c69e8d9c DH |
6033 | /* |
6034 | * check the target process has a UID that matches the current process's | |
6035 | */ | |
6036 | static bool check_same_owner(struct task_struct *p) | |
6037 | { | |
6038 | const struct cred *cred = current_cred(), *pcred; | |
6039 | bool match; | |
6040 | ||
6041 | rcu_read_lock(); | |
6042 | pcred = __task_cred(p); | |
6043 | match = (cred->euid == pcred->euid || | |
6044 | cred->euid == pcred->uid); | |
6045 | rcu_read_unlock(); | |
6046 | return match; | |
6047 | } | |
6048 | ||
961ccddd RR |
6049 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6050 | struct sched_param *param, bool user) | |
1da177e4 | 6051 | { |
83b699ed | 6052 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6053 | unsigned long flags; |
cb469845 | 6054 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6055 | struct rq *rq; |
1da177e4 | 6056 | |
66e5393a SR |
6057 | /* may grab non-irq protected spin_locks */ |
6058 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6059 | recheck: |
6060 | /* double check policy once rq lock held */ | |
6061 | if (policy < 0) | |
6062 | policy = oldpolicy = p->policy; | |
6063 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
6064 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
6065 | policy != SCHED_IDLE) | |
b0a9499c | 6066 | return -EINVAL; |
1da177e4 LT |
6067 | /* |
6068 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6069 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6070 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6071 | */ |
6072 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6073 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6074 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6075 | return -EINVAL; |
e05606d3 | 6076 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6077 | return -EINVAL; |
6078 | ||
37e4ab3f OC |
6079 | /* |
6080 | * Allow unprivileged RT tasks to decrease priority: | |
6081 | */ | |
961ccddd | 6082 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6083 | if (rt_policy(policy)) { |
8dc3e909 | 6084 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6085 | |
6086 | if (!lock_task_sighand(p, &flags)) | |
6087 | return -ESRCH; | |
6088 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6089 | unlock_task_sighand(p, &flags); | |
6090 | ||
6091 | /* can't set/change the rt policy */ | |
6092 | if (policy != p->policy && !rlim_rtprio) | |
6093 | return -EPERM; | |
6094 | ||
6095 | /* can't increase priority */ | |
6096 | if (param->sched_priority > p->rt_priority && | |
6097 | param->sched_priority > rlim_rtprio) | |
6098 | return -EPERM; | |
6099 | } | |
dd41f596 IM |
6100 | /* |
6101 | * Like positive nice levels, dont allow tasks to | |
6102 | * move out of SCHED_IDLE either: | |
6103 | */ | |
6104 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6105 | return -EPERM; | |
5fe1d75f | 6106 | |
37e4ab3f | 6107 | /* can't change other user's priorities */ |
c69e8d9c | 6108 | if (!check_same_owner(p)) |
37e4ab3f OC |
6109 | return -EPERM; |
6110 | } | |
1da177e4 | 6111 | |
725aad24 | 6112 | if (user) { |
b68aa230 | 6113 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6114 | /* |
6115 | * Do not allow realtime tasks into groups that have no runtime | |
6116 | * assigned. | |
6117 | */ | |
9a7e0b18 PZ |
6118 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6119 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6120 | return -EPERM; |
b68aa230 PZ |
6121 | #endif |
6122 | ||
725aad24 JF |
6123 | retval = security_task_setscheduler(p, policy, param); |
6124 | if (retval) | |
6125 | return retval; | |
6126 | } | |
6127 | ||
b29739f9 IM |
6128 | /* |
6129 | * make sure no PI-waiters arrive (or leave) while we are | |
6130 | * changing the priority of the task: | |
6131 | */ | |
6132 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6133 | /* |
6134 | * To be able to change p->policy safely, the apropriate | |
6135 | * runqueue lock must be held. | |
6136 | */ | |
b29739f9 | 6137 | rq = __task_rq_lock(p); |
1da177e4 LT |
6138 | /* recheck policy now with rq lock held */ |
6139 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6140 | policy = oldpolicy = -1; | |
b29739f9 IM |
6141 | __task_rq_unlock(rq); |
6142 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6143 | goto recheck; |
6144 | } | |
2daa3577 | 6145 | update_rq_clock(rq); |
dd41f596 | 6146 | on_rq = p->se.on_rq; |
051a1d1a | 6147 | running = task_current(rq, p); |
0e1f3483 | 6148 | if (on_rq) |
2e1cb74a | 6149 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6150 | if (running) |
6151 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6152 | |
1da177e4 | 6153 | oldprio = p->prio; |
dd41f596 | 6154 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6155 | |
0e1f3483 HS |
6156 | if (running) |
6157 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6158 | if (on_rq) { |
6159 | activate_task(rq, p, 0); | |
cb469845 SR |
6160 | |
6161 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6162 | } |
b29739f9 IM |
6163 | __task_rq_unlock(rq); |
6164 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6165 | ||
95e02ca9 TG |
6166 | rt_mutex_adjust_pi(p); |
6167 | ||
1da177e4 LT |
6168 | return 0; |
6169 | } | |
961ccddd RR |
6170 | |
6171 | /** | |
6172 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6173 | * @p: the task in question. | |
6174 | * @policy: new policy. | |
6175 | * @param: structure containing the new RT priority. | |
6176 | * | |
6177 | * NOTE that the task may be already dead. | |
6178 | */ | |
6179 | int sched_setscheduler(struct task_struct *p, int policy, | |
6180 | struct sched_param *param) | |
6181 | { | |
6182 | return __sched_setscheduler(p, policy, param, true); | |
6183 | } | |
1da177e4 LT |
6184 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6185 | ||
961ccddd RR |
6186 | /** |
6187 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6188 | * @p: the task in question. | |
6189 | * @policy: new policy. | |
6190 | * @param: structure containing the new RT priority. | |
6191 | * | |
6192 | * Just like sched_setscheduler, only don't bother checking if the | |
6193 | * current context has permission. For example, this is needed in | |
6194 | * stop_machine(): we create temporary high priority worker threads, | |
6195 | * but our caller might not have that capability. | |
6196 | */ | |
6197 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6198 | struct sched_param *param) | |
6199 | { | |
6200 | return __sched_setscheduler(p, policy, param, false); | |
6201 | } | |
6202 | ||
95cdf3b7 IM |
6203 | static int |
6204 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6205 | { |
1da177e4 LT |
6206 | struct sched_param lparam; |
6207 | struct task_struct *p; | |
36c8b586 | 6208 | int retval; |
1da177e4 LT |
6209 | |
6210 | if (!param || pid < 0) | |
6211 | return -EINVAL; | |
6212 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6213 | return -EFAULT; | |
5fe1d75f ON |
6214 | |
6215 | rcu_read_lock(); | |
6216 | retval = -ESRCH; | |
1da177e4 | 6217 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6218 | if (p != NULL) |
6219 | retval = sched_setscheduler(p, policy, &lparam); | |
6220 | rcu_read_unlock(); | |
36c8b586 | 6221 | |
1da177e4 LT |
6222 | return retval; |
6223 | } | |
6224 | ||
6225 | /** | |
6226 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6227 | * @pid: the pid in question. | |
6228 | * @policy: new policy. | |
6229 | * @param: structure containing the new RT priority. | |
6230 | */ | |
5add95d4 HC |
6231 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6232 | struct sched_param __user *, param) | |
1da177e4 | 6233 | { |
c21761f1 JB |
6234 | /* negative values for policy are not valid */ |
6235 | if (policy < 0) | |
6236 | return -EINVAL; | |
6237 | ||
1da177e4 LT |
6238 | return do_sched_setscheduler(pid, policy, param); |
6239 | } | |
6240 | ||
6241 | /** | |
6242 | * sys_sched_setparam - set/change the RT priority of a thread | |
6243 | * @pid: the pid in question. | |
6244 | * @param: structure containing the new RT priority. | |
6245 | */ | |
5add95d4 | 6246 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6247 | { |
6248 | return do_sched_setscheduler(pid, -1, param); | |
6249 | } | |
6250 | ||
6251 | /** | |
6252 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6253 | * @pid: the pid in question. | |
6254 | */ | |
5add95d4 | 6255 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6256 | { |
36c8b586 | 6257 | struct task_struct *p; |
3a5c359a | 6258 | int retval; |
1da177e4 LT |
6259 | |
6260 | if (pid < 0) | |
3a5c359a | 6261 | return -EINVAL; |
1da177e4 LT |
6262 | |
6263 | retval = -ESRCH; | |
6264 | read_lock(&tasklist_lock); | |
6265 | p = find_process_by_pid(pid); | |
6266 | if (p) { | |
6267 | retval = security_task_getscheduler(p); | |
6268 | if (!retval) | |
6269 | retval = p->policy; | |
6270 | } | |
6271 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6272 | return retval; |
6273 | } | |
6274 | ||
6275 | /** | |
6276 | * sys_sched_getscheduler - get the RT priority of a thread | |
6277 | * @pid: the pid in question. | |
6278 | * @param: structure containing the RT priority. | |
6279 | */ | |
5add95d4 | 6280 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6281 | { |
6282 | struct sched_param lp; | |
36c8b586 | 6283 | struct task_struct *p; |
3a5c359a | 6284 | int retval; |
1da177e4 LT |
6285 | |
6286 | if (!param || pid < 0) | |
3a5c359a | 6287 | return -EINVAL; |
1da177e4 LT |
6288 | |
6289 | read_lock(&tasklist_lock); | |
6290 | p = find_process_by_pid(pid); | |
6291 | retval = -ESRCH; | |
6292 | if (!p) | |
6293 | goto out_unlock; | |
6294 | ||
6295 | retval = security_task_getscheduler(p); | |
6296 | if (retval) | |
6297 | goto out_unlock; | |
6298 | ||
6299 | lp.sched_priority = p->rt_priority; | |
6300 | read_unlock(&tasklist_lock); | |
6301 | ||
6302 | /* | |
6303 | * This one might sleep, we cannot do it with a spinlock held ... | |
6304 | */ | |
6305 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6306 | ||
1da177e4 LT |
6307 | return retval; |
6308 | ||
6309 | out_unlock: | |
6310 | read_unlock(&tasklist_lock); | |
6311 | return retval; | |
6312 | } | |
6313 | ||
96f874e2 | 6314 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6315 | { |
5a16f3d3 | 6316 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6317 | struct task_struct *p; |
6318 | int retval; | |
1da177e4 | 6319 | |
95402b38 | 6320 | get_online_cpus(); |
1da177e4 LT |
6321 | read_lock(&tasklist_lock); |
6322 | ||
6323 | p = find_process_by_pid(pid); | |
6324 | if (!p) { | |
6325 | read_unlock(&tasklist_lock); | |
95402b38 | 6326 | put_online_cpus(); |
1da177e4 LT |
6327 | return -ESRCH; |
6328 | } | |
6329 | ||
6330 | /* | |
6331 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6332 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6333 | * usage count and then drop tasklist_lock. |
6334 | */ | |
6335 | get_task_struct(p); | |
6336 | read_unlock(&tasklist_lock); | |
6337 | ||
5a16f3d3 RR |
6338 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6339 | retval = -ENOMEM; | |
6340 | goto out_put_task; | |
6341 | } | |
6342 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6343 | retval = -ENOMEM; | |
6344 | goto out_free_cpus_allowed; | |
6345 | } | |
1da177e4 | 6346 | retval = -EPERM; |
c69e8d9c | 6347 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6348 | goto out_unlock; |
6349 | ||
e7834f8f DQ |
6350 | retval = security_task_setscheduler(p, 0, NULL); |
6351 | if (retval) | |
6352 | goto out_unlock; | |
6353 | ||
5a16f3d3 RR |
6354 | cpuset_cpus_allowed(p, cpus_allowed); |
6355 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6356 | again: |
5a16f3d3 | 6357 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6358 | |
8707d8b8 | 6359 | if (!retval) { |
5a16f3d3 RR |
6360 | cpuset_cpus_allowed(p, cpus_allowed); |
6361 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6362 | /* |
6363 | * We must have raced with a concurrent cpuset | |
6364 | * update. Just reset the cpus_allowed to the | |
6365 | * cpuset's cpus_allowed | |
6366 | */ | |
5a16f3d3 | 6367 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6368 | goto again; |
6369 | } | |
6370 | } | |
1da177e4 | 6371 | out_unlock: |
5a16f3d3 RR |
6372 | free_cpumask_var(new_mask); |
6373 | out_free_cpus_allowed: | |
6374 | free_cpumask_var(cpus_allowed); | |
6375 | out_put_task: | |
1da177e4 | 6376 | put_task_struct(p); |
95402b38 | 6377 | put_online_cpus(); |
1da177e4 LT |
6378 | return retval; |
6379 | } | |
6380 | ||
6381 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6382 | struct cpumask *new_mask) |
1da177e4 | 6383 | { |
96f874e2 RR |
6384 | if (len < cpumask_size()) |
6385 | cpumask_clear(new_mask); | |
6386 | else if (len > cpumask_size()) | |
6387 | len = cpumask_size(); | |
6388 | ||
1da177e4 LT |
6389 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6390 | } | |
6391 | ||
6392 | /** | |
6393 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6394 | * @pid: pid of the process | |
6395 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6396 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6397 | */ | |
5add95d4 HC |
6398 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6399 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6400 | { |
5a16f3d3 | 6401 | cpumask_var_t new_mask; |
1da177e4 LT |
6402 | int retval; |
6403 | ||
5a16f3d3 RR |
6404 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6405 | return -ENOMEM; | |
1da177e4 | 6406 | |
5a16f3d3 RR |
6407 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6408 | if (retval == 0) | |
6409 | retval = sched_setaffinity(pid, new_mask); | |
6410 | free_cpumask_var(new_mask); | |
6411 | return retval; | |
1da177e4 LT |
6412 | } |
6413 | ||
96f874e2 | 6414 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6415 | { |
36c8b586 | 6416 | struct task_struct *p; |
1da177e4 | 6417 | int retval; |
1da177e4 | 6418 | |
95402b38 | 6419 | get_online_cpus(); |
1da177e4 LT |
6420 | read_lock(&tasklist_lock); |
6421 | ||
6422 | retval = -ESRCH; | |
6423 | p = find_process_by_pid(pid); | |
6424 | if (!p) | |
6425 | goto out_unlock; | |
6426 | ||
e7834f8f DQ |
6427 | retval = security_task_getscheduler(p); |
6428 | if (retval) | |
6429 | goto out_unlock; | |
6430 | ||
96f874e2 | 6431 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6432 | |
6433 | out_unlock: | |
6434 | read_unlock(&tasklist_lock); | |
95402b38 | 6435 | put_online_cpus(); |
1da177e4 | 6436 | |
9531b62f | 6437 | return retval; |
1da177e4 LT |
6438 | } |
6439 | ||
6440 | /** | |
6441 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6442 | * @pid: pid of the process | |
6443 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6444 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6445 | */ | |
5add95d4 HC |
6446 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6447 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6448 | { |
6449 | int ret; | |
f17c8607 | 6450 | cpumask_var_t mask; |
1da177e4 | 6451 | |
f17c8607 | 6452 | if (len < cpumask_size()) |
1da177e4 LT |
6453 | return -EINVAL; |
6454 | ||
f17c8607 RR |
6455 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6456 | return -ENOMEM; | |
1da177e4 | 6457 | |
f17c8607 RR |
6458 | ret = sched_getaffinity(pid, mask); |
6459 | if (ret == 0) { | |
6460 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6461 | ret = -EFAULT; | |
6462 | else | |
6463 | ret = cpumask_size(); | |
6464 | } | |
6465 | free_cpumask_var(mask); | |
1da177e4 | 6466 | |
f17c8607 | 6467 | return ret; |
1da177e4 LT |
6468 | } |
6469 | ||
6470 | /** | |
6471 | * sys_sched_yield - yield the current processor to other threads. | |
6472 | * | |
dd41f596 IM |
6473 | * This function yields the current CPU to other tasks. If there are no |
6474 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6475 | */ |
5add95d4 | 6476 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6477 | { |
70b97a7f | 6478 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6479 | |
2d72376b | 6480 | schedstat_inc(rq, yld_count); |
4530d7ab | 6481 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6482 | |
6483 | /* | |
6484 | * Since we are going to call schedule() anyway, there's | |
6485 | * no need to preempt or enable interrupts: | |
6486 | */ | |
6487 | __release(rq->lock); | |
8a25d5de | 6488 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6489 | _raw_spin_unlock(&rq->lock); |
6490 | preempt_enable_no_resched(); | |
6491 | ||
6492 | schedule(); | |
6493 | ||
6494 | return 0; | |
6495 | } | |
6496 | ||
e7b38404 | 6497 | static void __cond_resched(void) |
1da177e4 | 6498 | { |
8e0a43d8 IM |
6499 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
6500 | __might_sleep(__FILE__, __LINE__); | |
6501 | #endif | |
5bbcfd90 IM |
6502 | /* |
6503 | * The BKS might be reacquired before we have dropped | |
6504 | * PREEMPT_ACTIVE, which could trigger a second | |
6505 | * cond_resched() call. | |
6506 | */ | |
1da177e4 LT |
6507 | do { |
6508 | add_preempt_count(PREEMPT_ACTIVE); | |
6509 | schedule(); | |
6510 | sub_preempt_count(PREEMPT_ACTIVE); | |
6511 | } while (need_resched()); | |
6512 | } | |
6513 | ||
02b67cc3 | 6514 | int __sched _cond_resched(void) |
1da177e4 | 6515 | { |
9414232f IM |
6516 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
6517 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
6518 | __cond_resched(); |
6519 | return 1; | |
6520 | } | |
6521 | return 0; | |
6522 | } | |
02b67cc3 | 6523 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6524 | |
6525 | /* | |
6526 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
6527 | * call schedule, and on return reacquire the lock. | |
6528 | * | |
41a2d6cf | 6529 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6530 | * operations here to prevent schedule() from being called twice (once via |
6531 | * spin_unlock(), once by hand). | |
6532 | */ | |
95cdf3b7 | 6533 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6534 | { |
95c354fe | 6535 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
6536 | int ret = 0; |
6537 | ||
95c354fe | 6538 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6539 | spin_unlock(lock); |
95c354fe NP |
6540 | if (resched && need_resched()) |
6541 | __cond_resched(); | |
6542 | else | |
6543 | cpu_relax(); | |
6df3cecb | 6544 | ret = 1; |
1da177e4 | 6545 | spin_lock(lock); |
1da177e4 | 6546 | } |
6df3cecb | 6547 | return ret; |
1da177e4 | 6548 | } |
1da177e4 LT |
6549 | EXPORT_SYMBOL(cond_resched_lock); |
6550 | ||
6551 | int __sched cond_resched_softirq(void) | |
6552 | { | |
6553 | BUG_ON(!in_softirq()); | |
6554 | ||
9414232f | 6555 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 6556 | local_bh_enable(); |
1da177e4 LT |
6557 | __cond_resched(); |
6558 | local_bh_disable(); | |
6559 | return 1; | |
6560 | } | |
6561 | return 0; | |
6562 | } | |
1da177e4 LT |
6563 | EXPORT_SYMBOL(cond_resched_softirq); |
6564 | ||
1da177e4 LT |
6565 | /** |
6566 | * yield - yield the current processor to other threads. | |
6567 | * | |
72fd4a35 | 6568 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6569 | * thread runnable and calls sys_sched_yield(). |
6570 | */ | |
6571 | void __sched yield(void) | |
6572 | { | |
6573 | set_current_state(TASK_RUNNING); | |
6574 | sys_sched_yield(); | |
6575 | } | |
1da177e4 LT |
6576 | EXPORT_SYMBOL(yield); |
6577 | ||
6578 | /* | |
41a2d6cf | 6579 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6580 | * that process accounting knows that this is a task in IO wait state. |
6581 | * | |
6582 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6583 | * has set its backing_dev_info: the queue against which it should throttle) | |
6584 | */ | |
6585 | void __sched io_schedule(void) | |
6586 | { | |
70b97a7f | 6587 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 6588 | |
0ff92245 | 6589 | delayacct_blkio_start(); |
1da177e4 LT |
6590 | atomic_inc(&rq->nr_iowait); |
6591 | schedule(); | |
6592 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6593 | delayacct_blkio_end(); |
1da177e4 | 6594 | } |
1da177e4 LT |
6595 | EXPORT_SYMBOL(io_schedule); |
6596 | ||
6597 | long __sched io_schedule_timeout(long timeout) | |
6598 | { | |
70b97a7f | 6599 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
6600 | long ret; |
6601 | ||
0ff92245 | 6602 | delayacct_blkio_start(); |
1da177e4 LT |
6603 | atomic_inc(&rq->nr_iowait); |
6604 | ret = schedule_timeout(timeout); | |
6605 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6606 | delayacct_blkio_end(); |
1da177e4 LT |
6607 | return ret; |
6608 | } | |
6609 | ||
6610 | /** | |
6611 | * sys_sched_get_priority_max - return maximum RT priority. | |
6612 | * @policy: scheduling class. | |
6613 | * | |
6614 | * this syscall returns the maximum rt_priority that can be used | |
6615 | * by a given scheduling class. | |
6616 | */ | |
5add95d4 | 6617 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6618 | { |
6619 | int ret = -EINVAL; | |
6620 | ||
6621 | switch (policy) { | |
6622 | case SCHED_FIFO: | |
6623 | case SCHED_RR: | |
6624 | ret = MAX_USER_RT_PRIO-1; | |
6625 | break; | |
6626 | case SCHED_NORMAL: | |
b0a9499c | 6627 | case SCHED_BATCH: |
dd41f596 | 6628 | case SCHED_IDLE: |
1da177e4 LT |
6629 | ret = 0; |
6630 | break; | |
6631 | } | |
6632 | return ret; | |
6633 | } | |
6634 | ||
6635 | /** | |
6636 | * sys_sched_get_priority_min - return minimum RT priority. | |
6637 | * @policy: scheduling class. | |
6638 | * | |
6639 | * this syscall returns the minimum rt_priority that can be used | |
6640 | * by a given scheduling class. | |
6641 | */ | |
5add95d4 | 6642 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6643 | { |
6644 | int ret = -EINVAL; | |
6645 | ||
6646 | switch (policy) { | |
6647 | case SCHED_FIFO: | |
6648 | case SCHED_RR: | |
6649 | ret = 1; | |
6650 | break; | |
6651 | case SCHED_NORMAL: | |
b0a9499c | 6652 | case SCHED_BATCH: |
dd41f596 | 6653 | case SCHED_IDLE: |
1da177e4 LT |
6654 | ret = 0; |
6655 | } | |
6656 | return ret; | |
6657 | } | |
6658 | ||
6659 | /** | |
6660 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6661 | * @pid: pid of the process. | |
6662 | * @interval: userspace pointer to the timeslice value. | |
6663 | * | |
6664 | * this syscall writes the default timeslice value of a given process | |
6665 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6666 | */ | |
17da2bd9 | 6667 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6668 | struct timespec __user *, interval) |
1da177e4 | 6669 | { |
36c8b586 | 6670 | struct task_struct *p; |
a4ec24b4 | 6671 | unsigned int time_slice; |
3a5c359a | 6672 | int retval; |
1da177e4 | 6673 | struct timespec t; |
1da177e4 LT |
6674 | |
6675 | if (pid < 0) | |
3a5c359a | 6676 | return -EINVAL; |
1da177e4 LT |
6677 | |
6678 | retval = -ESRCH; | |
6679 | read_lock(&tasklist_lock); | |
6680 | p = find_process_by_pid(pid); | |
6681 | if (!p) | |
6682 | goto out_unlock; | |
6683 | ||
6684 | retval = security_task_getscheduler(p); | |
6685 | if (retval) | |
6686 | goto out_unlock; | |
6687 | ||
77034937 IM |
6688 | /* |
6689 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6690 | * tasks that are on an otherwise idle runqueue: | |
6691 | */ | |
6692 | time_slice = 0; | |
6693 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6694 | time_slice = DEF_TIMESLICE; |
1868f958 | 6695 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6696 | struct sched_entity *se = &p->se; |
6697 | unsigned long flags; | |
6698 | struct rq *rq; | |
6699 | ||
6700 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6701 | if (rq->cfs.load.weight) |
6702 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6703 | task_rq_unlock(rq, &flags); |
6704 | } | |
1da177e4 | 6705 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6706 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6707 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6708 | return retval; |
3a5c359a | 6709 | |
1da177e4 LT |
6710 | out_unlock: |
6711 | read_unlock(&tasklist_lock); | |
6712 | return retval; | |
6713 | } | |
6714 | ||
7c731e0a | 6715 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6716 | |
82a1fcb9 | 6717 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6718 | { |
1da177e4 | 6719 | unsigned long free = 0; |
36c8b586 | 6720 | unsigned state; |
1da177e4 | 6721 | |
1da177e4 | 6722 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6723 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6724 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6725 | #if BITS_PER_LONG == 32 |
1da177e4 | 6726 | if (state == TASK_RUNNING) |
cc4ea795 | 6727 | printk(KERN_CONT " running "); |
1da177e4 | 6728 | else |
cc4ea795 | 6729 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6730 | #else |
6731 | if (state == TASK_RUNNING) | |
cc4ea795 | 6732 | printk(KERN_CONT " running task "); |
1da177e4 | 6733 | else |
cc4ea795 | 6734 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6735 | #endif |
6736 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6737 | free = stack_not_used(p); |
1da177e4 | 6738 | #endif |
aa47b7e0 DR |
6739 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
6740 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
6741 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6742 | |
5fb5e6de | 6743 | show_stack(p, NULL); |
1da177e4 LT |
6744 | } |
6745 | ||
e59e2ae2 | 6746 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6747 | { |
36c8b586 | 6748 | struct task_struct *g, *p; |
1da177e4 | 6749 | |
4bd77321 IM |
6750 | #if BITS_PER_LONG == 32 |
6751 | printk(KERN_INFO | |
6752 | " task PC stack pid father\n"); | |
1da177e4 | 6753 | #else |
4bd77321 IM |
6754 | printk(KERN_INFO |
6755 | " task PC stack pid father\n"); | |
1da177e4 LT |
6756 | #endif |
6757 | read_lock(&tasklist_lock); | |
6758 | do_each_thread(g, p) { | |
6759 | /* | |
6760 | * reset the NMI-timeout, listing all files on a slow | |
6761 | * console might take alot of time: | |
6762 | */ | |
6763 | touch_nmi_watchdog(); | |
39bc89fd | 6764 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6765 | sched_show_task(p); |
1da177e4 LT |
6766 | } while_each_thread(g, p); |
6767 | ||
04c9167f JF |
6768 | touch_all_softlockup_watchdogs(); |
6769 | ||
dd41f596 IM |
6770 | #ifdef CONFIG_SCHED_DEBUG |
6771 | sysrq_sched_debug_show(); | |
6772 | #endif | |
1da177e4 | 6773 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6774 | /* |
6775 | * Only show locks if all tasks are dumped: | |
6776 | */ | |
6777 | if (state_filter == -1) | |
6778 | debug_show_all_locks(); | |
1da177e4 LT |
6779 | } |
6780 | ||
1df21055 IM |
6781 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6782 | { | |
dd41f596 | 6783 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6784 | } |
6785 | ||
f340c0d1 IM |
6786 | /** |
6787 | * init_idle - set up an idle thread for a given CPU | |
6788 | * @idle: task in question | |
6789 | * @cpu: cpu the idle task belongs to | |
6790 | * | |
6791 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6792 | * flag, to make booting more robust. | |
6793 | */ | |
5c1e1767 | 6794 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6795 | { |
70b97a7f | 6796 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6797 | unsigned long flags; |
6798 | ||
5cbd54ef IM |
6799 | spin_lock_irqsave(&rq->lock, flags); |
6800 | ||
dd41f596 IM |
6801 | __sched_fork(idle); |
6802 | idle->se.exec_start = sched_clock(); | |
6803 | ||
b29739f9 | 6804 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6805 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6806 | __set_task_cpu(idle, cpu); |
1da177e4 | 6807 | |
1da177e4 | 6808 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6809 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6810 | idle->oncpu = 1; | |
6811 | #endif | |
1da177e4 LT |
6812 | spin_unlock_irqrestore(&rq->lock, flags); |
6813 | ||
6814 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6815 | #if defined(CONFIG_PREEMPT) |
6816 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6817 | #else | |
a1261f54 | 6818 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6819 | #endif |
dd41f596 IM |
6820 | /* |
6821 | * The idle tasks have their own, simple scheduling class: | |
6822 | */ | |
6823 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6824 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6825 | } |
6826 | ||
6827 | /* | |
6828 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6829 | * indicates which cpus entered this state. This is used | |
6830 | * in the rcu update to wait only for active cpus. For system | |
6831 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6832 | * always be CPU_BITS_NONE. |
1da177e4 | 6833 | */ |
6a7b3dc3 | 6834 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6835 | |
19978ca6 IM |
6836 | /* |
6837 | * Increase the granularity value when there are more CPUs, | |
6838 | * because with more CPUs the 'effective latency' as visible | |
6839 | * to users decreases. But the relationship is not linear, | |
6840 | * so pick a second-best guess by going with the log2 of the | |
6841 | * number of CPUs. | |
6842 | * | |
6843 | * This idea comes from the SD scheduler of Con Kolivas: | |
6844 | */ | |
6845 | static inline void sched_init_granularity(void) | |
6846 | { | |
6847 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6848 | const unsigned long limit = 200000000; | |
6849 | ||
6850 | sysctl_sched_min_granularity *= factor; | |
6851 | if (sysctl_sched_min_granularity > limit) | |
6852 | sysctl_sched_min_granularity = limit; | |
6853 | ||
6854 | sysctl_sched_latency *= factor; | |
6855 | if (sysctl_sched_latency > limit) | |
6856 | sysctl_sched_latency = limit; | |
6857 | ||
6858 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6859 | |
6860 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6861 | } |
6862 | ||
1da177e4 LT |
6863 | #ifdef CONFIG_SMP |
6864 | /* | |
6865 | * This is how migration works: | |
6866 | * | |
70b97a7f | 6867 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6868 | * runqueue and wake up that CPU's migration thread. |
6869 | * 2) we down() the locked semaphore => thread blocks. | |
6870 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6871 | * thread off the CPU) | |
6872 | * 4) it gets the migration request and checks whether the migrated | |
6873 | * task is still in the wrong runqueue. | |
6874 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6875 | * it and puts it into the right queue. | |
6876 | * 6) migration thread up()s the semaphore. | |
6877 | * 7) we wake up and the migration is done. | |
6878 | */ | |
6879 | ||
6880 | /* | |
6881 | * Change a given task's CPU affinity. Migrate the thread to a | |
6882 | * proper CPU and schedule it away if the CPU it's executing on | |
6883 | * is removed from the allowed bitmask. | |
6884 | * | |
6885 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6886 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6887 | * call is not atomic; no spinlocks may be held. |
6888 | */ | |
96f874e2 | 6889 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6890 | { |
70b97a7f | 6891 | struct migration_req req; |
1da177e4 | 6892 | unsigned long flags; |
70b97a7f | 6893 | struct rq *rq; |
48f24c4d | 6894 | int ret = 0; |
1da177e4 LT |
6895 | |
6896 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6897 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6898 | ret = -EINVAL; |
6899 | goto out; | |
6900 | } | |
6901 | ||
9985b0ba | 6902 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6903 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6904 | ret = -EINVAL; |
6905 | goto out; | |
6906 | } | |
6907 | ||
73fe6aae | 6908 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6909 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6910 | else { |
96f874e2 RR |
6911 | cpumask_copy(&p->cpus_allowed, new_mask); |
6912 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6913 | } |
6914 | ||
1da177e4 | 6915 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6916 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6917 | goto out; |
6918 | ||
1e5ce4f4 | 6919 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6920 | /* Need help from migration thread: drop lock and wait. */ |
6921 | task_rq_unlock(rq, &flags); | |
6922 | wake_up_process(rq->migration_thread); | |
6923 | wait_for_completion(&req.done); | |
6924 | tlb_migrate_finish(p->mm); | |
6925 | return 0; | |
6926 | } | |
6927 | out: | |
6928 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6929 | |
1da177e4 LT |
6930 | return ret; |
6931 | } | |
cd8ba7cd | 6932 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6933 | |
6934 | /* | |
41a2d6cf | 6935 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6936 | * this because either it can't run here any more (set_cpus_allowed() |
6937 | * away from this CPU, or CPU going down), or because we're | |
6938 | * attempting to rebalance this task on exec (sched_exec). | |
6939 | * | |
6940 | * So we race with normal scheduler movements, but that's OK, as long | |
6941 | * as the task is no longer on this CPU. | |
efc30814 KK |
6942 | * |
6943 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6944 | */ |
efc30814 | 6945 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6946 | { |
70b97a7f | 6947 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6948 | int ret = 0, on_rq; |
1da177e4 | 6949 | |
e761b772 | 6950 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6951 | return ret; |
1da177e4 LT |
6952 | |
6953 | rq_src = cpu_rq(src_cpu); | |
6954 | rq_dest = cpu_rq(dest_cpu); | |
6955 | ||
6956 | double_rq_lock(rq_src, rq_dest); | |
6957 | /* Already moved. */ | |
6958 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6959 | goto done; |
1da177e4 | 6960 | /* Affinity changed (again). */ |
96f874e2 | 6961 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6962 | goto fail; |
1da177e4 | 6963 | |
dd41f596 | 6964 | on_rq = p->se.on_rq; |
6e82a3be | 6965 | if (on_rq) |
2e1cb74a | 6966 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6967 | |
1da177e4 | 6968 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6969 | if (on_rq) { |
6970 | activate_task(rq_dest, p, 0); | |
15afe09b | 6971 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6972 | } |
b1e38734 | 6973 | done: |
efc30814 | 6974 | ret = 1; |
b1e38734 | 6975 | fail: |
1da177e4 | 6976 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6977 | return ret; |
1da177e4 LT |
6978 | } |
6979 | ||
6980 | /* | |
6981 | * migration_thread - this is a highprio system thread that performs | |
6982 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6983 | * another runqueue. | |
6984 | */ | |
95cdf3b7 | 6985 | static int migration_thread(void *data) |
1da177e4 | 6986 | { |
1da177e4 | 6987 | int cpu = (long)data; |
70b97a7f | 6988 | struct rq *rq; |
1da177e4 LT |
6989 | |
6990 | rq = cpu_rq(cpu); | |
6991 | BUG_ON(rq->migration_thread != current); | |
6992 | ||
6993 | set_current_state(TASK_INTERRUPTIBLE); | |
6994 | while (!kthread_should_stop()) { | |
70b97a7f | 6995 | struct migration_req *req; |
1da177e4 | 6996 | struct list_head *head; |
1da177e4 | 6997 | |
1da177e4 LT |
6998 | spin_lock_irq(&rq->lock); |
6999 | ||
7000 | if (cpu_is_offline(cpu)) { | |
7001 | spin_unlock_irq(&rq->lock); | |
7002 | goto wait_to_die; | |
7003 | } | |
7004 | ||
7005 | if (rq->active_balance) { | |
7006 | active_load_balance(rq, cpu); | |
7007 | rq->active_balance = 0; | |
7008 | } | |
7009 | ||
7010 | head = &rq->migration_queue; | |
7011 | ||
7012 | if (list_empty(head)) { | |
7013 | spin_unlock_irq(&rq->lock); | |
7014 | schedule(); | |
7015 | set_current_state(TASK_INTERRUPTIBLE); | |
7016 | continue; | |
7017 | } | |
70b97a7f | 7018 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7019 | list_del_init(head->next); |
7020 | ||
674311d5 NP |
7021 | spin_unlock(&rq->lock); |
7022 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7023 | local_irq_enable(); | |
1da177e4 LT |
7024 | |
7025 | complete(&req->done); | |
7026 | } | |
7027 | __set_current_state(TASK_RUNNING); | |
7028 | return 0; | |
7029 | ||
7030 | wait_to_die: | |
7031 | /* Wait for kthread_stop */ | |
7032 | set_current_state(TASK_INTERRUPTIBLE); | |
7033 | while (!kthread_should_stop()) { | |
7034 | schedule(); | |
7035 | set_current_state(TASK_INTERRUPTIBLE); | |
7036 | } | |
7037 | __set_current_state(TASK_RUNNING); | |
7038 | return 0; | |
7039 | } | |
7040 | ||
7041 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7042 | |
7043 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7044 | { | |
7045 | int ret; | |
7046 | ||
7047 | local_irq_disable(); | |
7048 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7049 | local_irq_enable(); | |
7050 | return ret; | |
7051 | } | |
7052 | ||
054b9108 | 7053 | /* |
3a4fa0a2 | 7054 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7055 | */ |
48f24c4d | 7056 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7057 | { |
70b97a7f | 7058 | int dest_cpu; |
6ca09dfc | 7059 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7060 | |
7061 | again: | |
7062 | /* Look for allowed, online CPU in same node. */ | |
7063 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7064 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7065 | goto move; | |
7066 | ||
7067 | /* Any allowed, online CPU? */ | |
7068 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7069 | if (dest_cpu < nr_cpu_ids) | |
7070 | goto move; | |
7071 | ||
7072 | /* No more Mr. Nice Guy. */ | |
7073 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7074 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7075 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7076 | |
e76bd8d9 RR |
7077 | /* |
7078 | * Don't tell them about moving exiting tasks or | |
7079 | * kernel threads (both mm NULL), since they never | |
7080 | * leave kernel. | |
7081 | */ | |
7082 | if (p->mm && printk_ratelimit()) { | |
7083 | printk(KERN_INFO "process %d (%s) no " | |
7084 | "longer affine to cpu%d\n", | |
7085 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7086 | } |
e76bd8d9 RR |
7087 | } |
7088 | ||
7089 | move: | |
7090 | /* It can have affinity changed while we were choosing. */ | |
7091 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7092 | goto again; | |
1da177e4 LT |
7093 | } |
7094 | ||
7095 | /* | |
7096 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7097 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7098 | * for performance reasons the counter is not stricly tracking tasks to | |
7099 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7100 | * to keep the global sum constant after CPU-down: | |
7101 | */ | |
70b97a7f | 7102 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7103 | { |
1e5ce4f4 | 7104 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7105 | unsigned long flags; |
7106 | ||
7107 | local_irq_save(flags); | |
7108 | double_rq_lock(rq_src, rq_dest); | |
7109 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7110 | rq_src->nr_uninterruptible = 0; | |
7111 | double_rq_unlock(rq_src, rq_dest); | |
7112 | local_irq_restore(flags); | |
7113 | } | |
7114 | ||
7115 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7116 | static void migrate_live_tasks(int src_cpu) | |
7117 | { | |
48f24c4d | 7118 | struct task_struct *p, *t; |
1da177e4 | 7119 | |
f7b4cddc | 7120 | read_lock(&tasklist_lock); |
1da177e4 | 7121 | |
48f24c4d IM |
7122 | do_each_thread(t, p) { |
7123 | if (p == current) | |
1da177e4 LT |
7124 | continue; |
7125 | ||
48f24c4d IM |
7126 | if (task_cpu(p) == src_cpu) |
7127 | move_task_off_dead_cpu(src_cpu, p); | |
7128 | } while_each_thread(t, p); | |
1da177e4 | 7129 | |
f7b4cddc | 7130 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7131 | } |
7132 | ||
dd41f596 IM |
7133 | /* |
7134 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7135 | * It does so by boosting its priority to highest possible. |
7136 | * Used by CPU offline code. | |
1da177e4 LT |
7137 | */ |
7138 | void sched_idle_next(void) | |
7139 | { | |
48f24c4d | 7140 | int this_cpu = smp_processor_id(); |
70b97a7f | 7141 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7142 | struct task_struct *p = rq->idle; |
7143 | unsigned long flags; | |
7144 | ||
7145 | /* cpu has to be offline */ | |
48f24c4d | 7146 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7147 | |
48f24c4d IM |
7148 | /* |
7149 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7150 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7151 | */ |
7152 | spin_lock_irqsave(&rq->lock, flags); | |
7153 | ||
dd41f596 | 7154 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7155 | |
94bc9a7b DA |
7156 | update_rq_clock(rq); |
7157 | activate_task(rq, p, 0); | |
1da177e4 LT |
7158 | |
7159 | spin_unlock_irqrestore(&rq->lock, flags); | |
7160 | } | |
7161 | ||
48f24c4d IM |
7162 | /* |
7163 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7164 | * offline. |
7165 | */ | |
7166 | void idle_task_exit(void) | |
7167 | { | |
7168 | struct mm_struct *mm = current->active_mm; | |
7169 | ||
7170 | BUG_ON(cpu_online(smp_processor_id())); | |
7171 | ||
7172 | if (mm != &init_mm) | |
7173 | switch_mm(mm, &init_mm, current); | |
7174 | mmdrop(mm); | |
7175 | } | |
7176 | ||
054b9108 | 7177 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7178 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7179 | { |
70b97a7f | 7180 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7181 | |
7182 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7183 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7184 | |
7185 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7186 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7187 | |
48f24c4d | 7188 | get_task_struct(p); |
1da177e4 LT |
7189 | |
7190 | /* | |
7191 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7192 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7193 | * fine. |
7194 | */ | |
f7b4cddc | 7195 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7196 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7197 | spin_lock_irq(&rq->lock); |
1da177e4 | 7198 | |
48f24c4d | 7199 | put_task_struct(p); |
1da177e4 LT |
7200 | } |
7201 | ||
7202 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7203 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7204 | { | |
70b97a7f | 7205 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7206 | struct task_struct *next; |
48f24c4d | 7207 | |
dd41f596 IM |
7208 | for ( ; ; ) { |
7209 | if (!rq->nr_running) | |
7210 | break; | |
a8e504d2 | 7211 | update_rq_clock(rq); |
b67802ea | 7212 | next = pick_next_task(rq); |
dd41f596 IM |
7213 | if (!next) |
7214 | break; | |
79c53799 | 7215 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7216 | migrate_dead(dead_cpu, next); |
e692ab53 | 7217 | |
1da177e4 LT |
7218 | } |
7219 | } | |
dce48a84 TG |
7220 | |
7221 | /* | |
7222 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7223 | */ | |
7224 | static void calc_global_load_remove(struct rq *rq) | |
7225 | { | |
7226 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
7227 | } | |
1da177e4 LT |
7228 | #endif /* CONFIG_HOTPLUG_CPU */ |
7229 | ||
e692ab53 NP |
7230 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7231 | ||
7232 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7233 | { |
7234 | .procname = "sched_domain", | |
c57baf1e | 7235 | .mode = 0555, |
e0361851 | 7236 | }, |
38605cae | 7237 | {0, }, |
e692ab53 NP |
7238 | }; |
7239 | ||
7240 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7241 | { |
c57baf1e | 7242 | .ctl_name = CTL_KERN, |
e0361851 | 7243 | .procname = "kernel", |
c57baf1e | 7244 | .mode = 0555, |
e0361851 AD |
7245 | .child = sd_ctl_dir, |
7246 | }, | |
38605cae | 7247 | {0, }, |
e692ab53 NP |
7248 | }; |
7249 | ||
7250 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7251 | { | |
7252 | struct ctl_table *entry = | |
5cf9f062 | 7253 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7254 | |
e692ab53 NP |
7255 | return entry; |
7256 | } | |
7257 | ||
6382bc90 MM |
7258 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7259 | { | |
cd790076 | 7260 | struct ctl_table *entry; |
6382bc90 | 7261 | |
cd790076 MM |
7262 | /* |
7263 | * In the intermediate directories, both the child directory and | |
7264 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7265 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7266 | * static strings and all have proc handlers. |
7267 | */ | |
7268 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7269 | if (entry->child) |
7270 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7271 | if (entry->proc_handler == NULL) |
7272 | kfree(entry->procname); | |
7273 | } | |
6382bc90 MM |
7274 | |
7275 | kfree(*tablep); | |
7276 | *tablep = NULL; | |
7277 | } | |
7278 | ||
e692ab53 | 7279 | static void |
e0361851 | 7280 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7281 | const char *procname, void *data, int maxlen, |
7282 | mode_t mode, proc_handler *proc_handler) | |
7283 | { | |
e692ab53 NP |
7284 | entry->procname = procname; |
7285 | entry->data = data; | |
7286 | entry->maxlen = maxlen; | |
7287 | entry->mode = mode; | |
7288 | entry->proc_handler = proc_handler; | |
7289 | } | |
7290 | ||
7291 | static struct ctl_table * | |
7292 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7293 | { | |
a5d8c348 | 7294 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7295 | |
ad1cdc1d MM |
7296 | if (table == NULL) |
7297 | return NULL; | |
7298 | ||
e0361851 | 7299 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7300 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7301 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7302 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7303 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7304 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7305 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7306 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7307 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7308 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7309 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7310 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7311 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7312 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7313 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7314 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7315 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7316 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7317 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7318 | &sd->cache_nice_tries, |
7319 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7320 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7321 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7322 | set_table_entry(&table[11], "name", sd->name, |
7323 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7324 | /* &table[12] is terminator */ | |
e692ab53 NP |
7325 | |
7326 | return table; | |
7327 | } | |
7328 | ||
9a4e7159 | 7329 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7330 | { |
7331 | struct ctl_table *entry, *table; | |
7332 | struct sched_domain *sd; | |
7333 | int domain_num = 0, i; | |
7334 | char buf[32]; | |
7335 | ||
7336 | for_each_domain(cpu, sd) | |
7337 | domain_num++; | |
7338 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7339 | if (table == NULL) |
7340 | return NULL; | |
e692ab53 NP |
7341 | |
7342 | i = 0; | |
7343 | for_each_domain(cpu, sd) { | |
7344 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7345 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7346 | entry->mode = 0555; |
e692ab53 NP |
7347 | entry->child = sd_alloc_ctl_domain_table(sd); |
7348 | entry++; | |
7349 | i++; | |
7350 | } | |
7351 | return table; | |
7352 | } | |
7353 | ||
7354 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7355 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7356 | { |
7357 | int i, cpu_num = num_online_cpus(); | |
7358 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7359 | char buf[32]; | |
7360 | ||
7378547f MM |
7361 | WARN_ON(sd_ctl_dir[0].child); |
7362 | sd_ctl_dir[0].child = entry; | |
7363 | ||
ad1cdc1d MM |
7364 | if (entry == NULL) |
7365 | return; | |
7366 | ||
97b6ea7b | 7367 | for_each_online_cpu(i) { |
e692ab53 | 7368 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7369 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7370 | entry->mode = 0555; |
e692ab53 | 7371 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7372 | entry++; |
e692ab53 | 7373 | } |
7378547f MM |
7374 | |
7375 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7376 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7377 | } | |
6382bc90 | 7378 | |
7378547f | 7379 | /* may be called multiple times per register */ |
6382bc90 MM |
7380 | static void unregister_sched_domain_sysctl(void) |
7381 | { | |
7378547f MM |
7382 | if (sd_sysctl_header) |
7383 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7384 | sd_sysctl_header = NULL; |
7378547f MM |
7385 | if (sd_ctl_dir[0].child) |
7386 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7387 | } |
e692ab53 | 7388 | #else |
6382bc90 MM |
7389 | static void register_sched_domain_sysctl(void) |
7390 | { | |
7391 | } | |
7392 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7393 | { |
7394 | } | |
7395 | #endif | |
7396 | ||
1f11eb6a GH |
7397 | static void set_rq_online(struct rq *rq) |
7398 | { | |
7399 | if (!rq->online) { | |
7400 | const struct sched_class *class; | |
7401 | ||
c6c4927b | 7402 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7403 | rq->online = 1; |
7404 | ||
7405 | for_each_class(class) { | |
7406 | if (class->rq_online) | |
7407 | class->rq_online(rq); | |
7408 | } | |
7409 | } | |
7410 | } | |
7411 | ||
7412 | static void set_rq_offline(struct rq *rq) | |
7413 | { | |
7414 | if (rq->online) { | |
7415 | const struct sched_class *class; | |
7416 | ||
7417 | for_each_class(class) { | |
7418 | if (class->rq_offline) | |
7419 | class->rq_offline(rq); | |
7420 | } | |
7421 | ||
c6c4927b | 7422 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7423 | rq->online = 0; |
7424 | } | |
7425 | } | |
7426 | ||
1da177e4 LT |
7427 | /* |
7428 | * migration_call - callback that gets triggered when a CPU is added. | |
7429 | * Here we can start up the necessary migration thread for the new CPU. | |
7430 | */ | |
48f24c4d IM |
7431 | static int __cpuinit |
7432 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7433 | { |
1da177e4 | 7434 | struct task_struct *p; |
48f24c4d | 7435 | int cpu = (long)hcpu; |
1da177e4 | 7436 | unsigned long flags; |
70b97a7f | 7437 | struct rq *rq; |
1da177e4 LT |
7438 | |
7439 | switch (action) { | |
5be9361c | 7440 | |
1da177e4 | 7441 | case CPU_UP_PREPARE: |
8bb78442 | 7442 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7443 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7444 | if (IS_ERR(p)) |
7445 | return NOTIFY_BAD; | |
1da177e4 LT |
7446 | kthread_bind(p, cpu); |
7447 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7448 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7449 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
7450 | task_rq_unlock(rq, &flags); |
7451 | cpu_rq(cpu)->migration_thread = p; | |
7452 | break; | |
48f24c4d | 7453 | |
1da177e4 | 7454 | case CPU_ONLINE: |
8bb78442 | 7455 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7456 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7457 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7458 | |
7459 | /* Update our root-domain */ | |
7460 | rq = cpu_rq(cpu); | |
7461 | spin_lock_irqsave(&rq->lock, flags); | |
dce48a84 TG |
7462 | rq->calc_load_update = calc_load_update; |
7463 | rq->calc_load_active = 0; | |
1f94ef59 | 7464 | if (rq->rd) { |
c6c4927b | 7465 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7466 | |
7467 | set_rq_online(rq); | |
1f94ef59 GH |
7468 | } |
7469 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7470 | break; |
48f24c4d | 7471 | |
1da177e4 LT |
7472 | #ifdef CONFIG_HOTPLUG_CPU |
7473 | case CPU_UP_CANCELED: | |
8bb78442 | 7474 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7475 | if (!cpu_rq(cpu)->migration_thread) |
7476 | break; | |
41a2d6cf | 7477 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7478 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7479 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7480 | kthread_stop(cpu_rq(cpu)->migration_thread); |
7481 | cpu_rq(cpu)->migration_thread = NULL; | |
7482 | break; | |
48f24c4d | 7483 | |
1da177e4 | 7484 | case CPU_DEAD: |
8bb78442 | 7485 | case CPU_DEAD_FROZEN: |
470fd646 | 7486 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7487 | migrate_live_tasks(cpu); |
7488 | rq = cpu_rq(cpu); | |
7489 | kthread_stop(rq->migration_thread); | |
7490 | rq->migration_thread = NULL; | |
7491 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7492 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7493 | update_rq_clock(rq); |
2e1cb74a | 7494 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7495 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7496 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7497 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7498 | migrate_dead_tasks(cpu); |
d2da272a | 7499 | spin_unlock_irq(&rq->lock); |
470fd646 | 7500 | cpuset_unlock(); |
1da177e4 LT |
7501 | migrate_nr_uninterruptible(rq); |
7502 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7503 | calc_global_load_remove(rq); |
41a2d6cf IM |
7504 | /* |
7505 | * No need to migrate the tasks: it was best-effort if | |
7506 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7507 | * the requestors. | |
7508 | */ | |
1da177e4 LT |
7509 | spin_lock_irq(&rq->lock); |
7510 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7511 | struct migration_req *req; |
7512 | ||
1da177e4 | 7513 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7514 | struct migration_req, list); |
1da177e4 | 7515 | list_del_init(&req->list); |
9a2bd244 | 7516 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7517 | complete(&req->done); |
9a2bd244 | 7518 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7519 | } |
7520 | spin_unlock_irq(&rq->lock); | |
7521 | break; | |
57d885fe | 7522 | |
08f503b0 GH |
7523 | case CPU_DYING: |
7524 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7525 | /* Update our root-domain */ |
7526 | rq = cpu_rq(cpu); | |
7527 | spin_lock_irqsave(&rq->lock, flags); | |
7528 | if (rq->rd) { | |
c6c4927b | 7529 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7530 | set_rq_offline(rq); |
57d885fe GH |
7531 | } |
7532 | spin_unlock_irqrestore(&rq->lock, flags); | |
7533 | break; | |
1da177e4 LT |
7534 | #endif |
7535 | } | |
7536 | return NOTIFY_OK; | |
7537 | } | |
7538 | ||
7539 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
7540 | * happens before everything else. | |
7541 | */ | |
26c2143b | 7542 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7543 | .notifier_call = migration_call, |
7544 | .priority = 10 | |
7545 | }; | |
7546 | ||
7babe8db | 7547 | static int __init migration_init(void) |
1da177e4 LT |
7548 | { |
7549 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7550 | int err; |
48f24c4d IM |
7551 | |
7552 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7553 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7554 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7555 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7556 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7557 | |
7558 | return err; | |
1da177e4 | 7559 | } |
7babe8db | 7560 | early_initcall(migration_init); |
1da177e4 LT |
7561 | #endif |
7562 | ||
7563 | #ifdef CONFIG_SMP | |
476f3534 | 7564 | |
3e9830dc | 7565 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7566 | |
7c16ec58 | 7567 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7568 | struct cpumask *groupmask) |
1da177e4 | 7569 | { |
4dcf6aff | 7570 | struct sched_group *group = sd->groups; |
434d53b0 | 7571 | char str[256]; |
1da177e4 | 7572 | |
968ea6d8 | 7573 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7574 | cpumask_clear(groupmask); |
4dcf6aff IM |
7575 | |
7576 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7577 | ||
7578 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7579 | printk("does not load-balance\n"); | |
7580 | if (sd->parent) | |
7581 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7582 | " has parent"); | |
7583 | return -1; | |
41c7ce9a NP |
7584 | } |
7585 | ||
eefd796a | 7586 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7587 | |
758b2cdc | 7588 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7589 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7590 | "CPU%d\n", cpu); | |
7591 | } | |
758b2cdc | 7592 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7593 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7594 | " CPU%d\n", cpu); | |
7595 | } | |
1da177e4 | 7596 | |
4dcf6aff | 7597 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7598 | do { |
4dcf6aff IM |
7599 | if (!group) { |
7600 | printk("\n"); | |
7601 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7602 | break; |
7603 | } | |
7604 | ||
4dcf6aff IM |
7605 | if (!group->__cpu_power) { |
7606 | printk(KERN_CONT "\n"); | |
7607 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7608 | "set\n"); | |
7609 | break; | |
7610 | } | |
1da177e4 | 7611 | |
758b2cdc | 7612 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7613 | printk(KERN_CONT "\n"); |
7614 | printk(KERN_ERR "ERROR: empty group\n"); | |
7615 | break; | |
7616 | } | |
1da177e4 | 7617 | |
758b2cdc | 7618 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7619 | printk(KERN_CONT "\n"); |
7620 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7621 | break; | |
7622 | } | |
1da177e4 | 7623 | |
758b2cdc | 7624 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7625 | |
968ea6d8 | 7626 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7627 | |
7628 | printk(KERN_CONT " %s", str); | |
7629 | if (group->__cpu_power != SCHED_LOAD_SCALE) { | |
7630 | printk(KERN_CONT " (__cpu_power = %d)", | |
7631 | group->__cpu_power); | |
7632 | } | |
1da177e4 | 7633 | |
4dcf6aff IM |
7634 | group = group->next; |
7635 | } while (group != sd->groups); | |
7636 | printk(KERN_CONT "\n"); | |
1da177e4 | 7637 | |
758b2cdc | 7638 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7639 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7640 | |
758b2cdc RR |
7641 | if (sd->parent && |
7642 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7643 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7644 | "of domain->span\n"); | |
7645 | return 0; | |
7646 | } | |
1da177e4 | 7647 | |
4dcf6aff IM |
7648 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7649 | { | |
d5dd3db1 | 7650 | cpumask_var_t groupmask; |
4dcf6aff | 7651 | int level = 0; |
1da177e4 | 7652 | |
4dcf6aff IM |
7653 | if (!sd) { |
7654 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7655 | return; | |
7656 | } | |
1da177e4 | 7657 | |
4dcf6aff IM |
7658 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7659 | ||
d5dd3db1 | 7660 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7661 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7662 | return; | |
7663 | } | |
7664 | ||
4dcf6aff | 7665 | for (;;) { |
7c16ec58 | 7666 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7667 | break; |
1da177e4 LT |
7668 | level++; |
7669 | sd = sd->parent; | |
33859f7f | 7670 | if (!sd) |
4dcf6aff IM |
7671 | break; |
7672 | } | |
d5dd3db1 | 7673 | free_cpumask_var(groupmask); |
1da177e4 | 7674 | } |
6d6bc0ad | 7675 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7676 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7677 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7678 | |
1a20ff27 | 7679 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7680 | { |
758b2cdc | 7681 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7682 | return 1; |
7683 | ||
7684 | /* Following flags need at least 2 groups */ | |
7685 | if (sd->flags & (SD_LOAD_BALANCE | | |
7686 | SD_BALANCE_NEWIDLE | | |
7687 | SD_BALANCE_FORK | | |
89c4710e SS |
7688 | SD_BALANCE_EXEC | |
7689 | SD_SHARE_CPUPOWER | | |
7690 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7691 | if (sd->groups != sd->groups->next) |
7692 | return 0; | |
7693 | } | |
7694 | ||
7695 | /* Following flags don't use groups */ | |
7696 | if (sd->flags & (SD_WAKE_IDLE | | |
7697 | SD_WAKE_AFFINE | | |
7698 | SD_WAKE_BALANCE)) | |
7699 | return 0; | |
7700 | ||
7701 | return 1; | |
7702 | } | |
7703 | ||
48f24c4d IM |
7704 | static int |
7705 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7706 | { |
7707 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7708 | ||
7709 | if (sd_degenerate(parent)) | |
7710 | return 1; | |
7711 | ||
758b2cdc | 7712 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7713 | return 0; |
7714 | ||
7715 | /* Does parent contain flags not in child? */ | |
7716 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7717 | if (cflags & SD_WAKE_AFFINE) | |
7718 | pflags &= ~SD_WAKE_BALANCE; | |
7719 | /* Flags needing groups don't count if only 1 group in parent */ | |
7720 | if (parent->groups == parent->groups->next) { | |
7721 | pflags &= ~(SD_LOAD_BALANCE | | |
7722 | SD_BALANCE_NEWIDLE | | |
7723 | SD_BALANCE_FORK | | |
89c4710e SS |
7724 | SD_BALANCE_EXEC | |
7725 | SD_SHARE_CPUPOWER | | |
7726 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7727 | if (nr_node_ids == 1) |
7728 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7729 | } |
7730 | if (~cflags & pflags) | |
7731 | return 0; | |
7732 | ||
7733 | return 1; | |
7734 | } | |
7735 | ||
c6c4927b RR |
7736 | static void free_rootdomain(struct root_domain *rd) |
7737 | { | |
68e74568 RR |
7738 | cpupri_cleanup(&rd->cpupri); |
7739 | ||
c6c4927b RR |
7740 | free_cpumask_var(rd->rto_mask); |
7741 | free_cpumask_var(rd->online); | |
7742 | free_cpumask_var(rd->span); | |
7743 | kfree(rd); | |
7744 | } | |
7745 | ||
57d885fe GH |
7746 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7747 | { | |
a0490fa3 | 7748 | struct root_domain *old_rd = NULL; |
57d885fe | 7749 | unsigned long flags; |
57d885fe GH |
7750 | |
7751 | spin_lock_irqsave(&rq->lock, flags); | |
7752 | ||
7753 | if (rq->rd) { | |
a0490fa3 | 7754 | old_rd = rq->rd; |
57d885fe | 7755 | |
c6c4927b | 7756 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7757 | set_rq_offline(rq); |
57d885fe | 7758 | |
c6c4927b | 7759 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7760 | |
a0490fa3 IM |
7761 | /* |
7762 | * If we dont want to free the old_rt yet then | |
7763 | * set old_rd to NULL to skip the freeing later | |
7764 | * in this function: | |
7765 | */ | |
7766 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7767 | old_rd = NULL; | |
57d885fe GH |
7768 | } |
7769 | ||
7770 | atomic_inc(&rd->refcount); | |
7771 | rq->rd = rd; | |
7772 | ||
c6c4927b RR |
7773 | cpumask_set_cpu(rq->cpu, rd->span); |
7774 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7775 | set_rq_online(rq); |
57d885fe GH |
7776 | |
7777 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7778 | |
7779 | if (old_rd) | |
7780 | free_rootdomain(old_rd); | |
57d885fe GH |
7781 | } |
7782 | ||
db2f59c8 | 7783 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe GH |
7784 | { |
7785 | memset(rd, 0, sizeof(*rd)); | |
7786 | ||
c6c4927b RR |
7787 | if (bootmem) { |
7788 | alloc_bootmem_cpumask_var(&def_root_domain.span); | |
7789 | alloc_bootmem_cpumask_var(&def_root_domain.online); | |
7790 | alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); | |
68e74568 | 7791 | cpupri_init(&rd->cpupri, true); |
c6c4927b RR |
7792 | return 0; |
7793 | } | |
7794 | ||
7795 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | |
0c910d28 | 7796 | goto out; |
c6c4927b RR |
7797 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
7798 | goto free_span; | |
7799 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | |
7800 | goto free_online; | |
6e0534f2 | 7801 | |
68e74568 RR |
7802 | if (cpupri_init(&rd->cpupri, false) != 0) |
7803 | goto free_rto_mask; | |
c6c4927b | 7804 | return 0; |
6e0534f2 | 7805 | |
68e74568 RR |
7806 | free_rto_mask: |
7807 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7808 | free_online: |
7809 | free_cpumask_var(rd->online); | |
7810 | free_span: | |
7811 | free_cpumask_var(rd->span); | |
0c910d28 | 7812 | out: |
c6c4927b | 7813 | return -ENOMEM; |
57d885fe GH |
7814 | } |
7815 | ||
7816 | static void init_defrootdomain(void) | |
7817 | { | |
c6c4927b RR |
7818 | init_rootdomain(&def_root_domain, true); |
7819 | ||
57d885fe GH |
7820 | atomic_set(&def_root_domain.refcount, 1); |
7821 | } | |
7822 | ||
dc938520 | 7823 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7824 | { |
7825 | struct root_domain *rd; | |
7826 | ||
7827 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7828 | if (!rd) | |
7829 | return NULL; | |
7830 | ||
c6c4927b RR |
7831 | if (init_rootdomain(rd, false) != 0) { |
7832 | kfree(rd); | |
7833 | return NULL; | |
7834 | } | |
57d885fe GH |
7835 | |
7836 | return rd; | |
7837 | } | |
7838 | ||
1da177e4 | 7839 | /* |
0eab9146 | 7840 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7841 | * hold the hotplug lock. |
7842 | */ | |
0eab9146 IM |
7843 | static void |
7844 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7845 | { |
70b97a7f | 7846 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7847 | struct sched_domain *tmp; |
7848 | ||
7849 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7850 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7851 | struct sched_domain *parent = tmp->parent; |
7852 | if (!parent) | |
7853 | break; | |
f29c9b1c | 7854 | |
1a848870 | 7855 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7856 | tmp->parent = parent->parent; |
1a848870 SS |
7857 | if (parent->parent) |
7858 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7859 | } else |
7860 | tmp = tmp->parent; | |
245af2c7 SS |
7861 | } |
7862 | ||
1a848870 | 7863 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7864 | sd = sd->parent; |
1a848870 SS |
7865 | if (sd) |
7866 | sd->child = NULL; | |
7867 | } | |
1da177e4 LT |
7868 | |
7869 | sched_domain_debug(sd, cpu); | |
7870 | ||
57d885fe | 7871 | rq_attach_root(rq, rd); |
674311d5 | 7872 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7873 | } |
7874 | ||
7875 | /* cpus with isolated domains */ | |
dcc30a35 | 7876 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7877 | |
7878 | /* Setup the mask of cpus configured for isolated domains */ | |
7879 | static int __init isolated_cpu_setup(char *str) | |
7880 | { | |
968ea6d8 | 7881 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7882 | return 1; |
7883 | } | |
7884 | ||
8927f494 | 7885 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7886 | |
7887 | /* | |
6711cab4 SS |
7888 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7889 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7890 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7891 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7892 | * |
7893 | * init_sched_build_groups will build a circular linked list of the groups | |
7894 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7895 | * and ->cpu_power to 0. | |
7896 | */ | |
a616058b | 7897 | static void |
96f874e2 RR |
7898 | init_sched_build_groups(const struct cpumask *span, |
7899 | const struct cpumask *cpu_map, | |
7900 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7901 | struct sched_group **sg, |
96f874e2 RR |
7902 | struct cpumask *tmpmask), |
7903 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7904 | { |
7905 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7906 | int i; |
7907 | ||
96f874e2 | 7908 | cpumask_clear(covered); |
7c16ec58 | 7909 | |
abcd083a | 7910 | for_each_cpu(i, span) { |
6711cab4 | 7911 | struct sched_group *sg; |
7c16ec58 | 7912 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7913 | int j; |
7914 | ||
758b2cdc | 7915 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7916 | continue; |
7917 | ||
758b2cdc | 7918 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7919 | sg->__cpu_power = 0; |
1da177e4 | 7920 | |
abcd083a | 7921 | for_each_cpu(j, span) { |
7c16ec58 | 7922 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7923 | continue; |
7924 | ||
96f874e2 | 7925 | cpumask_set_cpu(j, covered); |
758b2cdc | 7926 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7927 | } |
7928 | if (!first) | |
7929 | first = sg; | |
7930 | if (last) | |
7931 | last->next = sg; | |
7932 | last = sg; | |
7933 | } | |
7934 | last->next = first; | |
7935 | } | |
7936 | ||
9c1cfda2 | 7937 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7938 | |
9c1cfda2 | 7939 | #ifdef CONFIG_NUMA |
198e2f18 | 7940 | |
9c1cfda2 JH |
7941 | /** |
7942 | * find_next_best_node - find the next node to include in a sched_domain | |
7943 | * @node: node whose sched_domain we're building | |
7944 | * @used_nodes: nodes already in the sched_domain | |
7945 | * | |
41a2d6cf | 7946 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7947 | * finds the closest node not already in the @used_nodes map. |
7948 | * | |
7949 | * Should use nodemask_t. | |
7950 | */ | |
c5f59f08 | 7951 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7952 | { |
7953 | int i, n, val, min_val, best_node = 0; | |
7954 | ||
7955 | min_val = INT_MAX; | |
7956 | ||
076ac2af | 7957 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7958 | /* Start at @node */ |
076ac2af | 7959 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7960 | |
7961 | if (!nr_cpus_node(n)) | |
7962 | continue; | |
7963 | ||
7964 | /* Skip already used nodes */ | |
c5f59f08 | 7965 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7966 | continue; |
7967 | ||
7968 | /* Simple min distance search */ | |
7969 | val = node_distance(node, n); | |
7970 | ||
7971 | if (val < min_val) { | |
7972 | min_val = val; | |
7973 | best_node = n; | |
7974 | } | |
7975 | } | |
7976 | ||
c5f59f08 | 7977 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7978 | return best_node; |
7979 | } | |
7980 | ||
7981 | /** | |
7982 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7983 | * @node: node whose cpumask we're constructing | |
73486722 | 7984 | * @span: resulting cpumask |
9c1cfda2 | 7985 | * |
41a2d6cf | 7986 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7987 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7988 | * out optimally. | |
7989 | */ | |
96f874e2 | 7990 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7991 | { |
c5f59f08 | 7992 | nodemask_t used_nodes; |
48f24c4d | 7993 | int i; |
9c1cfda2 | 7994 | |
6ca09dfc | 7995 | cpumask_clear(span); |
c5f59f08 | 7996 | nodes_clear(used_nodes); |
9c1cfda2 | 7997 | |
6ca09dfc | 7998 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7999 | node_set(node, used_nodes); |
9c1cfda2 JH |
8000 | |
8001 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8002 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8003 | |
6ca09dfc | 8004 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8005 | } |
9c1cfda2 | 8006 | } |
6d6bc0ad | 8007 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8008 | |
5c45bf27 | 8009 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8010 | |
6c99e9ad RR |
8011 | /* |
8012 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8013 | * |
8014 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8015 | * and struct sched_domain. ) | |
6c99e9ad RR |
8016 | */ |
8017 | struct static_sched_group { | |
8018 | struct sched_group sg; | |
8019 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8020 | }; | |
8021 | ||
8022 | struct static_sched_domain { | |
8023 | struct sched_domain sd; | |
8024 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8025 | }; | |
8026 | ||
9c1cfda2 | 8027 | /* |
48f24c4d | 8028 | * SMT sched-domains: |
9c1cfda2 | 8029 | */ |
1da177e4 | 8030 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8031 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8032 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8033 | |
41a2d6cf | 8034 | static int |
96f874e2 RR |
8035 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8036 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8037 | { |
6711cab4 | 8038 | if (sg) |
6c99e9ad | 8039 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8040 | return cpu; |
8041 | } | |
6d6bc0ad | 8042 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8043 | |
48f24c4d IM |
8044 | /* |
8045 | * multi-core sched-domains: | |
8046 | */ | |
1e9f28fa | 8047 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8048 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8049 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8050 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8051 | |
8052 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8053 | static int |
96f874e2 RR |
8054 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8055 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8056 | { |
6711cab4 | 8057 | int group; |
7c16ec58 | 8058 | |
c69fc56d | 8059 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8060 | group = cpumask_first(mask); |
6711cab4 | 8061 | if (sg) |
6c99e9ad | 8062 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8063 | return group; |
1e9f28fa SS |
8064 | } |
8065 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8066 | static int |
96f874e2 RR |
8067 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8068 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8069 | { |
6711cab4 | 8070 | if (sg) |
6c99e9ad | 8071 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8072 | return cpu; |
8073 | } | |
8074 | #endif | |
8075 | ||
6c99e9ad RR |
8076 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8077 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8078 | |
41a2d6cf | 8079 | static int |
96f874e2 RR |
8080 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8081 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8082 | { |
6711cab4 | 8083 | int group; |
48f24c4d | 8084 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8085 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8086 | group = cpumask_first(mask); |
1e9f28fa | 8087 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8088 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8089 | group = cpumask_first(mask); |
1da177e4 | 8090 | #else |
6711cab4 | 8091 | group = cpu; |
1da177e4 | 8092 | #endif |
6711cab4 | 8093 | if (sg) |
6c99e9ad | 8094 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8095 | return group; |
1da177e4 LT |
8096 | } |
8097 | ||
8098 | #ifdef CONFIG_NUMA | |
1da177e4 | 8099 | /* |
9c1cfda2 JH |
8100 | * The init_sched_build_groups can't handle what we want to do with node |
8101 | * groups, so roll our own. Now each node has its own list of groups which | |
8102 | * gets dynamically allocated. | |
1da177e4 | 8103 | */ |
62ea9ceb | 8104 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8105 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8106 | |
62ea9ceb | 8107 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8108 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8109 | |
96f874e2 RR |
8110 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8111 | struct sched_group **sg, | |
8112 | struct cpumask *nodemask) | |
9c1cfda2 | 8113 | { |
6711cab4 SS |
8114 | int group; |
8115 | ||
6ca09dfc | 8116 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8117 | group = cpumask_first(nodemask); |
6711cab4 SS |
8118 | |
8119 | if (sg) | |
6c99e9ad | 8120 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8121 | return group; |
1da177e4 | 8122 | } |
6711cab4 | 8123 | |
08069033 SS |
8124 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8125 | { | |
8126 | struct sched_group *sg = group_head; | |
8127 | int j; | |
8128 | ||
8129 | if (!sg) | |
8130 | return; | |
3a5c359a | 8131 | do { |
758b2cdc | 8132 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8133 | struct sched_domain *sd; |
08069033 | 8134 | |
6c99e9ad | 8135 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8136 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8137 | /* |
8138 | * Only add "power" once for each | |
8139 | * physical package. | |
8140 | */ | |
8141 | continue; | |
8142 | } | |
08069033 | 8143 | |
3a5c359a AK |
8144 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
8145 | } | |
8146 | sg = sg->next; | |
8147 | } while (sg != group_head); | |
08069033 | 8148 | } |
6d6bc0ad | 8149 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8150 | |
a616058b | 8151 | #ifdef CONFIG_NUMA |
51888ca2 | 8152 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8153 | static void free_sched_groups(const struct cpumask *cpu_map, |
8154 | struct cpumask *nodemask) | |
51888ca2 | 8155 | { |
a616058b | 8156 | int cpu, i; |
51888ca2 | 8157 | |
abcd083a | 8158 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8159 | struct sched_group **sched_group_nodes |
8160 | = sched_group_nodes_bycpu[cpu]; | |
8161 | ||
51888ca2 SV |
8162 | if (!sched_group_nodes) |
8163 | continue; | |
8164 | ||
076ac2af | 8165 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8166 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8167 | ||
6ca09dfc | 8168 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8169 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8170 | continue; |
8171 | ||
8172 | if (sg == NULL) | |
8173 | continue; | |
8174 | sg = sg->next; | |
8175 | next_sg: | |
8176 | oldsg = sg; | |
8177 | sg = sg->next; | |
8178 | kfree(oldsg); | |
8179 | if (oldsg != sched_group_nodes[i]) | |
8180 | goto next_sg; | |
8181 | } | |
8182 | kfree(sched_group_nodes); | |
8183 | sched_group_nodes_bycpu[cpu] = NULL; | |
8184 | } | |
51888ca2 | 8185 | } |
6d6bc0ad | 8186 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8187 | static void free_sched_groups(const struct cpumask *cpu_map, |
8188 | struct cpumask *nodemask) | |
a616058b SS |
8189 | { |
8190 | } | |
6d6bc0ad | 8191 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8192 | |
89c4710e SS |
8193 | /* |
8194 | * Initialize sched groups cpu_power. | |
8195 | * | |
8196 | * cpu_power indicates the capacity of sched group, which is used while | |
8197 | * distributing the load between different sched groups in a sched domain. | |
8198 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8199 | * there are asymmetries in the topology. If there are asymmetries, group | |
8200 | * having more cpu_power will pickup more load compared to the group having | |
8201 | * less cpu_power. | |
8202 | * | |
8203 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
8204 | * the maximum number of tasks a group can handle in the presence of other idle | |
8205 | * or lightly loaded groups in the same sched domain. | |
8206 | */ | |
8207 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8208 | { | |
8209 | struct sched_domain *child; | |
8210 | struct sched_group *group; | |
8211 | ||
8212 | WARN_ON(!sd || !sd->groups); | |
8213 | ||
13318a71 | 8214 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8215 | return; |
8216 | ||
8217 | child = sd->child; | |
8218 | ||
5517d86b ED |
8219 | sd->groups->__cpu_power = 0; |
8220 | ||
89c4710e SS |
8221 | /* |
8222 | * For perf policy, if the groups in child domain share resources | |
8223 | * (for example cores sharing some portions of the cache hierarchy | |
8224 | * or SMT), then set this domain groups cpu_power such that each group | |
8225 | * can handle only one task, when there are other idle groups in the | |
8226 | * same sched domain. | |
8227 | */ | |
8228 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
8229 | (child->flags & | |
8230 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 8231 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
8232 | return; |
8233 | } | |
8234 | ||
89c4710e SS |
8235 | /* |
8236 | * add cpu_power of each child group to this groups cpu_power | |
8237 | */ | |
8238 | group = child->groups; | |
8239 | do { | |
5517d86b | 8240 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8241 | group = group->next; |
8242 | } while (group != child->groups); | |
8243 | } | |
8244 | ||
7c16ec58 MT |
8245 | /* |
8246 | * Initializers for schedule domains | |
8247 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8248 | */ | |
8249 | ||
a5d8c348 IM |
8250 | #ifdef CONFIG_SCHED_DEBUG |
8251 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8252 | #else | |
8253 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8254 | #endif | |
8255 | ||
7c16ec58 | 8256 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8257 | |
7c16ec58 MT |
8258 | #define SD_INIT_FUNC(type) \ |
8259 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8260 | { \ | |
8261 | memset(sd, 0, sizeof(*sd)); \ | |
8262 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8263 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8264 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8265 | } |
8266 | ||
8267 | SD_INIT_FUNC(CPU) | |
8268 | #ifdef CONFIG_NUMA | |
8269 | SD_INIT_FUNC(ALLNODES) | |
8270 | SD_INIT_FUNC(NODE) | |
8271 | #endif | |
8272 | #ifdef CONFIG_SCHED_SMT | |
8273 | SD_INIT_FUNC(SIBLING) | |
8274 | #endif | |
8275 | #ifdef CONFIG_SCHED_MC | |
8276 | SD_INIT_FUNC(MC) | |
8277 | #endif | |
8278 | ||
1d3504fc HS |
8279 | static int default_relax_domain_level = -1; |
8280 | ||
8281 | static int __init setup_relax_domain_level(char *str) | |
8282 | { | |
30e0e178 LZ |
8283 | unsigned long val; |
8284 | ||
8285 | val = simple_strtoul(str, NULL, 0); | |
8286 | if (val < SD_LV_MAX) | |
8287 | default_relax_domain_level = val; | |
8288 | ||
1d3504fc HS |
8289 | return 1; |
8290 | } | |
8291 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8292 | ||
8293 | static void set_domain_attribute(struct sched_domain *sd, | |
8294 | struct sched_domain_attr *attr) | |
8295 | { | |
8296 | int request; | |
8297 | ||
8298 | if (!attr || attr->relax_domain_level < 0) { | |
8299 | if (default_relax_domain_level < 0) | |
8300 | return; | |
8301 | else | |
8302 | request = default_relax_domain_level; | |
8303 | } else | |
8304 | request = attr->relax_domain_level; | |
8305 | if (request < sd->level) { | |
8306 | /* turn off idle balance on this domain */ | |
8307 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8308 | } else { | |
8309 | /* turn on idle balance on this domain */ | |
8310 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8311 | } | |
8312 | } | |
8313 | ||
1da177e4 | 8314 | /* |
1a20ff27 DG |
8315 | * Build sched domains for a given set of cpus and attach the sched domains |
8316 | * to the individual cpus | |
1da177e4 | 8317 | */ |
96f874e2 | 8318 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 8319 | struct sched_domain_attr *attr) |
1da177e4 | 8320 | { |
3404c8d9 | 8321 | int i, err = -ENOMEM; |
57d885fe | 8322 | struct root_domain *rd; |
3404c8d9 RR |
8323 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
8324 | tmpmask; | |
d1b55138 | 8325 | #ifdef CONFIG_NUMA |
3404c8d9 | 8326 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 8327 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 8328 | int sd_allnodes = 0; |
d1b55138 | 8329 | |
3404c8d9 RR |
8330 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
8331 | goto out; | |
8332 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
8333 | goto free_domainspan; | |
8334 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
8335 | goto free_covered; | |
8336 | #endif | |
8337 | ||
8338 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
8339 | goto free_notcovered; | |
8340 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
8341 | goto free_nodemask; | |
8342 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
8343 | goto free_this_sibling_map; | |
8344 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
8345 | goto free_this_core_map; | |
8346 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
8347 | goto free_send_covered; | |
8348 | ||
8349 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
8350 | /* |
8351 | * Allocate the per-node list of sched groups | |
8352 | */ | |
076ac2af | 8353 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 8354 | GFP_KERNEL); |
d1b55138 JH |
8355 | if (!sched_group_nodes) { |
8356 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 8357 | goto free_tmpmask; |
d1b55138 | 8358 | } |
d1b55138 | 8359 | #endif |
1da177e4 | 8360 | |
dc938520 | 8361 | rd = alloc_rootdomain(); |
57d885fe GH |
8362 | if (!rd) { |
8363 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 8364 | goto free_sched_groups; |
57d885fe GH |
8365 | } |
8366 | ||
7c16ec58 | 8367 | #ifdef CONFIG_NUMA |
96f874e2 | 8368 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
8369 | #endif |
8370 | ||
1da177e4 | 8371 | /* |
1a20ff27 | 8372 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8373 | */ |
abcd083a | 8374 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8375 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 8376 | |
6ca09dfc | 8377 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
8378 | |
8379 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
8380 | if (cpumask_weight(cpu_map) > |
8381 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 8382 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 8383 | SD_INIT(sd, ALLNODES); |
1d3504fc | 8384 | set_domain_attribute(sd, attr); |
758b2cdc | 8385 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 8386 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 8387 | p = sd; |
6711cab4 | 8388 | sd_allnodes = 1; |
9c1cfda2 JH |
8389 | } else |
8390 | p = NULL; | |
8391 | ||
62ea9ceb | 8392 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 8393 | SD_INIT(sd, NODE); |
1d3504fc | 8394 | set_domain_attribute(sd, attr); |
758b2cdc | 8395 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 8396 | sd->parent = p; |
1a848870 SS |
8397 | if (p) |
8398 | p->child = sd; | |
758b2cdc RR |
8399 | cpumask_and(sched_domain_span(sd), |
8400 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
8401 | #endif |
8402 | ||
8403 | p = sd; | |
6c99e9ad | 8404 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 8405 | SD_INIT(sd, CPU); |
1d3504fc | 8406 | set_domain_attribute(sd, attr); |
758b2cdc | 8407 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8408 | sd->parent = p; |
1a848870 SS |
8409 | if (p) |
8410 | p->child = sd; | |
7c16ec58 | 8411 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8412 | |
1e9f28fa SS |
8413 | #ifdef CONFIG_SCHED_MC |
8414 | p = sd; | |
6c99e9ad | 8415 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8416 | SD_INIT(sd, MC); |
1d3504fc | 8417 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8418 | cpumask_and(sched_domain_span(sd), cpu_map, |
8419 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8420 | sd->parent = p; |
1a848870 | 8421 | p->child = sd; |
7c16ec58 | 8422 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8423 | #endif |
8424 | ||
1da177e4 LT |
8425 | #ifdef CONFIG_SCHED_SMT |
8426 | p = sd; | |
6c99e9ad | 8427 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8428 | SD_INIT(sd, SIBLING); |
1d3504fc | 8429 | set_domain_attribute(sd, attr); |
758b2cdc | 8430 | cpumask_and(sched_domain_span(sd), |
c69fc56d | 8431 | topology_thread_cpumask(i), cpu_map); |
1da177e4 | 8432 | sd->parent = p; |
1a848870 | 8433 | p->child = sd; |
7c16ec58 | 8434 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8435 | #endif |
8436 | } | |
8437 | ||
8438 | #ifdef CONFIG_SCHED_SMT | |
8439 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8440 | for_each_cpu(i, cpu_map) { |
96f874e2 | 8441 | cpumask_and(this_sibling_map, |
c69fc56d | 8442 | topology_thread_cpumask(i), cpu_map); |
96f874e2 | 8443 | if (i != cpumask_first(this_sibling_map)) |
1da177e4 LT |
8444 | continue; |
8445 | ||
dd41f596 | 8446 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8447 | &cpu_to_cpu_group, |
8448 | send_covered, tmpmask); | |
1da177e4 LT |
8449 | } |
8450 | #endif | |
8451 | ||
1e9f28fa SS |
8452 | #ifdef CONFIG_SCHED_MC |
8453 | /* Set up multi-core groups */ | |
abcd083a | 8454 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8455 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8456 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8457 | continue; |
7c16ec58 | 8458 | |
dd41f596 | 8459 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8460 | &cpu_to_core_group, |
8461 | send_covered, tmpmask); | |
1e9f28fa SS |
8462 | } |
8463 | #endif | |
8464 | ||
1da177e4 | 8465 | /* Set up physical groups */ |
076ac2af | 8466 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8467 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8468 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8469 | continue; |
8470 | ||
7c16ec58 MT |
8471 | init_sched_build_groups(nodemask, cpu_map, |
8472 | &cpu_to_phys_group, | |
8473 | send_covered, tmpmask); | |
1da177e4 LT |
8474 | } |
8475 | ||
8476 | #ifdef CONFIG_NUMA | |
8477 | /* Set up node groups */ | |
7c16ec58 | 8478 | if (sd_allnodes) { |
7c16ec58 MT |
8479 | init_sched_build_groups(cpu_map, cpu_map, |
8480 | &cpu_to_allnodes_group, | |
8481 | send_covered, tmpmask); | |
8482 | } | |
9c1cfda2 | 8483 | |
076ac2af | 8484 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8485 | /* Set up node groups */ |
8486 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8487 | int j; |
8488 | ||
96f874e2 | 8489 | cpumask_clear(covered); |
6ca09dfc | 8490 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8491 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8492 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8493 | continue; |
d1b55138 | 8494 | } |
9c1cfda2 | 8495 | |
4bdbaad3 | 8496 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8497 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8498 | |
6c99e9ad RR |
8499 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8500 | GFP_KERNEL, i); | |
51888ca2 SV |
8501 | if (!sg) { |
8502 | printk(KERN_WARNING "Can not alloc domain group for " | |
8503 | "node %d\n", i); | |
8504 | goto error; | |
8505 | } | |
9c1cfda2 | 8506 | sched_group_nodes[i] = sg; |
abcd083a | 8507 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8508 | struct sched_domain *sd; |
9761eea8 | 8509 | |
62ea9ceb | 8510 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8511 | sd->groups = sg; |
9c1cfda2 | 8512 | } |
5517d86b | 8513 | sg->__cpu_power = 0; |
758b2cdc | 8514 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8515 | sg->next = sg; |
96f874e2 | 8516 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8517 | prev = sg; |
8518 | ||
076ac2af | 8519 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8520 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8521 | |
96f874e2 RR |
8522 | cpumask_complement(notcovered, covered); |
8523 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8524 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8525 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8526 | break; |
8527 | ||
6ca09dfc | 8528 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8529 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8530 | continue; |
8531 | ||
6c99e9ad RR |
8532 | sg = kmalloc_node(sizeof(struct sched_group) + |
8533 | cpumask_size(), | |
15f0b676 | 8534 | GFP_KERNEL, i); |
9c1cfda2 JH |
8535 | if (!sg) { |
8536 | printk(KERN_WARNING | |
8537 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8538 | goto error; |
9c1cfda2 | 8539 | } |
5517d86b | 8540 | sg->__cpu_power = 0; |
758b2cdc | 8541 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8542 | sg->next = prev->next; |
96f874e2 | 8543 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8544 | prev->next = sg; |
8545 | prev = sg; | |
8546 | } | |
9c1cfda2 | 8547 | } |
1da177e4 LT |
8548 | #endif |
8549 | ||
8550 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8551 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8552 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8553 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8554 | |
89c4710e | 8555 | init_sched_groups_power(i, sd); |
5c45bf27 | 8556 | } |
1da177e4 | 8557 | #endif |
1e9f28fa | 8558 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8559 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8560 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8561 | |
89c4710e | 8562 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8563 | } |
8564 | #endif | |
1e9f28fa | 8565 | |
abcd083a | 8566 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8567 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8568 | |
89c4710e | 8569 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8570 | } |
8571 | ||
9c1cfda2 | 8572 | #ifdef CONFIG_NUMA |
076ac2af | 8573 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8574 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8575 | |
6711cab4 SS |
8576 | if (sd_allnodes) { |
8577 | struct sched_group *sg; | |
f712c0c7 | 8578 | |
96f874e2 | 8579 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8580 | tmpmask); |
f712c0c7 SS |
8581 | init_numa_sched_groups_power(sg); |
8582 | } | |
9c1cfda2 JH |
8583 | #endif |
8584 | ||
1da177e4 | 8585 | /* Attach the domains */ |
abcd083a | 8586 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8587 | struct sched_domain *sd; |
8588 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8589 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8590 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8591 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8592 | #else |
6c99e9ad | 8593 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8594 | #endif |
57d885fe | 8595 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8596 | } |
51888ca2 | 8597 | |
3404c8d9 RR |
8598 | err = 0; |
8599 | ||
8600 | free_tmpmask: | |
8601 | free_cpumask_var(tmpmask); | |
8602 | free_send_covered: | |
8603 | free_cpumask_var(send_covered); | |
8604 | free_this_core_map: | |
8605 | free_cpumask_var(this_core_map); | |
8606 | free_this_sibling_map: | |
8607 | free_cpumask_var(this_sibling_map); | |
8608 | free_nodemask: | |
8609 | free_cpumask_var(nodemask); | |
8610 | free_notcovered: | |
8611 | #ifdef CONFIG_NUMA | |
8612 | free_cpumask_var(notcovered); | |
8613 | free_covered: | |
8614 | free_cpumask_var(covered); | |
8615 | free_domainspan: | |
8616 | free_cpumask_var(domainspan); | |
8617 | out: | |
8618 | #endif | |
8619 | return err; | |
8620 | ||
8621 | free_sched_groups: | |
8622 | #ifdef CONFIG_NUMA | |
8623 | kfree(sched_group_nodes); | |
8624 | #endif | |
8625 | goto free_tmpmask; | |
51888ca2 | 8626 | |
a616058b | 8627 | #ifdef CONFIG_NUMA |
51888ca2 | 8628 | error: |
7c16ec58 | 8629 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8630 | free_rootdomain(rd); |
3404c8d9 | 8631 | goto free_tmpmask; |
a616058b | 8632 | #endif |
1da177e4 | 8633 | } |
029190c5 | 8634 | |
96f874e2 | 8635 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8636 | { |
8637 | return __build_sched_domains(cpu_map, NULL); | |
8638 | } | |
8639 | ||
96f874e2 | 8640 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8641 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8642 | static struct sched_domain_attr *dattr_cur; |
8643 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8644 | |
8645 | /* | |
8646 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8647 | * cpumask) fails, then fallback to a single sched domain, |
8648 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8649 | */ |
4212823f | 8650 | static cpumask_var_t fallback_doms; |
029190c5 | 8651 | |
ee79d1bd HC |
8652 | /* |
8653 | * arch_update_cpu_topology lets virtualized architectures update the | |
8654 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8655 | * or 0 if it stayed the same. | |
8656 | */ | |
8657 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8658 | { |
ee79d1bd | 8659 | return 0; |
22e52b07 HC |
8660 | } |
8661 | ||
1a20ff27 | 8662 | /* |
41a2d6cf | 8663 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8664 | * For now this just excludes isolated cpus, but could be used to |
8665 | * exclude other special cases in the future. | |
1a20ff27 | 8666 | */ |
96f874e2 | 8667 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8668 | { |
7378547f MM |
8669 | int err; |
8670 | ||
22e52b07 | 8671 | arch_update_cpu_topology(); |
029190c5 | 8672 | ndoms_cur = 1; |
96f874e2 | 8673 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8674 | if (!doms_cur) |
4212823f | 8675 | doms_cur = fallback_doms; |
dcc30a35 | 8676 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8677 | dattr_cur = NULL; |
7378547f | 8678 | err = build_sched_domains(doms_cur); |
6382bc90 | 8679 | register_sched_domain_sysctl(); |
7378547f MM |
8680 | |
8681 | return err; | |
1a20ff27 DG |
8682 | } |
8683 | ||
96f874e2 RR |
8684 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8685 | struct cpumask *tmpmask) | |
1da177e4 | 8686 | { |
7c16ec58 | 8687 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8688 | } |
1da177e4 | 8689 | |
1a20ff27 DG |
8690 | /* |
8691 | * Detach sched domains from a group of cpus specified in cpu_map | |
8692 | * These cpus will now be attached to the NULL domain | |
8693 | */ | |
96f874e2 | 8694 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8695 | { |
96f874e2 RR |
8696 | /* Save because hotplug lock held. */ |
8697 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8698 | int i; |
8699 | ||
abcd083a | 8700 | for_each_cpu(i, cpu_map) |
57d885fe | 8701 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8702 | synchronize_sched(); |
96f874e2 | 8703 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8704 | } |
8705 | ||
1d3504fc HS |
8706 | /* handle null as "default" */ |
8707 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8708 | struct sched_domain_attr *new, int idx_new) | |
8709 | { | |
8710 | struct sched_domain_attr tmp; | |
8711 | ||
8712 | /* fast path */ | |
8713 | if (!new && !cur) | |
8714 | return 1; | |
8715 | ||
8716 | tmp = SD_ATTR_INIT; | |
8717 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8718 | new ? (new + idx_new) : &tmp, | |
8719 | sizeof(struct sched_domain_attr)); | |
8720 | } | |
8721 | ||
029190c5 PJ |
8722 | /* |
8723 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8724 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8725 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8726 | * It destroys each deleted domain and builds each new domain. | |
8727 | * | |
96f874e2 | 8728 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8729 | * The masks don't intersect (don't overlap.) We should setup one |
8730 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8731 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8732 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8733 | * it as it is. | |
8734 | * | |
41a2d6cf IM |
8735 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8736 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8737 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8738 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8739 | * the single partition 'fallback_doms', it also forces the domains | |
8740 | * to be rebuilt. | |
029190c5 | 8741 | * |
96f874e2 | 8742 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8743 | * ndoms_new == 0 is a special case for destroying existing domains, |
8744 | * and it will not create the default domain. | |
dfb512ec | 8745 | * |
029190c5 PJ |
8746 | * Call with hotplug lock held |
8747 | */ | |
96f874e2 RR |
8748 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8749 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8750 | struct sched_domain_attr *dattr_new) |
029190c5 | 8751 | { |
dfb512ec | 8752 | int i, j, n; |
d65bd5ec | 8753 | int new_topology; |
029190c5 | 8754 | |
712555ee | 8755 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8756 | |
7378547f MM |
8757 | /* always unregister in case we don't destroy any domains */ |
8758 | unregister_sched_domain_sysctl(); | |
8759 | ||
d65bd5ec HC |
8760 | /* Let architecture update cpu core mappings. */ |
8761 | new_topology = arch_update_cpu_topology(); | |
8762 | ||
dfb512ec | 8763 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8764 | |
8765 | /* Destroy deleted domains */ | |
8766 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8767 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8768 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8769 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8770 | goto match1; |
8771 | } | |
8772 | /* no match - a current sched domain not in new doms_new[] */ | |
8773 | detach_destroy_domains(doms_cur + i); | |
8774 | match1: | |
8775 | ; | |
8776 | } | |
8777 | ||
e761b772 MK |
8778 | if (doms_new == NULL) { |
8779 | ndoms_cur = 0; | |
4212823f | 8780 | doms_new = fallback_doms; |
dcc30a35 | 8781 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8782 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8783 | } |
8784 | ||
029190c5 PJ |
8785 | /* Build new domains */ |
8786 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8787 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8788 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8789 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8790 | goto match2; |
8791 | } | |
8792 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8793 | __build_sched_domains(doms_new + i, |
8794 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8795 | match2: |
8796 | ; | |
8797 | } | |
8798 | ||
8799 | /* Remember the new sched domains */ | |
4212823f | 8800 | if (doms_cur != fallback_doms) |
029190c5 | 8801 | kfree(doms_cur); |
1d3504fc | 8802 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8803 | doms_cur = doms_new; |
1d3504fc | 8804 | dattr_cur = dattr_new; |
029190c5 | 8805 | ndoms_cur = ndoms_new; |
7378547f MM |
8806 | |
8807 | register_sched_domain_sysctl(); | |
a1835615 | 8808 | |
712555ee | 8809 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8810 | } |
8811 | ||
5c45bf27 | 8812 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8813 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8814 | { |
95402b38 | 8815 | get_online_cpus(); |
dfb512ec MK |
8816 | |
8817 | /* Destroy domains first to force the rebuild */ | |
8818 | partition_sched_domains(0, NULL, NULL); | |
8819 | ||
e761b772 | 8820 | rebuild_sched_domains(); |
95402b38 | 8821 | put_online_cpus(); |
5c45bf27 SS |
8822 | } |
8823 | ||
8824 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8825 | { | |
afb8a9b7 | 8826 | unsigned int level = 0; |
5c45bf27 | 8827 | |
afb8a9b7 GS |
8828 | if (sscanf(buf, "%u", &level) != 1) |
8829 | return -EINVAL; | |
8830 | ||
8831 | /* | |
8832 | * level is always be positive so don't check for | |
8833 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8834 | * What happens on 0 or 1 byte write, | |
8835 | * need to check for count as well? | |
8836 | */ | |
8837 | ||
8838 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8839 | return -EINVAL; |
8840 | ||
8841 | if (smt) | |
afb8a9b7 | 8842 | sched_smt_power_savings = level; |
5c45bf27 | 8843 | else |
afb8a9b7 | 8844 | sched_mc_power_savings = level; |
5c45bf27 | 8845 | |
c70f22d2 | 8846 | arch_reinit_sched_domains(); |
5c45bf27 | 8847 | |
c70f22d2 | 8848 | return count; |
5c45bf27 SS |
8849 | } |
8850 | ||
5c45bf27 | 8851 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8852 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8853 | char *page) | |
5c45bf27 SS |
8854 | { |
8855 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8856 | } | |
f718cd4a | 8857 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8858 | const char *buf, size_t count) |
5c45bf27 SS |
8859 | { |
8860 | return sched_power_savings_store(buf, count, 0); | |
8861 | } | |
f718cd4a AK |
8862 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8863 | sched_mc_power_savings_show, | |
8864 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8865 | #endif |
8866 | ||
8867 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8868 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8869 | char *page) | |
5c45bf27 SS |
8870 | { |
8871 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8872 | } | |
f718cd4a | 8873 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8874 | const char *buf, size_t count) |
5c45bf27 SS |
8875 | { |
8876 | return sched_power_savings_store(buf, count, 1); | |
8877 | } | |
f718cd4a AK |
8878 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8879 | sched_smt_power_savings_show, | |
6707de00 AB |
8880 | sched_smt_power_savings_store); |
8881 | #endif | |
8882 | ||
39aac648 | 8883 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8884 | { |
8885 | int err = 0; | |
8886 | ||
8887 | #ifdef CONFIG_SCHED_SMT | |
8888 | if (smt_capable()) | |
8889 | err = sysfs_create_file(&cls->kset.kobj, | |
8890 | &attr_sched_smt_power_savings.attr); | |
8891 | #endif | |
8892 | #ifdef CONFIG_SCHED_MC | |
8893 | if (!err && mc_capable()) | |
8894 | err = sysfs_create_file(&cls->kset.kobj, | |
8895 | &attr_sched_mc_power_savings.attr); | |
8896 | #endif | |
8897 | return err; | |
8898 | } | |
6d6bc0ad | 8899 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8900 | |
e761b772 | 8901 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8902 | /* |
e761b772 MK |
8903 | * Add online and remove offline CPUs from the scheduler domains. |
8904 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8905 | */ |
8906 | static int update_sched_domains(struct notifier_block *nfb, | |
8907 | unsigned long action, void *hcpu) | |
e761b772 MK |
8908 | { |
8909 | switch (action) { | |
8910 | case CPU_ONLINE: | |
8911 | case CPU_ONLINE_FROZEN: | |
8912 | case CPU_DEAD: | |
8913 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8914 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8915 | return NOTIFY_OK; |
8916 | ||
8917 | default: | |
8918 | return NOTIFY_DONE; | |
8919 | } | |
8920 | } | |
8921 | #endif | |
8922 | ||
8923 | static int update_runtime(struct notifier_block *nfb, | |
8924 | unsigned long action, void *hcpu) | |
1da177e4 | 8925 | { |
7def2be1 PZ |
8926 | int cpu = (int)(long)hcpu; |
8927 | ||
1da177e4 | 8928 | switch (action) { |
1da177e4 | 8929 | case CPU_DOWN_PREPARE: |
8bb78442 | 8930 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8931 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8932 | return NOTIFY_OK; |
8933 | ||
1da177e4 | 8934 | case CPU_DOWN_FAILED: |
8bb78442 | 8935 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8936 | case CPU_ONLINE: |
8bb78442 | 8937 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8938 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8939 | return NOTIFY_OK; |
8940 | ||
1da177e4 LT |
8941 | default: |
8942 | return NOTIFY_DONE; | |
8943 | } | |
1da177e4 | 8944 | } |
1da177e4 LT |
8945 | |
8946 | void __init sched_init_smp(void) | |
8947 | { | |
dcc30a35 RR |
8948 | cpumask_var_t non_isolated_cpus; |
8949 | ||
8950 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8951 | |
434d53b0 MT |
8952 | #if defined(CONFIG_NUMA) |
8953 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8954 | GFP_KERNEL); | |
8955 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8956 | #endif | |
95402b38 | 8957 | get_online_cpus(); |
712555ee | 8958 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8959 | arch_init_sched_domains(cpu_online_mask); |
8960 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
8961 | if (cpumask_empty(non_isolated_cpus)) | |
8962 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8963 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8964 | put_online_cpus(); |
e761b772 MK |
8965 | |
8966 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8967 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8968 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8969 | #endif |
8970 | ||
8971 | /* RT runtime code needs to handle some hotplug events */ | |
8972 | hotcpu_notifier(update_runtime, 0); | |
8973 | ||
b328ca18 | 8974 | init_hrtick(); |
5c1e1767 NP |
8975 | |
8976 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8977 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8978 | BUG(); |
19978ca6 | 8979 | sched_init_granularity(); |
dcc30a35 | 8980 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
8981 | |
8982 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 8983 | init_sched_rt_class(); |
1da177e4 LT |
8984 | } |
8985 | #else | |
8986 | void __init sched_init_smp(void) | |
8987 | { | |
19978ca6 | 8988 | sched_init_granularity(); |
1da177e4 LT |
8989 | } |
8990 | #endif /* CONFIG_SMP */ | |
8991 | ||
8992 | int in_sched_functions(unsigned long addr) | |
8993 | { | |
1da177e4 LT |
8994 | return in_lock_functions(addr) || |
8995 | (addr >= (unsigned long)__sched_text_start | |
8996 | && addr < (unsigned long)__sched_text_end); | |
8997 | } | |
8998 | ||
a9957449 | 8999 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9000 | { |
9001 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9002 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9003 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9004 | cfs_rq->rq = rq; | |
9005 | #endif | |
67e9fb2a | 9006 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9007 | } |
9008 | ||
fa85ae24 PZ |
9009 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9010 | { | |
9011 | struct rt_prio_array *array; | |
9012 | int i; | |
9013 | ||
9014 | array = &rt_rq->active; | |
9015 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9016 | INIT_LIST_HEAD(array->queue + i); | |
9017 | __clear_bit(i, array->bitmap); | |
9018 | } | |
9019 | /* delimiter for bitsearch: */ | |
9020 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9021 | ||
052f1dc7 | 9022 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9023 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9024 | #ifdef CONFIG_SMP |
e864c499 | 9025 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9026 | #endif |
48d5e258 | 9027 | #endif |
fa85ae24 PZ |
9028 | #ifdef CONFIG_SMP |
9029 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9030 | rt_rq->overloaded = 0; |
917b627d | 9031 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9032 | #endif |
9033 | ||
9034 | rt_rq->rt_time = 0; | |
9035 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9036 | rt_rq->rt_runtime = 0; |
9037 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9038 | |
052f1dc7 | 9039 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9040 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9041 | rt_rq->rq = rq; |
9042 | #endif | |
fa85ae24 PZ |
9043 | } |
9044 | ||
6f505b16 | 9045 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9046 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9047 | struct sched_entity *se, int cpu, int add, | |
9048 | struct sched_entity *parent) | |
6f505b16 | 9049 | { |
ec7dc8ac | 9050 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9051 | tg->cfs_rq[cpu] = cfs_rq; |
9052 | init_cfs_rq(cfs_rq, rq); | |
9053 | cfs_rq->tg = tg; | |
9054 | if (add) | |
9055 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9056 | ||
9057 | tg->se[cpu] = se; | |
354d60c2 DG |
9058 | /* se could be NULL for init_task_group */ |
9059 | if (!se) | |
9060 | return; | |
9061 | ||
ec7dc8ac DG |
9062 | if (!parent) |
9063 | se->cfs_rq = &rq->cfs; | |
9064 | else | |
9065 | se->cfs_rq = parent->my_q; | |
9066 | ||
6f505b16 PZ |
9067 | se->my_q = cfs_rq; |
9068 | se->load.weight = tg->shares; | |
e05510d0 | 9069 | se->load.inv_weight = 0; |
ec7dc8ac | 9070 | se->parent = parent; |
6f505b16 | 9071 | } |
052f1dc7 | 9072 | #endif |
6f505b16 | 9073 | |
052f1dc7 | 9074 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9075 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9076 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9077 | struct sched_rt_entity *parent) | |
6f505b16 | 9078 | { |
ec7dc8ac DG |
9079 | struct rq *rq = cpu_rq(cpu); |
9080 | ||
6f505b16 PZ |
9081 | tg->rt_rq[cpu] = rt_rq; |
9082 | init_rt_rq(rt_rq, rq); | |
9083 | rt_rq->tg = tg; | |
9084 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9085 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9086 | if (add) |
9087 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9088 | ||
9089 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9090 | if (!rt_se) |
9091 | return; | |
9092 | ||
ec7dc8ac DG |
9093 | if (!parent) |
9094 | rt_se->rt_rq = &rq->rt; | |
9095 | else | |
9096 | rt_se->rt_rq = parent->my_q; | |
9097 | ||
6f505b16 | 9098 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9099 | rt_se->parent = parent; |
6f505b16 PZ |
9100 | INIT_LIST_HEAD(&rt_se->run_list); |
9101 | } | |
9102 | #endif | |
9103 | ||
1da177e4 LT |
9104 | void __init sched_init(void) |
9105 | { | |
dd41f596 | 9106 | int i, j; |
434d53b0 MT |
9107 | unsigned long alloc_size = 0, ptr; |
9108 | ||
9109 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9110 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9111 | #endif | |
9112 | #ifdef CONFIG_RT_GROUP_SCHED | |
9113 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9114 | #endif |
9115 | #ifdef CONFIG_USER_SCHED | |
9116 | alloc_size *= 2; | |
df7c8e84 RR |
9117 | #endif |
9118 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9119 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9120 | #endif |
9121 | /* | |
9122 | * As sched_init() is called before page_alloc is setup, | |
9123 | * we use alloc_bootmem(). | |
9124 | */ | |
9125 | if (alloc_size) { | |
5a9d3225 | 9126 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
9127 | |
9128 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9129 | init_task_group.se = (struct sched_entity **)ptr; | |
9130 | ptr += nr_cpu_ids * sizeof(void **); | |
9131 | ||
9132 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9133 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9134 | |
9135 | #ifdef CONFIG_USER_SCHED | |
9136 | root_task_group.se = (struct sched_entity **)ptr; | |
9137 | ptr += nr_cpu_ids * sizeof(void **); | |
9138 | ||
9139 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9140 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9141 | #endif /* CONFIG_USER_SCHED */ |
9142 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9143 | #ifdef CONFIG_RT_GROUP_SCHED |
9144 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9145 | ptr += nr_cpu_ids * sizeof(void **); | |
9146 | ||
9147 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9148 | ptr += nr_cpu_ids * sizeof(void **); |
9149 | ||
9150 | #ifdef CONFIG_USER_SCHED | |
9151 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9152 | ptr += nr_cpu_ids * sizeof(void **); | |
9153 | ||
9154 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9155 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9156 | #endif /* CONFIG_USER_SCHED */ |
9157 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9158 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9159 | for_each_possible_cpu(i) { | |
9160 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9161 | ptr += cpumask_size(); | |
9162 | } | |
9163 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9164 | } |
dd41f596 | 9165 | |
57d885fe GH |
9166 | #ifdef CONFIG_SMP |
9167 | init_defrootdomain(); | |
9168 | #endif | |
9169 | ||
d0b27fa7 PZ |
9170 | init_rt_bandwidth(&def_rt_bandwidth, |
9171 | global_rt_period(), global_rt_runtime()); | |
9172 | ||
9173 | #ifdef CONFIG_RT_GROUP_SCHED | |
9174 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9175 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9176 | #ifdef CONFIG_USER_SCHED |
9177 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9178 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9179 | #endif /* CONFIG_USER_SCHED */ |
9180 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9181 | |
052f1dc7 | 9182 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9183 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9184 | INIT_LIST_HEAD(&init_task_group.children); |
9185 | ||
9186 | #ifdef CONFIG_USER_SCHED | |
9187 | INIT_LIST_HEAD(&root_task_group.children); | |
9188 | init_task_group.parent = &root_task_group; | |
9189 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9190 | #endif /* CONFIG_USER_SCHED */ |
9191 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9192 | |
0a945022 | 9193 | for_each_possible_cpu(i) { |
70b97a7f | 9194 | struct rq *rq; |
1da177e4 LT |
9195 | |
9196 | rq = cpu_rq(i); | |
9197 | spin_lock_init(&rq->lock); | |
7897986b | 9198 | rq->nr_running = 0; |
dce48a84 TG |
9199 | rq->calc_load_active = 0; |
9200 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9201 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9202 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9203 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9204 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9205 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9206 | #ifdef CONFIG_CGROUP_SCHED |
9207 | /* | |
9208 | * How much cpu bandwidth does init_task_group get? | |
9209 | * | |
9210 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9211 | * gets 100% of the cpu resources in the system. This overall | |
9212 | * system cpu resource is divided among the tasks of | |
9213 | * init_task_group and its child task-groups in a fair manner, | |
9214 | * based on each entity's (task or task-group's) weight | |
9215 | * (se->load.weight). | |
9216 | * | |
9217 | * In other words, if init_task_group has 10 tasks of weight | |
9218 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9219 | * then A0's share of the cpu resource is: | |
9220 | * | |
9221 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
9222 | * | |
9223 | * We achieve this by letting init_task_group's tasks sit | |
9224 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9225 | */ | |
ec7dc8ac | 9226 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9227 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9228 | root_task_group.shares = NICE_0_LOAD; |
9229 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9230 | /* |
9231 | * In case of task-groups formed thr' the user id of tasks, | |
9232 | * init_task_group represents tasks belonging to root user. | |
9233 | * Hence it forms a sibling of all subsequent groups formed. | |
9234 | * In this case, init_task_group gets only a fraction of overall | |
9235 | * system cpu resource, based on the weight assigned to root | |
9236 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9237 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
9238 | * (init_cfs_rq) and having one entity represent this group of | |
9239 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
9240 | */ | |
ec7dc8ac | 9241 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 9242 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
9243 | &per_cpu(init_sched_entity, i), i, 1, |
9244 | root_task_group.se[i]); | |
6f505b16 | 9245 | |
052f1dc7 | 9246 | #endif |
354d60c2 DG |
9247 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9248 | ||
9249 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9250 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9251 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9252 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9253 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9254 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9255 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9256 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9257 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9258 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9259 | root_task_group.rt_se[i]); | |
354d60c2 | 9260 | #endif |
dd41f596 | 9261 | #endif |
1da177e4 | 9262 | |
dd41f596 IM |
9263 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9264 | rq->cpu_load[j] = 0; | |
1da177e4 | 9265 | #ifdef CONFIG_SMP |
41c7ce9a | 9266 | rq->sd = NULL; |
57d885fe | 9267 | rq->rd = NULL; |
1da177e4 | 9268 | rq->active_balance = 0; |
dd41f596 | 9269 | rq->next_balance = jiffies; |
1da177e4 | 9270 | rq->push_cpu = 0; |
0a2966b4 | 9271 | rq->cpu = i; |
1f11eb6a | 9272 | rq->online = 0; |
1da177e4 LT |
9273 | rq->migration_thread = NULL; |
9274 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9275 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9276 | #endif |
8f4d37ec | 9277 | init_rq_hrtick(rq); |
1da177e4 | 9278 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9279 | } |
9280 | ||
2dd73a4f | 9281 | set_load_weight(&init_task); |
b50f60ce | 9282 | |
e107be36 AK |
9283 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9284 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9285 | #endif | |
9286 | ||
c9819f45 | 9287 | #ifdef CONFIG_SMP |
962cf36c | 9288 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9289 | #endif |
9290 | ||
b50f60ce HC |
9291 | #ifdef CONFIG_RT_MUTEXES |
9292 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9293 | #endif | |
9294 | ||
1da177e4 LT |
9295 | /* |
9296 | * The boot idle thread does lazy MMU switching as well: | |
9297 | */ | |
9298 | atomic_inc(&init_mm.mm_count); | |
9299 | enter_lazy_tlb(&init_mm, current); | |
9300 | ||
9301 | /* | |
9302 | * Make us the idle thread. Technically, schedule() should not be | |
9303 | * called from this thread, however somewhere below it might be, | |
9304 | * but because we are the idle thread, we just pick up running again | |
9305 | * when this runqueue becomes "idle". | |
9306 | */ | |
9307 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9308 | |
9309 | calc_load_update = jiffies + LOAD_FREQ; | |
9310 | ||
dd41f596 IM |
9311 | /* |
9312 | * During early bootup we pretend to be a normal task: | |
9313 | */ | |
9314 | current->sched_class = &fair_sched_class; | |
6892b75e | 9315 | |
6a7b3dc3 RR |
9316 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
9317 | alloc_bootmem_cpumask_var(&nohz_cpu_mask); | |
bf4d83f6 | 9318 | #ifdef CONFIG_SMP |
7d1e6a9b RR |
9319 | #ifdef CONFIG_NO_HZ |
9320 | alloc_bootmem_cpumask_var(&nohz.cpu_mask); | |
f711f609 | 9321 | alloc_bootmem_cpumask_var(&nohz.ilb_grp_nohz_mask); |
7d1e6a9b | 9322 | #endif |
dcc30a35 | 9323 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
bf4d83f6 | 9324 | #endif /* SMP */ |
6a7b3dc3 | 9325 | |
6892b75e | 9326 | scheduler_running = 1; |
1da177e4 LT |
9327 | } |
9328 | ||
9329 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
9330 | void __might_sleep(char *file, int line) | |
9331 | { | |
48f24c4d | 9332 | #ifdef in_atomic |
1da177e4 LT |
9333 | static unsigned long prev_jiffy; /* ratelimiting */ |
9334 | ||
aef745fc IM |
9335 | if ((!in_atomic() && !irqs_disabled()) || |
9336 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
9337 | return; | |
9338 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9339 | return; | |
9340 | prev_jiffy = jiffies; | |
9341 | ||
9342 | printk(KERN_ERR | |
9343 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9344 | file, line); | |
9345 | printk(KERN_ERR | |
9346 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9347 | in_atomic(), irqs_disabled(), | |
9348 | current->pid, current->comm); | |
9349 | ||
9350 | debug_show_held_locks(current); | |
9351 | if (irqs_disabled()) | |
9352 | print_irqtrace_events(current); | |
9353 | dump_stack(); | |
1da177e4 LT |
9354 | #endif |
9355 | } | |
9356 | EXPORT_SYMBOL(__might_sleep); | |
9357 | #endif | |
9358 | ||
9359 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9360 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9361 | { | |
9362 | int on_rq; | |
3e51f33f | 9363 | |
3a5e4dc1 AK |
9364 | update_rq_clock(rq); |
9365 | on_rq = p->se.on_rq; | |
9366 | if (on_rq) | |
9367 | deactivate_task(rq, p, 0); | |
9368 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9369 | if (on_rq) { | |
9370 | activate_task(rq, p, 0); | |
9371 | resched_task(rq->curr); | |
9372 | } | |
9373 | } | |
9374 | ||
1da177e4 LT |
9375 | void normalize_rt_tasks(void) |
9376 | { | |
a0f98a1c | 9377 | struct task_struct *g, *p; |
1da177e4 | 9378 | unsigned long flags; |
70b97a7f | 9379 | struct rq *rq; |
1da177e4 | 9380 | |
4cf5d77a | 9381 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9382 | do_each_thread(g, p) { |
178be793 IM |
9383 | /* |
9384 | * Only normalize user tasks: | |
9385 | */ | |
9386 | if (!p->mm) | |
9387 | continue; | |
9388 | ||
6cfb0d5d | 9389 | p->se.exec_start = 0; |
6cfb0d5d | 9390 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9391 | p->se.wait_start = 0; |
dd41f596 | 9392 | p->se.sleep_start = 0; |
dd41f596 | 9393 | p->se.block_start = 0; |
6cfb0d5d | 9394 | #endif |
dd41f596 IM |
9395 | |
9396 | if (!rt_task(p)) { | |
9397 | /* | |
9398 | * Renice negative nice level userspace | |
9399 | * tasks back to 0: | |
9400 | */ | |
9401 | if (TASK_NICE(p) < 0 && p->mm) | |
9402 | set_user_nice(p, 0); | |
1da177e4 | 9403 | continue; |
dd41f596 | 9404 | } |
1da177e4 | 9405 | |
4cf5d77a | 9406 | spin_lock(&p->pi_lock); |
b29739f9 | 9407 | rq = __task_rq_lock(p); |
1da177e4 | 9408 | |
178be793 | 9409 | normalize_task(rq, p); |
3a5e4dc1 | 9410 | |
b29739f9 | 9411 | __task_rq_unlock(rq); |
4cf5d77a | 9412 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9413 | } while_each_thread(g, p); |
9414 | ||
4cf5d77a | 9415 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9416 | } |
9417 | ||
9418 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9419 | |
9420 | #ifdef CONFIG_IA64 | |
9421 | /* | |
9422 | * These functions are only useful for the IA64 MCA handling. | |
9423 | * | |
9424 | * They can only be called when the whole system has been | |
9425 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9426 | * activity can take place. Using them for anything else would | |
9427 | * be a serious bug, and as a result, they aren't even visible | |
9428 | * under any other configuration. | |
9429 | */ | |
9430 | ||
9431 | /** | |
9432 | * curr_task - return the current task for a given cpu. | |
9433 | * @cpu: the processor in question. | |
9434 | * | |
9435 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9436 | */ | |
36c8b586 | 9437 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9438 | { |
9439 | return cpu_curr(cpu); | |
9440 | } | |
9441 | ||
9442 | /** | |
9443 | * set_curr_task - set the current task for a given cpu. | |
9444 | * @cpu: the processor in question. | |
9445 | * @p: the task pointer to set. | |
9446 | * | |
9447 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9448 | * are serviced on a separate stack. It allows the architecture to switch the |
9449 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9450 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9451 | * and caller must save the original value of the current task (see | |
9452 | * curr_task() above) and restore that value before reenabling interrupts and | |
9453 | * re-starting the system. | |
9454 | * | |
9455 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9456 | */ | |
36c8b586 | 9457 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9458 | { |
9459 | cpu_curr(cpu) = p; | |
9460 | } | |
9461 | ||
9462 | #endif | |
29f59db3 | 9463 | |
bccbe08a PZ |
9464 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9465 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9466 | { |
9467 | int i; | |
9468 | ||
9469 | for_each_possible_cpu(i) { | |
9470 | if (tg->cfs_rq) | |
9471 | kfree(tg->cfs_rq[i]); | |
9472 | if (tg->se) | |
9473 | kfree(tg->se[i]); | |
6f505b16 PZ |
9474 | } |
9475 | ||
9476 | kfree(tg->cfs_rq); | |
9477 | kfree(tg->se); | |
6f505b16 PZ |
9478 | } |
9479 | ||
ec7dc8ac DG |
9480 | static |
9481 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9482 | { |
29f59db3 | 9483 | struct cfs_rq *cfs_rq; |
eab17229 | 9484 | struct sched_entity *se; |
9b5b7751 | 9485 | struct rq *rq; |
29f59db3 SV |
9486 | int i; |
9487 | ||
434d53b0 | 9488 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9489 | if (!tg->cfs_rq) |
9490 | goto err; | |
434d53b0 | 9491 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9492 | if (!tg->se) |
9493 | goto err; | |
052f1dc7 PZ |
9494 | |
9495 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9496 | |
9497 | for_each_possible_cpu(i) { | |
9b5b7751 | 9498 | rq = cpu_rq(i); |
29f59db3 | 9499 | |
eab17229 LZ |
9500 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9501 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9502 | if (!cfs_rq) |
9503 | goto err; | |
9504 | ||
eab17229 LZ |
9505 | se = kzalloc_node(sizeof(struct sched_entity), |
9506 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9507 | if (!se) |
9508 | goto err; | |
9509 | ||
eab17229 | 9510 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9511 | } |
9512 | ||
9513 | return 1; | |
9514 | ||
9515 | err: | |
9516 | return 0; | |
9517 | } | |
9518 | ||
9519 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9520 | { | |
9521 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9522 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9523 | } | |
9524 | ||
9525 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9526 | { | |
9527 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9528 | } | |
6d6bc0ad | 9529 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9530 | static inline void free_fair_sched_group(struct task_group *tg) |
9531 | { | |
9532 | } | |
9533 | ||
ec7dc8ac DG |
9534 | static inline |
9535 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9536 | { |
9537 | return 1; | |
9538 | } | |
9539 | ||
9540 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9541 | { | |
9542 | } | |
9543 | ||
9544 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9545 | { | |
9546 | } | |
6d6bc0ad | 9547 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9548 | |
9549 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9550 | static void free_rt_sched_group(struct task_group *tg) |
9551 | { | |
9552 | int i; | |
9553 | ||
d0b27fa7 PZ |
9554 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9555 | ||
bccbe08a PZ |
9556 | for_each_possible_cpu(i) { |
9557 | if (tg->rt_rq) | |
9558 | kfree(tg->rt_rq[i]); | |
9559 | if (tg->rt_se) | |
9560 | kfree(tg->rt_se[i]); | |
9561 | } | |
9562 | ||
9563 | kfree(tg->rt_rq); | |
9564 | kfree(tg->rt_se); | |
9565 | } | |
9566 | ||
ec7dc8ac DG |
9567 | static |
9568 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9569 | { |
9570 | struct rt_rq *rt_rq; | |
eab17229 | 9571 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9572 | struct rq *rq; |
9573 | int i; | |
9574 | ||
434d53b0 | 9575 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9576 | if (!tg->rt_rq) |
9577 | goto err; | |
434d53b0 | 9578 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9579 | if (!tg->rt_se) |
9580 | goto err; | |
9581 | ||
d0b27fa7 PZ |
9582 | init_rt_bandwidth(&tg->rt_bandwidth, |
9583 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9584 | |
9585 | for_each_possible_cpu(i) { | |
9586 | rq = cpu_rq(i); | |
9587 | ||
eab17229 LZ |
9588 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9589 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9590 | if (!rt_rq) |
9591 | goto err; | |
29f59db3 | 9592 | |
eab17229 LZ |
9593 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9594 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9595 | if (!rt_se) |
9596 | goto err; | |
29f59db3 | 9597 | |
eab17229 | 9598 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9599 | } |
9600 | ||
bccbe08a PZ |
9601 | return 1; |
9602 | ||
9603 | err: | |
9604 | return 0; | |
9605 | } | |
9606 | ||
9607 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9608 | { | |
9609 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9610 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9611 | } | |
9612 | ||
9613 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9614 | { | |
9615 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9616 | } | |
6d6bc0ad | 9617 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9618 | static inline void free_rt_sched_group(struct task_group *tg) |
9619 | { | |
9620 | } | |
9621 | ||
ec7dc8ac DG |
9622 | static inline |
9623 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9624 | { |
9625 | return 1; | |
9626 | } | |
9627 | ||
9628 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9629 | { | |
9630 | } | |
9631 | ||
9632 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9633 | { | |
9634 | } | |
6d6bc0ad | 9635 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9636 | |
d0b27fa7 | 9637 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9638 | static void free_sched_group(struct task_group *tg) |
9639 | { | |
9640 | free_fair_sched_group(tg); | |
9641 | free_rt_sched_group(tg); | |
9642 | kfree(tg); | |
9643 | } | |
9644 | ||
9645 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9646 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9647 | { |
9648 | struct task_group *tg; | |
9649 | unsigned long flags; | |
9650 | int i; | |
9651 | ||
9652 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9653 | if (!tg) | |
9654 | return ERR_PTR(-ENOMEM); | |
9655 | ||
ec7dc8ac | 9656 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9657 | goto err; |
9658 | ||
ec7dc8ac | 9659 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9660 | goto err; |
9661 | ||
8ed36996 | 9662 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9663 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9664 | register_fair_sched_group(tg, i); |
9665 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9666 | } |
6f505b16 | 9667 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9668 | |
9669 | WARN_ON(!parent); /* root should already exist */ | |
9670 | ||
9671 | tg->parent = parent; | |
f473aa5e | 9672 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9673 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9674 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9675 | |
9b5b7751 | 9676 | return tg; |
29f59db3 SV |
9677 | |
9678 | err: | |
6f505b16 | 9679 | free_sched_group(tg); |
29f59db3 SV |
9680 | return ERR_PTR(-ENOMEM); |
9681 | } | |
9682 | ||
9b5b7751 | 9683 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9684 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9685 | { |
29f59db3 | 9686 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9687 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9688 | } |
9689 | ||
9b5b7751 | 9690 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9691 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9692 | { |
8ed36996 | 9693 | unsigned long flags; |
9b5b7751 | 9694 | int i; |
29f59db3 | 9695 | |
8ed36996 | 9696 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9697 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9698 | unregister_fair_sched_group(tg, i); |
9699 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9700 | } |
6f505b16 | 9701 | list_del_rcu(&tg->list); |
f473aa5e | 9702 | list_del_rcu(&tg->siblings); |
8ed36996 | 9703 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9704 | |
9b5b7751 | 9705 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9706 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9707 | } |
9708 | ||
9b5b7751 | 9709 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9710 | * The caller of this function should have put the task in its new group |
9711 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9712 | * reflect its new group. | |
9b5b7751 SV |
9713 | */ |
9714 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9715 | { |
9716 | int on_rq, running; | |
9717 | unsigned long flags; | |
9718 | struct rq *rq; | |
9719 | ||
9720 | rq = task_rq_lock(tsk, &flags); | |
9721 | ||
29f59db3 SV |
9722 | update_rq_clock(rq); |
9723 | ||
051a1d1a | 9724 | running = task_current(rq, tsk); |
29f59db3 SV |
9725 | on_rq = tsk->se.on_rq; |
9726 | ||
0e1f3483 | 9727 | if (on_rq) |
29f59db3 | 9728 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9729 | if (unlikely(running)) |
9730 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9731 | |
6f505b16 | 9732 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9733 | |
810b3817 PZ |
9734 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9735 | if (tsk->sched_class->moved_group) | |
9736 | tsk->sched_class->moved_group(tsk); | |
9737 | #endif | |
9738 | ||
0e1f3483 HS |
9739 | if (unlikely(running)) |
9740 | tsk->sched_class->set_curr_task(rq); | |
9741 | if (on_rq) | |
7074badb | 9742 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9743 | |
29f59db3 SV |
9744 | task_rq_unlock(rq, &flags); |
9745 | } | |
6d6bc0ad | 9746 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9747 | |
052f1dc7 | 9748 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9749 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9750 | { |
9751 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9752 | int on_rq; |
9753 | ||
29f59db3 | 9754 | on_rq = se->on_rq; |
62fb1851 | 9755 | if (on_rq) |
29f59db3 SV |
9756 | dequeue_entity(cfs_rq, se, 0); |
9757 | ||
9758 | se->load.weight = shares; | |
e05510d0 | 9759 | se->load.inv_weight = 0; |
29f59db3 | 9760 | |
62fb1851 | 9761 | if (on_rq) |
29f59db3 | 9762 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9763 | } |
62fb1851 | 9764 | |
c09595f6 PZ |
9765 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9766 | { | |
9767 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9768 | struct rq *rq = cfs_rq->rq; | |
9769 | unsigned long flags; | |
9770 | ||
9771 | spin_lock_irqsave(&rq->lock, flags); | |
9772 | __set_se_shares(se, shares); | |
9773 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9774 | } |
9775 | ||
8ed36996 PZ |
9776 | static DEFINE_MUTEX(shares_mutex); |
9777 | ||
4cf86d77 | 9778 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9779 | { |
9780 | int i; | |
8ed36996 | 9781 | unsigned long flags; |
c61935fd | 9782 | |
ec7dc8ac DG |
9783 | /* |
9784 | * We can't change the weight of the root cgroup. | |
9785 | */ | |
9786 | if (!tg->se[0]) | |
9787 | return -EINVAL; | |
9788 | ||
18d95a28 PZ |
9789 | if (shares < MIN_SHARES) |
9790 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9791 | else if (shares > MAX_SHARES) |
9792 | shares = MAX_SHARES; | |
62fb1851 | 9793 | |
8ed36996 | 9794 | mutex_lock(&shares_mutex); |
9b5b7751 | 9795 | if (tg->shares == shares) |
5cb350ba | 9796 | goto done; |
29f59db3 | 9797 | |
8ed36996 | 9798 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9799 | for_each_possible_cpu(i) |
9800 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9801 | list_del_rcu(&tg->siblings); |
8ed36996 | 9802 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9803 | |
9804 | /* wait for any ongoing reference to this group to finish */ | |
9805 | synchronize_sched(); | |
9806 | ||
9807 | /* | |
9808 | * Now we are free to modify the group's share on each cpu | |
9809 | * w/o tripping rebalance_share or load_balance_fair. | |
9810 | */ | |
9b5b7751 | 9811 | tg->shares = shares; |
c09595f6 PZ |
9812 | for_each_possible_cpu(i) { |
9813 | /* | |
9814 | * force a rebalance | |
9815 | */ | |
9816 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9817 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9818 | } |
29f59db3 | 9819 | |
6b2d7700 SV |
9820 | /* |
9821 | * Enable load balance activity on this group, by inserting it back on | |
9822 | * each cpu's rq->leaf_cfs_rq_list. | |
9823 | */ | |
8ed36996 | 9824 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9825 | for_each_possible_cpu(i) |
9826 | register_fair_sched_group(tg, i); | |
f473aa5e | 9827 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9828 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9829 | done: |
8ed36996 | 9830 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9831 | return 0; |
29f59db3 SV |
9832 | } |
9833 | ||
5cb350ba DG |
9834 | unsigned long sched_group_shares(struct task_group *tg) |
9835 | { | |
9836 | return tg->shares; | |
9837 | } | |
052f1dc7 | 9838 | #endif |
5cb350ba | 9839 | |
052f1dc7 | 9840 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9841 | /* |
9f0c1e56 | 9842 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9843 | */ |
9f0c1e56 PZ |
9844 | static DEFINE_MUTEX(rt_constraints_mutex); |
9845 | ||
9846 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9847 | { | |
9848 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9849 | return 1ULL << 20; |
9f0c1e56 | 9850 | |
9a7e0b18 | 9851 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9852 | } |
9853 | ||
9a7e0b18 PZ |
9854 | /* Must be called with tasklist_lock held */ |
9855 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9856 | { |
9a7e0b18 | 9857 | struct task_struct *g, *p; |
b40b2e8e | 9858 | |
9a7e0b18 PZ |
9859 | do_each_thread(g, p) { |
9860 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9861 | return 1; | |
9862 | } while_each_thread(g, p); | |
b40b2e8e | 9863 | |
9a7e0b18 PZ |
9864 | return 0; |
9865 | } | |
b40b2e8e | 9866 | |
9a7e0b18 PZ |
9867 | struct rt_schedulable_data { |
9868 | struct task_group *tg; | |
9869 | u64 rt_period; | |
9870 | u64 rt_runtime; | |
9871 | }; | |
b40b2e8e | 9872 | |
9a7e0b18 PZ |
9873 | static int tg_schedulable(struct task_group *tg, void *data) |
9874 | { | |
9875 | struct rt_schedulable_data *d = data; | |
9876 | struct task_group *child; | |
9877 | unsigned long total, sum = 0; | |
9878 | u64 period, runtime; | |
b40b2e8e | 9879 | |
9a7e0b18 PZ |
9880 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9881 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9882 | |
9a7e0b18 PZ |
9883 | if (tg == d->tg) { |
9884 | period = d->rt_period; | |
9885 | runtime = d->rt_runtime; | |
b40b2e8e | 9886 | } |
b40b2e8e | 9887 | |
98a4826b PZ |
9888 | #ifdef CONFIG_USER_SCHED |
9889 | if (tg == &root_task_group) { | |
9890 | period = global_rt_period(); | |
9891 | runtime = global_rt_runtime(); | |
9892 | } | |
9893 | #endif | |
9894 | ||
4653f803 PZ |
9895 | /* |
9896 | * Cannot have more runtime than the period. | |
9897 | */ | |
9898 | if (runtime > period && runtime != RUNTIME_INF) | |
9899 | return -EINVAL; | |
6f505b16 | 9900 | |
4653f803 PZ |
9901 | /* |
9902 | * Ensure we don't starve existing RT tasks. | |
9903 | */ | |
9a7e0b18 PZ |
9904 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9905 | return -EBUSY; | |
6f505b16 | 9906 | |
9a7e0b18 | 9907 | total = to_ratio(period, runtime); |
6f505b16 | 9908 | |
4653f803 PZ |
9909 | /* |
9910 | * Nobody can have more than the global setting allows. | |
9911 | */ | |
9912 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9913 | return -EINVAL; | |
6f505b16 | 9914 | |
4653f803 PZ |
9915 | /* |
9916 | * The sum of our children's runtime should not exceed our own. | |
9917 | */ | |
9a7e0b18 PZ |
9918 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9919 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9920 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9921 | |
9a7e0b18 PZ |
9922 | if (child == d->tg) { |
9923 | period = d->rt_period; | |
9924 | runtime = d->rt_runtime; | |
9925 | } | |
6f505b16 | 9926 | |
9a7e0b18 | 9927 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9928 | } |
6f505b16 | 9929 | |
9a7e0b18 PZ |
9930 | if (sum > total) |
9931 | return -EINVAL; | |
9932 | ||
9933 | return 0; | |
6f505b16 PZ |
9934 | } |
9935 | ||
9a7e0b18 | 9936 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9937 | { |
9a7e0b18 PZ |
9938 | struct rt_schedulable_data data = { |
9939 | .tg = tg, | |
9940 | .rt_period = period, | |
9941 | .rt_runtime = runtime, | |
9942 | }; | |
9943 | ||
9944 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9945 | } |
9946 | ||
d0b27fa7 PZ |
9947 | static int tg_set_bandwidth(struct task_group *tg, |
9948 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9949 | { |
ac086bc2 | 9950 | int i, err = 0; |
9f0c1e56 | 9951 | |
9f0c1e56 | 9952 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9953 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9954 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9955 | if (err) | |
9f0c1e56 | 9956 | goto unlock; |
ac086bc2 PZ |
9957 | |
9958 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
9959 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
9960 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
9961 | |
9962 | for_each_possible_cpu(i) { | |
9963 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
9964 | ||
9965 | spin_lock(&rt_rq->rt_runtime_lock); | |
9966 | rt_rq->rt_runtime = rt_runtime; | |
9967 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9968 | } | |
9969 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 9970 | unlock: |
521f1a24 | 9971 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
9972 | mutex_unlock(&rt_constraints_mutex); |
9973 | ||
9974 | return err; | |
6f505b16 PZ |
9975 | } |
9976 | ||
d0b27fa7 PZ |
9977 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
9978 | { | |
9979 | u64 rt_runtime, rt_period; | |
9980 | ||
9981 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9982 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
9983 | if (rt_runtime_us < 0) | |
9984 | rt_runtime = RUNTIME_INF; | |
9985 | ||
9986 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
9987 | } | |
9988 | ||
9f0c1e56 PZ |
9989 | long sched_group_rt_runtime(struct task_group *tg) |
9990 | { | |
9991 | u64 rt_runtime_us; | |
9992 | ||
d0b27fa7 | 9993 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9994 | return -1; |
9995 | ||
d0b27fa7 | 9996 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9997 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9998 | return rt_runtime_us; | |
9999 | } | |
d0b27fa7 PZ |
10000 | |
10001 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10002 | { | |
10003 | u64 rt_runtime, rt_period; | |
10004 | ||
10005 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10006 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10007 | ||
619b0488 R |
10008 | if (rt_period == 0) |
10009 | return -EINVAL; | |
10010 | ||
d0b27fa7 PZ |
10011 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10012 | } | |
10013 | ||
10014 | long sched_group_rt_period(struct task_group *tg) | |
10015 | { | |
10016 | u64 rt_period_us; | |
10017 | ||
10018 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10019 | do_div(rt_period_us, NSEC_PER_USEC); | |
10020 | return rt_period_us; | |
10021 | } | |
10022 | ||
10023 | static int sched_rt_global_constraints(void) | |
10024 | { | |
4653f803 | 10025 | u64 runtime, period; |
d0b27fa7 PZ |
10026 | int ret = 0; |
10027 | ||
ec5d4989 HS |
10028 | if (sysctl_sched_rt_period <= 0) |
10029 | return -EINVAL; | |
10030 | ||
4653f803 PZ |
10031 | runtime = global_rt_runtime(); |
10032 | period = global_rt_period(); | |
10033 | ||
10034 | /* | |
10035 | * Sanity check on the sysctl variables. | |
10036 | */ | |
10037 | if (runtime > period && runtime != RUNTIME_INF) | |
10038 | return -EINVAL; | |
10b612f4 | 10039 | |
d0b27fa7 | 10040 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10041 | read_lock(&tasklist_lock); |
4653f803 | 10042 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10043 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10044 | mutex_unlock(&rt_constraints_mutex); |
10045 | ||
10046 | return ret; | |
10047 | } | |
54e99124 DG |
10048 | |
10049 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10050 | { | |
10051 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10052 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10053 | return 0; | |
10054 | ||
10055 | return 1; | |
10056 | } | |
10057 | ||
6d6bc0ad | 10058 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10059 | static int sched_rt_global_constraints(void) |
10060 | { | |
ac086bc2 PZ |
10061 | unsigned long flags; |
10062 | int i; | |
10063 | ||
ec5d4989 HS |
10064 | if (sysctl_sched_rt_period <= 0) |
10065 | return -EINVAL; | |
10066 | ||
60aa605d PZ |
10067 | /* |
10068 | * There's always some RT tasks in the root group | |
10069 | * -- migration, kstopmachine etc.. | |
10070 | */ | |
10071 | if (sysctl_sched_rt_runtime == 0) | |
10072 | return -EBUSY; | |
10073 | ||
ac086bc2 PZ |
10074 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10075 | for_each_possible_cpu(i) { | |
10076 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10077 | ||
10078 | spin_lock(&rt_rq->rt_runtime_lock); | |
10079 | rt_rq->rt_runtime = global_rt_runtime(); | |
10080 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10081 | } | |
10082 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10083 | ||
d0b27fa7 PZ |
10084 | return 0; |
10085 | } | |
6d6bc0ad | 10086 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10087 | |
10088 | int sched_rt_handler(struct ctl_table *table, int write, | |
10089 | struct file *filp, void __user *buffer, size_t *lenp, | |
10090 | loff_t *ppos) | |
10091 | { | |
10092 | int ret; | |
10093 | int old_period, old_runtime; | |
10094 | static DEFINE_MUTEX(mutex); | |
10095 | ||
10096 | mutex_lock(&mutex); | |
10097 | old_period = sysctl_sched_rt_period; | |
10098 | old_runtime = sysctl_sched_rt_runtime; | |
10099 | ||
10100 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10101 | ||
10102 | if (!ret && write) { | |
10103 | ret = sched_rt_global_constraints(); | |
10104 | if (ret) { | |
10105 | sysctl_sched_rt_period = old_period; | |
10106 | sysctl_sched_rt_runtime = old_runtime; | |
10107 | } else { | |
10108 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10109 | def_rt_bandwidth.rt_period = | |
10110 | ns_to_ktime(global_rt_period()); | |
10111 | } | |
10112 | } | |
10113 | mutex_unlock(&mutex); | |
10114 | ||
10115 | return ret; | |
10116 | } | |
68318b8e | 10117 | |
052f1dc7 | 10118 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10119 | |
10120 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10121 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10122 | { |
2b01dfe3 PM |
10123 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10124 | struct task_group, css); | |
68318b8e SV |
10125 | } |
10126 | ||
10127 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10128 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10129 | { |
ec7dc8ac | 10130 | struct task_group *tg, *parent; |
68318b8e | 10131 | |
2b01dfe3 | 10132 | if (!cgrp->parent) { |
68318b8e | 10133 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10134 | return &init_task_group.css; |
10135 | } | |
10136 | ||
ec7dc8ac DG |
10137 | parent = cgroup_tg(cgrp->parent); |
10138 | tg = sched_create_group(parent); | |
68318b8e SV |
10139 | if (IS_ERR(tg)) |
10140 | return ERR_PTR(-ENOMEM); | |
10141 | ||
68318b8e SV |
10142 | return &tg->css; |
10143 | } | |
10144 | ||
41a2d6cf IM |
10145 | static void |
10146 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10147 | { |
2b01dfe3 | 10148 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10149 | |
10150 | sched_destroy_group(tg); | |
10151 | } | |
10152 | ||
41a2d6cf IM |
10153 | static int |
10154 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10155 | struct task_struct *tsk) | |
68318b8e | 10156 | { |
b68aa230 | 10157 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10158 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10159 | return -EINVAL; |
10160 | #else | |
68318b8e SV |
10161 | /* We don't support RT-tasks being in separate groups */ |
10162 | if (tsk->sched_class != &fair_sched_class) | |
10163 | return -EINVAL; | |
b68aa230 | 10164 | #endif |
68318b8e SV |
10165 | |
10166 | return 0; | |
10167 | } | |
10168 | ||
10169 | static void | |
2b01dfe3 | 10170 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10171 | struct cgroup *old_cont, struct task_struct *tsk) |
10172 | { | |
10173 | sched_move_task(tsk); | |
10174 | } | |
10175 | ||
052f1dc7 | 10176 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10177 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10178 | u64 shareval) |
68318b8e | 10179 | { |
2b01dfe3 | 10180 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10181 | } |
10182 | ||
f4c753b7 | 10183 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10184 | { |
2b01dfe3 | 10185 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10186 | |
10187 | return (u64) tg->shares; | |
10188 | } | |
6d6bc0ad | 10189 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10190 | |
052f1dc7 | 10191 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10192 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10193 | s64 val) |
6f505b16 | 10194 | { |
06ecb27c | 10195 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10196 | } |
10197 | ||
06ecb27c | 10198 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10199 | { |
06ecb27c | 10200 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10201 | } |
d0b27fa7 PZ |
10202 | |
10203 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10204 | u64 rt_period_us) | |
10205 | { | |
10206 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10207 | } | |
10208 | ||
10209 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10210 | { | |
10211 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10212 | } | |
6d6bc0ad | 10213 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10214 | |
fe5c7cc2 | 10215 | static struct cftype cpu_files[] = { |
052f1dc7 | 10216 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10217 | { |
10218 | .name = "shares", | |
f4c753b7 PM |
10219 | .read_u64 = cpu_shares_read_u64, |
10220 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10221 | }, |
052f1dc7 PZ |
10222 | #endif |
10223 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10224 | { |
9f0c1e56 | 10225 | .name = "rt_runtime_us", |
06ecb27c PM |
10226 | .read_s64 = cpu_rt_runtime_read, |
10227 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10228 | }, |
d0b27fa7 PZ |
10229 | { |
10230 | .name = "rt_period_us", | |
f4c753b7 PM |
10231 | .read_u64 = cpu_rt_period_read_uint, |
10232 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10233 | }, |
052f1dc7 | 10234 | #endif |
68318b8e SV |
10235 | }; |
10236 | ||
10237 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10238 | { | |
fe5c7cc2 | 10239 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10240 | } |
10241 | ||
10242 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10243 | .name = "cpu", |
10244 | .create = cpu_cgroup_create, | |
10245 | .destroy = cpu_cgroup_destroy, | |
10246 | .can_attach = cpu_cgroup_can_attach, | |
10247 | .attach = cpu_cgroup_attach, | |
10248 | .populate = cpu_cgroup_populate, | |
10249 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10250 | .early_init = 1, |
10251 | }; | |
10252 | ||
052f1dc7 | 10253 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10254 | |
10255 | #ifdef CONFIG_CGROUP_CPUACCT | |
10256 | ||
10257 | /* | |
10258 | * CPU accounting code for task groups. | |
10259 | * | |
10260 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10261 | * (balbir@in.ibm.com). | |
10262 | */ | |
10263 | ||
934352f2 | 10264 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10265 | struct cpuacct { |
10266 | struct cgroup_subsys_state css; | |
10267 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10268 | u64 *cpuusage; | |
ef12fefa | 10269 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10270 | struct cpuacct *parent; |
d842de87 SV |
10271 | }; |
10272 | ||
10273 | struct cgroup_subsys cpuacct_subsys; | |
10274 | ||
10275 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10276 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10277 | { |
32cd756a | 10278 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10279 | struct cpuacct, css); |
10280 | } | |
10281 | ||
10282 | /* return cpu accounting group to which this task belongs */ | |
10283 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10284 | { | |
10285 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10286 | struct cpuacct, css); | |
10287 | } | |
10288 | ||
10289 | /* create a new cpu accounting group */ | |
10290 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10291 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10292 | { |
10293 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10294 | int i; |
d842de87 SV |
10295 | |
10296 | if (!ca) | |
ef12fefa | 10297 | goto out; |
d842de87 SV |
10298 | |
10299 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10300 | if (!ca->cpuusage) |
10301 | goto out_free_ca; | |
10302 | ||
10303 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10304 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10305 | goto out_free_counters; | |
d842de87 | 10306 | |
934352f2 BR |
10307 | if (cgrp->parent) |
10308 | ca->parent = cgroup_ca(cgrp->parent); | |
10309 | ||
d842de87 | 10310 | return &ca->css; |
ef12fefa BR |
10311 | |
10312 | out_free_counters: | |
10313 | while (--i >= 0) | |
10314 | percpu_counter_destroy(&ca->cpustat[i]); | |
10315 | free_percpu(ca->cpuusage); | |
10316 | out_free_ca: | |
10317 | kfree(ca); | |
10318 | out: | |
10319 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10320 | } |
10321 | ||
10322 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10323 | static void |
32cd756a | 10324 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10325 | { |
32cd756a | 10326 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10327 | int i; |
d842de87 | 10328 | |
ef12fefa BR |
10329 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10330 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10331 | free_percpu(ca->cpuusage); |
10332 | kfree(ca); | |
10333 | } | |
10334 | ||
720f5498 KC |
10335 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10336 | { | |
b36128c8 | 10337 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10338 | u64 data; |
10339 | ||
10340 | #ifndef CONFIG_64BIT | |
10341 | /* | |
10342 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10343 | */ | |
10344 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10345 | data = *cpuusage; | |
10346 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10347 | #else | |
10348 | data = *cpuusage; | |
10349 | #endif | |
10350 | ||
10351 | return data; | |
10352 | } | |
10353 | ||
10354 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10355 | { | |
b36128c8 | 10356 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10357 | |
10358 | #ifndef CONFIG_64BIT | |
10359 | /* | |
10360 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10361 | */ | |
10362 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10363 | *cpuusage = val; | |
10364 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10365 | #else | |
10366 | *cpuusage = val; | |
10367 | #endif | |
10368 | } | |
10369 | ||
d842de87 | 10370 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10371 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10372 | { |
32cd756a | 10373 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10374 | u64 totalcpuusage = 0; |
10375 | int i; | |
10376 | ||
720f5498 KC |
10377 | for_each_present_cpu(i) |
10378 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10379 | |
10380 | return totalcpuusage; | |
10381 | } | |
10382 | ||
0297b803 DG |
10383 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10384 | u64 reset) | |
10385 | { | |
10386 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10387 | int err = 0; | |
10388 | int i; | |
10389 | ||
10390 | if (reset) { | |
10391 | err = -EINVAL; | |
10392 | goto out; | |
10393 | } | |
10394 | ||
720f5498 KC |
10395 | for_each_present_cpu(i) |
10396 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10397 | |
0297b803 DG |
10398 | out: |
10399 | return err; | |
10400 | } | |
10401 | ||
e9515c3c KC |
10402 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10403 | struct seq_file *m) | |
10404 | { | |
10405 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10406 | u64 percpu; | |
10407 | int i; | |
10408 | ||
10409 | for_each_present_cpu(i) { | |
10410 | percpu = cpuacct_cpuusage_read(ca, i); | |
10411 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10412 | } | |
10413 | seq_printf(m, "\n"); | |
10414 | return 0; | |
10415 | } | |
10416 | ||
ef12fefa BR |
10417 | static const char *cpuacct_stat_desc[] = { |
10418 | [CPUACCT_STAT_USER] = "user", | |
10419 | [CPUACCT_STAT_SYSTEM] = "system", | |
10420 | }; | |
10421 | ||
10422 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10423 | struct cgroup_map_cb *cb) | |
10424 | { | |
10425 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10426 | int i; | |
10427 | ||
10428 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10429 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10430 | val = cputime64_to_clock_t(val); | |
10431 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10432 | } | |
10433 | return 0; | |
10434 | } | |
10435 | ||
d842de87 SV |
10436 | static struct cftype files[] = { |
10437 | { | |
10438 | .name = "usage", | |
f4c753b7 PM |
10439 | .read_u64 = cpuusage_read, |
10440 | .write_u64 = cpuusage_write, | |
d842de87 | 10441 | }, |
e9515c3c KC |
10442 | { |
10443 | .name = "usage_percpu", | |
10444 | .read_seq_string = cpuacct_percpu_seq_read, | |
10445 | }, | |
ef12fefa BR |
10446 | { |
10447 | .name = "stat", | |
10448 | .read_map = cpuacct_stats_show, | |
10449 | }, | |
d842de87 SV |
10450 | }; |
10451 | ||
32cd756a | 10452 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10453 | { |
32cd756a | 10454 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10455 | } |
10456 | ||
10457 | /* | |
10458 | * charge this task's execution time to its accounting group. | |
10459 | * | |
10460 | * called with rq->lock held. | |
10461 | */ | |
10462 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10463 | { | |
10464 | struct cpuacct *ca; | |
934352f2 | 10465 | int cpu; |
d842de87 | 10466 | |
c40c6f85 | 10467 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10468 | return; |
10469 | ||
934352f2 | 10470 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10471 | |
10472 | rcu_read_lock(); | |
10473 | ||
d842de87 | 10474 | ca = task_ca(tsk); |
d842de87 | 10475 | |
934352f2 | 10476 | for (; ca; ca = ca->parent) { |
b36128c8 | 10477 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10478 | *cpuusage += cputime; |
10479 | } | |
a18b83b7 BR |
10480 | |
10481 | rcu_read_unlock(); | |
d842de87 SV |
10482 | } |
10483 | ||
ef12fefa BR |
10484 | /* |
10485 | * Charge the system/user time to the task's accounting group. | |
10486 | */ | |
10487 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10488 | enum cpuacct_stat_index idx, cputime_t val) | |
10489 | { | |
10490 | struct cpuacct *ca; | |
10491 | ||
10492 | if (unlikely(!cpuacct_subsys.active)) | |
10493 | return; | |
10494 | ||
10495 | rcu_read_lock(); | |
10496 | ca = task_ca(tsk); | |
10497 | ||
10498 | do { | |
10499 | percpu_counter_add(&ca->cpustat[idx], val); | |
10500 | ca = ca->parent; | |
10501 | } while (ca); | |
10502 | rcu_read_unlock(); | |
10503 | } | |
10504 | ||
d842de87 SV |
10505 | struct cgroup_subsys cpuacct_subsys = { |
10506 | .name = "cpuacct", | |
10507 | .create = cpuacct_create, | |
10508 | .destroy = cpuacct_destroy, | |
10509 | .populate = cpuacct_populate, | |
10510 | .subsys_id = cpuacct_subsys_id, | |
10511 | }; | |
10512 | #endif /* CONFIG_CGROUP_CPUACCT */ |