]>
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> |
0d905bca | 42 | #include <linux/perf_counter.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
58 | #include <linux/kthread.h> | |
b5aadf7f | 59 | #include <linux/proc_fs.h> |
1da177e4 | 60 | #include <linux/seq_file.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
5517d86b | 67 | #include <linux/reciprocal_div.h> |
dff06c15 | 68 | #include <linux/unistd.h> |
f5ff8422 | 69 | #include <linux/pagemap.h> |
8f4d37ec | 70 | #include <linux/hrtimer.h> |
30914a58 | 71 | #include <linux/tick.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, | |
5c333864 | 243 | HRTIMER_MODE_ABS_PINNED, 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; |
a1ba4d8b | 496 | unsigned long rt_nr_total; |
a22d7fc1 | 497 | int overloaded; |
917b627d | 498 | struct plist_head pushable_tasks; |
fa85ae24 | 499 | #endif |
6f505b16 | 500 | int rt_throttled; |
fa85ae24 | 501 | u64 rt_time; |
ac086bc2 | 502 | u64 rt_runtime; |
ea736ed5 | 503 | /* Nests inside the rq lock: */ |
ac086bc2 | 504 | spinlock_t rt_runtime_lock; |
6f505b16 | 505 | |
052f1dc7 | 506 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
507 | unsigned long rt_nr_boosted; |
508 | ||
6f505b16 PZ |
509 | struct rq *rq; |
510 | struct list_head leaf_rt_rq_list; | |
511 | struct task_group *tg; | |
512 | struct sched_rt_entity *rt_se; | |
513 | #endif | |
6aa645ea IM |
514 | }; |
515 | ||
57d885fe GH |
516 | #ifdef CONFIG_SMP |
517 | ||
518 | /* | |
519 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
520 | * variables. Each exclusive cpuset essentially defines an island domain by |
521 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
522 | * exclusive cpuset is created, we also create and attach a new root-domain |
523 | * object. | |
524 | * | |
57d885fe GH |
525 | */ |
526 | struct root_domain { | |
527 | atomic_t refcount; | |
c6c4927b RR |
528 | cpumask_var_t span; |
529 | cpumask_var_t online; | |
637f5085 | 530 | |
0eab9146 | 531 | /* |
637f5085 GH |
532 | * The "RT overload" flag: it gets set if a CPU has more than |
533 | * one runnable RT task. | |
534 | */ | |
c6c4927b | 535 | cpumask_var_t rto_mask; |
0eab9146 | 536 | atomic_t rto_count; |
6e0534f2 GH |
537 | #ifdef CONFIG_SMP |
538 | struct cpupri cpupri; | |
539 | #endif | |
7a09b1a2 VS |
540 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
541 | /* | |
542 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
543 | * used when most cpus are idle in the system indicating overall very | |
544 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
545 | */ | |
546 | unsigned int sched_mc_preferred_wakeup_cpu; | |
547 | #endif | |
57d885fe GH |
548 | }; |
549 | ||
dc938520 GH |
550 | /* |
551 | * By default the system creates a single root-domain with all cpus as | |
552 | * members (mimicking the global state we have today). | |
553 | */ | |
57d885fe GH |
554 | static struct root_domain def_root_domain; |
555 | ||
556 | #endif | |
557 | ||
1da177e4 LT |
558 | /* |
559 | * This is the main, per-CPU runqueue data structure. | |
560 | * | |
561 | * Locking rule: those places that want to lock multiple runqueues | |
562 | * (such as the load balancing or the thread migration code), lock | |
563 | * acquire operations must be ordered by ascending &runqueue. | |
564 | */ | |
70b97a7f | 565 | struct rq { |
d8016491 IM |
566 | /* runqueue lock: */ |
567 | spinlock_t lock; | |
1da177e4 LT |
568 | |
569 | /* | |
570 | * nr_running and cpu_load should be in the same cacheline because | |
571 | * remote CPUs use both these fields when doing load calculation. | |
572 | */ | |
573 | unsigned long nr_running; | |
6aa645ea IM |
574 | #define CPU_LOAD_IDX_MAX 5 |
575 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 576 | #ifdef CONFIG_NO_HZ |
15934a37 | 577 | unsigned long last_tick_seen; |
46cb4b7c SS |
578 | unsigned char in_nohz_recently; |
579 | #endif | |
d8016491 IM |
580 | /* capture load from *all* tasks on this cpu: */ |
581 | struct load_weight load; | |
6aa645ea IM |
582 | unsigned long nr_load_updates; |
583 | u64 nr_switches; | |
23a185ca | 584 | u64 nr_migrations_in; |
6aa645ea IM |
585 | |
586 | struct cfs_rq cfs; | |
6f505b16 | 587 | struct rt_rq rt; |
6f505b16 | 588 | |
6aa645ea | 589 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
590 | /* list of leaf cfs_rq on this cpu: */ |
591 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
592 | #endif |
593 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 594 | struct list_head leaf_rt_rq_list; |
1da177e4 | 595 | #endif |
1da177e4 LT |
596 | |
597 | /* | |
598 | * This is part of a global counter where only the total sum | |
599 | * over all CPUs matters. A task can increase this counter on | |
600 | * one CPU and if it got migrated afterwards it may decrease | |
601 | * it on another CPU. Always updated under the runqueue lock: | |
602 | */ | |
603 | unsigned long nr_uninterruptible; | |
604 | ||
36c8b586 | 605 | struct task_struct *curr, *idle; |
c9819f45 | 606 | unsigned long next_balance; |
1da177e4 | 607 | struct mm_struct *prev_mm; |
6aa645ea | 608 | |
3e51f33f | 609 | u64 clock; |
6aa645ea | 610 | |
1da177e4 LT |
611 | atomic_t nr_iowait; |
612 | ||
613 | #ifdef CONFIG_SMP | |
0eab9146 | 614 | struct root_domain *rd; |
1da177e4 LT |
615 | struct sched_domain *sd; |
616 | ||
a0a522ce | 617 | unsigned char idle_at_tick; |
1da177e4 LT |
618 | /* For active balancing */ |
619 | int active_balance; | |
620 | int push_cpu; | |
d8016491 IM |
621 | /* cpu of this runqueue: */ |
622 | int cpu; | |
1f11eb6a | 623 | int online; |
1da177e4 | 624 | |
a8a51d5e | 625 | unsigned long avg_load_per_task; |
1da177e4 | 626 | |
36c8b586 | 627 | struct task_struct *migration_thread; |
1da177e4 LT |
628 | struct list_head migration_queue; |
629 | #endif | |
630 | ||
dce48a84 TG |
631 | /* calc_load related fields */ |
632 | unsigned long calc_load_update; | |
633 | long calc_load_active; | |
634 | ||
8f4d37ec | 635 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
636 | #ifdef CONFIG_SMP |
637 | int hrtick_csd_pending; | |
638 | struct call_single_data hrtick_csd; | |
639 | #endif | |
8f4d37ec PZ |
640 | struct hrtimer hrtick_timer; |
641 | #endif | |
642 | ||
1da177e4 LT |
643 | #ifdef CONFIG_SCHEDSTATS |
644 | /* latency stats */ | |
645 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
646 | unsigned long long rq_cpu_time; |
647 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
648 | |
649 | /* sys_sched_yield() stats */ | |
480b9434 | 650 | unsigned int yld_count; |
1da177e4 LT |
651 | |
652 | /* schedule() stats */ | |
480b9434 KC |
653 | unsigned int sched_switch; |
654 | unsigned int sched_count; | |
655 | unsigned int sched_goidle; | |
1da177e4 LT |
656 | |
657 | /* try_to_wake_up() stats */ | |
480b9434 KC |
658 | unsigned int ttwu_count; |
659 | unsigned int ttwu_local; | |
b8efb561 IM |
660 | |
661 | /* BKL stats */ | |
480b9434 | 662 | unsigned int bkl_count; |
1da177e4 LT |
663 | #endif |
664 | }; | |
665 | ||
f34e3b61 | 666 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 667 | |
15afe09b | 668 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 669 | { |
15afe09b | 670 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
671 | } |
672 | ||
0a2966b4 CL |
673 | static inline int cpu_of(struct rq *rq) |
674 | { | |
675 | #ifdef CONFIG_SMP | |
676 | return rq->cpu; | |
677 | #else | |
678 | return 0; | |
679 | #endif | |
680 | } | |
681 | ||
674311d5 NP |
682 | /* |
683 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 684 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
685 | * |
686 | * The domain tree of any CPU may only be accessed from within | |
687 | * preempt-disabled sections. | |
688 | */ | |
48f24c4d IM |
689 | #define for_each_domain(cpu, __sd) \ |
690 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
691 | |
692 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
693 | #define this_rq() (&__get_cpu_var(runqueues)) | |
694 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
695 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 696 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 697 | |
aa9c4c0f | 698 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
699 | { |
700 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
701 | } | |
702 | ||
bf5c91ba IM |
703 | /* |
704 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
705 | */ | |
706 | #ifdef CONFIG_SCHED_DEBUG | |
707 | # define const_debug __read_mostly | |
708 | #else | |
709 | # define const_debug static const | |
710 | #endif | |
711 | ||
017730c1 IM |
712 | /** |
713 | * runqueue_is_locked | |
714 | * | |
715 | * Returns true if the current cpu runqueue is locked. | |
716 | * This interface allows printk to be called with the runqueue lock | |
717 | * held and know whether or not it is OK to wake up the klogd. | |
718 | */ | |
719 | int runqueue_is_locked(void) | |
720 | { | |
721 | int cpu = get_cpu(); | |
722 | struct rq *rq = cpu_rq(cpu); | |
723 | int ret; | |
724 | ||
725 | ret = spin_is_locked(&rq->lock); | |
726 | put_cpu(); | |
727 | return ret; | |
728 | } | |
729 | ||
bf5c91ba IM |
730 | /* |
731 | * Debugging: various feature bits | |
732 | */ | |
f00b45c1 PZ |
733 | |
734 | #define SCHED_FEAT(name, enabled) \ | |
735 | __SCHED_FEAT_##name , | |
736 | ||
bf5c91ba | 737 | enum { |
f00b45c1 | 738 | #include "sched_features.h" |
bf5c91ba IM |
739 | }; |
740 | ||
f00b45c1 PZ |
741 | #undef SCHED_FEAT |
742 | ||
743 | #define SCHED_FEAT(name, enabled) \ | |
744 | (1UL << __SCHED_FEAT_##name) * enabled | | |
745 | ||
bf5c91ba | 746 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
747 | #include "sched_features.h" |
748 | 0; | |
749 | ||
750 | #undef SCHED_FEAT | |
751 | ||
752 | #ifdef CONFIG_SCHED_DEBUG | |
753 | #define SCHED_FEAT(name, enabled) \ | |
754 | #name , | |
755 | ||
983ed7a6 | 756 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
757 | #include "sched_features.h" |
758 | NULL | |
759 | }; | |
760 | ||
761 | #undef SCHED_FEAT | |
762 | ||
34f3a814 | 763 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 764 | { |
f00b45c1 PZ |
765 | int i; |
766 | ||
767 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
768 | if (!(sysctl_sched_features & (1UL << i))) |
769 | seq_puts(m, "NO_"); | |
770 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 771 | } |
34f3a814 | 772 | seq_puts(m, "\n"); |
f00b45c1 | 773 | |
34f3a814 | 774 | return 0; |
f00b45c1 PZ |
775 | } |
776 | ||
777 | static ssize_t | |
778 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
779 | size_t cnt, loff_t *ppos) | |
780 | { | |
781 | char buf[64]; | |
782 | char *cmp = buf; | |
783 | int neg = 0; | |
784 | int i; | |
785 | ||
786 | if (cnt > 63) | |
787 | cnt = 63; | |
788 | ||
789 | if (copy_from_user(&buf, ubuf, cnt)) | |
790 | return -EFAULT; | |
791 | ||
792 | buf[cnt] = 0; | |
793 | ||
c24b7c52 | 794 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
795 | neg = 1; |
796 | cmp += 3; | |
797 | } | |
798 | ||
799 | for (i = 0; sched_feat_names[i]; i++) { | |
800 | int len = strlen(sched_feat_names[i]); | |
801 | ||
802 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
803 | if (neg) | |
804 | sysctl_sched_features &= ~(1UL << i); | |
805 | else | |
806 | sysctl_sched_features |= (1UL << i); | |
807 | break; | |
808 | } | |
809 | } | |
810 | ||
811 | if (!sched_feat_names[i]) | |
812 | return -EINVAL; | |
813 | ||
814 | filp->f_pos += cnt; | |
815 | ||
816 | return cnt; | |
817 | } | |
818 | ||
34f3a814 LZ |
819 | static int sched_feat_open(struct inode *inode, struct file *filp) |
820 | { | |
821 | return single_open(filp, sched_feat_show, NULL); | |
822 | } | |
823 | ||
f00b45c1 | 824 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
825 | .open = sched_feat_open, |
826 | .write = sched_feat_write, | |
827 | .read = seq_read, | |
828 | .llseek = seq_lseek, | |
829 | .release = single_release, | |
f00b45c1 PZ |
830 | }; |
831 | ||
832 | static __init int sched_init_debug(void) | |
833 | { | |
f00b45c1 PZ |
834 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
835 | &sched_feat_fops); | |
836 | ||
837 | return 0; | |
838 | } | |
839 | late_initcall(sched_init_debug); | |
840 | ||
841 | #endif | |
842 | ||
843 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 844 | |
b82d9fdd PZ |
845 | /* |
846 | * Number of tasks to iterate in a single balance run. | |
847 | * Limited because this is done with IRQs disabled. | |
848 | */ | |
849 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
850 | ||
2398f2c6 PZ |
851 | /* |
852 | * ratelimit for updating the group shares. | |
55cd5340 | 853 | * default: 0.25ms |
2398f2c6 | 854 | */ |
55cd5340 | 855 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 856 | |
ffda12a1 PZ |
857 | /* |
858 | * Inject some fuzzyness into changing the per-cpu group shares | |
859 | * this avoids remote rq-locks at the expense of fairness. | |
860 | * default: 4 | |
861 | */ | |
862 | unsigned int sysctl_sched_shares_thresh = 4; | |
863 | ||
fa85ae24 | 864 | /* |
9f0c1e56 | 865 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
866 | * default: 1s |
867 | */ | |
9f0c1e56 | 868 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 869 | |
6892b75e IM |
870 | static __read_mostly int scheduler_running; |
871 | ||
9f0c1e56 PZ |
872 | /* |
873 | * part of the period that we allow rt tasks to run in us. | |
874 | * default: 0.95s | |
875 | */ | |
876 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 877 | |
d0b27fa7 PZ |
878 | static inline u64 global_rt_period(void) |
879 | { | |
880 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
881 | } | |
882 | ||
883 | static inline u64 global_rt_runtime(void) | |
884 | { | |
e26873bb | 885 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
886 | return RUNTIME_INF; |
887 | ||
888 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
889 | } | |
fa85ae24 | 890 | |
1da177e4 | 891 | #ifndef prepare_arch_switch |
4866cde0 NP |
892 | # define prepare_arch_switch(next) do { } while (0) |
893 | #endif | |
894 | #ifndef finish_arch_switch | |
895 | # define finish_arch_switch(prev) do { } while (0) | |
896 | #endif | |
897 | ||
051a1d1a DA |
898 | static inline int task_current(struct rq *rq, struct task_struct *p) |
899 | { | |
900 | return rq->curr == p; | |
901 | } | |
902 | ||
4866cde0 | 903 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 904 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 905 | { |
051a1d1a | 906 | return task_current(rq, p); |
4866cde0 NP |
907 | } |
908 | ||
70b97a7f | 909 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
910 | { |
911 | } | |
912 | ||
70b97a7f | 913 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 914 | { |
da04c035 IM |
915 | #ifdef CONFIG_DEBUG_SPINLOCK |
916 | /* this is a valid case when another task releases the spinlock */ | |
917 | rq->lock.owner = current; | |
918 | #endif | |
8a25d5de IM |
919 | /* |
920 | * If we are tracking spinlock dependencies then we have to | |
921 | * fix up the runqueue lock - which gets 'carried over' from | |
922 | * prev into current: | |
923 | */ | |
924 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
925 | ||
4866cde0 NP |
926 | spin_unlock_irq(&rq->lock); |
927 | } | |
928 | ||
929 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 930 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
931 | { |
932 | #ifdef CONFIG_SMP | |
933 | return p->oncpu; | |
934 | #else | |
051a1d1a | 935 | return task_current(rq, p); |
4866cde0 NP |
936 | #endif |
937 | } | |
938 | ||
70b97a7f | 939 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
940 | { |
941 | #ifdef CONFIG_SMP | |
942 | /* | |
943 | * We can optimise this out completely for !SMP, because the | |
944 | * SMP rebalancing from interrupt is the only thing that cares | |
945 | * here. | |
946 | */ | |
947 | next->oncpu = 1; | |
948 | #endif | |
949 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
950 | spin_unlock_irq(&rq->lock); | |
951 | #else | |
952 | spin_unlock(&rq->lock); | |
953 | #endif | |
954 | } | |
955 | ||
70b97a7f | 956 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
957 | { |
958 | #ifdef CONFIG_SMP | |
959 | /* | |
960 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
961 | * We must ensure this doesn't happen until the switch is completely | |
962 | * finished. | |
963 | */ | |
964 | smp_wmb(); | |
965 | prev->oncpu = 0; | |
966 | #endif | |
967 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
968 | local_irq_enable(); | |
1da177e4 | 969 | #endif |
4866cde0 NP |
970 | } |
971 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 972 | |
b29739f9 IM |
973 | /* |
974 | * __task_rq_lock - lock the runqueue a given task resides on. | |
975 | * Must be called interrupts disabled. | |
976 | */ | |
70b97a7f | 977 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
978 | __acquires(rq->lock) |
979 | { | |
3a5c359a AK |
980 | for (;;) { |
981 | struct rq *rq = task_rq(p); | |
982 | spin_lock(&rq->lock); | |
983 | if (likely(rq == task_rq(p))) | |
984 | return rq; | |
b29739f9 | 985 | spin_unlock(&rq->lock); |
b29739f9 | 986 | } |
b29739f9 IM |
987 | } |
988 | ||
1da177e4 LT |
989 | /* |
990 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 991 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
992 | * explicitly disabling preemption. |
993 | */ | |
70b97a7f | 994 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
995 | __acquires(rq->lock) |
996 | { | |
70b97a7f | 997 | struct rq *rq; |
1da177e4 | 998 | |
3a5c359a AK |
999 | for (;;) { |
1000 | local_irq_save(*flags); | |
1001 | rq = task_rq(p); | |
1002 | spin_lock(&rq->lock); | |
1003 | if (likely(rq == task_rq(p))) | |
1004 | return rq; | |
1da177e4 | 1005 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 1006 | } |
1da177e4 LT |
1007 | } |
1008 | ||
ad474cac ON |
1009 | void task_rq_unlock_wait(struct task_struct *p) |
1010 | { | |
1011 | struct rq *rq = task_rq(p); | |
1012 | ||
1013 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1014 | spin_unlock_wait(&rq->lock); | |
1015 | } | |
1016 | ||
a9957449 | 1017 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1018 | __releases(rq->lock) |
1019 | { | |
1020 | spin_unlock(&rq->lock); | |
1021 | } | |
1022 | ||
70b97a7f | 1023 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1024 | __releases(rq->lock) |
1025 | { | |
1026 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1027 | } | |
1028 | ||
1da177e4 | 1029 | /* |
cc2a73b5 | 1030 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1031 | */ |
a9957449 | 1032 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1033 | __acquires(rq->lock) |
1034 | { | |
70b97a7f | 1035 | struct rq *rq; |
1da177e4 LT |
1036 | |
1037 | local_irq_disable(); | |
1038 | rq = this_rq(); | |
1039 | spin_lock(&rq->lock); | |
1040 | ||
1041 | return rq; | |
1042 | } | |
1043 | ||
8f4d37ec PZ |
1044 | #ifdef CONFIG_SCHED_HRTICK |
1045 | /* | |
1046 | * Use HR-timers to deliver accurate preemption points. | |
1047 | * | |
1048 | * Its all a bit involved since we cannot program an hrt while holding the | |
1049 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1050 | * reschedule event. | |
1051 | * | |
1052 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1053 | * rq->lock. | |
1054 | */ | |
8f4d37ec PZ |
1055 | |
1056 | /* | |
1057 | * Use hrtick when: | |
1058 | * - enabled by features | |
1059 | * - hrtimer is actually high res | |
1060 | */ | |
1061 | static inline int hrtick_enabled(struct rq *rq) | |
1062 | { | |
1063 | if (!sched_feat(HRTICK)) | |
1064 | return 0; | |
ba42059f | 1065 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1066 | return 0; |
8f4d37ec PZ |
1067 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1068 | } | |
1069 | ||
8f4d37ec PZ |
1070 | static void hrtick_clear(struct rq *rq) |
1071 | { | |
1072 | if (hrtimer_active(&rq->hrtick_timer)) | |
1073 | hrtimer_cancel(&rq->hrtick_timer); | |
1074 | } | |
1075 | ||
8f4d37ec PZ |
1076 | /* |
1077 | * High-resolution timer tick. | |
1078 | * Runs from hardirq context with interrupts disabled. | |
1079 | */ | |
1080 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1081 | { | |
1082 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1083 | ||
1084 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1085 | ||
1086 | spin_lock(&rq->lock); | |
3e51f33f | 1087 | update_rq_clock(rq); |
8f4d37ec PZ |
1088 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1089 | spin_unlock(&rq->lock); | |
1090 | ||
1091 | return HRTIMER_NORESTART; | |
1092 | } | |
1093 | ||
95e904c7 | 1094 | #ifdef CONFIG_SMP |
31656519 PZ |
1095 | /* |
1096 | * called from hardirq (IPI) context | |
1097 | */ | |
1098 | static void __hrtick_start(void *arg) | |
b328ca18 | 1099 | { |
31656519 | 1100 | struct rq *rq = arg; |
b328ca18 | 1101 | |
31656519 PZ |
1102 | spin_lock(&rq->lock); |
1103 | hrtimer_restart(&rq->hrtick_timer); | |
1104 | rq->hrtick_csd_pending = 0; | |
1105 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1106 | } |
1107 | ||
31656519 PZ |
1108 | /* |
1109 | * Called to set the hrtick timer state. | |
1110 | * | |
1111 | * called with rq->lock held and irqs disabled | |
1112 | */ | |
1113 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1114 | { |
31656519 PZ |
1115 | struct hrtimer *timer = &rq->hrtick_timer; |
1116 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1117 | |
cc584b21 | 1118 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1119 | |
1120 | if (rq == this_rq()) { | |
1121 | hrtimer_restart(timer); | |
1122 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1123 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1124 | rq->hrtick_csd_pending = 1; |
1125 | } | |
b328ca18 PZ |
1126 | } |
1127 | ||
1128 | static int | |
1129 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1130 | { | |
1131 | int cpu = (int)(long)hcpu; | |
1132 | ||
1133 | switch (action) { | |
1134 | case CPU_UP_CANCELED: | |
1135 | case CPU_UP_CANCELED_FROZEN: | |
1136 | case CPU_DOWN_PREPARE: | |
1137 | case CPU_DOWN_PREPARE_FROZEN: | |
1138 | case CPU_DEAD: | |
1139 | case CPU_DEAD_FROZEN: | |
31656519 | 1140 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1141 | return NOTIFY_OK; |
1142 | } | |
1143 | ||
1144 | return NOTIFY_DONE; | |
1145 | } | |
1146 | ||
fa748203 | 1147 | static __init void init_hrtick(void) |
b328ca18 PZ |
1148 | { |
1149 | hotcpu_notifier(hotplug_hrtick, 0); | |
1150 | } | |
31656519 PZ |
1151 | #else |
1152 | /* | |
1153 | * Called to set the hrtick timer state. | |
1154 | * | |
1155 | * called with rq->lock held and irqs disabled | |
1156 | */ | |
1157 | static void hrtick_start(struct rq *rq, u64 delay) | |
1158 | { | |
7f1e2ca9 | 1159 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1160 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1161 | } |
b328ca18 | 1162 | |
006c75f1 | 1163 | static inline void init_hrtick(void) |
8f4d37ec | 1164 | { |
8f4d37ec | 1165 | } |
31656519 | 1166 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1167 | |
31656519 | 1168 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1169 | { |
31656519 PZ |
1170 | #ifdef CONFIG_SMP |
1171 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1172 | |
31656519 PZ |
1173 | rq->hrtick_csd.flags = 0; |
1174 | rq->hrtick_csd.func = __hrtick_start; | |
1175 | rq->hrtick_csd.info = rq; | |
1176 | #endif | |
8f4d37ec | 1177 | |
31656519 PZ |
1178 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1179 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1180 | } |
006c75f1 | 1181 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1182 | static inline void hrtick_clear(struct rq *rq) |
1183 | { | |
1184 | } | |
1185 | ||
8f4d37ec PZ |
1186 | static inline void init_rq_hrtick(struct rq *rq) |
1187 | { | |
1188 | } | |
1189 | ||
b328ca18 PZ |
1190 | static inline void init_hrtick(void) |
1191 | { | |
1192 | } | |
006c75f1 | 1193 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1194 | |
c24d20db IM |
1195 | /* |
1196 | * resched_task - mark a task 'to be rescheduled now'. | |
1197 | * | |
1198 | * On UP this means the setting of the need_resched flag, on SMP it | |
1199 | * might also involve a cross-CPU call to trigger the scheduler on | |
1200 | * the target CPU. | |
1201 | */ | |
1202 | #ifdef CONFIG_SMP | |
1203 | ||
1204 | #ifndef tsk_is_polling | |
1205 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1206 | #endif | |
1207 | ||
31656519 | 1208 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1209 | { |
1210 | int cpu; | |
1211 | ||
1212 | assert_spin_locked(&task_rq(p)->lock); | |
1213 | ||
5ed0cec0 | 1214 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1215 | return; |
1216 | ||
5ed0cec0 | 1217 | set_tsk_need_resched(p); |
c24d20db IM |
1218 | |
1219 | cpu = task_cpu(p); | |
1220 | if (cpu == smp_processor_id()) | |
1221 | return; | |
1222 | ||
1223 | /* NEED_RESCHED must be visible before we test polling */ | |
1224 | smp_mb(); | |
1225 | if (!tsk_is_polling(p)) | |
1226 | smp_send_reschedule(cpu); | |
1227 | } | |
1228 | ||
1229 | static void resched_cpu(int cpu) | |
1230 | { | |
1231 | struct rq *rq = cpu_rq(cpu); | |
1232 | unsigned long flags; | |
1233 | ||
1234 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1235 | return; | |
1236 | resched_task(cpu_curr(cpu)); | |
1237 | spin_unlock_irqrestore(&rq->lock, flags); | |
1238 | } | |
06d8308c TG |
1239 | |
1240 | #ifdef CONFIG_NO_HZ | |
1241 | /* | |
1242 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1243 | * idle CPU then this timer might expire before the next timer event | |
1244 | * which is scheduled to wake up that CPU. In case of a completely | |
1245 | * idle system the next event might even be infinite time into the | |
1246 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1247 | * leaves the inner idle loop so the newly added timer is taken into | |
1248 | * account when the CPU goes back to idle and evaluates the timer | |
1249 | * wheel for the next timer event. | |
1250 | */ | |
1251 | void wake_up_idle_cpu(int cpu) | |
1252 | { | |
1253 | struct rq *rq = cpu_rq(cpu); | |
1254 | ||
1255 | if (cpu == smp_processor_id()) | |
1256 | return; | |
1257 | ||
1258 | /* | |
1259 | * This is safe, as this function is called with the timer | |
1260 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1261 | * to idle and has not yet set rq->curr to idle then it will | |
1262 | * be serialized on the timer wheel base lock and take the new | |
1263 | * timer into account automatically. | |
1264 | */ | |
1265 | if (rq->curr != rq->idle) | |
1266 | return; | |
1267 | ||
1268 | /* | |
1269 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1270 | * lockless. The worst case is that the other CPU runs the | |
1271 | * idle task through an additional NOOP schedule() | |
1272 | */ | |
5ed0cec0 | 1273 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1274 | |
1275 | /* NEED_RESCHED must be visible before we test polling */ | |
1276 | smp_mb(); | |
1277 | if (!tsk_is_polling(rq->idle)) | |
1278 | smp_send_reschedule(cpu); | |
1279 | } | |
6d6bc0ad | 1280 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1281 | |
6d6bc0ad | 1282 | #else /* !CONFIG_SMP */ |
31656519 | 1283 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1284 | { |
1285 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1286 | set_tsk_need_resched(p); |
c24d20db | 1287 | } |
6d6bc0ad | 1288 | #endif /* CONFIG_SMP */ |
c24d20db | 1289 | |
45bf76df IM |
1290 | #if BITS_PER_LONG == 32 |
1291 | # define WMULT_CONST (~0UL) | |
1292 | #else | |
1293 | # define WMULT_CONST (1UL << 32) | |
1294 | #endif | |
1295 | ||
1296 | #define WMULT_SHIFT 32 | |
1297 | ||
194081eb IM |
1298 | /* |
1299 | * Shift right and round: | |
1300 | */ | |
cf2ab469 | 1301 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1302 | |
a7be37ac PZ |
1303 | /* |
1304 | * delta *= weight / lw | |
1305 | */ | |
cb1c4fc9 | 1306 | static unsigned long |
45bf76df IM |
1307 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1308 | struct load_weight *lw) | |
1309 | { | |
1310 | u64 tmp; | |
1311 | ||
7a232e03 LJ |
1312 | if (!lw->inv_weight) { |
1313 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1314 | lw->inv_weight = 1; | |
1315 | else | |
1316 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1317 | / (lw->weight+1); | |
1318 | } | |
45bf76df IM |
1319 | |
1320 | tmp = (u64)delta_exec * weight; | |
1321 | /* | |
1322 | * Check whether we'd overflow the 64-bit multiplication: | |
1323 | */ | |
194081eb | 1324 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1325 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1326 | WMULT_SHIFT/2); |
1327 | else | |
cf2ab469 | 1328 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1329 | |
ecf691da | 1330 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1331 | } |
1332 | ||
1091985b | 1333 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1334 | { |
1335 | lw->weight += inc; | |
e89996ae | 1336 | lw->inv_weight = 0; |
45bf76df IM |
1337 | } |
1338 | ||
1091985b | 1339 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1340 | { |
1341 | lw->weight -= dec; | |
e89996ae | 1342 | lw->inv_weight = 0; |
45bf76df IM |
1343 | } |
1344 | ||
2dd73a4f PW |
1345 | /* |
1346 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1347 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1348 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1349 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1350 | * scaled version of the new time slice allocation that they receive on time |
1351 | * slice expiry etc. | |
1352 | */ | |
1353 | ||
cce7ade8 PZ |
1354 | #define WEIGHT_IDLEPRIO 3 |
1355 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1356 | |
1357 | /* | |
1358 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1359 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1360 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1361 | * that remained on nice 0. | |
1362 | * | |
1363 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1364 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1365 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1366 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1367 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1368 | */ |
1369 | static const int prio_to_weight[40] = { | |
254753dc IM |
1370 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1371 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1372 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1373 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1374 | /* 0 */ 1024, 820, 655, 526, 423, | |
1375 | /* 5 */ 335, 272, 215, 172, 137, | |
1376 | /* 10 */ 110, 87, 70, 56, 45, | |
1377 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1378 | }; |
1379 | ||
5714d2de IM |
1380 | /* |
1381 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1382 | * | |
1383 | * In cases where the weight does not change often, we can use the | |
1384 | * precalculated inverse to speed up arithmetics by turning divisions | |
1385 | * into multiplications: | |
1386 | */ | |
dd41f596 | 1387 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1388 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1389 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1390 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1391 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1392 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1393 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1394 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1395 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1396 | }; |
2dd73a4f | 1397 | |
dd41f596 IM |
1398 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1399 | ||
1400 | /* | |
1401 | * runqueue iterator, to support SMP load-balancing between different | |
1402 | * scheduling classes, without having to expose their internal data | |
1403 | * structures to the load-balancing proper: | |
1404 | */ | |
1405 | struct rq_iterator { | |
1406 | void *arg; | |
1407 | struct task_struct *(*start)(void *); | |
1408 | struct task_struct *(*next)(void *); | |
1409 | }; | |
1410 | ||
e1d1484f PW |
1411 | #ifdef CONFIG_SMP |
1412 | static unsigned long | |
1413 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1414 | unsigned long max_load_move, struct sched_domain *sd, | |
1415 | enum cpu_idle_type idle, int *all_pinned, | |
1416 | int *this_best_prio, struct rq_iterator *iterator); | |
1417 | ||
1418 | static int | |
1419 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1420 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1421 | struct rq_iterator *iterator); | |
e1d1484f | 1422 | #endif |
dd41f596 | 1423 | |
ef12fefa BR |
1424 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1425 | enum cpuacct_stat_index { | |
1426 | CPUACCT_STAT_USER, /* ... user mode */ | |
1427 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1428 | ||
1429 | CPUACCT_STAT_NSTATS, | |
1430 | }; | |
1431 | ||
d842de87 SV |
1432 | #ifdef CONFIG_CGROUP_CPUACCT |
1433 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1434 | static void cpuacct_update_stats(struct task_struct *tsk, |
1435 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1436 | #else |
1437 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1438 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1439 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1440 | #endif |
1441 | ||
18d95a28 PZ |
1442 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1443 | { | |
1444 | update_load_add(&rq->load, load); | |
1445 | } | |
1446 | ||
1447 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1448 | { | |
1449 | update_load_sub(&rq->load, load); | |
1450 | } | |
1451 | ||
7940ca36 | 1452 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1453 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1454 | |
1455 | /* | |
1456 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1457 | * leaving it for the final time. | |
1458 | */ | |
eb755805 | 1459 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1460 | { |
1461 | struct task_group *parent, *child; | |
eb755805 | 1462 | int ret; |
c09595f6 PZ |
1463 | |
1464 | rcu_read_lock(); | |
1465 | parent = &root_task_group; | |
1466 | down: | |
eb755805 PZ |
1467 | ret = (*down)(parent, data); |
1468 | if (ret) | |
1469 | goto out_unlock; | |
c09595f6 PZ |
1470 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1471 | parent = child; | |
1472 | goto down; | |
1473 | ||
1474 | up: | |
1475 | continue; | |
1476 | } | |
eb755805 PZ |
1477 | ret = (*up)(parent, data); |
1478 | if (ret) | |
1479 | goto out_unlock; | |
c09595f6 PZ |
1480 | |
1481 | child = parent; | |
1482 | parent = parent->parent; | |
1483 | if (parent) | |
1484 | goto up; | |
eb755805 | 1485 | out_unlock: |
c09595f6 | 1486 | rcu_read_unlock(); |
eb755805 PZ |
1487 | |
1488 | return ret; | |
c09595f6 PZ |
1489 | } |
1490 | ||
eb755805 PZ |
1491 | static int tg_nop(struct task_group *tg, void *data) |
1492 | { | |
1493 | return 0; | |
c09595f6 | 1494 | } |
eb755805 PZ |
1495 | #endif |
1496 | ||
1497 | #ifdef CONFIG_SMP | |
1498 | static unsigned long source_load(int cpu, int type); | |
1499 | static unsigned long target_load(int cpu, int type); | |
1500 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1501 | ||
1502 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1503 | { | |
1504 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1505 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1506 | |
4cd42620 SR |
1507 | if (nr_running) |
1508 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1509 | else |
1510 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1511 | |
1512 | return rq->avg_load_per_task; | |
1513 | } | |
1514 | ||
1515 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1516 | |
c09595f6 PZ |
1517 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1518 | ||
1519 | /* | |
1520 | * Calculate and set the cpu's group shares. | |
1521 | */ | |
1522 | static void | |
ffda12a1 PZ |
1523 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1524 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1525 | { |
c09595f6 PZ |
1526 | unsigned long shares; |
1527 | unsigned long rq_weight; | |
1528 | ||
c8cba857 | 1529 | if (!tg->se[cpu]) |
c09595f6 PZ |
1530 | return; |
1531 | ||
ec4e0e2f | 1532 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1533 | |
c09595f6 PZ |
1534 | /* |
1535 | * \Sum shares * rq_weight | |
1536 | * shares = ----------------------- | |
1537 | * \Sum rq_weight | |
1538 | * | |
1539 | */ | |
ec4e0e2f | 1540 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1541 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1542 | |
ffda12a1 PZ |
1543 | if (abs(shares - tg->se[cpu]->load.weight) > |
1544 | sysctl_sched_shares_thresh) { | |
1545 | struct rq *rq = cpu_rq(cpu); | |
1546 | unsigned long flags; | |
c09595f6 | 1547 | |
ffda12a1 | 1548 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1549 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1550 | |
ffda12a1 PZ |
1551 | __set_se_shares(tg->se[cpu], shares); |
1552 | spin_unlock_irqrestore(&rq->lock, flags); | |
1553 | } | |
18d95a28 | 1554 | } |
c09595f6 PZ |
1555 | |
1556 | /* | |
c8cba857 PZ |
1557 | * Re-compute the task group their per cpu shares over the given domain. |
1558 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1559 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1560 | */ |
eb755805 | 1561 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1562 | { |
ec4e0e2f | 1563 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1564 | unsigned long shares = 0; |
eb755805 | 1565 | struct sched_domain *sd = data; |
c8cba857 | 1566 | int i; |
c09595f6 | 1567 | |
758b2cdc | 1568 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1569 | /* |
1570 | * If there are currently no tasks on the cpu pretend there | |
1571 | * is one of average load so that when a new task gets to | |
1572 | * run here it will not get delayed by group starvation. | |
1573 | */ | |
1574 | weight = tg->cfs_rq[i]->load.weight; | |
1575 | if (!weight) | |
1576 | weight = NICE_0_LOAD; | |
1577 | ||
1578 | tg->cfs_rq[i]->rq_weight = weight; | |
1579 | rq_weight += weight; | |
c8cba857 | 1580 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1581 | } |
c09595f6 | 1582 | |
c8cba857 PZ |
1583 | if ((!shares && rq_weight) || shares > tg->shares) |
1584 | shares = tg->shares; | |
1585 | ||
1586 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1587 | shares = tg->shares; | |
c09595f6 | 1588 | |
758b2cdc | 1589 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1590 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1591 | |
1592 | return 0; | |
c09595f6 PZ |
1593 | } |
1594 | ||
1595 | /* | |
c8cba857 PZ |
1596 | * Compute the cpu's hierarchical load factor for each task group. |
1597 | * This needs to be done in a top-down fashion because the load of a child | |
1598 | * group is a fraction of its parents load. | |
c09595f6 | 1599 | */ |
eb755805 | 1600 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1601 | { |
c8cba857 | 1602 | unsigned long load; |
eb755805 | 1603 | long cpu = (long)data; |
c09595f6 | 1604 | |
c8cba857 PZ |
1605 | if (!tg->parent) { |
1606 | load = cpu_rq(cpu)->load.weight; | |
1607 | } else { | |
1608 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1609 | load *= tg->cfs_rq[cpu]->shares; | |
1610 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1611 | } | |
c09595f6 | 1612 | |
c8cba857 | 1613 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1614 | |
eb755805 | 1615 | return 0; |
c09595f6 PZ |
1616 | } |
1617 | ||
c8cba857 | 1618 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1619 | { |
2398f2c6 PZ |
1620 | u64 now = cpu_clock(raw_smp_processor_id()); |
1621 | s64 elapsed = now - sd->last_update; | |
1622 | ||
1623 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1624 | sd->last_update = now; | |
eb755805 | 1625 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1626 | } |
4d8d595d PZ |
1627 | } |
1628 | ||
3e5459b4 PZ |
1629 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1630 | { | |
1631 | spin_unlock(&rq->lock); | |
1632 | update_shares(sd); | |
1633 | spin_lock(&rq->lock); | |
1634 | } | |
1635 | ||
eb755805 | 1636 | static void update_h_load(long cpu) |
c09595f6 | 1637 | { |
eb755805 | 1638 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1639 | } |
1640 | ||
c09595f6 PZ |
1641 | #else |
1642 | ||
c8cba857 | 1643 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1644 | { |
1645 | } | |
1646 | ||
3e5459b4 PZ |
1647 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1648 | { | |
1649 | } | |
1650 | ||
18d95a28 PZ |
1651 | #endif |
1652 | ||
8f45e2b5 GH |
1653 | #ifdef CONFIG_PREEMPT |
1654 | ||
70574a99 | 1655 | /* |
8f45e2b5 GH |
1656 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1657 | * way at the expense of forcing extra atomic operations in all | |
1658 | * invocations. This assures that the double_lock is acquired using the | |
1659 | * same underlying policy as the spinlock_t on this architecture, which | |
1660 | * reduces latency compared to the unfair variant below. However, it | |
1661 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1662 | */ |
8f45e2b5 GH |
1663 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1664 | __releases(this_rq->lock) | |
1665 | __acquires(busiest->lock) | |
1666 | __acquires(this_rq->lock) | |
1667 | { | |
1668 | spin_unlock(&this_rq->lock); | |
1669 | double_rq_lock(this_rq, busiest); | |
1670 | ||
1671 | return 1; | |
1672 | } | |
1673 | ||
1674 | #else | |
1675 | /* | |
1676 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1677 | * latency by eliminating extra atomic operations when the locks are | |
1678 | * already in proper order on entry. This favors lower cpu-ids and will | |
1679 | * grant the double lock to lower cpus over higher ids under contention, | |
1680 | * regardless of entry order into the function. | |
1681 | */ | |
1682 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1683 | __releases(this_rq->lock) |
1684 | __acquires(busiest->lock) | |
1685 | __acquires(this_rq->lock) | |
1686 | { | |
1687 | int ret = 0; | |
1688 | ||
70574a99 AD |
1689 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1690 | if (busiest < this_rq) { | |
1691 | spin_unlock(&this_rq->lock); | |
1692 | spin_lock(&busiest->lock); | |
1693 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1694 | ret = 1; | |
1695 | } else | |
1696 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1697 | } | |
1698 | return ret; | |
1699 | } | |
1700 | ||
8f45e2b5 GH |
1701 | #endif /* CONFIG_PREEMPT */ |
1702 | ||
1703 | /* | |
1704 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1705 | */ | |
1706 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1707 | { | |
1708 | if (unlikely(!irqs_disabled())) { | |
1709 | /* printk() doesn't work good under rq->lock */ | |
1710 | spin_unlock(&this_rq->lock); | |
1711 | BUG_ON(1); | |
1712 | } | |
1713 | ||
1714 | return _double_lock_balance(this_rq, busiest); | |
1715 | } | |
1716 | ||
70574a99 AD |
1717 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1718 | __releases(busiest->lock) | |
1719 | { | |
1720 | spin_unlock(&busiest->lock); | |
1721 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1722 | } | |
18d95a28 PZ |
1723 | #endif |
1724 | ||
30432094 | 1725 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1726 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1727 | { | |
30432094 | 1728 | #ifdef CONFIG_SMP |
34e83e85 IM |
1729 | cfs_rq->shares = shares; |
1730 | #endif | |
1731 | } | |
30432094 | 1732 | #endif |
e7693a36 | 1733 | |
dce48a84 TG |
1734 | static void calc_load_account_active(struct rq *this_rq); |
1735 | ||
dd41f596 | 1736 | #include "sched_stats.h" |
dd41f596 | 1737 | #include "sched_idletask.c" |
5522d5d5 IM |
1738 | #include "sched_fair.c" |
1739 | #include "sched_rt.c" | |
dd41f596 IM |
1740 | #ifdef CONFIG_SCHED_DEBUG |
1741 | # include "sched_debug.c" | |
1742 | #endif | |
1743 | ||
1744 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1745 | #define for_each_class(class) \ |
1746 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1747 | |
c09595f6 | 1748 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1749 | { |
1750 | rq->nr_running++; | |
9c217245 IM |
1751 | } |
1752 | ||
c09595f6 | 1753 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1754 | { |
1755 | rq->nr_running--; | |
9c217245 IM |
1756 | } |
1757 | ||
45bf76df IM |
1758 | static void set_load_weight(struct task_struct *p) |
1759 | { | |
1760 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1761 | p->se.load.weight = prio_to_weight[0] * 2; |
1762 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1763 | return; | |
1764 | } | |
45bf76df | 1765 | |
dd41f596 IM |
1766 | /* |
1767 | * SCHED_IDLE tasks get minimal weight: | |
1768 | */ | |
1769 | if (p->policy == SCHED_IDLE) { | |
1770 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1771 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1772 | return; | |
1773 | } | |
71f8bd46 | 1774 | |
dd41f596 IM |
1775 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1776 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1777 | } |
1778 | ||
2087a1ad GH |
1779 | static void update_avg(u64 *avg, u64 sample) |
1780 | { | |
1781 | s64 diff = sample - *avg; | |
1782 | *avg += diff >> 3; | |
1783 | } | |
1784 | ||
8159f87e | 1785 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1786 | { |
831451ac PZ |
1787 | if (wakeup) |
1788 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1789 | ||
dd41f596 | 1790 | sched_info_queued(p); |
fd390f6a | 1791 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1792 | p->se.on_rq = 1; |
71f8bd46 IM |
1793 | } |
1794 | ||
69be72c1 | 1795 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1796 | { |
831451ac PZ |
1797 | if (sleep) { |
1798 | if (p->se.last_wakeup) { | |
1799 | update_avg(&p->se.avg_overlap, | |
1800 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1801 | p->se.last_wakeup = 0; | |
1802 | } else { | |
1803 | update_avg(&p->se.avg_wakeup, | |
1804 | sysctl_sched_wakeup_granularity); | |
1805 | } | |
2087a1ad GH |
1806 | } |
1807 | ||
46ac22ba | 1808 | sched_info_dequeued(p); |
f02231e5 | 1809 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1810 | p->se.on_rq = 0; |
71f8bd46 IM |
1811 | } |
1812 | ||
14531189 | 1813 | /* |
dd41f596 | 1814 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1815 | */ |
14531189 IM |
1816 | static inline int __normal_prio(struct task_struct *p) |
1817 | { | |
dd41f596 | 1818 | return p->static_prio; |
14531189 IM |
1819 | } |
1820 | ||
b29739f9 IM |
1821 | /* |
1822 | * Calculate the expected normal priority: i.e. priority | |
1823 | * without taking RT-inheritance into account. Might be | |
1824 | * boosted by interactivity modifiers. Changes upon fork, | |
1825 | * setprio syscalls, and whenever the interactivity | |
1826 | * estimator recalculates. | |
1827 | */ | |
36c8b586 | 1828 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1829 | { |
1830 | int prio; | |
1831 | ||
e05606d3 | 1832 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1833 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1834 | else | |
1835 | prio = __normal_prio(p); | |
1836 | return prio; | |
1837 | } | |
1838 | ||
1839 | /* | |
1840 | * Calculate the current priority, i.e. the priority | |
1841 | * taken into account by the scheduler. This value might | |
1842 | * be boosted by RT tasks, or might be boosted by | |
1843 | * interactivity modifiers. Will be RT if the task got | |
1844 | * RT-boosted. If not then it returns p->normal_prio. | |
1845 | */ | |
36c8b586 | 1846 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1847 | { |
1848 | p->normal_prio = normal_prio(p); | |
1849 | /* | |
1850 | * If we are RT tasks or we were boosted to RT priority, | |
1851 | * keep the priority unchanged. Otherwise, update priority | |
1852 | * to the normal priority: | |
1853 | */ | |
1854 | if (!rt_prio(p->prio)) | |
1855 | return p->normal_prio; | |
1856 | return p->prio; | |
1857 | } | |
1858 | ||
1da177e4 | 1859 | /* |
dd41f596 | 1860 | * activate_task - move a task to the runqueue. |
1da177e4 | 1861 | */ |
dd41f596 | 1862 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1863 | { |
d9514f6c | 1864 | if (task_contributes_to_load(p)) |
dd41f596 | 1865 | rq->nr_uninterruptible--; |
1da177e4 | 1866 | |
8159f87e | 1867 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1868 | inc_nr_running(rq); |
1da177e4 LT |
1869 | } |
1870 | ||
1da177e4 LT |
1871 | /* |
1872 | * deactivate_task - remove a task from the runqueue. | |
1873 | */ | |
2e1cb74a | 1874 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1875 | { |
d9514f6c | 1876 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1877 | rq->nr_uninterruptible++; |
1878 | ||
69be72c1 | 1879 | dequeue_task(rq, p, sleep); |
c09595f6 | 1880 | dec_nr_running(rq); |
1da177e4 LT |
1881 | } |
1882 | ||
1da177e4 LT |
1883 | /** |
1884 | * task_curr - is this task currently executing on a CPU? | |
1885 | * @p: the task in question. | |
1886 | */ | |
36c8b586 | 1887 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1888 | { |
1889 | return cpu_curr(task_cpu(p)) == p; | |
1890 | } | |
1891 | ||
dd41f596 IM |
1892 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1893 | { | |
6f505b16 | 1894 | set_task_rq(p, cpu); |
dd41f596 | 1895 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1896 | /* |
1897 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1898 | * successfuly executed on another CPU. We must ensure that updates of | |
1899 | * per-task data have been completed by this moment. | |
1900 | */ | |
1901 | smp_wmb(); | |
dd41f596 | 1902 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1903 | #endif |
2dd73a4f PW |
1904 | } |
1905 | ||
cb469845 SR |
1906 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1907 | const struct sched_class *prev_class, | |
1908 | int oldprio, int running) | |
1909 | { | |
1910 | if (prev_class != p->sched_class) { | |
1911 | if (prev_class->switched_from) | |
1912 | prev_class->switched_from(rq, p, running); | |
1913 | p->sched_class->switched_to(rq, p, running); | |
1914 | } else | |
1915 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1916 | } | |
1917 | ||
1da177e4 | 1918 | #ifdef CONFIG_SMP |
c65cc870 | 1919 | |
e958b360 TG |
1920 | /* Used instead of source_load when we know the type == 0 */ |
1921 | static unsigned long weighted_cpuload(const int cpu) | |
1922 | { | |
1923 | return cpu_rq(cpu)->load.weight; | |
1924 | } | |
1925 | ||
cc367732 IM |
1926 | /* |
1927 | * Is this task likely cache-hot: | |
1928 | */ | |
e7693a36 | 1929 | static int |
cc367732 IM |
1930 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1931 | { | |
1932 | s64 delta; | |
1933 | ||
f540a608 IM |
1934 | /* |
1935 | * Buddy candidates are cache hot: | |
1936 | */ | |
4793241b PZ |
1937 | if (sched_feat(CACHE_HOT_BUDDY) && |
1938 | (&p->se == cfs_rq_of(&p->se)->next || | |
1939 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1940 | return 1; |
1941 | ||
cc367732 IM |
1942 | if (p->sched_class != &fair_sched_class) |
1943 | return 0; | |
1944 | ||
6bc1665b IM |
1945 | if (sysctl_sched_migration_cost == -1) |
1946 | return 1; | |
1947 | if (sysctl_sched_migration_cost == 0) | |
1948 | return 0; | |
1949 | ||
cc367732 IM |
1950 | delta = now - p->se.exec_start; |
1951 | ||
1952 | return delta < (s64)sysctl_sched_migration_cost; | |
1953 | } | |
1954 | ||
1955 | ||
dd41f596 | 1956 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1957 | { |
dd41f596 IM |
1958 | int old_cpu = task_cpu(p); |
1959 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1960 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1961 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1962 | u64 clock_offset; |
dd41f596 IM |
1963 | |
1964 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1965 | |
de1d7286 | 1966 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 1967 | |
6cfb0d5d IM |
1968 | #ifdef CONFIG_SCHEDSTATS |
1969 | if (p->se.wait_start) | |
1970 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1971 | if (p->se.sleep_start) |
1972 | p->se.sleep_start -= clock_offset; | |
1973 | if (p->se.block_start) | |
1974 | p->se.block_start -= clock_offset; | |
6c594c21 | 1975 | #endif |
cc367732 | 1976 | if (old_cpu != new_cpu) { |
6c594c21 | 1977 | p->se.nr_migrations++; |
23a185ca | 1978 | new_rq->nr_migrations_in++; |
6c594c21 | 1979 | #ifdef CONFIG_SCHEDSTATS |
cc367732 IM |
1980 | if (task_hot(p, old_rq->clock, NULL)) |
1981 | schedstat_inc(p, se.nr_forced2_migrations); | |
6cfb0d5d | 1982 | #endif |
e5289d4a PZ |
1983 | perf_swcounter_event(PERF_COUNT_SW_CPU_MIGRATIONS, |
1984 | 1, 1, NULL, 0); | |
6c594c21 | 1985 | } |
2830cf8c SV |
1986 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1987 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1988 | |
1989 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1990 | } |
1991 | ||
70b97a7f | 1992 | struct migration_req { |
1da177e4 | 1993 | struct list_head list; |
1da177e4 | 1994 | |
36c8b586 | 1995 | struct task_struct *task; |
1da177e4 LT |
1996 | int dest_cpu; |
1997 | ||
1da177e4 | 1998 | struct completion done; |
70b97a7f | 1999 | }; |
1da177e4 LT |
2000 | |
2001 | /* | |
2002 | * The task's runqueue lock must be held. | |
2003 | * Returns true if you have to wait for migration thread. | |
2004 | */ | |
36c8b586 | 2005 | static int |
70b97a7f | 2006 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2007 | { |
70b97a7f | 2008 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2009 | |
2010 | /* | |
2011 | * If the task is not on a runqueue (and not running), then | |
2012 | * it is sufficient to simply update the task's cpu field. | |
2013 | */ | |
dd41f596 | 2014 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2015 | set_task_cpu(p, dest_cpu); |
2016 | return 0; | |
2017 | } | |
2018 | ||
2019 | init_completion(&req->done); | |
1da177e4 LT |
2020 | req->task = p; |
2021 | req->dest_cpu = dest_cpu; | |
2022 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2023 | |
1da177e4 LT |
2024 | return 1; |
2025 | } | |
2026 | ||
a26b89f0 MM |
2027 | /* |
2028 | * wait_task_context_switch - wait for a thread to complete at least one | |
2029 | * context switch. | |
2030 | * | |
2031 | * @p must not be current. | |
2032 | */ | |
2033 | void wait_task_context_switch(struct task_struct *p) | |
2034 | { | |
2035 | unsigned long nvcsw, nivcsw, flags; | |
2036 | int running; | |
2037 | struct rq *rq; | |
2038 | ||
2039 | nvcsw = p->nvcsw; | |
2040 | nivcsw = p->nivcsw; | |
2041 | for (;;) { | |
2042 | /* | |
2043 | * The runqueue is assigned before the actual context | |
2044 | * switch. We need to take the runqueue lock. | |
2045 | * | |
2046 | * We could check initially without the lock but it is | |
2047 | * very likely that we need to take the lock in every | |
2048 | * iteration. | |
2049 | */ | |
2050 | rq = task_rq_lock(p, &flags); | |
2051 | running = task_running(rq, p); | |
2052 | task_rq_unlock(rq, &flags); | |
2053 | ||
2054 | if (likely(!running)) | |
2055 | break; | |
2056 | /* | |
2057 | * The switch count is incremented before the actual | |
2058 | * context switch. We thus wait for two switches to be | |
2059 | * sure at least one completed. | |
2060 | */ | |
2061 | if ((p->nvcsw - nvcsw) > 1) | |
2062 | break; | |
2063 | if ((p->nivcsw - nivcsw) > 1) | |
2064 | break; | |
2065 | ||
2066 | cpu_relax(); | |
2067 | } | |
2068 | } | |
2069 | ||
1da177e4 LT |
2070 | /* |
2071 | * wait_task_inactive - wait for a thread to unschedule. | |
2072 | * | |
85ba2d86 RM |
2073 | * If @match_state is nonzero, it's the @p->state value just checked and |
2074 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2075 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2076 | * we return a positive number (its total switch count). If a second call | |
2077 | * a short while later returns the same number, the caller can be sure that | |
2078 | * @p has remained unscheduled the whole time. | |
2079 | * | |
1da177e4 LT |
2080 | * The caller must ensure that the task *will* unschedule sometime soon, |
2081 | * else this function might spin for a *long* time. This function can't | |
2082 | * be called with interrupts off, or it may introduce deadlock with | |
2083 | * smp_call_function() if an IPI is sent by the same process we are | |
2084 | * waiting to become inactive. | |
2085 | */ | |
85ba2d86 | 2086 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2087 | { |
2088 | unsigned long flags; | |
dd41f596 | 2089 | int running, on_rq; |
85ba2d86 | 2090 | unsigned long ncsw; |
70b97a7f | 2091 | struct rq *rq; |
1da177e4 | 2092 | |
3a5c359a AK |
2093 | for (;;) { |
2094 | /* | |
2095 | * We do the initial early heuristics without holding | |
2096 | * any task-queue locks at all. We'll only try to get | |
2097 | * the runqueue lock when things look like they will | |
2098 | * work out! | |
2099 | */ | |
2100 | rq = task_rq(p); | |
fa490cfd | 2101 | |
3a5c359a AK |
2102 | /* |
2103 | * If the task is actively running on another CPU | |
2104 | * still, just relax and busy-wait without holding | |
2105 | * any locks. | |
2106 | * | |
2107 | * NOTE! Since we don't hold any locks, it's not | |
2108 | * even sure that "rq" stays as the right runqueue! | |
2109 | * But we don't care, since "task_running()" will | |
2110 | * return false if the runqueue has changed and p | |
2111 | * is actually now running somewhere else! | |
2112 | */ | |
85ba2d86 RM |
2113 | while (task_running(rq, p)) { |
2114 | if (match_state && unlikely(p->state != match_state)) | |
2115 | return 0; | |
3a5c359a | 2116 | cpu_relax(); |
85ba2d86 | 2117 | } |
fa490cfd | 2118 | |
3a5c359a AK |
2119 | /* |
2120 | * Ok, time to look more closely! We need the rq | |
2121 | * lock now, to be *sure*. If we're wrong, we'll | |
2122 | * just go back and repeat. | |
2123 | */ | |
2124 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2125 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2126 | running = task_running(rq, p); |
2127 | on_rq = p->se.on_rq; | |
85ba2d86 | 2128 | ncsw = 0; |
f31e11d8 | 2129 | if (!match_state || p->state == match_state) |
93dcf55f | 2130 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2131 | task_rq_unlock(rq, &flags); |
fa490cfd | 2132 | |
85ba2d86 RM |
2133 | /* |
2134 | * If it changed from the expected state, bail out now. | |
2135 | */ | |
2136 | if (unlikely(!ncsw)) | |
2137 | break; | |
2138 | ||
3a5c359a AK |
2139 | /* |
2140 | * Was it really running after all now that we | |
2141 | * checked with the proper locks actually held? | |
2142 | * | |
2143 | * Oops. Go back and try again.. | |
2144 | */ | |
2145 | if (unlikely(running)) { | |
2146 | cpu_relax(); | |
2147 | continue; | |
2148 | } | |
fa490cfd | 2149 | |
3a5c359a AK |
2150 | /* |
2151 | * It's not enough that it's not actively running, | |
2152 | * it must be off the runqueue _entirely_, and not | |
2153 | * preempted! | |
2154 | * | |
80dd99b3 | 2155 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2156 | * running right now), it's preempted, and we should |
2157 | * yield - it could be a while. | |
2158 | */ | |
2159 | if (unlikely(on_rq)) { | |
2160 | schedule_timeout_uninterruptible(1); | |
2161 | continue; | |
2162 | } | |
fa490cfd | 2163 | |
3a5c359a AK |
2164 | /* |
2165 | * Ahh, all good. It wasn't running, and it wasn't | |
2166 | * runnable, which means that it will never become | |
2167 | * running in the future either. We're all done! | |
2168 | */ | |
2169 | break; | |
2170 | } | |
85ba2d86 RM |
2171 | |
2172 | return ncsw; | |
1da177e4 LT |
2173 | } |
2174 | ||
2175 | /*** | |
2176 | * kick_process - kick a running thread to enter/exit the kernel | |
2177 | * @p: the to-be-kicked thread | |
2178 | * | |
2179 | * Cause a process which is running on another CPU to enter | |
2180 | * kernel-mode, without any delay. (to get signals handled.) | |
2181 | * | |
2182 | * NOTE: this function doesnt have to take the runqueue lock, | |
2183 | * because all it wants to ensure is that the remote task enters | |
2184 | * the kernel. If the IPI races and the task has been migrated | |
2185 | * to another CPU then no harm is done and the purpose has been | |
2186 | * achieved as well. | |
2187 | */ | |
36c8b586 | 2188 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2189 | { |
2190 | int cpu; | |
2191 | ||
2192 | preempt_disable(); | |
2193 | cpu = task_cpu(p); | |
2194 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2195 | smp_send_reschedule(cpu); | |
2196 | preempt_enable(); | |
2197 | } | |
b43e3521 | 2198 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 LT |
2199 | |
2200 | /* | |
2dd73a4f PW |
2201 | * Return a low guess at the load of a migration-source cpu weighted |
2202 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2203 | * |
2204 | * We want to under-estimate the load of migration sources, to | |
2205 | * balance conservatively. | |
2206 | */ | |
a9957449 | 2207 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2208 | { |
70b97a7f | 2209 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2210 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2211 | |
93b75217 | 2212 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2213 | return total; |
b910472d | 2214 | |
dd41f596 | 2215 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2216 | } |
2217 | ||
2218 | /* | |
2dd73a4f PW |
2219 | * Return a high guess at the load of a migration-target cpu weighted |
2220 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2221 | */ |
a9957449 | 2222 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2223 | { |
70b97a7f | 2224 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2225 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2226 | |
93b75217 | 2227 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2228 | return total; |
3b0bd9bc | 2229 | |
dd41f596 | 2230 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2231 | } |
2232 | ||
147cbb4b NP |
2233 | /* |
2234 | * find_idlest_group finds and returns the least busy CPU group within the | |
2235 | * domain. | |
2236 | */ | |
2237 | static struct sched_group * | |
2238 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2239 | { | |
2240 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2241 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2242 | int load_idx = sd->forkexec_idx; | |
2243 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2244 | ||
2245 | do { | |
2246 | unsigned long load, avg_load; | |
2247 | int local_group; | |
2248 | int i; | |
2249 | ||
da5a5522 | 2250 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2251 | if (!cpumask_intersects(sched_group_cpus(group), |
2252 | &p->cpus_allowed)) | |
3a5c359a | 2253 | continue; |
da5a5522 | 2254 | |
758b2cdc RR |
2255 | local_group = cpumask_test_cpu(this_cpu, |
2256 | sched_group_cpus(group)); | |
147cbb4b NP |
2257 | |
2258 | /* Tally up the load of all CPUs in the group */ | |
2259 | avg_load = 0; | |
2260 | ||
758b2cdc | 2261 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2262 | /* Bias balancing toward cpus of our domain */ |
2263 | if (local_group) | |
2264 | load = source_load(i, load_idx); | |
2265 | else | |
2266 | load = target_load(i, load_idx); | |
2267 | ||
2268 | avg_load += load; | |
2269 | } | |
2270 | ||
2271 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2272 | avg_load = sg_div_cpu_power(group, |
2273 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2274 | |
2275 | if (local_group) { | |
2276 | this_load = avg_load; | |
2277 | this = group; | |
2278 | } else if (avg_load < min_load) { | |
2279 | min_load = avg_load; | |
2280 | idlest = group; | |
2281 | } | |
3a5c359a | 2282 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2283 | |
2284 | if (!idlest || 100*this_load < imbalance*min_load) | |
2285 | return NULL; | |
2286 | return idlest; | |
2287 | } | |
2288 | ||
2289 | /* | |
0feaece9 | 2290 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2291 | */ |
95cdf3b7 | 2292 | static int |
758b2cdc | 2293 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2294 | { |
2295 | unsigned long load, min_load = ULONG_MAX; | |
2296 | int idlest = -1; | |
2297 | int i; | |
2298 | ||
da5a5522 | 2299 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2300 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2301 | load = weighted_cpuload(i); |
147cbb4b NP |
2302 | |
2303 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2304 | min_load = load; | |
2305 | idlest = i; | |
2306 | } | |
2307 | } | |
2308 | ||
2309 | return idlest; | |
2310 | } | |
2311 | ||
476d139c NP |
2312 | /* |
2313 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2314 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2315 | * SD_BALANCE_EXEC. | |
2316 | * | |
2317 | * Balance, ie. select the least loaded group. | |
2318 | * | |
2319 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2320 | * | |
2321 | * preempt must be disabled. | |
2322 | */ | |
2323 | static int sched_balance_self(int cpu, int flag) | |
2324 | { | |
2325 | struct task_struct *t = current; | |
2326 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2327 | |
c96d145e | 2328 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2329 | /* |
2330 | * If power savings logic is enabled for a domain, stop there. | |
2331 | */ | |
5c45bf27 SS |
2332 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2333 | break; | |
476d139c NP |
2334 | if (tmp->flags & flag) |
2335 | sd = tmp; | |
c96d145e | 2336 | } |
476d139c | 2337 | |
039a1c41 PZ |
2338 | if (sd) |
2339 | update_shares(sd); | |
2340 | ||
476d139c | 2341 | while (sd) { |
476d139c | 2342 | struct sched_group *group; |
1a848870 SS |
2343 | int new_cpu, weight; |
2344 | ||
2345 | if (!(sd->flags & flag)) { | |
2346 | sd = sd->child; | |
2347 | continue; | |
2348 | } | |
476d139c | 2349 | |
476d139c | 2350 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2351 | if (!group) { |
2352 | sd = sd->child; | |
2353 | continue; | |
2354 | } | |
476d139c | 2355 | |
758b2cdc | 2356 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2357 | if (new_cpu == -1 || new_cpu == cpu) { |
2358 | /* Now try balancing at a lower domain level of cpu */ | |
2359 | sd = sd->child; | |
2360 | continue; | |
2361 | } | |
476d139c | 2362 | |
1a848870 | 2363 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2364 | cpu = new_cpu; |
758b2cdc | 2365 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2366 | sd = NULL; |
476d139c | 2367 | for_each_domain(cpu, tmp) { |
758b2cdc | 2368 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2369 | break; |
2370 | if (tmp->flags & flag) | |
2371 | sd = tmp; | |
2372 | } | |
2373 | /* while loop will break here if sd == NULL */ | |
2374 | } | |
2375 | ||
2376 | return cpu; | |
2377 | } | |
2378 | ||
2379 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2380 | |
0793a61d TG |
2381 | /** |
2382 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2383 | * @p: the task to evaluate | |
2384 | * @func: the function to be called | |
2385 | * @info: the function call argument | |
2386 | * | |
2387 | * Calls the function @func when the task is currently running. This might | |
2388 | * be on the current CPU, which just calls the function directly | |
2389 | */ | |
2390 | void task_oncpu_function_call(struct task_struct *p, | |
2391 | void (*func) (void *info), void *info) | |
2392 | { | |
2393 | int cpu; | |
2394 | ||
2395 | preempt_disable(); | |
2396 | cpu = task_cpu(p); | |
2397 | if (task_curr(p)) | |
2398 | smp_call_function_single(cpu, func, info, 1); | |
2399 | preempt_enable(); | |
2400 | } | |
2401 | ||
1da177e4 LT |
2402 | /*** |
2403 | * try_to_wake_up - wake up a thread | |
2404 | * @p: the to-be-woken-up thread | |
2405 | * @state: the mask of task states that can be woken | |
2406 | * @sync: do a synchronous wakeup? | |
2407 | * | |
2408 | * Put it on the run-queue if it's not already there. The "current" | |
2409 | * thread is always on the run-queue (except when the actual | |
2410 | * re-schedule is in progress), and as such you're allowed to do | |
2411 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2412 | * runnable without the overhead of this. | |
2413 | * | |
2414 | * returns failure only if the task is already active. | |
2415 | */ | |
36c8b586 | 2416 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2417 | { |
cc367732 | 2418 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2419 | unsigned long flags; |
2420 | long old_state; | |
70b97a7f | 2421 | struct rq *rq; |
1da177e4 | 2422 | |
b85d0667 IM |
2423 | if (!sched_feat(SYNC_WAKEUPS)) |
2424 | sync = 0; | |
2425 | ||
2398f2c6 | 2426 | #ifdef CONFIG_SMP |
57310a98 | 2427 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2428 | struct sched_domain *sd; |
2429 | ||
2430 | this_cpu = raw_smp_processor_id(); | |
2431 | cpu = task_cpu(p); | |
2432 | ||
2433 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2434 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2435 | update_shares(sd); |
2436 | break; | |
2437 | } | |
2438 | } | |
2439 | } | |
2440 | #endif | |
2441 | ||
04e2f174 | 2442 | smp_wmb(); |
1da177e4 | 2443 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2444 | update_rq_clock(rq); |
1da177e4 LT |
2445 | old_state = p->state; |
2446 | if (!(old_state & state)) | |
2447 | goto out; | |
2448 | ||
dd41f596 | 2449 | if (p->se.on_rq) |
1da177e4 LT |
2450 | goto out_running; |
2451 | ||
2452 | cpu = task_cpu(p); | |
cc367732 | 2453 | orig_cpu = cpu; |
1da177e4 LT |
2454 | this_cpu = smp_processor_id(); |
2455 | ||
2456 | #ifdef CONFIG_SMP | |
2457 | if (unlikely(task_running(rq, p))) | |
2458 | goto out_activate; | |
2459 | ||
5d2f5a61 DA |
2460 | cpu = p->sched_class->select_task_rq(p, sync); |
2461 | if (cpu != orig_cpu) { | |
2462 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2463 | task_rq_unlock(rq, &flags); |
2464 | /* might preempt at this point */ | |
2465 | rq = task_rq_lock(p, &flags); | |
2466 | old_state = p->state; | |
2467 | if (!(old_state & state)) | |
2468 | goto out; | |
dd41f596 | 2469 | if (p->se.on_rq) |
1da177e4 LT |
2470 | goto out_running; |
2471 | ||
2472 | this_cpu = smp_processor_id(); | |
2473 | cpu = task_cpu(p); | |
2474 | } | |
2475 | ||
e7693a36 GH |
2476 | #ifdef CONFIG_SCHEDSTATS |
2477 | schedstat_inc(rq, ttwu_count); | |
2478 | if (cpu == this_cpu) | |
2479 | schedstat_inc(rq, ttwu_local); | |
2480 | else { | |
2481 | struct sched_domain *sd; | |
2482 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2483 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2484 | schedstat_inc(sd, ttwu_wake_remote); |
2485 | break; | |
2486 | } | |
2487 | } | |
2488 | } | |
6d6bc0ad | 2489 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2490 | |
1da177e4 LT |
2491 | out_activate: |
2492 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2493 | schedstat_inc(p, se.nr_wakeups); |
2494 | if (sync) | |
2495 | schedstat_inc(p, se.nr_wakeups_sync); | |
2496 | if (orig_cpu != cpu) | |
2497 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2498 | if (cpu == this_cpu) | |
2499 | schedstat_inc(p, se.nr_wakeups_local); | |
2500 | else | |
2501 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2502 | activate_task(rq, p, 1); |
1da177e4 LT |
2503 | success = 1; |
2504 | ||
831451ac PZ |
2505 | /* |
2506 | * Only attribute actual wakeups done by this task. | |
2507 | */ | |
2508 | if (!in_interrupt()) { | |
2509 | struct sched_entity *se = ¤t->se; | |
2510 | u64 sample = se->sum_exec_runtime; | |
2511 | ||
2512 | if (se->last_wakeup) | |
2513 | sample -= se->last_wakeup; | |
2514 | else | |
2515 | sample -= se->start_runtime; | |
2516 | update_avg(&se->avg_wakeup, sample); | |
2517 | ||
2518 | se->last_wakeup = se->sum_exec_runtime; | |
2519 | } | |
2520 | ||
1da177e4 | 2521 | out_running: |
468a15bb | 2522 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2523 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2524 | |
1da177e4 | 2525 | p->state = TASK_RUNNING; |
9a897c5a SR |
2526 | #ifdef CONFIG_SMP |
2527 | if (p->sched_class->task_wake_up) | |
2528 | p->sched_class->task_wake_up(rq, p); | |
2529 | #endif | |
1da177e4 LT |
2530 | out: |
2531 | task_rq_unlock(rq, &flags); | |
2532 | ||
2533 | return success; | |
2534 | } | |
2535 | ||
50fa610a DH |
2536 | /** |
2537 | * wake_up_process - Wake up a specific process | |
2538 | * @p: The process to be woken up. | |
2539 | * | |
2540 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2541 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2542 | * running. | |
2543 | * | |
2544 | * It may be assumed that this function implies a write memory barrier before | |
2545 | * changing the task state if and only if any tasks are woken up. | |
2546 | */ | |
7ad5b3a5 | 2547 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2548 | { |
d9514f6c | 2549 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2550 | } |
1da177e4 LT |
2551 | EXPORT_SYMBOL(wake_up_process); |
2552 | ||
7ad5b3a5 | 2553 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2554 | { |
2555 | return try_to_wake_up(p, state, 0); | |
2556 | } | |
2557 | ||
1da177e4 LT |
2558 | /* |
2559 | * Perform scheduler related setup for a newly forked process p. | |
2560 | * p is forked by current. | |
dd41f596 IM |
2561 | * |
2562 | * __sched_fork() is basic setup used by init_idle() too: | |
2563 | */ | |
2564 | static void __sched_fork(struct task_struct *p) | |
2565 | { | |
dd41f596 IM |
2566 | p->se.exec_start = 0; |
2567 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2568 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2569 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2570 | p->se.last_wakeup = 0; |
2571 | p->se.avg_overlap = 0; | |
831451ac PZ |
2572 | p->se.start_runtime = 0; |
2573 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2574 | |
2575 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2576 | p->se.wait_start = 0; |
2577 | p->se.wait_max = 0; | |
2578 | p->se.wait_count = 0; | |
2579 | p->se.wait_sum = 0; | |
2580 | ||
2581 | p->se.sleep_start = 0; | |
2582 | p->se.sleep_max = 0; | |
2583 | p->se.sum_sleep_runtime = 0; | |
2584 | ||
2585 | p->se.block_start = 0; | |
2586 | p->se.block_max = 0; | |
2587 | p->se.exec_max = 0; | |
2588 | p->se.slice_max = 0; | |
2589 | ||
2590 | p->se.nr_migrations_cold = 0; | |
2591 | p->se.nr_failed_migrations_affine = 0; | |
2592 | p->se.nr_failed_migrations_running = 0; | |
2593 | p->se.nr_failed_migrations_hot = 0; | |
2594 | p->se.nr_forced_migrations = 0; | |
2595 | p->se.nr_forced2_migrations = 0; | |
2596 | ||
2597 | p->se.nr_wakeups = 0; | |
2598 | p->se.nr_wakeups_sync = 0; | |
2599 | p->se.nr_wakeups_migrate = 0; | |
2600 | p->se.nr_wakeups_local = 0; | |
2601 | p->se.nr_wakeups_remote = 0; | |
2602 | p->se.nr_wakeups_affine = 0; | |
2603 | p->se.nr_wakeups_affine_attempts = 0; | |
2604 | p->se.nr_wakeups_passive = 0; | |
2605 | p->se.nr_wakeups_idle = 0; | |
2606 | ||
6cfb0d5d | 2607 | #endif |
476d139c | 2608 | |
fa717060 | 2609 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2610 | p->se.on_rq = 0; |
4a55bd5e | 2611 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2612 | |
e107be36 AK |
2613 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2614 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2615 | #endif | |
2616 | ||
1da177e4 LT |
2617 | /* |
2618 | * We mark the process as running here, but have not actually | |
2619 | * inserted it onto the runqueue yet. This guarantees that | |
2620 | * nobody will actually run it, and a signal or other external | |
2621 | * event cannot wake it up and insert it on the runqueue either. | |
2622 | */ | |
2623 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2624 | } |
2625 | ||
2626 | /* | |
2627 | * fork()/clone()-time setup: | |
2628 | */ | |
2629 | void sched_fork(struct task_struct *p, int clone_flags) | |
2630 | { | |
2631 | int cpu = get_cpu(); | |
2632 | ||
2633 | __sched_fork(p); | |
2634 | ||
2635 | #ifdef CONFIG_SMP | |
2636 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2637 | #endif | |
02e4bac2 | 2638 | set_task_cpu(p, cpu); |
b29739f9 IM |
2639 | |
2640 | /* | |
b9dc29e7 | 2641 | * Make sure we do not leak PI boosting priority to the child. |
b29739f9 | 2642 | */ |
b9dc29e7 | 2643 | p->prio = current->normal_prio; |
ca94c442 | 2644 | |
b9dc29e7 MG |
2645 | /* |
2646 | * Revert to default priority/policy on fork if requested. | |
2647 | */ | |
2648 | if (unlikely(p->sched_reset_on_fork)) { | |
2649 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) | |
2650 | p->policy = SCHED_NORMAL; | |
2651 | ||
2652 | if (p->normal_prio < DEFAULT_PRIO) | |
2653 | p->prio = DEFAULT_PRIO; | |
2654 | ||
6c697bdf MG |
2655 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2656 | p->static_prio = NICE_TO_PRIO(0); | |
2657 | set_load_weight(p); | |
2658 | } | |
2659 | ||
b9dc29e7 MG |
2660 | /* |
2661 | * We don't need the reset flag anymore after the fork. It has | |
2662 | * fulfilled its duty: | |
2663 | */ | |
2664 | p->sched_reset_on_fork = 0; | |
2665 | } | |
ca94c442 | 2666 | |
2ddbf952 HS |
2667 | if (!rt_prio(p->prio)) |
2668 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2669 | |
52f17b6c | 2670 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2671 | if (likely(sched_info_on())) |
52f17b6c | 2672 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2673 | #endif |
d6077cb8 | 2674 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2675 | p->oncpu = 0; |
2676 | #endif | |
1da177e4 | 2677 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2678 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2679 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2680 | #endif |
917b627d GH |
2681 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2682 | ||
476d139c | 2683 | put_cpu(); |
1da177e4 LT |
2684 | } |
2685 | ||
2686 | /* | |
2687 | * wake_up_new_task - wake up a newly created task for the first time. | |
2688 | * | |
2689 | * This function will do some initial scheduler statistics housekeeping | |
2690 | * that must be done for every newly created context, then puts the task | |
2691 | * on the runqueue and wakes it. | |
2692 | */ | |
7ad5b3a5 | 2693 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2694 | { |
2695 | unsigned long flags; | |
dd41f596 | 2696 | struct rq *rq; |
1da177e4 LT |
2697 | |
2698 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2699 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2700 | update_rq_clock(rq); |
1da177e4 LT |
2701 | |
2702 | p->prio = effective_prio(p); | |
2703 | ||
b9dca1e0 | 2704 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2705 | activate_task(rq, p, 0); |
1da177e4 | 2706 | } else { |
1da177e4 | 2707 | /* |
dd41f596 IM |
2708 | * Let the scheduling class do new task startup |
2709 | * management (if any): | |
1da177e4 | 2710 | */ |
ee0827d8 | 2711 | p->sched_class->task_new(rq, p); |
c09595f6 | 2712 | inc_nr_running(rq); |
1da177e4 | 2713 | } |
c71dd42d | 2714 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2715 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2716 | #ifdef CONFIG_SMP |
2717 | if (p->sched_class->task_wake_up) | |
2718 | p->sched_class->task_wake_up(rq, p); | |
2719 | #endif | |
dd41f596 | 2720 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2721 | } |
2722 | ||
e107be36 AK |
2723 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2724 | ||
2725 | /** | |
80dd99b3 | 2726 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2727 | * @notifier: notifier struct to register |
e107be36 AK |
2728 | */ |
2729 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2730 | { | |
2731 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2732 | } | |
2733 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2734 | ||
2735 | /** | |
2736 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2737 | * @notifier: notifier struct to unregister |
e107be36 AK |
2738 | * |
2739 | * This is safe to call from within a preemption notifier. | |
2740 | */ | |
2741 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2742 | { | |
2743 | hlist_del(¬ifier->link); | |
2744 | } | |
2745 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2746 | ||
2747 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2748 | { | |
2749 | struct preempt_notifier *notifier; | |
2750 | struct hlist_node *node; | |
2751 | ||
2752 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2753 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2754 | } | |
2755 | ||
2756 | static void | |
2757 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2758 | struct task_struct *next) | |
2759 | { | |
2760 | struct preempt_notifier *notifier; | |
2761 | struct hlist_node *node; | |
2762 | ||
2763 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2764 | notifier->ops->sched_out(notifier, next); | |
2765 | } | |
2766 | ||
6d6bc0ad | 2767 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2768 | |
2769 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2770 | { | |
2771 | } | |
2772 | ||
2773 | static void | |
2774 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2775 | struct task_struct *next) | |
2776 | { | |
2777 | } | |
2778 | ||
6d6bc0ad | 2779 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2780 | |
4866cde0 NP |
2781 | /** |
2782 | * prepare_task_switch - prepare to switch tasks | |
2783 | * @rq: the runqueue preparing to switch | |
421cee29 | 2784 | * @prev: the current task that is being switched out |
4866cde0 NP |
2785 | * @next: the task we are going to switch to. |
2786 | * | |
2787 | * This is called with the rq lock held and interrupts off. It must | |
2788 | * be paired with a subsequent finish_task_switch after the context | |
2789 | * switch. | |
2790 | * | |
2791 | * prepare_task_switch sets up locking and calls architecture specific | |
2792 | * hooks. | |
2793 | */ | |
e107be36 AK |
2794 | static inline void |
2795 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2796 | struct task_struct *next) | |
4866cde0 | 2797 | { |
e107be36 | 2798 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2799 | prepare_lock_switch(rq, next); |
2800 | prepare_arch_switch(next); | |
2801 | } | |
2802 | ||
1da177e4 LT |
2803 | /** |
2804 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2805 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2806 | * @prev: the thread we just switched away from. |
2807 | * | |
4866cde0 NP |
2808 | * finish_task_switch must be called after the context switch, paired |
2809 | * with a prepare_task_switch call before the context switch. | |
2810 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2811 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2812 | * |
2813 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2814 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2815 | * with the lock held can cause deadlocks; see schedule() for |
2816 | * details.) | |
2817 | */ | |
a9957449 | 2818 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2819 | __releases(rq->lock) |
2820 | { | |
1da177e4 | 2821 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2822 | long prev_state; |
967fc046 GH |
2823 | #ifdef CONFIG_SMP |
2824 | int post_schedule = 0; | |
2825 | ||
2826 | if (current->sched_class->needs_post_schedule) | |
2827 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2828 | #endif | |
1da177e4 LT |
2829 | |
2830 | rq->prev_mm = NULL; | |
2831 | ||
2832 | /* | |
2833 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2834 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2835 | * schedule one last time. The schedule call will never return, and |
2836 | * the scheduled task must drop that reference. | |
c394cc9f | 2837 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2838 | * still held, otherwise prev could be scheduled on another cpu, die |
2839 | * there before we look at prev->state, and then the reference would | |
2840 | * be dropped twice. | |
2841 | * Manfred Spraul <manfred@colorfullife.com> | |
2842 | */ | |
55a101f8 | 2843 | prev_state = prev->state; |
4866cde0 | 2844 | finish_arch_switch(prev); |
0793a61d | 2845 | perf_counter_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2846 | finish_lock_switch(rq, prev); |
9a897c5a | 2847 | #ifdef CONFIG_SMP |
967fc046 | 2848 | if (post_schedule) |
9a897c5a SR |
2849 | current->sched_class->post_schedule(rq); |
2850 | #endif | |
e8fa1362 | 2851 | |
e107be36 | 2852 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2853 | if (mm) |
2854 | mmdrop(mm); | |
c394cc9f | 2855 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2856 | /* |
2857 | * Remove function-return probe instances associated with this | |
2858 | * task and put them back on the free list. | |
9761eea8 | 2859 | */ |
c6fd91f0 | 2860 | kprobe_flush_task(prev); |
1da177e4 | 2861 | put_task_struct(prev); |
c6fd91f0 | 2862 | } |
1da177e4 LT |
2863 | } |
2864 | ||
2865 | /** | |
2866 | * schedule_tail - first thing a freshly forked thread must call. | |
2867 | * @prev: the thread we just switched away from. | |
2868 | */ | |
36c8b586 | 2869 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2870 | __releases(rq->lock) |
2871 | { | |
70b97a7f IM |
2872 | struct rq *rq = this_rq(); |
2873 | ||
4866cde0 NP |
2874 | finish_task_switch(rq, prev); |
2875 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2876 | /* In this case, finish_task_switch does not reenable preemption */ | |
2877 | preempt_enable(); | |
2878 | #endif | |
1da177e4 | 2879 | if (current->set_child_tid) |
b488893a | 2880 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2881 | } |
2882 | ||
2883 | /* | |
2884 | * context_switch - switch to the new MM and the new | |
2885 | * thread's register state. | |
2886 | */ | |
dd41f596 | 2887 | static inline void |
70b97a7f | 2888 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2889 | struct task_struct *next) |
1da177e4 | 2890 | { |
dd41f596 | 2891 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2892 | |
e107be36 | 2893 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2894 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2895 | mm = next->mm; |
2896 | oldmm = prev->active_mm; | |
9226d125 ZA |
2897 | /* |
2898 | * For paravirt, this is coupled with an exit in switch_to to | |
2899 | * combine the page table reload and the switch backend into | |
2900 | * one hypercall. | |
2901 | */ | |
224101ed | 2902 | arch_start_context_switch(prev); |
9226d125 | 2903 | |
dd41f596 | 2904 | if (unlikely(!mm)) { |
1da177e4 LT |
2905 | next->active_mm = oldmm; |
2906 | atomic_inc(&oldmm->mm_count); | |
2907 | enter_lazy_tlb(oldmm, next); | |
2908 | } else | |
2909 | switch_mm(oldmm, mm, next); | |
2910 | ||
dd41f596 | 2911 | if (unlikely(!prev->mm)) { |
1da177e4 | 2912 | prev->active_mm = NULL; |
1da177e4 LT |
2913 | rq->prev_mm = oldmm; |
2914 | } | |
3a5f5e48 IM |
2915 | /* |
2916 | * Since the runqueue lock will be released by the next | |
2917 | * task (which is an invalid locking op but in the case | |
2918 | * of the scheduler it's an obvious special-case), so we | |
2919 | * do an early lockdep release here: | |
2920 | */ | |
2921 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2922 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2923 | #endif |
1da177e4 LT |
2924 | |
2925 | /* Here we just switch the register state and the stack. */ | |
2926 | switch_to(prev, next, prev); | |
2927 | ||
dd41f596 IM |
2928 | barrier(); |
2929 | /* | |
2930 | * this_rq must be evaluated again because prev may have moved | |
2931 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2932 | * frame will be invalid. | |
2933 | */ | |
2934 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2935 | } |
2936 | ||
2937 | /* | |
2938 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2939 | * | |
2940 | * externally visible scheduler statistics: current number of runnable | |
2941 | * threads, current number of uninterruptible-sleeping threads, total | |
2942 | * number of context switches performed since bootup. | |
2943 | */ | |
2944 | unsigned long nr_running(void) | |
2945 | { | |
2946 | unsigned long i, sum = 0; | |
2947 | ||
2948 | for_each_online_cpu(i) | |
2949 | sum += cpu_rq(i)->nr_running; | |
2950 | ||
2951 | return sum; | |
2952 | } | |
2953 | ||
2954 | unsigned long nr_uninterruptible(void) | |
2955 | { | |
2956 | unsigned long i, sum = 0; | |
2957 | ||
0a945022 | 2958 | for_each_possible_cpu(i) |
1da177e4 LT |
2959 | sum += cpu_rq(i)->nr_uninterruptible; |
2960 | ||
2961 | /* | |
2962 | * Since we read the counters lockless, it might be slightly | |
2963 | * inaccurate. Do not allow it to go below zero though: | |
2964 | */ | |
2965 | if (unlikely((long)sum < 0)) | |
2966 | sum = 0; | |
2967 | ||
2968 | return sum; | |
2969 | } | |
2970 | ||
2971 | unsigned long long nr_context_switches(void) | |
2972 | { | |
cc94abfc SR |
2973 | int i; |
2974 | unsigned long long sum = 0; | |
1da177e4 | 2975 | |
0a945022 | 2976 | for_each_possible_cpu(i) |
1da177e4 LT |
2977 | sum += cpu_rq(i)->nr_switches; |
2978 | ||
2979 | return sum; | |
2980 | } | |
2981 | ||
2982 | unsigned long nr_iowait(void) | |
2983 | { | |
2984 | unsigned long i, sum = 0; | |
2985 | ||
0a945022 | 2986 | for_each_possible_cpu(i) |
1da177e4 LT |
2987 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2988 | ||
2989 | return sum; | |
2990 | } | |
2991 | ||
dce48a84 TG |
2992 | /* Variables and functions for calc_load */ |
2993 | static atomic_long_t calc_load_tasks; | |
2994 | static unsigned long calc_load_update; | |
2995 | unsigned long avenrun[3]; | |
2996 | EXPORT_SYMBOL(avenrun); | |
2997 | ||
2d02494f TG |
2998 | /** |
2999 | * get_avenrun - get the load average array | |
3000 | * @loads: pointer to dest load array | |
3001 | * @offset: offset to add | |
3002 | * @shift: shift count to shift the result left | |
3003 | * | |
3004 | * These values are estimates at best, so no need for locking. | |
3005 | */ | |
3006 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3007 | { | |
3008 | loads[0] = (avenrun[0] + offset) << shift; | |
3009 | loads[1] = (avenrun[1] + offset) << shift; | |
3010 | loads[2] = (avenrun[2] + offset) << shift; | |
3011 | } | |
3012 | ||
dce48a84 TG |
3013 | static unsigned long |
3014 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3015 | { |
dce48a84 TG |
3016 | load *= exp; |
3017 | load += active * (FIXED_1 - exp); | |
3018 | return load >> FSHIFT; | |
3019 | } | |
db1b1fef | 3020 | |
dce48a84 TG |
3021 | /* |
3022 | * calc_load - update the avenrun load estimates 10 ticks after the | |
3023 | * CPUs have updated calc_load_tasks. | |
3024 | */ | |
3025 | void calc_global_load(void) | |
3026 | { | |
3027 | unsigned long upd = calc_load_update + 10; | |
3028 | long active; | |
3029 | ||
3030 | if (time_before(jiffies, upd)) | |
3031 | return; | |
db1b1fef | 3032 | |
dce48a84 TG |
3033 | active = atomic_long_read(&calc_load_tasks); |
3034 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 3035 | |
dce48a84 TG |
3036 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3037 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3038 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
3039 | ||
3040 | calc_load_update += LOAD_FREQ; | |
3041 | } | |
3042 | ||
3043 | /* | |
3044 | * Either called from update_cpu_load() or from a cpu going idle | |
3045 | */ | |
3046 | static void calc_load_account_active(struct rq *this_rq) | |
3047 | { | |
3048 | long nr_active, delta; | |
3049 | ||
3050 | nr_active = this_rq->nr_running; | |
3051 | nr_active += (long) this_rq->nr_uninterruptible; | |
3052 | ||
3053 | if (nr_active != this_rq->calc_load_active) { | |
3054 | delta = nr_active - this_rq->calc_load_active; | |
3055 | this_rq->calc_load_active = nr_active; | |
3056 | atomic_long_add(delta, &calc_load_tasks); | |
3057 | } | |
db1b1fef JS |
3058 | } |
3059 | ||
23a185ca PM |
3060 | /* |
3061 | * Externally visible per-cpu scheduler statistics: | |
23a185ca PM |
3062 | * cpu_nr_migrations(cpu) - number of migrations into that cpu |
3063 | */ | |
23a185ca PM |
3064 | u64 cpu_nr_migrations(int cpu) |
3065 | { | |
3066 | return cpu_rq(cpu)->nr_migrations_in; | |
3067 | } | |
3068 | ||
48f24c4d | 3069 | /* |
dd41f596 IM |
3070 | * Update rq->cpu_load[] statistics. This function is usually called every |
3071 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3072 | */ |
dd41f596 | 3073 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3074 | { |
495eca49 | 3075 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3076 | int i, scale; |
3077 | ||
3078 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3079 | |
3080 | /* Update our load: */ | |
3081 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3082 | unsigned long old_load, new_load; | |
3083 | ||
3084 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3085 | ||
3086 | old_load = this_rq->cpu_load[i]; | |
3087 | new_load = this_load; | |
a25707f3 IM |
3088 | /* |
3089 | * Round up the averaging division if load is increasing. This | |
3090 | * prevents us from getting stuck on 9 if the load is 10, for | |
3091 | * example. | |
3092 | */ | |
3093 | if (new_load > old_load) | |
3094 | new_load += scale-1; | |
dd41f596 IM |
3095 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3096 | } | |
dce48a84 TG |
3097 | |
3098 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3099 | this_rq->calc_load_update += LOAD_FREQ; | |
3100 | calc_load_account_active(this_rq); | |
3101 | } | |
48f24c4d IM |
3102 | } |
3103 | ||
dd41f596 IM |
3104 | #ifdef CONFIG_SMP |
3105 | ||
1da177e4 LT |
3106 | /* |
3107 | * double_rq_lock - safely lock two runqueues | |
3108 | * | |
3109 | * Note this does not disable interrupts like task_rq_lock, | |
3110 | * you need to do so manually before calling. | |
3111 | */ | |
70b97a7f | 3112 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3113 | __acquires(rq1->lock) |
3114 | __acquires(rq2->lock) | |
3115 | { | |
054b9108 | 3116 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3117 | if (rq1 == rq2) { |
3118 | spin_lock(&rq1->lock); | |
3119 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3120 | } else { | |
c96d145e | 3121 | if (rq1 < rq2) { |
1da177e4 | 3122 | spin_lock(&rq1->lock); |
5e710e37 | 3123 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3124 | } else { |
3125 | spin_lock(&rq2->lock); | |
5e710e37 | 3126 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3127 | } |
3128 | } | |
6e82a3be IM |
3129 | update_rq_clock(rq1); |
3130 | update_rq_clock(rq2); | |
1da177e4 LT |
3131 | } |
3132 | ||
3133 | /* | |
3134 | * double_rq_unlock - safely unlock two runqueues | |
3135 | * | |
3136 | * Note this does not restore interrupts like task_rq_unlock, | |
3137 | * you need to do so manually after calling. | |
3138 | */ | |
70b97a7f | 3139 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3140 | __releases(rq1->lock) |
3141 | __releases(rq2->lock) | |
3142 | { | |
3143 | spin_unlock(&rq1->lock); | |
3144 | if (rq1 != rq2) | |
3145 | spin_unlock(&rq2->lock); | |
3146 | else | |
3147 | __release(rq2->lock); | |
3148 | } | |
3149 | ||
1da177e4 LT |
3150 | /* |
3151 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3152 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3153 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3154 | * the cpu_allowed mask is restored. |
3155 | */ | |
36c8b586 | 3156 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3157 | { |
70b97a7f | 3158 | struct migration_req req; |
1da177e4 | 3159 | unsigned long flags; |
70b97a7f | 3160 | struct rq *rq; |
1da177e4 LT |
3161 | |
3162 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3163 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3164 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3165 | goto out; |
3166 | ||
3167 | /* force the process onto the specified CPU */ | |
3168 | if (migrate_task(p, dest_cpu, &req)) { | |
3169 | /* Need to wait for migration thread (might exit: take ref). */ | |
3170 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3171 | |
1da177e4 LT |
3172 | get_task_struct(mt); |
3173 | task_rq_unlock(rq, &flags); | |
3174 | wake_up_process(mt); | |
3175 | put_task_struct(mt); | |
3176 | wait_for_completion(&req.done); | |
36c8b586 | 3177 | |
1da177e4 LT |
3178 | return; |
3179 | } | |
3180 | out: | |
3181 | task_rq_unlock(rq, &flags); | |
3182 | } | |
3183 | ||
3184 | /* | |
476d139c NP |
3185 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3186 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3187 | */ |
3188 | void sched_exec(void) | |
3189 | { | |
1da177e4 | 3190 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3191 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3192 | put_cpu(); |
476d139c NP |
3193 | if (new_cpu != this_cpu) |
3194 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3195 | } |
3196 | ||
3197 | /* | |
3198 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3199 | * Both runqueues must be locked. | |
3200 | */ | |
dd41f596 IM |
3201 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3202 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3203 | { |
2e1cb74a | 3204 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3205 | set_task_cpu(p, this_cpu); |
dd41f596 | 3206 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3207 | /* |
3208 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3209 | * to be always true for them. | |
3210 | */ | |
15afe09b | 3211 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3212 | } |
3213 | ||
3214 | /* | |
3215 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3216 | */ | |
858119e1 | 3217 | static |
70b97a7f | 3218 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3219 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3220 | int *all_pinned) |
1da177e4 | 3221 | { |
708dc512 | 3222 | int tsk_cache_hot = 0; |
1da177e4 LT |
3223 | /* |
3224 | * We do not migrate tasks that are: | |
3225 | * 1) running (obviously), or | |
3226 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3227 | * 3) are cache-hot on their current CPU. | |
3228 | */ | |
96f874e2 | 3229 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3230 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3231 | return 0; |
cc367732 | 3232 | } |
81026794 NP |
3233 | *all_pinned = 0; |
3234 | ||
cc367732 IM |
3235 | if (task_running(rq, p)) { |
3236 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3237 | return 0; |
cc367732 | 3238 | } |
1da177e4 | 3239 | |
da84d961 IM |
3240 | /* |
3241 | * Aggressive migration if: | |
3242 | * 1) task is cache cold, or | |
3243 | * 2) too many balance attempts have failed. | |
3244 | */ | |
3245 | ||
708dc512 LH |
3246 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3247 | if (!tsk_cache_hot || | |
3248 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3249 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3250 | if (tsk_cache_hot) { |
da84d961 | 3251 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3252 | schedstat_inc(p, se.nr_forced_migrations); |
3253 | } | |
da84d961 IM |
3254 | #endif |
3255 | return 1; | |
3256 | } | |
3257 | ||
708dc512 | 3258 | if (tsk_cache_hot) { |
cc367732 | 3259 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3260 | return 0; |
cc367732 | 3261 | } |
1da177e4 LT |
3262 | return 1; |
3263 | } | |
3264 | ||
e1d1484f PW |
3265 | static unsigned long |
3266 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3267 | unsigned long max_load_move, struct sched_domain *sd, | |
3268 | enum cpu_idle_type idle, int *all_pinned, | |
3269 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3270 | { |
051c6764 | 3271 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3272 | struct task_struct *p; |
3273 | long rem_load_move = max_load_move; | |
1da177e4 | 3274 | |
e1d1484f | 3275 | if (max_load_move == 0) |
1da177e4 LT |
3276 | goto out; |
3277 | ||
81026794 NP |
3278 | pinned = 1; |
3279 | ||
1da177e4 | 3280 | /* |
dd41f596 | 3281 | * Start the load-balancing iterator: |
1da177e4 | 3282 | */ |
dd41f596 IM |
3283 | p = iterator->start(iterator->arg); |
3284 | next: | |
b82d9fdd | 3285 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3286 | goto out; |
051c6764 PZ |
3287 | |
3288 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3289 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3290 | p = iterator->next(iterator->arg); |
3291 | goto next; | |
1da177e4 LT |
3292 | } |
3293 | ||
dd41f596 | 3294 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3295 | pulled++; |
dd41f596 | 3296 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3297 | |
7e96fa58 GH |
3298 | #ifdef CONFIG_PREEMPT |
3299 | /* | |
3300 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3301 | * will stop after the first task is pulled to minimize the critical | |
3302 | * section. | |
3303 | */ | |
3304 | if (idle == CPU_NEWLY_IDLE) | |
3305 | goto out; | |
3306 | #endif | |
3307 | ||
2dd73a4f | 3308 | /* |
b82d9fdd | 3309 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3310 | */ |
e1d1484f | 3311 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3312 | if (p->prio < *this_best_prio) |
3313 | *this_best_prio = p->prio; | |
dd41f596 IM |
3314 | p = iterator->next(iterator->arg); |
3315 | goto next; | |
1da177e4 LT |
3316 | } |
3317 | out: | |
3318 | /* | |
e1d1484f | 3319 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3320 | * so we can safely collect pull_task() stats here rather than |
3321 | * inside pull_task(). | |
3322 | */ | |
3323 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3324 | |
3325 | if (all_pinned) | |
3326 | *all_pinned = pinned; | |
e1d1484f PW |
3327 | |
3328 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3329 | } |
3330 | ||
dd41f596 | 3331 | /* |
43010659 PW |
3332 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3333 | * this_rq, as part of a balancing operation within domain "sd". | |
3334 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3335 | * |
3336 | * Called with both runqueues locked. | |
3337 | */ | |
3338 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3339 | unsigned long max_load_move, |
dd41f596 IM |
3340 | struct sched_domain *sd, enum cpu_idle_type idle, |
3341 | int *all_pinned) | |
3342 | { | |
5522d5d5 | 3343 | const struct sched_class *class = sched_class_highest; |
43010659 | 3344 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3345 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3346 | |
3347 | do { | |
43010659 PW |
3348 | total_load_moved += |
3349 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3350 | max_load_move - total_load_moved, |
a4ac01c3 | 3351 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3352 | class = class->next; |
c4acb2c0 | 3353 | |
7e96fa58 GH |
3354 | #ifdef CONFIG_PREEMPT |
3355 | /* | |
3356 | * NEWIDLE balancing is a source of latency, so preemptible | |
3357 | * kernels will stop after the first task is pulled to minimize | |
3358 | * the critical section. | |
3359 | */ | |
c4acb2c0 GH |
3360 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3361 | break; | |
7e96fa58 | 3362 | #endif |
43010659 | 3363 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3364 | |
43010659 PW |
3365 | return total_load_moved > 0; |
3366 | } | |
3367 | ||
e1d1484f PW |
3368 | static int |
3369 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3370 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3371 | struct rq_iterator *iterator) | |
3372 | { | |
3373 | struct task_struct *p = iterator->start(iterator->arg); | |
3374 | int pinned = 0; | |
3375 | ||
3376 | while (p) { | |
3377 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3378 | pull_task(busiest, p, this_rq, this_cpu); | |
3379 | /* | |
3380 | * Right now, this is only the second place pull_task() | |
3381 | * is called, so we can safely collect pull_task() | |
3382 | * stats here rather than inside pull_task(). | |
3383 | */ | |
3384 | schedstat_inc(sd, lb_gained[idle]); | |
3385 | ||
3386 | return 1; | |
3387 | } | |
3388 | p = iterator->next(iterator->arg); | |
3389 | } | |
3390 | ||
3391 | return 0; | |
3392 | } | |
3393 | ||
43010659 PW |
3394 | /* |
3395 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3396 | * part of active balancing operations within "domain". | |
3397 | * Returns 1 if successful and 0 otherwise. | |
3398 | * | |
3399 | * Called with both runqueues locked. | |
3400 | */ | |
3401 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3402 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3403 | { | |
5522d5d5 | 3404 | const struct sched_class *class; |
43010659 PW |
3405 | |
3406 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3407 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3408 | return 1; |
3409 | ||
3410 | return 0; | |
dd41f596 | 3411 | } |
67bb6c03 | 3412 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3413 | /* |
222d656d GS |
3414 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3415 | * during load balancing. | |
1da177e4 | 3416 | */ |
222d656d GS |
3417 | struct sd_lb_stats { |
3418 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3419 | struct sched_group *this; /* Local group in this sd */ | |
3420 | unsigned long total_load; /* Total load of all groups in sd */ | |
3421 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3422 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3423 | ||
3424 | /** Statistics of this group */ | |
3425 | unsigned long this_load; | |
3426 | unsigned long this_load_per_task; | |
3427 | unsigned long this_nr_running; | |
3428 | ||
3429 | /* Statistics of the busiest group */ | |
3430 | unsigned long max_load; | |
3431 | unsigned long busiest_load_per_task; | |
3432 | unsigned long busiest_nr_running; | |
3433 | ||
3434 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3435 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3436 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3437 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3438 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3439 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3440 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3441 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3442 | #endif |
222d656d | 3443 | }; |
1da177e4 | 3444 | |
d5ac537e | 3445 | /* |
381be78f GS |
3446 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3447 | */ | |
3448 | struct sg_lb_stats { | |
3449 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3450 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3451 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3452 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3453 | unsigned long group_capacity; | |
3454 | int group_imb; /* Is there an imbalance in the group ? */ | |
3455 | }; | |
408ed066 | 3456 | |
67bb6c03 GS |
3457 | /** |
3458 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3459 | * @group: The group whose first cpu is to be returned. | |
3460 | */ | |
3461 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3462 | { | |
3463 | return cpumask_first(sched_group_cpus(group)); | |
3464 | } | |
3465 | ||
3466 | /** | |
3467 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3468 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3469 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3470 | */ | |
3471 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3472 | enum cpu_idle_type idle) | |
3473 | { | |
3474 | int load_idx; | |
3475 | ||
3476 | switch (idle) { | |
3477 | case CPU_NOT_IDLE: | |
7897986b | 3478 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3479 | break; |
3480 | ||
3481 | case CPU_NEWLY_IDLE: | |
7897986b | 3482 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3483 | break; |
3484 | default: | |
7897986b | 3485 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3486 | break; |
3487 | } | |
1da177e4 | 3488 | |
67bb6c03 GS |
3489 | return load_idx; |
3490 | } | |
1da177e4 | 3491 | |
1da177e4 | 3492 | |
c071df18 GS |
3493 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3494 | /** | |
3495 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3496 | * the given sched_domain, during load balancing. | |
3497 | * | |
3498 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3499 | * @sds: Variable containing the statistics for sd. | |
3500 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3501 | */ | |
3502 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3503 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3504 | { | |
3505 | /* | |
3506 | * Busy processors will not participate in power savings | |
3507 | * balance. | |
3508 | */ | |
3509 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3510 | sds->power_savings_balance = 0; | |
3511 | else { | |
3512 | sds->power_savings_balance = 1; | |
3513 | sds->min_nr_running = ULONG_MAX; | |
3514 | sds->leader_nr_running = 0; | |
3515 | } | |
3516 | } | |
783609c6 | 3517 | |
c071df18 GS |
3518 | /** |
3519 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3520 | * sched_domain while performing load balancing. | |
3521 | * | |
3522 | * @group: sched_group belonging to the sched_domain under consideration. | |
3523 | * @sds: Variable containing the statistics of the sched_domain | |
3524 | * @local_group: Does group contain the CPU for which we're performing | |
3525 | * load balancing ? | |
3526 | * @sgs: Variable containing the statistics of the group. | |
3527 | */ | |
3528 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3529 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3530 | { | |
408ed066 | 3531 | |
c071df18 GS |
3532 | if (!sds->power_savings_balance) |
3533 | return; | |
1da177e4 | 3534 | |
c071df18 GS |
3535 | /* |
3536 | * If the local group is idle or completely loaded | |
3537 | * no need to do power savings balance at this domain | |
3538 | */ | |
3539 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3540 | !sds->this_nr_running)) | |
3541 | sds->power_savings_balance = 0; | |
2dd73a4f | 3542 | |
c071df18 GS |
3543 | /* |
3544 | * If a group is already running at full capacity or idle, | |
3545 | * don't include that group in power savings calculations | |
3546 | */ | |
3547 | if (!sds->power_savings_balance || | |
3548 | sgs->sum_nr_running >= sgs->group_capacity || | |
3549 | !sgs->sum_nr_running) | |
3550 | return; | |
5969fe06 | 3551 | |
c071df18 GS |
3552 | /* |
3553 | * Calculate the group which has the least non-idle load. | |
3554 | * This is the group from where we need to pick up the load | |
3555 | * for saving power | |
3556 | */ | |
3557 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3558 | (sgs->sum_nr_running == sds->min_nr_running && | |
3559 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3560 | sds->group_min = group; | |
3561 | sds->min_nr_running = sgs->sum_nr_running; | |
3562 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3563 | sgs->sum_nr_running; | |
3564 | } | |
783609c6 | 3565 | |
c071df18 GS |
3566 | /* |
3567 | * Calculate the group which is almost near its | |
3568 | * capacity but still has some space to pick up some load | |
3569 | * from other group and save more power | |
3570 | */ | |
3571 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3572 | return; | |
1da177e4 | 3573 | |
c071df18 GS |
3574 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3575 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3576 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3577 | sds->group_leader = group; | |
3578 | sds->leader_nr_running = sgs->sum_nr_running; | |
3579 | } | |
3580 | } | |
408ed066 | 3581 | |
c071df18 | 3582 | /** |
d5ac537e | 3583 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3584 | * @sds: Variable containing the statistics of the sched_domain |
3585 | * under consideration. | |
3586 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3587 | * @imbalance: Variable to store the imbalance. | |
3588 | * | |
d5ac537e RD |
3589 | * Description: |
3590 | * Check if we have potential to perform some power-savings balance. | |
3591 | * If yes, set the busiest group to be the least loaded group in the | |
3592 | * sched_domain, so that it's CPUs can be put to idle. | |
3593 | * | |
c071df18 GS |
3594 | * Returns 1 if there is potential to perform power-savings balance. |
3595 | * Else returns 0. | |
3596 | */ | |
3597 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3598 | int this_cpu, unsigned long *imbalance) | |
3599 | { | |
3600 | if (!sds->power_savings_balance) | |
3601 | return 0; | |
1da177e4 | 3602 | |
c071df18 GS |
3603 | if (sds->this != sds->group_leader || |
3604 | sds->group_leader == sds->group_min) | |
3605 | return 0; | |
783609c6 | 3606 | |
c071df18 GS |
3607 | *imbalance = sds->min_load_per_task; |
3608 | sds->busiest = sds->group_min; | |
1da177e4 | 3609 | |
c071df18 GS |
3610 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3611 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3612 | group_first_cpu(sds->group_leader); | |
3613 | } | |
3614 | ||
3615 | return 1; | |
1da177e4 | 3616 | |
c071df18 GS |
3617 | } |
3618 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3619 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3620 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3621 | { | |
3622 | return; | |
3623 | } | |
408ed066 | 3624 | |
c071df18 GS |
3625 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3626 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3627 | { | |
3628 | return; | |
3629 | } | |
3630 | ||
3631 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3632 | int this_cpu, unsigned long *imbalance) | |
3633 | { | |
3634 | return 0; | |
3635 | } | |
3636 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3637 | ||
3638 | ||
1f8c553d GS |
3639 | /** |
3640 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3641 | * @group: sched_group whose statistics are to be updated. | |
3642 | * @this_cpu: Cpu for which load balance is currently performed. | |
3643 | * @idle: Idle status of this_cpu | |
3644 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3645 | * @sd_idle: Idle status of the sched_domain containing group. | |
3646 | * @local_group: Does group contain this_cpu. | |
3647 | * @cpus: Set of cpus considered for load balancing. | |
3648 | * @balance: Should we balance. | |
3649 | * @sgs: variable to hold the statistics for this group. | |
3650 | */ | |
3651 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3652 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3653 | int local_group, const struct cpumask *cpus, | |
3654 | int *balance, struct sg_lb_stats *sgs) | |
3655 | { | |
3656 | unsigned long load, max_cpu_load, min_cpu_load; | |
3657 | int i; | |
3658 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3659 | unsigned long sum_avg_load_per_task; | |
3660 | unsigned long avg_load_per_task; | |
3661 | ||
3662 | if (local_group) | |
3663 | balance_cpu = group_first_cpu(group); | |
3664 | ||
3665 | /* Tally up the load of all CPUs in the group */ | |
3666 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3667 | max_cpu_load = 0; | |
3668 | min_cpu_load = ~0UL; | |
408ed066 | 3669 | |
1f8c553d GS |
3670 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3671 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3672 | |
1f8c553d GS |
3673 | if (*sd_idle && rq->nr_running) |
3674 | *sd_idle = 0; | |
5c45bf27 | 3675 | |
1f8c553d | 3676 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3677 | if (local_group) { |
1f8c553d GS |
3678 | if (idle_cpu(i) && !first_idle_cpu) { |
3679 | first_idle_cpu = 1; | |
3680 | balance_cpu = i; | |
3681 | } | |
3682 | ||
3683 | load = target_load(i, load_idx); | |
3684 | } else { | |
3685 | load = source_load(i, load_idx); | |
3686 | if (load > max_cpu_load) | |
3687 | max_cpu_load = load; | |
3688 | if (min_cpu_load > load) | |
3689 | min_cpu_load = load; | |
1da177e4 | 3690 | } |
5c45bf27 | 3691 | |
1f8c553d GS |
3692 | sgs->group_load += load; |
3693 | sgs->sum_nr_running += rq->nr_running; | |
3694 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3695 | |
1f8c553d GS |
3696 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3697 | } | |
5c45bf27 | 3698 | |
1f8c553d GS |
3699 | /* |
3700 | * First idle cpu or the first cpu(busiest) in this sched group | |
3701 | * is eligible for doing load balancing at this and above | |
3702 | * domains. In the newly idle case, we will allow all the cpu's | |
3703 | * to do the newly idle load balance. | |
3704 | */ | |
3705 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3706 | balance_cpu != this_cpu && balance) { | |
3707 | *balance = 0; | |
3708 | return; | |
3709 | } | |
5c45bf27 | 3710 | |
1f8c553d GS |
3711 | /* Adjust by relative CPU power of the group */ |
3712 | sgs->avg_load = sg_div_cpu_power(group, | |
3713 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3714 | |
1f8c553d GS |
3715 | |
3716 | /* | |
3717 | * Consider the group unbalanced when the imbalance is larger | |
3718 | * than the average weight of two tasks. | |
3719 | * | |
3720 | * APZ: with cgroup the avg task weight can vary wildly and | |
3721 | * might not be a suitable number - should we keep a | |
3722 | * normalized nr_running number somewhere that negates | |
3723 | * the hierarchy? | |
3724 | */ | |
3725 | avg_load_per_task = sg_div_cpu_power(group, | |
3726 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3727 | ||
3728 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3729 | sgs->group_imb = 1; | |
3730 | ||
3731 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3732 | ||
3733 | } | |
dd41f596 | 3734 | |
37abe198 GS |
3735 | /** |
3736 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3737 | * @sd: sched_domain whose statistics are to be updated. | |
3738 | * @this_cpu: Cpu for which load balance is currently performed. | |
3739 | * @idle: Idle status of this_cpu | |
3740 | * @sd_idle: Idle status of the sched_domain containing group. | |
3741 | * @cpus: Set of cpus considered for load balancing. | |
3742 | * @balance: Should we balance. | |
3743 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3744 | */ |
37abe198 GS |
3745 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3746 | enum cpu_idle_type idle, int *sd_idle, | |
3747 | const struct cpumask *cpus, int *balance, | |
3748 | struct sd_lb_stats *sds) | |
1da177e4 | 3749 | { |
222d656d | 3750 | struct sched_group *group = sd->groups; |
37abe198 | 3751 | struct sg_lb_stats sgs; |
222d656d GS |
3752 | int load_idx; |
3753 | ||
c071df18 | 3754 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3755 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3756 | |
3757 | do { | |
1da177e4 | 3758 | int local_group; |
1da177e4 | 3759 | |
758b2cdc RR |
3760 | local_group = cpumask_test_cpu(this_cpu, |
3761 | sched_group_cpus(group)); | |
381be78f | 3762 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3763 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3764 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3765 | |
37abe198 GS |
3766 | if (local_group && balance && !(*balance)) |
3767 | return; | |
783609c6 | 3768 | |
37abe198 GS |
3769 | sds->total_load += sgs.group_load; |
3770 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3771 | |
1da177e4 | 3772 | if (local_group) { |
37abe198 GS |
3773 | sds->this_load = sgs.avg_load; |
3774 | sds->this = group; | |
3775 | sds->this_nr_running = sgs.sum_nr_running; | |
3776 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3777 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3778 | (sgs.sum_nr_running > sgs.group_capacity || |
3779 | sgs.group_imb)) { | |
37abe198 GS |
3780 | sds->max_load = sgs.avg_load; |
3781 | sds->busiest = group; | |
3782 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3783 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3784 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3785 | } |
5c45bf27 | 3786 | |
c071df18 | 3787 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3788 | group = group->next; |
3789 | } while (group != sd->groups); | |
3790 | ||
37abe198 | 3791 | } |
1da177e4 | 3792 | |
2e6f44ae GS |
3793 | /** |
3794 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3795 | * amongst the groups of a sched_domain, during |
3796 | * load balancing. | |
2e6f44ae GS |
3797 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3798 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3799 | * @imbalance: Variable to store the imbalance. | |
3800 | */ | |
3801 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3802 | int this_cpu, unsigned long *imbalance) | |
3803 | { | |
3804 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3805 | unsigned int imbn = 2; | |
3806 | ||
3807 | if (sds->this_nr_running) { | |
3808 | sds->this_load_per_task /= sds->this_nr_running; | |
3809 | if (sds->busiest_load_per_task > | |
3810 | sds->this_load_per_task) | |
3811 | imbn = 1; | |
3812 | } else | |
3813 | sds->this_load_per_task = | |
3814 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3815 | |
2e6f44ae GS |
3816 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3817 | sds->busiest_load_per_task * imbn) { | |
3818 | *imbalance = sds->busiest_load_per_task; | |
3819 | return; | |
3820 | } | |
908a7c1b | 3821 | |
1da177e4 | 3822 | /* |
2e6f44ae GS |
3823 | * OK, we don't have enough imbalance to justify moving tasks, |
3824 | * however we may be able to increase total CPU power used by | |
3825 | * moving them. | |
1da177e4 | 3826 | */ |
2dd73a4f | 3827 | |
2e6f44ae GS |
3828 | pwr_now += sds->busiest->__cpu_power * |
3829 | min(sds->busiest_load_per_task, sds->max_load); | |
3830 | pwr_now += sds->this->__cpu_power * | |
3831 | min(sds->this_load_per_task, sds->this_load); | |
3832 | pwr_now /= SCHED_LOAD_SCALE; | |
3833 | ||
3834 | /* Amount of load we'd subtract */ | |
3835 | tmp = sg_div_cpu_power(sds->busiest, | |
3836 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3837 | if (sds->max_load > tmp) | |
3838 | pwr_move += sds->busiest->__cpu_power * | |
3839 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3840 | ||
3841 | /* Amount of load we'd add */ | |
3842 | if (sds->max_load * sds->busiest->__cpu_power < | |
3843 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3844 | tmp = sg_div_cpu_power(sds->this, | |
3845 | sds->max_load * sds->busiest->__cpu_power); | |
3846 | else | |
3847 | tmp = sg_div_cpu_power(sds->this, | |
3848 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3849 | pwr_move += sds->this->__cpu_power * | |
3850 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3851 | pwr_move /= SCHED_LOAD_SCALE; | |
3852 | ||
3853 | /* Move if we gain throughput */ | |
3854 | if (pwr_move > pwr_now) | |
3855 | *imbalance = sds->busiest_load_per_task; | |
3856 | } | |
dbc523a3 GS |
3857 | |
3858 | /** | |
3859 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3860 | * groups of a given sched_domain during load balance. | |
3861 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3862 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3863 | * @imbalance: The variable to store the imbalance. | |
3864 | */ | |
3865 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3866 | unsigned long *imbalance) | |
3867 | { | |
3868 | unsigned long max_pull; | |
2dd73a4f PW |
3869 | /* |
3870 | * In the presence of smp nice balancing, certain scenarios can have | |
3871 | * max load less than avg load(as we skip the groups at or below | |
3872 | * its cpu_power, while calculating max_load..) | |
3873 | */ | |
dbc523a3 | 3874 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3875 | *imbalance = 0; |
dbc523a3 | 3876 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3877 | } |
0c117f1b SS |
3878 | |
3879 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3880 | max_pull = min(sds->max_load - sds->avg_load, |
3881 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3882 | |
1da177e4 | 3883 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
3884 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
3885 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
3886 | / SCHED_LOAD_SCALE; |
3887 | ||
2dd73a4f PW |
3888 | /* |
3889 | * if *imbalance is less than the average load per runnable task | |
3890 | * there is no gaurantee that any tasks will be moved so we'll have | |
3891 | * a think about bumping its value to force at least one task to be | |
3892 | * moved | |
3893 | */ | |
dbc523a3 GS |
3894 | if (*imbalance < sds->busiest_load_per_task) |
3895 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3896 | |
dbc523a3 | 3897 | } |
37abe198 | 3898 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3899 | |
b7bb4c9b GS |
3900 | /** |
3901 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3902 | * if there is an imbalance. If there isn't an imbalance, and | |
3903 | * the user has opted for power-savings, it returns a group whose | |
3904 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3905 | * such a group exists. | |
3906 | * | |
3907 | * Also calculates the amount of weighted load which should be moved | |
3908 | * to restore balance. | |
3909 | * | |
3910 | * @sd: The sched_domain whose busiest group is to be returned. | |
3911 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3912 | * @imbalance: Variable which stores amount of weighted load which should | |
3913 | * be moved to restore balance/put a group to idle. | |
3914 | * @idle: The idle status of this_cpu. | |
3915 | * @sd_idle: The idleness of sd | |
3916 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3917 | * @balance: Pointer to a variable indicating if this_cpu | |
3918 | * is the appropriate cpu to perform load balancing at this_level. | |
3919 | * | |
3920 | * Returns: - the busiest group if imbalance exists. | |
3921 | * - If no imbalance and user has opted for power-savings balance, | |
3922 | * return the least loaded group whose CPUs can be | |
3923 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3924 | */ |
3925 | static struct sched_group * | |
3926 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3927 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3928 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3929 | { | |
3930 | struct sd_lb_stats sds; | |
1da177e4 | 3931 | |
37abe198 | 3932 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3933 | |
37abe198 GS |
3934 | /* |
3935 | * Compute the various statistics relavent for load balancing at | |
3936 | * this level. | |
3937 | */ | |
3938 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3939 | balance, &sds); | |
3940 | ||
b7bb4c9b GS |
3941 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3942 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3943 | * at this level. | |
3944 | * 2) There is no busy sibling group to pull from. | |
3945 | * 3) This group is the busiest group. | |
3946 | * 4) This group is more busy than the avg busieness at this | |
3947 | * sched_domain. | |
3948 | * 5) The imbalance is within the specified limit. | |
3949 | * 6) Any rebalance would lead to ping-pong | |
3950 | */ | |
37abe198 GS |
3951 | if (balance && !(*balance)) |
3952 | goto ret; | |
1da177e4 | 3953 | |
b7bb4c9b GS |
3954 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3955 | goto out_balanced; | |
1da177e4 | 3956 | |
b7bb4c9b | 3957 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3958 | goto out_balanced; |
1da177e4 | 3959 | |
222d656d | 3960 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3961 | |
b7bb4c9b GS |
3962 | if (sds.this_load >= sds.avg_load) |
3963 | goto out_balanced; | |
3964 | ||
3965 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3966 | goto out_balanced; |
3967 | ||
222d656d GS |
3968 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3969 | if (sds.group_imb) | |
3970 | sds.busiest_load_per_task = | |
3971 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3972 | |
1da177e4 LT |
3973 | /* |
3974 | * We're trying to get all the cpus to the average_load, so we don't | |
3975 | * want to push ourselves above the average load, nor do we wish to | |
3976 | * reduce the max loaded cpu below the average load, as either of these | |
3977 | * actions would just result in more rebalancing later, and ping-pong | |
3978 | * tasks around. Thus we look for the minimum possible imbalance. | |
3979 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3980 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3981 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3982 | * appear as very large values with unsigned longs. |
3983 | */ | |
222d656d | 3984 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3985 | goto out_balanced; |
3986 | ||
dbc523a3 GS |
3987 | /* Looks like there is an imbalance. Compute it */ |
3988 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3989 | return sds.busiest; |
1da177e4 LT |
3990 | |
3991 | out_balanced: | |
c071df18 GS |
3992 | /* |
3993 | * There is no obvious imbalance. But check if we can do some balancing | |
3994 | * to save power. | |
3995 | */ | |
3996 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3997 | return sds.busiest; | |
783609c6 | 3998 | ret: |
1da177e4 LT |
3999 | *imbalance = 0; |
4000 | return NULL; | |
4001 | } | |
4002 | ||
4003 | /* | |
4004 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
4005 | */ | |
70b97a7f | 4006 | static struct rq * |
d15bcfdb | 4007 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 4008 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 4009 | { |
70b97a7f | 4010 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 4011 | unsigned long max_load = 0; |
1da177e4 LT |
4012 | int i; |
4013 | ||
758b2cdc | 4014 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 4015 | unsigned long wl; |
0a2966b4 | 4016 | |
96f874e2 | 4017 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
4018 | continue; |
4019 | ||
48f24c4d | 4020 | rq = cpu_rq(i); |
dd41f596 | 4021 | wl = weighted_cpuload(i); |
2dd73a4f | 4022 | |
dd41f596 | 4023 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 4024 | continue; |
1da177e4 | 4025 | |
dd41f596 IM |
4026 | if (wl > max_load) { |
4027 | max_load = wl; | |
48f24c4d | 4028 | busiest = rq; |
1da177e4 LT |
4029 | } |
4030 | } | |
4031 | ||
4032 | return busiest; | |
4033 | } | |
4034 | ||
77391d71 NP |
4035 | /* |
4036 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
4037 | * so long as it is large enough. | |
4038 | */ | |
4039 | #define MAX_PINNED_INTERVAL 512 | |
4040 | ||
df7c8e84 RR |
4041 | /* Working cpumask for load_balance and load_balance_newidle. */ |
4042 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
4043 | ||
1da177e4 LT |
4044 | /* |
4045 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4046 | * tasks if there is an imbalance. | |
1da177e4 | 4047 | */ |
70b97a7f | 4048 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4049 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4050 | int *balance) |
1da177e4 | 4051 | { |
43010659 | 4052 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4053 | struct sched_group *group; |
1da177e4 | 4054 | unsigned long imbalance; |
70b97a7f | 4055 | struct rq *busiest; |
fe2eea3f | 4056 | unsigned long flags; |
df7c8e84 | 4057 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4058 | |
96f874e2 | 4059 | cpumask_setall(cpus); |
7c16ec58 | 4060 | |
89c4710e SS |
4061 | /* |
4062 | * When power savings policy is enabled for the parent domain, idle | |
4063 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4064 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4065 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4066 | */ |
d15bcfdb | 4067 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4068 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4069 | sd_idle = 1; |
1da177e4 | 4070 | |
2d72376b | 4071 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4072 | |
0a2966b4 | 4073 | redo: |
c8cba857 | 4074 | update_shares(sd); |
0a2966b4 | 4075 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4076 | cpus, balance); |
783609c6 | 4077 | |
06066714 | 4078 | if (*balance == 0) |
783609c6 | 4079 | goto out_balanced; |
783609c6 | 4080 | |
1da177e4 LT |
4081 | if (!group) { |
4082 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4083 | goto out_balanced; | |
4084 | } | |
4085 | ||
7c16ec58 | 4086 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4087 | if (!busiest) { |
4088 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4089 | goto out_balanced; | |
4090 | } | |
4091 | ||
db935dbd | 4092 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4093 | |
4094 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4095 | ||
43010659 | 4096 | ld_moved = 0; |
1da177e4 LT |
4097 | if (busiest->nr_running > 1) { |
4098 | /* | |
4099 | * Attempt to move tasks. If find_busiest_group has found | |
4100 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4101 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4102 | * correctly treated as an imbalance. |
4103 | */ | |
fe2eea3f | 4104 | local_irq_save(flags); |
e17224bf | 4105 | double_rq_lock(this_rq, busiest); |
43010659 | 4106 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4107 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4108 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4109 | local_irq_restore(flags); |
81026794 | 4110 | |
46cb4b7c SS |
4111 | /* |
4112 | * some other cpu did the load balance for us. | |
4113 | */ | |
43010659 | 4114 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4115 | resched_cpu(this_cpu); |
4116 | ||
81026794 | 4117 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4118 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4119 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4120 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4121 | goto redo; |
81026794 | 4122 | goto out_balanced; |
0a2966b4 | 4123 | } |
1da177e4 | 4124 | } |
81026794 | 4125 | |
43010659 | 4126 | if (!ld_moved) { |
1da177e4 LT |
4127 | schedstat_inc(sd, lb_failed[idle]); |
4128 | sd->nr_balance_failed++; | |
4129 | ||
4130 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4131 | |
fe2eea3f | 4132 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4133 | |
4134 | /* don't kick the migration_thread, if the curr | |
4135 | * task on busiest cpu can't be moved to this_cpu | |
4136 | */ | |
96f874e2 RR |
4137 | if (!cpumask_test_cpu(this_cpu, |
4138 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4139 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4140 | all_pinned = 1; |
4141 | goto out_one_pinned; | |
4142 | } | |
4143 | ||
1da177e4 LT |
4144 | if (!busiest->active_balance) { |
4145 | busiest->active_balance = 1; | |
4146 | busiest->push_cpu = this_cpu; | |
81026794 | 4147 | active_balance = 1; |
1da177e4 | 4148 | } |
fe2eea3f | 4149 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4150 | if (active_balance) |
1da177e4 LT |
4151 | wake_up_process(busiest->migration_thread); |
4152 | ||
4153 | /* | |
4154 | * We've kicked active balancing, reset the failure | |
4155 | * counter. | |
4156 | */ | |
39507451 | 4157 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4158 | } |
81026794 | 4159 | } else |
1da177e4 LT |
4160 | sd->nr_balance_failed = 0; |
4161 | ||
81026794 | 4162 | if (likely(!active_balance)) { |
1da177e4 LT |
4163 | /* We were unbalanced, so reset the balancing interval */ |
4164 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4165 | } else { |
4166 | /* | |
4167 | * If we've begun active balancing, start to back off. This | |
4168 | * case may not be covered by the all_pinned logic if there | |
4169 | * is only 1 task on the busy runqueue (because we don't call | |
4170 | * move_tasks). | |
4171 | */ | |
4172 | if (sd->balance_interval < sd->max_interval) | |
4173 | sd->balance_interval *= 2; | |
1da177e4 LT |
4174 | } |
4175 | ||
43010659 | 4176 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4177 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4178 | ld_moved = -1; |
4179 | ||
4180 | goto out; | |
1da177e4 LT |
4181 | |
4182 | out_balanced: | |
1da177e4 LT |
4183 | schedstat_inc(sd, lb_balanced[idle]); |
4184 | ||
16cfb1c0 | 4185 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4186 | |
4187 | out_one_pinned: | |
1da177e4 | 4188 | /* tune up the balancing interval */ |
77391d71 NP |
4189 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4190 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4191 | sd->balance_interval *= 2; |
4192 | ||
48f24c4d | 4193 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4194 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4195 | ld_moved = -1; |
4196 | else | |
4197 | ld_moved = 0; | |
4198 | out: | |
c8cba857 PZ |
4199 | if (ld_moved) |
4200 | update_shares(sd); | |
c09595f6 | 4201 | return ld_moved; |
1da177e4 LT |
4202 | } |
4203 | ||
4204 | /* | |
4205 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4206 | * tasks if there is an imbalance. | |
4207 | * | |
d15bcfdb | 4208 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4209 | * this_rq is locked. |
4210 | */ | |
48f24c4d | 4211 | static int |
df7c8e84 | 4212 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4213 | { |
4214 | struct sched_group *group; | |
70b97a7f | 4215 | struct rq *busiest = NULL; |
1da177e4 | 4216 | unsigned long imbalance; |
43010659 | 4217 | int ld_moved = 0; |
5969fe06 | 4218 | int sd_idle = 0; |
969bb4e4 | 4219 | int all_pinned = 0; |
df7c8e84 | 4220 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4221 | |
96f874e2 | 4222 | cpumask_setall(cpus); |
5969fe06 | 4223 | |
89c4710e SS |
4224 | /* |
4225 | * When power savings policy is enabled for the parent domain, idle | |
4226 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4227 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4228 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4229 | */ |
4230 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4231 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4232 | sd_idle = 1; |
1da177e4 | 4233 | |
2d72376b | 4234 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4235 | redo: |
3e5459b4 | 4236 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4237 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4238 | &sd_idle, cpus, NULL); |
1da177e4 | 4239 | if (!group) { |
d15bcfdb | 4240 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4241 | goto out_balanced; |
1da177e4 LT |
4242 | } |
4243 | ||
7c16ec58 | 4244 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4245 | if (!busiest) { |
d15bcfdb | 4246 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4247 | goto out_balanced; |
1da177e4 LT |
4248 | } |
4249 | ||
db935dbd NP |
4250 | BUG_ON(busiest == this_rq); |
4251 | ||
d15bcfdb | 4252 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4253 | |
43010659 | 4254 | ld_moved = 0; |
d6d5cfaf NP |
4255 | if (busiest->nr_running > 1) { |
4256 | /* Attempt to move tasks */ | |
4257 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4258 | /* this_rq->clock is already updated */ |
4259 | update_rq_clock(busiest); | |
43010659 | 4260 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4261 | imbalance, sd, CPU_NEWLY_IDLE, |
4262 | &all_pinned); | |
1b12bbc7 | 4263 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4264 | |
969bb4e4 | 4265 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4266 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4267 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4268 | goto redo; |
4269 | } | |
d6d5cfaf NP |
4270 | } |
4271 | ||
43010659 | 4272 | if (!ld_moved) { |
36dffab6 | 4273 | int active_balance = 0; |
ad273b32 | 4274 | |
d15bcfdb | 4275 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4276 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4277 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4278 | return -1; |
ad273b32 VS |
4279 | |
4280 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4281 | return -1; | |
4282 | ||
4283 | if (sd->nr_balance_failed++ < 2) | |
4284 | return -1; | |
4285 | ||
4286 | /* | |
4287 | * The only task running in a non-idle cpu can be moved to this | |
4288 | * cpu in an attempt to completely freeup the other CPU | |
4289 | * package. The same method used to move task in load_balance() | |
4290 | * have been extended for load_balance_newidle() to speedup | |
4291 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4292 | * | |
4293 | * The package power saving logic comes from | |
4294 | * find_busiest_group(). If there are no imbalance, then | |
4295 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4296 | * f_b_g() will select a group from which a running task may be | |
4297 | * pulled to this cpu in order to make the other package idle. | |
4298 | * If there is no opportunity to make a package idle and if | |
4299 | * there are no imbalance, then f_b_g() will return NULL and no | |
4300 | * action will be taken in load_balance_newidle(). | |
4301 | * | |
4302 | * Under normal task pull operation due to imbalance, there | |
4303 | * will be more than one task in the source run queue and | |
4304 | * move_tasks() will succeed. ld_moved will be true and this | |
4305 | * active balance code will not be triggered. | |
4306 | */ | |
4307 | ||
4308 | /* Lock busiest in correct order while this_rq is held */ | |
4309 | double_lock_balance(this_rq, busiest); | |
4310 | ||
4311 | /* | |
4312 | * don't kick the migration_thread, if the curr | |
4313 | * task on busiest cpu can't be moved to this_cpu | |
4314 | */ | |
6ca09dfc | 4315 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4316 | double_unlock_balance(this_rq, busiest); |
4317 | all_pinned = 1; | |
4318 | return ld_moved; | |
4319 | } | |
4320 | ||
4321 | if (!busiest->active_balance) { | |
4322 | busiest->active_balance = 1; | |
4323 | busiest->push_cpu = this_cpu; | |
4324 | active_balance = 1; | |
4325 | } | |
4326 | ||
4327 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4328 | /* |
4329 | * Should not call ttwu while holding a rq->lock | |
4330 | */ | |
4331 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4332 | if (active_balance) |
4333 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4334 | spin_lock(&this_rq->lock); |
ad273b32 | 4335 | |
5969fe06 | 4336 | } else |
16cfb1c0 | 4337 | sd->nr_balance_failed = 0; |
1da177e4 | 4338 | |
3e5459b4 | 4339 | update_shares_locked(this_rq, sd); |
43010659 | 4340 | return ld_moved; |
16cfb1c0 NP |
4341 | |
4342 | out_balanced: | |
d15bcfdb | 4343 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4344 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4345 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4346 | return -1; |
16cfb1c0 | 4347 | sd->nr_balance_failed = 0; |
48f24c4d | 4348 | |
16cfb1c0 | 4349 | return 0; |
1da177e4 LT |
4350 | } |
4351 | ||
4352 | /* | |
4353 | * idle_balance is called by schedule() if this_cpu is about to become | |
4354 | * idle. Attempts to pull tasks from other CPUs. | |
4355 | */ | |
70b97a7f | 4356 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4357 | { |
4358 | struct sched_domain *sd; | |
efbe027e | 4359 | int pulled_task = 0; |
dd41f596 | 4360 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4361 | |
4362 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4363 | unsigned long interval; |
4364 | ||
4365 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4366 | continue; | |
4367 | ||
4368 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4369 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4370 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4371 | sd); |
92c4ca5c CL |
4372 | |
4373 | interval = msecs_to_jiffies(sd->balance_interval); | |
4374 | if (time_after(next_balance, sd->last_balance + interval)) | |
4375 | next_balance = sd->last_balance + interval; | |
4376 | if (pulled_task) | |
4377 | break; | |
1da177e4 | 4378 | } |
dd41f596 | 4379 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4380 | /* |
4381 | * We are going idle. next_balance may be set based on | |
4382 | * a busy processor. So reset next_balance. | |
4383 | */ | |
4384 | this_rq->next_balance = next_balance; | |
dd41f596 | 4385 | } |
1da177e4 LT |
4386 | } |
4387 | ||
4388 | /* | |
4389 | * active_load_balance is run by migration threads. It pushes running tasks | |
4390 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4391 | * running on each physical CPU where possible, and avoids physical / | |
4392 | * logical imbalances. | |
4393 | * | |
4394 | * Called with busiest_rq locked. | |
4395 | */ | |
70b97a7f | 4396 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4397 | { |
39507451 | 4398 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4399 | struct sched_domain *sd; |
4400 | struct rq *target_rq; | |
39507451 | 4401 | |
48f24c4d | 4402 | /* Is there any task to move? */ |
39507451 | 4403 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4404 | return; |
4405 | ||
4406 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4407 | |
4408 | /* | |
39507451 | 4409 | * This condition is "impossible", if it occurs |
41a2d6cf | 4410 | * we need to fix it. Originally reported by |
39507451 | 4411 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4412 | */ |
39507451 | 4413 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4414 | |
39507451 NP |
4415 | /* move a task from busiest_rq to target_rq */ |
4416 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4417 | update_rq_clock(busiest_rq); |
4418 | update_rq_clock(target_rq); | |
39507451 NP |
4419 | |
4420 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4421 | for_each_domain(target_cpu, sd) { |
39507451 | 4422 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4423 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4424 | break; |
c96d145e | 4425 | } |
39507451 | 4426 | |
48f24c4d | 4427 | if (likely(sd)) { |
2d72376b | 4428 | schedstat_inc(sd, alb_count); |
39507451 | 4429 | |
43010659 PW |
4430 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4431 | sd, CPU_IDLE)) | |
48f24c4d IM |
4432 | schedstat_inc(sd, alb_pushed); |
4433 | else | |
4434 | schedstat_inc(sd, alb_failed); | |
4435 | } | |
1b12bbc7 | 4436 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4437 | } |
4438 | ||
46cb4b7c SS |
4439 | #ifdef CONFIG_NO_HZ |
4440 | static struct { | |
4441 | atomic_t load_balancer; | |
7d1e6a9b | 4442 | cpumask_var_t cpu_mask; |
f711f609 | 4443 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4444 | } nohz ____cacheline_aligned = { |
4445 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4446 | }; |
4447 | ||
eea08f32 AB |
4448 | int get_nohz_load_balancer(void) |
4449 | { | |
4450 | return atomic_read(&nohz.load_balancer); | |
4451 | } | |
4452 | ||
f711f609 GS |
4453 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4454 | /** | |
4455 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4456 | * @cpu: The cpu whose lowest level of sched domain is to | |
4457 | * be returned. | |
4458 | * @flag: The flag to check for the lowest sched_domain | |
4459 | * for the given cpu. | |
4460 | * | |
4461 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4462 | */ | |
4463 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4464 | { | |
4465 | struct sched_domain *sd; | |
4466 | ||
4467 | for_each_domain(cpu, sd) | |
4468 | if (sd && (sd->flags & flag)) | |
4469 | break; | |
4470 | ||
4471 | return sd; | |
4472 | } | |
4473 | ||
4474 | /** | |
4475 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4476 | * @cpu: The cpu whose domains we're iterating over. | |
4477 | * @sd: variable holding the value of the power_savings_sd | |
4478 | * for cpu. | |
4479 | * @flag: The flag to filter the sched_domains to be iterated. | |
4480 | * | |
4481 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4482 | * set, starting from the lowest sched_domain to the highest. | |
4483 | */ | |
4484 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4485 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4486 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4487 | ||
4488 | /** | |
4489 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4490 | * @ilb_group: group to be checked for semi-idleness | |
4491 | * | |
4492 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4493 | * | |
4494 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4495 | * and atleast one non-idle CPU. This helper function checks if the given | |
4496 | * sched_group is semi-idle or not. | |
4497 | */ | |
4498 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4499 | { | |
4500 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4501 | sched_group_cpus(ilb_group)); | |
4502 | ||
4503 | /* | |
4504 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4505 | * and atleast one idle cpu. | |
4506 | */ | |
4507 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4508 | return 0; | |
4509 | ||
4510 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4511 | return 0; | |
4512 | ||
4513 | return 1; | |
4514 | } | |
4515 | /** | |
4516 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4517 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4518 | * | |
4519 | * Returns: Returns the id of the idle load balancer if it exists, | |
4520 | * Else, returns >= nr_cpu_ids. | |
4521 | * | |
4522 | * This algorithm picks the idle load balancer such that it belongs to a | |
4523 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4524 | * completely idle packages/cores just for the purpose of idle load balancing | |
4525 | * when there are other idle cpu's which are better suited for that job. | |
4526 | */ | |
4527 | static int find_new_ilb(int cpu) | |
4528 | { | |
4529 | struct sched_domain *sd; | |
4530 | struct sched_group *ilb_group; | |
4531 | ||
4532 | /* | |
4533 | * Have idle load balancer selection from semi-idle packages only | |
4534 | * when power-aware load balancing is enabled | |
4535 | */ | |
4536 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4537 | goto out_done; | |
4538 | ||
4539 | /* | |
4540 | * Optimize for the case when we have no idle CPUs or only one | |
4541 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4542 | */ | |
4543 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4544 | goto out_done; | |
4545 | ||
4546 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4547 | ilb_group = sd->groups; | |
4548 | ||
4549 | do { | |
4550 | if (is_semi_idle_group(ilb_group)) | |
4551 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4552 | ||
4553 | ilb_group = ilb_group->next; | |
4554 | ||
4555 | } while (ilb_group != sd->groups); | |
4556 | } | |
4557 | ||
4558 | out_done: | |
4559 | return cpumask_first(nohz.cpu_mask); | |
4560 | } | |
4561 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4562 | static inline int find_new_ilb(int call_cpu) | |
4563 | { | |
6e29ec57 | 4564 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4565 | } |
4566 | #endif | |
4567 | ||
7835b98b | 4568 | /* |
46cb4b7c SS |
4569 | * This routine will try to nominate the ilb (idle load balancing) |
4570 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4571 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4572 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4573 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4574 | * arrives... | |
4575 | * | |
4576 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4577 | * for idle load balancing. ilb owner will still be part of | |
4578 | * nohz.cpu_mask.. | |
7835b98b | 4579 | * |
46cb4b7c SS |
4580 | * While stopping the tick, this cpu will become the ilb owner if there |
4581 | * is no other owner. And will be the owner till that cpu becomes busy | |
4582 | * or if all cpus in the system stop their ticks at which point | |
4583 | * there is no need for ilb owner. | |
4584 | * | |
4585 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4586 | * next busy scheduler_tick() | |
4587 | */ | |
4588 | int select_nohz_load_balancer(int stop_tick) | |
4589 | { | |
4590 | int cpu = smp_processor_id(); | |
4591 | ||
4592 | if (stop_tick) { | |
46cb4b7c SS |
4593 | cpu_rq(cpu)->in_nohz_recently = 1; |
4594 | ||
483b4ee6 SS |
4595 | if (!cpu_active(cpu)) { |
4596 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4597 | return 0; | |
4598 | ||
4599 | /* | |
4600 | * If we are going offline and still the leader, | |
4601 | * give up! | |
4602 | */ | |
46cb4b7c SS |
4603 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4604 | BUG(); | |
483b4ee6 | 4605 | |
46cb4b7c SS |
4606 | return 0; |
4607 | } | |
4608 | ||
483b4ee6 SS |
4609 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4610 | ||
46cb4b7c | 4611 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4612 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4613 | if (atomic_read(&nohz.load_balancer) == cpu) |
4614 | atomic_set(&nohz.load_balancer, -1); | |
4615 | return 0; | |
4616 | } | |
4617 | ||
4618 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4619 | /* make me the ilb owner */ | |
4620 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4621 | return 1; | |
e790fb0b GS |
4622 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4623 | int new_ilb; | |
4624 | ||
4625 | if (!(sched_smt_power_savings || | |
4626 | sched_mc_power_savings)) | |
4627 | return 1; | |
4628 | /* | |
4629 | * Check to see if there is a more power-efficient | |
4630 | * ilb. | |
4631 | */ | |
4632 | new_ilb = find_new_ilb(cpu); | |
4633 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4634 | atomic_set(&nohz.load_balancer, -1); | |
4635 | resched_cpu(new_ilb); | |
4636 | return 0; | |
4637 | } | |
46cb4b7c | 4638 | return 1; |
e790fb0b | 4639 | } |
46cb4b7c | 4640 | } else { |
7d1e6a9b | 4641 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4642 | return 0; |
4643 | ||
7d1e6a9b | 4644 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4645 | |
4646 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4647 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4648 | BUG(); | |
4649 | } | |
4650 | return 0; | |
4651 | } | |
4652 | #endif | |
4653 | ||
4654 | static DEFINE_SPINLOCK(balancing); | |
4655 | ||
4656 | /* | |
7835b98b CL |
4657 | * It checks each scheduling domain to see if it is due to be balanced, |
4658 | * and initiates a balancing operation if so. | |
4659 | * | |
4660 | * Balancing parameters are set up in arch_init_sched_domains. | |
4661 | */ | |
a9957449 | 4662 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4663 | { |
46cb4b7c SS |
4664 | int balance = 1; |
4665 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4666 | unsigned long interval; |
4667 | struct sched_domain *sd; | |
46cb4b7c | 4668 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4669 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4670 | int update_next_balance = 0; |
d07355f5 | 4671 | int need_serialize; |
1da177e4 | 4672 | |
46cb4b7c | 4673 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4674 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4675 | continue; | |
4676 | ||
4677 | interval = sd->balance_interval; | |
d15bcfdb | 4678 | if (idle != CPU_IDLE) |
1da177e4 LT |
4679 | interval *= sd->busy_factor; |
4680 | ||
4681 | /* scale ms to jiffies */ | |
4682 | interval = msecs_to_jiffies(interval); | |
4683 | if (unlikely(!interval)) | |
4684 | interval = 1; | |
dd41f596 IM |
4685 | if (interval > HZ*NR_CPUS/10) |
4686 | interval = HZ*NR_CPUS/10; | |
4687 | ||
d07355f5 | 4688 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4689 | |
d07355f5 | 4690 | if (need_serialize) { |
08c183f3 CL |
4691 | if (!spin_trylock(&balancing)) |
4692 | goto out; | |
4693 | } | |
4694 | ||
c9819f45 | 4695 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4696 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4697 | /* |
4698 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4699 | * longer idle, or one of our SMT siblings is |
4700 | * not idle. | |
4701 | */ | |
d15bcfdb | 4702 | idle = CPU_NOT_IDLE; |
1da177e4 | 4703 | } |
1bd77f2d | 4704 | sd->last_balance = jiffies; |
1da177e4 | 4705 | } |
d07355f5 | 4706 | if (need_serialize) |
08c183f3 CL |
4707 | spin_unlock(&balancing); |
4708 | out: | |
f549da84 | 4709 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4710 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4711 | update_next_balance = 1; |
4712 | } | |
783609c6 SS |
4713 | |
4714 | /* | |
4715 | * Stop the load balance at this level. There is another | |
4716 | * CPU in our sched group which is doing load balancing more | |
4717 | * actively. | |
4718 | */ | |
4719 | if (!balance) | |
4720 | break; | |
1da177e4 | 4721 | } |
f549da84 SS |
4722 | |
4723 | /* | |
4724 | * next_balance will be updated only when there is a need. | |
4725 | * When the cpu is attached to null domain for ex, it will not be | |
4726 | * updated. | |
4727 | */ | |
4728 | if (likely(update_next_balance)) | |
4729 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4730 | } |
4731 | ||
4732 | /* | |
4733 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4734 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4735 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4736 | */ | |
4737 | static void run_rebalance_domains(struct softirq_action *h) | |
4738 | { | |
dd41f596 IM |
4739 | int this_cpu = smp_processor_id(); |
4740 | struct rq *this_rq = cpu_rq(this_cpu); | |
4741 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4742 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4743 | |
dd41f596 | 4744 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4745 | |
4746 | #ifdef CONFIG_NO_HZ | |
4747 | /* | |
4748 | * If this cpu is the owner for idle load balancing, then do the | |
4749 | * balancing on behalf of the other idle cpus whose ticks are | |
4750 | * stopped. | |
4751 | */ | |
dd41f596 IM |
4752 | if (this_rq->idle_at_tick && |
4753 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4754 | struct rq *rq; |
4755 | int balance_cpu; | |
4756 | ||
7d1e6a9b RR |
4757 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4758 | if (balance_cpu == this_cpu) | |
4759 | continue; | |
4760 | ||
46cb4b7c SS |
4761 | /* |
4762 | * If this cpu gets work to do, stop the load balancing | |
4763 | * work being done for other cpus. Next load | |
4764 | * balancing owner will pick it up. | |
4765 | */ | |
4766 | if (need_resched()) | |
4767 | break; | |
4768 | ||
de0cf899 | 4769 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4770 | |
4771 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4772 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4773 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4774 | } |
4775 | } | |
4776 | #endif | |
4777 | } | |
4778 | ||
8a0be9ef FW |
4779 | static inline int on_null_domain(int cpu) |
4780 | { | |
4781 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4782 | } | |
4783 | ||
46cb4b7c SS |
4784 | /* |
4785 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4786 | * | |
4787 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4788 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4789 | * if the whole system is idle. | |
4790 | */ | |
dd41f596 | 4791 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4792 | { |
46cb4b7c SS |
4793 | #ifdef CONFIG_NO_HZ |
4794 | /* | |
4795 | * If we were in the nohz mode recently and busy at the current | |
4796 | * scheduler tick, then check if we need to nominate new idle | |
4797 | * load balancer. | |
4798 | */ | |
4799 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4800 | rq->in_nohz_recently = 0; | |
4801 | ||
4802 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4803 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4804 | atomic_set(&nohz.load_balancer, -1); |
4805 | } | |
4806 | ||
4807 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 4808 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 4809 | |
434d53b0 | 4810 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4811 | resched_cpu(ilb); |
4812 | } | |
4813 | } | |
4814 | ||
4815 | /* | |
4816 | * If this cpu is idle and doing idle load balancing for all the | |
4817 | * cpus with ticks stopped, is it time for that to stop? | |
4818 | */ | |
4819 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4820 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4821 | resched_cpu(cpu); |
4822 | return; | |
4823 | } | |
4824 | ||
4825 | /* | |
4826 | * If this cpu is idle and the idle load balancing is done by | |
4827 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4828 | */ | |
4829 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4830 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4831 | return; |
4832 | #endif | |
8a0be9ef FW |
4833 | /* Don't need to rebalance while attached to NULL domain */ |
4834 | if (time_after_eq(jiffies, rq->next_balance) && | |
4835 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4836 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4837 | } |
dd41f596 IM |
4838 | |
4839 | #else /* CONFIG_SMP */ | |
4840 | ||
1da177e4 LT |
4841 | /* |
4842 | * on UP we do not need to balance between CPUs: | |
4843 | */ | |
70b97a7f | 4844 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4845 | { |
4846 | } | |
dd41f596 | 4847 | |
1da177e4 LT |
4848 | #endif |
4849 | ||
1da177e4 LT |
4850 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4851 | ||
4852 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4853 | ||
4854 | /* | |
c5f8d995 | 4855 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4856 | * @p in case that task is currently running. |
c5f8d995 HS |
4857 | * |
4858 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4859 | */ |
c5f8d995 HS |
4860 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4861 | { | |
4862 | u64 ns = 0; | |
4863 | ||
4864 | if (task_current(rq, p)) { | |
4865 | update_rq_clock(rq); | |
4866 | ns = rq->clock - p->se.exec_start; | |
4867 | if ((s64)ns < 0) | |
4868 | ns = 0; | |
4869 | } | |
4870 | ||
4871 | return ns; | |
4872 | } | |
4873 | ||
bb34d92f | 4874 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4875 | { |
1da177e4 | 4876 | unsigned long flags; |
41b86e9c | 4877 | struct rq *rq; |
bb34d92f | 4878 | u64 ns = 0; |
48f24c4d | 4879 | |
41b86e9c | 4880 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4881 | ns = do_task_delta_exec(p, rq); |
4882 | task_rq_unlock(rq, &flags); | |
1508487e | 4883 | |
c5f8d995 HS |
4884 | return ns; |
4885 | } | |
f06febc9 | 4886 | |
c5f8d995 HS |
4887 | /* |
4888 | * Return accounted runtime for the task. | |
4889 | * In case the task is currently running, return the runtime plus current's | |
4890 | * pending runtime that have not been accounted yet. | |
4891 | */ | |
4892 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4893 | { | |
4894 | unsigned long flags; | |
4895 | struct rq *rq; | |
4896 | u64 ns = 0; | |
4897 | ||
4898 | rq = task_rq_lock(p, &flags); | |
4899 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4900 | task_rq_unlock(rq, &flags); | |
4901 | ||
4902 | return ns; | |
4903 | } | |
48f24c4d | 4904 | |
c5f8d995 HS |
4905 | /* |
4906 | * Return sum_exec_runtime for the thread group. | |
4907 | * In case the task is currently running, return the sum plus current's | |
4908 | * pending runtime that have not been accounted yet. | |
4909 | * | |
4910 | * Note that the thread group might have other running tasks as well, | |
4911 | * so the return value not includes other pending runtime that other | |
4912 | * running tasks might have. | |
4913 | */ | |
4914 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4915 | { | |
4916 | struct task_cputime totals; | |
4917 | unsigned long flags; | |
4918 | struct rq *rq; | |
4919 | u64 ns; | |
4920 | ||
4921 | rq = task_rq_lock(p, &flags); | |
4922 | thread_group_cputime(p, &totals); | |
4923 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4924 | task_rq_unlock(rq, &flags); |
48f24c4d | 4925 | |
1da177e4 LT |
4926 | return ns; |
4927 | } | |
4928 | ||
1da177e4 LT |
4929 | /* |
4930 | * Account user cpu time to a process. | |
4931 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4932 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4933 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4934 | */ |
457533a7 MS |
4935 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4936 | cputime_t cputime_scaled) | |
1da177e4 LT |
4937 | { |
4938 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4939 | cputime64_t tmp; | |
4940 | ||
457533a7 | 4941 | /* Add user time to process. */ |
1da177e4 | 4942 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4943 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4944 | account_group_user_time(p, cputime); |
1da177e4 LT |
4945 | |
4946 | /* Add user time to cpustat. */ | |
4947 | tmp = cputime_to_cputime64(cputime); | |
4948 | if (TASK_NICE(p) > 0) | |
4949 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4950 | else | |
4951 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
4952 | |
4953 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
4954 | /* Account for user time used */ |
4955 | acct_update_integrals(p); | |
1da177e4 LT |
4956 | } |
4957 | ||
94886b84 LV |
4958 | /* |
4959 | * Account guest cpu time to a process. | |
4960 | * @p: the process that the cpu time gets accounted to | |
4961 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4962 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4963 | */ |
457533a7 MS |
4964 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4965 | cputime_t cputime_scaled) | |
94886b84 LV |
4966 | { |
4967 | cputime64_t tmp; | |
4968 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4969 | ||
4970 | tmp = cputime_to_cputime64(cputime); | |
4971 | ||
457533a7 | 4972 | /* Add guest time to process. */ |
94886b84 | 4973 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4974 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4975 | account_group_user_time(p, cputime); |
94886b84 LV |
4976 | p->gtime = cputime_add(p->gtime, cputime); |
4977 | ||
457533a7 | 4978 | /* Add guest time to cpustat. */ |
94886b84 LV |
4979 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4980 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4981 | } | |
4982 | ||
1da177e4 LT |
4983 | /* |
4984 | * Account system cpu time to a process. | |
4985 | * @p: the process that the cpu time gets accounted to | |
4986 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4987 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4988 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4989 | */ |
4990 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4991 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4992 | { |
4993 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4994 | cputime64_t tmp; |
4995 | ||
983ed7a6 | 4996 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4997 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4998 | return; |
4999 | } | |
94886b84 | 5000 | |
457533a7 | 5001 | /* Add system time to process. */ |
1da177e4 | 5002 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 5003 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 5004 | account_group_system_time(p, cputime); |
1da177e4 LT |
5005 | |
5006 | /* Add system time to cpustat. */ | |
5007 | tmp = cputime_to_cputime64(cputime); | |
5008 | if (hardirq_count() - hardirq_offset) | |
5009 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
5010 | else if (softirq_count()) | |
5011 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 5012 | else |
79741dd3 MS |
5013 | cpustat->system = cputime64_add(cpustat->system, tmp); |
5014 | ||
ef12fefa BR |
5015 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
5016 | ||
1da177e4 LT |
5017 | /* Account for system time used */ |
5018 | acct_update_integrals(p); | |
1da177e4 LT |
5019 | } |
5020 | ||
c66f08be | 5021 | /* |
1da177e4 | 5022 | * Account for involuntary wait time. |
1da177e4 | 5023 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 5024 | */ |
79741dd3 | 5025 | void account_steal_time(cputime_t cputime) |
c66f08be | 5026 | { |
79741dd3 MS |
5027 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
5028 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
5029 | ||
5030 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
5031 | } |
5032 | ||
1da177e4 | 5033 | /* |
79741dd3 MS |
5034 | * Account for idle time. |
5035 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 5036 | */ |
79741dd3 | 5037 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
5038 | { |
5039 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 5040 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 5041 | struct rq *rq = this_rq(); |
1da177e4 | 5042 | |
79741dd3 MS |
5043 | if (atomic_read(&rq->nr_iowait) > 0) |
5044 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
5045 | else | |
5046 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
5047 | } |
5048 | ||
79741dd3 MS |
5049 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5050 | ||
5051 | /* | |
5052 | * Account a single tick of cpu time. | |
5053 | * @p: the process that the cpu time gets accounted to | |
5054 | * @user_tick: indicates if the tick is a user or a system tick | |
5055 | */ | |
5056 | void account_process_tick(struct task_struct *p, int user_tick) | |
5057 | { | |
5058 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
5059 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
5060 | struct rq *rq = this_rq(); | |
5061 | ||
5062 | if (user_tick) | |
5063 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 5064 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
5065 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
5066 | one_jiffy_scaled); | |
5067 | else | |
5068 | account_idle_time(one_jiffy); | |
5069 | } | |
5070 | ||
5071 | /* | |
5072 | * Account multiple ticks of steal time. | |
5073 | * @p: the process from which the cpu time has been stolen | |
5074 | * @ticks: number of stolen ticks | |
5075 | */ | |
5076 | void account_steal_ticks(unsigned long ticks) | |
5077 | { | |
5078 | account_steal_time(jiffies_to_cputime(ticks)); | |
5079 | } | |
5080 | ||
5081 | /* | |
5082 | * Account multiple ticks of idle time. | |
5083 | * @ticks: number of stolen ticks | |
5084 | */ | |
5085 | void account_idle_ticks(unsigned long ticks) | |
5086 | { | |
5087 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5088 | } |
5089 | ||
79741dd3 MS |
5090 | #endif |
5091 | ||
49048622 BS |
5092 | /* |
5093 | * Use precise platform statistics if available: | |
5094 | */ | |
5095 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5096 | cputime_t task_utime(struct task_struct *p) | |
5097 | { | |
5098 | return p->utime; | |
5099 | } | |
5100 | ||
5101 | cputime_t task_stime(struct task_struct *p) | |
5102 | { | |
5103 | return p->stime; | |
5104 | } | |
5105 | #else | |
5106 | cputime_t task_utime(struct task_struct *p) | |
5107 | { | |
5108 | clock_t utime = cputime_to_clock_t(p->utime), | |
5109 | total = utime + cputime_to_clock_t(p->stime); | |
5110 | u64 temp; | |
5111 | ||
5112 | /* | |
5113 | * Use CFS's precise accounting: | |
5114 | */ | |
5115 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
5116 | ||
5117 | if (total) { | |
5118 | temp *= utime; | |
5119 | do_div(temp, total); | |
5120 | } | |
5121 | utime = (clock_t)temp; | |
5122 | ||
5123 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
5124 | return p->prev_utime; | |
5125 | } | |
5126 | ||
5127 | cputime_t task_stime(struct task_struct *p) | |
5128 | { | |
5129 | clock_t stime; | |
5130 | ||
5131 | /* | |
5132 | * Use CFS's precise accounting. (we subtract utime from | |
5133 | * the total, to make sure the total observed by userspace | |
5134 | * grows monotonically - apps rely on that): | |
5135 | */ | |
5136 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5137 | cputime_to_clock_t(task_utime(p)); | |
5138 | ||
5139 | if (stime >= 0) | |
5140 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5141 | ||
5142 | return p->prev_stime; | |
5143 | } | |
5144 | #endif | |
5145 | ||
5146 | inline cputime_t task_gtime(struct task_struct *p) | |
5147 | { | |
5148 | return p->gtime; | |
5149 | } | |
5150 | ||
7835b98b CL |
5151 | /* |
5152 | * This function gets called by the timer code, with HZ frequency. | |
5153 | * We call it with interrupts disabled. | |
5154 | * | |
5155 | * It also gets called by the fork code, when changing the parent's | |
5156 | * timeslices. | |
5157 | */ | |
5158 | void scheduler_tick(void) | |
5159 | { | |
7835b98b CL |
5160 | int cpu = smp_processor_id(); |
5161 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5162 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5163 | |
5164 | sched_clock_tick(); | |
dd41f596 IM |
5165 | |
5166 | spin_lock(&rq->lock); | |
3e51f33f | 5167 | update_rq_clock(rq); |
f1a438d8 | 5168 | update_cpu_load(rq); |
fa85ae24 | 5169 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5170 | spin_unlock(&rq->lock); |
7835b98b | 5171 | |
e220d2dc PZ |
5172 | perf_counter_task_tick(curr, cpu); |
5173 | ||
e418e1c2 | 5174 | #ifdef CONFIG_SMP |
dd41f596 IM |
5175 | rq->idle_at_tick = idle_cpu(cpu); |
5176 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5177 | #endif |
1da177e4 LT |
5178 | } |
5179 | ||
132380a0 | 5180 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5181 | { |
5182 | if (in_lock_functions(addr)) { | |
5183 | addr = CALLER_ADDR2; | |
5184 | if (in_lock_functions(addr)) | |
5185 | addr = CALLER_ADDR3; | |
5186 | } | |
5187 | return addr; | |
5188 | } | |
1da177e4 | 5189 | |
7e49fcce SR |
5190 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5191 | defined(CONFIG_PREEMPT_TRACER)) | |
5192 | ||
43627582 | 5193 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5194 | { |
6cd8a4bb | 5195 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5196 | /* |
5197 | * Underflow? | |
5198 | */ | |
9a11b49a IM |
5199 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5200 | return; | |
6cd8a4bb | 5201 | #endif |
1da177e4 | 5202 | preempt_count() += val; |
6cd8a4bb | 5203 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5204 | /* |
5205 | * Spinlock count overflowing soon? | |
5206 | */ | |
33859f7f MOS |
5207 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5208 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5209 | #endif |
5210 | if (preempt_count() == val) | |
5211 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5212 | } |
5213 | EXPORT_SYMBOL(add_preempt_count); | |
5214 | ||
43627582 | 5215 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5216 | { |
6cd8a4bb | 5217 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5218 | /* |
5219 | * Underflow? | |
5220 | */ | |
01e3eb82 | 5221 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5222 | return; |
1da177e4 LT |
5223 | /* |
5224 | * Is the spinlock portion underflowing? | |
5225 | */ | |
9a11b49a IM |
5226 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5227 | !(preempt_count() & PREEMPT_MASK))) | |
5228 | return; | |
6cd8a4bb | 5229 | #endif |
9a11b49a | 5230 | |
6cd8a4bb SR |
5231 | if (preempt_count() == val) |
5232 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5233 | preempt_count() -= val; |
5234 | } | |
5235 | EXPORT_SYMBOL(sub_preempt_count); | |
5236 | ||
5237 | #endif | |
5238 | ||
5239 | /* | |
dd41f596 | 5240 | * Print scheduling while atomic bug: |
1da177e4 | 5241 | */ |
dd41f596 | 5242 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5243 | { |
838225b4 SS |
5244 | struct pt_regs *regs = get_irq_regs(); |
5245 | ||
5246 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5247 | prev->comm, prev->pid, preempt_count()); | |
5248 | ||
dd41f596 | 5249 | debug_show_held_locks(prev); |
e21f5b15 | 5250 | print_modules(); |
dd41f596 IM |
5251 | if (irqs_disabled()) |
5252 | print_irqtrace_events(prev); | |
838225b4 SS |
5253 | |
5254 | if (regs) | |
5255 | show_regs(regs); | |
5256 | else | |
5257 | dump_stack(); | |
dd41f596 | 5258 | } |
1da177e4 | 5259 | |
dd41f596 IM |
5260 | /* |
5261 | * Various schedule()-time debugging checks and statistics: | |
5262 | */ | |
5263 | static inline void schedule_debug(struct task_struct *prev) | |
5264 | { | |
1da177e4 | 5265 | /* |
41a2d6cf | 5266 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5267 | * schedule() atomically, we ignore that path for now. |
5268 | * Otherwise, whine if we are scheduling when we should not be. | |
5269 | */ | |
3f33a7ce | 5270 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5271 | __schedule_bug(prev); |
5272 | ||
1da177e4 LT |
5273 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5274 | ||
2d72376b | 5275 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5276 | #ifdef CONFIG_SCHEDSTATS |
5277 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5278 | schedstat_inc(this_rq(), bkl_count); |
5279 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5280 | } |
5281 | #endif | |
dd41f596 IM |
5282 | } |
5283 | ||
df1c99d4 MG |
5284 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5285 | { | |
5286 | if (prev->state == TASK_RUNNING) { | |
5287 | u64 runtime = prev->se.sum_exec_runtime; | |
5288 | ||
5289 | runtime -= prev->se.prev_sum_exec_runtime; | |
5290 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5291 | ||
5292 | /* | |
5293 | * In order to avoid avg_overlap growing stale when we are | |
5294 | * indeed overlapping and hence not getting put to sleep, grow | |
5295 | * the avg_overlap on preemption. | |
5296 | * | |
5297 | * We use the average preemption runtime because that | |
5298 | * correlates to the amount of cache footprint a task can | |
5299 | * build up. | |
5300 | */ | |
5301 | update_avg(&prev->se.avg_overlap, runtime); | |
5302 | } | |
5303 | prev->sched_class->put_prev_task(rq, prev); | |
5304 | } | |
5305 | ||
dd41f596 IM |
5306 | /* |
5307 | * Pick up the highest-prio task: | |
5308 | */ | |
5309 | static inline struct task_struct * | |
b67802ea | 5310 | pick_next_task(struct rq *rq) |
dd41f596 | 5311 | { |
5522d5d5 | 5312 | const struct sched_class *class; |
dd41f596 | 5313 | struct task_struct *p; |
1da177e4 LT |
5314 | |
5315 | /* | |
dd41f596 IM |
5316 | * Optimization: we know that if all tasks are in |
5317 | * the fair class we can call that function directly: | |
1da177e4 | 5318 | */ |
dd41f596 | 5319 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5320 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5321 | if (likely(p)) |
5322 | return p; | |
1da177e4 LT |
5323 | } |
5324 | ||
dd41f596 IM |
5325 | class = sched_class_highest; |
5326 | for ( ; ; ) { | |
fb8d4724 | 5327 | p = class->pick_next_task(rq); |
dd41f596 IM |
5328 | if (p) |
5329 | return p; | |
5330 | /* | |
5331 | * Will never be NULL as the idle class always | |
5332 | * returns a non-NULL p: | |
5333 | */ | |
5334 | class = class->next; | |
5335 | } | |
5336 | } | |
1da177e4 | 5337 | |
dd41f596 IM |
5338 | /* |
5339 | * schedule() is the main scheduler function. | |
5340 | */ | |
ff743345 | 5341 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5342 | { |
5343 | struct task_struct *prev, *next; | |
67ca7bde | 5344 | unsigned long *switch_count; |
dd41f596 | 5345 | struct rq *rq; |
31656519 | 5346 | int cpu; |
dd41f596 | 5347 | |
ff743345 PZ |
5348 | need_resched: |
5349 | preempt_disable(); | |
dd41f596 IM |
5350 | cpu = smp_processor_id(); |
5351 | rq = cpu_rq(cpu); | |
5352 | rcu_qsctr_inc(cpu); | |
5353 | prev = rq->curr; | |
5354 | switch_count = &prev->nivcsw; | |
5355 | ||
5356 | release_kernel_lock(prev); | |
5357 | need_resched_nonpreemptible: | |
5358 | ||
5359 | schedule_debug(prev); | |
1da177e4 | 5360 | |
31656519 | 5361 | if (sched_feat(HRTICK)) |
f333fdc9 | 5362 | hrtick_clear(rq); |
8f4d37ec | 5363 | |
8cd162ce | 5364 | spin_lock_irq(&rq->lock); |
3e51f33f | 5365 | update_rq_clock(rq); |
1e819950 | 5366 | clear_tsk_need_resched(prev); |
1da177e4 | 5367 | |
1da177e4 | 5368 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5369 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5370 | prev->state = TASK_RUNNING; |
16882c1e | 5371 | else |
2e1cb74a | 5372 | deactivate_task(rq, prev, 1); |
dd41f596 | 5373 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5374 | } |
5375 | ||
9a897c5a SR |
5376 | #ifdef CONFIG_SMP |
5377 | if (prev->sched_class->pre_schedule) | |
5378 | prev->sched_class->pre_schedule(rq, prev); | |
5379 | #endif | |
f65eda4f | 5380 | |
dd41f596 | 5381 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5382 | idle_balance(cpu, rq); |
1da177e4 | 5383 | |
df1c99d4 | 5384 | put_prev_task(rq, prev); |
b67802ea | 5385 | next = pick_next_task(rq); |
1da177e4 | 5386 | |
1da177e4 | 5387 | if (likely(prev != next)) { |
673a90a1 | 5388 | sched_info_switch(prev, next); |
564c2b21 | 5389 | perf_counter_task_sched_out(prev, next, cpu); |
673a90a1 | 5390 | |
1da177e4 LT |
5391 | rq->nr_switches++; |
5392 | rq->curr = next; | |
5393 | ++*switch_count; | |
5394 | ||
dd41f596 | 5395 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5396 | /* |
5397 | * the context switch might have flipped the stack from under | |
5398 | * us, hence refresh the local variables. | |
5399 | */ | |
5400 | cpu = smp_processor_id(); | |
5401 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5402 | } else |
5403 | spin_unlock_irq(&rq->lock); | |
5404 | ||
8f4d37ec | 5405 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5406 | goto need_resched_nonpreemptible; |
8f4d37ec | 5407 | |
1da177e4 | 5408 | preempt_enable_no_resched(); |
ff743345 | 5409 | if (need_resched()) |
1da177e4 LT |
5410 | goto need_resched; |
5411 | } | |
1da177e4 LT |
5412 | EXPORT_SYMBOL(schedule); |
5413 | ||
0d66bf6d PZ |
5414 | #ifdef CONFIG_SMP |
5415 | /* | |
5416 | * Look out! "owner" is an entirely speculative pointer | |
5417 | * access and not reliable. | |
5418 | */ | |
5419 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5420 | { | |
5421 | unsigned int cpu; | |
5422 | struct rq *rq; | |
5423 | ||
5424 | if (!sched_feat(OWNER_SPIN)) | |
5425 | return 0; | |
5426 | ||
5427 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5428 | /* | |
5429 | * Need to access the cpu field knowing that | |
5430 | * DEBUG_PAGEALLOC could have unmapped it if | |
5431 | * the mutex owner just released it and exited. | |
5432 | */ | |
5433 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5434 | goto out; | |
5435 | #else | |
5436 | cpu = owner->cpu; | |
5437 | #endif | |
5438 | ||
5439 | /* | |
5440 | * Even if the access succeeded (likely case), | |
5441 | * the cpu field may no longer be valid. | |
5442 | */ | |
5443 | if (cpu >= nr_cpumask_bits) | |
5444 | goto out; | |
5445 | ||
5446 | /* | |
5447 | * We need to validate that we can do a | |
5448 | * get_cpu() and that we have the percpu area. | |
5449 | */ | |
5450 | if (!cpu_online(cpu)) | |
5451 | goto out; | |
5452 | ||
5453 | rq = cpu_rq(cpu); | |
5454 | ||
5455 | for (;;) { | |
5456 | /* | |
5457 | * Owner changed, break to re-assess state. | |
5458 | */ | |
5459 | if (lock->owner != owner) | |
5460 | break; | |
5461 | ||
5462 | /* | |
5463 | * Is that owner really running on that cpu? | |
5464 | */ | |
5465 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5466 | return 0; | |
5467 | ||
5468 | cpu_relax(); | |
5469 | } | |
5470 | out: | |
5471 | return 1; | |
5472 | } | |
5473 | #endif | |
5474 | ||
1da177e4 LT |
5475 | #ifdef CONFIG_PREEMPT |
5476 | /* | |
2ed6e34f | 5477 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5478 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5479 | * occur there and call schedule directly. |
5480 | */ | |
5481 | asmlinkage void __sched preempt_schedule(void) | |
5482 | { | |
5483 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5484 | |
1da177e4 LT |
5485 | /* |
5486 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5487 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5488 | */ |
beed33a8 | 5489 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5490 | return; |
5491 | ||
3a5c359a AK |
5492 | do { |
5493 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5494 | schedule(); |
3a5c359a | 5495 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5496 | |
3a5c359a AK |
5497 | /* |
5498 | * Check again in case we missed a preemption opportunity | |
5499 | * between schedule and now. | |
5500 | */ | |
5501 | barrier(); | |
5ed0cec0 | 5502 | } while (need_resched()); |
1da177e4 | 5503 | } |
1da177e4 LT |
5504 | EXPORT_SYMBOL(preempt_schedule); |
5505 | ||
5506 | /* | |
2ed6e34f | 5507 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5508 | * off of irq context. |
5509 | * Note, that this is called and return with irqs disabled. This will | |
5510 | * protect us against recursive calling from irq. | |
5511 | */ | |
5512 | asmlinkage void __sched preempt_schedule_irq(void) | |
5513 | { | |
5514 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5515 | |
2ed6e34f | 5516 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5517 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5518 | ||
3a5c359a AK |
5519 | do { |
5520 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5521 | local_irq_enable(); |
5522 | schedule(); | |
5523 | local_irq_disable(); | |
3a5c359a | 5524 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5525 | |
3a5c359a AK |
5526 | /* |
5527 | * Check again in case we missed a preemption opportunity | |
5528 | * between schedule and now. | |
5529 | */ | |
5530 | barrier(); | |
5ed0cec0 | 5531 | } while (need_resched()); |
1da177e4 LT |
5532 | } |
5533 | ||
5534 | #endif /* CONFIG_PREEMPT */ | |
5535 | ||
95cdf3b7 IM |
5536 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5537 | void *key) | |
1da177e4 | 5538 | { |
48f24c4d | 5539 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5540 | } |
1da177e4 LT |
5541 | EXPORT_SYMBOL(default_wake_function); |
5542 | ||
5543 | /* | |
41a2d6cf IM |
5544 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5545 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5546 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5547 | * | |
5548 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5549 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5550 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5551 | */ | |
78ddb08f | 5552 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5553 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5554 | { |
2e45874c | 5555 | wait_queue_t *curr, *next; |
1da177e4 | 5556 | |
2e45874c | 5557 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5558 | unsigned flags = curr->flags; |
5559 | ||
1da177e4 | 5560 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5561 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5562 | break; |
5563 | } | |
5564 | } | |
5565 | ||
5566 | /** | |
5567 | * __wake_up - wake up threads blocked on a waitqueue. | |
5568 | * @q: the waitqueue | |
5569 | * @mode: which threads | |
5570 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5571 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5572 | * |
5573 | * It may be assumed that this function implies a write memory barrier before | |
5574 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5575 | */ |
7ad5b3a5 | 5576 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5577 | int nr_exclusive, void *key) |
1da177e4 LT |
5578 | { |
5579 | unsigned long flags; | |
5580 | ||
5581 | spin_lock_irqsave(&q->lock, flags); | |
5582 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5583 | spin_unlock_irqrestore(&q->lock, flags); | |
5584 | } | |
1da177e4 LT |
5585 | EXPORT_SYMBOL(__wake_up); |
5586 | ||
5587 | /* | |
5588 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5589 | */ | |
7ad5b3a5 | 5590 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5591 | { |
5592 | __wake_up_common(q, mode, 1, 0, NULL); | |
5593 | } | |
5594 | ||
4ede816a DL |
5595 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5596 | { | |
5597 | __wake_up_common(q, mode, 1, 0, key); | |
5598 | } | |
5599 | ||
1da177e4 | 5600 | /** |
4ede816a | 5601 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5602 | * @q: the waitqueue |
5603 | * @mode: which threads | |
5604 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5605 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5606 | * |
5607 | * The sync wakeup differs that the waker knows that it will schedule | |
5608 | * away soon, so while the target thread will be woken up, it will not | |
5609 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5610 | * with each other. This can prevent needless bouncing between CPUs. | |
5611 | * | |
5612 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5613 | * |
5614 | * It may be assumed that this function implies a write memory barrier before | |
5615 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5616 | */ |
4ede816a DL |
5617 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5618 | int nr_exclusive, void *key) | |
1da177e4 LT |
5619 | { |
5620 | unsigned long flags; | |
5621 | int sync = 1; | |
5622 | ||
5623 | if (unlikely(!q)) | |
5624 | return; | |
5625 | ||
5626 | if (unlikely(!nr_exclusive)) | |
5627 | sync = 0; | |
5628 | ||
5629 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5630 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5631 | spin_unlock_irqrestore(&q->lock, flags); |
5632 | } | |
4ede816a DL |
5633 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5634 | ||
5635 | /* | |
5636 | * __wake_up_sync - see __wake_up_sync_key() | |
5637 | */ | |
5638 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5639 | { | |
5640 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5641 | } | |
1da177e4 LT |
5642 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5643 | ||
65eb3dc6 KD |
5644 | /** |
5645 | * complete: - signals a single thread waiting on this completion | |
5646 | * @x: holds the state of this particular completion | |
5647 | * | |
5648 | * This will wake up a single thread waiting on this completion. Threads will be | |
5649 | * awakened in the same order in which they were queued. | |
5650 | * | |
5651 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5652 | * |
5653 | * It may be assumed that this function implies a write memory barrier before | |
5654 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5655 | */ |
b15136e9 | 5656 | void complete(struct completion *x) |
1da177e4 LT |
5657 | { |
5658 | unsigned long flags; | |
5659 | ||
5660 | spin_lock_irqsave(&x->wait.lock, flags); | |
5661 | x->done++; | |
d9514f6c | 5662 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5663 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5664 | } | |
5665 | EXPORT_SYMBOL(complete); | |
5666 | ||
65eb3dc6 KD |
5667 | /** |
5668 | * complete_all: - signals all threads waiting on this completion | |
5669 | * @x: holds the state of this particular completion | |
5670 | * | |
5671 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5672 | * |
5673 | * It may be assumed that this function implies a write memory barrier before | |
5674 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5675 | */ |
b15136e9 | 5676 | void complete_all(struct completion *x) |
1da177e4 LT |
5677 | { |
5678 | unsigned long flags; | |
5679 | ||
5680 | spin_lock_irqsave(&x->wait.lock, flags); | |
5681 | x->done += UINT_MAX/2; | |
d9514f6c | 5682 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5683 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5684 | } | |
5685 | EXPORT_SYMBOL(complete_all); | |
5686 | ||
8cbbe86d AK |
5687 | static inline long __sched |
5688 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5689 | { |
1da177e4 LT |
5690 | if (!x->done) { |
5691 | DECLARE_WAITQUEUE(wait, current); | |
5692 | ||
5693 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5694 | __add_wait_queue_tail(&x->wait, &wait); | |
5695 | do { | |
94d3d824 | 5696 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5697 | timeout = -ERESTARTSYS; |
5698 | break; | |
8cbbe86d AK |
5699 | } |
5700 | __set_current_state(state); | |
1da177e4 LT |
5701 | spin_unlock_irq(&x->wait.lock); |
5702 | timeout = schedule_timeout(timeout); | |
5703 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5704 | } while (!x->done && timeout); |
1da177e4 | 5705 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5706 | if (!x->done) |
5707 | return timeout; | |
1da177e4 LT |
5708 | } |
5709 | x->done--; | |
ea71a546 | 5710 | return timeout ?: 1; |
1da177e4 | 5711 | } |
1da177e4 | 5712 | |
8cbbe86d AK |
5713 | static long __sched |
5714 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5715 | { |
1da177e4 LT |
5716 | might_sleep(); |
5717 | ||
5718 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5719 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5720 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5721 | return timeout; |
5722 | } | |
1da177e4 | 5723 | |
65eb3dc6 KD |
5724 | /** |
5725 | * wait_for_completion: - waits for completion of a task | |
5726 | * @x: holds the state of this particular completion | |
5727 | * | |
5728 | * This waits to be signaled for completion of a specific task. It is NOT | |
5729 | * interruptible and there is no timeout. | |
5730 | * | |
5731 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5732 | * and interrupt capability. Also see complete(). | |
5733 | */ | |
b15136e9 | 5734 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5735 | { |
5736 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5737 | } |
8cbbe86d | 5738 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5739 | |
65eb3dc6 KD |
5740 | /** |
5741 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5742 | * @x: holds the state of this particular completion | |
5743 | * @timeout: timeout value in jiffies | |
5744 | * | |
5745 | * This waits for either a completion of a specific task to be signaled or for a | |
5746 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5747 | * interruptible. | |
5748 | */ | |
b15136e9 | 5749 | unsigned long __sched |
8cbbe86d | 5750 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5751 | { |
8cbbe86d | 5752 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5753 | } |
8cbbe86d | 5754 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5755 | |
65eb3dc6 KD |
5756 | /** |
5757 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5758 | * @x: holds the state of this particular completion | |
5759 | * | |
5760 | * This waits for completion of a specific task to be signaled. It is | |
5761 | * interruptible. | |
5762 | */ | |
8cbbe86d | 5763 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5764 | { |
51e97990 AK |
5765 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5766 | if (t == -ERESTARTSYS) | |
5767 | return t; | |
5768 | return 0; | |
0fec171c | 5769 | } |
8cbbe86d | 5770 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5771 | |
65eb3dc6 KD |
5772 | /** |
5773 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5774 | * @x: holds the state of this particular completion | |
5775 | * @timeout: timeout value in jiffies | |
5776 | * | |
5777 | * This waits for either a completion of a specific task to be signaled or for a | |
5778 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5779 | */ | |
b15136e9 | 5780 | unsigned long __sched |
8cbbe86d AK |
5781 | wait_for_completion_interruptible_timeout(struct completion *x, |
5782 | unsigned long timeout) | |
0fec171c | 5783 | { |
8cbbe86d | 5784 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5785 | } |
8cbbe86d | 5786 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5787 | |
65eb3dc6 KD |
5788 | /** |
5789 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5790 | * @x: holds the state of this particular completion | |
5791 | * | |
5792 | * This waits to be signaled for completion of a specific task. It can be | |
5793 | * interrupted by a kill signal. | |
5794 | */ | |
009e577e MW |
5795 | int __sched wait_for_completion_killable(struct completion *x) |
5796 | { | |
5797 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5798 | if (t == -ERESTARTSYS) | |
5799 | return t; | |
5800 | return 0; | |
5801 | } | |
5802 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5803 | ||
be4de352 DC |
5804 | /** |
5805 | * try_wait_for_completion - try to decrement a completion without blocking | |
5806 | * @x: completion structure | |
5807 | * | |
5808 | * Returns: 0 if a decrement cannot be done without blocking | |
5809 | * 1 if a decrement succeeded. | |
5810 | * | |
5811 | * If a completion is being used as a counting completion, | |
5812 | * attempt to decrement the counter without blocking. This | |
5813 | * enables us to avoid waiting if the resource the completion | |
5814 | * is protecting is not available. | |
5815 | */ | |
5816 | bool try_wait_for_completion(struct completion *x) | |
5817 | { | |
5818 | int ret = 1; | |
5819 | ||
5820 | spin_lock_irq(&x->wait.lock); | |
5821 | if (!x->done) | |
5822 | ret = 0; | |
5823 | else | |
5824 | x->done--; | |
5825 | spin_unlock_irq(&x->wait.lock); | |
5826 | return ret; | |
5827 | } | |
5828 | EXPORT_SYMBOL(try_wait_for_completion); | |
5829 | ||
5830 | /** | |
5831 | * completion_done - Test to see if a completion has any waiters | |
5832 | * @x: completion structure | |
5833 | * | |
5834 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5835 | * 1 if there are no waiters. | |
5836 | * | |
5837 | */ | |
5838 | bool completion_done(struct completion *x) | |
5839 | { | |
5840 | int ret = 1; | |
5841 | ||
5842 | spin_lock_irq(&x->wait.lock); | |
5843 | if (!x->done) | |
5844 | ret = 0; | |
5845 | spin_unlock_irq(&x->wait.lock); | |
5846 | return ret; | |
5847 | } | |
5848 | EXPORT_SYMBOL(completion_done); | |
5849 | ||
8cbbe86d AK |
5850 | static long __sched |
5851 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5852 | { |
0fec171c IM |
5853 | unsigned long flags; |
5854 | wait_queue_t wait; | |
5855 | ||
5856 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5857 | |
8cbbe86d | 5858 | __set_current_state(state); |
1da177e4 | 5859 | |
8cbbe86d AK |
5860 | spin_lock_irqsave(&q->lock, flags); |
5861 | __add_wait_queue(q, &wait); | |
5862 | spin_unlock(&q->lock); | |
5863 | timeout = schedule_timeout(timeout); | |
5864 | spin_lock_irq(&q->lock); | |
5865 | __remove_wait_queue(q, &wait); | |
5866 | spin_unlock_irqrestore(&q->lock, flags); | |
5867 | ||
5868 | return timeout; | |
5869 | } | |
5870 | ||
5871 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5872 | { | |
5873 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5874 | } |
1da177e4 LT |
5875 | EXPORT_SYMBOL(interruptible_sleep_on); |
5876 | ||
0fec171c | 5877 | long __sched |
95cdf3b7 | 5878 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5879 | { |
8cbbe86d | 5880 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5881 | } |
1da177e4 LT |
5882 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5883 | ||
0fec171c | 5884 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5885 | { |
8cbbe86d | 5886 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5887 | } |
1da177e4 LT |
5888 | EXPORT_SYMBOL(sleep_on); |
5889 | ||
0fec171c | 5890 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5891 | { |
8cbbe86d | 5892 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5893 | } |
1da177e4 LT |
5894 | EXPORT_SYMBOL(sleep_on_timeout); |
5895 | ||
b29739f9 IM |
5896 | #ifdef CONFIG_RT_MUTEXES |
5897 | ||
5898 | /* | |
5899 | * rt_mutex_setprio - set the current priority of a task | |
5900 | * @p: task | |
5901 | * @prio: prio value (kernel-internal form) | |
5902 | * | |
5903 | * This function changes the 'effective' priority of a task. It does | |
5904 | * not touch ->normal_prio like __setscheduler(). | |
5905 | * | |
5906 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5907 | */ | |
36c8b586 | 5908 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5909 | { |
5910 | unsigned long flags; | |
83b699ed | 5911 | int oldprio, on_rq, running; |
70b97a7f | 5912 | struct rq *rq; |
cb469845 | 5913 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5914 | |
5915 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5916 | ||
5917 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5918 | update_rq_clock(rq); |
b29739f9 | 5919 | |
d5f9f942 | 5920 | oldprio = p->prio; |
dd41f596 | 5921 | on_rq = p->se.on_rq; |
051a1d1a | 5922 | running = task_current(rq, p); |
0e1f3483 | 5923 | if (on_rq) |
69be72c1 | 5924 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5925 | if (running) |
5926 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5927 | |
5928 | if (rt_prio(prio)) | |
5929 | p->sched_class = &rt_sched_class; | |
5930 | else | |
5931 | p->sched_class = &fair_sched_class; | |
5932 | ||
b29739f9 IM |
5933 | p->prio = prio; |
5934 | ||
0e1f3483 HS |
5935 | if (running) |
5936 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5937 | if (on_rq) { |
8159f87e | 5938 | enqueue_task(rq, p, 0); |
cb469845 SR |
5939 | |
5940 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5941 | } |
5942 | task_rq_unlock(rq, &flags); | |
5943 | } | |
5944 | ||
5945 | #endif | |
5946 | ||
36c8b586 | 5947 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5948 | { |
dd41f596 | 5949 | int old_prio, delta, on_rq; |
1da177e4 | 5950 | unsigned long flags; |
70b97a7f | 5951 | struct rq *rq; |
1da177e4 LT |
5952 | |
5953 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5954 | return; | |
5955 | /* | |
5956 | * We have to be careful, if called from sys_setpriority(), | |
5957 | * the task might be in the middle of scheduling on another CPU. | |
5958 | */ | |
5959 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5960 | update_rq_clock(rq); |
1da177e4 LT |
5961 | /* |
5962 | * The RT priorities are set via sched_setscheduler(), but we still | |
5963 | * allow the 'normal' nice value to be set - but as expected | |
5964 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5965 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5966 | */ |
e05606d3 | 5967 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5968 | p->static_prio = NICE_TO_PRIO(nice); |
5969 | goto out_unlock; | |
5970 | } | |
dd41f596 | 5971 | on_rq = p->se.on_rq; |
c09595f6 | 5972 | if (on_rq) |
69be72c1 | 5973 | dequeue_task(rq, p, 0); |
1da177e4 | 5974 | |
1da177e4 | 5975 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5976 | set_load_weight(p); |
b29739f9 IM |
5977 | old_prio = p->prio; |
5978 | p->prio = effective_prio(p); | |
5979 | delta = p->prio - old_prio; | |
1da177e4 | 5980 | |
dd41f596 | 5981 | if (on_rq) { |
8159f87e | 5982 | enqueue_task(rq, p, 0); |
1da177e4 | 5983 | /* |
d5f9f942 AM |
5984 | * If the task increased its priority or is running and |
5985 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5986 | */ |
d5f9f942 | 5987 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5988 | resched_task(rq->curr); |
5989 | } | |
5990 | out_unlock: | |
5991 | task_rq_unlock(rq, &flags); | |
5992 | } | |
1da177e4 LT |
5993 | EXPORT_SYMBOL(set_user_nice); |
5994 | ||
e43379f1 MM |
5995 | /* |
5996 | * can_nice - check if a task can reduce its nice value | |
5997 | * @p: task | |
5998 | * @nice: nice value | |
5999 | */ | |
36c8b586 | 6000 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6001 | { |
024f4747 MM |
6002 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
6003 | int nice_rlim = 20 - nice; | |
48f24c4d | 6004 | |
e43379f1 MM |
6005 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
6006 | capable(CAP_SYS_NICE)); | |
6007 | } | |
6008 | ||
1da177e4 LT |
6009 | #ifdef __ARCH_WANT_SYS_NICE |
6010 | ||
6011 | /* | |
6012 | * sys_nice - change the priority of the current process. | |
6013 | * @increment: priority increment | |
6014 | * | |
6015 | * sys_setpriority is a more generic, but much slower function that | |
6016 | * does similar things. | |
6017 | */ | |
5add95d4 | 6018 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6019 | { |
48f24c4d | 6020 | long nice, retval; |
1da177e4 LT |
6021 | |
6022 | /* | |
6023 | * Setpriority might change our priority at the same moment. | |
6024 | * We don't have to worry. Conceptually one call occurs first | |
6025 | * and we have a single winner. | |
6026 | */ | |
e43379f1 MM |
6027 | if (increment < -40) |
6028 | increment = -40; | |
1da177e4 LT |
6029 | if (increment > 40) |
6030 | increment = 40; | |
6031 | ||
2b8f836f | 6032 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
6033 | if (nice < -20) |
6034 | nice = -20; | |
6035 | if (nice > 19) | |
6036 | nice = 19; | |
6037 | ||
e43379f1 MM |
6038 | if (increment < 0 && !can_nice(current, nice)) |
6039 | return -EPERM; | |
6040 | ||
1da177e4 LT |
6041 | retval = security_task_setnice(current, nice); |
6042 | if (retval) | |
6043 | return retval; | |
6044 | ||
6045 | set_user_nice(current, nice); | |
6046 | return 0; | |
6047 | } | |
6048 | ||
6049 | #endif | |
6050 | ||
6051 | /** | |
6052 | * task_prio - return the priority value of a given task. | |
6053 | * @p: the task in question. | |
6054 | * | |
6055 | * This is the priority value as seen by users in /proc. | |
6056 | * RT tasks are offset by -200. Normal tasks are centered | |
6057 | * around 0, value goes from -16 to +15. | |
6058 | */ | |
36c8b586 | 6059 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6060 | { |
6061 | return p->prio - MAX_RT_PRIO; | |
6062 | } | |
6063 | ||
6064 | /** | |
6065 | * task_nice - return the nice value of a given task. | |
6066 | * @p: the task in question. | |
6067 | */ | |
36c8b586 | 6068 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6069 | { |
6070 | return TASK_NICE(p); | |
6071 | } | |
150d8bed | 6072 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6073 | |
6074 | /** | |
6075 | * idle_cpu - is a given cpu idle currently? | |
6076 | * @cpu: the processor in question. | |
6077 | */ | |
6078 | int idle_cpu(int cpu) | |
6079 | { | |
6080 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6081 | } | |
6082 | ||
1da177e4 LT |
6083 | /** |
6084 | * idle_task - return the idle task for a given cpu. | |
6085 | * @cpu: the processor in question. | |
6086 | */ | |
36c8b586 | 6087 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6088 | { |
6089 | return cpu_rq(cpu)->idle; | |
6090 | } | |
6091 | ||
6092 | /** | |
6093 | * find_process_by_pid - find a process with a matching PID value. | |
6094 | * @pid: the pid in question. | |
6095 | */ | |
a9957449 | 6096 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6097 | { |
228ebcbe | 6098 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6099 | } |
6100 | ||
6101 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6102 | static void |
6103 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6104 | { |
dd41f596 | 6105 | BUG_ON(p->se.on_rq); |
48f24c4d | 6106 | |
1da177e4 | 6107 | p->policy = policy; |
dd41f596 IM |
6108 | switch (p->policy) { |
6109 | case SCHED_NORMAL: | |
6110 | case SCHED_BATCH: | |
6111 | case SCHED_IDLE: | |
6112 | p->sched_class = &fair_sched_class; | |
6113 | break; | |
6114 | case SCHED_FIFO: | |
6115 | case SCHED_RR: | |
6116 | p->sched_class = &rt_sched_class; | |
6117 | break; | |
6118 | } | |
6119 | ||
1da177e4 | 6120 | p->rt_priority = prio; |
b29739f9 IM |
6121 | p->normal_prio = normal_prio(p); |
6122 | /* we are holding p->pi_lock already */ | |
6123 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 6124 | set_load_weight(p); |
1da177e4 LT |
6125 | } |
6126 | ||
c69e8d9c DH |
6127 | /* |
6128 | * check the target process has a UID that matches the current process's | |
6129 | */ | |
6130 | static bool check_same_owner(struct task_struct *p) | |
6131 | { | |
6132 | const struct cred *cred = current_cred(), *pcred; | |
6133 | bool match; | |
6134 | ||
6135 | rcu_read_lock(); | |
6136 | pcred = __task_cred(p); | |
6137 | match = (cred->euid == pcred->euid || | |
6138 | cred->euid == pcred->uid); | |
6139 | rcu_read_unlock(); | |
6140 | return match; | |
6141 | } | |
6142 | ||
961ccddd RR |
6143 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6144 | struct sched_param *param, bool user) | |
1da177e4 | 6145 | { |
83b699ed | 6146 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6147 | unsigned long flags; |
cb469845 | 6148 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6149 | struct rq *rq; |
ca94c442 | 6150 | int reset_on_fork; |
1da177e4 | 6151 | |
66e5393a SR |
6152 | /* may grab non-irq protected spin_locks */ |
6153 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6154 | recheck: |
6155 | /* double check policy once rq lock held */ | |
ca94c442 LP |
6156 | if (policy < 0) { |
6157 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 6158 | policy = oldpolicy = p->policy; |
ca94c442 LP |
6159 | } else { |
6160 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
6161 | policy &= ~SCHED_RESET_ON_FORK; | |
6162 | ||
6163 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
6164 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
6165 | policy != SCHED_IDLE) | |
6166 | return -EINVAL; | |
6167 | } | |
6168 | ||
1da177e4 LT |
6169 | /* |
6170 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6171 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6172 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6173 | */ |
6174 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6175 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6176 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6177 | return -EINVAL; |
e05606d3 | 6178 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6179 | return -EINVAL; |
6180 | ||
37e4ab3f OC |
6181 | /* |
6182 | * Allow unprivileged RT tasks to decrease priority: | |
6183 | */ | |
961ccddd | 6184 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6185 | if (rt_policy(policy)) { |
8dc3e909 | 6186 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6187 | |
6188 | if (!lock_task_sighand(p, &flags)) | |
6189 | return -ESRCH; | |
6190 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6191 | unlock_task_sighand(p, &flags); | |
6192 | ||
6193 | /* can't set/change the rt policy */ | |
6194 | if (policy != p->policy && !rlim_rtprio) | |
6195 | return -EPERM; | |
6196 | ||
6197 | /* can't increase priority */ | |
6198 | if (param->sched_priority > p->rt_priority && | |
6199 | param->sched_priority > rlim_rtprio) | |
6200 | return -EPERM; | |
6201 | } | |
dd41f596 IM |
6202 | /* |
6203 | * Like positive nice levels, dont allow tasks to | |
6204 | * move out of SCHED_IDLE either: | |
6205 | */ | |
6206 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6207 | return -EPERM; | |
5fe1d75f | 6208 | |
37e4ab3f | 6209 | /* can't change other user's priorities */ |
c69e8d9c | 6210 | if (!check_same_owner(p)) |
37e4ab3f | 6211 | return -EPERM; |
ca94c442 LP |
6212 | |
6213 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
6214 | if (p->sched_reset_on_fork && !reset_on_fork) | |
6215 | return -EPERM; | |
37e4ab3f | 6216 | } |
1da177e4 | 6217 | |
725aad24 | 6218 | if (user) { |
b68aa230 | 6219 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6220 | /* |
6221 | * Do not allow realtime tasks into groups that have no runtime | |
6222 | * assigned. | |
6223 | */ | |
9a7e0b18 PZ |
6224 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6225 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6226 | return -EPERM; |
b68aa230 PZ |
6227 | #endif |
6228 | ||
725aad24 JF |
6229 | retval = security_task_setscheduler(p, policy, param); |
6230 | if (retval) | |
6231 | return retval; | |
6232 | } | |
6233 | ||
b29739f9 IM |
6234 | /* |
6235 | * make sure no PI-waiters arrive (or leave) while we are | |
6236 | * changing the priority of the task: | |
6237 | */ | |
6238 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6239 | /* |
6240 | * To be able to change p->policy safely, the apropriate | |
6241 | * runqueue lock must be held. | |
6242 | */ | |
b29739f9 | 6243 | rq = __task_rq_lock(p); |
1da177e4 LT |
6244 | /* recheck policy now with rq lock held */ |
6245 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6246 | policy = oldpolicy = -1; | |
b29739f9 IM |
6247 | __task_rq_unlock(rq); |
6248 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6249 | goto recheck; |
6250 | } | |
2daa3577 | 6251 | update_rq_clock(rq); |
dd41f596 | 6252 | on_rq = p->se.on_rq; |
051a1d1a | 6253 | running = task_current(rq, p); |
0e1f3483 | 6254 | if (on_rq) |
2e1cb74a | 6255 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6256 | if (running) |
6257 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6258 | |
ca94c442 LP |
6259 | p->sched_reset_on_fork = reset_on_fork; |
6260 | ||
1da177e4 | 6261 | oldprio = p->prio; |
dd41f596 | 6262 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6263 | |
0e1f3483 HS |
6264 | if (running) |
6265 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6266 | if (on_rq) { |
6267 | activate_task(rq, p, 0); | |
cb469845 SR |
6268 | |
6269 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6270 | } |
b29739f9 IM |
6271 | __task_rq_unlock(rq); |
6272 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6273 | ||
95e02ca9 TG |
6274 | rt_mutex_adjust_pi(p); |
6275 | ||
1da177e4 LT |
6276 | return 0; |
6277 | } | |
961ccddd RR |
6278 | |
6279 | /** | |
6280 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6281 | * @p: the task in question. | |
6282 | * @policy: new policy. | |
6283 | * @param: structure containing the new RT priority. | |
6284 | * | |
6285 | * NOTE that the task may be already dead. | |
6286 | */ | |
6287 | int sched_setscheduler(struct task_struct *p, int policy, | |
6288 | struct sched_param *param) | |
6289 | { | |
6290 | return __sched_setscheduler(p, policy, param, true); | |
6291 | } | |
1da177e4 LT |
6292 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6293 | ||
961ccddd RR |
6294 | /** |
6295 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6296 | * @p: the task in question. | |
6297 | * @policy: new policy. | |
6298 | * @param: structure containing the new RT priority. | |
6299 | * | |
6300 | * Just like sched_setscheduler, only don't bother checking if the | |
6301 | * current context has permission. For example, this is needed in | |
6302 | * stop_machine(): we create temporary high priority worker threads, | |
6303 | * but our caller might not have that capability. | |
6304 | */ | |
6305 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6306 | struct sched_param *param) | |
6307 | { | |
6308 | return __sched_setscheduler(p, policy, param, false); | |
6309 | } | |
6310 | ||
95cdf3b7 IM |
6311 | static int |
6312 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6313 | { |
1da177e4 LT |
6314 | struct sched_param lparam; |
6315 | struct task_struct *p; | |
36c8b586 | 6316 | int retval; |
1da177e4 LT |
6317 | |
6318 | if (!param || pid < 0) | |
6319 | return -EINVAL; | |
6320 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6321 | return -EFAULT; | |
5fe1d75f ON |
6322 | |
6323 | rcu_read_lock(); | |
6324 | retval = -ESRCH; | |
1da177e4 | 6325 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6326 | if (p != NULL) |
6327 | retval = sched_setscheduler(p, policy, &lparam); | |
6328 | rcu_read_unlock(); | |
36c8b586 | 6329 | |
1da177e4 LT |
6330 | return retval; |
6331 | } | |
6332 | ||
6333 | /** | |
6334 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6335 | * @pid: the pid in question. | |
6336 | * @policy: new policy. | |
6337 | * @param: structure containing the new RT priority. | |
6338 | */ | |
5add95d4 HC |
6339 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6340 | struct sched_param __user *, param) | |
1da177e4 | 6341 | { |
c21761f1 JB |
6342 | /* negative values for policy are not valid */ |
6343 | if (policy < 0) | |
6344 | return -EINVAL; | |
6345 | ||
1da177e4 LT |
6346 | return do_sched_setscheduler(pid, policy, param); |
6347 | } | |
6348 | ||
6349 | /** | |
6350 | * sys_sched_setparam - set/change the RT priority of a thread | |
6351 | * @pid: the pid in question. | |
6352 | * @param: structure containing the new RT priority. | |
6353 | */ | |
5add95d4 | 6354 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6355 | { |
6356 | return do_sched_setscheduler(pid, -1, param); | |
6357 | } | |
6358 | ||
6359 | /** | |
6360 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6361 | * @pid: the pid in question. | |
6362 | */ | |
5add95d4 | 6363 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6364 | { |
36c8b586 | 6365 | struct task_struct *p; |
3a5c359a | 6366 | int retval; |
1da177e4 LT |
6367 | |
6368 | if (pid < 0) | |
3a5c359a | 6369 | return -EINVAL; |
1da177e4 LT |
6370 | |
6371 | retval = -ESRCH; | |
6372 | read_lock(&tasklist_lock); | |
6373 | p = find_process_by_pid(pid); | |
6374 | if (p) { | |
6375 | retval = security_task_getscheduler(p); | |
6376 | if (!retval) | |
ca94c442 LP |
6377 | retval = p->policy |
6378 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 LT |
6379 | } |
6380 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6381 | return retval; |
6382 | } | |
6383 | ||
6384 | /** | |
ca94c442 | 6385 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6386 | * @pid: the pid in question. |
6387 | * @param: structure containing the RT priority. | |
6388 | */ | |
5add95d4 | 6389 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6390 | { |
6391 | struct sched_param lp; | |
36c8b586 | 6392 | struct task_struct *p; |
3a5c359a | 6393 | int retval; |
1da177e4 LT |
6394 | |
6395 | if (!param || pid < 0) | |
3a5c359a | 6396 | return -EINVAL; |
1da177e4 LT |
6397 | |
6398 | read_lock(&tasklist_lock); | |
6399 | p = find_process_by_pid(pid); | |
6400 | retval = -ESRCH; | |
6401 | if (!p) | |
6402 | goto out_unlock; | |
6403 | ||
6404 | retval = security_task_getscheduler(p); | |
6405 | if (retval) | |
6406 | goto out_unlock; | |
6407 | ||
6408 | lp.sched_priority = p->rt_priority; | |
6409 | read_unlock(&tasklist_lock); | |
6410 | ||
6411 | /* | |
6412 | * This one might sleep, we cannot do it with a spinlock held ... | |
6413 | */ | |
6414 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6415 | ||
1da177e4 LT |
6416 | return retval; |
6417 | ||
6418 | out_unlock: | |
6419 | read_unlock(&tasklist_lock); | |
6420 | return retval; | |
6421 | } | |
6422 | ||
96f874e2 | 6423 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6424 | { |
5a16f3d3 | 6425 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6426 | struct task_struct *p; |
6427 | int retval; | |
1da177e4 | 6428 | |
95402b38 | 6429 | get_online_cpus(); |
1da177e4 LT |
6430 | read_lock(&tasklist_lock); |
6431 | ||
6432 | p = find_process_by_pid(pid); | |
6433 | if (!p) { | |
6434 | read_unlock(&tasklist_lock); | |
95402b38 | 6435 | put_online_cpus(); |
1da177e4 LT |
6436 | return -ESRCH; |
6437 | } | |
6438 | ||
6439 | /* | |
6440 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6441 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6442 | * usage count and then drop tasklist_lock. |
6443 | */ | |
6444 | get_task_struct(p); | |
6445 | read_unlock(&tasklist_lock); | |
6446 | ||
5a16f3d3 RR |
6447 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6448 | retval = -ENOMEM; | |
6449 | goto out_put_task; | |
6450 | } | |
6451 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6452 | retval = -ENOMEM; | |
6453 | goto out_free_cpus_allowed; | |
6454 | } | |
1da177e4 | 6455 | retval = -EPERM; |
c69e8d9c | 6456 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6457 | goto out_unlock; |
6458 | ||
e7834f8f DQ |
6459 | retval = security_task_setscheduler(p, 0, NULL); |
6460 | if (retval) | |
6461 | goto out_unlock; | |
6462 | ||
5a16f3d3 RR |
6463 | cpuset_cpus_allowed(p, cpus_allowed); |
6464 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6465 | again: |
5a16f3d3 | 6466 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6467 | |
8707d8b8 | 6468 | if (!retval) { |
5a16f3d3 RR |
6469 | cpuset_cpus_allowed(p, cpus_allowed); |
6470 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6471 | /* |
6472 | * We must have raced with a concurrent cpuset | |
6473 | * update. Just reset the cpus_allowed to the | |
6474 | * cpuset's cpus_allowed | |
6475 | */ | |
5a16f3d3 | 6476 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6477 | goto again; |
6478 | } | |
6479 | } | |
1da177e4 | 6480 | out_unlock: |
5a16f3d3 RR |
6481 | free_cpumask_var(new_mask); |
6482 | out_free_cpus_allowed: | |
6483 | free_cpumask_var(cpus_allowed); | |
6484 | out_put_task: | |
1da177e4 | 6485 | put_task_struct(p); |
95402b38 | 6486 | put_online_cpus(); |
1da177e4 LT |
6487 | return retval; |
6488 | } | |
6489 | ||
6490 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6491 | struct cpumask *new_mask) |
1da177e4 | 6492 | { |
96f874e2 RR |
6493 | if (len < cpumask_size()) |
6494 | cpumask_clear(new_mask); | |
6495 | else if (len > cpumask_size()) | |
6496 | len = cpumask_size(); | |
6497 | ||
1da177e4 LT |
6498 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6499 | } | |
6500 | ||
6501 | /** | |
6502 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6503 | * @pid: pid of the process | |
6504 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6505 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6506 | */ | |
5add95d4 HC |
6507 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6508 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6509 | { |
5a16f3d3 | 6510 | cpumask_var_t new_mask; |
1da177e4 LT |
6511 | int retval; |
6512 | ||
5a16f3d3 RR |
6513 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6514 | return -ENOMEM; | |
1da177e4 | 6515 | |
5a16f3d3 RR |
6516 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6517 | if (retval == 0) | |
6518 | retval = sched_setaffinity(pid, new_mask); | |
6519 | free_cpumask_var(new_mask); | |
6520 | return retval; | |
1da177e4 LT |
6521 | } |
6522 | ||
96f874e2 | 6523 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6524 | { |
36c8b586 | 6525 | struct task_struct *p; |
1da177e4 | 6526 | int retval; |
1da177e4 | 6527 | |
95402b38 | 6528 | get_online_cpus(); |
1da177e4 LT |
6529 | read_lock(&tasklist_lock); |
6530 | ||
6531 | retval = -ESRCH; | |
6532 | p = find_process_by_pid(pid); | |
6533 | if (!p) | |
6534 | goto out_unlock; | |
6535 | ||
e7834f8f DQ |
6536 | retval = security_task_getscheduler(p); |
6537 | if (retval) | |
6538 | goto out_unlock; | |
6539 | ||
96f874e2 | 6540 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6541 | |
6542 | out_unlock: | |
6543 | read_unlock(&tasklist_lock); | |
95402b38 | 6544 | put_online_cpus(); |
1da177e4 | 6545 | |
9531b62f | 6546 | return retval; |
1da177e4 LT |
6547 | } |
6548 | ||
6549 | /** | |
6550 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6551 | * @pid: pid of the process | |
6552 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6553 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6554 | */ | |
5add95d4 HC |
6555 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6556 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6557 | { |
6558 | int ret; | |
f17c8607 | 6559 | cpumask_var_t mask; |
1da177e4 | 6560 | |
f17c8607 | 6561 | if (len < cpumask_size()) |
1da177e4 LT |
6562 | return -EINVAL; |
6563 | ||
f17c8607 RR |
6564 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6565 | return -ENOMEM; | |
1da177e4 | 6566 | |
f17c8607 RR |
6567 | ret = sched_getaffinity(pid, mask); |
6568 | if (ret == 0) { | |
6569 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6570 | ret = -EFAULT; | |
6571 | else | |
6572 | ret = cpumask_size(); | |
6573 | } | |
6574 | free_cpumask_var(mask); | |
1da177e4 | 6575 | |
f17c8607 | 6576 | return ret; |
1da177e4 LT |
6577 | } |
6578 | ||
6579 | /** | |
6580 | * sys_sched_yield - yield the current processor to other threads. | |
6581 | * | |
dd41f596 IM |
6582 | * This function yields the current CPU to other tasks. If there are no |
6583 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6584 | */ |
5add95d4 | 6585 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6586 | { |
70b97a7f | 6587 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6588 | |
2d72376b | 6589 | schedstat_inc(rq, yld_count); |
4530d7ab | 6590 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6591 | |
6592 | /* | |
6593 | * Since we are going to call schedule() anyway, there's | |
6594 | * no need to preempt or enable interrupts: | |
6595 | */ | |
6596 | __release(rq->lock); | |
8a25d5de | 6597 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6598 | _raw_spin_unlock(&rq->lock); |
6599 | preempt_enable_no_resched(); | |
6600 | ||
6601 | schedule(); | |
6602 | ||
6603 | return 0; | |
6604 | } | |
6605 | ||
d86ee480 PZ |
6606 | static inline int should_resched(void) |
6607 | { | |
6608 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
6609 | } | |
6610 | ||
e7b38404 | 6611 | static void __cond_resched(void) |
1da177e4 | 6612 | { |
e7aaaa69 FW |
6613 | add_preempt_count(PREEMPT_ACTIVE); |
6614 | schedule(); | |
6615 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
6616 | } |
6617 | ||
02b67cc3 | 6618 | int __sched _cond_resched(void) |
1da177e4 | 6619 | { |
d86ee480 | 6620 | if (should_resched()) { |
1da177e4 LT |
6621 | __cond_resched(); |
6622 | return 1; | |
6623 | } | |
6624 | return 0; | |
6625 | } | |
02b67cc3 | 6626 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6627 | |
6628 | /* | |
613afbf8 | 6629 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6630 | * call schedule, and on return reacquire the lock. |
6631 | * | |
41a2d6cf | 6632 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6633 | * operations here to prevent schedule() from being called twice (once via |
6634 | * spin_unlock(), once by hand). | |
6635 | */ | |
613afbf8 | 6636 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6637 | { |
d86ee480 | 6638 | int resched = should_resched(); |
6df3cecb JK |
6639 | int ret = 0; |
6640 | ||
95c354fe | 6641 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6642 | spin_unlock(lock); |
d86ee480 | 6643 | if (resched) |
95c354fe NP |
6644 | __cond_resched(); |
6645 | else | |
6646 | cpu_relax(); | |
6df3cecb | 6647 | ret = 1; |
1da177e4 | 6648 | spin_lock(lock); |
1da177e4 | 6649 | } |
6df3cecb | 6650 | return ret; |
1da177e4 | 6651 | } |
613afbf8 | 6652 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6653 | |
613afbf8 | 6654 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
6655 | { |
6656 | BUG_ON(!in_softirq()); | |
6657 | ||
d86ee480 | 6658 | if (should_resched()) { |
98d82567 | 6659 | local_bh_enable(); |
1da177e4 LT |
6660 | __cond_resched(); |
6661 | local_bh_disable(); | |
6662 | return 1; | |
6663 | } | |
6664 | return 0; | |
6665 | } | |
613afbf8 | 6666 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 6667 | |
1da177e4 LT |
6668 | /** |
6669 | * yield - yield the current processor to other threads. | |
6670 | * | |
72fd4a35 | 6671 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6672 | * thread runnable and calls sys_sched_yield(). |
6673 | */ | |
6674 | void __sched yield(void) | |
6675 | { | |
6676 | set_current_state(TASK_RUNNING); | |
6677 | sys_sched_yield(); | |
6678 | } | |
1da177e4 LT |
6679 | EXPORT_SYMBOL(yield); |
6680 | ||
6681 | /* | |
41a2d6cf | 6682 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6683 | * that process accounting knows that this is a task in IO wait state. |
6684 | * | |
6685 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6686 | * has set its backing_dev_info: the queue against which it should throttle) | |
6687 | */ | |
6688 | void __sched io_schedule(void) | |
6689 | { | |
54d35f29 | 6690 | struct rq *rq = raw_rq(); |
1da177e4 | 6691 | |
0ff92245 | 6692 | delayacct_blkio_start(); |
1da177e4 LT |
6693 | atomic_inc(&rq->nr_iowait); |
6694 | schedule(); | |
6695 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6696 | delayacct_blkio_end(); |
1da177e4 | 6697 | } |
1da177e4 LT |
6698 | EXPORT_SYMBOL(io_schedule); |
6699 | ||
6700 | long __sched io_schedule_timeout(long timeout) | |
6701 | { | |
54d35f29 | 6702 | struct rq *rq = raw_rq(); |
1da177e4 LT |
6703 | long ret; |
6704 | ||
0ff92245 | 6705 | delayacct_blkio_start(); |
1da177e4 LT |
6706 | atomic_inc(&rq->nr_iowait); |
6707 | ret = schedule_timeout(timeout); | |
6708 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6709 | delayacct_blkio_end(); |
1da177e4 LT |
6710 | return ret; |
6711 | } | |
6712 | ||
6713 | /** | |
6714 | * sys_sched_get_priority_max - return maximum RT priority. | |
6715 | * @policy: scheduling class. | |
6716 | * | |
6717 | * this syscall returns the maximum rt_priority that can be used | |
6718 | * by a given scheduling class. | |
6719 | */ | |
5add95d4 | 6720 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6721 | { |
6722 | int ret = -EINVAL; | |
6723 | ||
6724 | switch (policy) { | |
6725 | case SCHED_FIFO: | |
6726 | case SCHED_RR: | |
6727 | ret = MAX_USER_RT_PRIO-1; | |
6728 | break; | |
6729 | case SCHED_NORMAL: | |
b0a9499c | 6730 | case SCHED_BATCH: |
dd41f596 | 6731 | case SCHED_IDLE: |
1da177e4 LT |
6732 | ret = 0; |
6733 | break; | |
6734 | } | |
6735 | return ret; | |
6736 | } | |
6737 | ||
6738 | /** | |
6739 | * sys_sched_get_priority_min - return minimum RT priority. | |
6740 | * @policy: scheduling class. | |
6741 | * | |
6742 | * this syscall returns the minimum rt_priority that can be used | |
6743 | * by a given scheduling class. | |
6744 | */ | |
5add95d4 | 6745 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6746 | { |
6747 | int ret = -EINVAL; | |
6748 | ||
6749 | switch (policy) { | |
6750 | case SCHED_FIFO: | |
6751 | case SCHED_RR: | |
6752 | ret = 1; | |
6753 | break; | |
6754 | case SCHED_NORMAL: | |
b0a9499c | 6755 | case SCHED_BATCH: |
dd41f596 | 6756 | case SCHED_IDLE: |
1da177e4 LT |
6757 | ret = 0; |
6758 | } | |
6759 | return ret; | |
6760 | } | |
6761 | ||
6762 | /** | |
6763 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6764 | * @pid: pid of the process. | |
6765 | * @interval: userspace pointer to the timeslice value. | |
6766 | * | |
6767 | * this syscall writes the default timeslice value of a given process | |
6768 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6769 | */ | |
17da2bd9 | 6770 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6771 | struct timespec __user *, interval) |
1da177e4 | 6772 | { |
36c8b586 | 6773 | struct task_struct *p; |
a4ec24b4 | 6774 | unsigned int time_slice; |
3a5c359a | 6775 | int retval; |
1da177e4 | 6776 | struct timespec t; |
1da177e4 LT |
6777 | |
6778 | if (pid < 0) | |
3a5c359a | 6779 | return -EINVAL; |
1da177e4 LT |
6780 | |
6781 | retval = -ESRCH; | |
6782 | read_lock(&tasklist_lock); | |
6783 | p = find_process_by_pid(pid); | |
6784 | if (!p) | |
6785 | goto out_unlock; | |
6786 | ||
6787 | retval = security_task_getscheduler(p); | |
6788 | if (retval) | |
6789 | goto out_unlock; | |
6790 | ||
77034937 IM |
6791 | /* |
6792 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6793 | * tasks that are on an otherwise idle runqueue: | |
6794 | */ | |
6795 | time_slice = 0; | |
6796 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6797 | time_slice = DEF_TIMESLICE; |
1868f958 | 6798 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6799 | struct sched_entity *se = &p->se; |
6800 | unsigned long flags; | |
6801 | struct rq *rq; | |
6802 | ||
6803 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6804 | if (rq->cfs.load.weight) |
6805 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6806 | task_rq_unlock(rq, &flags); |
6807 | } | |
1da177e4 | 6808 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6809 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6810 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6811 | return retval; |
3a5c359a | 6812 | |
1da177e4 LT |
6813 | out_unlock: |
6814 | read_unlock(&tasklist_lock); | |
6815 | return retval; | |
6816 | } | |
6817 | ||
7c731e0a | 6818 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6819 | |
82a1fcb9 | 6820 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6821 | { |
1da177e4 | 6822 | unsigned long free = 0; |
36c8b586 | 6823 | unsigned state; |
1da177e4 | 6824 | |
1da177e4 | 6825 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6826 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6827 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6828 | #if BITS_PER_LONG == 32 |
1da177e4 | 6829 | if (state == TASK_RUNNING) |
cc4ea795 | 6830 | printk(KERN_CONT " running "); |
1da177e4 | 6831 | else |
cc4ea795 | 6832 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6833 | #else |
6834 | if (state == TASK_RUNNING) | |
cc4ea795 | 6835 | printk(KERN_CONT " running task "); |
1da177e4 | 6836 | else |
cc4ea795 | 6837 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6838 | #endif |
6839 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6840 | free = stack_not_used(p); |
1da177e4 | 6841 | #endif |
aa47b7e0 DR |
6842 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
6843 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
6844 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 6845 | |
5fb5e6de | 6846 | show_stack(p, NULL); |
1da177e4 LT |
6847 | } |
6848 | ||
e59e2ae2 | 6849 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6850 | { |
36c8b586 | 6851 | struct task_struct *g, *p; |
1da177e4 | 6852 | |
4bd77321 IM |
6853 | #if BITS_PER_LONG == 32 |
6854 | printk(KERN_INFO | |
6855 | " task PC stack pid father\n"); | |
1da177e4 | 6856 | #else |
4bd77321 IM |
6857 | printk(KERN_INFO |
6858 | " task PC stack pid father\n"); | |
1da177e4 LT |
6859 | #endif |
6860 | read_lock(&tasklist_lock); | |
6861 | do_each_thread(g, p) { | |
6862 | /* | |
6863 | * reset the NMI-timeout, listing all files on a slow | |
6864 | * console might take alot of time: | |
6865 | */ | |
6866 | touch_nmi_watchdog(); | |
39bc89fd | 6867 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6868 | sched_show_task(p); |
1da177e4 LT |
6869 | } while_each_thread(g, p); |
6870 | ||
04c9167f JF |
6871 | touch_all_softlockup_watchdogs(); |
6872 | ||
dd41f596 IM |
6873 | #ifdef CONFIG_SCHED_DEBUG |
6874 | sysrq_sched_debug_show(); | |
6875 | #endif | |
1da177e4 | 6876 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6877 | /* |
6878 | * Only show locks if all tasks are dumped: | |
6879 | */ | |
6880 | if (state_filter == -1) | |
6881 | debug_show_all_locks(); | |
1da177e4 LT |
6882 | } |
6883 | ||
1df21055 IM |
6884 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6885 | { | |
dd41f596 | 6886 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6887 | } |
6888 | ||
f340c0d1 IM |
6889 | /** |
6890 | * init_idle - set up an idle thread for a given CPU | |
6891 | * @idle: task in question | |
6892 | * @cpu: cpu the idle task belongs to | |
6893 | * | |
6894 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6895 | * flag, to make booting more robust. | |
6896 | */ | |
5c1e1767 | 6897 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6898 | { |
70b97a7f | 6899 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6900 | unsigned long flags; |
6901 | ||
5cbd54ef IM |
6902 | spin_lock_irqsave(&rq->lock, flags); |
6903 | ||
dd41f596 IM |
6904 | __sched_fork(idle); |
6905 | idle->se.exec_start = sched_clock(); | |
6906 | ||
b29739f9 | 6907 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6908 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6909 | __set_task_cpu(idle, cpu); |
1da177e4 | 6910 | |
1da177e4 | 6911 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6912 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6913 | idle->oncpu = 1; | |
6914 | #endif | |
1da177e4 LT |
6915 | spin_unlock_irqrestore(&rq->lock, flags); |
6916 | ||
6917 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6918 | #if defined(CONFIG_PREEMPT) |
6919 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6920 | #else | |
a1261f54 | 6921 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6922 | #endif |
dd41f596 IM |
6923 | /* |
6924 | * The idle tasks have their own, simple scheduling class: | |
6925 | */ | |
6926 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6927 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6928 | } |
6929 | ||
6930 | /* | |
6931 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6932 | * indicates which cpus entered this state. This is used | |
6933 | * in the rcu update to wait only for active cpus. For system | |
6934 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6935 | * always be CPU_BITS_NONE. |
1da177e4 | 6936 | */ |
6a7b3dc3 | 6937 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6938 | |
19978ca6 IM |
6939 | /* |
6940 | * Increase the granularity value when there are more CPUs, | |
6941 | * because with more CPUs the 'effective latency' as visible | |
6942 | * to users decreases. But the relationship is not linear, | |
6943 | * so pick a second-best guess by going with the log2 of the | |
6944 | * number of CPUs. | |
6945 | * | |
6946 | * This idea comes from the SD scheduler of Con Kolivas: | |
6947 | */ | |
6948 | static inline void sched_init_granularity(void) | |
6949 | { | |
6950 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6951 | const unsigned long limit = 200000000; | |
6952 | ||
6953 | sysctl_sched_min_granularity *= factor; | |
6954 | if (sysctl_sched_min_granularity > limit) | |
6955 | sysctl_sched_min_granularity = limit; | |
6956 | ||
6957 | sysctl_sched_latency *= factor; | |
6958 | if (sysctl_sched_latency > limit) | |
6959 | sysctl_sched_latency = limit; | |
6960 | ||
6961 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6962 | |
6963 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6964 | } |
6965 | ||
1da177e4 LT |
6966 | #ifdef CONFIG_SMP |
6967 | /* | |
6968 | * This is how migration works: | |
6969 | * | |
70b97a7f | 6970 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6971 | * runqueue and wake up that CPU's migration thread. |
6972 | * 2) we down() the locked semaphore => thread blocks. | |
6973 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6974 | * thread off the CPU) | |
6975 | * 4) it gets the migration request and checks whether the migrated | |
6976 | * task is still in the wrong runqueue. | |
6977 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6978 | * it and puts it into the right queue. | |
6979 | * 6) migration thread up()s the semaphore. | |
6980 | * 7) we wake up and the migration is done. | |
6981 | */ | |
6982 | ||
6983 | /* | |
6984 | * Change a given task's CPU affinity. Migrate the thread to a | |
6985 | * proper CPU and schedule it away if the CPU it's executing on | |
6986 | * is removed from the allowed bitmask. | |
6987 | * | |
6988 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6989 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6990 | * call is not atomic; no spinlocks may be held. |
6991 | */ | |
96f874e2 | 6992 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6993 | { |
70b97a7f | 6994 | struct migration_req req; |
1da177e4 | 6995 | unsigned long flags; |
70b97a7f | 6996 | struct rq *rq; |
48f24c4d | 6997 | int ret = 0; |
1da177e4 LT |
6998 | |
6999 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 7000 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
7001 | ret = -EINVAL; |
7002 | goto out; | |
7003 | } | |
7004 | ||
9985b0ba | 7005 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 7006 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
7007 | ret = -EINVAL; |
7008 | goto out; | |
7009 | } | |
7010 | ||
73fe6aae | 7011 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 7012 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 7013 | else { |
96f874e2 RR |
7014 | cpumask_copy(&p->cpus_allowed, new_mask); |
7015 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
7016 | } |
7017 | ||
1da177e4 | 7018 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 7019 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
7020 | goto out; |
7021 | ||
1e5ce4f4 | 7022 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
7023 | /* Need help from migration thread: drop lock and wait. */ |
7024 | task_rq_unlock(rq, &flags); | |
7025 | wake_up_process(rq->migration_thread); | |
7026 | wait_for_completion(&req.done); | |
7027 | tlb_migrate_finish(p->mm); | |
7028 | return 0; | |
7029 | } | |
7030 | out: | |
7031 | task_rq_unlock(rq, &flags); | |
48f24c4d | 7032 | |
1da177e4 LT |
7033 | return ret; |
7034 | } | |
cd8ba7cd | 7035 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
7036 | |
7037 | /* | |
41a2d6cf | 7038 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
7039 | * this because either it can't run here any more (set_cpus_allowed() |
7040 | * away from this CPU, or CPU going down), or because we're | |
7041 | * attempting to rebalance this task on exec (sched_exec). | |
7042 | * | |
7043 | * So we race with normal scheduler movements, but that's OK, as long | |
7044 | * as the task is no longer on this CPU. | |
efc30814 KK |
7045 | * |
7046 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 7047 | */ |
efc30814 | 7048 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 7049 | { |
70b97a7f | 7050 | struct rq *rq_dest, *rq_src; |
dd41f596 | 7051 | int ret = 0, on_rq; |
1da177e4 | 7052 | |
e761b772 | 7053 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 7054 | return ret; |
1da177e4 LT |
7055 | |
7056 | rq_src = cpu_rq(src_cpu); | |
7057 | rq_dest = cpu_rq(dest_cpu); | |
7058 | ||
7059 | double_rq_lock(rq_src, rq_dest); | |
7060 | /* Already moved. */ | |
7061 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7062 | goto done; |
1da177e4 | 7063 | /* Affinity changed (again). */ |
96f874e2 | 7064 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7065 | goto fail; |
1da177e4 | 7066 | |
dd41f596 | 7067 | on_rq = p->se.on_rq; |
6e82a3be | 7068 | if (on_rq) |
2e1cb74a | 7069 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7070 | |
1da177e4 | 7071 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7072 | if (on_rq) { |
7073 | activate_task(rq_dest, p, 0); | |
15afe09b | 7074 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7075 | } |
b1e38734 | 7076 | done: |
efc30814 | 7077 | ret = 1; |
b1e38734 | 7078 | fail: |
1da177e4 | 7079 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7080 | return ret; |
1da177e4 LT |
7081 | } |
7082 | ||
7083 | /* | |
7084 | * migration_thread - this is a highprio system thread that performs | |
7085 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7086 | * another runqueue. | |
7087 | */ | |
95cdf3b7 | 7088 | static int migration_thread(void *data) |
1da177e4 | 7089 | { |
1da177e4 | 7090 | int cpu = (long)data; |
70b97a7f | 7091 | struct rq *rq; |
1da177e4 LT |
7092 | |
7093 | rq = cpu_rq(cpu); | |
7094 | BUG_ON(rq->migration_thread != current); | |
7095 | ||
7096 | set_current_state(TASK_INTERRUPTIBLE); | |
7097 | while (!kthread_should_stop()) { | |
70b97a7f | 7098 | struct migration_req *req; |
1da177e4 | 7099 | struct list_head *head; |
1da177e4 | 7100 | |
1da177e4 LT |
7101 | spin_lock_irq(&rq->lock); |
7102 | ||
7103 | if (cpu_is_offline(cpu)) { | |
7104 | spin_unlock_irq(&rq->lock); | |
371cbb38 | 7105 | break; |
1da177e4 LT |
7106 | } |
7107 | ||
7108 | if (rq->active_balance) { | |
7109 | active_load_balance(rq, cpu); | |
7110 | rq->active_balance = 0; | |
7111 | } | |
7112 | ||
7113 | head = &rq->migration_queue; | |
7114 | ||
7115 | if (list_empty(head)) { | |
7116 | spin_unlock_irq(&rq->lock); | |
7117 | schedule(); | |
7118 | set_current_state(TASK_INTERRUPTIBLE); | |
7119 | continue; | |
7120 | } | |
70b97a7f | 7121 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7122 | list_del_init(head->next); |
7123 | ||
674311d5 NP |
7124 | spin_unlock(&rq->lock); |
7125 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7126 | local_irq_enable(); | |
1da177e4 LT |
7127 | |
7128 | complete(&req->done); | |
7129 | } | |
7130 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 7131 | |
1da177e4 LT |
7132 | return 0; |
7133 | } | |
7134 | ||
7135 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7136 | |
7137 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7138 | { | |
7139 | int ret; | |
7140 | ||
7141 | local_irq_disable(); | |
7142 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7143 | local_irq_enable(); | |
7144 | return ret; | |
7145 | } | |
7146 | ||
054b9108 | 7147 | /* |
3a4fa0a2 | 7148 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7149 | */ |
48f24c4d | 7150 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7151 | { |
70b97a7f | 7152 | int dest_cpu; |
6ca09dfc | 7153 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7154 | |
7155 | again: | |
7156 | /* Look for allowed, online CPU in same node. */ | |
7157 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7158 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7159 | goto move; | |
7160 | ||
7161 | /* Any allowed, online CPU? */ | |
7162 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7163 | if (dest_cpu < nr_cpu_ids) | |
7164 | goto move; | |
7165 | ||
7166 | /* No more Mr. Nice Guy. */ | |
7167 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7168 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7169 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7170 | |
e76bd8d9 RR |
7171 | /* |
7172 | * Don't tell them about moving exiting tasks or | |
7173 | * kernel threads (both mm NULL), since they never | |
7174 | * leave kernel. | |
7175 | */ | |
7176 | if (p->mm && printk_ratelimit()) { | |
7177 | printk(KERN_INFO "process %d (%s) no " | |
7178 | "longer affine to cpu%d\n", | |
7179 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7180 | } |
e76bd8d9 RR |
7181 | } |
7182 | ||
7183 | move: | |
7184 | /* It can have affinity changed while we were choosing. */ | |
7185 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7186 | goto again; | |
1da177e4 LT |
7187 | } |
7188 | ||
7189 | /* | |
7190 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7191 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7192 | * for performance reasons the counter is not stricly tracking tasks to | |
7193 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7194 | * to keep the global sum constant after CPU-down: | |
7195 | */ | |
70b97a7f | 7196 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7197 | { |
1e5ce4f4 | 7198 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7199 | unsigned long flags; |
7200 | ||
7201 | local_irq_save(flags); | |
7202 | double_rq_lock(rq_src, rq_dest); | |
7203 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7204 | rq_src->nr_uninterruptible = 0; | |
7205 | double_rq_unlock(rq_src, rq_dest); | |
7206 | local_irq_restore(flags); | |
7207 | } | |
7208 | ||
7209 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7210 | static void migrate_live_tasks(int src_cpu) | |
7211 | { | |
48f24c4d | 7212 | struct task_struct *p, *t; |
1da177e4 | 7213 | |
f7b4cddc | 7214 | read_lock(&tasklist_lock); |
1da177e4 | 7215 | |
48f24c4d IM |
7216 | do_each_thread(t, p) { |
7217 | if (p == current) | |
1da177e4 LT |
7218 | continue; |
7219 | ||
48f24c4d IM |
7220 | if (task_cpu(p) == src_cpu) |
7221 | move_task_off_dead_cpu(src_cpu, p); | |
7222 | } while_each_thread(t, p); | |
1da177e4 | 7223 | |
f7b4cddc | 7224 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7225 | } |
7226 | ||
dd41f596 IM |
7227 | /* |
7228 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7229 | * It does so by boosting its priority to highest possible. |
7230 | * Used by CPU offline code. | |
1da177e4 LT |
7231 | */ |
7232 | void sched_idle_next(void) | |
7233 | { | |
48f24c4d | 7234 | int this_cpu = smp_processor_id(); |
70b97a7f | 7235 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7236 | struct task_struct *p = rq->idle; |
7237 | unsigned long flags; | |
7238 | ||
7239 | /* cpu has to be offline */ | |
48f24c4d | 7240 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7241 | |
48f24c4d IM |
7242 | /* |
7243 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7244 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7245 | */ |
7246 | spin_lock_irqsave(&rq->lock, flags); | |
7247 | ||
dd41f596 | 7248 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7249 | |
94bc9a7b DA |
7250 | update_rq_clock(rq); |
7251 | activate_task(rq, p, 0); | |
1da177e4 LT |
7252 | |
7253 | spin_unlock_irqrestore(&rq->lock, flags); | |
7254 | } | |
7255 | ||
48f24c4d IM |
7256 | /* |
7257 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7258 | * offline. |
7259 | */ | |
7260 | void idle_task_exit(void) | |
7261 | { | |
7262 | struct mm_struct *mm = current->active_mm; | |
7263 | ||
7264 | BUG_ON(cpu_online(smp_processor_id())); | |
7265 | ||
7266 | if (mm != &init_mm) | |
7267 | switch_mm(mm, &init_mm, current); | |
7268 | mmdrop(mm); | |
7269 | } | |
7270 | ||
054b9108 | 7271 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7272 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7273 | { |
70b97a7f | 7274 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7275 | |
7276 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7277 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7278 | |
7279 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7280 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7281 | |
48f24c4d | 7282 | get_task_struct(p); |
1da177e4 LT |
7283 | |
7284 | /* | |
7285 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7286 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7287 | * fine. |
7288 | */ | |
f7b4cddc | 7289 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7290 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7291 | spin_lock_irq(&rq->lock); |
1da177e4 | 7292 | |
48f24c4d | 7293 | put_task_struct(p); |
1da177e4 LT |
7294 | } |
7295 | ||
7296 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7297 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7298 | { | |
70b97a7f | 7299 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7300 | struct task_struct *next; |
48f24c4d | 7301 | |
dd41f596 IM |
7302 | for ( ; ; ) { |
7303 | if (!rq->nr_running) | |
7304 | break; | |
a8e504d2 | 7305 | update_rq_clock(rq); |
b67802ea | 7306 | next = pick_next_task(rq); |
dd41f596 IM |
7307 | if (!next) |
7308 | break; | |
79c53799 | 7309 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7310 | migrate_dead(dead_cpu, next); |
e692ab53 | 7311 | |
1da177e4 LT |
7312 | } |
7313 | } | |
dce48a84 TG |
7314 | |
7315 | /* | |
7316 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7317 | */ | |
7318 | static void calc_global_load_remove(struct rq *rq) | |
7319 | { | |
7320 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
7321 | } | |
1da177e4 LT |
7322 | #endif /* CONFIG_HOTPLUG_CPU */ |
7323 | ||
e692ab53 NP |
7324 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7325 | ||
7326 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7327 | { |
7328 | .procname = "sched_domain", | |
c57baf1e | 7329 | .mode = 0555, |
e0361851 | 7330 | }, |
38605cae | 7331 | {0, }, |
e692ab53 NP |
7332 | }; |
7333 | ||
7334 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7335 | { |
c57baf1e | 7336 | .ctl_name = CTL_KERN, |
e0361851 | 7337 | .procname = "kernel", |
c57baf1e | 7338 | .mode = 0555, |
e0361851 AD |
7339 | .child = sd_ctl_dir, |
7340 | }, | |
38605cae | 7341 | {0, }, |
e692ab53 NP |
7342 | }; |
7343 | ||
7344 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7345 | { | |
7346 | struct ctl_table *entry = | |
5cf9f062 | 7347 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7348 | |
e692ab53 NP |
7349 | return entry; |
7350 | } | |
7351 | ||
6382bc90 MM |
7352 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7353 | { | |
cd790076 | 7354 | struct ctl_table *entry; |
6382bc90 | 7355 | |
cd790076 MM |
7356 | /* |
7357 | * In the intermediate directories, both the child directory and | |
7358 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7359 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7360 | * static strings and all have proc handlers. |
7361 | */ | |
7362 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7363 | if (entry->child) |
7364 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7365 | if (entry->proc_handler == NULL) |
7366 | kfree(entry->procname); | |
7367 | } | |
6382bc90 MM |
7368 | |
7369 | kfree(*tablep); | |
7370 | *tablep = NULL; | |
7371 | } | |
7372 | ||
e692ab53 | 7373 | static void |
e0361851 | 7374 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7375 | const char *procname, void *data, int maxlen, |
7376 | mode_t mode, proc_handler *proc_handler) | |
7377 | { | |
e692ab53 NP |
7378 | entry->procname = procname; |
7379 | entry->data = data; | |
7380 | entry->maxlen = maxlen; | |
7381 | entry->mode = mode; | |
7382 | entry->proc_handler = proc_handler; | |
7383 | } | |
7384 | ||
7385 | static struct ctl_table * | |
7386 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7387 | { | |
a5d8c348 | 7388 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7389 | |
ad1cdc1d MM |
7390 | if (table == NULL) |
7391 | return NULL; | |
7392 | ||
e0361851 | 7393 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7394 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7395 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7396 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7397 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7398 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7399 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7400 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7401 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7402 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7403 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7404 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7405 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7406 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7407 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7408 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7409 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7410 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7411 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7412 | &sd->cache_nice_tries, |
7413 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7414 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7415 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7416 | set_table_entry(&table[11], "name", sd->name, |
7417 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7418 | /* &table[12] is terminator */ | |
e692ab53 NP |
7419 | |
7420 | return table; | |
7421 | } | |
7422 | ||
9a4e7159 | 7423 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7424 | { |
7425 | struct ctl_table *entry, *table; | |
7426 | struct sched_domain *sd; | |
7427 | int domain_num = 0, i; | |
7428 | char buf[32]; | |
7429 | ||
7430 | for_each_domain(cpu, sd) | |
7431 | domain_num++; | |
7432 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7433 | if (table == NULL) |
7434 | return NULL; | |
e692ab53 NP |
7435 | |
7436 | i = 0; | |
7437 | for_each_domain(cpu, sd) { | |
7438 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7439 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7440 | entry->mode = 0555; |
e692ab53 NP |
7441 | entry->child = sd_alloc_ctl_domain_table(sd); |
7442 | entry++; | |
7443 | i++; | |
7444 | } | |
7445 | return table; | |
7446 | } | |
7447 | ||
7448 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7449 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7450 | { |
7451 | int i, cpu_num = num_online_cpus(); | |
7452 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7453 | char buf[32]; | |
7454 | ||
7378547f MM |
7455 | WARN_ON(sd_ctl_dir[0].child); |
7456 | sd_ctl_dir[0].child = entry; | |
7457 | ||
ad1cdc1d MM |
7458 | if (entry == NULL) |
7459 | return; | |
7460 | ||
97b6ea7b | 7461 | for_each_online_cpu(i) { |
e692ab53 | 7462 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7463 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7464 | entry->mode = 0555; |
e692ab53 | 7465 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7466 | entry++; |
e692ab53 | 7467 | } |
7378547f MM |
7468 | |
7469 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7470 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7471 | } | |
6382bc90 | 7472 | |
7378547f | 7473 | /* may be called multiple times per register */ |
6382bc90 MM |
7474 | static void unregister_sched_domain_sysctl(void) |
7475 | { | |
7378547f MM |
7476 | if (sd_sysctl_header) |
7477 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7478 | sd_sysctl_header = NULL; |
7378547f MM |
7479 | if (sd_ctl_dir[0].child) |
7480 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7481 | } |
e692ab53 | 7482 | #else |
6382bc90 MM |
7483 | static void register_sched_domain_sysctl(void) |
7484 | { | |
7485 | } | |
7486 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7487 | { |
7488 | } | |
7489 | #endif | |
7490 | ||
1f11eb6a GH |
7491 | static void set_rq_online(struct rq *rq) |
7492 | { | |
7493 | if (!rq->online) { | |
7494 | const struct sched_class *class; | |
7495 | ||
c6c4927b | 7496 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7497 | rq->online = 1; |
7498 | ||
7499 | for_each_class(class) { | |
7500 | if (class->rq_online) | |
7501 | class->rq_online(rq); | |
7502 | } | |
7503 | } | |
7504 | } | |
7505 | ||
7506 | static void set_rq_offline(struct rq *rq) | |
7507 | { | |
7508 | if (rq->online) { | |
7509 | const struct sched_class *class; | |
7510 | ||
7511 | for_each_class(class) { | |
7512 | if (class->rq_offline) | |
7513 | class->rq_offline(rq); | |
7514 | } | |
7515 | ||
c6c4927b | 7516 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7517 | rq->online = 0; |
7518 | } | |
7519 | } | |
7520 | ||
1da177e4 LT |
7521 | /* |
7522 | * migration_call - callback that gets triggered when a CPU is added. | |
7523 | * Here we can start up the necessary migration thread for the new CPU. | |
7524 | */ | |
48f24c4d IM |
7525 | static int __cpuinit |
7526 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7527 | { |
1da177e4 | 7528 | struct task_struct *p; |
48f24c4d | 7529 | int cpu = (long)hcpu; |
1da177e4 | 7530 | unsigned long flags; |
70b97a7f | 7531 | struct rq *rq; |
1da177e4 LT |
7532 | |
7533 | switch (action) { | |
5be9361c | 7534 | |
1da177e4 | 7535 | case CPU_UP_PREPARE: |
8bb78442 | 7536 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7537 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7538 | if (IS_ERR(p)) |
7539 | return NOTIFY_BAD; | |
1da177e4 LT |
7540 | kthread_bind(p, cpu); |
7541 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7542 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7543 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 7544 | task_rq_unlock(rq, &flags); |
371cbb38 | 7545 | get_task_struct(p); |
1da177e4 LT |
7546 | cpu_rq(cpu)->migration_thread = p; |
7547 | break; | |
48f24c4d | 7548 | |
1da177e4 | 7549 | case CPU_ONLINE: |
8bb78442 | 7550 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7551 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7552 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7553 | |
7554 | /* Update our root-domain */ | |
7555 | rq = cpu_rq(cpu); | |
7556 | spin_lock_irqsave(&rq->lock, flags); | |
dce48a84 TG |
7557 | rq->calc_load_update = calc_load_update; |
7558 | rq->calc_load_active = 0; | |
1f94ef59 | 7559 | if (rq->rd) { |
c6c4927b | 7560 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7561 | |
7562 | set_rq_online(rq); | |
1f94ef59 GH |
7563 | } |
7564 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7565 | break; |
48f24c4d | 7566 | |
1da177e4 LT |
7567 | #ifdef CONFIG_HOTPLUG_CPU |
7568 | case CPU_UP_CANCELED: | |
8bb78442 | 7569 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7570 | if (!cpu_rq(cpu)->migration_thread) |
7571 | break; | |
41a2d6cf | 7572 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7573 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7574 | cpumask_any(cpu_online_mask)); |
1da177e4 | 7575 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 7576 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
7577 | cpu_rq(cpu)->migration_thread = NULL; |
7578 | break; | |
48f24c4d | 7579 | |
1da177e4 | 7580 | case CPU_DEAD: |
8bb78442 | 7581 | case CPU_DEAD_FROZEN: |
470fd646 | 7582 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7583 | migrate_live_tasks(cpu); |
7584 | rq = cpu_rq(cpu); | |
7585 | kthread_stop(rq->migration_thread); | |
371cbb38 | 7586 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
7587 | rq->migration_thread = NULL; |
7588 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7589 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7590 | update_rq_clock(rq); |
2e1cb74a | 7591 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7592 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7593 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7594 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7595 | migrate_dead_tasks(cpu); |
d2da272a | 7596 | spin_unlock_irq(&rq->lock); |
470fd646 | 7597 | cpuset_unlock(); |
1da177e4 LT |
7598 | migrate_nr_uninterruptible(rq); |
7599 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7600 | calc_global_load_remove(rq); |
41a2d6cf IM |
7601 | /* |
7602 | * No need to migrate the tasks: it was best-effort if | |
7603 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7604 | * the requestors. | |
7605 | */ | |
1da177e4 LT |
7606 | spin_lock_irq(&rq->lock); |
7607 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7608 | struct migration_req *req; |
7609 | ||
1da177e4 | 7610 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7611 | struct migration_req, list); |
1da177e4 | 7612 | list_del_init(&req->list); |
9a2bd244 | 7613 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7614 | complete(&req->done); |
9a2bd244 | 7615 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7616 | } |
7617 | spin_unlock_irq(&rq->lock); | |
7618 | break; | |
57d885fe | 7619 | |
08f503b0 GH |
7620 | case CPU_DYING: |
7621 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7622 | /* Update our root-domain */ |
7623 | rq = cpu_rq(cpu); | |
7624 | spin_lock_irqsave(&rq->lock, flags); | |
7625 | if (rq->rd) { | |
c6c4927b | 7626 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7627 | set_rq_offline(rq); |
57d885fe GH |
7628 | } |
7629 | spin_unlock_irqrestore(&rq->lock, flags); | |
7630 | break; | |
1da177e4 LT |
7631 | #endif |
7632 | } | |
7633 | return NOTIFY_OK; | |
7634 | } | |
7635 | ||
f38b0820 PM |
7636 | /* |
7637 | * Register at high priority so that task migration (migrate_all_tasks) | |
7638 | * happens before everything else. This has to be lower priority than | |
7639 | * the notifier in the perf_counter subsystem, though. | |
1da177e4 | 7640 | */ |
26c2143b | 7641 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7642 | .notifier_call = migration_call, |
7643 | .priority = 10 | |
7644 | }; | |
7645 | ||
7babe8db | 7646 | static int __init migration_init(void) |
1da177e4 LT |
7647 | { |
7648 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7649 | int err; |
48f24c4d IM |
7650 | |
7651 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7652 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7653 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7654 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7655 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7656 | |
7657 | return err; | |
1da177e4 | 7658 | } |
7babe8db | 7659 | early_initcall(migration_init); |
1da177e4 LT |
7660 | #endif |
7661 | ||
7662 | #ifdef CONFIG_SMP | |
476f3534 | 7663 | |
3e9830dc | 7664 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7665 | |
7c16ec58 | 7666 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7667 | struct cpumask *groupmask) |
1da177e4 | 7668 | { |
4dcf6aff | 7669 | struct sched_group *group = sd->groups; |
434d53b0 | 7670 | char str[256]; |
1da177e4 | 7671 | |
968ea6d8 | 7672 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7673 | cpumask_clear(groupmask); |
4dcf6aff IM |
7674 | |
7675 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7676 | ||
7677 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7678 | printk("does not load-balance\n"); | |
7679 | if (sd->parent) | |
7680 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7681 | " has parent"); | |
7682 | return -1; | |
41c7ce9a NP |
7683 | } |
7684 | ||
eefd796a | 7685 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7686 | |
758b2cdc | 7687 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7688 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7689 | "CPU%d\n", cpu); | |
7690 | } | |
758b2cdc | 7691 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7692 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7693 | " CPU%d\n", cpu); | |
7694 | } | |
1da177e4 | 7695 | |
4dcf6aff | 7696 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7697 | do { |
4dcf6aff IM |
7698 | if (!group) { |
7699 | printk("\n"); | |
7700 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7701 | break; |
7702 | } | |
7703 | ||
4dcf6aff IM |
7704 | if (!group->__cpu_power) { |
7705 | printk(KERN_CONT "\n"); | |
7706 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7707 | "set\n"); | |
7708 | break; | |
7709 | } | |
1da177e4 | 7710 | |
758b2cdc | 7711 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7712 | printk(KERN_CONT "\n"); |
7713 | printk(KERN_ERR "ERROR: empty group\n"); | |
7714 | break; | |
7715 | } | |
1da177e4 | 7716 | |
758b2cdc | 7717 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7718 | printk(KERN_CONT "\n"); |
7719 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7720 | break; | |
7721 | } | |
1da177e4 | 7722 | |
758b2cdc | 7723 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7724 | |
968ea6d8 | 7725 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7726 | |
7727 | printk(KERN_CONT " %s", str); | |
7728 | if (group->__cpu_power != SCHED_LOAD_SCALE) { | |
7729 | printk(KERN_CONT " (__cpu_power = %d)", | |
7730 | group->__cpu_power); | |
7731 | } | |
1da177e4 | 7732 | |
4dcf6aff IM |
7733 | group = group->next; |
7734 | } while (group != sd->groups); | |
7735 | printk(KERN_CONT "\n"); | |
1da177e4 | 7736 | |
758b2cdc | 7737 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7738 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7739 | |
758b2cdc RR |
7740 | if (sd->parent && |
7741 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7742 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7743 | "of domain->span\n"); | |
7744 | return 0; | |
7745 | } | |
1da177e4 | 7746 | |
4dcf6aff IM |
7747 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7748 | { | |
d5dd3db1 | 7749 | cpumask_var_t groupmask; |
4dcf6aff | 7750 | int level = 0; |
1da177e4 | 7751 | |
4dcf6aff IM |
7752 | if (!sd) { |
7753 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7754 | return; | |
7755 | } | |
1da177e4 | 7756 | |
4dcf6aff IM |
7757 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7758 | ||
d5dd3db1 | 7759 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7760 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7761 | return; | |
7762 | } | |
7763 | ||
4dcf6aff | 7764 | for (;;) { |
7c16ec58 | 7765 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7766 | break; |
1da177e4 LT |
7767 | level++; |
7768 | sd = sd->parent; | |
33859f7f | 7769 | if (!sd) |
4dcf6aff IM |
7770 | break; |
7771 | } | |
d5dd3db1 | 7772 | free_cpumask_var(groupmask); |
1da177e4 | 7773 | } |
6d6bc0ad | 7774 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7775 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7776 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7777 | |
1a20ff27 | 7778 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7779 | { |
758b2cdc | 7780 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7781 | return 1; |
7782 | ||
7783 | /* Following flags need at least 2 groups */ | |
7784 | if (sd->flags & (SD_LOAD_BALANCE | | |
7785 | SD_BALANCE_NEWIDLE | | |
7786 | SD_BALANCE_FORK | | |
89c4710e SS |
7787 | SD_BALANCE_EXEC | |
7788 | SD_SHARE_CPUPOWER | | |
7789 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7790 | if (sd->groups != sd->groups->next) |
7791 | return 0; | |
7792 | } | |
7793 | ||
7794 | /* Following flags don't use groups */ | |
7795 | if (sd->flags & (SD_WAKE_IDLE | | |
7796 | SD_WAKE_AFFINE | | |
7797 | SD_WAKE_BALANCE)) | |
7798 | return 0; | |
7799 | ||
7800 | return 1; | |
7801 | } | |
7802 | ||
48f24c4d IM |
7803 | static int |
7804 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7805 | { |
7806 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7807 | ||
7808 | if (sd_degenerate(parent)) | |
7809 | return 1; | |
7810 | ||
758b2cdc | 7811 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7812 | return 0; |
7813 | ||
7814 | /* Does parent contain flags not in child? */ | |
7815 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7816 | if (cflags & SD_WAKE_AFFINE) | |
7817 | pflags &= ~SD_WAKE_BALANCE; | |
7818 | /* Flags needing groups don't count if only 1 group in parent */ | |
7819 | if (parent->groups == parent->groups->next) { | |
7820 | pflags &= ~(SD_LOAD_BALANCE | | |
7821 | SD_BALANCE_NEWIDLE | | |
7822 | SD_BALANCE_FORK | | |
89c4710e SS |
7823 | SD_BALANCE_EXEC | |
7824 | SD_SHARE_CPUPOWER | | |
7825 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7826 | if (nr_node_ids == 1) |
7827 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7828 | } |
7829 | if (~cflags & pflags) | |
7830 | return 0; | |
7831 | ||
7832 | return 1; | |
7833 | } | |
7834 | ||
c6c4927b RR |
7835 | static void free_rootdomain(struct root_domain *rd) |
7836 | { | |
68e74568 RR |
7837 | cpupri_cleanup(&rd->cpupri); |
7838 | ||
c6c4927b RR |
7839 | free_cpumask_var(rd->rto_mask); |
7840 | free_cpumask_var(rd->online); | |
7841 | free_cpumask_var(rd->span); | |
7842 | kfree(rd); | |
7843 | } | |
7844 | ||
57d885fe GH |
7845 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7846 | { | |
a0490fa3 | 7847 | struct root_domain *old_rd = NULL; |
57d885fe | 7848 | unsigned long flags; |
57d885fe GH |
7849 | |
7850 | spin_lock_irqsave(&rq->lock, flags); | |
7851 | ||
7852 | if (rq->rd) { | |
a0490fa3 | 7853 | old_rd = rq->rd; |
57d885fe | 7854 | |
c6c4927b | 7855 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7856 | set_rq_offline(rq); |
57d885fe | 7857 | |
c6c4927b | 7858 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7859 | |
a0490fa3 IM |
7860 | /* |
7861 | * If we dont want to free the old_rt yet then | |
7862 | * set old_rd to NULL to skip the freeing later | |
7863 | * in this function: | |
7864 | */ | |
7865 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7866 | old_rd = NULL; | |
57d885fe GH |
7867 | } |
7868 | ||
7869 | atomic_inc(&rd->refcount); | |
7870 | rq->rd = rd; | |
7871 | ||
c6c4927b RR |
7872 | cpumask_set_cpu(rq->cpu, rd->span); |
7873 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7874 | set_rq_online(rq); |
57d885fe GH |
7875 | |
7876 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7877 | |
7878 | if (old_rd) | |
7879 | free_rootdomain(old_rd); | |
57d885fe GH |
7880 | } |
7881 | ||
fd5e1b5d | 7882 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 7883 | { |
36b7b6d4 PE |
7884 | gfp_t gfp = GFP_KERNEL; |
7885 | ||
57d885fe GH |
7886 | memset(rd, 0, sizeof(*rd)); |
7887 | ||
36b7b6d4 PE |
7888 | if (bootmem) |
7889 | gfp = GFP_NOWAIT; | |
c6c4927b | 7890 | |
36b7b6d4 | 7891 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 7892 | goto out; |
36b7b6d4 | 7893 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 7894 | goto free_span; |
36b7b6d4 | 7895 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 7896 | goto free_online; |
6e0534f2 | 7897 | |
0fb53029 | 7898 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 7899 | goto free_rto_mask; |
c6c4927b | 7900 | return 0; |
6e0534f2 | 7901 | |
68e74568 RR |
7902 | free_rto_mask: |
7903 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7904 | free_online: |
7905 | free_cpumask_var(rd->online); | |
7906 | free_span: | |
7907 | free_cpumask_var(rd->span); | |
0c910d28 | 7908 | out: |
c6c4927b | 7909 | return -ENOMEM; |
57d885fe GH |
7910 | } |
7911 | ||
7912 | static void init_defrootdomain(void) | |
7913 | { | |
c6c4927b RR |
7914 | init_rootdomain(&def_root_domain, true); |
7915 | ||
57d885fe GH |
7916 | atomic_set(&def_root_domain.refcount, 1); |
7917 | } | |
7918 | ||
dc938520 | 7919 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7920 | { |
7921 | struct root_domain *rd; | |
7922 | ||
7923 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7924 | if (!rd) | |
7925 | return NULL; | |
7926 | ||
c6c4927b RR |
7927 | if (init_rootdomain(rd, false) != 0) { |
7928 | kfree(rd); | |
7929 | return NULL; | |
7930 | } | |
57d885fe GH |
7931 | |
7932 | return rd; | |
7933 | } | |
7934 | ||
1da177e4 | 7935 | /* |
0eab9146 | 7936 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7937 | * hold the hotplug lock. |
7938 | */ | |
0eab9146 IM |
7939 | static void |
7940 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7941 | { |
70b97a7f | 7942 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7943 | struct sched_domain *tmp; |
7944 | ||
7945 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7946 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7947 | struct sched_domain *parent = tmp->parent; |
7948 | if (!parent) | |
7949 | break; | |
f29c9b1c | 7950 | |
1a848870 | 7951 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7952 | tmp->parent = parent->parent; |
1a848870 SS |
7953 | if (parent->parent) |
7954 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7955 | } else |
7956 | tmp = tmp->parent; | |
245af2c7 SS |
7957 | } |
7958 | ||
1a848870 | 7959 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7960 | sd = sd->parent; |
1a848870 SS |
7961 | if (sd) |
7962 | sd->child = NULL; | |
7963 | } | |
1da177e4 LT |
7964 | |
7965 | sched_domain_debug(sd, cpu); | |
7966 | ||
57d885fe | 7967 | rq_attach_root(rq, rd); |
674311d5 | 7968 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7969 | } |
7970 | ||
7971 | /* cpus with isolated domains */ | |
dcc30a35 | 7972 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7973 | |
7974 | /* Setup the mask of cpus configured for isolated domains */ | |
7975 | static int __init isolated_cpu_setup(char *str) | |
7976 | { | |
968ea6d8 | 7977 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7978 | return 1; |
7979 | } | |
7980 | ||
8927f494 | 7981 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7982 | |
7983 | /* | |
6711cab4 SS |
7984 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7985 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7986 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7987 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7988 | * |
7989 | * init_sched_build_groups will build a circular linked list of the groups | |
7990 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7991 | * and ->cpu_power to 0. | |
7992 | */ | |
a616058b | 7993 | static void |
96f874e2 RR |
7994 | init_sched_build_groups(const struct cpumask *span, |
7995 | const struct cpumask *cpu_map, | |
7996 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7997 | struct sched_group **sg, |
96f874e2 RR |
7998 | struct cpumask *tmpmask), |
7999 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
8000 | { |
8001 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
8002 | int i; |
8003 | ||
96f874e2 | 8004 | cpumask_clear(covered); |
7c16ec58 | 8005 | |
abcd083a | 8006 | for_each_cpu(i, span) { |
6711cab4 | 8007 | struct sched_group *sg; |
7c16ec58 | 8008 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
8009 | int j; |
8010 | ||
758b2cdc | 8011 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
8012 | continue; |
8013 | ||
758b2cdc | 8014 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 8015 | sg->__cpu_power = 0; |
1da177e4 | 8016 | |
abcd083a | 8017 | for_each_cpu(j, span) { |
7c16ec58 | 8018 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
8019 | continue; |
8020 | ||
96f874e2 | 8021 | cpumask_set_cpu(j, covered); |
758b2cdc | 8022 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
8023 | } |
8024 | if (!first) | |
8025 | first = sg; | |
8026 | if (last) | |
8027 | last->next = sg; | |
8028 | last = sg; | |
8029 | } | |
8030 | last->next = first; | |
8031 | } | |
8032 | ||
9c1cfda2 | 8033 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 8034 | |
9c1cfda2 | 8035 | #ifdef CONFIG_NUMA |
198e2f18 | 8036 | |
9c1cfda2 JH |
8037 | /** |
8038 | * find_next_best_node - find the next node to include in a sched_domain | |
8039 | * @node: node whose sched_domain we're building | |
8040 | * @used_nodes: nodes already in the sched_domain | |
8041 | * | |
41a2d6cf | 8042 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
8043 | * finds the closest node not already in the @used_nodes map. |
8044 | * | |
8045 | * Should use nodemask_t. | |
8046 | */ | |
c5f59f08 | 8047 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
8048 | { |
8049 | int i, n, val, min_val, best_node = 0; | |
8050 | ||
8051 | min_val = INT_MAX; | |
8052 | ||
076ac2af | 8053 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 8054 | /* Start at @node */ |
076ac2af | 8055 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8056 | |
8057 | if (!nr_cpus_node(n)) | |
8058 | continue; | |
8059 | ||
8060 | /* Skip already used nodes */ | |
c5f59f08 | 8061 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8062 | continue; |
8063 | ||
8064 | /* Simple min distance search */ | |
8065 | val = node_distance(node, n); | |
8066 | ||
8067 | if (val < min_val) { | |
8068 | min_val = val; | |
8069 | best_node = n; | |
8070 | } | |
8071 | } | |
8072 | ||
c5f59f08 | 8073 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8074 | return best_node; |
8075 | } | |
8076 | ||
8077 | /** | |
8078 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8079 | * @node: node whose cpumask we're constructing | |
73486722 | 8080 | * @span: resulting cpumask |
9c1cfda2 | 8081 | * |
41a2d6cf | 8082 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8083 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8084 | * out optimally. | |
8085 | */ | |
96f874e2 | 8086 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8087 | { |
c5f59f08 | 8088 | nodemask_t used_nodes; |
48f24c4d | 8089 | int i; |
9c1cfda2 | 8090 | |
6ca09dfc | 8091 | cpumask_clear(span); |
c5f59f08 | 8092 | nodes_clear(used_nodes); |
9c1cfda2 | 8093 | |
6ca09dfc | 8094 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8095 | node_set(node, used_nodes); |
9c1cfda2 JH |
8096 | |
8097 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8098 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8099 | |
6ca09dfc | 8100 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8101 | } |
9c1cfda2 | 8102 | } |
6d6bc0ad | 8103 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8104 | |
5c45bf27 | 8105 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8106 | |
6c99e9ad RR |
8107 | /* |
8108 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8109 | * |
8110 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8111 | * and struct sched_domain. ) | |
6c99e9ad RR |
8112 | */ |
8113 | struct static_sched_group { | |
8114 | struct sched_group sg; | |
8115 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8116 | }; | |
8117 | ||
8118 | struct static_sched_domain { | |
8119 | struct sched_domain sd; | |
8120 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8121 | }; | |
8122 | ||
9c1cfda2 | 8123 | /* |
48f24c4d | 8124 | * SMT sched-domains: |
9c1cfda2 | 8125 | */ |
1da177e4 | 8126 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8127 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8128 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8129 | |
41a2d6cf | 8130 | static int |
96f874e2 RR |
8131 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8132 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8133 | { |
6711cab4 | 8134 | if (sg) |
6c99e9ad | 8135 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8136 | return cpu; |
8137 | } | |
6d6bc0ad | 8138 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8139 | |
48f24c4d IM |
8140 | /* |
8141 | * multi-core sched-domains: | |
8142 | */ | |
1e9f28fa | 8143 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8144 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8145 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8146 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8147 | |
8148 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8149 | static int |
96f874e2 RR |
8150 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8151 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8152 | { |
6711cab4 | 8153 | int group; |
7c16ec58 | 8154 | |
c69fc56d | 8155 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8156 | group = cpumask_first(mask); |
6711cab4 | 8157 | if (sg) |
6c99e9ad | 8158 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8159 | return group; |
1e9f28fa SS |
8160 | } |
8161 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8162 | static int |
96f874e2 RR |
8163 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8164 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8165 | { |
6711cab4 | 8166 | if (sg) |
6c99e9ad | 8167 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8168 | return cpu; |
8169 | } | |
8170 | #endif | |
8171 | ||
6c99e9ad RR |
8172 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8173 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8174 | |
41a2d6cf | 8175 | static int |
96f874e2 RR |
8176 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8177 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8178 | { |
6711cab4 | 8179 | int group; |
48f24c4d | 8180 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8181 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8182 | group = cpumask_first(mask); |
1e9f28fa | 8183 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8184 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8185 | group = cpumask_first(mask); |
1da177e4 | 8186 | #else |
6711cab4 | 8187 | group = cpu; |
1da177e4 | 8188 | #endif |
6711cab4 | 8189 | if (sg) |
6c99e9ad | 8190 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8191 | return group; |
1da177e4 LT |
8192 | } |
8193 | ||
8194 | #ifdef CONFIG_NUMA | |
1da177e4 | 8195 | /* |
9c1cfda2 JH |
8196 | * The init_sched_build_groups can't handle what we want to do with node |
8197 | * groups, so roll our own. Now each node has its own list of groups which | |
8198 | * gets dynamically allocated. | |
1da177e4 | 8199 | */ |
62ea9ceb | 8200 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8201 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8202 | |
62ea9ceb | 8203 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8204 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8205 | |
96f874e2 RR |
8206 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8207 | struct sched_group **sg, | |
8208 | struct cpumask *nodemask) | |
9c1cfda2 | 8209 | { |
6711cab4 SS |
8210 | int group; |
8211 | ||
6ca09dfc | 8212 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8213 | group = cpumask_first(nodemask); |
6711cab4 SS |
8214 | |
8215 | if (sg) | |
6c99e9ad | 8216 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8217 | return group; |
1da177e4 | 8218 | } |
6711cab4 | 8219 | |
08069033 SS |
8220 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8221 | { | |
8222 | struct sched_group *sg = group_head; | |
8223 | int j; | |
8224 | ||
8225 | if (!sg) | |
8226 | return; | |
3a5c359a | 8227 | do { |
758b2cdc | 8228 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8229 | struct sched_domain *sd; |
08069033 | 8230 | |
6c99e9ad | 8231 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8232 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8233 | /* |
8234 | * Only add "power" once for each | |
8235 | * physical package. | |
8236 | */ | |
8237 | continue; | |
8238 | } | |
08069033 | 8239 | |
3a5c359a AK |
8240 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
8241 | } | |
8242 | sg = sg->next; | |
8243 | } while (sg != group_head); | |
08069033 | 8244 | } |
6d6bc0ad | 8245 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8246 | |
a616058b | 8247 | #ifdef CONFIG_NUMA |
51888ca2 | 8248 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8249 | static void free_sched_groups(const struct cpumask *cpu_map, |
8250 | struct cpumask *nodemask) | |
51888ca2 | 8251 | { |
a616058b | 8252 | int cpu, i; |
51888ca2 | 8253 | |
abcd083a | 8254 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8255 | struct sched_group **sched_group_nodes |
8256 | = sched_group_nodes_bycpu[cpu]; | |
8257 | ||
51888ca2 SV |
8258 | if (!sched_group_nodes) |
8259 | continue; | |
8260 | ||
076ac2af | 8261 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8262 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8263 | ||
6ca09dfc | 8264 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8265 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8266 | continue; |
8267 | ||
8268 | if (sg == NULL) | |
8269 | continue; | |
8270 | sg = sg->next; | |
8271 | next_sg: | |
8272 | oldsg = sg; | |
8273 | sg = sg->next; | |
8274 | kfree(oldsg); | |
8275 | if (oldsg != sched_group_nodes[i]) | |
8276 | goto next_sg; | |
8277 | } | |
8278 | kfree(sched_group_nodes); | |
8279 | sched_group_nodes_bycpu[cpu] = NULL; | |
8280 | } | |
51888ca2 | 8281 | } |
6d6bc0ad | 8282 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8283 | static void free_sched_groups(const struct cpumask *cpu_map, |
8284 | struct cpumask *nodemask) | |
a616058b SS |
8285 | { |
8286 | } | |
6d6bc0ad | 8287 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8288 | |
89c4710e SS |
8289 | /* |
8290 | * Initialize sched groups cpu_power. | |
8291 | * | |
8292 | * cpu_power indicates the capacity of sched group, which is used while | |
8293 | * distributing the load between different sched groups in a sched domain. | |
8294 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8295 | * there are asymmetries in the topology. If there are asymmetries, group | |
8296 | * having more cpu_power will pickup more load compared to the group having | |
8297 | * less cpu_power. | |
8298 | * | |
8299 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
8300 | * the maximum number of tasks a group can handle in the presence of other idle | |
8301 | * or lightly loaded groups in the same sched domain. | |
8302 | */ | |
8303 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8304 | { | |
8305 | struct sched_domain *child; | |
8306 | struct sched_group *group; | |
8307 | ||
8308 | WARN_ON(!sd || !sd->groups); | |
8309 | ||
13318a71 | 8310 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8311 | return; |
8312 | ||
8313 | child = sd->child; | |
8314 | ||
5517d86b ED |
8315 | sd->groups->__cpu_power = 0; |
8316 | ||
89c4710e SS |
8317 | /* |
8318 | * For perf policy, if the groups in child domain share resources | |
8319 | * (for example cores sharing some portions of the cache hierarchy | |
8320 | * or SMT), then set this domain groups cpu_power such that each group | |
8321 | * can handle only one task, when there are other idle groups in the | |
8322 | * same sched domain. | |
8323 | */ | |
8324 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
8325 | (child->flags & | |
8326 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 8327 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
8328 | return; |
8329 | } | |
8330 | ||
89c4710e SS |
8331 | /* |
8332 | * add cpu_power of each child group to this groups cpu_power | |
8333 | */ | |
8334 | group = child->groups; | |
8335 | do { | |
5517d86b | 8336 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8337 | group = group->next; |
8338 | } while (group != child->groups); | |
8339 | } | |
8340 | ||
7c16ec58 MT |
8341 | /* |
8342 | * Initializers for schedule domains | |
8343 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8344 | */ | |
8345 | ||
a5d8c348 IM |
8346 | #ifdef CONFIG_SCHED_DEBUG |
8347 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8348 | #else | |
8349 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8350 | #endif | |
8351 | ||
7c16ec58 | 8352 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8353 | |
7c16ec58 MT |
8354 | #define SD_INIT_FUNC(type) \ |
8355 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8356 | { \ | |
8357 | memset(sd, 0, sizeof(*sd)); \ | |
8358 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8359 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8360 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8361 | } |
8362 | ||
8363 | SD_INIT_FUNC(CPU) | |
8364 | #ifdef CONFIG_NUMA | |
8365 | SD_INIT_FUNC(ALLNODES) | |
8366 | SD_INIT_FUNC(NODE) | |
8367 | #endif | |
8368 | #ifdef CONFIG_SCHED_SMT | |
8369 | SD_INIT_FUNC(SIBLING) | |
8370 | #endif | |
8371 | #ifdef CONFIG_SCHED_MC | |
8372 | SD_INIT_FUNC(MC) | |
8373 | #endif | |
8374 | ||
1d3504fc HS |
8375 | static int default_relax_domain_level = -1; |
8376 | ||
8377 | static int __init setup_relax_domain_level(char *str) | |
8378 | { | |
30e0e178 LZ |
8379 | unsigned long val; |
8380 | ||
8381 | val = simple_strtoul(str, NULL, 0); | |
8382 | if (val < SD_LV_MAX) | |
8383 | default_relax_domain_level = val; | |
8384 | ||
1d3504fc HS |
8385 | return 1; |
8386 | } | |
8387 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8388 | ||
8389 | static void set_domain_attribute(struct sched_domain *sd, | |
8390 | struct sched_domain_attr *attr) | |
8391 | { | |
8392 | int request; | |
8393 | ||
8394 | if (!attr || attr->relax_domain_level < 0) { | |
8395 | if (default_relax_domain_level < 0) | |
8396 | return; | |
8397 | else | |
8398 | request = default_relax_domain_level; | |
8399 | } else | |
8400 | request = attr->relax_domain_level; | |
8401 | if (request < sd->level) { | |
8402 | /* turn off idle balance on this domain */ | |
8403 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8404 | } else { | |
8405 | /* turn on idle balance on this domain */ | |
8406 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8407 | } | |
8408 | } | |
8409 | ||
1da177e4 | 8410 | /* |
1a20ff27 DG |
8411 | * Build sched domains for a given set of cpus and attach the sched domains |
8412 | * to the individual cpus | |
1da177e4 | 8413 | */ |
96f874e2 | 8414 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 8415 | struct sched_domain_attr *attr) |
1da177e4 | 8416 | { |
3404c8d9 | 8417 | int i, err = -ENOMEM; |
57d885fe | 8418 | struct root_domain *rd; |
3404c8d9 RR |
8419 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
8420 | tmpmask; | |
d1b55138 | 8421 | #ifdef CONFIG_NUMA |
3404c8d9 | 8422 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 8423 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 8424 | int sd_allnodes = 0; |
d1b55138 | 8425 | |
3404c8d9 RR |
8426 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
8427 | goto out; | |
8428 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
8429 | goto free_domainspan; | |
8430 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
8431 | goto free_covered; | |
8432 | #endif | |
8433 | ||
8434 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
8435 | goto free_notcovered; | |
8436 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
8437 | goto free_nodemask; | |
8438 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
8439 | goto free_this_sibling_map; | |
8440 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
8441 | goto free_this_core_map; | |
8442 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
8443 | goto free_send_covered; | |
8444 | ||
8445 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
8446 | /* |
8447 | * Allocate the per-node list of sched groups | |
8448 | */ | |
076ac2af | 8449 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 8450 | GFP_KERNEL); |
d1b55138 JH |
8451 | if (!sched_group_nodes) { |
8452 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 8453 | goto free_tmpmask; |
d1b55138 | 8454 | } |
d1b55138 | 8455 | #endif |
1da177e4 | 8456 | |
dc938520 | 8457 | rd = alloc_rootdomain(); |
57d885fe GH |
8458 | if (!rd) { |
8459 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 8460 | goto free_sched_groups; |
57d885fe GH |
8461 | } |
8462 | ||
7c16ec58 | 8463 | #ifdef CONFIG_NUMA |
96f874e2 | 8464 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
8465 | #endif |
8466 | ||
1da177e4 | 8467 | /* |
1a20ff27 | 8468 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8469 | */ |
abcd083a | 8470 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8471 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 8472 | |
6ca09dfc | 8473 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
8474 | |
8475 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
8476 | if (cpumask_weight(cpu_map) > |
8477 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 8478 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 8479 | SD_INIT(sd, ALLNODES); |
1d3504fc | 8480 | set_domain_attribute(sd, attr); |
758b2cdc | 8481 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 8482 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 8483 | p = sd; |
6711cab4 | 8484 | sd_allnodes = 1; |
9c1cfda2 JH |
8485 | } else |
8486 | p = NULL; | |
8487 | ||
62ea9ceb | 8488 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 8489 | SD_INIT(sd, NODE); |
1d3504fc | 8490 | set_domain_attribute(sd, attr); |
758b2cdc | 8491 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 8492 | sd->parent = p; |
1a848870 SS |
8493 | if (p) |
8494 | p->child = sd; | |
758b2cdc RR |
8495 | cpumask_and(sched_domain_span(sd), |
8496 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
8497 | #endif |
8498 | ||
8499 | p = sd; | |
6c99e9ad | 8500 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 8501 | SD_INIT(sd, CPU); |
1d3504fc | 8502 | set_domain_attribute(sd, attr); |
758b2cdc | 8503 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8504 | sd->parent = p; |
1a848870 SS |
8505 | if (p) |
8506 | p->child = sd; | |
7c16ec58 | 8507 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8508 | |
1e9f28fa SS |
8509 | #ifdef CONFIG_SCHED_MC |
8510 | p = sd; | |
6c99e9ad | 8511 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8512 | SD_INIT(sd, MC); |
1d3504fc | 8513 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8514 | cpumask_and(sched_domain_span(sd), cpu_map, |
8515 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8516 | sd->parent = p; |
1a848870 | 8517 | p->child = sd; |
7c16ec58 | 8518 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8519 | #endif |
8520 | ||
1da177e4 LT |
8521 | #ifdef CONFIG_SCHED_SMT |
8522 | p = sd; | |
6c99e9ad | 8523 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8524 | SD_INIT(sd, SIBLING); |
1d3504fc | 8525 | set_domain_attribute(sd, attr); |
758b2cdc | 8526 | cpumask_and(sched_domain_span(sd), |
c69fc56d | 8527 | topology_thread_cpumask(i), cpu_map); |
1da177e4 | 8528 | sd->parent = p; |
1a848870 | 8529 | p->child = sd; |
7c16ec58 | 8530 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8531 | #endif |
8532 | } | |
8533 | ||
8534 | #ifdef CONFIG_SCHED_SMT | |
8535 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8536 | for_each_cpu(i, cpu_map) { |
96f874e2 | 8537 | cpumask_and(this_sibling_map, |
c69fc56d | 8538 | topology_thread_cpumask(i), cpu_map); |
96f874e2 | 8539 | if (i != cpumask_first(this_sibling_map)) |
1da177e4 LT |
8540 | continue; |
8541 | ||
dd41f596 | 8542 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8543 | &cpu_to_cpu_group, |
8544 | send_covered, tmpmask); | |
1da177e4 LT |
8545 | } |
8546 | #endif | |
8547 | ||
1e9f28fa SS |
8548 | #ifdef CONFIG_SCHED_MC |
8549 | /* Set up multi-core groups */ | |
abcd083a | 8550 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8551 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8552 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8553 | continue; |
7c16ec58 | 8554 | |
dd41f596 | 8555 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8556 | &cpu_to_core_group, |
8557 | send_covered, tmpmask); | |
1e9f28fa SS |
8558 | } |
8559 | #endif | |
8560 | ||
1da177e4 | 8561 | /* Set up physical groups */ |
076ac2af | 8562 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8563 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8564 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8565 | continue; |
8566 | ||
7c16ec58 MT |
8567 | init_sched_build_groups(nodemask, cpu_map, |
8568 | &cpu_to_phys_group, | |
8569 | send_covered, tmpmask); | |
1da177e4 LT |
8570 | } |
8571 | ||
8572 | #ifdef CONFIG_NUMA | |
8573 | /* Set up node groups */ | |
7c16ec58 | 8574 | if (sd_allnodes) { |
7c16ec58 MT |
8575 | init_sched_build_groups(cpu_map, cpu_map, |
8576 | &cpu_to_allnodes_group, | |
8577 | send_covered, tmpmask); | |
8578 | } | |
9c1cfda2 | 8579 | |
076ac2af | 8580 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8581 | /* Set up node groups */ |
8582 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8583 | int j; |
8584 | ||
96f874e2 | 8585 | cpumask_clear(covered); |
6ca09dfc | 8586 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8587 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8588 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8589 | continue; |
d1b55138 | 8590 | } |
9c1cfda2 | 8591 | |
4bdbaad3 | 8592 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8593 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8594 | |
6c99e9ad RR |
8595 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8596 | GFP_KERNEL, i); | |
51888ca2 SV |
8597 | if (!sg) { |
8598 | printk(KERN_WARNING "Can not alloc domain group for " | |
8599 | "node %d\n", i); | |
8600 | goto error; | |
8601 | } | |
9c1cfda2 | 8602 | sched_group_nodes[i] = sg; |
abcd083a | 8603 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8604 | struct sched_domain *sd; |
9761eea8 | 8605 | |
62ea9ceb | 8606 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8607 | sd->groups = sg; |
9c1cfda2 | 8608 | } |
5517d86b | 8609 | sg->__cpu_power = 0; |
758b2cdc | 8610 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8611 | sg->next = sg; |
96f874e2 | 8612 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8613 | prev = sg; |
8614 | ||
076ac2af | 8615 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8616 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8617 | |
96f874e2 RR |
8618 | cpumask_complement(notcovered, covered); |
8619 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8620 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8621 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8622 | break; |
8623 | ||
6ca09dfc | 8624 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8625 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8626 | continue; |
8627 | ||
6c99e9ad RR |
8628 | sg = kmalloc_node(sizeof(struct sched_group) + |
8629 | cpumask_size(), | |
15f0b676 | 8630 | GFP_KERNEL, i); |
9c1cfda2 JH |
8631 | if (!sg) { |
8632 | printk(KERN_WARNING | |
8633 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8634 | goto error; |
9c1cfda2 | 8635 | } |
5517d86b | 8636 | sg->__cpu_power = 0; |
758b2cdc | 8637 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8638 | sg->next = prev->next; |
96f874e2 | 8639 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8640 | prev->next = sg; |
8641 | prev = sg; | |
8642 | } | |
9c1cfda2 | 8643 | } |
1da177e4 LT |
8644 | #endif |
8645 | ||
8646 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8647 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8648 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8649 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8650 | |
89c4710e | 8651 | init_sched_groups_power(i, sd); |
5c45bf27 | 8652 | } |
1da177e4 | 8653 | #endif |
1e9f28fa | 8654 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8655 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8656 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8657 | |
89c4710e | 8658 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8659 | } |
8660 | #endif | |
1e9f28fa | 8661 | |
abcd083a | 8662 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8663 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8664 | |
89c4710e | 8665 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8666 | } |
8667 | ||
9c1cfda2 | 8668 | #ifdef CONFIG_NUMA |
076ac2af | 8669 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8670 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8671 | |
6711cab4 SS |
8672 | if (sd_allnodes) { |
8673 | struct sched_group *sg; | |
f712c0c7 | 8674 | |
96f874e2 | 8675 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8676 | tmpmask); |
f712c0c7 SS |
8677 | init_numa_sched_groups_power(sg); |
8678 | } | |
9c1cfda2 JH |
8679 | #endif |
8680 | ||
1da177e4 | 8681 | /* Attach the domains */ |
abcd083a | 8682 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8683 | struct sched_domain *sd; |
8684 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8685 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8686 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8687 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8688 | #else |
6c99e9ad | 8689 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8690 | #endif |
57d885fe | 8691 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8692 | } |
51888ca2 | 8693 | |
3404c8d9 RR |
8694 | err = 0; |
8695 | ||
8696 | free_tmpmask: | |
8697 | free_cpumask_var(tmpmask); | |
8698 | free_send_covered: | |
8699 | free_cpumask_var(send_covered); | |
8700 | free_this_core_map: | |
8701 | free_cpumask_var(this_core_map); | |
8702 | free_this_sibling_map: | |
8703 | free_cpumask_var(this_sibling_map); | |
8704 | free_nodemask: | |
8705 | free_cpumask_var(nodemask); | |
8706 | free_notcovered: | |
8707 | #ifdef CONFIG_NUMA | |
8708 | free_cpumask_var(notcovered); | |
8709 | free_covered: | |
8710 | free_cpumask_var(covered); | |
8711 | free_domainspan: | |
8712 | free_cpumask_var(domainspan); | |
8713 | out: | |
8714 | #endif | |
8715 | return err; | |
8716 | ||
8717 | free_sched_groups: | |
8718 | #ifdef CONFIG_NUMA | |
8719 | kfree(sched_group_nodes); | |
8720 | #endif | |
8721 | goto free_tmpmask; | |
51888ca2 | 8722 | |
a616058b | 8723 | #ifdef CONFIG_NUMA |
51888ca2 | 8724 | error: |
7c16ec58 | 8725 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8726 | free_rootdomain(rd); |
3404c8d9 | 8727 | goto free_tmpmask; |
a616058b | 8728 | #endif |
1da177e4 | 8729 | } |
029190c5 | 8730 | |
96f874e2 | 8731 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8732 | { |
8733 | return __build_sched_domains(cpu_map, NULL); | |
8734 | } | |
8735 | ||
96f874e2 | 8736 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8737 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8738 | static struct sched_domain_attr *dattr_cur; |
8739 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8740 | |
8741 | /* | |
8742 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8743 | * cpumask) fails, then fallback to a single sched domain, |
8744 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8745 | */ |
4212823f | 8746 | static cpumask_var_t fallback_doms; |
029190c5 | 8747 | |
ee79d1bd HC |
8748 | /* |
8749 | * arch_update_cpu_topology lets virtualized architectures update the | |
8750 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8751 | * or 0 if it stayed the same. | |
8752 | */ | |
8753 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8754 | { |
ee79d1bd | 8755 | return 0; |
22e52b07 HC |
8756 | } |
8757 | ||
1a20ff27 | 8758 | /* |
41a2d6cf | 8759 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8760 | * For now this just excludes isolated cpus, but could be used to |
8761 | * exclude other special cases in the future. | |
1a20ff27 | 8762 | */ |
96f874e2 | 8763 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8764 | { |
7378547f MM |
8765 | int err; |
8766 | ||
22e52b07 | 8767 | arch_update_cpu_topology(); |
029190c5 | 8768 | ndoms_cur = 1; |
96f874e2 | 8769 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8770 | if (!doms_cur) |
4212823f | 8771 | doms_cur = fallback_doms; |
dcc30a35 | 8772 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8773 | dattr_cur = NULL; |
7378547f | 8774 | err = build_sched_domains(doms_cur); |
6382bc90 | 8775 | register_sched_domain_sysctl(); |
7378547f MM |
8776 | |
8777 | return err; | |
1a20ff27 DG |
8778 | } |
8779 | ||
96f874e2 RR |
8780 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8781 | struct cpumask *tmpmask) | |
1da177e4 | 8782 | { |
7c16ec58 | 8783 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8784 | } |
1da177e4 | 8785 | |
1a20ff27 DG |
8786 | /* |
8787 | * Detach sched domains from a group of cpus specified in cpu_map | |
8788 | * These cpus will now be attached to the NULL domain | |
8789 | */ | |
96f874e2 | 8790 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8791 | { |
96f874e2 RR |
8792 | /* Save because hotplug lock held. */ |
8793 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8794 | int i; |
8795 | ||
abcd083a | 8796 | for_each_cpu(i, cpu_map) |
57d885fe | 8797 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8798 | synchronize_sched(); |
96f874e2 | 8799 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8800 | } |
8801 | ||
1d3504fc HS |
8802 | /* handle null as "default" */ |
8803 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8804 | struct sched_domain_attr *new, int idx_new) | |
8805 | { | |
8806 | struct sched_domain_attr tmp; | |
8807 | ||
8808 | /* fast path */ | |
8809 | if (!new && !cur) | |
8810 | return 1; | |
8811 | ||
8812 | tmp = SD_ATTR_INIT; | |
8813 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8814 | new ? (new + idx_new) : &tmp, | |
8815 | sizeof(struct sched_domain_attr)); | |
8816 | } | |
8817 | ||
029190c5 PJ |
8818 | /* |
8819 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8820 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8821 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8822 | * It destroys each deleted domain and builds each new domain. | |
8823 | * | |
96f874e2 | 8824 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8825 | * The masks don't intersect (don't overlap.) We should setup one |
8826 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8827 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8828 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8829 | * it as it is. | |
8830 | * | |
41a2d6cf IM |
8831 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8832 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8833 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8834 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8835 | * the single partition 'fallback_doms', it also forces the domains | |
8836 | * to be rebuilt. | |
029190c5 | 8837 | * |
96f874e2 | 8838 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8839 | * ndoms_new == 0 is a special case for destroying existing domains, |
8840 | * and it will not create the default domain. | |
dfb512ec | 8841 | * |
029190c5 PJ |
8842 | * Call with hotplug lock held |
8843 | */ | |
96f874e2 RR |
8844 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8845 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8846 | struct sched_domain_attr *dattr_new) |
029190c5 | 8847 | { |
dfb512ec | 8848 | int i, j, n; |
d65bd5ec | 8849 | int new_topology; |
029190c5 | 8850 | |
712555ee | 8851 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8852 | |
7378547f MM |
8853 | /* always unregister in case we don't destroy any domains */ |
8854 | unregister_sched_domain_sysctl(); | |
8855 | ||
d65bd5ec HC |
8856 | /* Let architecture update cpu core mappings. */ |
8857 | new_topology = arch_update_cpu_topology(); | |
8858 | ||
dfb512ec | 8859 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8860 | |
8861 | /* Destroy deleted domains */ | |
8862 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8863 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8864 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8865 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8866 | goto match1; |
8867 | } | |
8868 | /* no match - a current sched domain not in new doms_new[] */ | |
8869 | detach_destroy_domains(doms_cur + i); | |
8870 | match1: | |
8871 | ; | |
8872 | } | |
8873 | ||
e761b772 MK |
8874 | if (doms_new == NULL) { |
8875 | ndoms_cur = 0; | |
4212823f | 8876 | doms_new = fallback_doms; |
dcc30a35 | 8877 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8878 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8879 | } |
8880 | ||
029190c5 PJ |
8881 | /* Build new domains */ |
8882 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8883 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8884 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8885 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8886 | goto match2; |
8887 | } | |
8888 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8889 | __build_sched_domains(doms_new + i, |
8890 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8891 | match2: |
8892 | ; | |
8893 | } | |
8894 | ||
8895 | /* Remember the new sched domains */ | |
4212823f | 8896 | if (doms_cur != fallback_doms) |
029190c5 | 8897 | kfree(doms_cur); |
1d3504fc | 8898 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8899 | doms_cur = doms_new; |
1d3504fc | 8900 | dattr_cur = dattr_new; |
029190c5 | 8901 | ndoms_cur = ndoms_new; |
7378547f MM |
8902 | |
8903 | register_sched_domain_sysctl(); | |
a1835615 | 8904 | |
712555ee | 8905 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8906 | } |
8907 | ||
5c45bf27 | 8908 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8909 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8910 | { |
95402b38 | 8911 | get_online_cpus(); |
dfb512ec MK |
8912 | |
8913 | /* Destroy domains first to force the rebuild */ | |
8914 | partition_sched_domains(0, NULL, NULL); | |
8915 | ||
e761b772 | 8916 | rebuild_sched_domains(); |
95402b38 | 8917 | put_online_cpus(); |
5c45bf27 SS |
8918 | } |
8919 | ||
8920 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8921 | { | |
afb8a9b7 | 8922 | unsigned int level = 0; |
5c45bf27 | 8923 | |
afb8a9b7 GS |
8924 | if (sscanf(buf, "%u", &level) != 1) |
8925 | return -EINVAL; | |
8926 | ||
8927 | /* | |
8928 | * level is always be positive so don't check for | |
8929 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8930 | * What happens on 0 or 1 byte write, | |
8931 | * need to check for count as well? | |
8932 | */ | |
8933 | ||
8934 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
8935 | return -EINVAL; |
8936 | ||
8937 | if (smt) | |
afb8a9b7 | 8938 | sched_smt_power_savings = level; |
5c45bf27 | 8939 | else |
afb8a9b7 | 8940 | sched_mc_power_savings = level; |
5c45bf27 | 8941 | |
c70f22d2 | 8942 | arch_reinit_sched_domains(); |
5c45bf27 | 8943 | |
c70f22d2 | 8944 | return count; |
5c45bf27 SS |
8945 | } |
8946 | ||
5c45bf27 | 8947 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8948 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8949 | char *page) | |
5c45bf27 SS |
8950 | { |
8951 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8952 | } | |
f718cd4a | 8953 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8954 | const char *buf, size_t count) |
5c45bf27 SS |
8955 | { |
8956 | return sched_power_savings_store(buf, count, 0); | |
8957 | } | |
f718cd4a AK |
8958 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8959 | sched_mc_power_savings_show, | |
8960 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8961 | #endif |
8962 | ||
8963 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8964 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8965 | char *page) | |
5c45bf27 SS |
8966 | { |
8967 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8968 | } | |
f718cd4a | 8969 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8970 | const char *buf, size_t count) |
5c45bf27 SS |
8971 | { |
8972 | return sched_power_savings_store(buf, count, 1); | |
8973 | } | |
f718cd4a AK |
8974 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8975 | sched_smt_power_savings_show, | |
6707de00 AB |
8976 | sched_smt_power_savings_store); |
8977 | #endif | |
8978 | ||
39aac648 | 8979 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8980 | { |
8981 | int err = 0; | |
8982 | ||
8983 | #ifdef CONFIG_SCHED_SMT | |
8984 | if (smt_capable()) | |
8985 | err = sysfs_create_file(&cls->kset.kobj, | |
8986 | &attr_sched_smt_power_savings.attr); | |
8987 | #endif | |
8988 | #ifdef CONFIG_SCHED_MC | |
8989 | if (!err && mc_capable()) | |
8990 | err = sysfs_create_file(&cls->kset.kobj, | |
8991 | &attr_sched_mc_power_savings.attr); | |
8992 | #endif | |
8993 | return err; | |
8994 | } | |
6d6bc0ad | 8995 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8996 | |
e761b772 | 8997 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8998 | /* |
e761b772 MK |
8999 | * Add online and remove offline CPUs from the scheduler domains. |
9000 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
9001 | */ |
9002 | static int update_sched_domains(struct notifier_block *nfb, | |
9003 | unsigned long action, void *hcpu) | |
e761b772 MK |
9004 | { |
9005 | switch (action) { | |
9006 | case CPU_ONLINE: | |
9007 | case CPU_ONLINE_FROZEN: | |
9008 | case CPU_DEAD: | |
9009 | case CPU_DEAD_FROZEN: | |
dfb512ec | 9010 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
9011 | return NOTIFY_OK; |
9012 | ||
9013 | default: | |
9014 | return NOTIFY_DONE; | |
9015 | } | |
9016 | } | |
9017 | #endif | |
9018 | ||
9019 | static int update_runtime(struct notifier_block *nfb, | |
9020 | unsigned long action, void *hcpu) | |
1da177e4 | 9021 | { |
7def2be1 PZ |
9022 | int cpu = (int)(long)hcpu; |
9023 | ||
1da177e4 | 9024 | switch (action) { |
1da177e4 | 9025 | case CPU_DOWN_PREPARE: |
8bb78442 | 9026 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 9027 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
9028 | return NOTIFY_OK; |
9029 | ||
1da177e4 | 9030 | case CPU_DOWN_FAILED: |
8bb78442 | 9031 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 9032 | case CPU_ONLINE: |
8bb78442 | 9033 | case CPU_ONLINE_FROZEN: |
7def2be1 | 9034 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
9035 | return NOTIFY_OK; |
9036 | ||
1da177e4 LT |
9037 | default: |
9038 | return NOTIFY_DONE; | |
9039 | } | |
1da177e4 | 9040 | } |
1da177e4 LT |
9041 | |
9042 | void __init sched_init_smp(void) | |
9043 | { | |
dcc30a35 RR |
9044 | cpumask_var_t non_isolated_cpus; |
9045 | ||
9046 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 9047 | |
434d53b0 MT |
9048 | #if defined(CONFIG_NUMA) |
9049 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
9050 | GFP_KERNEL); | |
9051 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
9052 | #endif | |
95402b38 | 9053 | get_online_cpus(); |
712555ee | 9054 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
9055 | arch_init_sched_domains(cpu_online_mask); |
9056 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
9057 | if (cpumask_empty(non_isolated_cpus)) | |
9058 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9059 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9060 | put_online_cpus(); |
e761b772 MK |
9061 | |
9062 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9063 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9064 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9065 | #endif |
9066 | ||
9067 | /* RT runtime code needs to handle some hotplug events */ | |
9068 | hotcpu_notifier(update_runtime, 0); | |
9069 | ||
b328ca18 | 9070 | init_hrtick(); |
5c1e1767 NP |
9071 | |
9072 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9073 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9074 | BUG(); |
19978ca6 | 9075 | sched_init_granularity(); |
dcc30a35 | 9076 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
9077 | |
9078 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 9079 | init_sched_rt_class(); |
1da177e4 LT |
9080 | } |
9081 | #else | |
9082 | void __init sched_init_smp(void) | |
9083 | { | |
19978ca6 | 9084 | sched_init_granularity(); |
1da177e4 LT |
9085 | } |
9086 | #endif /* CONFIG_SMP */ | |
9087 | ||
cd1bb94b AB |
9088 | const_debug unsigned int sysctl_timer_migration = 1; |
9089 | ||
1da177e4 LT |
9090 | int in_sched_functions(unsigned long addr) |
9091 | { | |
1da177e4 LT |
9092 | return in_lock_functions(addr) || |
9093 | (addr >= (unsigned long)__sched_text_start | |
9094 | && addr < (unsigned long)__sched_text_end); | |
9095 | } | |
9096 | ||
a9957449 | 9097 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9098 | { |
9099 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9100 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9101 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9102 | cfs_rq->rq = rq; | |
9103 | #endif | |
67e9fb2a | 9104 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9105 | } |
9106 | ||
fa85ae24 PZ |
9107 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9108 | { | |
9109 | struct rt_prio_array *array; | |
9110 | int i; | |
9111 | ||
9112 | array = &rt_rq->active; | |
9113 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9114 | INIT_LIST_HEAD(array->queue + i); | |
9115 | __clear_bit(i, array->bitmap); | |
9116 | } | |
9117 | /* delimiter for bitsearch: */ | |
9118 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9119 | ||
052f1dc7 | 9120 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9121 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9122 | #ifdef CONFIG_SMP |
e864c499 | 9123 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9124 | #endif |
48d5e258 | 9125 | #endif |
fa85ae24 PZ |
9126 | #ifdef CONFIG_SMP |
9127 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9128 | rt_rq->overloaded = 0; |
c20b08e3 | 9129 | plist_head_init(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9130 | #endif |
9131 | ||
9132 | rt_rq->rt_time = 0; | |
9133 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9134 | rt_rq->rt_runtime = 0; |
9135 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9136 | |
052f1dc7 | 9137 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9138 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9139 | rt_rq->rq = rq; |
9140 | #endif | |
fa85ae24 PZ |
9141 | } |
9142 | ||
6f505b16 | 9143 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9144 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9145 | struct sched_entity *se, int cpu, int add, | |
9146 | struct sched_entity *parent) | |
6f505b16 | 9147 | { |
ec7dc8ac | 9148 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9149 | tg->cfs_rq[cpu] = cfs_rq; |
9150 | init_cfs_rq(cfs_rq, rq); | |
9151 | cfs_rq->tg = tg; | |
9152 | if (add) | |
9153 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9154 | ||
9155 | tg->se[cpu] = se; | |
354d60c2 DG |
9156 | /* se could be NULL for init_task_group */ |
9157 | if (!se) | |
9158 | return; | |
9159 | ||
ec7dc8ac DG |
9160 | if (!parent) |
9161 | se->cfs_rq = &rq->cfs; | |
9162 | else | |
9163 | se->cfs_rq = parent->my_q; | |
9164 | ||
6f505b16 PZ |
9165 | se->my_q = cfs_rq; |
9166 | se->load.weight = tg->shares; | |
e05510d0 | 9167 | se->load.inv_weight = 0; |
ec7dc8ac | 9168 | se->parent = parent; |
6f505b16 | 9169 | } |
052f1dc7 | 9170 | #endif |
6f505b16 | 9171 | |
052f1dc7 | 9172 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9173 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9174 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9175 | struct sched_rt_entity *parent) | |
6f505b16 | 9176 | { |
ec7dc8ac DG |
9177 | struct rq *rq = cpu_rq(cpu); |
9178 | ||
6f505b16 PZ |
9179 | tg->rt_rq[cpu] = rt_rq; |
9180 | init_rt_rq(rt_rq, rq); | |
9181 | rt_rq->tg = tg; | |
9182 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9183 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9184 | if (add) |
9185 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9186 | ||
9187 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9188 | if (!rt_se) |
9189 | return; | |
9190 | ||
ec7dc8ac DG |
9191 | if (!parent) |
9192 | rt_se->rt_rq = &rq->rt; | |
9193 | else | |
9194 | rt_se->rt_rq = parent->my_q; | |
9195 | ||
6f505b16 | 9196 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9197 | rt_se->parent = parent; |
6f505b16 PZ |
9198 | INIT_LIST_HEAD(&rt_se->run_list); |
9199 | } | |
9200 | #endif | |
9201 | ||
1da177e4 LT |
9202 | void __init sched_init(void) |
9203 | { | |
dd41f596 | 9204 | int i, j; |
434d53b0 MT |
9205 | unsigned long alloc_size = 0, ptr; |
9206 | ||
9207 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9208 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9209 | #endif | |
9210 | #ifdef CONFIG_RT_GROUP_SCHED | |
9211 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9212 | #endif |
9213 | #ifdef CONFIG_USER_SCHED | |
9214 | alloc_size *= 2; | |
df7c8e84 RR |
9215 | #endif |
9216 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9217 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9218 | #endif |
9219 | /* | |
9220 | * As sched_init() is called before page_alloc is setup, | |
9221 | * we use alloc_bootmem(). | |
9222 | */ | |
9223 | if (alloc_size) { | |
36b7b6d4 | 9224 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9225 | |
9226 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9227 | init_task_group.se = (struct sched_entity **)ptr; | |
9228 | ptr += nr_cpu_ids * sizeof(void **); | |
9229 | ||
9230 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9231 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9232 | |
9233 | #ifdef CONFIG_USER_SCHED | |
9234 | root_task_group.se = (struct sched_entity **)ptr; | |
9235 | ptr += nr_cpu_ids * sizeof(void **); | |
9236 | ||
9237 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9238 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9239 | #endif /* CONFIG_USER_SCHED */ |
9240 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9241 | #ifdef CONFIG_RT_GROUP_SCHED |
9242 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9243 | ptr += nr_cpu_ids * sizeof(void **); | |
9244 | ||
9245 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9246 | ptr += nr_cpu_ids * sizeof(void **); |
9247 | ||
9248 | #ifdef CONFIG_USER_SCHED | |
9249 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9250 | ptr += nr_cpu_ids * sizeof(void **); | |
9251 | ||
9252 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9253 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9254 | #endif /* CONFIG_USER_SCHED */ |
9255 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9256 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9257 | for_each_possible_cpu(i) { | |
9258 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9259 | ptr += cpumask_size(); | |
9260 | } | |
9261 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9262 | } |
dd41f596 | 9263 | |
57d885fe GH |
9264 | #ifdef CONFIG_SMP |
9265 | init_defrootdomain(); | |
9266 | #endif | |
9267 | ||
d0b27fa7 PZ |
9268 | init_rt_bandwidth(&def_rt_bandwidth, |
9269 | global_rt_period(), global_rt_runtime()); | |
9270 | ||
9271 | #ifdef CONFIG_RT_GROUP_SCHED | |
9272 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9273 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9274 | #ifdef CONFIG_USER_SCHED |
9275 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9276 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9277 | #endif /* CONFIG_USER_SCHED */ |
9278 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9279 | |
052f1dc7 | 9280 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9281 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9282 | INIT_LIST_HEAD(&init_task_group.children); |
9283 | ||
9284 | #ifdef CONFIG_USER_SCHED | |
9285 | INIT_LIST_HEAD(&root_task_group.children); | |
9286 | init_task_group.parent = &root_task_group; | |
9287 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9288 | #endif /* CONFIG_USER_SCHED */ |
9289 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9290 | |
0a945022 | 9291 | for_each_possible_cpu(i) { |
70b97a7f | 9292 | struct rq *rq; |
1da177e4 LT |
9293 | |
9294 | rq = cpu_rq(i); | |
9295 | spin_lock_init(&rq->lock); | |
7897986b | 9296 | rq->nr_running = 0; |
dce48a84 TG |
9297 | rq->calc_load_active = 0; |
9298 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9299 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9300 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9301 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9302 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9303 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9304 | #ifdef CONFIG_CGROUP_SCHED |
9305 | /* | |
9306 | * How much cpu bandwidth does init_task_group get? | |
9307 | * | |
9308 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9309 | * gets 100% of the cpu resources in the system. This overall | |
9310 | * system cpu resource is divided among the tasks of | |
9311 | * init_task_group and its child task-groups in a fair manner, | |
9312 | * based on each entity's (task or task-group's) weight | |
9313 | * (se->load.weight). | |
9314 | * | |
9315 | * In other words, if init_task_group has 10 tasks of weight | |
9316 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9317 | * then A0's share of the cpu resource is: | |
9318 | * | |
0d905bca | 9319 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9320 | * |
9321 | * We achieve this by letting init_task_group's tasks sit | |
9322 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9323 | */ | |
ec7dc8ac | 9324 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9325 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9326 | root_task_group.shares = NICE_0_LOAD; |
9327 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9328 | /* |
9329 | * In case of task-groups formed thr' the user id of tasks, | |
9330 | * init_task_group represents tasks belonging to root user. | |
9331 | * Hence it forms a sibling of all subsequent groups formed. | |
9332 | * In this case, init_task_group gets only a fraction of overall | |
9333 | * system cpu resource, based on the weight assigned to root | |
9334 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9335 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
9336 | * (init_cfs_rq) and having one entity represent this group of | |
9337 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
9338 | */ | |
ec7dc8ac | 9339 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 9340 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
9341 | &per_cpu(init_sched_entity, i), i, 1, |
9342 | root_task_group.se[i]); | |
6f505b16 | 9343 | |
052f1dc7 | 9344 | #endif |
354d60c2 DG |
9345 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9346 | ||
9347 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9348 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9349 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9350 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9351 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9352 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9353 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9354 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9355 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9356 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9357 | root_task_group.rt_se[i]); | |
354d60c2 | 9358 | #endif |
dd41f596 | 9359 | #endif |
1da177e4 | 9360 | |
dd41f596 IM |
9361 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9362 | rq->cpu_load[j] = 0; | |
1da177e4 | 9363 | #ifdef CONFIG_SMP |
41c7ce9a | 9364 | rq->sd = NULL; |
57d885fe | 9365 | rq->rd = NULL; |
1da177e4 | 9366 | rq->active_balance = 0; |
dd41f596 | 9367 | rq->next_balance = jiffies; |
1da177e4 | 9368 | rq->push_cpu = 0; |
0a2966b4 | 9369 | rq->cpu = i; |
1f11eb6a | 9370 | rq->online = 0; |
1da177e4 LT |
9371 | rq->migration_thread = NULL; |
9372 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9373 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9374 | #endif |
8f4d37ec | 9375 | init_rq_hrtick(rq); |
1da177e4 | 9376 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9377 | } |
9378 | ||
2dd73a4f | 9379 | set_load_weight(&init_task); |
b50f60ce | 9380 | |
e107be36 AK |
9381 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9382 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9383 | #endif | |
9384 | ||
c9819f45 | 9385 | #ifdef CONFIG_SMP |
962cf36c | 9386 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9387 | #endif |
9388 | ||
b50f60ce HC |
9389 | #ifdef CONFIG_RT_MUTEXES |
9390 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9391 | #endif | |
9392 | ||
1da177e4 LT |
9393 | /* |
9394 | * The boot idle thread does lazy MMU switching as well: | |
9395 | */ | |
9396 | atomic_inc(&init_mm.mm_count); | |
9397 | enter_lazy_tlb(&init_mm, current); | |
9398 | ||
9399 | /* | |
9400 | * Make us the idle thread. Technically, schedule() should not be | |
9401 | * called from this thread, however somewhere below it might be, | |
9402 | * but because we are the idle thread, we just pick up running again | |
9403 | * when this runqueue becomes "idle". | |
9404 | */ | |
9405 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9406 | |
9407 | calc_load_update = jiffies + LOAD_FREQ; | |
9408 | ||
dd41f596 IM |
9409 | /* |
9410 | * During early bootup we pretend to be a normal task: | |
9411 | */ | |
9412 | current->sched_class = &fair_sched_class; | |
6892b75e | 9413 | |
6a7b3dc3 | 9414 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
4bdddf8f | 9415 | alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9416 | #ifdef CONFIG_SMP |
7d1e6a9b | 9417 | #ifdef CONFIG_NO_HZ |
4bdddf8f PE |
9418 | alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
9419 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); | |
7d1e6a9b | 9420 | #endif |
4bdddf8f | 9421 | alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
bf4d83f6 | 9422 | #endif /* SMP */ |
6a7b3dc3 | 9423 | |
0d905bca IM |
9424 | perf_counter_init(); |
9425 | ||
6892b75e | 9426 | scheduler_running = 1; |
1da177e4 LT |
9427 | } |
9428 | ||
9429 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
9430 | static inline int preempt_count_equals(int preempt_offset) |
9431 | { | |
9432 | int nested = preempt_count() & ~PREEMPT_ACTIVE; | |
9433 | ||
9434 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
9435 | } | |
9436 | ||
9437 | void __might_sleep(char *file, int line, int preempt_offset) | |
1da177e4 | 9438 | { |
48f24c4d | 9439 | #ifdef in_atomic |
1da177e4 LT |
9440 | static unsigned long prev_jiffy; /* ratelimiting */ |
9441 | ||
e4aafea2 FW |
9442 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
9443 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
9444 | return; |
9445 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9446 | return; | |
9447 | prev_jiffy = jiffies; | |
9448 | ||
9449 | printk(KERN_ERR | |
9450 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9451 | file, line); | |
9452 | printk(KERN_ERR | |
9453 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9454 | in_atomic(), irqs_disabled(), | |
9455 | current->pid, current->comm); | |
9456 | ||
9457 | debug_show_held_locks(current); | |
9458 | if (irqs_disabled()) | |
9459 | print_irqtrace_events(current); | |
9460 | dump_stack(); | |
1da177e4 LT |
9461 | #endif |
9462 | } | |
9463 | EXPORT_SYMBOL(__might_sleep); | |
9464 | #endif | |
9465 | ||
9466 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9467 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9468 | { | |
9469 | int on_rq; | |
3e51f33f | 9470 | |
3a5e4dc1 AK |
9471 | update_rq_clock(rq); |
9472 | on_rq = p->se.on_rq; | |
9473 | if (on_rq) | |
9474 | deactivate_task(rq, p, 0); | |
9475 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9476 | if (on_rq) { | |
9477 | activate_task(rq, p, 0); | |
9478 | resched_task(rq->curr); | |
9479 | } | |
9480 | } | |
9481 | ||
1da177e4 LT |
9482 | void normalize_rt_tasks(void) |
9483 | { | |
a0f98a1c | 9484 | struct task_struct *g, *p; |
1da177e4 | 9485 | unsigned long flags; |
70b97a7f | 9486 | struct rq *rq; |
1da177e4 | 9487 | |
4cf5d77a | 9488 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9489 | do_each_thread(g, p) { |
178be793 IM |
9490 | /* |
9491 | * Only normalize user tasks: | |
9492 | */ | |
9493 | if (!p->mm) | |
9494 | continue; | |
9495 | ||
6cfb0d5d | 9496 | p->se.exec_start = 0; |
6cfb0d5d | 9497 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9498 | p->se.wait_start = 0; |
dd41f596 | 9499 | p->se.sleep_start = 0; |
dd41f596 | 9500 | p->se.block_start = 0; |
6cfb0d5d | 9501 | #endif |
dd41f596 IM |
9502 | |
9503 | if (!rt_task(p)) { | |
9504 | /* | |
9505 | * Renice negative nice level userspace | |
9506 | * tasks back to 0: | |
9507 | */ | |
9508 | if (TASK_NICE(p) < 0 && p->mm) | |
9509 | set_user_nice(p, 0); | |
1da177e4 | 9510 | continue; |
dd41f596 | 9511 | } |
1da177e4 | 9512 | |
4cf5d77a | 9513 | spin_lock(&p->pi_lock); |
b29739f9 | 9514 | rq = __task_rq_lock(p); |
1da177e4 | 9515 | |
178be793 | 9516 | normalize_task(rq, p); |
3a5e4dc1 | 9517 | |
b29739f9 | 9518 | __task_rq_unlock(rq); |
4cf5d77a | 9519 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9520 | } while_each_thread(g, p); |
9521 | ||
4cf5d77a | 9522 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9523 | } |
9524 | ||
9525 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9526 | |
9527 | #ifdef CONFIG_IA64 | |
9528 | /* | |
9529 | * These functions are only useful for the IA64 MCA handling. | |
9530 | * | |
9531 | * They can only be called when the whole system has been | |
9532 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9533 | * activity can take place. Using them for anything else would | |
9534 | * be a serious bug, and as a result, they aren't even visible | |
9535 | * under any other configuration. | |
9536 | */ | |
9537 | ||
9538 | /** | |
9539 | * curr_task - return the current task for a given cpu. | |
9540 | * @cpu: the processor in question. | |
9541 | * | |
9542 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9543 | */ | |
36c8b586 | 9544 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9545 | { |
9546 | return cpu_curr(cpu); | |
9547 | } | |
9548 | ||
9549 | /** | |
9550 | * set_curr_task - set the current task for a given cpu. | |
9551 | * @cpu: the processor in question. | |
9552 | * @p: the task pointer to set. | |
9553 | * | |
9554 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9555 | * are serviced on a separate stack. It allows the architecture to switch the |
9556 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9557 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9558 | * and caller must save the original value of the current task (see | |
9559 | * curr_task() above) and restore that value before reenabling interrupts and | |
9560 | * re-starting the system. | |
9561 | * | |
9562 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9563 | */ | |
36c8b586 | 9564 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9565 | { |
9566 | cpu_curr(cpu) = p; | |
9567 | } | |
9568 | ||
9569 | #endif | |
29f59db3 | 9570 | |
bccbe08a PZ |
9571 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9572 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9573 | { |
9574 | int i; | |
9575 | ||
9576 | for_each_possible_cpu(i) { | |
9577 | if (tg->cfs_rq) | |
9578 | kfree(tg->cfs_rq[i]); | |
9579 | if (tg->se) | |
9580 | kfree(tg->se[i]); | |
6f505b16 PZ |
9581 | } |
9582 | ||
9583 | kfree(tg->cfs_rq); | |
9584 | kfree(tg->se); | |
6f505b16 PZ |
9585 | } |
9586 | ||
ec7dc8ac DG |
9587 | static |
9588 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9589 | { |
29f59db3 | 9590 | struct cfs_rq *cfs_rq; |
eab17229 | 9591 | struct sched_entity *se; |
9b5b7751 | 9592 | struct rq *rq; |
29f59db3 SV |
9593 | int i; |
9594 | ||
434d53b0 | 9595 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9596 | if (!tg->cfs_rq) |
9597 | goto err; | |
434d53b0 | 9598 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9599 | if (!tg->se) |
9600 | goto err; | |
052f1dc7 PZ |
9601 | |
9602 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9603 | |
9604 | for_each_possible_cpu(i) { | |
9b5b7751 | 9605 | rq = cpu_rq(i); |
29f59db3 | 9606 | |
eab17229 LZ |
9607 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9608 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9609 | if (!cfs_rq) |
9610 | goto err; | |
9611 | ||
eab17229 LZ |
9612 | se = kzalloc_node(sizeof(struct sched_entity), |
9613 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9614 | if (!se) |
9615 | goto err; | |
9616 | ||
eab17229 | 9617 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9618 | } |
9619 | ||
9620 | return 1; | |
9621 | ||
9622 | err: | |
9623 | return 0; | |
9624 | } | |
9625 | ||
9626 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9627 | { | |
9628 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9629 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9630 | } | |
9631 | ||
9632 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9633 | { | |
9634 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9635 | } | |
6d6bc0ad | 9636 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9637 | static inline void free_fair_sched_group(struct task_group *tg) |
9638 | { | |
9639 | } | |
9640 | ||
ec7dc8ac DG |
9641 | static inline |
9642 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9643 | { |
9644 | return 1; | |
9645 | } | |
9646 | ||
9647 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9648 | { | |
9649 | } | |
9650 | ||
9651 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9652 | { | |
9653 | } | |
6d6bc0ad | 9654 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9655 | |
9656 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9657 | static void free_rt_sched_group(struct task_group *tg) |
9658 | { | |
9659 | int i; | |
9660 | ||
d0b27fa7 PZ |
9661 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9662 | ||
bccbe08a PZ |
9663 | for_each_possible_cpu(i) { |
9664 | if (tg->rt_rq) | |
9665 | kfree(tg->rt_rq[i]); | |
9666 | if (tg->rt_se) | |
9667 | kfree(tg->rt_se[i]); | |
9668 | } | |
9669 | ||
9670 | kfree(tg->rt_rq); | |
9671 | kfree(tg->rt_se); | |
9672 | } | |
9673 | ||
ec7dc8ac DG |
9674 | static |
9675 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9676 | { |
9677 | struct rt_rq *rt_rq; | |
eab17229 | 9678 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9679 | struct rq *rq; |
9680 | int i; | |
9681 | ||
434d53b0 | 9682 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9683 | if (!tg->rt_rq) |
9684 | goto err; | |
434d53b0 | 9685 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9686 | if (!tg->rt_se) |
9687 | goto err; | |
9688 | ||
d0b27fa7 PZ |
9689 | init_rt_bandwidth(&tg->rt_bandwidth, |
9690 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9691 | |
9692 | for_each_possible_cpu(i) { | |
9693 | rq = cpu_rq(i); | |
9694 | ||
eab17229 LZ |
9695 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9696 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9697 | if (!rt_rq) |
9698 | goto err; | |
29f59db3 | 9699 | |
eab17229 LZ |
9700 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9701 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9702 | if (!rt_se) |
9703 | goto err; | |
29f59db3 | 9704 | |
eab17229 | 9705 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9706 | } |
9707 | ||
bccbe08a PZ |
9708 | return 1; |
9709 | ||
9710 | err: | |
9711 | return 0; | |
9712 | } | |
9713 | ||
9714 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9715 | { | |
9716 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9717 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9718 | } | |
9719 | ||
9720 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9721 | { | |
9722 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9723 | } | |
6d6bc0ad | 9724 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9725 | static inline void free_rt_sched_group(struct task_group *tg) |
9726 | { | |
9727 | } | |
9728 | ||
ec7dc8ac DG |
9729 | static inline |
9730 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9731 | { |
9732 | return 1; | |
9733 | } | |
9734 | ||
9735 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9736 | { | |
9737 | } | |
9738 | ||
9739 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9740 | { | |
9741 | } | |
6d6bc0ad | 9742 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9743 | |
d0b27fa7 | 9744 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9745 | static void free_sched_group(struct task_group *tg) |
9746 | { | |
9747 | free_fair_sched_group(tg); | |
9748 | free_rt_sched_group(tg); | |
9749 | kfree(tg); | |
9750 | } | |
9751 | ||
9752 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9753 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9754 | { |
9755 | struct task_group *tg; | |
9756 | unsigned long flags; | |
9757 | int i; | |
9758 | ||
9759 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9760 | if (!tg) | |
9761 | return ERR_PTR(-ENOMEM); | |
9762 | ||
ec7dc8ac | 9763 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9764 | goto err; |
9765 | ||
ec7dc8ac | 9766 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9767 | goto err; |
9768 | ||
8ed36996 | 9769 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9770 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9771 | register_fair_sched_group(tg, i); |
9772 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9773 | } |
6f505b16 | 9774 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9775 | |
9776 | WARN_ON(!parent); /* root should already exist */ | |
9777 | ||
9778 | tg->parent = parent; | |
f473aa5e | 9779 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9780 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9781 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9782 | |
9b5b7751 | 9783 | return tg; |
29f59db3 SV |
9784 | |
9785 | err: | |
6f505b16 | 9786 | free_sched_group(tg); |
29f59db3 SV |
9787 | return ERR_PTR(-ENOMEM); |
9788 | } | |
9789 | ||
9b5b7751 | 9790 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9791 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9792 | { |
29f59db3 | 9793 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9794 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9795 | } |
9796 | ||
9b5b7751 | 9797 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9798 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9799 | { |
8ed36996 | 9800 | unsigned long flags; |
9b5b7751 | 9801 | int i; |
29f59db3 | 9802 | |
8ed36996 | 9803 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9804 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9805 | unregister_fair_sched_group(tg, i); |
9806 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9807 | } |
6f505b16 | 9808 | list_del_rcu(&tg->list); |
f473aa5e | 9809 | list_del_rcu(&tg->siblings); |
8ed36996 | 9810 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9811 | |
9b5b7751 | 9812 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9813 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9814 | } |
9815 | ||
9b5b7751 | 9816 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9817 | * The caller of this function should have put the task in its new group |
9818 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9819 | * reflect its new group. | |
9b5b7751 SV |
9820 | */ |
9821 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9822 | { |
9823 | int on_rq, running; | |
9824 | unsigned long flags; | |
9825 | struct rq *rq; | |
9826 | ||
9827 | rq = task_rq_lock(tsk, &flags); | |
9828 | ||
29f59db3 SV |
9829 | update_rq_clock(rq); |
9830 | ||
051a1d1a | 9831 | running = task_current(rq, tsk); |
29f59db3 SV |
9832 | on_rq = tsk->se.on_rq; |
9833 | ||
0e1f3483 | 9834 | if (on_rq) |
29f59db3 | 9835 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9836 | if (unlikely(running)) |
9837 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9838 | |
6f505b16 | 9839 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9840 | |
810b3817 PZ |
9841 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9842 | if (tsk->sched_class->moved_group) | |
9843 | tsk->sched_class->moved_group(tsk); | |
9844 | #endif | |
9845 | ||
0e1f3483 HS |
9846 | if (unlikely(running)) |
9847 | tsk->sched_class->set_curr_task(rq); | |
9848 | if (on_rq) | |
7074badb | 9849 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9850 | |
29f59db3 SV |
9851 | task_rq_unlock(rq, &flags); |
9852 | } | |
6d6bc0ad | 9853 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9854 | |
052f1dc7 | 9855 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9856 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9857 | { |
9858 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9859 | int on_rq; |
9860 | ||
29f59db3 | 9861 | on_rq = se->on_rq; |
62fb1851 | 9862 | if (on_rq) |
29f59db3 SV |
9863 | dequeue_entity(cfs_rq, se, 0); |
9864 | ||
9865 | se->load.weight = shares; | |
e05510d0 | 9866 | se->load.inv_weight = 0; |
29f59db3 | 9867 | |
62fb1851 | 9868 | if (on_rq) |
29f59db3 | 9869 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9870 | } |
62fb1851 | 9871 | |
c09595f6 PZ |
9872 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9873 | { | |
9874 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9875 | struct rq *rq = cfs_rq->rq; | |
9876 | unsigned long flags; | |
9877 | ||
9878 | spin_lock_irqsave(&rq->lock, flags); | |
9879 | __set_se_shares(se, shares); | |
9880 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9881 | } |
9882 | ||
8ed36996 PZ |
9883 | static DEFINE_MUTEX(shares_mutex); |
9884 | ||
4cf86d77 | 9885 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9886 | { |
9887 | int i; | |
8ed36996 | 9888 | unsigned long flags; |
c61935fd | 9889 | |
ec7dc8ac DG |
9890 | /* |
9891 | * We can't change the weight of the root cgroup. | |
9892 | */ | |
9893 | if (!tg->se[0]) | |
9894 | return -EINVAL; | |
9895 | ||
18d95a28 PZ |
9896 | if (shares < MIN_SHARES) |
9897 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9898 | else if (shares > MAX_SHARES) |
9899 | shares = MAX_SHARES; | |
62fb1851 | 9900 | |
8ed36996 | 9901 | mutex_lock(&shares_mutex); |
9b5b7751 | 9902 | if (tg->shares == shares) |
5cb350ba | 9903 | goto done; |
29f59db3 | 9904 | |
8ed36996 | 9905 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9906 | for_each_possible_cpu(i) |
9907 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9908 | list_del_rcu(&tg->siblings); |
8ed36996 | 9909 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9910 | |
9911 | /* wait for any ongoing reference to this group to finish */ | |
9912 | synchronize_sched(); | |
9913 | ||
9914 | /* | |
9915 | * Now we are free to modify the group's share on each cpu | |
9916 | * w/o tripping rebalance_share or load_balance_fair. | |
9917 | */ | |
9b5b7751 | 9918 | tg->shares = shares; |
c09595f6 PZ |
9919 | for_each_possible_cpu(i) { |
9920 | /* | |
9921 | * force a rebalance | |
9922 | */ | |
9923 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9924 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9925 | } |
29f59db3 | 9926 | |
6b2d7700 SV |
9927 | /* |
9928 | * Enable load balance activity on this group, by inserting it back on | |
9929 | * each cpu's rq->leaf_cfs_rq_list. | |
9930 | */ | |
8ed36996 | 9931 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9932 | for_each_possible_cpu(i) |
9933 | register_fair_sched_group(tg, i); | |
f473aa5e | 9934 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9935 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9936 | done: |
8ed36996 | 9937 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9938 | return 0; |
29f59db3 SV |
9939 | } |
9940 | ||
5cb350ba DG |
9941 | unsigned long sched_group_shares(struct task_group *tg) |
9942 | { | |
9943 | return tg->shares; | |
9944 | } | |
052f1dc7 | 9945 | #endif |
5cb350ba | 9946 | |
052f1dc7 | 9947 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9948 | /* |
9f0c1e56 | 9949 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9950 | */ |
9f0c1e56 PZ |
9951 | static DEFINE_MUTEX(rt_constraints_mutex); |
9952 | ||
9953 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9954 | { | |
9955 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9956 | return 1ULL << 20; |
9f0c1e56 | 9957 | |
9a7e0b18 | 9958 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9959 | } |
9960 | ||
9a7e0b18 PZ |
9961 | /* Must be called with tasklist_lock held */ |
9962 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9963 | { |
9a7e0b18 | 9964 | struct task_struct *g, *p; |
b40b2e8e | 9965 | |
9a7e0b18 PZ |
9966 | do_each_thread(g, p) { |
9967 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9968 | return 1; | |
9969 | } while_each_thread(g, p); | |
b40b2e8e | 9970 | |
9a7e0b18 PZ |
9971 | return 0; |
9972 | } | |
b40b2e8e | 9973 | |
9a7e0b18 PZ |
9974 | struct rt_schedulable_data { |
9975 | struct task_group *tg; | |
9976 | u64 rt_period; | |
9977 | u64 rt_runtime; | |
9978 | }; | |
b40b2e8e | 9979 | |
9a7e0b18 PZ |
9980 | static int tg_schedulable(struct task_group *tg, void *data) |
9981 | { | |
9982 | struct rt_schedulable_data *d = data; | |
9983 | struct task_group *child; | |
9984 | unsigned long total, sum = 0; | |
9985 | u64 period, runtime; | |
b40b2e8e | 9986 | |
9a7e0b18 PZ |
9987 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9988 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9989 | |
9a7e0b18 PZ |
9990 | if (tg == d->tg) { |
9991 | period = d->rt_period; | |
9992 | runtime = d->rt_runtime; | |
b40b2e8e | 9993 | } |
b40b2e8e | 9994 | |
98a4826b PZ |
9995 | #ifdef CONFIG_USER_SCHED |
9996 | if (tg == &root_task_group) { | |
9997 | period = global_rt_period(); | |
9998 | runtime = global_rt_runtime(); | |
9999 | } | |
10000 | #endif | |
10001 | ||
4653f803 PZ |
10002 | /* |
10003 | * Cannot have more runtime than the period. | |
10004 | */ | |
10005 | if (runtime > period && runtime != RUNTIME_INF) | |
10006 | return -EINVAL; | |
6f505b16 | 10007 | |
4653f803 PZ |
10008 | /* |
10009 | * Ensure we don't starve existing RT tasks. | |
10010 | */ | |
9a7e0b18 PZ |
10011 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
10012 | return -EBUSY; | |
6f505b16 | 10013 | |
9a7e0b18 | 10014 | total = to_ratio(period, runtime); |
6f505b16 | 10015 | |
4653f803 PZ |
10016 | /* |
10017 | * Nobody can have more than the global setting allows. | |
10018 | */ | |
10019 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
10020 | return -EINVAL; | |
6f505b16 | 10021 | |
4653f803 PZ |
10022 | /* |
10023 | * The sum of our children's runtime should not exceed our own. | |
10024 | */ | |
9a7e0b18 PZ |
10025 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
10026 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
10027 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 10028 | |
9a7e0b18 PZ |
10029 | if (child == d->tg) { |
10030 | period = d->rt_period; | |
10031 | runtime = d->rt_runtime; | |
10032 | } | |
6f505b16 | 10033 | |
9a7e0b18 | 10034 | sum += to_ratio(period, runtime); |
9f0c1e56 | 10035 | } |
6f505b16 | 10036 | |
9a7e0b18 PZ |
10037 | if (sum > total) |
10038 | return -EINVAL; | |
10039 | ||
10040 | return 0; | |
6f505b16 PZ |
10041 | } |
10042 | ||
9a7e0b18 | 10043 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 10044 | { |
9a7e0b18 PZ |
10045 | struct rt_schedulable_data data = { |
10046 | .tg = tg, | |
10047 | .rt_period = period, | |
10048 | .rt_runtime = runtime, | |
10049 | }; | |
10050 | ||
10051 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
10052 | } |
10053 | ||
d0b27fa7 PZ |
10054 | static int tg_set_bandwidth(struct task_group *tg, |
10055 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 10056 | { |
ac086bc2 | 10057 | int i, err = 0; |
9f0c1e56 | 10058 | |
9f0c1e56 | 10059 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 10060 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
10061 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10062 | if (err) | |
9f0c1e56 | 10063 | goto unlock; |
ac086bc2 PZ |
10064 | |
10065 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10066 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10067 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10068 | |
10069 | for_each_possible_cpu(i) { | |
10070 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10071 | ||
10072 | spin_lock(&rt_rq->rt_runtime_lock); | |
10073 | rt_rq->rt_runtime = rt_runtime; | |
10074 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10075 | } | |
10076 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10077 | unlock: |
521f1a24 | 10078 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10079 | mutex_unlock(&rt_constraints_mutex); |
10080 | ||
10081 | return err; | |
6f505b16 PZ |
10082 | } |
10083 | ||
d0b27fa7 PZ |
10084 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10085 | { | |
10086 | u64 rt_runtime, rt_period; | |
10087 | ||
10088 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10089 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10090 | if (rt_runtime_us < 0) | |
10091 | rt_runtime = RUNTIME_INF; | |
10092 | ||
10093 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10094 | } | |
10095 | ||
9f0c1e56 PZ |
10096 | long sched_group_rt_runtime(struct task_group *tg) |
10097 | { | |
10098 | u64 rt_runtime_us; | |
10099 | ||
d0b27fa7 | 10100 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10101 | return -1; |
10102 | ||
d0b27fa7 | 10103 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10104 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10105 | return rt_runtime_us; | |
10106 | } | |
d0b27fa7 PZ |
10107 | |
10108 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10109 | { | |
10110 | u64 rt_runtime, rt_period; | |
10111 | ||
10112 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10113 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10114 | ||
619b0488 R |
10115 | if (rt_period == 0) |
10116 | return -EINVAL; | |
10117 | ||
d0b27fa7 PZ |
10118 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10119 | } | |
10120 | ||
10121 | long sched_group_rt_period(struct task_group *tg) | |
10122 | { | |
10123 | u64 rt_period_us; | |
10124 | ||
10125 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10126 | do_div(rt_period_us, NSEC_PER_USEC); | |
10127 | return rt_period_us; | |
10128 | } | |
10129 | ||
10130 | static int sched_rt_global_constraints(void) | |
10131 | { | |
4653f803 | 10132 | u64 runtime, period; |
d0b27fa7 PZ |
10133 | int ret = 0; |
10134 | ||
ec5d4989 HS |
10135 | if (sysctl_sched_rt_period <= 0) |
10136 | return -EINVAL; | |
10137 | ||
4653f803 PZ |
10138 | runtime = global_rt_runtime(); |
10139 | period = global_rt_period(); | |
10140 | ||
10141 | /* | |
10142 | * Sanity check on the sysctl variables. | |
10143 | */ | |
10144 | if (runtime > period && runtime != RUNTIME_INF) | |
10145 | return -EINVAL; | |
10b612f4 | 10146 | |
d0b27fa7 | 10147 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10148 | read_lock(&tasklist_lock); |
4653f803 | 10149 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10150 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10151 | mutex_unlock(&rt_constraints_mutex); |
10152 | ||
10153 | return ret; | |
10154 | } | |
54e99124 DG |
10155 | |
10156 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10157 | { | |
10158 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10159 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10160 | return 0; | |
10161 | ||
10162 | return 1; | |
10163 | } | |
10164 | ||
6d6bc0ad | 10165 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10166 | static int sched_rt_global_constraints(void) |
10167 | { | |
ac086bc2 PZ |
10168 | unsigned long flags; |
10169 | int i; | |
10170 | ||
ec5d4989 HS |
10171 | if (sysctl_sched_rt_period <= 0) |
10172 | return -EINVAL; | |
10173 | ||
60aa605d PZ |
10174 | /* |
10175 | * There's always some RT tasks in the root group | |
10176 | * -- migration, kstopmachine etc.. | |
10177 | */ | |
10178 | if (sysctl_sched_rt_runtime == 0) | |
10179 | return -EBUSY; | |
10180 | ||
ac086bc2 PZ |
10181 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10182 | for_each_possible_cpu(i) { | |
10183 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10184 | ||
10185 | spin_lock(&rt_rq->rt_runtime_lock); | |
10186 | rt_rq->rt_runtime = global_rt_runtime(); | |
10187 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10188 | } | |
10189 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10190 | ||
d0b27fa7 PZ |
10191 | return 0; |
10192 | } | |
6d6bc0ad | 10193 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10194 | |
10195 | int sched_rt_handler(struct ctl_table *table, int write, | |
10196 | struct file *filp, void __user *buffer, size_t *lenp, | |
10197 | loff_t *ppos) | |
10198 | { | |
10199 | int ret; | |
10200 | int old_period, old_runtime; | |
10201 | static DEFINE_MUTEX(mutex); | |
10202 | ||
10203 | mutex_lock(&mutex); | |
10204 | old_period = sysctl_sched_rt_period; | |
10205 | old_runtime = sysctl_sched_rt_runtime; | |
10206 | ||
10207 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10208 | ||
10209 | if (!ret && write) { | |
10210 | ret = sched_rt_global_constraints(); | |
10211 | if (ret) { | |
10212 | sysctl_sched_rt_period = old_period; | |
10213 | sysctl_sched_rt_runtime = old_runtime; | |
10214 | } else { | |
10215 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10216 | def_rt_bandwidth.rt_period = | |
10217 | ns_to_ktime(global_rt_period()); | |
10218 | } | |
10219 | } | |
10220 | mutex_unlock(&mutex); | |
10221 | ||
10222 | return ret; | |
10223 | } | |
68318b8e | 10224 | |
052f1dc7 | 10225 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10226 | |
10227 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10228 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10229 | { |
2b01dfe3 PM |
10230 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10231 | struct task_group, css); | |
68318b8e SV |
10232 | } |
10233 | ||
10234 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10235 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10236 | { |
ec7dc8ac | 10237 | struct task_group *tg, *parent; |
68318b8e | 10238 | |
2b01dfe3 | 10239 | if (!cgrp->parent) { |
68318b8e | 10240 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10241 | return &init_task_group.css; |
10242 | } | |
10243 | ||
ec7dc8ac DG |
10244 | parent = cgroup_tg(cgrp->parent); |
10245 | tg = sched_create_group(parent); | |
68318b8e SV |
10246 | if (IS_ERR(tg)) |
10247 | return ERR_PTR(-ENOMEM); | |
10248 | ||
68318b8e SV |
10249 | return &tg->css; |
10250 | } | |
10251 | ||
41a2d6cf IM |
10252 | static void |
10253 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10254 | { |
2b01dfe3 | 10255 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10256 | |
10257 | sched_destroy_group(tg); | |
10258 | } | |
10259 | ||
41a2d6cf IM |
10260 | static int |
10261 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10262 | struct task_struct *tsk) | |
68318b8e | 10263 | { |
b68aa230 | 10264 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10265 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10266 | return -EINVAL; |
10267 | #else | |
68318b8e SV |
10268 | /* We don't support RT-tasks being in separate groups */ |
10269 | if (tsk->sched_class != &fair_sched_class) | |
10270 | return -EINVAL; | |
b68aa230 | 10271 | #endif |
68318b8e SV |
10272 | |
10273 | return 0; | |
10274 | } | |
10275 | ||
10276 | static void | |
2b01dfe3 | 10277 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10278 | struct cgroup *old_cont, struct task_struct *tsk) |
10279 | { | |
10280 | sched_move_task(tsk); | |
10281 | } | |
10282 | ||
052f1dc7 | 10283 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10284 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10285 | u64 shareval) |
68318b8e | 10286 | { |
2b01dfe3 | 10287 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10288 | } |
10289 | ||
f4c753b7 | 10290 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10291 | { |
2b01dfe3 | 10292 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10293 | |
10294 | return (u64) tg->shares; | |
10295 | } | |
6d6bc0ad | 10296 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10297 | |
052f1dc7 | 10298 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10299 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10300 | s64 val) |
6f505b16 | 10301 | { |
06ecb27c | 10302 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10303 | } |
10304 | ||
06ecb27c | 10305 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10306 | { |
06ecb27c | 10307 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10308 | } |
d0b27fa7 PZ |
10309 | |
10310 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10311 | u64 rt_period_us) | |
10312 | { | |
10313 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10314 | } | |
10315 | ||
10316 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10317 | { | |
10318 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10319 | } | |
6d6bc0ad | 10320 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10321 | |
fe5c7cc2 | 10322 | static struct cftype cpu_files[] = { |
052f1dc7 | 10323 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10324 | { |
10325 | .name = "shares", | |
f4c753b7 PM |
10326 | .read_u64 = cpu_shares_read_u64, |
10327 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10328 | }, |
052f1dc7 PZ |
10329 | #endif |
10330 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10331 | { |
9f0c1e56 | 10332 | .name = "rt_runtime_us", |
06ecb27c PM |
10333 | .read_s64 = cpu_rt_runtime_read, |
10334 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10335 | }, |
d0b27fa7 PZ |
10336 | { |
10337 | .name = "rt_period_us", | |
f4c753b7 PM |
10338 | .read_u64 = cpu_rt_period_read_uint, |
10339 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10340 | }, |
052f1dc7 | 10341 | #endif |
68318b8e SV |
10342 | }; |
10343 | ||
10344 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10345 | { | |
fe5c7cc2 | 10346 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10347 | } |
10348 | ||
10349 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10350 | .name = "cpu", |
10351 | .create = cpu_cgroup_create, | |
10352 | .destroy = cpu_cgroup_destroy, | |
10353 | .can_attach = cpu_cgroup_can_attach, | |
10354 | .attach = cpu_cgroup_attach, | |
10355 | .populate = cpu_cgroup_populate, | |
10356 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10357 | .early_init = 1, |
10358 | }; | |
10359 | ||
052f1dc7 | 10360 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10361 | |
10362 | #ifdef CONFIG_CGROUP_CPUACCT | |
10363 | ||
10364 | /* | |
10365 | * CPU accounting code for task groups. | |
10366 | * | |
10367 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10368 | * (balbir@in.ibm.com). | |
10369 | */ | |
10370 | ||
934352f2 | 10371 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10372 | struct cpuacct { |
10373 | struct cgroup_subsys_state css; | |
10374 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10375 | u64 *cpuusage; | |
ef12fefa | 10376 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10377 | struct cpuacct *parent; |
d842de87 SV |
10378 | }; |
10379 | ||
10380 | struct cgroup_subsys cpuacct_subsys; | |
10381 | ||
10382 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10383 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10384 | { |
32cd756a | 10385 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10386 | struct cpuacct, css); |
10387 | } | |
10388 | ||
10389 | /* return cpu accounting group to which this task belongs */ | |
10390 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10391 | { | |
10392 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10393 | struct cpuacct, css); | |
10394 | } | |
10395 | ||
10396 | /* create a new cpu accounting group */ | |
10397 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10398 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10399 | { |
10400 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10401 | int i; |
d842de87 SV |
10402 | |
10403 | if (!ca) | |
ef12fefa | 10404 | goto out; |
d842de87 SV |
10405 | |
10406 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10407 | if (!ca->cpuusage) |
10408 | goto out_free_ca; | |
10409 | ||
10410 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10411 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10412 | goto out_free_counters; | |
d842de87 | 10413 | |
934352f2 BR |
10414 | if (cgrp->parent) |
10415 | ca->parent = cgroup_ca(cgrp->parent); | |
10416 | ||
d842de87 | 10417 | return &ca->css; |
ef12fefa BR |
10418 | |
10419 | out_free_counters: | |
10420 | while (--i >= 0) | |
10421 | percpu_counter_destroy(&ca->cpustat[i]); | |
10422 | free_percpu(ca->cpuusage); | |
10423 | out_free_ca: | |
10424 | kfree(ca); | |
10425 | out: | |
10426 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10427 | } |
10428 | ||
10429 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10430 | static void |
32cd756a | 10431 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10432 | { |
32cd756a | 10433 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10434 | int i; |
d842de87 | 10435 | |
ef12fefa BR |
10436 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10437 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10438 | free_percpu(ca->cpuusage); |
10439 | kfree(ca); | |
10440 | } | |
10441 | ||
720f5498 KC |
10442 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10443 | { | |
b36128c8 | 10444 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10445 | u64 data; |
10446 | ||
10447 | #ifndef CONFIG_64BIT | |
10448 | /* | |
10449 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10450 | */ | |
10451 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10452 | data = *cpuusage; | |
10453 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10454 | #else | |
10455 | data = *cpuusage; | |
10456 | #endif | |
10457 | ||
10458 | return data; | |
10459 | } | |
10460 | ||
10461 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10462 | { | |
b36128c8 | 10463 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10464 | |
10465 | #ifndef CONFIG_64BIT | |
10466 | /* | |
10467 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10468 | */ | |
10469 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10470 | *cpuusage = val; | |
10471 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10472 | #else | |
10473 | *cpuusage = val; | |
10474 | #endif | |
10475 | } | |
10476 | ||
d842de87 | 10477 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10478 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10479 | { |
32cd756a | 10480 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10481 | u64 totalcpuusage = 0; |
10482 | int i; | |
10483 | ||
720f5498 KC |
10484 | for_each_present_cpu(i) |
10485 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10486 | |
10487 | return totalcpuusage; | |
10488 | } | |
10489 | ||
0297b803 DG |
10490 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10491 | u64 reset) | |
10492 | { | |
10493 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10494 | int err = 0; | |
10495 | int i; | |
10496 | ||
10497 | if (reset) { | |
10498 | err = -EINVAL; | |
10499 | goto out; | |
10500 | } | |
10501 | ||
720f5498 KC |
10502 | for_each_present_cpu(i) |
10503 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10504 | |
0297b803 DG |
10505 | out: |
10506 | return err; | |
10507 | } | |
10508 | ||
e9515c3c KC |
10509 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10510 | struct seq_file *m) | |
10511 | { | |
10512 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10513 | u64 percpu; | |
10514 | int i; | |
10515 | ||
10516 | for_each_present_cpu(i) { | |
10517 | percpu = cpuacct_cpuusage_read(ca, i); | |
10518 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10519 | } | |
10520 | seq_printf(m, "\n"); | |
10521 | return 0; | |
10522 | } | |
10523 | ||
ef12fefa BR |
10524 | static const char *cpuacct_stat_desc[] = { |
10525 | [CPUACCT_STAT_USER] = "user", | |
10526 | [CPUACCT_STAT_SYSTEM] = "system", | |
10527 | }; | |
10528 | ||
10529 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10530 | struct cgroup_map_cb *cb) | |
10531 | { | |
10532 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10533 | int i; | |
10534 | ||
10535 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10536 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10537 | val = cputime64_to_clock_t(val); | |
10538 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10539 | } | |
10540 | return 0; | |
10541 | } | |
10542 | ||
d842de87 SV |
10543 | static struct cftype files[] = { |
10544 | { | |
10545 | .name = "usage", | |
f4c753b7 PM |
10546 | .read_u64 = cpuusage_read, |
10547 | .write_u64 = cpuusage_write, | |
d842de87 | 10548 | }, |
e9515c3c KC |
10549 | { |
10550 | .name = "usage_percpu", | |
10551 | .read_seq_string = cpuacct_percpu_seq_read, | |
10552 | }, | |
ef12fefa BR |
10553 | { |
10554 | .name = "stat", | |
10555 | .read_map = cpuacct_stats_show, | |
10556 | }, | |
d842de87 SV |
10557 | }; |
10558 | ||
32cd756a | 10559 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10560 | { |
32cd756a | 10561 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10562 | } |
10563 | ||
10564 | /* | |
10565 | * charge this task's execution time to its accounting group. | |
10566 | * | |
10567 | * called with rq->lock held. | |
10568 | */ | |
10569 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10570 | { | |
10571 | struct cpuacct *ca; | |
934352f2 | 10572 | int cpu; |
d842de87 | 10573 | |
c40c6f85 | 10574 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10575 | return; |
10576 | ||
934352f2 | 10577 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10578 | |
10579 | rcu_read_lock(); | |
10580 | ||
d842de87 | 10581 | ca = task_ca(tsk); |
d842de87 | 10582 | |
934352f2 | 10583 | for (; ca; ca = ca->parent) { |
b36128c8 | 10584 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10585 | *cpuusage += cputime; |
10586 | } | |
a18b83b7 BR |
10587 | |
10588 | rcu_read_unlock(); | |
d842de87 SV |
10589 | } |
10590 | ||
ef12fefa BR |
10591 | /* |
10592 | * Charge the system/user time to the task's accounting group. | |
10593 | */ | |
10594 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10595 | enum cpuacct_stat_index idx, cputime_t val) | |
10596 | { | |
10597 | struct cpuacct *ca; | |
10598 | ||
10599 | if (unlikely(!cpuacct_subsys.active)) | |
10600 | return; | |
10601 | ||
10602 | rcu_read_lock(); | |
10603 | ca = task_ca(tsk); | |
10604 | ||
10605 | do { | |
10606 | percpu_counter_add(&ca->cpustat[idx], val); | |
10607 | ca = ca->parent; | |
10608 | } while (ca); | |
10609 | rcu_read_unlock(); | |
10610 | } | |
10611 | ||
d842de87 SV |
10612 | struct cgroup_subsys cpuacct_subsys = { |
10613 | .name = "cpuacct", | |
10614 | .create = cpuacct_create, | |
10615 | .destroy = cpuacct_destroy, | |
10616 | .populate = cpuacct_populate, | |
10617 | .subsys_id = cpuacct_subsys_id, | |
10618 | }; | |
10619 | #endif /* CONFIG_CGROUP_CPUACCT */ |