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> |
434d53b0 | 72 | #include <linux/bootmem.h> |
f00b45c1 PZ |
73 | #include <linux/debugfs.h> |
74 | #include <linux/ctype.h> | |
6cd8a4bb | 75 | #include <linux/ftrace.h> |
0a16b607 | 76 | #include <trace/sched.h> |
1da177e4 | 77 | |
5517d86b | 78 | #include <asm/tlb.h> |
838225b4 | 79 | #include <asm/irq_regs.h> |
1da177e4 | 80 | |
6e0534f2 GH |
81 | #include "sched_cpupri.h" |
82 | ||
1da177e4 LT |
83 | /* |
84 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
85 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
86 | * and back. | |
87 | */ | |
88 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
89 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
90 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
91 | ||
92 | /* | |
93 | * 'User priority' is the nice value converted to something we | |
94 | * can work with better when scaling various scheduler parameters, | |
95 | * it's a [ 0 ... 39 ] range. | |
96 | */ | |
97 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
98 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
99 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
100 | ||
101 | /* | |
d7876a08 | 102 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 103 | */ |
d6322faf | 104 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 105 | |
6aa645ea IM |
106 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
107 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
108 | ||
1da177e4 LT |
109 | /* |
110 | * These are the 'tuning knobs' of the scheduler: | |
111 | * | |
a4ec24b4 | 112 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
113 | * Timeslices get refilled after they expire. |
114 | */ | |
1da177e4 | 115 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 116 | |
d0b27fa7 PZ |
117 | /* |
118 | * single value that denotes runtime == period, ie unlimited time. | |
119 | */ | |
120 | #define RUNTIME_INF ((u64)~0ULL) | |
121 | ||
7e066fb8 MD |
122 | DEFINE_TRACE(sched_wait_task); |
123 | DEFINE_TRACE(sched_wakeup); | |
124 | DEFINE_TRACE(sched_wakeup_new); | |
125 | DEFINE_TRACE(sched_switch); | |
126 | DEFINE_TRACE(sched_migrate_task); | |
127 | ||
5517d86b | 128 | #ifdef CONFIG_SMP |
fd2ab30b SN |
129 | |
130 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
131 | ||
5517d86b ED |
132 | /* |
133 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
134 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
135 | */ | |
136 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
137 | { | |
138 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
139 | } | |
140 | ||
141 | /* | |
142 | * Each time a sched group cpu_power is changed, | |
143 | * we must compute its reciprocal value | |
144 | */ | |
145 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
146 | { | |
147 | sg->__cpu_power += val; | |
148 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
149 | } | |
150 | #endif | |
151 | ||
e05606d3 IM |
152 | static inline int rt_policy(int policy) |
153 | { | |
3f33a7ce | 154 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
155 | return 1; |
156 | return 0; | |
157 | } | |
158 | ||
159 | static inline int task_has_rt_policy(struct task_struct *p) | |
160 | { | |
161 | return rt_policy(p->policy); | |
162 | } | |
163 | ||
1da177e4 | 164 | /* |
6aa645ea | 165 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 166 | */ |
6aa645ea IM |
167 | struct rt_prio_array { |
168 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
169 | struct list_head queue[MAX_RT_PRIO]; | |
170 | }; | |
171 | ||
d0b27fa7 | 172 | struct rt_bandwidth { |
ea736ed5 IM |
173 | /* nests inside the rq lock: */ |
174 | spinlock_t rt_runtime_lock; | |
175 | ktime_t rt_period; | |
176 | u64 rt_runtime; | |
177 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
178 | }; |
179 | ||
180 | static struct rt_bandwidth def_rt_bandwidth; | |
181 | ||
182 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
183 | ||
184 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
185 | { | |
186 | struct rt_bandwidth *rt_b = | |
187 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
188 | ktime_t now; | |
189 | int overrun; | |
190 | int idle = 0; | |
191 | ||
192 | for (;;) { | |
193 | now = hrtimer_cb_get_time(timer); | |
194 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
195 | ||
196 | if (!overrun) | |
197 | break; | |
198 | ||
199 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
200 | } | |
201 | ||
202 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
203 | } | |
204 | ||
205 | static | |
206 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
207 | { | |
208 | rt_b->rt_period = ns_to_ktime(period); | |
209 | rt_b->rt_runtime = runtime; | |
210 | ||
ac086bc2 PZ |
211 | spin_lock_init(&rt_b->rt_runtime_lock); |
212 | ||
d0b27fa7 PZ |
213 | hrtimer_init(&rt_b->rt_period_timer, |
214 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
215 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
216 | } |
217 | ||
c8bfff6d KH |
218 | static inline int rt_bandwidth_enabled(void) |
219 | { | |
220 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
221 | } |
222 | ||
223 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
224 | { | |
225 | ktime_t now; | |
226 | ||
cac64d00 | 227 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
228 | return; |
229 | ||
230 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
231 | return; | |
232 | ||
233 | spin_lock(&rt_b->rt_runtime_lock); | |
234 | for (;;) { | |
7f1e2ca9 PZ |
235 | unsigned long delta; |
236 | ktime_t soft, hard; | |
237 | ||
d0b27fa7 PZ |
238 | if (hrtimer_active(&rt_b->rt_period_timer)) |
239 | break; | |
240 | ||
241 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
242 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
243 | |
244 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
245 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
246 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
247 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
248 | HRTIMER_MODE_ABS, 0); | |
d0b27fa7 PZ |
249 | } |
250 | spin_unlock(&rt_b->rt_runtime_lock); | |
251 | } | |
252 | ||
253 | #ifdef CONFIG_RT_GROUP_SCHED | |
254 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
255 | { | |
256 | hrtimer_cancel(&rt_b->rt_period_timer); | |
257 | } | |
258 | #endif | |
259 | ||
712555ee HC |
260 | /* |
261 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
262 | * detach_destroy_domains and partition_sched_domains. | |
263 | */ | |
264 | static DEFINE_MUTEX(sched_domains_mutex); | |
265 | ||
052f1dc7 | 266 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 267 | |
68318b8e SV |
268 | #include <linux/cgroup.h> |
269 | ||
29f59db3 SV |
270 | struct cfs_rq; |
271 | ||
6f505b16 PZ |
272 | static LIST_HEAD(task_groups); |
273 | ||
29f59db3 | 274 | /* task group related information */ |
4cf86d77 | 275 | struct task_group { |
052f1dc7 | 276 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
277 | struct cgroup_subsys_state css; |
278 | #endif | |
052f1dc7 | 279 | |
6c415b92 AB |
280 | #ifdef CONFIG_USER_SCHED |
281 | uid_t uid; | |
282 | #endif | |
283 | ||
052f1dc7 | 284 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
285 | /* schedulable entities of this group on each cpu */ |
286 | struct sched_entity **se; | |
287 | /* runqueue "owned" by this group on each cpu */ | |
288 | struct cfs_rq **cfs_rq; | |
289 | unsigned long shares; | |
052f1dc7 PZ |
290 | #endif |
291 | ||
292 | #ifdef CONFIG_RT_GROUP_SCHED | |
293 | struct sched_rt_entity **rt_se; | |
294 | struct rt_rq **rt_rq; | |
295 | ||
d0b27fa7 | 296 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 297 | #endif |
6b2d7700 | 298 | |
ae8393e5 | 299 | struct rcu_head rcu; |
6f505b16 | 300 | struct list_head list; |
f473aa5e PZ |
301 | |
302 | struct task_group *parent; | |
303 | struct list_head siblings; | |
304 | struct list_head children; | |
29f59db3 SV |
305 | }; |
306 | ||
354d60c2 | 307 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 308 | |
6c415b92 AB |
309 | /* Helper function to pass uid information to create_sched_user() */ |
310 | void set_tg_uid(struct user_struct *user) | |
311 | { | |
312 | user->tg->uid = user->uid; | |
313 | } | |
314 | ||
eff766a6 PZ |
315 | /* |
316 | * Root task group. | |
317 | * Every UID task group (including init_task_group aka UID-0) will | |
318 | * be a child to this group. | |
319 | */ | |
320 | struct task_group root_task_group; | |
321 | ||
052f1dc7 | 322 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
323 | /* Default task group's sched entity on each cpu */ |
324 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
325 | /* Default task group's cfs_rq on each cpu */ | |
326 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 327 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
328 | |
329 | #ifdef CONFIG_RT_GROUP_SCHED | |
330 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
331 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 332 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 333 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 334 | #define root_task_group init_task_group |
9a7e0b18 | 335 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 336 | |
8ed36996 | 337 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
338 | * a task group's cpu shares. |
339 | */ | |
8ed36996 | 340 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 341 | |
57310a98 PZ |
342 | #ifdef CONFIG_SMP |
343 | static int root_task_group_empty(void) | |
344 | { | |
345 | return list_empty(&root_task_group.children); | |
346 | } | |
347 | #endif | |
348 | ||
052f1dc7 | 349 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
350 | #ifdef CONFIG_USER_SCHED |
351 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 352 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 353 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 354 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 355 | |
cb4ad1ff | 356 | /* |
2e084786 LJ |
357 | * A weight of 0 or 1 can cause arithmetics problems. |
358 | * A weight of a cfs_rq is the sum of weights of which entities | |
359 | * are queued on this cfs_rq, so a weight of a entity should not be | |
360 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
361 | * (The default weight is 1024 - so there's no practical |
362 | * limitation from this.) | |
363 | */ | |
18d95a28 | 364 | #define MIN_SHARES 2 |
2e084786 | 365 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 366 | |
052f1dc7 PZ |
367 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
368 | #endif | |
369 | ||
29f59db3 | 370 | /* Default task group. |
3a252015 | 371 | * Every task in system belong to this group at bootup. |
29f59db3 | 372 | */ |
434d53b0 | 373 | struct task_group init_task_group; |
29f59db3 SV |
374 | |
375 | /* return group to which a task belongs */ | |
4cf86d77 | 376 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 377 | { |
4cf86d77 | 378 | struct task_group *tg; |
9b5b7751 | 379 | |
052f1dc7 | 380 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
381 | rcu_read_lock(); |
382 | tg = __task_cred(p)->user->tg; | |
383 | rcu_read_unlock(); | |
052f1dc7 | 384 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
385 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
386 | struct task_group, css); | |
24e377a8 | 387 | #else |
41a2d6cf | 388 | tg = &init_task_group; |
24e377a8 | 389 | #endif |
9b5b7751 | 390 | return tg; |
29f59db3 SV |
391 | } |
392 | ||
393 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 394 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 395 | { |
052f1dc7 | 396 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
397 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
398 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 399 | #endif |
6f505b16 | 400 | |
052f1dc7 | 401 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
402 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
403 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 404 | #endif |
29f59db3 SV |
405 | } |
406 | ||
407 | #else | |
408 | ||
57310a98 PZ |
409 | #ifdef CONFIG_SMP |
410 | static int root_task_group_empty(void) | |
411 | { | |
412 | return 1; | |
413 | } | |
414 | #endif | |
415 | ||
6f505b16 | 416 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
417 | static inline struct task_group *task_group(struct task_struct *p) |
418 | { | |
419 | return NULL; | |
420 | } | |
29f59db3 | 421 | |
052f1dc7 | 422 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 423 | |
6aa645ea IM |
424 | /* CFS-related fields in a runqueue */ |
425 | struct cfs_rq { | |
426 | struct load_weight load; | |
427 | unsigned long nr_running; | |
428 | ||
6aa645ea | 429 | u64 exec_clock; |
e9acbff6 | 430 | u64 min_vruntime; |
6aa645ea IM |
431 | |
432 | struct rb_root tasks_timeline; | |
433 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
434 | |
435 | struct list_head tasks; | |
436 | struct list_head *balance_iterator; | |
437 | ||
438 | /* | |
439 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
440 | * It is set to NULL otherwise (i.e when none are currently running). |
441 | */ | |
4793241b | 442 | struct sched_entity *curr, *next, *last; |
ddc97297 | 443 | |
5ac5c4d6 | 444 | unsigned int nr_spread_over; |
ddc97297 | 445 | |
62160e3f | 446 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
447 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
448 | ||
41a2d6cf IM |
449 | /* |
450 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
451 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
452 | * (like users, containers etc.) | |
453 | * | |
454 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
455 | * list is used during load balance. | |
456 | */ | |
41a2d6cf IM |
457 | struct list_head leaf_cfs_rq_list; |
458 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
459 | |
460 | #ifdef CONFIG_SMP | |
c09595f6 | 461 | /* |
c8cba857 | 462 | * the part of load.weight contributed by tasks |
c09595f6 | 463 | */ |
c8cba857 | 464 | unsigned long task_weight; |
c09595f6 | 465 | |
c8cba857 PZ |
466 | /* |
467 | * h_load = weight * f(tg) | |
468 | * | |
469 | * Where f(tg) is the recursive weight fraction assigned to | |
470 | * this group. | |
471 | */ | |
472 | unsigned long h_load; | |
c09595f6 | 473 | |
c8cba857 PZ |
474 | /* |
475 | * this cpu's part of tg->shares | |
476 | */ | |
477 | unsigned long shares; | |
f1d239f7 PZ |
478 | |
479 | /* | |
480 | * load.weight at the time we set shares | |
481 | */ | |
482 | unsigned long rq_weight; | |
c09595f6 | 483 | #endif |
6aa645ea IM |
484 | #endif |
485 | }; | |
1da177e4 | 486 | |
6aa645ea IM |
487 | /* Real-Time classes' related field in a runqueue: */ |
488 | struct rt_rq { | |
489 | struct rt_prio_array active; | |
63489e45 | 490 | unsigned long rt_nr_running; |
052f1dc7 | 491 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
492 | struct { |
493 | int curr; /* highest queued rt task prio */ | |
398a153b | 494 | #ifdef CONFIG_SMP |
e864c499 | 495 | int next; /* next highest */ |
398a153b | 496 | #endif |
e864c499 | 497 | } highest_prio; |
6f505b16 | 498 | #endif |
fa85ae24 | 499 | #ifdef CONFIG_SMP |
73fe6aae | 500 | unsigned long rt_nr_migratory; |
a22d7fc1 | 501 | int overloaded; |
917b627d | 502 | struct plist_head pushable_tasks; |
fa85ae24 | 503 | #endif |
6f505b16 | 504 | int rt_throttled; |
fa85ae24 | 505 | u64 rt_time; |
ac086bc2 | 506 | u64 rt_runtime; |
ea736ed5 | 507 | /* Nests inside the rq lock: */ |
ac086bc2 | 508 | spinlock_t rt_runtime_lock; |
6f505b16 | 509 | |
052f1dc7 | 510 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
511 | unsigned long rt_nr_boosted; |
512 | ||
6f505b16 PZ |
513 | struct rq *rq; |
514 | struct list_head leaf_rt_rq_list; | |
515 | struct task_group *tg; | |
516 | struct sched_rt_entity *rt_se; | |
517 | #endif | |
6aa645ea IM |
518 | }; |
519 | ||
57d885fe GH |
520 | #ifdef CONFIG_SMP |
521 | ||
522 | /* | |
523 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
524 | * variables. Each exclusive cpuset essentially defines an island domain by |
525 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
526 | * exclusive cpuset is created, we also create and attach a new root-domain |
527 | * object. | |
528 | * | |
57d885fe GH |
529 | */ |
530 | struct root_domain { | |
531 | atomic_t refcount; | |
c6c4927b RR |
532 | cpumask_var_t span; |
533 | cpumask_var_t online; | |
637f5085 | 534 | |
0eab9146 | 535 | /* |
637f5085 GH |
536 | * The "RT overload" flag: it gets set if a CPU has more than |
537 | * one runnable RT task. | |
538 | */ | |
c6c4927b | 539 | cpumask_var_t rto_mask; |
0eab9146 | 540 | atomic_t rto_count; |
6e0534f2 GH |
541 | #ifdef CONFIG_SMP |
542 | struct cpupri cpupri; | |
543 | #endif | |
7a09b1a2 VS |
544 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
545 | /* | |
546 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
547 | * used when most cpus are idle in the system indicating overall very | |
548 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
549 | */ | |
550 | unsigned int sched_mc_preferred_wakeup_cpu; | |
551 | #endif | |
57d885fe GH |
552 | }; |
553 | ||
dc938520 GH |
554 | /* |
555 | * By default the system creates a single root-domain with all cpus as | |
556 | * members (mimicking the global state we have today). | |
557 | */ | |
57d885fe GH |
558 | static struct root_domain def_root_domain; |
559 | ||
560 | #endif | |
561 | ||
1da177e4 LT |
562 | /* |
563 | * This is the main, per-CPU runqueue data structure. | |
564 | * | |
565 | * Locking rule: those places that want to lock multiple runqueues | |
566 | * (such as the load balancing or the thread migration code), lock | |
567 | * acquire operations must be ordered by ascending &runqueue. | |
568 | */ | |
70b97a7f | 569 | struct rq { |
d8016491 IM |
570 | /* runqueue lock: */ |
571 | spinlock_t lock; | |
1da177e4 LT |
572 | |
573 | /* | |
574 | * nr_running and cpu_load should be in the same cacheline because | |
575 | * remote CPUs use both these fields when doing load calculation. | |
576 | */ | |
577 | unsigned long nr_running; | |
6aa645ea IM |
578 | #define CPU_LOAD_IDX_MAX 5 |
579 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 580 | #ifdef CONFIG_NO_HZ |
15934a37 | 581 | unsigned long last_tick_seen; |
46cb4b7c SS |
582 | unsigned char in_nohz_recently; |
583 | #endif | |
d8016491 IM |
584 | /* capture load from *all* tasks on this cpu: */ |
585 | struct load_weight load; | |
6aa645ea IM |
586 | unsigned long nr_load_updates; |
587 | u64 nr_switches; | |
23a185ca | 588 | u64 nr_migrations_in; |
6aa645ea IM |
589 | |
590 | struct cfs_rq cfs; | |
6f505b16 | 591 | struct rt_rq rt; |
6f505b16 | 592 | |
6aa645ea | 593 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
594 | /* list of leaf cfs_rq on this cpu: */ |
595 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
596 | #endif |
597 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 598 | struct list_head leaf_rt_rq_list; |
1da177e4 | 599 | #endif |
1da177e4 LT |
600 | |
601 | /* | |
602 | * This is part of a global counter where only the total sum | |
603 | * over all CPUs matters. A task can increase this counter on | |
604 | * one CPU and if it got migrated afterwards it may decrease | |
605 | * it on another CPU. Always updated under the runqueue lock: | |
606 | */ | |
607 | unsigned long nr_uninterruptible; | |
608 | ||
36c8b586 | 609 | struct task_struct *curr, *idle; |
c9819f45 | 610 | unsigned long next_balance; |
1da177e4 | 611 | struct mm_struct *prev_mm; |
6aa645ea | 612 | |
3e51f33f | 613 | u64 clock; |
6aa645ea | 614 | |
1da177e4 LT |
615 | atomic_t nr_iowait; |
616 | ||
617 | #ifdef CONFIG_SMP | |
0eab9146 | 618 | struct root_domain *rd; |
1da177e4 LT |
619 | struct sched_domain *sd; |
620 | ||
a0a522ce | 621 | unsigned char idle_at_tick; |
1da177e4 LT |
622 | /* For active balancing */ |
623 | int active_balance; | |
624 | int push_cpu; | |
d8016491 IM |
625 | /* cpu of this runqueue: */ |
626 | int cpu; | |
1f11eb6a | 627 | int online; |
1da177e4 | 628 | |
a8a51d5e | 629 | unsigned long avg_load_per_task; |
1da177e4 | 630 | |
36c8b586 | 631 | struct task_struct *migration_thread; |
1da177e4 LT |
632 | struct list_head migration_queue; |
633 | #endif | |
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) | |
696 | ||
aa9c4c0f | 697 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
698 | { |
699 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
700 | } | |
701 | ||
bf5c91ba IM |
702 | /* |
703 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
704 | */ | |
705 | #ifdef CONFIG_SCHED_DEBUG | |
706 | # define const_debug __read_mostly | |
707 | #else | |
708 | # define const_debug static const | |
709 | #endif | |
710 | ||
017730c1 IM |
711 | /** |
712 | * runqueue_is_locked | |
713 | * | |
714 | * Returns true if the current cpu runqueue is locked. | |
715 | * This interface allows printk to be called with the runqueue lock | |
716 | * held and know whether or not it is OK to wake up the klogd. | |
717 | */ | |
718 | int runqueue_is_locked(void) | |
719 | { | |
720 | int cpu = get_cpu(); | |
721 | struct rq *rq = cpu_rq(cpu); | |
722 | int ret; | |
723 | ||
724 | ret = spin_is_locked(&rq->lock); | |
725 | put_cpu(); | |
726 | return ret; | |
727 | } | |
728 | ||
bf5c91ba IM |
729 | /* |
730 | * Debugging: various feature bits | |
731 | */ | |
f00b45c1 PZ |
732 | |
733 | #define SCHED_FEAT(name, enabled) \ | |
734 | __SCHED_FEAT_##name , | |
735 | ||
bf5c91ba | 736 | enum { |
f00b45c1 | 737 | #include "sched_features.h" |
bf5c91ba IM |
738 | }; |
739 | ||
f00b45c1 PZ |
740 | #undef SCHED_FEAT |
741 | ||
742 | #define SCHED_FEAT(name, enabled) \ | |
743 | (1UL << __SCHED_FEAT_##name) * enabled | | |
744 | ||
bf5c91ba | 745 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
746 | #include "sched_features.h" |
747 | 0; | |
748 | ||
749 | #undef SCHED_FEAT | |
750 | ||
751 | #ifdef CONFIG_SCHED_DEBUG | |
752 | #define SCHED_FEAT(name, enabled) \ | |
753 | #name , | |
754 | ||
983ed7a6 | 755 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
756 | #include "sched_features.h" |
757 | NULL | |
758 | }; | |
759 | ||
760 | #undef SCHED_FEAT | |
761 | ||
34f3a814 | 762 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 763 | { |
f00b45c1 PZ |
764 | int i; |
765 | ||
766 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
767 | if (!(sysctl_sched_features & (1UL << i))) |
768 | seq_puts(m, "NO_"); | |
769 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 770 | } |
34f3a814 | 771 | seq_puts(m, "\n"); |
f00b45c1 | 772 | |
34f3a814 | 773 | return 0; |
f00b45c1 PZ |
774 | } |
775 | ||
776 | static ssize_t | |
777 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
778 | size_t cnt, loff_t *ppos) | |
779 | { | |
780 | char buf[64]; | |
781 | char *cmp = buf; | |
782 | int neg = 0; | |
783 | int i; | |
784 | ||
785 | if (cnt > 63) | |
786 | cnt = 63; | |
787 | ||
788 | if (copy_from_user(&buf, ubuf, cnt)) | |
789 | return -EFAULT; | |
790 | ||
791 | buf[cnt] = 0; | |
792 | ||
c24b7c52 | 793 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
794 | neg = 1; |
795 | cmp += 3; | |
796 | } | |
797 | ||
798 | for (i = 0; sched_feat_names[i]; i++) { | |
799 | int len = strlen(sched_feat_names[i]); | |
800 | ||
801 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
802 | if (neg) | |
803 | sysctl_sched_features &= ~(1UL << i); | |
804 | else | |
805 | sysctl_sched_features |= (1UL << i); | |
806 | break; | |
807 | } | |
808 | } | |
809 | ||
810 | if (!sched_feat_names[i]) | |
811 | return -EINVAL; | |
812 | ||
813 | filp->f_pos += cnt; | |
814 | ||
815 | return cnt; | |
816 | } | |
817 | ||
34f3a814 LZ |
818 | static int sched_feat_open(struct inode *inode, struct file *filp) |
819 | { | |
820 | return single_open(filp, sched_feat_show, NULL); | |
821 | } | |
822 | ||
f00b45c1 | 823 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
824 | .open = sched_feat_open, |
825 | .write = sched_feat_write, | |
826 | .read = seq_read, | |
827 | .llseek = seq_lseek, | |
828 | .release = single_release, | |
f00b45c1 PZ |
829 | }; |
830 | ||
831 | static __init int sched_init_debug(void) | |
832 | { | |
f00b45c1 PZ |
833 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
834 | &sched_feat_fops); | |
835 | ||
836 | return 0; | |
837 | } | |
838 | late_initcall(sched_init_debug); | |
839 | ||
840 | #endif | |
841 | ||
842 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 843 | |
b82d9fdd PZ |
844 | /* |
845 | * Number of tasks to iterate in a single balance run. | |
846 | * Limited because this is done with IRQs disabled. | |
847 | */ | |
848 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
849 | ||
2398f2c6 PZ |
850 | /* |
851 | * ratelimit for updating the group shares. | |
55cd5340 | 852 | * default: 0.25ms |
2398f2c6 | 853 | */ |
55cd5340 | 854 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 855 | |
ffda12a1 PZ |
856 | /* |
857 | * Inject some fuzzyness into changing the per-cpu group shares | |
858 | * this avoids remote rq-locks at the expense of fairness. | |
859 | * default: 4 | |
860 | */ | |
861 | unsigned int sysctl_sched_shares_thresh = 4; | |
862 | ||
fa85ae24 | 863 | /* |
9f0c1e56 | 864 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
865 | * default: 1s |
866 | */ | |
9f0c1e56 | 867 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 868 | |
6892b75e IM |
869 | static __read_mostly int scheduler_running; |
870 | ||
9f0c1e56 PZ |
871 | /* |
872 | * part of the period that we allow rt tasks to run in us. | |
873 | * default: 0.95s | |
874 | */ | |
875 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 876 | |
d0b27fa7 PZ |
877 | static inline u64 global_rt_period(void) |
878 | { | |
879 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
880 | } | |
881 | ||
882 | static inline u64 global_rt_runtime(void) | |
883 | { | |
e26873bb | 884 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
885 | return RUNTIME_INF; |
886 | ||
887 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
888 | } | |
fa85ae24 | 889 | |
1da177e4 | 890 | #ifndef prepare_arch_switch |
4866cde0 NP |
891 | # define prepare_arch_switch(next) do { } while (0) |
892 | #endif | |
893 | #ifndef finish_arch_switch | |
894 | # define finish_arch_switch(prev) do { } while (0) | |
895 | #endif | |
896 | ||
051a1d1a DA |
897 | static inline int task_current(struct rq *rq, struct task_struct *p) |
898 | { | |
899 | return rq->curr == p; | |
900 | } | |
901 | ||
4866cde0 | 902 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 903 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 904 | { |
051a1d1a | 905 | return task_current(rq, p); |
4866cde0 NP |
906 | } |
907 | ||
70b97a7f | 908 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
909 | { |
910 | } | |
911 | ||
70b97a7f | 912 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 913 | { |
da04c035 IM |
914 | #ifdef CONFIG_DEBUG_SPINLOCK |
915 | /* this is a valid case when another task releases the spinlock */ | |
916 | rq->lock.owner = current; | |
917 | #endif | |
8a25d5de IM |
918 | /* |
919 | * If we are tracking spinlock dependencies then we have to | |
920 | * fix up the runqueue lock - which gets 'carried over' from | |
921 | * prev into current: | |
922 | */ | |
923 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
924 | ||
4866cde0 NP |
925 | spin_unlock_irq(&rq->lock); |
926 | } | |
927 | ||
928 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 929 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
930 | { |
931 | #ifdef CONFIG_SMP | |
932 | return p->oncpu; | |
933 | #else | |
051a1d1a | 934 | return task_current(rq, p); |
4866cde0 NP |
935 | #endif |
936 | } | |
937 | ||
70b97a7f | 938 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
939 | { |
940 | #ifdef CONFIG_SMP | |
941 | /* | |
942 | * We can optimise this out completely for !SMP, because the | |
943 | * SMP rebalancing from interrupt is the only thing that cares | |
944 | * here. | |
945 | */ | |
946 | next->oncpu = 1; | |
947 | #endif | |
948 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
949 | spin_unlock_irq(&rq->lock); | |
950 | #else | |
951 | spin_unlock(&rq->lock); | |
952 | #endif | |
953 | } | |
954 | ||
70b97a7f | 955 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
956 | { |
957 | #ifdef CONFIG_SMP | |
958 | /* | |
959 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
960 | * We must ensure this doesn't happen until the switch is completely | |
961 | * finished. | |
962 | */ | |
963 | smp_wmb(); | |
964 | prev->oncpu = 0; | |
965 | #endif | |
966 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
967 | local_irq_enable(); | |
1da177e4 | 968 | #endif |
4866cde0 NP |
969 | } |
970 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 971 | |
b29739f9 IM |
972 | /* |
973 | * __task_rq_lock - lock the runqueue a given task resides on. | |
974 | * Must be called interrupts disabled. | |
975 | */ | |
70b97a7f | 976 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
977 | __acquires(rq->lock) |
978 | { | |
3a5c359a AK |
979 | for (;;) { |
980 | struct rq *rq = task_rq(p); | |
981 | spin_lock(&rq->lock); | |
982 | if (likely(rq == task_rq(p))) | |
983 | return rq; | |
b29739f9 | 984 | spin_unlock(&rq->lock); |
b29739f9 | 985 | } |
b29739f9 IM |
986 | } |
987 | ||
1da177e4 LT |
988 | /* |
989 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 990 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
991 | * explicitly disabling preemption. |
992 | */ | |
70b97a7f | 993 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
994 | __acquires(rq->lock) |
995 | { | |
70b97a7f | 996 | struct rq *rq; |
1da177e4 | 997 | |
3a5c359a AK |
998 | for (;;) { |
999 | local_irq_save(*flags); | |
1000 | rq = task_rq(p); | |
1001 | spin_lock(&rq->lock); | |
1002 | if (likely(rq == task_rq(p))) | |
1003 | return rq; | |
1da177e4 | 1004 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 1005 | } |
1da177e4 LT |
1006 | } |
1007 | ||
ad474cac ON |
1008 | void task_rq_unlock_wait(struct task_struct *p) |
1009 | { | |
1010 | struct rq *rq = task_rq(p); | |
1011 | ||
1012 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1013 | spin_unlock_wait(&rq->lock); | |
1014 | } | |
1015 | ||
a9957449 | 1016 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1017 | __releases(rq->lock) |
1018 | { | |
1019 | spin_unlock(&rq->lock); | |
1020 | } | |
1021 | ||
70b97a7f | 1022 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1023 | __releases(rq->lock) |
1024 | { | |
1025 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1026 | } | |
1027 | ||
1da177e4 | 1028 | /* |
cc2a73b5 | 1029 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1030 | */ |
a9957449 | 1031 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1032 | __acquires(rq->lock) |
1033 | { | |
70b97a7f | 1034 | struct rq *rq; |
1da177e4 LT |
1035 | |
1036 | local_irq_disable(); | |
1037 | rq = this_rq(); | |
1038 | spin_lock(&rq->lock); | |
1039 | ||
1040 | return rq; | |
1041 | } | |
1042 | ||
8f4d37ec PZ |
1043 | #ifdef CONFIG_SCHED_HRTICK |
1044 | /* | |
1045 | * Use HR-timers to deliver accurate preemption points. | |
1046 | * | |
1047 | * Its all a bit involved since we cannot program an hrt while holding the | |
1048 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1049 | * reschedule event. | |
1050 | * | |
1051 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1052 | * rq->lock. | |
1053 | */ | |
8f4d37ec PZ |
1054 | |
1055 | /* | |
1056 | * Use hrtick when: | |
1057 | * - enabled by features | |
1058 | * - hrtimer is actually high res | |
1059 | */ | |
1060 | static inline int hrtick_enabled(struct rq *rq) | |
1061 | { | |
1062 | if (!sched_feat(HRTICK)) | |
1063 | return 0; | |
ba42059f | 1064 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1065 | return 0; |
8f4d37ec PZ |
1066 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1067 | } | |
1068 | ||
8f4d37ec PZ |
1069 | static void hrtick_clear(struct rq *rq) |
1070 | { | |
1071 | if (hrtimer_active(&rq->hrtick_timer)) | |
1072 | hrtimer_cancel(&rq->hrtick_timer); | |
1073 | } | |
1074 | ||
8f4d37ec PZ |
1075 | /* |
1076 | * High-resolution timer tick. | |
1077 | * Runs from hardirq context with interrupts disabled. | |
1078 | */ | |
1079 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1080 | { | |
1081 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1082 | ||
1083 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1084 | ||
1085 | spin_lock(&rq->lock); | |
3e51f33f | 1086 | update_rq_clock(rq); |
8f4d37ec PZ |
1087 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1088 | spin_unlock(&rq->lock); | |
1089 | ||
1090 | return HRTIMER_NORESTART; | |
1091 | } | |
1092 | ||
95e904c7 | 1093 | #ifdef CONFIG_SMP |
31656519 PZ |
1094 | /* |
1095 | * called from hardirq (IPI) context | |
1096 | */ | |
1097 | static void __hrtick_start(void *arg) | |
b328ca18 | 1098 | { |
31656519 | 1099 | struct rq *rq = arg; |
b328ca18 | 1100 | |
31656519 PZ |
1101 | spin_lock(&rq->lock); |
1102 | hrtimer_restart(&rq->hrtick_timer); | |
1103 | rq->hrtick_csd_pending = 0; | |
1104 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1105 | } |
1106 | ||
31656519 PZ |
1107 | /* |
1108 | * Called to set the hrtick timer state. | |
1109 | * | |
1110 | * called with rq->lock held and irqs disabled | |
1111 | */ | |
1112 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1113 | { |
31656519 PZ |
1114 | struct hrtimer *timer = &rq->hrtick_timer; |
1115 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1116 | |
cc584b21 | 1117 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1118 | |
1119 | if (rq == this_rq()) { | |
1120 | hrtimer_restart(timer); | |
1121 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1122 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1123 | rq->hrtick_csd_pending = 1; |
1124 | } | |
b328ca18 PZ |
1125 | } |
1126 | ||
1127 | static int | |
1128 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1129 | { | |
1130 | int cpu = (int)(long)hcpu; | |
1131 | ||
1132 | switch (action) { | |
1133 | case CPU_UP_CANCELED: | |
1134 | case CPU_UP_CANCELED_FROZEN: | |
1135 | case CPU_DOWN_PREPARE: | |
1136 | case CPU_DOWN_PREPARE_FROZEN: | |
1137 | case CPU_DEAD: | |
1138 | case CPU_DEAD_FROZEN: | |
31656519 | 1139 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1140 | return NOTIFY_OK; |
1141 | } | |
1142 | ||
1143 | return NOTIFY_DONE; | |
1144 | } | |
1145 | ||
fa748203 | 1146 | static __init void init_hrtick(void) |
b328ca18 PZ |
1147 | { |
1148 | hotcpu_notifier(hotplug_hrtick, 0); | |
1149 | } | |
31656519 PZ |
1150 | #else |
1151 | /* | |
1152 | * Called to set the hrtick timer state. | |
1153 | * | |
1154 | * called with rq->lock held and irqs disabled | |
1155 | */ | |
1156 | static void hrtick_start(struct rq *rq, u64 delay) | |
1157 | { | |
7f1e2ca9 PZ |
1158 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
1159 | HRTIMER_MODE_REL, 0); | |
31656519 | 1160 | } |
b328ca18 | 1161 | |
006c75f1 | 1162 | static inline void init_hrtick(void) |
8f4d37ec | 1163 | { |
8f4d37ec | 1164 | } |
31656519 | 1165 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1166 | |
31656519 | 1167 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1168 | { |
31656519 PZ |
1169 | #ifdef CONFIG_SMP |
1170 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1171 | |
31656519 PZ |
1172 | rq->hrtick_csd.flags = 0; |
1173 | rq->hrtick_csd.func = __hrtick_start; | |
1174 | rq->hrtick_csd.info = rq; | |
1175 | #endif | |
8f4d37ec | 1176 | |
31656519 PZ |
1177 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1178 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1179 | } |
006c75f1 | 1180 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1181 | static inline void hrtick_clear(struct rq *rq) |
1182 | { | |
1183 | } | |
1184 | ||
8f4d37ec PZ |
1185 | static inline void init_rq_hrtick(struct rq *rq) |
1186 | { | |
1187 | } | |
1188 | ||
b328ca18 PZ |
1189 | static inline void init_hrtick(void) |
1190 | { | |
1191 | } | |
006c75f1 | 1192 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1193 | |
c24d20db IM |
1194 | /* |
1195 | * resched_task - mark a task 'to be rescheduled now'. | |
1196 | * | |
1197 | * On UP this means the setting of the need_resched flag, on SMP it | |
1198 | * might also involve a cross-CPU call to trigger the scheduler on | |
1199 | * the target CPU. | |
1200 | */ | |
1201 | #ifdef CONFIG_SMP | |
1202 | ||
1203 | #ifndef tsk_is_polling | |
1204 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1205 | #endif | |
1206 | ||
31656519 | 1207 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1208 | { |
1209 | int cpu; | |
1210 | ||
1211 | assert_spin_locked(&task_rq(p)->lock); | |
1212 | ||
5ed0cec0 | 1213 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1214 | return; |
1215 | ||
5ed0cec0 | 1216 | set_tsk_need_resched(p); |
c24d20db IM |
1217 | |
1218 | cpu = task_cpu(p); | |
1219 | if (cpu == smp_processor_id()) | |
1220 | return; | |
1221 | ||
1222 | /* NEED_RESCHED must be visible before we test polling */ | |
1223 | smp_mb(); | |
1224 | if (!tsk_is_polling(p)) | |
1225 | smp_send_reschedule(cpu); | |
1226 | } | |
1227 | ||
1228 | static void resched_cpu(int cpu) | |
1229 | { | |
1230 | struct rq *rq = cpu_rq(cpu); | |
1231 | unsigned long flags; | |
1232 | ||
1233 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1234 | return; | |
1235 | resched_task(cpu_curr(cpu)); | |
1236 | spin_unlock_irqrestore(&rq->lock, flags); | |
1237 | } | |
06d8308c TG |
1238 | |
1239 | #ifdef CONFIG_NO_HZ | |
1240 | /* | |
1241 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1242 | * idle CPU then this timer might expire before the next timer event | |
1243 | * which is scheduled to wake up that CPU. In case of a completely | |
1244 | * idle system the next event might even be infinite time into the | |
1245 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1246 | * leaves the inner idle loop so the newly added timer is taken into | |
1247 | * account when the CPU goes back to idle and evaluates the timer | |
1248 | * wheel for the next timer event. | |
1249 | */ | |
1250 | void wake_up_idle_cpu(int cpu) | |
1251 | { | |
1252 | struct rq *rq = cpu_rq(cpu); | |
1253 | ||
1254 | if (cpu == smp_processor_id()) | |
1255 | return; | |
1256 | ||
1257 | /* | |
1258 | * This is safe, as this function is called with the timer | |
1259 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1260 | * to idle and has not yet set rq->curr to idle then it will | |
1261 | * be serialized on the timer wheel base lock and take the new | |
1262 | * timer into account automatically. | |
1263 | */ | |
1264 | if (rq->curr != rq->idle) | |
1265 | return; | |
1266 | ||
1267 | /* | |
1268 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1269 | * lockless. The worst case is that the other CPU runs the | |
1270 | * idle task through an additional NOOP schedule() | |
1271 | */ | |
5ed0cec0 | 1272 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1273 | |
1274 | /* NEED_RESCHED must be visible before we test polling */ | |
1275 | smp_mb(); | |
1276 | if (!tsk_is_polling(rq->idle)) | |
1277 | smp_send_reschedule(cpu); | |
1278 | } | |
6d6bc0ad | 1279 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1280 | |
6d6bc0ad | 1281 | #else /* !CONFIG_SMP */ |
31656519 | 1282 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1283 | { |
1284 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1285 | set_tsk_need_resched(p); |
c24d20db | 1286 | } |
6d6bc0ad | 1287 | #endif /* CONFIG_SMP */ |
c24d20db | 1288 | |
45bf76df IM |
1289 | #if BITS_PER_LONG == 32 |
1290 | # define WMULT_CONST (~0UL) | |
1291 | #else | |
1292 | # define WMULT_CONST (1UL << 32) | |
1293 | #endif | |
1294 | ||
1295 | #define WMULT_SHIFT 32 | |
1296 | ||
194081eb IM |
1297 | /* |
1298 | * Shift right and round: | |
1299 | */ | |
cf2ab469 | 1300 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1301 | |
a7be37ac PZ |
1302 | /* |
1303 | * delta *= weight / lw | |
1304 | */ | |
cb1c4fc9 | 1305 | static unsigned long |
45bf76df IM |
1306 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1307 | struct load_weight *lw) | |
1308 | { | |
1309 | u64 tmp; | |
1310 | ||
7a232e03 LJ |
1311 | if (!lw->inv_weight) { |
1312 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1313 | lw->inv_weight = 1; | |
1314 | else | |
1315 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1316 | / (lw->weight+1); | |
1317 | } | |
45bf76df IM |
1318 | |
1319 | tmp = (u64)delta_exec * weight; | |
1320 | /* | |
1321 | * Check whether we'd overflow the 64-bit multiplication: | |
1322 | */ | |
194081eb | 1323 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1324 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1325 | WMULT_SHIFT/2); |
1326 | else | |
cf2ab469 | 1327 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1328 | |
ecf691da | 1329 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1330 | } |
1331 | ||
1091985b | 1332 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1333 | { |
1334 | lw->weight += inc; | |
e89996ae | 1335 | lw->inv_weight = 0; |
45bf76df IM |
1336 | } |
1337 | ||
1091985b | 1338 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1339 | { |
1340 | lw->weight -= dec; | |
e89996ae | 1341 | lw->inv_weight = 0; |
45bf76df IM |
1342 | } |
1343 | ||
2dd73a4f PW |
1344 | /* |
1345 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1346 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1347 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1348 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1349 | * scaled version of the new time slice allocation that they receive on time |
1350 | * slice expiry etc. | |
1351 | */ | |
1352 | ||
cce7ade8 PZ |
1353 | #define WEIGHT_IDLEPRIO 3 |
1354 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1355 | |
1356 | /* | |
1357 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1358 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1359 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1360 | * that remained on nice 0. | |
1361 | * | |
1362 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1363 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1364 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1365 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1366 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1367 | */ |
1368 | static const int prio_to_weight[40] = { | |
254753dc IM |
1369 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1370 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1371 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1372 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1373 | /* 0 */ 1024, 820, 655, 526, 423, | |
1374 | /* 5 */ 335, 272, 215, 172, 137, | |
1375 | /* 10 */ 110, 87, 70, 56, 45, | |
1376 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1377 | }; |
1378 | ||
5714d2de IM |
1379 | /* |
1380 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1381 | * | |
1382 | * In cases where the weight does not change often, we can use the | |
1383 | * precalculated inverse to speed up arithmetics by turning divisions | |
1384 | * into multiplications: | |
1385 | */ | |
dd41f596 | 1386 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1387 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1388 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1389 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1390 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1391 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1392 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1393 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1394 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1395 | }; |
2dd73a4f | 1396 | |
dd41f596 IM |
1397 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1398 | ||
1399 | /* | |
1400 | * runqueue iterator, to support SMP load-balancing between different | |
1401 | * scheduling classes, without having to expose their internal data | |
1402 | * structures to the load-balancing proper: | |
1403 | */ | |
1404 | struct rq_iterator { | |
1405 | void *arg; | |
1406 | struct task_struct *(*start)(void *); | |
1407 | struct task_struct *(*next)(void *); | |
1408 | }; | |
1409 | ||
e1d1484f PW |
1410 | #ifdef CONFIG_SMP |
1411 | static unsigned long | |
1412 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1413 | unsigned long max_load_move, struct sched_domain *sd, | |
1414 | enum cpu_idle_type idle, int *all_pinned, | |
1415 | int *this_best_prio, struct rq_iterator *iterator); | |
1416 | ||
1417 | static int | |
1418 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1419 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1420 | struct rq_iterator *iterator); | |
e1d1484f | 1421 | #endif |
dd41f596 | 1422 | |
ef12fefa BR |
1423 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1424 | enum cpuacct_stat_index { | |
1425 | CPUACCT_STAT_USER, /* ... user mode */ | |
1426 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1427 | ||
1428 | CPUACCT_STAT_NSTATS, | |
1429 | }; | |
1430 | ||
d842de87 SV |
1431 | #ifdef CONFIG_CGROUP_CPUACCT |
1432 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1433 | static void cpuacct_update_stats(struct task_struct *tsk, |
1434 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1435 | #else |
1436 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1437 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1438 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1439 | #endif |
1440 | ||
18d95a28 PZ |
1441 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1442 | { | |
1443 | update_load_add(&rq->load, load); | |
1444 | } | |
1445 | ||
1446 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1447 | { | |
1448 | update_load_sub(&rq->load, load); | |
1449 | } | |
1450 | ||
7940ca36 | 1451 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1452 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1453 | |
1454 | /* | |
1455 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1456 | * leaving it for the final time. | |
1457 | */ | |
eb755805 | 1458 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1459 | { |
1460 | struct task_group *parent, *child; | |
eb755805 | 1461 | int ret; |
c09595f6 PZ |
1462 | |
1463 | rcu_read_lock(); | |
1464 | parent = &root_task_group; | |
1465 | down: | |
eb755805 PZ |
1466 | ret = (*down)(parent, data); |
1467 | if (ret) | |
1468 | goto out_unlock; | |
c09595f6 PZ |
1469 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1470 | parent = child; | |
1471 | goto down; | |
1472 | ||
1473 | up: | |
1474 | continue; | |
1475 | } | |
eb755805 PZ |
1476 | ret = (*up)(parent, data); |
1477 | if (ret) | |
1478 | goto out_unlock; | |
c09595f6 PZ |
1479 | |
1480 | child = parent; | |
1481 | parent = parent->parent; | |
1482 | if (parent) | |
1483 | goto up; | |
eb755805 | 1484 | out_unlock: |
c09595f6 | 1485 | rcu_read_unlock(); |
eb755805 PZ |
1486 | |
1487 | return ret; | |
c09595f6 PZ |
1488 | } |
1489 | ||
eb755805 PZ |
1490 | static int tg_nop(struct task_group *tg, void *data) |
1491 | { | |
1492 | return 0; | |
c09595f6 | 1493 | } |
eb755805 PZ |
1494 | #endif |
1495 | ||
1496 | #ifdef CONFIG_SMP | |
1497 | static unsigned long source_load(int cpu, int type); | |
1498 | static unsigned long target_load(int cpu, int type); | |
1499 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1500 | ||
1501 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1502 | { | |
1503 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1504 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1505 | |
4cd42620 SR |
1506 | if (nr_running) |
1507 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1508 | else |
1509 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1510 | |
1511 | return rq->avg_load_per_task; | |
1512 | } | |
1513 | ||
1514 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1515 | |
c09595f6 PZ |
1516 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1517 | ||
1518 | /* | |
1519 | * Calculate and set the cpu's group shares. | |
1520 | */ | |
1521 | static void | |
ffda12a1 PZ |
1522 | update_group_shares_cpu(struct task_group *tg, int cpu, |
1523 | unsigned long sd_shares, unsigned long sd_rq_weight) | |
18d95a28 | 1524 | { |
c09595f6 PZ |
1525 | unsigned long shares; |
1526 | unsigned long rq_weight; | |
1527 | ||
c8cba857 | 1528 | if (!tg->se[cpu]) |
c09595f6 PZ |
1529 | return; |
1530 | ||
ec4e0e2f | 1531 | rq_weight = tg->cfs_rq[cpu]->rq_weight; |
c8cba857 | 1532 | |
c09595f6 PZ |
1533 | /* |
1534 | * \Sum shares * rq_weight | |
1535 | * shares = ----------------------- | |
1536 | * \Sum rq_weight | |
1537 | * | |
1538 | */ | |
ec4e0e2f | 1539 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1540 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1541 | |
ffda12a1 PZ |
1542 | if (abs(shares - tg->se[cpu]->load.weight) > |
1543 | sysctl_sched_shares_thresh) { | |
1544 | struct rq *rq = cpu_rq(cpu); | |
1545 | unsigned long flags; | |
c09595f6 | 1546 | |
ffda12a1 | 1547 | spin_lock_irqsave(&rq->lock, flags); |
ec4e0e2f | 1548 | tg->cfs_rq[cpu]->shares = shares; |
c09595f6 | 1549 | |
ffda12a1 PZ |
1550 | __set_se_shares(tg->se[cpu], shares); |
1551 | spin_unlock_irqrestore(&rq->lock, flags); | |
1552 | } | |
18d95a28 | 1553 | } |
c09595f6 PZ |
1554 | |
1555 | /* | |
c8cba857 PZ |
1556 | * Re-compute the task group their per cpu shares over the given domain. |
1557 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1558 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1559 | */ |
eb755805 | 1560 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1561 | { |
ec4e0e2f | 1562 | unsigned long weight, rq_weight = 0; |
c8cba857 | 1563 | unsigned long shares = 0; |
eb755805 | 1564 | struct sched_domain *sd = data; |
c8cba857 | 1565 | int i; |
c09595f6 | 1566 | |
758b2cdc | 1567 | for_each_cpu(i, sched_domain_span(sd)) { |
ec4e0e2f KC |
1568 | /* |
1569 | * If there are currently no tasks on the cpu pretend there | |
1570 | * is one of average load so that when a new task gets to | |
1571 | * run here it will not get delayed by group starvation. | |
1572 | */ | |
1573 | weight = tg->cfs_rq[i]->load.weight; | |
1574 | if (!weight) | |
1575 | weight = NICE_0_LOAD; | |
1576 | ||
1577 | tg->cfs_rq[i]->rq_weight = weight; | |
1578 | rq_weight += weight; | |
c8cba857 | 1579 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1580 | } |
c09595f6 | 1581 | |
c8cba857 PZ |
1582 | if ((!shares && rq_weight) || shares > tg->shares) |
1583 | shares = tg->shares; | |
1584 | ||
1585 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1586 | shares = tg->shares; | |
c09595f6 | 1587 | |
758b2cdc | 1588 | for_each_cpu(i, sched_domain_span(sd)) |
ffda12a1 | 1589 | update_group_shares_cpu(tg, i, shares, rq_weight); |
eb755805 PZ |
1590 | |
1591 | return 0; | |
c09595f6 PZ |
1592 | } |
1593 | ||
1594 | /* | |
c8cba857 PZ |
1595 | * Compute the cpu's hierarchical load factor for each task group. |
1596 | * This needs to be done in a top-down fashion because the load of a child | |
1597 | * group is a fraction of its parents load. | |
c09595f6 | 1598 | */ |
eb755805 | 1599 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1600 | { |
c8cba857 | 1601 | unsigned long load; |
eb755805 | 1602 | long cpu = (long)data; |
c09595f6 | 1603 | |
c8cba857 PZ |
1604 | if (!tg->parent) { |
1605 | load = cpu_rq(cpu)->load.weight; | |
1606 | } else { | |
1607 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1608 | load *= tg->cfs_rq[cpu]->shares; | |
1609 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1610 | } | |
c09595f6 | 1611 | |
c8cba857 | 1612 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1613 | |
eb755805 | 1614 | return 0; |
c09595f6 PZ |
1615 | } |
1616 | ||
c8cba857 | 1617 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1618 | { |
2398f2c6 PZ |
1619 | u64 now = cpu_clock(raw_smp_processor_id()); |
1620 | s64 elapsed = now - sd->last_update; | |
1621 | ||
1622 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1623 | sd->last_update = now; | |
eb755805 | 1624 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1625 | } |
4d8d595d PZ |
1626 | } |
1627 | ||
3e5459b4 PZ |
1628 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1629 | { | |
1630 | spin_unlock(&rq->lock); | |
1631 | update_shares(sd); | |
1632 | spin_lock(&rq->lock); | |
1633 | } | |
1634 | ||
eb755805 | 1635 | static void update_h_load(long cpu) |
c09595f6 | 1636 | { |
eb755805 | 1637 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1638 | } |
1639 | ||
c09595f6 PZ |
1640 | #else |
1641 | ||
c8cba857 | 1642 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1643 | { |
1644 | } | |
1645 | ||
3e5459b4 PZ |
1646 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1647 | { | |
1648 | } | |
1649 | ||
18d95a28 PZ |
1650 | #endif |
1651 | ||
8f45e2b5 GH |
1652 | #ifdef CONFIG_PREEMPT |
1653 | ||
70574a99 | 1654 | /* |
8f45e2b5 GH |
1655 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1656 | * way at the expense of forcing extra atomic operations in all | |
1657 | * invocations. This assures that the double_lock is acquired using the | |
1658 | * same underlying policy as the spinlock_t on this architecture, which | |
1659 | * reduces latency compared to the unfair variant below. However, it | |
1660 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1661 | */ |
8f45e2b5 GH |
1662 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1663 | __releases(this_rq->lock) | |
1664 | __acquires(busiest->lock) | |
1665 | __acquires(this_rq->lock) | |
1666 | { | |
1667 | spin_unlock(&this_rq->lock); | |
1668 | double_rq_lock(this_rq, busiest); | |
1669 | ||
1670 | return 1; | |
1671 | } | |
1672 | ||
1673 | #else | |
1674 | /* | |
1675 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1676 | * latency by eliminating extra atomic operations when the locks are | |
1677 | * already in proper order on entry. This favors lower cpu-ids and will | |
1678 | * grant the double lock to lower cpus over higher ids under contention, | |
1679 | * regardless of entry order into the function. | |
1680 | */ | |
1681 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1682 | __releases(this_rq->lock) |
1683 | __acquires(busiest->lock) | |
1684 | __acquires(this_rq->lock) | |
1685 | { | |
1686 | int ret = 0; | |
1687 | ||
70574a99 AD |
1688 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1689 | if (busiest < this_rq) { | |
1690 | spin_unlock(&this_rq->lock); | |
1691 | spin_lock(&busiest->lock); | |
1692 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1693 | ret = 1; | |
1694 | } else | |
1695 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1696 | } | |
1697 | return ret; | |
1698 | } | |
1699 | ||
8f45e2b5 GH |
1700 | #endif /* CONFIG_PREEMPT */ |
1701 | ||
1702 | /* | |
1703 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1704 | */ | |
1705 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1706 | { | |
1707 | if (unlikely(!irqs_disabled())) { | |
1708 | /* printk() doesn't work good under rq->lock */ | |
1709 | spin_unlock(&this_rq->lock); | |
1710 | BUG_ON(1); | |
1711 | } | |
1712 | ||
1713 | return _double_lock_balance(this_rq, busiest); | |
1714 | } | |
1715 | ||
70574a99 AD |
1716 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1717 | __releases(busiest->lock) | |
1718 | { | |
1719 | spin_unlock(&busiest->lock); | |
1720 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1721 | } | |
18d95a28 PZ |
1722 | #endif |
1723 | ||
30432094 | 1724 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1725 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1726 | { | |
30432094 | 1727 | #ifdef CONFIG_SMP |
34e83e85 IM |
1728 | cfs_rq->shares = shares; |
1729 | #endif | |
1730 | } | |
30432094 | 1731 | #endif |
e7693a36 | 1732 | |
dd41f596 | 1733 | #include "sched_stats.h" |
dd41f596 | 1734 | #include "sched_idletask.c" |
5522d5d5 IM |
1735 | #include "sched_fair.c" |
1736 | #include "sched_rt.c" | |
dd41f596 IM |
1737 | #ifdef CONFIG_SCHED_DEBUG |
1738 | # include "sched_debug.c" | |
1739 | #endif | |
1740 | ||
1741 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1742 | #define for_each_class(class) \ |
1743 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1744 | |
c09595f6 | 1745 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1746 | { |
1747 | rq->nr_running++; | |
9c217245 IM |
1748 | } |
1749 | ||
c09595f6 | 1750 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1751 | { |
1752 | rq->nr_running--; | |
9c217245 IM |
1753 | } |
1754 | ||
45bf76df IM |
1755 | static void set_load_weight(struct task_struct *p) |
1756 | { | |
1757 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1758 | p->se.load.weight = prio_to_weight[0] * 2; |
1759 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1760 | return; | |
1761 | } | |
45bf76df | 1762 | |
dd41f596 IM |
1763 | /* |
1764 | * SCHED_IDLE tasks get minimal weight: | |
1765 | */ | |
1766 | if (p->policy == SCHED_IDLE) { | |
1767 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1768 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1769 | return; | |
1770 | } | |
71f8bd46 | 1771 | |
dd41f596 IM |
1772 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1773 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1774 | } |
1775 | ||
2087a1ad GH |
1776 | static void update_avg(u64 *avg, u64 sample) |
1777 | { | |
1778 | s64 diff = sample - *avg; | |
1779 | *avg += diff >> 3; | |
1780 | } | |
1781 | ||
8159f87e | 1782 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1783 | { |
831451ac PZ |
1784 | if (wakeup) |
1785 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1786 | ||
dd41f596 | 1787 | sched_info_queued(p); |
fd390f6a | 1788 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1789 | p->se.on_rq = 1; |
71f8bd46 IM |
1790 | } |
1791 | ||
69be72c1 | 1792 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1793 | { |
831451ac PZ |
1794 | if (sleep) { |
1795 | if (p->se.last_wakeup) { | |
1796 | update_avg(&p->se.avg_overlap, | |
1797 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1798 | p->se.last_wakeup = 0; | |
1799 | } else { | |
1800 | update_avg(&p->se.avg_wakeup, | |
1801 | sysctl_sched_wakeup_granularity); | |
1802 | } | |
2087a1ad GH |
1803 | } |
1804 | ||
46ac22ba | 1805 | sched_info_dequeued(p); |
f02231e5 | 1806 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1807 | p->se.on_rq = 0; |
71f8bd46 IM |
1808 | } |
1809 | ||
14531189 | 1810 | /* |
dd41f596 | 1811 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1812 | */ |
14531189 IM |
1813 | static inline int __normal_prio(struct task_struct *p) |
1814 | { | |
dd41f596 | 1815 | return p->static_prio; |
14531189 IM |
1816 | } |
1817 | ||
b29739f9 IM |
1818 | /* |
1819 | * Calculate the expected normal priority: i.e. priority | |
1820 | * without taking RT-inheritance into account. Might be | |
1821 | * boosted by interactivity modifiers. Changes upon fork, | |
1822 | * setprio syscalls, and whenever the interactivity | |
1823 | * estimator recalculates. | |
1824 | */ | |
36c8b586 | 1825 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1826 | { |
1827 | int prio; | |
1828 | ||
e05606d3 | 1829 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1830 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1831 | else | |
1832 | prio = __normal_prio(p); | |
1833 | return prio; | |
1834 | } | |
1835 | ||
1836 | /* | |
1837 | * Calculate the current priority, i.e. the priority | |
1838 | * taken into account by the scheduler. This value might | |
1839 | * be boosted by RT tasks, or might be boosted by | |
1840 | * interactivity modifiers. Will be RT if the task got | |
1841 | * RT-boosted. If not then it returns p->normal_prio. | |
1842 | */ | |
36c8b586 | 1843 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1844 | { |
1845 | p->normal_prio = normal_prio(p); | |
1846 | /* | |
1847 | * If we are RT tasks or we were boosted to RT priority, | |
1848 | * keep the priority unchanged. Otherwise, update priority | |
1849 | * to the normal priority: | |
1850 | */ | |
1851 | if (!rt_prio(p->prio)) | |
1852 | return p->normal_prio; | |
1853 | return p->prio; | |
1854 | } | |
1855 | ||
1da177e4 | 1856 | /* |
dd41f596 | 1857 | * activate_task - move a task to the runqueue. |
1da177e4 | 1858 | */ |
dd41f596 | 1859 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1860 | { |
d9514f6c | 1861 | if (task_contributes_to_load(p)) |
dd41f596 | 1862 | rq->nr_uninterruptible--; |
1da177e4 | 1863 | |
8159f87e | 1864 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1865 | inc_nr_running(rq); |
1da177e4 LT |
1866 | } |
1867 | ||
1da177e4 LT |
1868 | /* |
1869 | * deactivate_task - remove a task from the runqueue. | |
1870 | */ | |
2e1cb74a | 1871 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1872 | { |
d9514f6c | 1873 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1874 | rq->nr_uninterruptible++; |
1875 | ||
69be72c1 | 1876 | dequeue_task(rq, p, sleep); |
c09595f6 | 1877 | dec_nr_running(rq); |
1da177e4 LT |
1878 | } |
1879 | ||
1da177e4 LT |
1880 | /** |
1881 | * task_curr - is this task currently executing on a CPU? | |
1882 | * @p: the task in question. | |
1883 | */ | |
36c8b586 | 1884 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1885 | { |
1886 | return cpu_curr(task_cpu(p)) == p; | |
1887 | } | |
1888 | ||
dd41f596 IM |
1889 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1890 | { | |
6f505b16 | 1891 | set_task_rq(p, cpu); |
dd41f596 | 1892 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1893 | /* |
1894 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1895 | * successfuly executed on another CPU. We must ensure that updates of | |
1896 | * per-task data have been completed by this moment. | |
1897 | */ | |
1898 | smp_wmb(); | |
dd41f596 | 1899 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1900 | #endif |
2dd73a4f PW |
1901 | } |
1902 | ||
cb469845 SR |
1903 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1904 | const struct sched_class *prev_class, | |
1905 | int oldprio, int running) | |
1906 | { | |
1907 | if (prev_class != p->sched_class) { | |
1908 | if (prev_class->switched_from) | |
1909 | prev_class->switched_from(rq, p, running); | |
1910 | p->sched_class->switched_to(rq, p, running); | |
1911 | } else | |
1912 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1913 | } | |
1914 | ||
1da177e4 | 1915 | #ifdef CONFIG_SMP |
c65cc870 | 1916 | |
e958b360 TG |
1917 | /* Used instead of source_load when we know the type == 0 */ |
1918 | static unsigned long weighted_cpuload(const int cpu) | |
1919 | { | |
1920 | return cpu_rq(cpu)->load.weight; | |
1921 | } | |
1922 | ||
cc367732 IM |
1923 | /* |
1924 | * Is this task likely cache-hot: | |
1925 | */ | |
e7693a36 | 1926 | static int |
cc367732 IM |
1927 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1928 | { | |
1929 | s64 delta; | |
1930 | ||
f540a608 IM |
1931 | /* |
1932 | * Buddy candidates are cache hot: | |
1933 | */ | |
4793241b PZ |
1934 | if (sched_feat(CACHE_HOT_BUDDY) && |
1935 | (&p->se == cfs_rq_of(&p->se)->next || | |
1936 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1937 | return 1; |
1938 | ||
cc367732 IM |
1939 | if (p->sched_class != &fair_sched_class) |
1940 | return 0; | |
1941 | ||
6bc1665b IM |
1942 | if (sysctl_sched_migration_cost == -1) |
1943 | return 1; | |
1944 | if (sysctl_sched_migration_cost == 0) | |
1945 | return 0; | |
1946 | ||
cc367732 IM |
1947 | delta = now - p->se.exec_start; |
1948 | ||
1949 | return delta < (s64)sysctl_sched_migration_cost; | |
1950 | } | |
1951 | ||
1952 | ||
dd41f596 | 1953 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1954 | { |
dd41f596 IM |
1955 | int old_cpu = task_cpu(p); |
1956 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1957 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1958 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1959 | u64 clock_offset; |
dd41f596 IM |
1960 | |
1961 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 1962 | |
cbc34ed1 PZ |
1963 | trace_sched_migrate_task(p, task_cpu(p), new_cpu); |
1964 | ||
6cfb0d5d IM |
1965 | #ifdef CONFIG_SCHEDSTATS |
1966 | if (p->se.wait_start) | |
1967 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1968 | if (p->se.sleep_start) |
1969 | p->se.sleep_start -= clock_offset; | |
1970 | if (p->se.block_start) | |
1971 | p->se.block_start -= clock_offset; | |
6c594c21 | 1972 | #endif |
cc367732 | 1973 | if (old_cpu != new_cpu) { |
6c594c21 | 1974 | p->se.nr_migrations++; |
23a185ca | 1975 | new_rq->nr_migrations_in++; |
6c594c21 | 1976 | #ifdef CONFIG_SCHEDSTATS |
cc367732 IM |
1977 | if (task_hot(p, old_rq->clock, NULL)) |
1978 | schedstat_inc(p, se.nr_forced2_migrations); | |
6cfb0d5d | 1979 | #endif |
3f731ca6 | 1980 | perf_counter_task_migration(p, new_cpu); |
6c594c21 | 1981 | } |
2830cf8c SV |
1982 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1983 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1984 | |
1985 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1986 | } |
1987 | ||
70b97a7f | 1988 | struct migration_req { |
1da177e4 | 1989 | struct list_head list; |
1da177e4 | 1990 | |
36c8b586 | 1991 | struct task_struct *task; |
1da177e4 LT |
1992 | int dest_cpu; |
1993 | ||
1da177e4 | 1994 | struct completion done; |
70b97a7f | 1995 | }; |
1da177e4 LT |
1996 | |
1997 | /* | |
1998 | * The task's runqueue lock must be held. | |
1999 | * Returns true if you have to wait for migration thread. | |
2000 | */ | |
36c8b586 | 2001 | static int |
70b97a7f | 2002 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2003 | { |
70b97a7f | 2004 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2005 | |
2006 | /* | |
2007 | * If the task is not on a runqueue (and not running), then | |
2008 | * it is sufficient to simply update the task's cpu field. | |
2009 | */ | |
dd41f596 | 2010 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2011 | set_task_cpu(p, dest_cpu); |
2012 | return 0; | |
2013 | } | |
2014 | ||
2015 | init_completion(&req->done); | |
1da177e4 LT |
2016 | req->task = p; |
2017 | req->dest_cpu = dest_cpu; | |
2018 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2019 | |
1da177e4 LT |
2020 | return 1; |
2021 | } | |
2022 | ||
2023 | /* | |
2024 | * wait_task_inactive - wait for a thread to unschedule. | |
2025 | * | |
85ba2d86 RM |
2026 | * If @match_state is nonzero, it's the @p->state value just checked and |
2027 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2028 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2029 | * we return a positive number (its total switch count). If a second call | |
2030 | * a short while later returns the same number, the caller can be sure that | |
2031 | * @p has remained unscheduled the whole time. | |
2032 | * | |
1da177e4 LT |
2033 | * The caller must ensure that the task *will* unschedule sometime soon, |
2034 | * else this function might spin for a *long* time. This function can't | |
2035 | * be called with interrupts off, or it may introduce deadlock with | |
2036 | * smp_call_function() if an IPI is sent by the same process we are | |
2037 | * waiting to become inactive. | |
2038 | */ | |
85ba2d86 | 2039 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2040 | { |
2041 | unsigned long flags; | |
dd41f596 | 2042 | int running, on_rq; |
85ba2d86 | 2043 | unsigned long ncsw; |
70b97a7f | 2044 | struct rq *rq; |
1da177e4 | 2045 | |
3a5c359a AK |
2046 | for (;;) { |
2047 | /* | |
2048 | * We do the initial early heuristics without holding | |
2049 | * any task-queue locks at all. We'll only try to get | |
2050 | * the runqueue lock when things look like they will | |
2051 | * work out! | |
2052 | */ | |
2053 | rq = task_rq(p); | |
fa490cfd | 2054 | |
3a5c359a AK |
2055 | /* |
2056 | * If the task is actively running on another CPU | |
2057 | * still, just relax and busy-wait without holding | |
2058 | * any locks. | |
2059 | * | |
2060 | * NOTE! Since we don't hold any locks, it's not | |
2061 | * even sure that "rq" stays as the right runqueue! | |
2062 | * But we don't care, since "task_running()" will | |
2063 | * return false if the runqueue has changed and p | |
2064 | * is actually now running somewhere else! | |
2065 | */ | |
85ba2d86 RM |
2066 | while (task_running(rq, p)) { |
2067 | if (match_state && unlikely(p->state != match_state)) | |
2068 | return 0; | |
3a5c359a | 2069 | cpu_relax(); |
85ba2d86 | 2070 | } |
fa490cfd | 2071 | |
3a5c359a AK |
2072 | /* |
2073 | * Ok, time to look more closely! We need the rq | |
2074 | * lock now, to be *sure*. If we're wrong, we'll | |
2075 | * just go back and repeat. | |
2076 | */ | |
2077 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2078 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2079 | running = task_running(rq, p); |
2080 | on_rq = p->se.on_rq; | |
85ba2d86 | 2081 | ncsw = 0; |
f31e11d8 | 2082 | if (!match_state || p->state == match_state) |
93dcf55f | 2083 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2084 | task_rq_unlock(rq, &flags); |
fa490cfd | 2085 | |
85ba2d86 RM |
2086 | /* |
2087 | * If it changed from the expected state, bail out now. | |
2088 | */ | |
2089 | if (unlikely(!ncsw)) | |
2090 | break; | |
2091 | ||
3a5c359a AK |
2092 | /* |
2093 | * Was it really running after all now that we | |
2094 | * checked with the proper locks actually held? | |
2095 | * | |
2096 | * Oops. Go back and try again.. | |
2097 | */ | |
2098 | if (unlikely(running)) { | |
2099 | cpu_relax(); | |
2100 | continue; | |
2101 | } | |
fa490cfd | 2102 | |
3a5c359a AK |
2103 | /* |
2104 | * It's not enough that it's not actively running, | |
2105 | * it must be off the runqueue _entirely_, and not | |
2106 | * preempted! | |
2107 | * | |
80dd99b3 | 2108 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2109 | * running right now), it's preempted, and we should |
2110 | * yield - it could be a while. | |
2111 | */ | |
2112 | if (unlikely(on_rq)) { | |
2113 | schedule_timeout_uninterruptible(1); | |
2114 | continue; | |
2115 | } | |
fa490cfd | 2116 | |
3a5c359a AK |
2117 | /* |
2118 | * Ahh, all good. It wasn't running, and it wasn't | |
2119 | * runnable, which means that it will never become | |
2120 | * running in the future either. We're all done! | |
2121 | */ | |
2122 | break; | |
2123 | } | |
85ba2d86 RM |
2124 | |
2125 | return ncsw; | |
1da177e4 LT |
2126 | } |
2127 | ||
2128 | /*** | |
2129 | * kick_process - kick a running thread to enter/exit the kernel | |
2130 | * @p: the to-be-kicked thread | |
2131 | * | |
2132 | * Cause a process which is running on another CPU to enter | |
2133 | * kernel-mode, without any delay. (to get signals handled.) | |
2134 | * | |
2135 | * NOTE: this function doesnt have to take the runqueue lock, | |
2136 | * because all it wants to ensure is that the remote task enters | |
2137 | * the kernel. If the IPI races and the task has been migrated | |
2138 | * to another CPU then no harm is done and the purpose has been | |
2139 | * achieved as well. | |
2140 | */ | |
36c8b586 | 2141 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2142 | { |
2143 | int cpu; | |
2144 | ||
2145 | preempt_disable(); | |
2146 | cpu = task_cpu(p); | |
2147 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2148 | smp_send_reschedule(cpu); | |
2149 | preempt_enable(); | |
2150 | } | |
2151 | ||
2152 | /* | |
2dd73a4f PW |
2153 | * Return a low guess at the load of a migration-source cpu weighted |
2154 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2155 | * |
2156 | * We want to under-estimate the load of migration sources, to | |
2157 | * balance conservatively. | |
2158 | */ | |
a9957449 | 2159 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2160 | { |
70b97a7f | 2161 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2162 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2163 | |
93b75217 | 2164 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2165 | return total; |
b910472d | 2166 | |
dd41f596 | 2167 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2168 | } |
2169 | ||
2170 | /* | |
2dd73a4f PW |
2171 | * Return a high guess at the load of a migration-target cpu weighted |
2172 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2173 | */ |
a9957449 | 2174 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2175 | { |
70b97a7f | 2176 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2177 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2178 | |
93b75217 | 2179 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2180 | return total; |
3b0bd9bc | 2181 | |
dd41f596 | 2182 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2183 | } |
2184 | ||
147cbb4b NP |
2185 | /* |
2186 | * find_idlest_group finds and returns the least busy CPU group within the | |
2187 | * domain. | |
2188 | */ | |
2189 | static struct sched_group * | |
2190 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2191 | { | |
2192 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2193 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2194 | int load_idx = sd->forkexec_idx; | |
2195 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2196 | ||
2197 | do { | |
2198 | unsigned long load, avg_load; | |
2199 | int local_group; | |
2200 | int i; | |
2201 | ||
da5a5522 | 2202 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2203 | if (!cpumask_intersects(sched_group_cpus(group), |
2204 | &p->cpus_allowed)) | |
3a5c359a | 2205 | continue; |
da5a5522 | 2206 | |
758b2cdc RR |
2207 | local_group = cpumask_test_cpu(this_cpu, |
2208 | sched_group_cpus(group)); | |
147cbb4b NP |
2209 | |
2210 | /* Tally up the load of all CPUs in the group */ | |
2211 | avg_load = 0; | |
2212 | ||
758b2cdc | 2213 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2214 | /* Bias balancing toward cpus of our domain */ |
2215 | if (local_group) | |
2216 | load = source_load(i, load_idx); | |
2217 | else | |
2218 | load = target_load(i, load_idx); | |
2219 | ||
2220 | avg_load += load; | |
2221 | } | |
2222 | ||
2223 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2224 | avg_load = sg_div_cpu_power(group, |
2225 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2226 | |
2227 | if (local_group) { | |
2228 | this_load = avg_load; | |
2229 | this = group; | |
2230 | } else if (avg_load < min_load) { | |
2231 | min_load = avg_load; | |
2232 | idlest = group; | |
2233 | } | |
3a5c359a | 2234 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2235 | |
2236 | if (!idlest || 100*this_load < imbalance*min_load) | |
2237 | return NULL; | |
2238 | return idlest; | |
2239 | } | |
2240 | ||
2241 | /* | |
0feaece9 | 2242 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2243 | */ |
95cdf3b7 | 2244 | static int |
758b2cdc | 2245 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2246 | { |
2247 | unsigned long load, min_load = ULONG_MAX; | |
2248 | int idlest = -1; | |
2249 | int i; | |
2250 | ||
da5a5522 | 2251 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2252 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2253 | load = weighted_cpuload(i); |
147cbb4b NP |
2254 | |
2255 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2256 | min_load = load; | |
2257 | idlest = i; | |
2258 | } | |
2259 | } | |
2260 | ||
2261 | return idlest; | |
2262 | } | |
2263 | ||
476d139c NP |
2264 | /* |
2265 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2266 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2267 | * SD_BALANCE_EXEC. | |
2268 | * | |
2269 | * Balance, ie. select the least loaded group. | |
2270 | * | |
2271 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2272 | * | |
2273 | * preempt must be disabled. | |
2274 | */ | |
2275 | static int sched_balance_self(int cpu, int flag) | |
2276 | { | |
2277 | struct task_struct *t = current; | |
2278 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2279 | |
c96d145e | 2280 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2281 | /* |
2282 | * If power savings logic is enabled for a domain, stop there. | |
2283 | */ | |
5c45bf27 SS |
2284 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2285 | break; | |
476d139c NP |
2286 | if (tmp->flags & flag) |
2287 | sd = tmp; | |
c96d145e | 2288 | } |
476d139c | 2289 | |
039a1c41 PZ |
2290 | if (sd) |
2291 | update_shares(sd); | |
2292 | ||
476d139c | 2293 | while (sd) { |
476d139c | 2294 | struct sched_group *group; |
1a848870 SS |
2295 | int new_cpu, weight; |
2296 | ||
2297 | if (!(sd->flags & flag)) { | |
2298 | sd = sd->child; | |
2299 | continue; | |
2300 | } | |
476d139c | 2301 | |
476d139c | 2302 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2303 | if (!group) { |
2304 | sd = sd->child; | |
2305 | continue; | |
2306 | } | |
476d139c | 2307 | |
758b2cdc | 2308 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2309 | if (new_cpu == -1 || new_cpu == cpu) { |
2310 | /* Now try balancing at a lower domain level of cpu */ | |
2311 | sd = sd->child; | |
2312 | continue; | |
2313 | } | |
476d139c | 2314 | |
1a848870 | 2315 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2316 | cpu = new_cpu; |
758b2cdc | 2317 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2318 | sd = NULL; |
476d139c | 2319 | for_each_domain(cpu, tmp) { |
758b2cdc | 2320 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2321 | break; |
2322 | if (tmp->flags & flag) | |
2323 | sd = tmp; | |
2324 | } | |
2325 | /* while loop will break here if sd == NULL */ | |
2326 | } | |
2327 | ||
2328 | return cpu; | |
2329 | } | |
2330 | ||
2331 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2332 | |
0793a61d TG |
2333 | /** |
2334 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2335 | * @p: the task to evaluate | |
2336 | * @func: the function to be called | |
2337 | * @info: the function call argument | |
2338 | * | |
2339 | * Calls the function @func when the task is currently running. This might | |
2340 | * be on the current CPU, which just calls the function directly | |
2341 | */ | |
2342 | void task_oncpu_function_call(struct task_struct *p, | |
2343 | void (*func) (void *info), void *info) | |
2344 | { | |
2345 | int cpu; | |
2346 | ||
2347 | preempt_disable(); | |
2348 | cpu = task_cpu(p); | |
2349 | if (task_curr(p)) | |
2350 | smp_call_function_single(cpu, func, info, 1); | |
2351 | preempt_enable(); | |
2352 | } | |
2353 | ||
1da177e4 LT |
2354 | /*** |
2355 | * try_to_wake_up - wake up a thread | |
2356 | * @p: the to-be-woken-up thread | |
2357 | * @state: the mask of task states that can be woken | |
2358 | * @sync: do a synchronous wakeup? | |
2359 | * | |
2360 | * Put it on the run-queue if it's not already there. The "current" | |
2361 | * thread is always on the run-queue (except when the actual | |
2362 | * re-schedule is in progress), and as such you're allowed to do | |
2363 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2364 | * runnable without the overhead of this. | |
2365 | * | |
2366 | * returns failure only if the task is already active. | |
2367 | */ | |
36c8b586 | 2368 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2369 | { |
cc367732 | 2370 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2371 | unsigned long flags; |
2372 | long old_state; | |
70b97a7f | 2373 | struct rq *rq; |
1da177e4 | 2374 | |
b85d0667 IM |
2375 | if (!sched_feat(SYNC_WAKEUPS)) |
2376 | sync = 0; | |
2377 | ||
2398f2c6 | 2378 | #ifdef CONFIG_SMP |
57310a98 | 2379 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2380 | struct sched_domain *sd; |
2381 | ||
2382 | this_cpu = raw_smp_processor_id(); | |
2383 | cpu = task_cpu(p); | |
2384 | ||
2385 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2386 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2387 | update_shares(sd); |
2388 | break; | |
2389 | } | |
2390 | } | |
2391 | } | |
2392 | #endif | |
2393 | ||
04e2f174 | 2394 | smp_wmb(); |
1da177e4 | 2395 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2396 | update_rq_clock(rq); |
1da177e4 LT |
2397 | old_state = p->state; |
2398 | if (!(old_state & state)) | |
2399 | goto out; | |
2400 | ||
dd41f596 | 2401 | if (p->se.on_rq) |
1da177e4 LT |
2402 | goto out_running; |
2403 | ||
2404 | cpu = task_cpu(p); | |
cc367732 | 2405 | orig_cpu = cpu; |
1da177e4 LT |
2406 | this_cpu = smp_processor_id(); |
2407 | ||
2408 | #ifdef CONFIG_SMP | |
2409 | if (unlikely(task_running(rq, p))) | |
2410 | goto out_activate; | |
2411 | ||
5d2f5a61 DA |
2412 | cpu = p->sched_class->select_task_rq(p, sync); |
2413 | if (cpu != orig_cpu) { | |
2414 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2415 | task_rq_unlock(rq, &flags); |
2416 | /* might preempt at this point */ | |
2417 | rq = task_rq_lock(p, &flags); | |
2418 | old_state = p->state; | |
2419 | if (!(old_state & state)) | |
2420 | goto out; | |
dd41f596 | 2421 | if (p->se.on_rq) |
1da177e4 LT |
2422 | goto out_running; |
2423 | ||
2424 | this_cpu = smp_processor_id(); | |
2425 | cpu = task_cpu(p); | |
2426 | } | |
2427 | ||
e7693a36 GH |
2428 | #ifdef CONFIG_SCHEDSTATS |
2429 | schedstat_inc(rq, ttwu_count); | |
2430 | if (cpu == this_cpu) | |
2431 | schedstat_inc(rq, ttwu_local); | |
2432 | else { | |
2433 | struct sched_domain *sd; | |
2434 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2435 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2436 | schedstat_inc(sd, ttwu_wake_remote); |
2437 | break; | |
2438 | } | |
2439 | } | |
2440 | } | |
6d6bc0ad | 2441 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2442 | |
1da177e4 LT |
2443 | out_activate: |
2444 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2445 | schedstat_inc(p, se.nr_wakeups); |
2446 | if (sync) | |
2447 | schedstat_inc(p, se.nr_wakeups_sync); | |
2448 | if (orig_cpu != cpu) | |
2449 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2450 | if (cpu == this_cpu) | |
2451 | schedstat_inc(p, se.nr_wakeups_local); | |
2452 | else | |
2453 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2454 | activate_task(rq, p, 1); |
1da177e4 LT |
2455 | success = 1; |
2456 | ||
831451ac PZ |
2457 | /* |
2458 | * Only attribute actual wakeups done by this task. | |
2459 | */ | |
2460 | if (!in_interrupt()) { | |
2461 | struct sched_entity *se = ¤t->se; | |
2462 | u64 sample = se->sum_exec_runtime; | |
2463 | ||
2464 | if (se->last_wakeup) | |
2465 | sample -= se->last_wakeup; | |
2466 | else | |
2467 | sample -= se->start_runtime; | |
2468 | update_avg(&se->avg_wakeup, sample); | |
2469 | ||
2470 | se->last_wakeup = se->sum_exec_runtime; | |
2471 | } | |
2472 | ||
1da177e4 | 2473 | out_running: |
468a15bb | 2474 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2475 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2476 | |
1da177e4 | 2477 | p->state = TASK_RUNNING; |
9a897c5a SR |
2478 | #ifdef CONFIG_SMP |
2479 | if (p->sched_class->task_wake_up) | |
2480 | p->sched_class->task_wake_up(rq, p); | |
2481 | #endif | |
1da177e4 LT |
2482 | out: |
2483 | task_rq_unlock(rq, &flags); | |
2484 | ||
2485 | return success; | |
2486 | } | |
2487 | ||
7ad5b3a5 | 2488 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2489 | { |
d9514f6c | 2490 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2491 | } |
1da177e4 LT |
2492 | EXPORT_SYMBOL(wake_up_process); |
2493 | ||
7ad5b3a5 | 2494 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2495 | { |
2496 | return try_to_wake_up(p, state, 0); | |
2497 | } | |
2498 | ||
1da177e4 LT |
2499 | /* |
2500 | * Perform scheduler related setup for a newly forked process p. | |
2501 | * p is forked by current. | |
dd41f596 IM |
2502 | * |
2503 | * __sched_fork() is basic setup used by init_idle() too: | |
2504 | */ | |
2505 | static void __sched_fork(struct task_struct *p) | |
2506 | { | |
dd41f596 IM |
2507 | p->se.exec_start = 0; |
2508 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2509 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2510 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2511 | p->se.last_wakeup = 0; |
2512 | p->se.avg_overlap = 0; | |
831451ac PZ |
2513 | p->se.start_runtime = 0; |
2514 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2515 | |
2516 | #ifdef CONFIG_SCHEDSTATS | |
2517 | p->se.wait_start = 0; | |
dd41f596 IM |
2518 | p->se.sum_sleep_runtime = 0; |
2519 | p->se.sleep_start = 0; | |
dd41f596 IM |
2520 | p->se.block_start = 0; |
2521 | p->se.sleep_max = 0; | |
2522 | p->se.block_max = 0; | |
2523 | p->se.exec_max = 0; | |
eba1ed4b | 2524 | p->se.slice_max = 0; |
dd41f596 | 2525 | p->se.wait_max = 0; |
6cfb0d5d | 2526 | #endif |
476d139c | 2527 | |
fa717060 | 2528 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2529 | p->se.on_rq = 0; |
4a55bd5e | 2530 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2531 | |
e107be36 AK |
2532 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2533 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2534 | #endif | |
2535 | ||
1da177e4 LT |
2536 | /* |
2537 | * We mark the process as running here, but have not actually | |
2538 | * inserted it onto the runqueue yet. This guarantees that | |
2539 | * nobody will actually run it, and a signal or other external | |
2540 | * event cannot wake it up and insert it on the runqueue either. | |
2541 | */ | |
2542 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2543 | } |
2544 | ||
2545 | /* | |
2546 | * fork()/clone()-time setup: | |
2547 | */ | |
2548 | void sched_fork(struct task_struct *p, int clone_flags) | |
2549 | { | |
2550 | int cpu = get_cpu(); | |
2551 | ||
2552 | __sched_fork(p); | |
2553 | ||
2554 | #ifdef CONFIG_SMP | |
2555 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2556 | #endif | |
02e4bac2 | 2557 | set_task_cpu(p, cpu); |
b29739f9 IM |
2558 | |
2559 | /* | |
2560 | * Make sure we do not leak PI boosting priority to the child: | |
2561 | */ | |
2562 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2563 | if (!rt_prio(p->prio)) |
2564 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2565 | |
52f17b6c | 2566 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2567 | if (likely(sched_info_on())) |
52f17b6c | 2568 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2569 | #endif |
d6077cb8 | 2570 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2571 | p->oncpu = 0; |
2572 | #endif | |
1da177e4 | 2573 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2574 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2575 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2576 | #endif |
917b627d GH |
2577 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2578 | ||
476d139c | 2579 | put_cpu(); |
1da177e4 LT |
2580 | } |
2581 | ||
2582 | /* | |
2583 | * wake_up_new_task - wake up a newly created task for the first time. | |
2584 | * | |
2585 | * This function will do some initial scheduler statistics housekeeping | |
2586 | * that must be done for every newly created context, then puts the task | |
2587 | * on the runqueue and wakes it. | |
2588 | */ | |
7ad5b3a5 | 2589 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2590 | { |
2591 | unsigned long flags; | |
dd41f596 | 2592 | struct rq *rq; |
1da177e4 LT |
2593 | |
2594 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2595 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2596 | update_rq_clock(rq); |
1da177e4 LT |
2597 | |
2598 | p->prio = effective_prio(p); | |
2599 | ||
b9dca1e0 | 2600 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2601 | activate_task(rq, p, 0); |
1da177e4 | 2602 | } else { |
1da177e4 | 2603 | /* |
dd41f596 IM |
2604 | * Let the scheduling class do new task startup |
2605 | * management (if any): | |
1da177e4 | 2606 | */ |
ee0827d8 | 2607 | p->sched_class->task_new(rq, p); |
c09595f6 | 2608 | inc_nr_running(rq); |
1da177e4 | 2609 | } |
c71dd42d | 2610 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2611 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2612 | #ifdef CONFIG_SMP |
2613 | if (p->sched_class->task_wake_up) | |
2614 | p->sched_class->task_wake_up(rq, p); | |
2615 | #endif | |
dd41f596 | 2616 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2617 | } |
2618 | ||
e107be36 AK |
2619 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2620 | ||
2621 | /** | |
80dd99b3 | 2622 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2623 | * @notifier: notifier struct to register |
e107be36 AK |
2624 | */ |
2625 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2626 | { | |
2627 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2628 | } | |
2629 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2630 | ||
2631 | /** | |
2632 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2633 | * @notifier: notifier struct to unregister |
e107be36 AK |
2634 | * |
2635 | * This is safe to call from within a preemption notifier. | |
2636 | */ | |
2637 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2638 | { | |
2639 | hlist_del(¬ifier->link); | |
2640 | } | |
2641 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2642 | ||
2643 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2644 | { | |
2645 | struct preempt_notifier *notifier; | |
2646 | struct hlist_node *node; | |
2647 | ||
2648 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2649 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2650 | } | |
2651 | ||
2652 | static void | |
2653 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2654 | struct task_struct *next) | |
2655 | { | |
2656 | struct preempt_notifier *notifier; | |
2657 | struct hlist_node *node; | |
2658 | ||
2659 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2660 | notifier->ops->sched_out(notifier, next); | |
2661 | } | |
2662 | ||
6d6bc0ad | 2663 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2664 | |
2665 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2666 | { | |
2667 | } | |
2668 | ||
2669 | static void | |
2670 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2671 | struct task_struct *next) | |
2672 | { | |
2673 | } | |
2674 | ||
6d6bc0ad | 2675 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2676 | |
4866cde0 NP |
2677 | /** |
2678 | * prepare_task_switch - prepare to switch tasks | |
2679 | * @rq: the runqueue preparing to switch | |
421cee29 | 2680 | * @prev: the current task that is being switched out |
4866cde0 NP |
2681 | * @next: the task we are going to switch to. |
2682 | * | |
2683 | * This is called with the rq lock held and interrupts off. It must | |
2684 | * be paired with a subsequent finish_task_switch after the context | |
2685 | * switch. | |
2686 | * | |
2687 | * prepare_task_switch sets up locking and calls architecture specific | |
2688 | * hooks. | |
2689 | */ | |
e107be36 AK |
2690 | static inline void |
2691 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2692 | struct task_struct *next) | |
4866cde0 | 2693 | { |
e107be36 | 2694 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2695 | prepare_lock_switch(rq, next); |
2696 | prepare_arch_switch(next); | |
2697 | } | |
2698 | ||
1da177e4 LT |
2699 | /** |
2700 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2701 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2702 | * @prev: the thread we just switched away from. |
2703 | * | |
4866cde0 NP |
2704 | * finish_task_switch must be called after the context switch, paired |
2705 | * with a prepare_task_switch call before the context switch. | |
2706 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2707 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2708 | * |
2709 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2710 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2711 | * with the lock held can cause deadlocks; see schedule() for |
2712 | * details.) | |
2713 | */ | |
a9957449 | 2714 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2715 | __releases(rq->lock) |
2716 | { | |
1da177e4 | 2717 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2718 | long prev_state; |
967fc046 GH |
2719 | #ifdef CONFIG_SMP |
2720 | int post_schedule = 0; | |
2721 | ||
2722 | if (current->sched_class->needs_post_schedule) | |
2723 | post_schedule = current->sched_class->needs_post_schedule(rq); | |
2724 | #endif | |
1da177e4 LT |
2725 | |
2726 | rq->prev_mm = NULL; | |
2727 | ||
2728 | /* | |
2729 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2730 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2731 | * schedule one last time. The schedule call will never return, and |
2732 | * the scheduled task must drop that reference. | |
c394cc9f | 2733 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2734 | * still held, otherwise prev could be scheduled on another cpu, die |
2735 | * there before we look at prev->state, and then the reference would | |
2736 | * be dropped twice. | |
2737 | * Manfred Spraul <manfred@colorfullife.com> | |
2738 | */ | |
55a101f8 | 2739 | prev_state = prev->state; |
4866cde0 | 2740 | finish_arch_switch(prev); |
0793a61d | 2741 | perf_counter_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2742 | finish_lock_switch(rq, prev); |
9a897c5a | 2743 | #ifdef CONFIG_SMP |
967fc046 | 2744 | if (post_schedule) |
9a897c5a SR |
2745 | current->sched_class->post_schedule(rq); |
2746 | #endif | |
e8fa1362 | 2747 | |
e107be36 | 2748 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2749 | if (mm) |
2750 | mmdrop(mm); | |
c394cc9f | 2751 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2752 | /* |
2753 | * Remove function-return probe instances associated with this | |
2754 | * task and put them back on the free list. | |
9761eea8 | 2755 | */ |
c6fd91f0 | 2756 | kprobe_flush_task(prev); |
1da177e4 | 2757 | put_task_struct(prev); |
c6fd91f0 | 2758 | } |
1da177e4 LT |
2759 | } |
2760 | ||
2761 | /** | |
2762 | * schedule_tail - first thing a freshly forked thread must call. | |
2763 | * @prev: the thread we just switched away from. | |
2764 | */ | |
36c8b586 | 2765 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2766 | __releases(rq->lock) |
2767 | { | |
70b97a7f IM |
2768 | struct rq *rq = this_rq(); |
2769 | ||
4866cde0 NP |
2770 | finish_task_switch(rq, prev); |
2771 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2772 | /* In this case, finish_task_switch does not reenable preemption */ | |
2773 | preempt_enable(); | |
2774 | #endif | |
1da177e4 | 2775 | if (current->set_child_tid) |
b488893a | 2776 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2777 | } |
2778 | ||
2779 | /* | |
2780 | * context_switch - switch to the new MM and the new | |
2781 | * thread's register state. | |
2782 | */ | |
dd41f596 | 2783 | static inline void |
70b97a7f | 2784 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2785 | struct task_struct *next) |
1da177e4 | 2786 | { |
dd41f596 | 2787 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2788 | |
e107be36 | 2789 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2790 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2791 | mm = next->mm; |
2792 | oldmm = prev->active_mm; | |
9226d125 ZA |
2793 | /* |
2794 | * For paravirt, this is coupled with an exit in switch_to to | |
2795 | * combine the page table reload and the switch backend into | |
2796 | * one hypercall. | |
2797 | */ | |
2798 | arch_enter_lazy_cpu_mode(); | |
2799 | ||
dd41f596 | 2800 | if (unlikely(!mm)) { |
1da177e4 LT |
2801 | next->active_mm = oldmm; |
2802 | atomic_inc(&oldmm->mm_count); | |
2803 | enter_lazy_tlb(oldmm, next); | |
2804 | } else | |
2805 | switch_mm(oldmm, mm, next); | |
2806 | ||
dd41f596 | 2807 | if (unlikely(!prev->mm)) { |
1da177e4 | 2808 | prev->active_mm = NULL; |
1da177e4 LT |
2809 | rq->prev_mm = oldmm; |
2810 | } | |
3a5f5e48 IM |
2811 | /* |
2812 | * Since the runqueue lock will be released by the next | |
2813 | * task (which is an invalid locking op but in the case | |
2814 | * of the scheduler it's an obvious special-case), so we | |
2815 | * do an early lockdep release here: | |
2816 | */ | |
2817 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2818 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2819 | #endif |
1da177e4 LT |
2820 | |
2821 | /* Here we just switch the register state and the stack. */ | |
2822 | switch_to(prev, next, prev); | |
2823 | ||
dd41f596 IM |
2824 | barrier(); |
2825 | /* | |
2826 | * this_rq must be evaluated again because prev may have moved | |
2827 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2828 | * frame will be invalid. | |
2829 | */ | |
2830 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2831 | } |
2832 | ||
2833 | /* | |
2834 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2835 | * | |
2836 | * externally visible scheduler statistics: current number of runnable | |
2837 | * threads, current number of uninterruptible-sleeping threads, total | |
2838 | * number of context switches performed since bootup. | |
2839 | */ | |
2840 | unsigned long nr_running(void) | |
2841 | { | |
2842 | unsigned long i, sum = 0; | |
2843 | ||
2844 | for_each_online_cpu(i) | |
2845 | sum += cpu_rq(i)->nr_running; | |
2846 | ||
2847 | return sum; | |
2848 | } | |
2849 | ||
2850 | unsigned long nr_uninterruptible(void) | |
2851 | { | |
2852 | unsigned long i, sum = 0; | |
2853 | ||
0a945022 | 2854 | for_each_possible_cpu(i) |
1da177e4 LT |
2855 | sum += cpu_rq(i)->nr_uninterruptible; |
2856 | ||
2857 | /* | |
2858 | * Since we read the counters lockless, it might be slightly | |
2859 | * inaccurate. Do not allow it to go below zero though: | |
2860 | */ | |
2861 | if (unlikely((long)sum < 0)) | |
2862 | sum = 0; | |
2863 | ||
2864 | return sum; | |
2865 | } | |
2866 | ||
2867 | unsigned long long nr_context_switches(void) | |
2868 | { | |
cc94abfc SR |
2869 | int i; |
2870 | unsigned long long sum = 0; | |
1da177e4 | 2871 | |
0a945022 | 2872 | for_each_possible_cpu(i) |
1da177e4 LT |
2873 | sum += cpu_rq(i)->nr_switches; |
2874 | ||
2875 | return sum; | |
2876 | } | |
2877 | ||
2878 | unsigned long nr_iowait(void) | |
2879 | { | |
2880 | unsigned long i, sum = 0; | |
2881 | ||
0a945022 | 2882 | for_each_possible_cpu(i) |
1da177e4 LT |
2883 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2884 | ||
2885 | return sum; | |
2886 | } | |
2887 | ||
db1b1fef JS |
2888 | unsigned long nr_active(void) |
2889 | { | |
2890 | unsigned long i, running = 0, uninterruptible = 0; | |
2891 | ||
2892 | for_each_online_cpu(i) { | |
2893 | running += cpu_rq(i)->nr_running; | |
2894 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2895 | } | |
2896 | ||
2897 | if (unlikely((long)uninterruptible < 0)) | |
2898 | uninterruptible = 0; | |
2899 | ||
2900 | return running + uninterruptible; | |
2901 | } | |
2902 | ||
23a185ca PM |
2903 | /* |
2904 | * Externally visible per-cpu scheduler statistics: | |
23a185ca PM |
2905 | * cpu_nr_migrations(cpu) - number of migrations into that cpu |
2906 | */ | |
23a185ca PM |
2907 | u64 cpu_nr_migrations(int cpu) |
2908 | { | |
2909 | return cpu_rq(cpu)->nr_migrations_in; | |
2910 | } | |
2911 | ||
48f24c4d | 2912 | /* |
dd41f596 IM |
2913 | * Update rq->cpu_load[] statistics. This function is usually called every |
2914 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2915 | */ |
dd41f596 | 2916 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2917 | { |
495eca49 | 2918 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2919 | int i, scale; |
2920 | ||
2921 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2922 | |
2923 | /* Update our load: */ | |
2924 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2925 | unsigned long old_load, new_load; | |
2926 | ||
2927 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2928 | ||
2929 | old_load = this_rq->cpu_load[i]; | |
2930 | new_load = this_load; | |
a25707f3 IM |
2931 | /* |
2932 | * Round up the averaging division if load is increasing. This | |
2933 | * prevents us from getting stuck on 9 if the load is 10, for | |
2934 | * example. | |
2935 | */ | |
2936 | if (new_load > old_load) | |
2937 | new_load += scale-1; | |
dd41f596 IM |
2938 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2939 | } | |
48f24c4d IM |
2940 | } |
2941 | ||
dd41f596 IM |
2942 | #ifdef CONFIG_SMP |
2943 | ||
1da177e4 LT |
2944 | /* |
2945 | * double_rq_lock - safely lock two runqueues | |
2946 | * | |
2947 | * Note this does not disable interrupts like task_rq_lock, | |
2948 | * you need to do so manually before calling. | |
2949 | */ | |
70b97a7f | 2950 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2951 | __acquires(rq1->lock) |
2952 | __acquires(rq2->lock) | |
2953 | { | |
054b9108 | 2954 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2955 | if (rq1 == rq2) { |
2956 | spin_lock(&rq1->lock); | |
2957 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2958 | } else { | |
c96d145e | 2959 | if (rq1 < rq2) { |
1da177e4 | 2960 | spin_lock(&rq1->lock); |
5e710e37 | 2961 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2962 | } else { |
2963 | spin_lock(&rq2->lock); | |
5e710e37 | 2964 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
2965 | } |
2966 | } | |
6e82a3be IM |
2967 | update_rq_clock(rq1); |
2968 | update_rq_clock(rq2); | |
1da177e4 LT |
2969 | } |
2970 | ||
2971 | /* | |
2972 | * double_rq_unlock - safely unlock two runqueues | |
2973 | * | |
2974 | * Note this does not restore interrupts like task_rq_unlock, | |
2975 | * you need to do so manually after calling. | |
2976 | */ | |
70b97a7f | 2977 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2978 | __releases(rq1->lock) |
2979 | __releases(rq2->lock) | |
2980 | { | |
2981 | spin_unlock(&rq1->lock); | |
2982 | if (rq1 != rq2) | |
2983 | spin_unlock(&rq2->lock); | |
2984 | else | |
2985 | __release(rq2->lock); | |
2986 | } | |
2987 | ||
1da177e4 LT |
2988 | /* |
2989 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2990 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2991 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2992 | * the cpu_allowed mask is restored. |
2993 | */ | |
36c8b586 | 2994 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2995 | { |
70b97a7f | 2996 | struct migration_req req; |
1da177e4 | 2997 | unsigned long flags; |
70b97a7f | 2998 | struct rq *rq; |
1da177e4 LT |
2999 | |
3000 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3001 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3002 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3003 | goto out; |
3004 | ||
3005 | /* force the process onto the specified CPU */ | |
3006 | if (migrate_task(p, dest_cpu, &req)) { | |
3007 | /* Need to wait for migration thread (might exit: take ref). */ | |
3008 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3009 | |
1da177e4 LT |
3010 | get_task_struct(mt); |
3011 | task_rq_unlock(rq, &flags); | |
3012 | wake_up_process(mt); | |
3013 | put_task_struct(mt); | |
3014 | wait_for_completion(&req.done); | |
36c8b586 | 3015 | |
1da177e4 LT |
3016 | return; |
3017 | } | |
3018 | out: | |
3019 | task_rq_unlock(rq, &flags); | |
3020 | } | |
3021 | ||
3022 | /* | |
476d139c NP |
3023 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3024 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3025 | */ |
3026 | void sched_exec(void) | |
3027 | { | |
1da177e4 | 3028 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3029 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3030 | put_cpu(); |
476d139c NP |
3031 | if (new_cpu != this_cpu) |
3032 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3033 | } |
3034 | ||
3035 | /* | |
3036 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3037 | * Both runqueues must be locked. | |
3038 | */ | |
dd41f596 IM |
3039 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3040 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3041 | { |
2e1cb74a | 3042 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3043 | set_task_cpu(p, this_cpu); |
dd41f596 | 3044 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3045 | /* |
3046 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3047 | * to be always true for them. | |
3048 | */ | |
15afe09b | 3049 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3050 | } |
3051 | ||
3052 | /* | |
3053 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3054 | */ | |
858119e1 | 3055 | static |
70b97a7f | 3056 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3057 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3058 | int *all_pinned) |
1da177e4 | 3059 | { |
708dc512 | 3060 | int tsk_cache_hot = 0; |
1da177e4 LT |
3061 | /* |
3062 | * We do not migrate tasks that are: | |
3063 | * 1) running (obviously), or | |
3064 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3065 | * 3) are cache-hot on their current CPU. | |
3066 | */ | |
96f874e2 | 3067 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3068 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3069 | return 0; |
cc367732 | 3070 | } |
81026794 NP |
3071 | *all_pinned = 0; |
3072 | ||
cc367732 IM |
3073 | if (task_running(rq, p)) { |
3074 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3075 | return 0; |
cc367732 | 3076 | } |
1da177e4 | 3077 | |
da84d961 IM |
3078 | /* |
3079 | * Aggressive migration if: | |
3080 | * 1) task is cache cold, or | |
3081 | * 2) too many balance attempts have failed. | |
3082 | */ | |
3083 | ||
708dc512 LH |
3084 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3085 | if (!tsk_cache_hot || | |
3086 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3087 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3088 | if (tsk_cache_hot) { |
da84d961 | 3089 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3090 | schedstat_inc(p, se.nr_forced_migrations); |
3091 | } | |
da84d961 IM |
3092 | #endif |
3093 | return 1; | |
3094 | } | |
3095 | ||
708dc512 | 3096 | if (tsk_cache_hot) { |
cc367732 | 3097 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3098 | return 0; |
cc367732 | 3099 | } |
1da177e4 LT |
3100 | return 1; |
3101 | } | |
3102 | ||
e1d1484f PW |
3103 | static unsigned long |
3104 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3105 | unsigned long max_load_move, struct sched_domain *sd, | |
3106 | enum cpu_idle_type idle, int *all_pinned, | |
3107 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3108 | { |
051c6764 | 3109 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3110 | struct task_struct *p; |
3111 | long rem_load_move = max_load_move; | |
1da177e4 | 3112 | |
e1d1484f | 3113 | if (max_load_move == 0) |
1da177e4 LT |
3114 | goto out; |
3115 | ||
81026794 NP |
3116 | pinned = 1; |
3117 | ||
1da177e4 | 3118 | /* |
dd41f596 | 3119 | * Start the load-balancing iterator: |
1da177e4 | 3120 | */ |
dd41f596 IM |
3121 | p = iterator->start(iterator->arg); |
3122 | next: | |
b82d9fdd | 3123 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3124 | goto out; |
051c6764 PZ |
3125 | |
3126 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3127 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3128 | p = iterator->next(iterator->arg); |
3129 | goto next; | |
1da177e4 LT |
3130 | } |
3131 | ||
dd41f596 | 3132 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3133 | pulled++; |
dd41f596 | 3134 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3135 | |
7e96fa58 GH |
3136 | #ifdef CONFIG_PREEMPT |
3137 | /* | |
3138 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3139 | * will stop after the first task is pulled to minimize the critical | |
3140 | * section. | |
3141 | */ | |
3142 | if (idle == CPU_NEWLY_IDLE) | |
3143 | goto out; | |
3144 | #endif | |
3145 | ||
2dd73a4f | 3146 | /* |
b82d9fdd | 3147 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3148 | */ |
e1d1484f | 3149 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3150 | if (p->prio < *this_best_prio) |
3151 | *this_best_prio = p->prio; | |
dd41f596 IM |
3152 | p = iterator->next(iterator->arg); |
3153 | goto next; | |
1da177e4 LT |
3154 | } |
3155 | out: | |
3156 | /* | |
e1d1484f | 3157 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3158 | * so we can safely collect pull_task() stats here rather than |
3159 | * inside pull_task(). | |
3160 | */ | |
3161 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3162 | |
3163 | if (all_pinned) | |
3164 | *all_pinned = pinned; | |
e1d1484f PW |
3165 | |
3166 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3167 | } |
3168 | ||
dd41f596 | 3169 | /* |
43010659 PW |
3170 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3171 | * this_rq, as part of a balancing operation within domain "sd". | |
3172 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3173 | * |
3174 | * Called with both runqueues locked. | |
3175 | */ | |
3176 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3177 | unsigned long max_load_move, |
dd41f596 IM |
3178 | struct sched_domain *sd, enum cpu_idle_type idle, |
3179 | int *all_pinned) | |
3180 | { | |
5522d5d5 | 3181 | const struct sched_class *class = sched_class_highest; |
43010659 | 3182 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3183 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3184 | |
3185 | do { | |
43010659 PW |
3186 | total_load_moved += |
3187 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3188 | max_load_move - total_load_moved, |
a4ac01c3 | 3189 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3190 | class = class->next; |
c4acb2c0 | 3191 | |
7e96fa58 GH |
3192 | #ifdef CONFIG_PREEMPT |
3193 | /* | |
3194 | * NEWIDLE balancing is a source of latency, so preemptible | |
3195 | * kernels will stop after the first task is pulled to minimize | |
3196 | * the critical section. | |
3197 | */ | |
c4acb2c0 GH |
3198 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3199 | break; | |
7e96fa58 | 3200 | #endif |
43010659 | 3201 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3202 | |
43010659 PW |
3203 | return total_load_moved > 0; |
3204 | } | |
3205 | ||
e1d1484f PW |
3206 | static int |
3207 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3208 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3209 | struct rq_iterator *iterator) | |
3210 | { | |
3211 | struct task_struct *p = iterator->start(iterator->arg); | |
3212 | int pinned = 0; | |
3213 | ||
3214 | while (p) { | |
3215 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3216 | pull_task(busiest, p, this_rq, this_cpu); | |
3217 | /* | |
3218 | * Right now, this is only the second place pull_task() | |
3219 | * is called, so we can safely collect pull_task() | |
3220 | * stats here rather than inside pull_task(). | |
3221 | */ | |
3222 | schedstat_inc(sd, lb_gained[idle]); | |
3223 | ||
3224 | return 1; | |
3225 | } | |
3226 | p = iterator->next(iterator->arg); | |
3227 | } | |
3228 | ||
3229 | return 0; | |
3230 | } | |
3231 | ||
43010659 PW |
3232 | /* |
3233 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3234 | * part of active balancing operations within "domain". | |
3235 | * Returns 1 if successful and 0 otherwise. | |
3236 | * | |
3237 | * Called with both runqueues locked. | |
3238 | */ | |
3239 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3240 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3241 | { | |
5522d5d5 | 3242 | const struct sched_class *class; |
43010659 PW |
3243 | |
3244 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 3245 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
3246 | return 1; |
3247 | ||
3248 | return 0; | |
dd41f596 | 3249 | } |
67bb6c03 | 3250 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3251 | /* |
222d656d GS |
3252 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3253 | * during load balancing. | |
1da177e4 | 3254 | */ |
222d656d GS |
3255 | struct sd_lb_stats { |
3256 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3257 | struct sched_group *this; /* Local group in this sd */ | |
3258 | unsigned long total_load; /* Total load of all groups in sd */ | |
3259 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3260 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3261 | ||
3262 | /** Statistics of this group */ | |
3263 | unsigned long this_load; | |
3264 | unsigned long this_load_per_task; | |
3265 | unsigned long this_nr_running; | |
3266 | ||
3267 | /* Statistics of the busiest group */ | |
3268 | unsigned long max_load; | |
3269 | unsigned long busiest_load_per_task; | |
3270 | unsigned long busiest_nr_running; | |
3271 | ||
3272 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3273 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3274 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3275 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3276 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3277 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3278 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3279 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3280 | #endif |
222d656d | 3281 | }; |
1da177e4 | 3282 | |
d5ac537e | 3283 | /* |
381be78f GS |
3284 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3285 | */ | |
3286 | struct sg_lb_stats { | |
3287 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3288 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3289 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3290 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3291 | unsigned long group_capacity; | |
3292 | int group_imb; /* Is there an imbalance in the group ? */ | |
3293 | }; | |
3294 | ||
67bb6c03 GS |
3295 | /** |
3296 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3297 | * @group: The group whose first cpu is to be returned. | |
3298 | */ | |
3299 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3300 | { | |
3301 | return cpumask_first(sched_group_cpus(group)); | |
3302 | } | |
3303 | ||
3304 | /** | |
3305 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3306 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3307 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3308 | */ | |
3309 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3310 | enum cpu_idle_type idle) | |
3311 | { | |
3312 | int load_idx; | |
408ed066 | 3313 | |
67bb6c03 GS |
3314 | switch (idle) { |
3315 | case CPU_NOT_IDLE: | |
7897986b | 3316 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3317 | break; |
3318 | ||
3319 | case CPU_NEWLY_IDLE: | |
7897986b | 3320 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3321 | break; |
3322 | default: | |
7897986b | 3323 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3324 | break; |
3325 | } | |
1da177e4 | 3326 | |
67bb6c03 GS |
3327 | return load_idx; |
3328 | } | |
1da177e4 | 3329 | |
1da177e4 | 3330 | |
c071df18 GS |
3331 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3332 | /** | |
3333 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3334 | * the given sched_domain, during load balancing. | |
3335 | * | |
3336 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3337 | * @sds: Variable containing the statistics for sd. | |
3338 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3339 | */ | |
3340 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3341 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3342 | { | |
3343 | /* | |
3344 | * Busy processors will not participate in power savings | |
3345 | * balance. | |
3346 | */ | |
3347 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3348 | sds->power_savings_balance = 0; | |
3349 | else { | |
3350 | sds->power_savings_balance = 1; | |
3351 | sds->min_nr_running = ULONG_MAX; | |
3352 | sds->leader_nr_running = 0; | |
3353 | } | |
3354 | } | |
783609c6 | 3355 | |
c071df18 GS |
3356 | /** |
3357 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3358 | * sched_domain while performing load balancing. | |
3359 | * | |
3360 | * @group: sched_group belonging to the sched_domain under consideration. | |
3361 | * @sds: Variable containing the statistics of the sched_domain | |
3362 | * @local_group: Does group contain the CPU for which we're performing | |
3363 | * load balancing ? | |
3364 | * @sgs: Variable containing the statistics of the group. | |
3365 | */ | |
3366 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3367 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3368 | { | |
408ed066 | 3369 | |
c071df18 GS |
3370 | if (!sds->power_savings_balance) |
3371 | return; | |
1da177e4 | 3372 | |
c071df18 GS |
3373 | /* |
3374 | * If the local group is idle or completely loaded | |
3375 | * no need to do power savings balance at this domain | |
3376 | */ | |
3377 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3378 | !sds->this_nr_running)) | |
3379 | sds->power_savings_balance = 0; | |
2dd73a4f | 3380 | |
c071df18 GS |
3381 | /* |
3382 | * If a group is already running at full capacity or idle, | |
3383 | * don't include that group in power savings calculations | |
3384 | */ | |
3385 | if (!sds->power_savings_balance || | |
3386 | sgs->sum_nr_running >= sgs->group_capacity || | |
3387 | !sgs->sum_nr_running) | |
3388 | return; | |
5969fe06 | 3389 | |
c071df18 GS |
3390 | /* |
3391 | * Calculate the group which has the least non-idle load. | |
3392 | * This is the group from where we need to pick up the load | |
3393 | * for saving power | |
3394 | */ | |
3395 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3396 | (sgs->sum_nr_running == sds->min_nr_running && | |
3397 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3398 | sds->group_min = group; | |
3399 | sds->min_nr_running = sgs->sum_nr_running; | |
3400 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3401 | sgs->sum_nr_running; | |
3402 | } | |
783609c6 | 3403 | |
c071df18 GS |
3404 | /* |
3405 | * Calculate the group which is almost near its | |
3406 | * capacity but still has some space to pick up some load | |
3407 | * from other group and save more power | |
3408 | */ | |
3409 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3410 | return; | |
1da177e4 | 3411 | |
c071df18 GS |
3412 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3413 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3414 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3415 | sds->group_leader = group; | |
3416 | sds->leader_nr_running = sgs->sum_nr_running; | |
3417 | } | |
3418 | } | |
408ed066 | 3419 | |
c071df18 | 3420 | /** |
d5ac537e | 3421 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3422 | * @sds: Variable containing the statistics of the sched_domain |
3423 | * under consideration. | |
3424 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3425 | * @imbalance: Variable to store the imbalance. | |
3426 | * | |
d5ac537e RD |
3427 | * Description: |
3428 | * Check if we have potential to perform some power-savings balance. | |
3429 | * If yes, set the busiest group to be the least loaded group in the | |
3430 | * sched_domain, so that it's CPUs can be put to idle. | |
3431 | * | |
c071df18 GS |
3432 | * Returns 1 if there is potential to perform power-savings balance. |
3433 | * Else returns 0. | |
3434 | */ | |
3435 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3436 | int this_cpu, unsigned long *imbalance) | |
3437 | { | |
3438 | if (!sds->power_savings_balance) | |
3439 | return 0; | |
1da177e4 | 3440 | |
c071df18 GS |
3441 | if (sds->this != sds->group_leader || |
3442 | sds->group_leader == sds->group_min) | |
3443 | return 0; | |
783609c6 | 3444 | |
c071df18 GS |
3445 | *imbalance = sds->min_load_per_task; |
3446 | sds->busiest = sds->group_min; | |
1da177e4 | 3447 | |
c071df18 GS |
3448 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3449 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3450 | group_first_cpu(sds->group_leader); | |
3451 | } | |
1da177e4 | 3452 | |
c071df18 | 3453 | return 1; |
408ed066 | 3454 | |
c071df18 GS |
3455 | } |
3456 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3457 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3458 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3459 | { | |
3460 | return; | |
3461 | } | |
3462 | ||
3463 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3464 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3465 | { | |
3466 | return; | |
3467 | } | |
3468 | ||
3469 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3470 | int this_cpu, unsigned long *imbalance) | |
3471 | { | |
3472 | return 0; | |
3473 | } | |
3474 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3475 | ||
3476 | ||
1f8c553d GS |
3477 | /** |
3478 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3479 | * @group: sched_group whose statistics are to be updated. | |
3480 | * @this_cpu: Cpu for which load balance is currently performed. | |
3481 | * @idle: Idle status of this_cpu | |
3482 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3483 | * @sd_idle: Idle status of the sched_domain containing group. | |
3484 | * @local_group: Does group contain this_cpu. | |
3485 | * @cpus: Set of cpus considered for load balancing. | |
3486 | * @balance: Should we balance. | |
3487 | * @sgs: variable to hold the statistics for this group. | |
3488 | */ | |
3489 | static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu, | |
3490 | enum cpu_idle_type idle, int load_idx, int *sd_idle, | |
3491 | int local_group, const struct cpumask *cpus, | |
3492 | int *balance, struct sg_lb_stats *sgs) | |
3493 | { | |
3494 | unsigned long load, max_cpu_load, min_cpu_load; | |
3495 | int i; | |
3496 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3497 | unsigned long sum_avg_load_per_task; | |
3498 | unsigned long avg_load_per_task; | |
3499 | ||
3500 | if (local_group) | |
3501 | balance_cpu = group_first_cpu(group); | |
408ed066 | 3502 | |
1f8c553d GS |
3503 | /* Tally up the load of all CPUs in the group */ |
3504 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3505 | max_cpu_load = 0; | |
3506 | min_cpu_load = ~0UL; | |
908a7c1b | 3507 | |
1f8c553d GS |
3508 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3509 | struct rq *rq = cpu_rq(i); | |
5c45bf27 | 3510 | |
1f8c553d GS |
3511 | if (*sd_idle && rq->nr_running) |
3512 | *sd_idle = 0; | |
3513 | ||
3514 | /* Bias balancing toward cpus of our domain */ | |
1da177e4 | 3515 | if (local_group) { |
1f8c553d GS |
3516 | if (idle_cpu(i) && !first_idle_cpu) { |
3517 | first_idle_cpu = 1; | |
3518 | balance_cpu = i; | |
3519 | } | |
3520 | ||
3521 | load = target_load(i, load_idx); | |
3522 | } else { | |
3523 | load = source_load(i, load_idx); | |
3524 | if (load > max_cpu_load) | |
3525 | max_cpu_load = load; | |
3526 | if (min_cpu_load > load) | |
3527 | min_cpu_load = load; | |
1da177e4 | 3528 | } |
5c45bf27 | 3529 | |
1f8c553d GS |
3530 | sgs->group_load += load; |
3531 | sgs->sum_nr_running += rq->nr_running; | |
3532 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3533 | |
1f8c553d GS |
3534 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3535 | } | |
5c45bf27 | 3536 | |
1f8c553d GS |
3537 | /* |
3538 | * First idle cpu or the first cpu(busiest) in this sched group | |
3539 | * is eligible for doing load balancing at this and above | |
3540 | * domains. In the newly idle case, we will allow all the cpu's | |
3541 | * to do the newly idle load balance. | |
3542 | */ | |
3543 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3544 | balance_cpu != this_cpu && balance) { | |
3545 | *balance = 0; | |
3546 | return; | |
3547 | } | |
5c45bf27 | 3548 | |
1f8c553d GS |
3549 | /* Adjust by relative CPU power of the group */ |
3550 | sgs->avg_load = sg_div_cpu_power(group, | |
3551 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3552 | |
1f8c553d GS |
3553 | |
3554 | /* | |
3555 | * Consider the group unbalanced when the imbalance is larger | |
3556 | * than the average weight of two tasks. | |
3557 | * | |
3558 | * APZ: with cgroup the avg task weight can vary wildly and | |
3559 | * might not be a suitable number - should we keep a | |
3560 | * normalized nr_running number somewhere that negates | |
3561 | * the hierarchy? | |
3562 | */ | |
3563 | avg_load_per_task = sg_div_cpu_power(group, | |
3564 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3565 | ||
3566 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3567 | sgs->group_imb = 1; | |
3568 | ||
3569 | sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; | |
3570 | ||
3571 | } | |
dd41f596 | 3572 | |
37abe198 GS |
3573 | /** |
3574 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3575 | * @sd: sched_domain whose statistics are to be updated. | |
3576 | * @this_cpu: Cpu for which load balance is currently performed. | |
3577 | * @idle: Idle status of this_cpu | |
3578 | * @sd_idle: Idle status of the sched_domain containing group. | |
3579 | * @cpus: Set of cpus considered for load balancing. | |
3580 | * @balance: Should we balance. | |
3581 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3582 | */ |
37abe198 GS |
3583 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3584 | enum cpu_idle_type idle, int *sd_idle, | |
3585 | const struct cpumask *cpus, int *balance, | |
3586 | struct sd_lb_stats *sds) | |
1da177e4 | 3587 | { |
222d656d | 3588 | struct sched_group *group = sd->groups; |
37abe198 | 3589 | struct sg_lb_stats sgs; |
222d656d GS |
3590 | int load_idx; |
3591 | ||
c071df18 | 3592 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3593 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3594 | |
3595 | do { | |
1da177e4 | 3596 | int local_group; |
1da177e4 | 3597 | |
758b2cdc RR |
3598 | local_group = cpumask_test_cpu(this_cpu, |
3599 | sched_group_cpus(group)); | |
381be78f | 3600 | memset(&sgs, 0, sizeof(sgs)); |
1f8c553d GS |
3601 | update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle, |
3602 | local_group, cpus, balance, &sgs); | |
1da177e4 | 3603 | |
37abe198 GS |
3604 | if (local_group && balance && !(*balance)) |
3605 | return; | |
783609c6 | 3606 | |
37abe198 GS |
3607 | sds->total_load += sgs.group_load; |
3608 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3609 | |
1da177e4 | 3610 | if (local_group) { |
37abe198 GS |
3611 | sds->this_load = sgs.avg_load; |
3612 | sds->this = group; | |
3613 | sds->this_nr_running = sgs.sum_nr_running; | |
3614 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3615 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3616 | (sgs.sum_nr_running > sgs.group_capacity || |
3617 | sgs.group_imb)) { | |
37abe198 GS |
3618 | sds->max_load = sgs.avg_load; |
3619 | sds->busiest = group; | |
3620 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3621 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3622 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3623 | } |
5c45bf27 | 3624 | |
c071df18 | 3625 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3626 | group = group->next; |
3627 | } while (group != sd->groups); | |
3628 | ||
37abe198 | 3629 | } |
1da177e4 | 3630 | |
2e6f44ae GS |
3631 | /** |
3632 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3633 | * amongst the groups of a sched_domain, during |
3634 | * load balancing. | |
2e6f44ae GS |
3635 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3636 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3637 | * @imbalance: Variable to store the imbalance. | |
3638 | */ | |
3639 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3640 | int this_cpu, unsigned long *imbalance) | |
3641 | { | |
3642 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3643 | unsigned int imbn = 2; | |
3644 | ||
3645 | if (sds->this_nr_running) { | |
3646 | sds->this_load_per_task /= sds->this_nr_running; | |
3647 | if (sds->busiest_load_per_task > | |
3648 | sds->this_load_per_task) | |
3649 | imbn = 1; | |
3650 | } else | |
3651 | sds->this_load_per_task = | |
3652 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 3653 | |
2e6f44ae GS |
3654 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
3655 | sds->busiest_load_per_task * imbn) { | |
3656 | *imbalance = sds->busiest_load_per_task; | |
3657 | return; | |
3658 | } | |
908a7c1b | 3659 | |
1da177e4 | 3660 | /* |
2e6f44ae GS |
3661 | * OK, we don't have enough imbalance to justify moving tasks, |
3662 | * however we may be able to increase total CPU power used by | |
3663 | * moving them. | |
1da177e4 | 3664 | */ |
2dd73a4f | 3665 | |
2e6f44ae GS |
3666 | pwr_now += sds->busiest->__cpu_power * |
3667 | min(sds->busiest_load_per_task, sds->max_load); | |
3668 | pwr_now += sds->this->__cpu_power * | |
3669 | min(sds->this_load_per_task, sds->this_load); | |
3670 | pwr_now /= SCHED_LOAD_SCALE; | |
3671 | ||
3672 | /* Amount of load we'd subtract */ | |
3673 | tmp = sg_div_cpu_power(sds->busiest, | |
3674 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3675 | if (sds->max_load > tmp) | |
3676 | pwr_move += sds->busiest->__cpu_power * | |
3677 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
3678 | ||
3679 | /* Amount of load we'd add */ | |
3680 | if (sds->max_load * sds->busiest->__cpu_power < | |
3681 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
3682 | tmp = sg_div_cpu_power(sds->this, | |
3683 | sds->max_load * sds->busiest->__cpu_power); | |
3684 | else | |
3685 | tmp = sg_div_cpu_power(sds->this, | |
3686 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
3687 | pwr_move += sds->this->__cpu_power * | |
3688 | min(sds->this_load_per_task, sds->this_load + tmp); | |
3689 | pwr_move /= SCHED_LOAD_SCALE; | |
3690 | ||
3691 | /* Move if we gain throughput */ | |
3692 | if (pwr_move > pwr_now) | |
3693 | *imbalance = sds->busiest_load_per_task; | |
3694 | } | |
dbc523a3 GS |
3695 | |
3696 | /** | |
3697 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
3698 | * groups of a given sched_domain during load balance. | |
3699 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
3700 | * @this_cpu: Cpu for which currently load balance is being performed. | |
3701 | * @imbalance: The variable to store the imbalance. | |
3702 | */ | |
3703 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
3704 | unsigned long *imbalance) | |
3705 | { | |
3706 | unsigned long max_pull; | |
2dd73a4f PW |
3707 | /* |
3708 | * In the presence of smp nice balancing, certain scenarios can have | |
3709 | * max load less than avg load(as we skip the groups at or below | |
3710 | * its cpu_power, while calculating max_load..) | |
3711 | */ | |
dbc523a3 | 3712 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 3713 | *imbalance = 0; |
dbc523a3 | 3714 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 3715 | } |
0c117f1b SS |
3716 | |
3717 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
3718 | max_pull = min(sds->max_load - sds->avg_load, |
3719 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 3720 | |
1da177e4 | 3721 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
3722 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
3723 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
3724 | / SCHED_LOAD_SCALE; |
3725 | ||
2dd73a4f PW |
3726 | /* |
3727 | * if *imbalance is less than the average load per runnable task | |
3728 | * there is no gaurantee that any tasks will be moved so we'll have | |
3729 | * a think about bumping its value to force at least one task to be | |
3730 | * moved | |
3731 | */ | |
dbc523a3 GS |
3732 | if (*imbalance < sds->busiest_load_per_task) |
3733 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 3734 | |
dbc523a3 | 3735 | } |
37abe198 | 3736 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 3737 | |
b7bb4c9b GS |
3738 | /** |
3739 | * find_busiest_group - Returns the busiest group within the sched_domain | |
3740 | * if there is an imbalance. If there isn't an imbalance, and | |
3741 | * the user has opted for power-savings, it returns a group whose | |
3742 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
3743 | * such a group exists. | |
3744 | * | |
3745 | * Also calculates the amount of weighted load which should be moved | |
3746 | * to restore balance. | |
3747 | * | |
3748 | * @sd: The sched_domain whose busiest group is to be returned. | |
3749 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
3750 | * @imbalance: Variable which stores amount of weighted load which should | |
3751 | * be moved to restore balance/put a group to idle. | |
3752 | * @idle: The idle status of this_cpu. | |
3753 | * @sd_idle: The idleness of sd | |
3754 | * @cpus: The set of CPUs under consideration for load-balancing. | |
3755 | * @balance: Pointer to a variable indicating if this_cpu | |
3756 | * is the appropriate cpu to perform load balancing at this_level. | |
3757 | * | |
3758 | * Returns: - the busiest group if imbalance exists. | |
3759 | * - If no imbalance and user has opted for power-savings balance, | |
3760 | * return the least loaded group whose CPUs can be | |
3761 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
3762 | */ |
3763 | static struct sched_group * | |
3764 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
3765 | unsigned long *imbalance, enum cpu_idle_type idle, | |
3766 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
3767 | { | |
3768 | struct sd_lb_stats sds; | |
1da177e4 | 3769 | |
37abe198 | 3770 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 3771 | |
37abe198 GS |
3772 | /* |
3773 | * Compute the various statistics relavent for load balancing at | |
3774 | * this level. | |
3775 | */ | |
3776 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
3777 | balance, &sds); | |
3778 | ||
b7bb4c9b GS |
3779 | /* Cases where imbalance does not exist from POV of this_cpu */ |
3780 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
3781 | * at this level. | |
3782 | * 2) There is no busy sibling group to pull from. | |
3783 | * 3) This group is the busiest group. | |
3784 | * 4) This group is more busy than the avg busieness at this | |
3785 | * sched_domain. | |
3786 | * 5) The imbalance is within the specified limit. | |
3787 | * 6) Any rebalance would lead to ping-pong | |
3788 | */ | |
37abe198 GS |
3789 | if (balance && !(*balance)) |
3790 | goto ret; | |
1da177e4 | 3791 | |
b7bb4c9b GS |
3792 | if (!sds.busiest || sds.busiest_nr_running == 0) |
3793 | goto out_balanced; | |
1da177e4 | 3794 | |
b7bb4c9b | 3795 | if (sds.this_load >= sds.max_load) |
1da177e4 | 3796 | goto out_balanced; |
1da177e4 | 3797 | |
222d656d | 3798 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 3799 | |
b7bb4c9b GS |
3800 | if (sds.this_load >= sds.avg_load) |
3801 | goto out_balanced; | |
3802 | ||
3803 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
3804 | goto out_balanced; |
3805 | ||
222d656d GS |
3806 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
3807 | if (sds.group_imb) | |
3808 | sds.busiest_load_per_task = | |
3809 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 3810 | |
1da177e4 LT |
3811 | /* |
3812 | * We're trying to get all the cpus to the average_load, so we don't | |
3813 | * want to push ourselves above the average load, nor do we wish to | |
3814 | * reduce the max loaded cpu below the average load, as either of these | |
3815 | * actions would just result in more rebalancing later, and ping-pong | |
3816 | * tasks around. Thus we look for the minimum possible imbalance. | |
3817 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3818 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3819 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3820 | * appear as very large values with unsigned longs. |
3821 | */ | |
222d656d | 3822 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
3823 | goto out_balanced; |
3824 | ||
dbc523a3 GS |
3825 | /* Looks like there is an imbalance. Compute it */ |
3826 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 3827 | return sds.busiest; |
1da177e4 LT |
3828 | |
3829 | out_balanced: | |
c071df18 GS |
3830 | /* |
3831 | * There is no obvious imbalance. But check if we can do some balancing | |
3832 | * to save power. | |
3833 | */ | |
3834 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
3835 | return sds.busiest; | |
783609c6 | 3836 | ret: |
1da177e4 LT |
3837 | *imbalance = 0; |
3838 | return NULL; | |
3839 | } | |
3840 | ||
3841 | /* | |
3842 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3843 | */ | |
70b97a7f | 3844 | static struct rq * |
d15bcfdb | 3845 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 3846 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 3847 | { |
70b97a7f | 3848 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3849 | unsigned long max_load = 0; |
1da177e4 LT |
3850 | int i; |
3851 | ||
758b2cdc | 3852 | for_each_cpu(i, sched_group_cpus(group)) { |
dd41f596 | 3853 | unsigned long wl; |
0a2966b4 | 3854 | |
96f874e2 | 3855 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
3856 | continue; |
3857 | ||
48f24c4d | 3858 | rq = cpu_rq(i); |
dd41f596 | 3859 | wl = weighted_cpuload(i); |
2dd73a4f | 3860 | |
dd41f596 | 3861 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3862 | continue; |
1da177e4 | 3863 | |
dd41f596 IM |
3864 | if (wl > max_load) { |
3865 | max_load = wl; | |
48f24c4d | 3866 | busiest = rq; |
1da177e4 LT |
3867 | } |
3868 | } | |
3869 | ||
3870 | return busiest; | |
3871 | } | |
3872 | ||
77391d71 NP |
3873 | /* |
3874 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3875 | * so long as it is large enough. | |
3876 | */ | |
3877 | #define MAX_PINNED_INTERVAL 512 | |
3878 | ||
df7c8e84 RR |
3879 | /* Working cpumask for load_balance and load_balance_newidle. */ |
3880 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
3881 | ||
1da177e4 LT |
3882 | /* |
3883 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3884 | * tasks if there is an imbalance. | |
1da177e4 | 3885 | */ |
70b97a7f | 3886 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3887 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 3888 | int *balance) |
1da177e4 | 3889 | { |
43010659 | 3890 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3891 | struct sched_group *group; |
1da177e4 | 3892 | unsigned long imbalance; |
70b97a7f | 3893 | struct rq *busiest; |
fe2eea3f | 3894 | unsigned long flags; |
df7c8e84 | 3895 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 3896 | |
96f874e2 | 3897 | cpumask_setall(cpus); |
7c16ec58 | 3898 | |
89c4710e SS |
3899 | /* |
3900 | * When power savings policy is enabled for the parent domain, idle | |
3901 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3902 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3903 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3904 | */ |
d15bcfdb | 3905 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3906 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3907 | sd_idle = 1; |
1da177e4 | 3908 | |
2d72376b | 3909 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3910 | |
0a2966b4 | 3911 | redo: |
c8cba857 | 3912 | update_shares(sd); |
0a2966b4 | 3913 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 3914 | cpus, balance); |
783609c6 | 3915 | |
06066714 | 3916 | if (*balance == 0) |
783609c6 | 3917 | goto out_balanced; |
783609c6 | 3918 | |
1da177e4 LT |
3919 | if (!group) { |
3920 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3921 | goto out_balanced; | |
3922 | } | |
3923 | ||
7c16ec58 | 3924 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3925 | if (!busiest) { |
3926 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3927 | goto out_balanced; | |
3928 | } | |
3929 | ||
db935dbd | 3930 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3931 | |
3932 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3933 | ||
43010659 | 3934 | ld_moved = 0; |
1da177e4 LT |
3935 | if (busiest->nr_running > 1) { |
3936 | /* | |
3937 | * Attempt to move tasks. If find_busiest_group has found | |
3938 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3939 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3940 | * correctly treated as an imbalance. |
3941 | */ | |
fe2eea3f | 3942 | local_irq_save(flags); |
e17224bf | 3943 | double_rq_lock(this_rq, busiest); |
43010659 | 3944 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3945 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3946 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3947 | local_irq_restore(flags); |
81026794 | 3948 | |
46cb4b7c SS |
3949 | /* |
3950 | * some other cpu did the load balance for us. | |
3951 | */ | |
43010659 | 3952 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3953 | resched_cpu(this_cpu); |
3954 | ||
81026794 | 3955 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3956 | if (unlikely(all_pinned)) { |
96f874e2 RR |
3957 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
3958 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 3959 | goto redo; |
81026794 | 3960 | goto out_balanced; |
0a2966b4 | 3961 | } |
1da177e4 | 3962 | } |
81026794 | 3963 | |
43010659 | 3964 | if (!ld_moved) { |
1da177e4 LT |
3965 | schedstat_inc(sd, lb_failed[idle]); |
3966 | sd->nr_balance_failed++; | |
3967 | ||
3968 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3969 | |
fe2eea3f | 3970 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3971 | |
3972 | /* don't kick the migration_thread, if the curr | |
3973 | * task on busiest cpu can't be moved to this_cpu | |
3974 | */ | |
96f874e2 RR |
3975 | if (!cpumask_test_cpu(this_cpu, |
3976 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3977 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3978 | all_pinned = 1; |
3979 | goto out_one_pinned; | |
3980 | } | |
3981 | ||
1da177e4 LT |
3982 | if (!busiest->active_balance) { |
3983 | busiest->active_balance = 1; | |
3984 | busiest->push_cpu = this_cpu; | |
81026794 | 3985 | active_balance = 1; |
1da177e4 | 3986 | } |
fe2eea3f | 3987 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3988 | if (active_balance) |
1da177e4 LT |
3989 | wake_up_process(busiest->migration_thread); |
3990 | ||
3991 | /* | |
3992 | * We've kicked active balancing, reset the failure | |
3993 | * counter. | |
3994 | */ | |
39507451 | 3995 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3996 | } |
81026794 | 3997 | } else |
1da177e4 LT |
3998 | sd->nr_balance_failed = 0; |
3999 | ||
81026794 | 4000 | if (likely(!active_balance)) { |
1da177e4 LT |
4001 | /* We were unbalanced, so reset the balancing interval */ |
4002 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4003 | } else { |
4004 | /* | |
4005 | * If we've begun active balancing, start to back off. This | |
4006 | * case may not be covered by the all_pinned logic if there | |
4007 | * is only 1 task on the busy runqueue (because we don't call | |
4008 | * move_tasks). | |
4009 | */ | |
4010 | if (sd->balance_interval < sd->max_interval) | |
4011 | sd->balance_interval *= 2; | |
1da177e4 LT |
4012 | } |
4013 | ||
43010659 | 4014 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4015 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4016 | ld_moved = -1; |
4017 | ||
4018 | goto out; | |
1da177e4 LT |
4019 | |
4020 | out_balanced: | |
1da177e4 LT |
4021 | schedstat_inc(sd, lb_balanced[idle]); |
4022 | ||
16cfb1c0 | 4023 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4024 | |
4025 | out_one_pinned: | |
1da177e4 | 4026 | /* tune up the balancing interval */ |
77391d71 NP |
4027 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4028 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4029 | sd->balance_interval *= 2; |
4030 | ||
48f24c4d | 4031 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4032 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4033 | ld_moved = -1; |
4034 | else | |
4035 | ld_moved = 0; | |
4036 | out: | |
c8cba857 PZ |
4037 | if (ld_moved) |
4038 | update_shares(sd); | |
c09595f6 | 4039 | return ld_moved; |
1da177e4 LT |
4040 | } |
4041 | ||
4042 | /* | |
4043 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4044 | * tasks if there is an imbalance. | |
4045 | * | |
d15bcfdb | 4046 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4047 | * this_rq is locked. |
4048 | */ | |
48f24c4d | 4049 | static int |
df7c8e84 | 4050 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4051 | { |
4052 | struct sched_group *group; | |
70b97a7f | 4053 | struct rq *busiest = NULL; |
1da177e4 | 4054 | unsigned long imbalance; |
43010659 | 4055 | int ld_moved = 0; |
5969fe06 | 4056 | int sd_idle = 0; |
969bb4e4 | 4057 | int all_pinned = 0; |
df7c8e84 | 4058 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4059 | |
96f874e2 | 4060 | cpumask_setall(cpus); |
5969fe06 | 4061 | |
89c4710e SS |
4062 | /* |
4063 | * When power savings policy is enabled for the parent domain, idle | |
4064 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4065 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4066 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4067 | */ |
4068 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4069 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4070 | sd_idle = 1; |
1da177e4 | 4071 | |
2d72376b | 4072 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4073 | redo: |
3e5459b4 | 4074 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4075 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4076 | &sd_idle, cpus, NULL); |
1da177e4 | 4077 | if (!group) { |
d15bcfdb | 4078 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4079 | goto out_balanced; |
1da177e4 LT |
4080 | } |
4081 | ||
7c16ec58 | 4082 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4083 | if (!busiest) { |
d15bcfdb | 4084 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4085 | goto out_balanced; |
1da177e4 LT |
4086 | } |
4087 | ||
db935dbd NP |
4088 | BUG_ON(busiest == this_rq); |
4089 | ||
d15bcfdb | 4090 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4091 | |
43010659 | 4092 | ld_moved = 0; |
d6d5cfaf NP |
4093 | if (busiest->nr_running > 1) { |
4094 | /* Attempt to move tasks */ | |
4095 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4096 | /* this_rq->clock is already updated */ |
4097 | update_rq_clock(busiest); | |
43010659 | 4098 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4099 | imbalance, sd, CPU_NEWLY_IDLE, |
4100 | &all_pinned); | |
1b12bbc7 | 4101 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4102 | |
969bb4e4 | 4103 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4104 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4105 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4106 | goto redo; |
4107 | } | |
d6d5cfaf NP |
4108 | } |
4109 | ||
43010659 | 4110 | if (!ld_moved) { |
36dffab6 | 4111 | int active_balance = 0; |
ad273b32 | 4112 | |
d15bcfdb | 4113 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4114 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4115 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4116 | return -1; |
ad273b32 VS |
4117 | |
4118 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4119 | return -1; | |
4120 | ||
4121 | if (sd->nr_balance_failed++ < 2) | |
4122 | return -1; | |
4123 | ||
4124 | /* | |
4125 | * The only task running in a non-idle cpu can be moved to this | |
4126 | * cpu in an attempt to completely freeup the other CPU | |
4127 | * package. The same method used to move task in load_balance() | |
4128 | * have been extended for load_balance_newidle() to speedup | |
4129 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4130 | * | |
4131 | * The package power saving logic comes from | |
4132 | * find_busiest_group(). If there are no imbalance, then | |
4133 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4134 | * f_b_g() will select a group from which a running task may be | |
4135 | * pulled to this cpu in order to make the other package idle. | |
4136 | * If there is no opportunity to make a package idle and if | |
4137 | * there are no imbalance, then f_b_g() will return NULL and no | |
4138 | * action will be taken in load_balance_newidle(). | |
4139 | * | |
4140 | * Under normal task pull operation due to imbalance, there | |
4141 | * will be more than one task in the source run queue and | |
4142 | * move_tasks() will succeed. ld_moved will be true and this | |
4143 | * active balance code will not be triggered. | |
4144 | */ | |
4145 | ||
4146 | /* Lock busiest in correct order while this_rq is held */ | |
4147 | double_lock_balance(this_rq, busiest); | |
4148 | ||
4149 | /* | |
4150 | * don't kick the migration_thread, if the curr | |
4151 | * task on busiest cpu can't be moved to this_cpu | |
4152 | */ | |
6ca09dfc | 4153 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4154 | double_unlock_balance(this_rq, busiest); |
4155 | all_pinned = 1; | |
4156 | return ld_moved; | |
4157 | } | |
4158 | ||
4159 | if (!busiest->active_balance) { | |
4160 | busiest->active_balance = 1; | |
4161 | busiest->push_cpu = this_cpu; | |
4162 | active_balance = 1; | |
4163 | } | |
4164 | ||
4165 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4166 | /* |
4167 | * Should not call ttwu while holding a rq->lock | |
4168 | */ | |
4169 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4170 | if (active_balance) |
4171 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4172 | spin_lock(&this_rq->lock); |
ad273b32 | 4173 | |
5969fe06 | 4174 | } else |
16cfb1c0 | 4175 | sd->nr_balance_failed = 0; |
1da177e4 | 4176 | |
3e5459b4 | 4177 | update_shares_locked(this_rq, sd); |
43010659 | 4178 | return ld_moved; |
16cfb1c0 NP |
4179 | |
4180 | out_balanced: | |
d15bcfdb | 4181 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4182 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4183 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4184 | return -1; |
16cfb1c0 | 4185 | sd->nr_balance_failed = 0; |
48f24c4d | 4186 | |
16cfb1c0 | 4187 | return 0; |
1da177e4 LT |
4188 | } |
4189 | ||
4190 | /* | |
4191 | * idle_balance is called by schedule() if this_cpu is about to become | |
4192 | * idle. Attempts to pull tasks from other CPUs. | |
4193 | */ | |
70b97a7f | 4194 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4195 | { |
4196 | struct sched_domain *sd; | |
efbe027e | 4197 | int pulled_task = 0; |
dd41f596 | 4198 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4199 | |
4200 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4201 | unsigned long interval; |
4202 | ||
4203 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4204 | continue; | |
4205 | ||
4206 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4207 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4208 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4209 | sd); |
92c4ca5c CL |
4210 | |
4211 | interval = msecs_to_jiffies(sd->balance_interval); | |
4212 | if (time_after(next_balance, sd->last_balance + interval)) | |
4213 | next_balance = sd->last_balance + interval; | |
4214 | if (pulled_task) | |
4215 | break; | |
1da177e4 | 4216 | } |
dd41f596 | 4217 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4218 | /* |
4219 | * We are going idle. next_balance may be set based on | |
4220 | * a busy processor. So reset next_balance. | |
4221 | */ | |
4222 | this_rq->next_balance = next_balance; | |
dd41f596 | 4223 | } |
1da177e4 LT |
4224 | } |
4225 | ||
4226 | /* | |
4227 | * active_load_balance is run by migration threads. It pushes running tasks | |
4228 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4229 | * running on each physical CPU where possible, and avoids physical / | |
4230 | * logical imbalances. | |
4231 | * | |
4232 | * Called with busiest_rq locked. | |
4233 | */ | |
70b97a7f | 4234 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4235 | { |
39507451 | 4236 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4237 | struct sched_domain *sd; |
4238 | struct rq *target_rq; | |
39507451 | 4239 | |
48f24c4d | 4240 | /* Is there any task to move? */ |
39507451 | 4241 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4242 | return; |
4243 | ||
4244 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4245 | |
4246 | /* | |
39507451 | 4247 | * This condition is "impossible", if it occurs |
41a2d6cf | 4248 | * we need to fix it. Originally reported by |
39507451 | 4249 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4250 | */ |
39507451 | 4251 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4252 | |
39507451 NP |
4253 | /* move a task from busiest_rq to target_rq */ |
4254 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4255 | update_rq_clock(busiest_rq); |
4256 | update_rq_clock(target_rq); | |
39507451 NP |
4257 | |
4258 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4259 | for_each_domain(target_cpu, sd) { |
39507451 | 4260 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4261 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4262 | break; |
c96d145e | 4263 | } |
39507451 | 4264 | |
48f24c4d | 4265 | if (likely(sd)) { |
2d72376b | 4266 | schedstat_inc(sd, alb_count); |
39507451 | 4267 | |
43010659 PW |
4268 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4269 | sd, CPU_IDLE)) | |
48f24c4d IM |
4270 | schedstat_inc(sd, alb_pushed); |
4271 | else | |
4272 | schedstat_inc(sd, alb_failed); | |
4273 | } | |
1b12bbc7 | 4274 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4275 | } |
4276 | ||
46cb4b7c SS |
4277 | #ifdef CONFIG_NO_HZ |
4278 | static struct { | |
4279 | atomic_t load_balancer; | |
7d1e6a9b | 4280 | cpumask_var_t cpu_mask; |
46cb4b7c SS |
4281 | } nohz ____cacheline_aligned = { |
4282 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4283 | }; |
4284 | ||
7835b98b | 4285 | /* |
46cb4b7c SS |
4286 | * This routine will try to nominate the ilb (idle load balancing) |
4287 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4288 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4289 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4290 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4291 | * arrives... | |
4292 | * | |
4293 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4294 | * for idle load balancing. ilb owner will still be part of | |
4295 | * nohz.cpu_mask.. | |
7835b98b | 4296 | * |
46cb4b7c SS |
4297 | * While stopping the tick, this cpu will become the ilb owner if there |
4298 | * is no other owner. And will be the owner till that cpu becomes busy | |
4299 | * or if all cpus in the system stop their ticks at which point | |
4300 | * there is no need for ilb owner. | |
4301 | * | |
4302 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4303 | * next busy scheduler_tick() | |
4304 | */ | |
4305 | int select_nohz_load_balancer(int stop_tick) | |
4306 | { | |
4307 | int cpu = smp_processor_id(); | |
4308 | ||
4309 | if (stop_tick) { | |
46cb4b7c SS |
4310 | cpu_rq(cpu)->in_nohz_recently = 1; |
4311 | ||
483b4ee6 SS |
4312 | if (!cpu_active(cpu)) { |
4313 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4314 | return 0; | |
4315 | ||
4316 | /* | |
4317 | * If we are going offline and still the leader, | |
4318 | * give up! | |
4319 | */ | |
46cb4b7c SS |
4320 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4321 | BUG(); | |
483b4ee6 | 4322 | |
46cb4b7c SS |
4323 | return 0; |
4324 | } | |
4325 | ||
483b4ee6 SS |
4326 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4327 | ||
46cb4b7c | 4328 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4329 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4330 | if (atomic_read(&nohz.load_balancer) == cpu) |
4331 | atomic_set(&nohz.load_balancer, -1); | |
4332 | return 0; | |
4333 | } | |
4334 | ||
4335 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4336 | /* make me the ilb owner */ | |
4337 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4338 | return 1; | |
4339 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
4340 | return 1; | |
4341 | } else { | |
7d1e6a9b | 4342 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4343 | return 0; |
4344 | ||
7d1e6a9b | 4345 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4346 | |
4347 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4348 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4349 | BUG(); | |
4350 | } | |
4351 | return 0; | |
4352 | } | |
4353 | #endif | |
4354 | ||
4355 | static DEFINE_SPINLOCK(balancing); | |
4356 | ||
4357 | /* | |
7835b98b CL |
4358 | * It checks each scheduling domain to see if it is due to be balanced, |
4359 | * and initiates a balancing operation if so. | |
4360 | * | |
4361 | * Balancing parameters are set up in arch_init_sched_domains. | |
4362 | */ | |
a9957449 | 4363 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4364 | { |
46cb4b7c SS |
4365 | int balance = 1; |
4366 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4367 | unsigned long interval; |
4368 | struct sched_domain *sd; | |
46cb4b7c | 4369 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4370 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4371 | int update_next_balance = 0; |
d07355f5 | 4372 | int need_serialize; |
1da177e4 | 4373 | |
46cb4b7c | 4374 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4375 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4376 | continue; | |
4377 | ||
4378 | interval = sd->balance_interval; | |
d15bcfdb | 4379 | if (idle != CPU_IDLE) |
1da177e4 LT |
4380 | interval *= sd->busy_factor; |
4381 | ||
4382 | /* scale ms to jiffies */ | |
4383 | interval = msecs_to_jiffies(interval); | |
4384 | if (unlikely(!interval)) | |
4385 | interval = 1; | |
dd41f596 IM |
4386 | if (interval > HZ*NR_CPUS/10) |
4387 | interval = HZ*NR_CPUS/10; | |
4388 | ||
d07355f5 | 4389 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4390 | |
d07355f5 | 4391 | if (need_serialize) { |
08c183f3 CL |
4392 | if (!spin_trylock(&balancing)) |
4393 | goto out; | |
4394 | } | |
4395 | ||
c9819f45 | 4396 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4397 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4398 | /* |
4399 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4400 | * longer idle, or one of our SMT siblings is |
4401 | * not idle. | |
4402 | */ | |
d15bcfdb | 4403 | idle = CPU_NOT_IDLE; |
1da177e4 | 4404 | } |
1bd77f2d | 4405 | sd->last_balance = jiffies; |
1da177e4 | 4406 | } |
d07355f5 | 4407 | if (need_serialize) |
08c183f3 CL |
4408 | spin_unlock(&balancing); |
4409 | out: | |
f549da84 | 4410 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4411 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4412 | update_next_balance = 1; |
4413 | } | |
783609c6 SS |
4414 | |
4415 | /* | |
4416 | * Stop the load balance at this level. There is another | |
4417 | * CPU in our sched group which is doing load balancing more | |
4418 | * actively. | |
4419 | */ | |
4420 | if (!balance) | |
4421 | break; | |
1da177e4 | 4422 | } |
f549da84 SS |
4423 | |
4424 | /* | |
4425 | * next_balance will be updated only when there is a need. | |
4426 | * When the cpu is attached to null domain for ex, it will not be | |
4427 | * updated. | |
4428 | */ | |
4429 | if (likely(update_next_balance)) | |
4430 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4431 | } |
4432 | ||
4433 | /* | |
4434 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4435 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4436 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4437 | */ | |
4438 | static void run_rebalance_domains(struct softirq_action *h) | |
4439 | { | |
dd41f596 IM |
4440 | int this_cpu = smp_processor_id(); |
4441 | struct rq *this_rq = cpu_rq(this_cpu); | |
4442 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4443 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4444 | |
dd41f596 | 4445 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4446 | |
4447 | #ifdef CONFIG_NO_HZ | |
4448 | /* | |
4449 | * If this cpu is the owner for idle load balancing, then do the | |
4450 | * balancing on behalf of the other idle cpus whose ticks are | |
4451 | * stopped. | |
4452 | */ | |
dd41f596 IM |
4453 | if (this_rq->idle_at_tick && |
4454 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4455 | struct rq *rq; |
4456 | int balance_cpu; | |
4457 | ||
7d1e6a9b RR |
4458 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4459 | if (balance_cpu == this_cpu) | |
4460 | continue; | |
4461 | ||
46cb4b7c SS |
4462 | /* |
4463 | * If this cpu gets work to do, stop the load balancing | |
4464 | * work being done for other cpus. Next load | |
4465 | * balancing owner will pick it up. | |
4466 | */ | |
4467 | if (need_resched()) | |
4468 | break; | |
4469 | ||
de0cf899 | 4470 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4471 | |
4472 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4473 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4474 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4475 | } |
4476 | } | |
4477 | #endif | |
4478 | } | |
4479 | ||
8a0be9ef FW |
4480 | static inline int on_null_domain(int cpu) |
4481 | { | |
4482 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4483 | } | |
4484 | ||
46cb4b7c SS |
4485 | /* |
4486 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
4487 | * | |
4488 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
4489 | * idle load balancing owner or decide to stop the periodic load balancing, | |
4490 | * if the whole system is idle. | |
4491 | */ | |
dd41f596 | 4492 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 4493 | { |
46cb4b7c SS |
4494 | #ifdef CONFIG_NO_HZ |
4495 | /* | |
4496 | * If we were in the nohz mode recently and busy at the current | |
4497 | * scheduler tick, then check if we need to nominate new idle | |
4498 | * load balancer. | |
4499 | */ | |
4500 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
4501 | rq->in_nohz_recently = 0; | |
4502 | ||
4503 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 4504 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4505 | atomic_set(&nohz.load_balancer, -1); |
4506 | } | |
4507 | ||
4508 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4509 | /* | |
4510 | * simple selection for now: Nominate the | |
4511 | * first cpu in the nohz list to be the next | |
4512 | * ilb owner. | |
4513 | * | |
4514 | * TBD: Traverse the sched domains and nominate | |
4515 | * the nearest cpu in the nohz.cpu_mask. | |
4516 | */ | |
7d1e6a9b | 4517 | int ilb = cpumask_first(nohz.cpu_mask); |
46cb4b7c | 4518 | |
434d53b0 | 4519 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
4520 | resched_cpu(ilb); |
4521 | } | |
4522 | } | |
4523 | ||
4524 | /* | |
4525 | * If this cpu is idle and doing idle load balancing for all the | |
4526 | * cpus with ticks stopped, is it time for that to stop? | |
4527 | */ | |
4528 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 4529 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4530 | resched_cpu(cpu); |
4531 | return; | |
4532 | } | |
4533 | ||
4534 | /* | |
4535 | * If this cpu is idle and the idle load balancing is done by | |
4536 | * someone else, then no need raise the SCHED_SOFTIRQ | |
4537 | */ | |
4538 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 4539 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4540 | return; |
4541 | #endif | |
8a0be9ef FW |
4542 | /* Don't need to rebalance while attached to NULL domain */ |
4543 | if (time_after_eq(jiffies, rq->next_balance) && | |
4544 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 4545 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 4546 | } |
dd41f596 IM |
4547 | |
4548 | #else /* CONFIG_SMP */ | |
4549 | ||
1da177e4 LT |
4550 | /* |
4551 | * on UP we do not need to balance between CPUs: | |
4552 | */ | |
70b97a7f | 4553 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
4554 | { |
4555 | } | |
dd41f596 | 4556 | |
1da177e4 LT |
4557 | #endif |
4558 | ||
1da177e4 LT |
4559 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
4560 | ||
4561 | EXPORT_PER_CPU_SYMBOL(kstat); | |
4562 | ||
4563 | /* | |
c5f8d995 | 4564 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 4565 | * @p in case that task is currently running. |
c5f8d995 HS |
4566 | * |
4567 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 4568 | */ |
c5f8d995 HS |
4569 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
4570 | { | |
4571 | u64 ns = 0; | |
4572 | ||
4573 | if (task_current(rq, p)) { | |
4574 | update_rq_clock(rq); | |
4575 | ns = rq->clock - p->se.exec_start; | |
4576 | if ((s64)ns < 0) | |
4577 | ns = 0; | |
4578 | } | |
4579 | ||
4580 | return ns; | |
4581 | } | |
4582 | ||
bb34d92f | 4583 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 4584 | { |
1da177e4 | 4585 | unsigned long flags; |
41b86e9c | 4586 | struct rq *rq; |
bb34d92f | 4587 | u64 ns = 0; |
48f24c4d | 4588 | |
41b86e9c | 4589 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
4590 | ns = do_task_delta_exec(p, rq); |
4591 | task_rq_unlock(rq, &flags); | |
1508487e | 4592 | |
c5f8d995 HS |
4593 | return ns; |
4594 | } | |
f06febc9 | 4595 | |
c5f8d995 HS |
4596 | /* |
4597 | * Return accounted runtime for the task. | |
4598 | * In case the task is currently running, return the runtime plus current's | |
4599 | * pending runtime that have not been accounted yet. | |
4600 | */ | |
4601 | unsigned long long task_sched_runtime(struct task_struct *p) | |
4602 | { | |
4603 | unsigned long flags; | |
4604 | struct rq *rq; | |
4605 | u64 ns = 0; | |
4606 | ||
4607 | rq = task_rq_lock(p, &flags); | |
4608 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
4609 | task_rq_unlock(rq, &flags); | |
4610 | ||
4611 | return ns; | |
4612 | } | |
48f24c4d | 4613 | |
c5f8d995 HS |
4614 | /* |
4615 | * Return sum_exec_runtime for the thread group. | |
4616 | * In case the task is currently running, return the sum plus current's | |
4617 | * pending runtime that have not been accounted yet. | |
4618 | * | |
4619 | * Note that the thread group might have other running tasks as well, | |
4620 | * so the return value not includes other pending runtime that other | |
4621 | * running tasks might have. | |
4622 | */ | |
4623 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
4624 | { | |
4625 | struct task_cputime totals; | |
4626 | unsigned long flags; | |
4627 | struct rq *rq; | |
4628 | u64 ns; | |
48f24c4d | 4629 | |
c5f8d995 HS |
4630 | rq = task_rq_lock(p, &flags); |
4631 | thread_group_cputime(p, &totals); | |
4632 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 4633 | task_rq_unlock(rq, &flags); |
48f24c4d | 4634 | |
1da177e4 LT |
4635 | return ns; |
4636 | } | |
4637 | ||
1da177e4 LT |
4638 | /* |
4639 | * Account user cpu time to a process. | |
4640 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 4641 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 4642 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 4643 | */ |
457533a7 MS |
4644 | void account_user_time(struct task_struct *p, cputime_t cputime, |
4645 | cputime_t cputime_scaled) | |
1da177e4 LT |
4646 | { |
4647 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4648 | cputime64_t tmp; | |
4649 | ||
457533a7 | 4650 | /* Add user time to process. */ |
1da177e4 | 4651 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4652 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4653 | account_group_user_time(p, cputime); |
1da177e4 LT |
4654 | |
4655 | /* Add user time to cpustat. */ | |
4656 | tmp = cputime_to_cputime64(cputime); | |
4657 | if (TASK_NICE(p) > 0) | |
4658 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
4659 | else | |
4660 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
4661 | |
4662 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
4663 | /* Account for user time used */ |
4664 | acct_update_integrals(p); | |
1da177e4 LT |
4665 | } |
4666 | ||
94886b84 LV |
4667 | /* |
4668 | * Account guest cpu time to a process. | |
4669 | * @p: the process that the cpu time gets accounted to | |
4670 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 4671 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 4672 | */ |
457533a7 MS |
4673 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
4674 | cputime_t cputime_scaled) | |
94886b84 LV |
4675 | { |
4676 | cputime64_t tmp; | |
4677 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4678 | ||
4679 | tmp = cputime_to_cputime64(cputime); | |
4680 | ||
457533a7 | 4681 | /* Add guest time to process. */ |
94886b84 | 4682 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 4683 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 4684 | account_group_user_time(p, cputime); |
94886b84 LV |
4685 | p->gtime = cputime_add(p->gtime, cputime); |
4686 | ||
457533a7 | 4687 | /* Add guest time to cpustat. */ |
94886b84 LV |
4688 | cpustat->user = cputime64_add(cpustat->user, tmp); |
4689 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
4690 | } | |
4691 | ||
1da177e4 LT |
4692 | /* |
4693 | * Account system cpu time to a process. | |
4694 | * @p: the process that the cpu time gets accounted to | |
4695 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
4696 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 4697 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
4698 | */ |
4699 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 4700 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
4701 | { |
4702 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
4703 | cputime64_t tmp; |
4704 | ||
983ed7a6 | 4705 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 4706 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
4707 | return; |
4708 | } | |
94886b84 | 4709 | |
457533a7 | 4710 | /* Add system time to process. */ |
1da177e4 | 4711 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 4712 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 4713 | account_group_system_time(p, cputime); |
1da177e4 LT |
4714 | |
4715 | /* Add system time to cpustat. */ | |
4716 | tmp = cputime_to_cputime64(cputime); | |
4717 | if (hardirq_count() - hardirq_offset) | |
4718 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
4719 | else if (softirq_count()) | |
4720 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 4721 | else |
79741dd3 MS |
4722 | cpustat->system = cputime64_add(cpustat->system, tmp); |
4723 | ||
ef12fefa BR |
4724 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
4725 | ||
1da177e4 LT |
4726 | /* Account for system time used */ |
4727 | acct_update_integrals(p); | |
1da177e4 LT |
4728 | } |
4729 | ||
c66f08be | 4730 | /* |
1da177e4 | 4731 | * Account for involuntary wait time. |
1da177e4 | 4732 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 4733 | */ |
79741dd3 | 4734 | void account_steal_time(cputime_t cputime) |
c66f08be | 4735 | { |
79741dd3 MS |
4736 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
4737 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
4738 | ||
4739 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
4740 | } |
4741 | ||
1da177e4 | 4742 | /* |
79741dd3 MS |
4743 | * Account for idle time. |
4744 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 4745 | */ |
79741dd3 | 4746 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
4747 | { |
4748 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 4749 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 4750 | struct rq *rq = this_rq(); |
1da177e4 | 4751 | |
79741dd3 MS |
4752 | if (atomic_read(&rq->nr_iowait) > 0) |
4753 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
4754 | else | |
4755 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
4756 | } |
4757 | ||
79741dd3 MS |
4758 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
4759 | ||
4760 | /* | |
4761 | * Account a single tick of cpu time. | |
4762 | * @p: the process that the cpu time gets accounted to | |
4763 | * @user_tick: indicates if the tick is a user or a system tick | |
4764 | */ | |
4765 | void account_process_tick(struct task_struct *p, int user_tick) | |
4766 | { | |
4767 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
4768 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
4769 | struct rq *rq = this_rq(); | |
4770 | ||
4771 | if (user_tick) | |
4772 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 4773 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
4774 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
4775 | one_jiffy_scaled); | |
4776 | else | |
4777 | account_idle_time(one_jiffy); | |
4778 | } | |
4779 | ||
4780 | /* | |
4781 | * Account multiple ticks of steal time. | |
4782 | * @p: the process from which the cpu time has been stolen | |
4783 | * @ticks: number of stolen ticks | |
4784 | */ | |
4785 | void account_steal_ticks(unsigned long ticks) | |
4786 | { | |
4787 | account_steal_time(jiffies_to_cputime(ticks)); | |
4788 | } | |
4789 | ||
4790 | /* | |
4791 | * Account multiple ticks of idle time. | |
4792 | * @ticks: number of stolen ticks | |
4793 | */ | |
4794 | void account_idle_ticks(unsigned long ticks) | |
4795 | { | |
4796 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
4797 | } |
4798 | ||
79741dd3 MS |
4799 | #endif |
4800 | ||
49048622 BS |
4801 | /* |
4802 | * Use precise platform statistics if available: | |
4803 | */ | |
4804 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
4805 | cputime_t task_utime(struct task_struct *p) | |
4806 | { | |
4807 | return p->utime; | |
4808 | } | |
4809 | ||
4810 | cputime_t task_stime(struct task_struct *p) | |
4811 | { | |
4812 | return p->stime; | |
4813 | } | |
4814 | #else | |
4815 | cputime_t task_utime(struct task_struct *p) | |
4816 | { | |
4817 | clock_t utime = cputime_to_clock_t(p->utime), | |
4818 | total = utime + cputime_to_clock_t(p->stime); | |
4819 | u64 temp; | |
4820 | ||
4821 | /* | |
4822 | * Use CFS's precise accounting: | |
4823 | */ | |
4824 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
4825 | ||
4826 | if (total) { | |
4827 | temp *= utime; | |
4828 | do_div(temp, total); | |
4829 | } | |
4830 | utime = (clock_t)temp; | |
4831 | ||
4832 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
4833 | return p->prev_utime; | |
4834 | } | |
4835 | ||
4836 | cputime_t task_stime(struct task_struct *p) | |
4837 | { | |
4838 | clock_t stime; | |
4839 | ||
4840 | /* | |
4841 | * Use CFS's precise accounting. (we subtract utime from | |
4842 | * the total, to make sure the total observed by userspace | |
4843 | * grows monotonically - apps rely on that): | |
4844 | */ | |
4845 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
4846 | cputime_to_clock_t(task_utime(p)); | |
4847 | ||
4848 | if (stime >= 0) | |
4849 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
4850 | ||
4851 | return p->prev_stime; | |
4852 | } | |
4853 | #endif | |
4854 | ||
4855 | inline cputime_t task_gtime(struct task_struct *p) | |
4856 | { | |
4857 | return p->gtime; | |
4858 | } | |
4859 | ||
7835b98b CL |
4860 | /* |
4861 | * This function gets called by the timer code, with HZ frequency. | |
4862 | * We call it with interrupts disabled. | |
4863 | * | |
4864 | * It also gets called by the fork code, when changing the parent's | |
4865 | * timeslices. | |
4866 | */ | |
4867 | void scheduler_tick(void) | |
4868 | { | |
7835b98b CL |
4869 | int cpu = smp_processor_id(); |
4870 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4871 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4872 | |
4873 | sched_clock_tick(); | |
dd41f596 IM |
4874 | |
4875 | spin_lock(&rq->lock); | |
3e51f33f | 4876 | update_rq_clock(rq); |
f1a438d8 | 4877 | update_cpu_load(rq); |
fa85ae24 | 4878 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4879 | spin_unlock(&rq->lock); |
7835b98b | 4880 | |
e220d2dc PZ |
4881 | perf_counter_task_tick(curr, cpu); |
4882 | ||
e418e1c2 | 4883 | #ifdef CONFIG_SMP |
dd41f596 IM |
4884 | rq->idle_at_tick = idle_cpu(cpu); |
4885 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4886 | #endif |
1da177e4 LT |
4887 | } |
4888 | ||
132380a0 | 4889 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
4890 | { |
4891 | if (in_lock_functions(addr)) { | |
4892 | addr = CALLER_ADDR2; | |
4893 | if (in_lock_functions(addr)) | |
4894 | addr = CALLER_ADDR3; | |
4895 | } | |
4896 | return addr; | |
4897 | } | |
1da177e4 | 4898 | |
7e49fcce SR |
4899 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4900 | defined(CONFIG_PREEMPT_TRACER)) | |
4901 | ||
43627582 | 4902 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4903 | { |
6cd8a4bb | 4904 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4905 | /* |
4906 | * Underflow? | |
4907 | */ | |
9a11b49a IM |
4908 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4909 | return; | |
6cd8a4bb | 4910 | #endif |
1da177e4 | 4911 | preempt_count() += val; |
6cd8a4bb | 4912 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4913 | /* |
4914 | * Spinlock count overflowing soon? | |
4915 | */ | |
33859f7f MOS |
4916 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4917 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4918 | #endif |
4919 | if (preempt_count() == val) | |
4920 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4921 | } |
4922 | EXPORT_SYMBOL(add_preempt_count); | |
4923 | ||
43627582 | 4924 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4925 | { |
6cd8a4bb | 4926 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4927 | /* |
4928 | * Underflow? | |
4929 | */ | |
01e3eb82 | 4930 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4931 | return; |
1da177e4 LT |
4932 | /* |
4933 | * Is the spinlock portion underflowing? | |
4934 | */ | |
9a11b49a IM |
4935 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4936 | !(preempt_count() & PREEMPT_MASK))) | |
4937 | return; | |
6cd8a4bb | 4938 | #endif |
9a11b49a | 4939 | |
6cd8a4bb SR |
4940 | if (preempt_count() == val) |
4941 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4942 | preempt_count() -= val; |
4943 | } | |
4944 | EXPORT_SYMBOL(sub_preempt_count); | |
4945 | ||
4946 | #endif | |
4947 | ||
4948 | /* | |
dd41f596 | 4949 | * Print scheduling while atomic bug: |
1da177e4 | 4950 | */ |
dd41f596 | 4951 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4952 | { |
838225b4 SS |
4953 | struct pt_regs *regs = get_irq_regs(); |
4954 | ||
4955 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4956 | prev->comm, prev->pid, preempt_count()); | |
4957 | ||
dd41f596 | 4958 | debug_show_held_locks(prev); |
e21f5b15 | 4959 | print_modules(); |
dd41f596 IM |
4960 | if (irqs_disabled()) |
4961 | print_irqtrace_events(prev); | |
838225b4 SS |
4962 | |
4963 | if (regs) | |
4964 | show_regs(regs); | |
4965 | else | |
4966 | dump_stack(); | |
dd41f596 | 4967 | } |
1da177e4 | 4968 | |
dd41f596 IM |
4969 | /* |
4970 | * Various schedule()-time debugging checks and statistics: | |
4971 | */ | |
4972 | static inline void schedule_debug(struct task_struct *prev) | |
4973 | { | |
1da177e4 | 4974 | /* |
41a2d6cf | 4975 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4976 | * schedule() atomically, we ignore that path for now. |
4977 | * Otherwise, whine if we are scheduling when we should not be. | |
4978 | */ | |
3f33a7ce | 4979 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4980 | __schedule_bug(prev); |
4981 | ||
1da177e4 LT |
4982 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4983 | ||
2d72376b | 4984 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4985 | #ifdef CONFIG_SCHEDSTATS |
4986 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4987 | schedstat_inc(this_rq(), bkl_count); |
4988 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4989 | } |
4990 | #endif | |
dd41f596 IM |
4991 | } |
4992 | ||
df1c99d4 MG |
4993 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
4994 | { | |
4995 | if (prev->state == TASK_RUNNING) { | |
4996 | u64 runtime = prev->se.sum_exec_runtime; | |
4997 | ||
4998 | runtime -= prev->se.prev_sum_exec_runtime; | |
4999 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5000 | ||
5001 | /* | |
5002 | * In order to avoid avg_overlap growing stale when we are | |
5003 | * indeed overlapping and hence not getting put to sleep, grow | |
5004 | * the avg_overlap on preemption. | |
5005 | * | |
5006 | * We use the average preemption runtime because that | |
5007 | * correlates to the amount of cache footprint a task can | |
5008 | * build up. | |
5009 | */ | |
5010 | update_avg(&prev->se.avg_overlap, runtime); | |
5011 | } | |
5012 | prev->sched_class->put_prev_task(rq, prev); | |
5013 | } | |
5014 | ||
dd41f596 IM |
5015 | /* |
5016 | * Pick up the highest-prio task: | |
5017 | */ | |
5018 | static inline struct task_struct * | |
b67802ea | 5019 | pick_next_task(struct rq *rq) |
dd41f596 | 5020 | { |
5522d5d5 | 5021 | const struct sched_class *class; |
dd41f596 | 5022 | struct task_struct *p; |
1da177e4 LT |
5023 | |
5024 | /* | |
dd41f596 IM |
5025 | * Optimization: we know that if all tasks are in |
5026 | * the fair class we can call that function directly: | |
1da177e4 | 5027 | */ |
dd41f596 | 5028 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5029 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5030 | if (likely(p)) |
5031 | return p; | |
1da177e4 LT |
5032 | } |
5033 | ||
dd41f596 IM |
5034 | class = sched_class_highest; |
5035 | for ( ; ; ) { | |
fb8d4724 | 5036 | p = class->pick_next_task(rq); |
dd41f596 IM |
5037 | if (p) |
5038 | return p; | |
5039 | /* | |
5040 | * Will never be NULL as the idle class always | |
5041 | * returns a non-NULL p: | |
5042 | */ | |
5043 | class = class->next; | |
5044 | } | |
5045 | } | |
1da177e4 | 5046 | |
dd41f596 IM |
5047 | /* |
5048 | * schedule() is the main scheduler function. | |
5049 | */ | |
41719b03 | 5050 | asmlinkage void __sched __schedule(void) |
dd41f596 IM |
5051 | { |
5052 | struct task_struct *prev, *next; | |
67ca7bde | 5053 | unsigned long *switch_count; |
dd41f596 | 5054 | struct rq *rq; |
31656519 | 5055 | int cpu; |
dd41f596 | 5056 | |
dd41f596 IM |
5057 | cpu = smp_processor_id(); |
5058 | rq = cpu_rq(cpu); | |
5059 | rcu_qsctr_inc(cpu); | |
5060 | prev = rq->curr; | |
5061 | switch_count = &prev->nivcsw; | |
5062 | ||
5063 | release_kernel_lock(prev); | |
5064 | need_resched_nonpreemptible: | |
5065 | ||
5066 | schedule_debug(prev); | |
1da177e4 | 5067 | |
31656519 | 5068 | if (sched_feat(HRTICK)) |
f333fdc9 | 5069 | hrtick_clear(rq); |
8f4d37ec | 5070 | |
8cd162ce | 5071 | spin_lock_irq(&rq->lock); |
3e51f33f | 5072 | update_rq_clock(rq); |
1e819950 | 5073 | clear_tsk_need_resched(prev); |
1da177e4 | 5074 | |
1da177e4 | 5075 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5076 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5077 | prev->state = TASK_RUNNING; |
16882c1e | 5078 | else |
2e1cb74a | 5079 | deactivate_task(rq, prev, 1); |
dd41f596 | 5080 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5081 | } |
5082 | ||
9a897c5a SR |
5083 | #ifdef CONFIG_SMP |
5084 | if (prev->sched_class->pre_schedule) | |
5085 | prev->sched_class->pre_schedule(rq, prev); | |
5086 | #endif | |
f65eda4f | 5087 | |
dd41f596 | 5088 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5089 | idle_balance(cpu, rq); |
1da177e4 | 5090 | |
df1c99d4 | 5091 | put_prev_task(rq, prev); |
b67802ea | 5092 | next = pick_next_task(rq); |
1da177e4 | 5093 | |
1da177e4 | 5094 | if (likely(prev != next)) { |
673a90a1 | 5095 | sched_info_switch(prev, next); |
564c2b21 | 5096 | perf_counter_task_sched_out(prev, next, cpu); |
673a90a1 | 5097 | |
1da177e4 LT |
5098 | rq->nr_switches++; |
5099 | rq->curr = next; | |
5100 | ++*switch_count; | |
5101 | ||
dd41f596 | 5102 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5103 | /* |
5104 | * the context switch might have flipped the stack from under | |
5105 | * us, hence refresh the local variables. | |
5106 | */ | |
5107 | cpu = smp_processor_id(); | |
5108 | rq = cpu_rq(cpu); | |
1da177e4 LT |
5109 | } else |
5110 | spin_unlock_irq(&rq->lock); | |
5111 | ||
8f4d37ec | 5112 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5113 | goto need_resched_nonpreemptible; |
41719b03 | 5114 | } |
8f4d37ec | 5115 | |
41719b03 PZ |
5116 | asmlinkage void __sched schedule(void) |
5117 | { | |
5118 | need_resched: | |
5119 | preempt_disable(); | |
5120 | __schedule(); | |
1da177e4 LT |
5121 | preempt_enable_no_resched(); |
5122 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
5123 | goto need_resched; | |
5124 | } | |
1da177e4 LT |
5125 | EXPORT_SYMBOL(schedule); |
5126 | ||
0d66bf6d PZ |
5127 | #ifdef CONFIG_SMP |
5128 | /* | |
5129 | * Look out! "owner" is an entirely speculative pointer | |
5130 | * access and not reliable. | |
5131 | */ | |
5132 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5133 | { | |
5134 | unsigned int cpu; | |
5135 | struct rq *rq; | |
5136 | ||
5137 | if (!sched_feat(OWNER_SPIN)) | |
5138 | return 0; | |
5139 | ||
5140 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5141 | /* | |
5142 | * Need to access the cpu field knowing that | |
5143 | * DEBUG_PAGEALLOC could have unmapped it if | |
5144 | * the mutex owner just released it and exited. | |
5145 | */ | |
5146 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5147 | goto out; | |
5148 | #else | |
5149 | cpu = owner->cpu; | |
5150 | #endif | |
5151 | ||
5152 | /* | |
5153 | * Even if the access succeeded (likely case), | |
5154 | * the cpu field may no longer be valid. | |
5155 | */ | |
5156 | if (cpu >= nr_cpumask_bits) | |
5157 | goto out; | |
5158 | ||
5159 | /* | |
5160 | * We need to validate that we can do a | |
5161 | * get_cpu() and that we have the percpu area. | |
5162 | */ | |
5163 | if (!cpu_online(cpu)) | |
5164 | goto out; | |
5165 | ||
5166 | rq = cpu_rq(cpu); | |
5167 | ||
5168 | for (;;) { | |
5169 | /* | |
5170 | * Owner changed, break to re-assess state. | |
5171 | */ | |
5172 | if (lock->owner != owner) | |
5173 | break; | |
5174 | ||
5175 | /* | |
5176 | * Is that owner really running on that cpu? | |
5177 | */ | |
5178 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5179 | return 0; | |
5180 | ||
5181 | cpu_relax(); | |
5182 | } | |
5183 | out: | |
5184 | return 1; | |
5185 | } | |
5186 | #endif | |
5187 | ||
1da177e4 LT |
5188 | #ifdef CONFIG_PREEMPT |
5189 | /* | |
2ed6e34f | 5190 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5191 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5192 | * occur there and call schedule directly. |
5193 | */ | |
5194 | asmlinkage void __sched preempt_schedule(void) | |
5195 | { | |
5196 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5197 | |
1da177e4 LT |
5198 | /* |
5199 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5200 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5201 | */ |
beed33a8 | 5202 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5203 | return; |
5204 | ||
3a5c359a AK |
5205 | do { |
5206 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5207 | schedule(); |
3a5c359a | 5208 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5209 | |
3a5c359a AK |
5210 | /* |
5211 | * Check again in case we missed a preemption opportunity | |
5212 | * between schedule and now. | |
5213 | */ | |
5214 | barrier(); | |
5ed0cec0 | 5215 | } while (need_resched()); |
1da177e4 | 5216 | } |
1da177e4 LT |
5217 | EXPORT_SYMBOL(preempt_schedule); |
5218 | ||
5219 | /* | |
2ed6e34f | 5220 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5221 | * off of irq context. |
5222 | * Note, that this is called and return with irqs disabled. This will | |
5223 | * protect us against recursive calling from irq. | |
5224 | */ | |
5225 | asmlinkage void __sched preempt_schedule_irq(void) | |
5226 | { | |
5227 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5228 | |
2ed6e34f | 5229 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5230 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5231 | ||
3a5c359a AK |
5232 | do { |
5233 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5234 | local_irq_enable(); |
5235 | schedule(); | |
5236 | local_irq_disable(); | |
3a5c359a | 5237 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5238 | |
3a5c359a AK |
5239 | /* |
5240 | * Check again in case we missed a preemption opportunity | |
5241 | * between schedule and now. | |
5242 | */ | |
5243 | barrier(); | |
5ed0cec0 | 5244 | } while (need_resched()); |
1da177e4 LT |
5245 | } |
5246 | ||
5247 | #endif /* CONFIG_PREEMPT */ | |
5248 | ||
95cdf3b7 IM |
5249 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5250 | void *key) | |
1da177e4 | 5251 | { |
48f24c4d | 5252 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5253 | } |
1da177e4 LT |
5254 | EXPORT_SYMBOL(default_wake_function); |
5255 | ||
5256 | /* | |
41a2d6cf IM |
5257 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5258 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5259 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5260 | * | |
5261 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5262 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5263 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5264 | */ | |
777c6c5f JW |
5265 | void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
5266 | int nr_exclusive, int sync, void *key) | |
1da177e4 | 5267 | { |
2e45874c | 5268 | wait_queue_t *curr, *next; |
1da177e4 | 5269 | |
2e45874c | 5270 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5271 | unsigned flags = curr->flags; |
5272 | ||
1da177e4 | 5273 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5274 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5275 | break; |
5276 | } | |
5277 | } | |
5278 | ||
5279 | /** | |
5280 | * __wake_up - wake up threads blocked on a waitqueue. | |
5281 | * @q: the waitqueue | |
5282 | * @mode: which threads | |
5283 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5284 | * @key: is directly passed to the wakeup function |
1da177e4 | 5285 | */ |
7ad5b3a5 | 5286 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5287 | int nr_exclusive, void *key) |
1da177e4 LT |
5288 | { |
5289 | unsigned long flags; | |
5290 | ||
5291 | spin_lock_irqsave(&q->lock, flags); | |
5292 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5293 | spin_unlock_irqrestore(&q->lock, flags); | |
5294 | } | |
1da177e4 LT |
5295 | EXPORT_SYMBOL(__wake_up); |
5296 | ||
5297 | /* | |
5298 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5299 | */ | |
7ad5b3a5 | 5300 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5301 | { |
5302 | __wake_up_common(q, mode, 1, 0, NULL); | |
5303 | } | |
5304 | ||
4ede816a DL |
5305 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5306 | { | |
5307 | __wake_up_common(q, mode, 1, 0, key); | |
5308 | } | |
5309 | ||
1da177e4 | 5310 | /** |
4ede816a | 5311 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5312 | * @q: the waitqueue |
5313 | * @mode: which threads | |
5314 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5315 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5316 | * |
5317 | * The sync wakeup differs that the waker knows that it will schedule | |
5318 | * away soon, so while the target thread will be woken up, it will not | |
5319 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5320 | * with each other. This can prevent needless bouncing between CPUs. | |
5321 | * | |
5322 | * On UP it can prevent extra preemption. | |
5323 | */ | |
4ede816a DL |
5324 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5325 | int nr_exclusive, void *key) | |
1da177e4 LT |
5326 | { |
5327 | unsigned long flags; | |
5328 | int sync = 1; | |
5329 | ||
5330 | if (unlikely(!q)) | |
5331 | return; | |
5332 | ||
5333 | if (unlikely(!nr_exclusive)) | |
5334 | sync = 0; | |
5335 | ||
5336 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5337 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5338 | spin_unlock_irqrestore(&q->lock, flags); |
5339 | } | |
4ede816a DL |
5340 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5341 | ||
5342 | /* | |
5343 | * __wake_up_sync - see __wake_up_sync_key() | |
5344 | */ | |
5345 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5346 | { | |
5347 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5348 | } | |
1da177e4 LT |
5349 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5350 | ||
65eb3dc6 KD |
5351 | /** |
5352 | * complete: - signals a single thread waiting on this completion | |
5353 | * @x: holds the state of this particular completion | |
5354 | * | |
5355 | * This will wake up a single thread waiting on this completion. Threads will be | |
5356 | * awakened in the same order in which they were queued. | |
5357 | * | |
5358 | * See also complete_all(), wait_for_completion() and related routines. | |
5359 | */ | |
b15136e9 | 5360 | void complete(struct completion *x) |
1da177e4 LT |
5361 | { |
5362 | unsigned long flags; | |
5363 | ||
5364 | spin_lock_irqsave(&x->wait.lock, flags); | |
5365 | x->done++; | |
d9514f6c | 5366 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5367 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5368 | } | |
5369 | EXPORT_SYMBOL(complete); | |
5370 | ||
65eb3dc6 KD |
5371 | /** |
5372 | * complete_all: - signals all threads waiting on this completion | |
5373 | * @x: holds the state of this particular completion | |
5374 | * | |
5375 | * This will wake up all threads waiting on this particular completion event. | |
5376 | */ | |
b15136e9 | 5377 | void complete_all(struct completion *x) |
1da177e4 LT |
5378 | { |
5379 | unsigned long flags; | |
5380 | ||
5381 | spin_lock_irqsave(&x->wait.lock, flags); | |
5382 | x->done += UINT_MAX/2; | |
d9514f6c | 5383 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5384 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5385 | } | |
5386 | EXPORT_SYMBOL(complete_all); | |
5387 | ||
8cbbe86d AK |
5388 | static inline long __sched |
5389 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5390 | { |
1da177e4 LT |
5391 | if (!x->done) { |
5392 | DECLARE_WAITQUEUE(wait, current); | |
5393 | ||
5394 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5395 | __add_wait_queue_tail(&x->wait, &wait); | |
5396 | do { | |
94d3d824 | 5397 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5398 | timeout = -ERESTARTSYS; |
5399 | break; | |
8cbbe86d AK |
5400 | } |
5401 | __set_current_state(state); | |
1da177e4 LT |
5402 | spin_unlock_irq(&x->wait.lock); |
5403 | timeout = schedule_timeout(timeout); | |
5404 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5405 | } while (!x->done && timeout); |
1da177e4 | 5406 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5407 | if (!x->done) |
5408 | return timeout; | |
1da177e4 LT |
5409 | } |
5410 | x->done--; | |
ea71a546 | 5411 | return timeout ?: 1; |
1da177e4 | 5412 | } |
1da177e4 | 5413 | |
8cbbe86d AK |
5414 | static long __sched |
5415 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5416 | { |
1da177e4 LT |
5417 | might_sleep(); |
5418 | ||
5419 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5420 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5421 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5422 | return timeout; |
5423 | } | |
1da177e4 | 5424 | |
65eb3dc6 KD |
5425 | /** |
5426 | * wait_for_completion: - waits for completion of a task | |
5427 | * @x: holds the state of this particular completion | |
5428 | * | |
5429 | * This waits to be signaled for completion of a specific task. It is NOT | |
5430 | * interruptible and there is no timeout. | |
5431 | * | |
5432 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5433 | * and interrupt capability. Also see complete(). | |
5434 | */ | |
b15136e9 | 5435 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5436 | { |
5437 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5438 | } |
8cbbe86d | 5439 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5440 | |
65eb3dc6 KD |
5441 | /** |
5442 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5443 | * @x: holds the state of this particular completion | |
5444 | * @timeout: timeout value in jiffies | |
5445 | * | |
5446 | * This waits for either a completion of a specific task to be signaled or for a | |
5447 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5448 | * interruptible. | |
5449 | */ | |
b15136e9 | 5450 | unsigned long __sched |
8cbbe86d | 5451 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5452 | { |
8cbbe86d | 5453 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5454 | } |
8cbbe86d | 5455 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5456 | |
65eb3dc6 KD |
5457 | /** |
5458 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5459 | * @x: holds the state of this particular completion | |
5460 | * | |
5461 | * This waits for completion of a specific task to be signaled. It is | |
5462 | * interruptible. | |
5463 | */ | |
8cbbe86d | 5464 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5465 | { |
51e97990 AK |
5466 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5467 | if (t == -ERESTARTSYS) | |
5468 | return t; | |
5469 | return 0; | |
0fec171c | 5470 | } |
8cbbe86d | 5471 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5472 | |
65eb3dc6 KD |
5473 | /** |
5474 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5475 | * @x: holds the state of this particular completion | |
5476 | * @timeout: timeout value in jiffies | |
5477 | * | |
5478 | * This waits for either a completion of a specific task to be signaled or for a | |
5479 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5480 | */ | |
b15136e9 | 5481 | unsigned long __sched |
8cbbe86d AK |
5482 | wait_for_completion_interruptible_timeout(struct completion *x, |
5483 | unsigned long timeout) | |
0fec171c | 5484 | { |
8cbbe86d | 5485 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5486 | } |
8cbbe86d | 5487 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 5488 | |
65eb3dc6 KD |
5489 | /** |
5490 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
5491 | * @x: holds the state of this particular completion | |
5492 | * | |
5493 | * This waits to be signaled for completion of a specific task. It can be | |
5494 | * interrupted by a kill signal. | |
5495 | */ | |
009e577e MW |
5496 | int __sched wait_for_completion_killable(struct completion *x) |
5497 | { | |
5498 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
5499 | if (t == -ERESTARTSYS) | |
5500 | return t; | |
5501 | return 0; | |
5502 | } | |
5503 | EXPORT_SYMBOL(wait_for_completion_killable); | |
5504 | ||
be4de352 DC |
5505 | /** |
5506 | * try_wait_for_completion - try to decrement a completion without blocking | |
5507 | * @x: completion structure | |
5508 | * | |
5509 | * Returns: 0 if a decrement cannot be done without blocking | |
5510 | * 1 if a decrement succeeded. | |
5511 | * | |
5512 | * If a completion is being used as a counting completion, | |
5513 | * attempt to decrement the counter without blocking. This | |
5514 | * enables us to avoid waiting if the resource the completion | |
5515 | * is protecting is not available. | |
5516 | */ | |
5517 | bool try_wait_for_completion(struct completion *x) | |
5518 | { | |
5519 | int ret = 1; | |
5520 | ||
5521 | spin_lock_irq(&x->wait.lock); | |
5522 | if (!x->done) | |
5523 | ret = 0; | |
5524 | else | |
5525 | x->done--; | |
5526 | spin_unlock_irq(&x->wait.lock); | |
5527 | return ret; | |
5528 | } | |
5529 | EXPORT_SYMBOL(try_wait_for_completion); | |
5530 | ||
5531 | /** | |
5532 | * completion_done - Test to see if a completion has any waiters | |
5533 | * @x: completion structure | |
5534 | * | |
5535 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
5536 | * 1 if there are no waiters. | |
5537 | * | |
5538 | */ | |
5539 | bool completion_done(struct completion *x) | |
5540 | { | |
5541 | int ret = 1; | |
5542 | ||
5543 | spin_lock_irq(&x->wait.lock); | |
5544 | if (!x->done) | |
5545 | ret = 0; | |
5546 | spin_unlock_irq(&x->wait.lock); | |
5547 | return ret; | |
5548 | } | |
5549 | EXPORT_SYMBOL(completion_done); | |
5550 | ||
8cbbe86d AK |
5551 | static long __sched |
5552 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 5553 | { |
0fec171c IM |
5554 | unsigned long flags; |
5555 | wait_queue_t wait; | |
5556 | ||
5557 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 5558 | |
8cbbe86d | 5559 | __set_current_state(state); |
1da177e4 | 5560 | |
8cbbe86d AK |
5561 | spin_lock_irqsave(&q->lock, flags); |
5562 | __add_wait_queue(q, &wait); | |
5563 | spin_unlock(&q->lock); | |
5564 | timeout = schedule_timeout(timeout); | |
5565 | spin_lock_irq(&q->lock); | |
5566 | __remove_wait_queue(q, &wait); | |
5567 | spin_unlock_irqrestore(&q->lock, flags); | |
5568 | ||
5569 | return timeout; | |
5570 | } | |
5571 | ||
5572 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
5573 | { | |
5574 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 5575 | } |
1da177e4 LT |
5576 | EXPORT_SYMBOL(interruptible_sleep_on); |
5577 | ||
0fec171c | 5578 | long __sched |
95cdf3b7 | 5579 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5580 | { |
8cbbe86d | 5581 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 5582 | } |
1da177e4 LT |
5583 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
5584 | ||
0fec171c | 5585 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 5586 | { |
8cbbe86d | 5587 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 5588 | } |
1da177e4 LT |
5589 | EXPORT_SYMBOL(sleep_on); |
5590 | ||
0fec171c | 5591 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 5592 | { |
8cbbe86d | 5593 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 5594 | } |
1da177e4 LT |
5595 | EXPORT_SYMBOL(sleep_on_timeout); |
5596 | ||
b29739f9 IM |
5597 | #ifdef CONFIG_RT_MUTEXES |
5598 | ||
5599 | /* | |
5600 | * rt_mutex_setprio - set the current priority of a task | |
5601 | * @p: task | |
5602 | * @prio: prio value (kernel-internal form) | |
5603 | * | |
5604 | * This function changes the 'effective' priority of a task. It does | |
5605 | * not touch ->normal_prio like __setscheduler(). | |
5606 | * | |
5607 | * Used by the rt_mutex code to implement priority inheritance logic. | |
5608 | */ | |
36c8b586 | 5609 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
5610 | { |
5611 | unsigned long flags; | |
83b699ed | 5612 | int oldprio, on_rq, running; |
70b97a7f | 5613 | struct rq *rq; |
cb469845 | 5614 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
5615 | |
5616 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
5617 | ||
5618 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5619 | update_rq_clock(rq); |
b29739f9 | 5620 | |
d5f9f942 | 5621 | oldprio = p->prio; |
dd41f596 | 5622 | on_rq = p->se.on_rq; |
051a1d1a | 5623 | running = task_current(rq, p); |
0e1f3483 | 5624 | if (on_rq) |
69be72c1 | 5625 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
5626 | if (running) |
5627 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
5628 | |
5629 | if (rt_prio(prio)) | |
5630 | p->sched_class = &rt_sched_class; | |
5631 | else | |
5632 | p->sched_class = &fair_sched_class; | |
5633 | ||
b29739f9 IM |
5634 | p->prio = prio; |
5635 | ||
0e1f3483 HS |
5636 | if (running) |
5637 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 5638 | if (on_rq) { |
8159f87e | 5639 | enqueue_task(rq, p, 0); |
cb469845 SR |
5640 | |
5641 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
5642 | } |
5643 | task_rq_unlock(rq, &flags); | |
5644 | } | |
5645 | ||
5646 | #endif | |
5647 | ||
36c8b586 | 5648 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 5649 | { |
dd41f596 | 5650 | int old_prio, delta, on_rq; |
1da177e4 | 5651 | unsigned long flags; |
70b97a7f | 5652 | struct rq *rq; |
1da177e4 LT |
5653 | |
5654 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
5655 | return; | |
5656 | /* | |
5657 | * We have to be careful, if called from sys_setpriority(), | |
5658 | * the task might be in the middle of scheduling on another CPU. | |
5659 | */ | |
5660 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 5661 | update_rq_clock(rq); |
1da177e4 LT |
5662 | /* |
5663 | * The RT priorities are set via sched_setscheduler(), but we still | |
5664 | * allow the 'normal' nice value to be set - but as expected | |
5665 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 5666 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 5667 | */ |
e05606d3 | 5668 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
5669 | p->static_prio = NICE_TO_PRIO(nice); |
5670 | goto out_unlock; | |
5671 | } | |
dd41f596 | 5672 | on_rq = p->se.on_rq; |
c09595f6 | 5673 | if (on_rq) |
69be72c1 | 5674 | dequeue_task(rq, p, 0); |
1da177e4 | 5675 | |
1da177e4 | 5676 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 5677 | set_load_weight(p); |
b29739f9 IM |
5678 | old_prio = p->prio; |
5679 | p->prio = effective_prio(p); | |
5680 | delta = p->prio - old_prio; | |
1da177e4 | 5681 | |
dd41f596 | 5682 | if (on_rq) { |
8159f87e | 5683 | enqueue_task(rq, p, 0); |
1da177e4 | 5684 | /* |
d5f9f942 AM |
5685 | * If the task increased its priority or is running and |
5686 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 5687 | */ |
d5f9f942 | 5688 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
5689 | resched_task(rq->curr); |
5690 | } | |
5691 | out_unlock: | |
5692 | task_rq_unlock(rq, &flags); | |
5693 | } | |
1da177e4 LT |
5694 | EXPORT_SYMBOL(set_user_nice); |
5695 | ||
e43379f1 MM |
5696 | /* |
5697 | * can_nice - check if a task can reduce its nice value | |
5698 | * @p: task | |
5699 | * @nice: nice value | |
5700 | */ | |
36c8b586 | 5701 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 5702 | { |
024f4747 MM |
5703 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
5704 | int nice_rlim = 20 - nice; | |
48f24c4d | 5705 | |
e43379f1 MM |
5706 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
5707 | capable(CAP_SYS_NICE)); | |
5708 | } | |
5709 | ||
1da177e4 LT |
5710 | #ifdef __ARCH_WANT_SYS_NICE |
5711 | ||
5712 | /* | |
5713 | * sys_nice - change the priority of the current process. | |
5714 | * @increment: priority increment | |
5715 | * | |
5716 | * sys_setpriority is a more generic, but much slower function that | |
5717 | * does similar things. | |
5718 | */ | |
5add95d4 | 5719 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 5720 | { |
48f24c4d | 5721 | long nice, retval; |
1da177e4 LT |
5722 | |
5723 | /* | |
5724 | * Setpriority might change our priority at the same moment. | |
5725 | * We don't have to worry. Conceptually one call occurs first | |
5726 | * and we have a single winner. | |
5727 | */ | |
e43379f1 MM |
5728 | if (increment < -40) |
5729 | increment = -40; | |
1da177e4 LT |
5730 | if (increment > 40) |
5731 | increment = 40; | |
5732 | ||
2b8f836f | 5733 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
5734 | if (nice < -20) |
5735 | nice = -20; | |
5736 | if (nice > 19) | |
5737 | nice = 19; | |
5738 | ||
e43379f1 MM |
5739 | if (increment < 0 && !can_nice(current, nice)) |
5740 | return -EPERM; | |
5741 | ||
1da177e4 LT |
5742 | retval = security_task_setnice(current, nice); |
5743 | if (retval) | |
5744 | return retval; | |
5745 | ||
5746 | set_user_nice(current, nice); | |
5747 | return 0; | |
5748 | } | |
5749 | ||
5750 | #endif | |
5751 | ||
5752 | /** | |
5753 | * task_prio - return the priority value of a given task. | |
5754 | * @p: the task in question. | |
5755 | * | |
5756 | * This is the priority value as seen by users in /proc. | |
5757 | * RT tasks are offset by -200. Normal tasks are centered | |
5758 | * around 0, value goes from -16 to +15. | |
5759 | */ | |
36c8b586 | 5760 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
5761 | { |
5762 | return p->prio - MAX_RT_PRIO; | |
5763 | } | |
5764 | ||
5765 | /** | |
5766 | * task_nice - return the nice value of a given task. | |
5767 | * @p: the task in question. | |
5768 | */ | |
36c8b586 | 5769 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
5770 | { |
5771 | return TASK_NICE(p); | |
5772 | } | |
150d8bed | 5773 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
5774 | |
5775 | /** | |
5776 | * idle_cpu - is a given cpu idle currently? | |
5777 | * @cpu: the processor in question. | |
5778 | */ | |
5779 | int idle_cpu(int cpu) | |
5780 | { | |
5781 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
5782 | } | |
5783 | ||
1da177e4 LT |
5784 | /** |
5785 | * idle_task - return the idle task for a given cpu. | |
5786 | * @cpu: the processor in question. | |
5787 | */ | |
36c8b586 | 5788 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
5789 | { |
5790 | return cpu_rq(cpu)->idle; | |
5791 | } | |
5792 | ||
5793 | /** | |
5794 | * find_process_by_pid - find a process with a matching PID value. | |
5795 | * @pid: the pid in question. | |
5796 | */ | |
a9957449 | 5797 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 5798 | { |
228ebcbe | 5799 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
5800 | } |
5801 | ||
5802 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
5803 | static void |
5804 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 5805 | { |
dd41f596 | 5806 | BUG_ON(p->se.on_rq); |
48f24c4d | 5807 | |
1da177e4 | 5808 | p->policy = policy; |
dd41f596 IM |
5809 | switch (p->policy) { |
5810 | case SCHED_NORMAL: | |
5811 | case SCHED_BATCH: | |
5812 | case SCHED_IDLE: | |
5813 | p->sched_class = &fair_sched_class; | |
5814 | break; | |
5815 | case SCHED_FIFO: | |
5816 | case SCHED_RR: | |
5817 | p->sched_class = &rt_sched_class; | |
5818 | break; | |
5819 | } | |
5820 | ||
1da177e4 | 5821 | p->rt_priority = prio; |
b29739f9 IM |
5822 | p->normal_prio = normal_prio(p); |
5823 | /* we are holding p->pi_lock already */ | |
5824 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 5825 | set_load_weight(p); |
1da177e4 LT |
5826 | } |
5827 | ||
c69e8d9c DH |
5828 | /* |
5829 | * check the target process has a UID that matches the current process's | |
5830 | */ | |
5831 | static bool check_same_owner(struct task_struct *p) | |
5832 | { | |
5833 | const struct cred *cred = current_cred(), *pcred; | |
5834 | bool match; | |
5835 | ||
5836 | rcu_read_lock(); | |
5837 | pcred = __task_cred(p); | |
5838 | match = (cred->euid == pcred->euid || | |
5839 | cred->euid == pcred->uid); | |
5840 | rcu_read_unlock(); | |
5841 | return match; | |
5842 | } | |
5843 | ||
961ccddd RR |
5844 | static int __sched_setscheduler(struct task_struct *p, int policy, |
5845 | struct sched_param *param, bool user) | |
1da177e4 | 5846 | { |
83b699ed | 5847 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5848 | unsigned long flags; |
cb469845 | 5849 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 5850 | struct rq *rq; |
1da177e4 | 5851 | |
66e5393a SR |
5852 | /* may grab non-irq protected spin_locks */ |
5853 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5854 | recheck: |
5855 | /* double check policy once rq lock held */ | |
5856 | if (policy < 0) | |
5857 | policy = oldpolicy = p->policy; | |
5858 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
5859 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
5860 | policy != SCHED_IDLE) | |
b0a9499c | 5861 | return -EINVAL; |
1da177e4 LT |
5862 | /* |
5863 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5864 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5865 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5866 | */ |
5867 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5868 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5869 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5870 | return -EINVAL; |
e05606d3 | 5871 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5872 | return -EINVAL; |
5873 | ||
37e4ab3f OC |
5874 | /* |
5875 | * Allow unprivileged RT tasks to decrease priority: | |
5876 | */ | |
961ccddd | 5877 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5878 | if (rt_policy(policy)) { |
8dc3e909 | 5879 | unsigned long rlim_rtprio; |
8dc3e909 ON |
5880 | |
5881 | if (!lock_task_sighand(p, &flags)) | |
5882 | return -ESRCH; | |
5883 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
5884 | unlock_task_sighand(p, &flags); | |
5885 | ||
5886 | /* can't set/change the rt policy */ | |
5887 | if (policy != p->policy && !rlim_rtprio) | |
5888 | return -EPERM; | |
5889 | ||
5890 | /* can't increase priority */ | |
5891 | if (param->sched_priority > p->rt_priority && | |
5892 | param->sched_priority > rlim_rtprio) | |
5893 | return -EPERM; | |
5894 | } | |
dd41f596 IM |
5895 | /* |
5896 | * Like positive nice levels, dont allow tasks to | |
5897 | * move out of SCHED_IDLE either: | |
5898 | */ | |
5899 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
5900 | return -EPERM; | |
5fe1d75f | 5901 | |
37e4ab3f | 5902 | /* can't change other user's priorities */ |
c69e8d9c | 5903 | if (!check_same_owner(p)) |
37e4ab3f OC |
5904 | return -EPERM; |
5905 | } | |
1da177e4 | 5906 | |
725aad24 | 5907 | if (user) { |
b68aa230 | 5908 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
5909 | /* |
5910 | * Do not allow realtime tasks into groups that have no runtime | |
5911 | * assigned. | |
5912 | */ | |
9a7e0b18 PZ |
5913 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
5914 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 5915 | return -EPERM; |
b68aa230 PZ |
5916 | #endif |
5917 | ||
725aad24 JF |
5918 | retval = security_task_setscheduler(p, policy, param); |
5919 | if (retval) | |
5920 | return retval; | |
5921 | } | |
5922 | ||
b29739f9 IM |
5923 | /* |
5924 | * make sure no PI-waiters arrive (or leave) while we are | |
5925 | * changing the priority of the task: | |
5926 | */ | |
5927 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
5928 | /* |
5929 | * To be able to change p->policy safely, the apropriate | |
5930 | * runqueue lock must be held. | |
5931 | */ | |
b29739f9 | 5932 | rq = __task_rq_lock(p); |
1da177e4 LT |
5933 | /* recheck policy now with rq lock held */ |
5934 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5935 | policy = oldpolicy = -1; | |
b29739f9 IM |
5936 | __task_rq_unlock(rq); |
5937 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
5938 | goto recheck; |
5939 | } | |
2daa3577 | 5940 | update_rq_clock(rq); |
dd41f596 | 5941 | on_rq = p->se.on_rq; |
051a1d1a | 5942 | running = task_current(rq, p); |
0e1f3483 | 5943 | if (on_rq) |
2e1cb74a | 5944 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5945 | if (running) |
5946 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5947 | |
1da177e4 | 5948 | oldprio = p->prio; |
dd41f596 | 5949 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5950 | |
0e1f3483 HS |
5951 | if (running) |
5952 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5953 | if (on_rq) { |
5954 | activate_task(rq, p, 0); | |
cb469845 SR |
5955 | |
5956 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5957 | } |
b29739f9 IM |
5958 | __task_rq_unlock(rq); |
5959 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
5960 | ||
95e02ca9 TG |
5961 | rt_mutex_adjust_pi(p); |
5962 | ||
1da177e4 LT |
5963 | return 0; |
5964 | } | |
961ccddd RR |
5965 | |
5966 | /** | |
5967 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5968 | * @p: the task in question. | |
5969 | * @policy: new policy. | |
5970 | * @param: structure containing the new RT priority. | |
5971 | * | |
5972 | * NOTE that the task may be already dead. | |
5973 | */ | |
5974 | int sched_setscheduler(struct task_struct *p, int policy, | |
5975 | struct sched_param *param) | |
5976 | { | |
5977 | return __sched_setscheduler(p, policy, param, true); | |
5978 | } | |
1da177e4 LT |
5979 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5980 | ||
961ccddd RR |
5981 | /** |
5982 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5983 | * @p: the task in question. | |
5984 | * @policy: new policy. | |
5985 | * @param: structure containing the new RT priority. | |
5986 | * | |
5987 | * Just like sched_setscheduler, only don't bother checking if the | |
5988 | * current context has permission. For example, this is needed in | |
5989 | * stop_machine(): we create temporary high priority worker threads, | |
5990 | * but our caller might not have that capability. | |
5991 | */ | |
5992 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5993 | struct sched_param *param) | |
5994 | { | |
5995 | return __sched_setscheduler(p, policy, param, false); | |
5996 | } | |
5997 | ||
95cdf3b7 IM |
5998 | static int |
5999 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6000 | { |
1da177e4 LT |
6001 | struct sched_param lparam; |
6002 | struct task_struct *p; | |
36c8b586 | 6003 | int retval; |
1da177e4 LT |
6004 | |
6005 | if (!param || pid < 0) | |
6006 | return -EINVAL; | |
6007 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6008 | return -EFAULT; | |
5fe1d75f ON |
6009 | |
6010 | rcu_read_lock(); | |
6011 | retval = -ESRCH; | |
1da177e4 | 6012 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6013 | if (p != NULL) |
6014 | retval = sched_setscheduler(p, policy, &lparam); | |
6015 | rcu_read_unlock(); | |
36c8b586 | 6016 | |
1da177e4 LT |
6017 | return retval; |
6018 | } | |
6019 | ||
6020 | /** | |
6021 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6022 | * @pid: the pid in question. | |
6023 | * @policy: new policy. | |
6024 | * @param: structure containing the new RT priority. | |
6025 | */ | |
5add95d4 HC |
6026 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6027 | struct sched_param __user *, param) | |
1da177e4 | 6028 | { |
c21761f1 JB |
6029 | /* negative values for policy are not valid */ |
6030 | if (policy < 0) | |
6031 | return -EINVAL; | |
6032 | ||
1da177e4 LT |
6033 | return do_sched_setscheduler(pid, policy, param); |
6034 | } | |
6035 | ||
6036 | /** | |
6037 | * sys_sched_setparam - set/change the RT priority of a thread | |
6038 | * @pid: the pid in question. | |
6039 | * @param: structure containing the new RT priority. | |
6040 | */ | |
5add95d4 | 6041 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6042 | { |
6043 | return do_sched_setscheduler(pid, -1, param); | |
6044 | } | |
6045 | ||
6046 | /** | |
6047 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6048 | * @pid: the pid in question. | |
6049 | */ | |
5add95d4 | 6050 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6051 | { |
36c8b586 | 6052 | struct task_struct *p; |
3a5c359a | 6053 | int retval; |
1da177e4 LT |
6054 | |
6055 | if (pid < 0) | |
3a5c359a | 6056 | return -EINVAL; |
1da177e4 LT |
6057 | |
6058 | retval = -ESRCH; | |
6059 | read_lock(&tasklist_lock); | |
6060 | p = find_process_by_pid(pid); | |
6061 | if (p) { | |
6062 | retval = security_task_getscheduler(p); | |
6063 | if (!retval) | |
6064 | retval = p->policy; | |
6065 | } | |
6066 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6067 | return retval; |
6068 | } | |
6069 | ||
6070 | /** | |
6071 | * sys_sched_getscheduler - get the RT priority of a thread | |
6072 | * @pid: the pid in question. | |
6073 | * @param: structure containing the RT priority. | |
6074 | */ | |
5add95d4 | 6075 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6076 | { |
6077 | struct sched_param lp; | |
36c8b586 | 6078 | struct task_struct *p; |
3a5c359a | 6079 | int retval; |
1da177e4 LT |
6080 | |
6081 | if (!param || pid < 0) | |
3a5c359a | 6082 | return -EINVAL; |
1da177e4 LT |
6083 | |
6084 | read_lock(&tasklist_lock); | |
6085 | p = find_process_by_pid(pid); | |
6086 | retval = -ESRCH; | |
6087 | if (!p) | |
6088 | goto out_unlock; | |
6089 | ||
6090 | retval = security_task_getscheduler(p); | |
6091 | if (retval) | |
6092 | goto out_unlock; | |
6093 | ||
6094 | lp.sched_priority = p->rt_priority; | |
6095 | read_unlock(&tasklist_lock); | |
6096 | ||
6097 | /* | |
6098 | * This one might sleep, we cannot do it with a spinlock held ... | |
6099 | */ | |
6100 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6101 | ||
1da177e4 LT |
6102 | return retval; |
6103 | ||
6104 | out_unlock: | |
6105 | read_unlock(&tasklist_lock); | |
6106 | return retval; | |
6107 | } | |
6108 | ||
96f874e2 | 6109 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6110 | { |
5a16f3d3 | 6111 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6112 | struct task_struct *p; |
6113 | int retval; | |
1da177e4 | 6114 | |
95402b38 | 6115 | get_online_cpus(); |
1da177e4 LT |
6116 | read_lock(&tasklist_lock); |
6117 | ||
6118 | p = find_process_by_pid(pid); | |
6119 | if (!p) { | |
6120 | read_unlock(&tasklist_lock); | |
95402b38 | 6121 | put_online_cpus(); |
1da177e4 LT |
6122 | return -ESRCH; |
6123 | } | |
6124 | ||
6125 | /* | |
6126 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6127 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6128 | * usage count and then drop tasklist_lock. |
6129 | */ | |
6130 | get_task_struct(p); | |
6131 | read_unlock(&tasklist_lock); | |
6132 | ||
5a16f3d3 RR |
6133 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6134 | retval = -ENOMEM; | |
6135 | goto out_put_task; | |
6136 | } | |
6137 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6138 | retval = -ENOMEM; | |
6139 | goto out_free_cpus_allowed; | |
6140 | } | |
1da177e4 | 6141 | retval = -EPERM; |
c69e8d9c | 6142 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6143 | goto out_unlock; |
6144 | ||
e7834f8f DQ |
6145 | retval = security_task_setscheduler(p, 0, NULL); |
6146 | if (retval) | |
6147 | goto out_unlock; | |
6148 | ||
5a16f3d3 RR |
6149 | cpuset_cpus_allowed(p, cpus_allowed); |
6150 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6151 | again: |
5a16f3d3 | 6152 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6153 | |
8707d8b8 | 6154 | if (!retval) { |
5a16f3d3 RR |
6155 | cpuset_cpus_allowed(p, cpus_allowed); |
6156 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6157 | /* |
6158 | * We must have raced with a concurrent cpuset | |
6159 | * update. Just reset the cpus_allowed to the | |
6160 | * cpuset's cpus_allowed | |
6161 | */ | |
5a16f3d3 | 6162 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6163 | goto again; |
6164 | } | |
6165 | } | |
1da177e4 | 6166 | out_unlock: |
5a16f3d3 RR |
6167 | free_cpumask_var(new_mask); |
6168 | out_free_cpus_allowed: | |
6169 | free_cpumask_var(cpus_allowed); | |
6170 | out_put_task: | |
1da177e4 | 6171 | put_task_struct(p); |
95402b38 | 6172 | put_online_cpus(); |
1da177e4 LT |
6173 | return retval; |
6174 | } | |
6175 | ||
6176 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6177 | struct cpumask *new_mask) |
1da177e4 | 6178 | { |
96f874e2 RR |
6179 | if (len < cpumask_size()) |
6180 | cpumask_clear(new_mask); | |
6181 | else if (len > cpumask_size()) | |
6182 | len = cpumask_size(); | |
6183 | ||
1da177e4 LT |
6184 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6185 | } | |
6186 | ||
6187 | /** | |
6188 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6189 | * @pid: pid of the process | |
6190 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6191 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6192 | */ | |
5add95d4 HC |
6193 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6194 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6195 | { |
5a16f3d3 | 6196 | cpumask_var_t new_mask; |
1da177e4 LT |
6197 | int retval; |
6198 | ||
5a16f3d3 RR |
6199 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6200 | return -ENOMEM; | |
1da177e4 | 6201 | |
5a16f3d3 RR |
6202 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6203 | if (retval == 0) | |
6204 | retval = sched_setaffinity(pid, new_mask); | |
6205 | free_cpumask_var(new_mask); | |
6206 | return retval; | |
1da177e4 LT |
6207 | } |
6208 | ||
96f874e2 | 6209 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6210 | { |
36c8b586 | 6211 | struct task_struct *p; |
1da177e4 | 6212 | int retval; |
1da177e4 | 6213 | |
95402b38 | 6214 | get_online_cpus(); |
1da177e4 LT |
6215 | read_lock(&tasklist_lock); |
6216 | ||
6217 | retval = -ESRCH; | |
6218 | p = find_process_by_pid(pid); | |
6219 | if (!p) | |
6220 | goto out_unlock; | |
6221 | ||
e7834f8f DQ |
6222 | retval = security_task_getscheduler(p); |
6223 | if (retval) | |
6224 | goto out_unlock; | |
6225 | ||
96f874e2 | 6226 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6227 | |
6228 | out_unlock: | |
6229 | read_unlock(&tasklist_lock); | |
95402b38 | 6230 | put_online_cpus(); |
1da177e4 | 6231 | |
9531b62f | 6232 | return retval; |
1da177e4 LT |
6233 | } |
6234 | ||
6235 | /** | |
6236 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6237 | * @pid: pid of the process | |
6238 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6239 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6240 | */ | |
5add95d4 HC |
6241 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6242 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6243 | { |
6244 | int ret; | |
f17c8607 | 6245 | cpumask_var_t mask; |
1da177e4 | 6246 | |
f17c8607 | 6247 | if (len < cpumask_size()) |
1da177e4 LT |
6248 | return -EINVAL; |
6249 | ||
f17c8607 RR |
6250 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6251 | return -ENOMEM; | |
1da177e4 | 6252 | |
f17c8607 RR |
6253 | ret = sched_getaffinity(pid, mask); |
6254 | if (ret == 0) { | |
6255 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6256 | ret = -EFAULT; | |
6257 | else | |
6258 | ret = cpumask_size(); | |
6259 | } | |
6260 | free_cpumask_var(mask); | |
1da177e4 | 6261 | |
f17c8607 | 6262 | return ret; |
1da177e4 LT |
6263 | } |
6264 | ||
6265 | /** | |
6266 | * sys_sched_yield - yield the current processor to other threads. | |
6267 | * | |
dd41f596 IM |
6268 | * This function yields the current CPU to other tasks. If there are no |
6269 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6270 | */ |
5add95d4 | 6271 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6272 | { |
70b97a7f | 6273 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6274 | |
2d72376b | 6275 | schedstat_inc(rq, yld_count); |
4530d7ab | 6276 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6277 | |
6278 | /* | |
6279 | * Since we are going to call schedule() anyway, there's | |
6280 | * no need to preempt or enable interrupts: | |
6281 | */ | |
6282 | __release(rq->lock); | |
8a25d5de | 6283 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6284 | _raw_spin_unlock(&rq->lock); |
6285 | preempt_enable_no_resched(); | |
6286 | ||
6287 | schedule(); | |
6288 | ||
6289 | return 0; | |
6290 | } | |
6291 | ||
e7b38404 | 6292 | static void __cond_resched(void) |
1da177e4 | 6293 | { |
8e0a43d8 IM |
6294 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
6295 | __might_sleep(__FILE__, __LINE__); | |
6296 | #endif | |
5bbcfd90 IM |
6297 | /* |
6298 | * The BKS might be reacquired before we have dropped | |
6299 | * PREEMPT_ACTIVE, which could trigger a second | |
6300 | * cond_resched() call. | |
6301 | */ | |
1da177e4 LT |
6302 | do { |
6303 | add_preempt_count(PREEMPT_ACTIVE); | |
6304 | schedule(); | |
6305 | sub_preempt_count(PREEMPT_ACTIVE); | |
6306 | } while (need_resched()); | |
6307 | } | |
6308 | ||
02b67cc3 | 6309 | int __sched _cond_resched(void) |
1da177e4 | 6310 | { |
9414232f IM |
6311 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
6312 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
6313 | __cond_resched(); |
6314 | return 1; | |
6315 | } | |
6316 | return 0; | |
6317 | } | |
02b67cc3 | 6318 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6319 | |
6320 | /* | |
6321 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
6322 | * call schedule, and on return reacquire the lock. | |
6323 | * | |
41a2d6cf | 6324 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6325 | * operations here to prevent schedule() from being called twice (once via |
6326 | * spin_unlock(), once by hand). | |
6327 | */ | |
95cdf3b7 | 6328 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6329 | { |
95c354fe | 6330 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
6331 | int ret = 0; |
6332 | ||
95c354fe | 6333 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6334 | spin_unlock(lock); |
95c354fe NP |
6335 | if (resched && need_resched()) |
6336 | __cond_resched(); | |
6337 | else | |
6338 | cpu_relax(); | |
6df3cecb | 6339 | ret = 1; |
1da177e4 | 6340 | spin_lock(lock); |
1da177e4 | 6341 | } |
6df3cecb | 6342 | return ret; |
1da177e4 | 6343 | } |
1da177e4 LT |
6344 | EXPORT_SYMBOL(cond_resched_lock); |
6345 | ||
6346 | int __sched cond_resched_softirq(void) | |
6347 | { | |
6348 | BUG_ON(!in_softirq()); | |
6349 | ||
9414232f | 6350 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 6351 | local_bh_enable(); |
1da177e4 LT |
6352 | __cond_resched(); |
6353 | local_bh_disable(); | |
6354 | return 1; | |
6355 | } | |
6356 | return 0; | |
6357 | } | |
1da177e4 LT |
6358 | EXPORT_SYMBOL(cond_resched_softirq); |
6359 | ||
1da177e4 LT |
6360 | /** |
6361 | * yield - yield the current processor to other threads. | |
6362 | * | |
72fd4a35 | 6363 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6364 | * thread runnable and calls sys_sched_yield(). |
6365 | */ | |
6366 | void __sched yield(void) | |
6367 | { | |
6368 | set_current_state(TASK_RUNNING); | |
6369 | sys_sched_yield(); | |
6370 | } | |
1da177e4 LT |
6371 | EXPORT_SYMBOL(yield); |
6372 | ||
6373 | /* | |
41a2d6cf | 6374 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6375 | * that process accounting knows that this is a task in IO wait state. |
6376 | * | |
6377 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6378 | * has set its backing_dev_info: the queue against which it should throttle) | |
6379 | */ | |
6380 | void __sched io_schedule(void) | |
6381 | { | |
70b97a7f | 6382 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 6383 | |
0ff92245 | 6384 | delayacct_blkio_start(); |
1da177e4 LT |
6385 | atomic_inc(&rq->nr_iowait); |
6386 | schedule(); | |
6387 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6388 | delayacct_blkio_end(); |
1da177e4 | 6389 | } |
1da177e4 LT |
6390 | EXPORT_SYMBOL(io_schedule); |
6391 | ||
6392 | long __sched io_schedule_timeout(long timeout) | |
6393 | { | |
70b97a7f | 6394 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
6395 | long ret; |
6396 | ||
0ff92245 | 6397 | delayacct_blkio_start(); |
1da177e4 LT |
6398 | atomic_inc(&rq->nr_iowait); |
6399 | ret = schedule_timeout(timeout); | |
6400 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 6401 | delayacct_blkio_end(); |
1da177e4 LT |
6402 | return ret; |
6403 | } | |
6404 | ||
6405 | /** | |
6406 | * sys_sched_get_priority_max - return maximum RT priority. | |
6407 | * @policy: scheduling class. | |
6408 | * | |
6409 | * this syscall returns the maximum rt_priority that can be used | |
6410 | * by a given scheduling class. | |
6411 | */ | |
5add95d4 | 6412 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6413 | { |
6414 | int ret = -EINVAL; | |
6415 | ||
6416 | switch (policy) { | |
6417 | case SCHED_FIFO: | |
6418 | case SCHED_RR: | |
6419 | ret = MAX_USER_RT_PRIO-1; | |
6420 | break; | |
6421 | case SCHED_NORMAL: | |
b0a9499c | 6422 | case SCHED_BATCH: |
dd41f596 | 6423 | case SCHED_IDLE: |
1da177e4 LT |
6424 | ret = 0; |
6425 | break; | |
6426 | } | |
6427 | return ret; | |
6428 | } | |
6429 | ||
6430 | /** | |
6431 | * sys_sched_get_priority_min - return minimum RT priority. | |
6432 | * @policy: scheduling class. | |
6433 | * | |
6434 | * this syscall returns the minimum rt_priority that can be used | |
6435 | * by a given scheduling class. | |
6436 | */ | |
5add95d4 | 6437 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6438 | { |
6439 | int ret = -EINVAL; | |
6440 | ||
6441 | switch (policy) { | |
6442 | case SCHED_FIFO: | |
6443 | case SCHED_RR: | |
6444 | ret = 1; | |
6445 | break; | |
6446 | case SCHED_NORMAL: | |
b0a9499c | 6447 | case SCHED_BATCH: |
dd41f596 | 6448 | case SCHED_IDLE: |
1da177e4 LT |
6449 | ret = 0; |
6450 | } | |
6451 | return ret; | |
6452 | } | |
6453 | ||
6454 | /** | |
6455 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6456 | * @pid: pid of the process. | |
6457 | * @interval: userspace pointer to the timeslice value. | |
6458 | * | |
6459 | * this syscall writes the default timeslice value of a given process | |
6460 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6461 | */ | |
17da2bd9 | 6462 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6463 | struct timespec __user *, interval) |
1da177e4 | 6464 | { |
36c8b586 | 6465 | struct task_struct *p; |
a4ec24b4 | 6466 | unsigned int time_slice; |
3a5c359a | 6467 | int retval; |
1da177e4 | 6468 | struct timespec t; |
1da177e4 LT |
6469 | |
6470 | if (pid < 0) | |
3a5c359a | 6471 | return -EINVAL; |
1da177e4 LT |
6472 | |
6473 | retval = -ESRCH; | |
6474 | read_lock(&tasklist_lock); | |
6475 | p = find_process_by_pid(pid); | |
6476 | if (!p) | |
6477 | goto out_unlock; | |
6478 | ||
6479 | retval = security_task_getscheduler(p); | |
6480 | if (retval) | |
6481 | goto out_unlock; | |
6482 | ||
77034937 IM |
6483 | /* |
6484 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
6485 | * tasks that are on an otherwise idle runqueue: | |
6486 | */ | |
6487 | time_slice = 0; | |
6488 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 6489 | time_slice = DEF_TIMESLICE; |
1868f958 | 6490 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
6491 | struct sched_entity *se = &p->se; |
6492 | unsigned long flags; | |
6493 | struct rq *rq; | |
6494 | ||
6495 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
6496 | if (rq->cfs.load.weight) |
6497 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
6498 | task_rq_unlock(rq, &flags); |
6499 | } | |
1da177e4 | 6500 | read_unlock(&tasklist_lock); |
a4ec24b4 | 6501 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 6502 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 6503 | return retval; |
3a5c359a | 6504 | |
1da177e4 LT |
6505 | out_unlock: |
6506 | read_unlock(&tasklist_lock); | |
6507 | return retval; | |
6508 | } | |
6509 | ||
7c731e0a | 6510 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 6511 | |
82a1fcb9 | 6512 | void sched_show_task(struct task_struct *p) |
1da177e4 | 6513 | { |
1da177e4 | 6514 | unsigned long free = 0; |
36c8b586 | 6515 | unsigned state; |
1da177e4 | 6516 | |
1da177e4 | 6517 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 6518 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 6519 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 6520 | #if BITS_PER_LONG == 32 |
1da177e4 | 6521 | if (state == TASK_RUNNING) |
cc4ea795 | 6522 | printk(KERN_CONT " running "); |
1da177e4 | 6523 | else |
cc4ea795 | 6524 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
6525 | #else |
6526 | if (state == TASK_RUNNING) | |
cc4ea795 | 6527 | printk(KERN_CONT " running task "); |
1da177e4 | 6528 | else |
cc4ea795 | 6529 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
6530 | #endif |
6531 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 6532 | free = stack_not_used(p); |
1da177e4 | 6533 | #endif |
ba25f9dc | 6534 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 6535 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 6536 | |
5fb5e6de | 6537 | show_stack(p, NULL); |
1da177e4 LT |
6538 | } |
6539 | ||
e59e2ae2 | 6540 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 6541 | { |
36c8b586 | 6542 | struct task_struct *g, *p; |
1da177e4 | 6543 | |
4bd77321 IM |
6544 | #if BITS_PER_LONG == 32 |
6545 | printk(KERN_INFO | |
6546 | " task PC stack pid father\n"); | |
1da177e4 | 6547 | #else |
4bd77321 IM |
6548 | printk(KERN_INFO |
6549 | " task PC stack pid father\n"); | |
1da177e4 LT |
6550 | #endif |
6551 | read_lock(&tasklist_lock); | |
6552 | do_each_thread(g, p) { | |
6553 | /* | |
6554 | * reset the NMI-timeout, listing all files on a slow | |
6555 | * console might take alot of time: | |
6556 | */ | |
6557 | touch_nmi_watchdog(); | |
39bc89fd | 6558 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 6559 | sched_show_task(p); |
1da177e4 LT |
6560 | } while_each_thread(g, p); |
6561 | ||
04c9167f JF |
6562 | touch_all_softlockup_watchdogs(); |
6563 | ||
dd41f596 IM |
6564 | #ifdef CONFIG_SCHED_DEBUG |
6565 | sysrq_sched_debug_show(); | |
6566 | #endif | |
1da177e4 | 6567 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
6568 | /* |
6569 | * Only show locks if all tasks are dumped: | |
6570 | */ | |
6571 | if (state_filter == -1) | |
6572 | debug_show_all_locks(); | |
1da177e4 LT |
6573 | } |
6574 | ||
1df21055 IM |
6575 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
6576 | { | |
dd41f596 | 6577 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
6578 | } |
6579 | ||
f340c0d1 IM |
6580 | /** |
6581 | * init_idle - set up an idle thread for a given CPU | |
6582 | * @idle: task in question | |
6583 | * @cpu: cpu the idle task belongs to | |
6584 | * | |
6585 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
6586 | * flag, to make booting more robust. | |
6587 | */ | |
5c1e1767 | 6588 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 6589 | { |
70b97a7f | 6590 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6591 | unsigned long flags; |
6592 | ||
5cbd54ef IM |
6593 | spin_lock_irqsave(&rq->lock, flags); |
6594 | ||
dd41f596 IM |
6595 | __sched_fork(idle); |
6596 | idle->se.exec_start = sched_clock(); | |
6597 | ||
b29739f9 | 6598 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 6599 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 6600 | __set_task_cpu(idle, cpu); |
1da177e4 | 6601 | |
1da177e4 | 6602 | rq->curr = rq->idle = idle; |
4866cde0 NP |
6603 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
6604 | idle->oncpu = 1; | |
6605 | #endif | |
1da177e4 LT |
6606 | spin_unlock_irqrestore(&rq->lock, flags); |
6607 | ||
6608 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
6609 | #if defined(CONFIG_PREEMPT) |
6610 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
6611 | #else | |
a1261f54 | 6612 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 6613 | #endif |
dd41f596 IM |
6614 | /* |
6615 | * The idle tasks have their own, simple scheduling class: | |
6616 | */ | |
6617 | idle->sched_class = &idle_sched_class; | |
fb52607a | 6618 | ftrace_graph_init_task(idle); |
1da177e4 LT |
6619 | } |
6620 | ||
6621 | /* | |
6622 | * In a system that switches off the HZ timer nohz_cpu_mask | |
6623 | * indicates which cpus entered this state. This is used | |
6624 | * in the rcu update to wait only for active cpus. For system | |
6625 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 6626 | * always be CPU_BITS_NONE. |
1da177e4 | 6627 | */ |
6a7b3dc3 | 6628 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 6629 | |
19978ca6 IM |
6630 | /* |
6631 | * Increase the granularity value when there are more CPUs, | |
6632 | * because with more CPUs the 'effective latency' as visible | |
6633 | * to users decreases. But the relationship is not linear, | |
6634 | * so pick a second-best guess by going with the log2 of the | |
6635 | * number of CPUs. | |
6636 | * | |
6637 | * This idea comes from the SD scheduler of Con Kolivas: | |
6638 | */ | |
6639 | static inline void sched_init_granularity(void) | |
6640 | { | |
6641 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
6642 | const unsigned long limit = 200000000; | |
6643 | ||
6644 | sysctl_sched_min_granularity *= factor; | |
6645 | if (sysctl_sched_min_granularity > limit) | |
6646 | sysctl_sched_min_granularity = limit; | |
6647 | ||
6648 | sysctl_sched_latency *= factor; | |
6649 | if (sysctl_sched_latency > limit) | |
6650 | sysctl_sched_latency = limit; | |
6651 | ||
6652 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
6653 | |
6654 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
6655 | } |
6656 | ||
1da177e4 LT |
6657 | #ifdef CONFIG_SMP |
6658 | /* | |
6659 | * This is how migration works: | |
6660 | * | |
70b97a7f | 6661 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
6662 | * runqueue and wake up that CPU's migration thread. |
6663 | * 2) we down() the locked semaphore => thread blocks. | |
6664 | * 3) migration thread wakes up (implicitly it forces the migrated | |
6665 | * thread off the CPU) | |
6666 | * 4) it gets the migration request and checks whether the migrated | |
6667 | * task is still in the wrong runqueue. | |
6668 | * 5) if it's in the wrong runqueue then the migration thread removes | |
6669 | * it and puts it into the right queue. | |
6670 | * 6) migration thread up()s the semaphore. | |
6671 | * 7) we wake up and the migration is done. | |
6672 | */ | |
6673 | ||
6674 | /* | |
6675 | * Change a given task's CPU affinity. Migrate the thread to a | |
6676 | * proper CPU and schedule it away if the CPU it's executing on | |
6677 | * is removed from the allowed bitmask. | |
6678 | * | |
6679 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 6680 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
6681 | * call is not atomic; no spinlocks may be held. |
6682 | */ | |
96f874e2 | 6683 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 6684 | { |
70b97a7f | 6685 | struct migration_req req; |
1da177e4 | 6686 | unsigned long flags; |
70b97a7f | 6687 | struct rq *rq; |
48f24c4d | 6688 | int ret = 0; |
1da177e4 LT |
6689 | |
6690 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 6691 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
6692 | ret = -EINVAL; |
6693 | goto out; | |
6694 | } | |
6695 | ||
9985b0ba | 6696 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 6697 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
6698 | ret = -EINVAL; |
6699 | goto out; | |
6700 | } | |
6701 | ||
73fe6aae | 6702 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 6703 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 6704 | else { |
96f874e2 RR |
6705 | cpumask_copy(&p->cpus_allowed, new_mask); |
6706 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
6707 | } |
6708 | ||
1da177e4 | 6709 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6710 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6711 | goto out; |
6712 | ||
1e5ce4f4 | 6713 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 LT |
6714 | /* Need help from migration thread: drop lock and wait. */ |
6715 | task_rq_unlock(rq, &flags); | |
6716 | wake_up_process(rq->migration_thread); | |
6717 | wait_for_completion(&req.done); | |
6718 | tlb_migrate_finish(p->mm); | |
6719 | return 0; | |
6720 | } | |
6721 | out: | |
6722 | task_rq_unlock(rq, &flags); | |
48f24c4d | 6723 | |
1da177e4 LT |
6724 | return ret; |
6725 | } | |
cd8ba7cd | 6726 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6727 | |
6728 | /* | |
41a2d6cf | 6729 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6730 | * this because either it can't run here any more (set_cpus_allowed() |
6731 | * away from this CPU, or CPU going down), or because we're | |
6732 | * attempting to rebalance this task on exec (sched_exec). | |
6733 | * | |
6734 | * So we race with normal scheduler movements, but that's OK, as long | |
6735 | * as the task is no longer on this CPU. | |
efc30814 KK |
6736 | * |
6737 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6738 | */ |
efc30814 | 6739 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6740 | { |
70b97a7f | 6741 | struct rq *rq_dest, *rq_src; |
dd41f596 | 6742 | int ret = 0, on_rq; |
1da177e4 | 6743 | |
e761b772 | 6744 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6745 | return ret; |
1da177e4 LT |
6746 | |
6747 | rq_src = cpu_rq(src_cpu); | |
6748 | rq_dest = cpu_rq(dest_cpu); | |
6749 | ||
6750 | double_rq_lock(rq_src, rq_dest); | |
6751 | /* Already moved. */ | |
6752 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6753 | goto done; |
1da177e4 | 6754 | /* Affinity changed (again). */ |
96f874e2 | 6755 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6756 | goto fail; |
1da177e4 | 6757 | |
dd41f596 | 6758 | on_rq = p->se.on_rq; |
6e82a3be | 6759 | if (on_rq) |
2e1cb74a | 6760 | deactivate_task(rq_src, p, 0); |
6e82a3be | 6761 | |
1da177e4 | 6762 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
6763 | if (on_rq) { |
6764 | activate_task(rq_dest, p, 0); | |
15afe09b | 6765 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6766 | } |
b1e38734 | 6767 | done: |
efc30814 | 6768 | ret = 1; |
b1e38734 | 6769 | fail: |
1da177e4 | 6770 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 6771 | return ret; |
1da177e4 LT |
6772 | } |
6773 | ||
6774 | /* | |
6775 | * migration_thread - this is a highprio system thread that performs | |
6776 | * thread migration by bumping thread off CPU then 'pushing' onto | |
6777 | * another runqueue. | |
6778 | */ | |
95cdf3b7 | 6779 | static int migration_thread(void *data) |
1da177e4 | 6780 | { |
1da177e4 | 6781 | int cpu = (long)data; |
70b97a7f | 6782 | struct rq *rq; |
1da177e4 LT |
6783 | |
6784 | rq = cpu_rq(cpu); | |
6785 | BUG_ON(rq->migration_thread != current); | |
6786 | ||
6787 | set_current_state(TASK_INTERRUPTIBLE); | |
6788 | while (!kthread_should_stop()) { | |
70b97a7f | 6789 | struct migration_req *req; |
1da177e4 | 6790 | struct list_head *head; |
1da177e4 | 6791 | |
1da177e4 LT |
6792 | spin_lock_irq(&rq->lock); |
6793 | ||
6794 | if (cpu_is_offline(cpu)) { | |
6795 | spin_unlock_irq(&rq->lock); | |
6796 | goto wait_to_die; | |
6797 | } | |
6798 | ||
6799 | if (rq->active_balance) { | |
6800 | active_load_balance(rq, cpu); | |
6801 | rq->active_balance = 0; | |
6802 | } | |
6803 | ||
6804 | head = &rq->migration_queue; | |
6805 | ||
6806 | if (list_empty(head)) { | |
6807 | spin_unlock_irq(&rq->lock); | |
6808 | schedule(); | |
6809 | set_current_state(TASK_INTERRUPTIBLE); | |
6810 | continue; | |
6811 | } | |
70b97a7f | 6812 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
6813 | list_del_init(head->next); |
6814 | ||
674311d5 NP |
6815 | spin_unlock(&rq->lock); |
6816 | __migrate_task(req->task, cpu, req->dest_cpu); | |
6817 | local_irq_enable(); | |
1da177e4 LT |
6818 | |
6819 | complete(&req->done); | |
6820 | } | |
6821 | __set_current_state(TASK_RUNNING); | |
6822 | return 0; | |
6823 | ||
6824 | wait_to_die: | |
6825 | /* Wait for kthread_stop */ | |
6826 | set_current_state(TASK_INTERRUPTIBLE); | |
6827 | while (!kthread_should_stop()) { | |
6828 | schedule(); | |
6829 | set_current_state(TASK_INTERRUPTIBLE); | |
6830 | } | |
6831 | __set_current_state(TASK_RUNNING); | |
6832 | return 0; | |
6833 | } | |
6834 | ||
6835 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
6836 | |
6837 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
6838 | { | |
6839 | int ret; | |
6840 | ||
6841 | local_irq_disable(); | |
6842 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
6843 | local_irq_enable(); | |
6844 | return ret; | |
6845 | } | |
6846 | ||
054b9108 | 6847 | /* |
3a4fa0a2 | 6848 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 6849 | */ |
48f24c4d | 6850 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 6851 | { |
70b97a7f | 6852 | int dest_cpu; |
6ca09dfc | 6853 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
1da177e4 | 6854 | |
e76bd8d9 RR |
6855 | again: |
6856 | /* Look for allowed, online CPU in same node. */ | |
6857 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
6858 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
6859 | goto move; | |
3a5c359a | 6860 | |
e76bd8d9 RR |
6861 | /* Any allowed, online CPU? */ |
6862 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
6863 | if (dest_cpu < nr_cpu_ids) | |
6864 | goto move; | |
3a5c359a | 6865 | |
e76bd8d9 RR |
6866 | /* No more Mr. Nice Guy. */ |
6867 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
6868 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
6869 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 6870 | |
e76bd8d9 RR |
6871 | /* |
6872 | * Don't tell them about moving exiting tasks or | |
6873 | * kernel threads (both mm NULL), since they never | |
6874 | * leave kernel. | |
6875 | */ | |
6876 | if (p->mm && printk_ratelimit()) { | |
6877 | printk(KERN_INFO "process %d (%s) no " | |
6878 | "longer affine to cpu%d\n", | |
6879 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 6880 | } |
e76bd8d9 RR |
6881 | } |
6882 | ||
6883 | move: | |
6884 | /* It can have affinity changed while we were choosing. */ | |
6885 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
6886 | goto again; | |
1da177e4 LT |
6887 | } |
6888 | ||
6889 | /* | |
6890 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6891 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6892 | * for performance reasons the counter is not stricly tracking tasks to | |
6893 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6894 | * to keep the global sum constant after CPU-down: | |
6895 | */ | |
70b97a7f | 6896 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6897 | { |
1e5ce4f4 | 6898 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
6899 | unsigned long flags; |
6900 | ||
6901 | local_irq_save(flags); | |
6902 | double_rq_lock(rq_src, rq_dest); | |
6903 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
6904 | rq_src->nr_uninterruptible = 0; | |
6905 | double_rq_unlock(rq_src, rq_dest); | |
6906 | local_irq_restore(flags); | |
6907 | } | |
6908 | ||
6909 | /* Run through task list and migrate tasks from the dead cpu. */ | |
6910 | static void migrate_live_tasks(int src_cpu) | |
6911 | { | |
48f24c4d | 6912 | struct task_struct *p, *t; |
1da177e4 | 6913 | |
f7b4cddc | 6914 | read_lock(&tasklist_lock); |
1da177e4 | 6915 | |
48f24c4d IM |
6916 | do_each_thread(t, p) { |
6917 | if (p == current) | |
1da177e4 LT |
6918 | continue; |
6919 | ||
48f24c4d IM |
6920 | if (task_cpu(p) == src_cpu) |
6921 | move_task_off_dead_cpu(src_cpu, p); | |
6922 | } while_each_thread(t, p); | |
1da177e4 | 6923 | |
f7b4cddc | 6924 | read_unlock(&tasklist_lock); |
1da177e4 LT |
6925 | } |
6926 | ||
dd41f596 IM |
6927 | /* |
6928 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
6929 | * It does so by boosting its priority to highest possible. |
6930 | * Used by CPU offline code. | |
1da177e4 LT |
6931 | */ |
6932 | void sched_idle_next(void) | |
6933 | { | |
48f24c4d | 6934 | int this_cpu = smp_processor_id(); |
70b97a7f | 6935 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
6936 | struct task_struct *p = rq->idle; |
6937 | unsigned long flags; | |
6938 | ||
6939 | /* cpu has to be offline */ | |
48f24c4d | 6940 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 6941 | |
48f24c4d IM |
6942 | /* |
6943 | * Strictly not necessary since rest of the CPUs are stopped by now | |
6944 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
6945 | */ |
6946 | spin_lock_irqsave(&rq->lock, flags); | |
6947 | ||
dd41f596 | 6948 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6949 | |
94bc9a7b DA |
6950 | update_rq_clock(rq); |
6951 | activate_task(rq, p, 0); | |
1da177e4 LT |
6952 | |
6953 | spin_unlock_irqrestore(&rq->lock, flags); | |
6954 | } | |
6955 | ||
48f24c4d IM |
6956 | /* |
6957 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6958 | * offline. |
6959 | */ | |
6960 | void idle_task_exit(void) | |
6961 | { | |
6962 | struct mm_struct *mm = current->active_mm; | |
6963 | ||
6964 | BUG_ON(cpu_online(smp_processor_id())); | |
6965 | ||
6966 | if (mm != &init_mm) | |
6967 | switch_mm(mm, &init_mm, current); | |
6968 | mmdrop(mm); | |
6969 | } | |
6970 | ||
054b9108 | 6971 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6972 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6973 | { |
70b97a7f | 6974 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6975 | |
6976 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6977 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6978 | |
6979 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6980 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6981 | |
48f24c4d | 6982 | get_task_struct(p); |
1da177e4 LT |
6983 | |
6984 | /* | |
6985 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6986 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6987 | * fine. |
6988 | */ | |
f7b4cddc | 6989 | spin_unlock_irq(&rq->lock); |
48f24c4d | 6990 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 6991 | spin_lock_irq(&rq->lock); |
1da177e4 | 6992 | |
48f24c4d | 6993 | put_task_struct(p); |
1da177e4 LT |
6994 | } |
6995 | ||
6996 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6997 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6998 | { | |
70b97a7f | 6999 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7000 | struct task_struct *next; |
48f24c4d | 7001 | |
dd41f596 IM |
7002 | for ( ; ; ) { |
7003 | if (!rq->nr_running) | |
7004 | break; | |
a8e504d2 | 7005 | update_rq_clock(rq); |
b67802ea | 7006 | next = pick_next_task(rq); |
dd41f596 IM |
7007 | if (!next) |
7008 | break; | |
79c53799 | 7009 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7010 | migrate_dead(dead_cpu, next); |
e692ab53 | 7011 | |
1da177e4 LT |
7012 | } |
7013 | } | |
7014 | #endif /* CONFIG_HOTPLUG_CPU */ | |
7015 | ||
e692ab53 NP |
7016 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7017 | ||
7018 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7019 | { |
7020 | .procname = "sched_domain", | |
c57baf1e | 7021 | .mode = 0555, |
e0361851 | 7022 | }, |
38605cae | 7023 | {0, }, |
e692ab53 NP |
7024 | }; |
7025 | ||
7026 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7027 | { |
c57baf1e | 7028 | .ctl_name = CTL_KERN, |
e0361851 | 7029 | .procname = "kernel", |
c57baf1e | 7030 | .mode = 0555, |
e0361851 AD |
7031 | .child = sd_ctl_dir, |
7032 | }, | |
38605cae | 7033 | {0, }, |
e692ab53 NP |
7034 | }; |
7035 | ||
7036 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7037 | { | |
7038 | struct ctl_table *entry = | |
5cf9f062 | 7039 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7040 | |
e692ab53 NP |
7041 | return entry; |
7042 | } | |
7043 | ||
6382bc90 MM |
7044 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7045 | { | |
cd790076 | 7046 | struct ctl_table *entry; |
6382bc90 | 7047 | |
cd790076 MM |
7048 | /* |
7049 | * In the intermediate directories, both the child directory and | |
7050 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7051 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7052 | * static strings and all have proc handlers. |
7053 | */ | |
7054 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7055 | if (entry->child) |
7056 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7057 | if (entry->proc_handler == NULL) |
7058 | kfree(entry->procname); | |
7059 | } | |
6382bc90 MM |
7060 | |
7061 | kfree(*tablep); | |
7062 | *tablep = NULL; | |
7063 | } | |
7064 | ||
e692ab53 | 7065 | static void |
e0361851 | 7066 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7067 | const char *procname, void *data, int maxlen, |
7068 | mode_t mode, proc_handler *proc_handler) | |
7069 | { | |
e692ab53 NP |
7070 | entry->procname = procname; |
7071 | entry->data = data; | |
7072 | entry->maxlen = maxlen; | |
7073 | entry->mode = mode; | |
7074 | entry->proc_handler = proc_handler; | |
7075 | } | |
7076 | ||
7077 | static struct ctl_table * | |
7078 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7079 | { | |
a5d8c348 | 7080 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7081 | |
ad1cdc1d MM |
7082 | if (table == NULL) |
7083 | return NULL; | |
7084 | ||
e0361851 | 7085 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7086 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7087 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7088 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7089 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7090 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7091 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7092 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7093 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7094 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7095 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7096 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7097 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7098 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7099 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7100 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7101 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7102 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7103 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7104 | &sd->cache_nice_tries, |
7105 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7106 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7107 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7108 | set_table_entry(&table[11], "name", sd->name, |
7109 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7110 | /* &table[12] is terminator */ | |
e692ab53 NP |
7111 | |
7112 | return table; | |
7113 | } | |
7114 | ||
9a4e7159 | 7115 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7116 | { |
7117 | struct ctl_table *entry, *table; | |
7118 | struct sched_domain *sd; | |
7119 | int domain_num = 0, i; | |
7120 | char buf[32]; | |
7121 | ||
7122 | for_each_domain(cpu, sd) | |
7123 | domain_num++; | |
7124 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7125 | if (table == NULL) |
7126 | return NULL; | |
e692ab53 NP |
7127 | |
7128 | i = 0; | |
7129 | for_each_domain(cpu, sd) { | |
7130 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7131 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7132 | entry->mode = 0555; |
e692ab53 NP |
7133 | entry->child = sd_alloc_ctl_domain_table(sd); |
7134 | entry++; | |
7135 | i++; | |
7136 | } | |
7137 | return table; | |
7138 | } | |
7139 | ||
7140 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7141 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7142 | { |
7143 | int i, cpu_num = num_online_cpus(); | |
7144 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7145 | char buf[32]; | |
7146 | ||
7378547f MM |
7147 | WARN_ON(sd_ctl_dir[0].child); |
7148 | sd_ctl_dir[0].child = entry; | |
7149 | ||
ad1cdc1d MM |
7150 | if (entry == NULL) |
7151 | return; | |
7152 | ||
97b6ea7b | 7153 | for_each_online_cpu(i) { |
e692ab53 | 7154 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7155 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7156 | entry->mode = 0555; |
e692ab53 | 7157 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7158 | entry++; |
e692ab53 | 7159 | } |
7378547f MM |
7160 | |
7161 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7162 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7163 | } | |
6382bc90 | 7164 | |
7378547f | 7165 | /* may be called multiple times per register */ |
6382bc90 MM |
7166 | static void unregister_sched_domain_sysctl(void) |
7167 | { | |
7378547f MM |
7168 | if (sd_sysctl_header) |
7169 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7170 | sd_sysctl_header = NULL; |
7378547f MM |
7171 | if (sd_ctl_dir[0].child) |
7172 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7173 | } |
e692ab53 | 7174 | #else |
6382bc90 MM |
7175 | static void register_sched_domain_sysctl(void) |
7176 | { | |
7177 | } | |
7178 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7179 | { |
7180 | } | |
7181 | #endif | |
7182 | ||
1f11eb6a GH |
7183 | static void set_rq_online(struct rq *rq) |
7184 | { | |
7185 | if (!rq->online) { | |
7186 | const struct sched_class *class; | |
7187 | ||
c6c4927b | 7188 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7189 | rq->online = 1; |
7190 | ||
7191 | for_each_class(class) { | |
7192 | if (class->rq_online) | |
7193 | class->rq_online(rq); | |
7194 | } | |
7195 | } | |
7196 | } | |
7197 | ||
7198 | static void set_rq_offline(struct rq *rq) | |
7199 | { | |
7200 | if (rq->online) { | |
7201 | const struct sched_class *class; | |
7202 | ||
7203 | for_each_class(class) { | |
7204 | if (class->rq_offline) | |
7205 | class->rq_offline(rq); | |
7206 | } | |
7207 | ||
c6c4927b | 7208 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7209 | rq->online = 0; |
7210 | } | |
7211 | } | |
7212 | ||
1da177e4 LT |
7213 | /* |
7214 | * migration_call - callback that gets triggered when a CPU is added. | |
7215 | * Here we can start up the necessary migration thread for the new CPU. | |
7216 | */ | |
48f24c4d IM |
7217 | static int __cpuinit |
7218 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7219 | { |
1da177e4 | 7220 | struct task_struct *p; |
48f24c4d | 7221 | int cpu = (long)hcpu; |
1da177e4 | 7222 | unsigned long flags; |
70b97a7f | 7223 | struct rq *rq; |
1da177e4 LT |
7224 | |
7225 | switch (action) { | |
5be9361c | 7226 | |
1da177e4 | 7227 | case CPU_UP_PREPARE: |
8bb78442 | 7228 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7229 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7230 | if (IS_ERR(p)) |
7231 | return NOTIFY_BAD; | |
1da177e4 LT |
7232 | kthread_bind(p, cpu); |
7233 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7234 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7235 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
7236 | task_rq_unlock(rq, &flags); |
7237 | cpu_rq(cpu)->migration_thread = p; | |
7238 | break; | |
48f24c4d | 7239 | |
1da177e4 | 7240 | case CPU_ONLINE: |
8bb78442 | 7241 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7242 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7243 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7244 | |
7245 | /* Update our root-domain */ | |
7246 | rq = cpu_rq(cpu); | |
7247 | spin_lock_irqsave(&rq->lock, flags); | |
7248 | if (rq->rd) { | |
c6c4927b | 7249 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7250 | |
7251 | set_rq_online(rq); | |
1f94ef59 GH |
7252 | } |
7253 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7254 | break; |
48f24c4d | 7255 | |
1da177e4 LT |
7256 | #ifdef CONFIG_HOTPLUG_CPU |
7257 | case CPU_UP_CANCELED: | |
8bb78442 | 7258 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7259 | if (!cpu_rq(cpu)->migration_thread) |
7260 | break; | |
41a2d6cf | 7261 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7262 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7263 | cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7264 | kthread_stop(cpu_rq(cpu)->migration_thread); |
7265 | cpu_rq(cpu)->migration_thread = NULL; | |
7266 | break; | |
48f24c4d | 7267 | |
1da177e4 | 7268 | case CPU_DEAD: |
8bb78442 | 7269 | case CPU_DEAD_FROZEN: |
470fd646 | 7270 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7271 | migrate_live_tasks(cpu); |
7272 | rq = cpu_rq(cpu); | |
7273 | kthread_stop(rq->migration_thread); | |
7274 | rq->migration_thread = NULL; | |
7275 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7276 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7277 | update_rq_clock(rq); |
2e1cb74a | 7278 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7279 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7280 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7281 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7282 | migrate_dead_tasks(cpu); |
d2da272a | 7283 | spin_unlock_irq(&rq->lock); |
470fd646 | 7284 | cpuset_unlock(); |
1da177e4 LT |
7285 | migrate_nr_uninterruptible(rq); |
7286 | BUG_ON(rq->nr_running != 0); | |
7287 | ||
41a2d6cf IM |
7288 | /* |
7289 | * No need to migrate the tasks: it was best-effort if | |
7290 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7291 | * the requestors. | |
7292 | */ | |
1da177e4 LT |
7293 | spin_lock_irq(&rq->lock); |
7294 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7295 | struct migration_req *req; |
7296 | ||
1da177e4 | 7297 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7298 | struct migration_req, list); |
1da177e4 | 7299 | list_del_init(&req->list); |
9a2bd244 | 7300 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7301 | complete(&req->done); |
9a2bd244 | 7302 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7303 | } |
7304 | spin_unlock_irq(&rq->lock); | |
7305 | break; | |
57d885fe | 7306 | |
08f503b0 GH |
7307 | case CPU_DYING: |
7308 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7309 | /* Update our root-domain */ |
7310 | rq = cpu_rq(cpu); | |
7311 | spin_lock_irqsave(&rq->lock, flags); | |
7312 | if (rq->rd) { | |
c6c4927b | 7313 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7314 | set_rq_offline(rq); |
57d885fe GH |
7315 | } |
7316 | spin_unlock_irqrestore(&rq->lock, flags); | |
7317 | break; | |
1da177e4 LT |
7318 | #endif |
7319 | } | |
7320 | return NOTIFY_OK; | |
7321 | } | |
7322 | ||
f38b0820 PM |
7323 | /* |
7324 | * Register at high priority so that task migration (migrate_all_tasks) | |
7325 | * happens before everything else. This has to be lower priority than | |
7326 | * the notifier in the perf_counter subsystem, though. | |
1da177e4 | 7327 | */ |
26c2143b | 7328 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7329 | .notifier_call = migration_call, |
7330 | .priority = 10 | |
7331 | }; | |
7332 | ||
7babe8db | 7333 | static int __init migration_init(void) |
1da177e4 LT |
7334 | { |
7335 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7336 | int err; |
48f24c4d IM |
7337 | |
7338 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7339 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7340 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7341 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7342 | register_cpu_notifier(&migration_notifier); | |
7babe8db EGM |
7343 | |
7344 | return err; | |
1da177e4 | 7345 | } |
7babe8db | 7346 | early_initcall(migration_init); |
1da177e4 LT |
7347 | #endif |
7348 | ||
7349 | #ifdef CONFIG_SMP | |
476f3534 | 7350 | |
3e9830dc | 7351 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7352 | |
7c16ec58 | 7353 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7354 | struct cpumask *groupmask) |
1da177e4 | 7355 | { |
4dcf6aff | 7356 | struct sched_group *group = sd->groups; |
434d53b0 | 7357 | char str[256]; |
1da177e4 | 7358 | |
968ea6d8 | 7359 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7360 | cpumask_clear(groupmask); |
4dcf6aff IM |
7361 | |
7362 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7363 | ||
7364 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7365 | printk("does not load-balance\n"); | |
7366 | if (sd->parent) | |
7367 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7368 | " has parent"); | |
7369 | return -1; | |
41c7ce9a NP |
7370 | } |
7371 | ||
eefd796a | 7372 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7373 | |
758b2cdc | 7374 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7375 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7376 | "CPU%d\n", cpu); | |
7377 | } | |
758b2cdc | 7378 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7379 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7380 | " CPU%d\n", cpu); | |
7381 | } | |
1da177e4 | 7382 | |
4dcf6aff | 7383 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7384 | do { |
4dcf6aff IM |
7385 | if (!group) { |
7386 | printk("\n"); | |
7387 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7388 | break; |
7389 | } | |
7390 | ||
4dcf6aff IM |
7391 | if (!group->__cpu_power) { |
7392 | printk(KERN_CONT "\n"); | |
7393 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7394 | "set\n"); | |
7395 | break; | |
7396 | } | |
1da177e4 | 7397 | |
758b2cdc | 7398 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7399 | printk(KERN_CONT "\n"); |
7400 | printk(KERN_ERR "ERROR: empty group\n"); | |
7401 | break; | |
7402 | } | |
1da177e4 | 7403 | |
758b2cdc | 7404 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7405 | printk(KERN_CONT "\n"); |
7406 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7407 | break; | |
7408 | } | |
1da177e4 | 7409 | |
758b2cdc | 7410 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7411 | |
968ea6d8 | 7412 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 7413 | |
4dcf6aff | 7414 | printk(KERN_CONT " %s", str); |
381512cf GS |
7415 | if (group->__cpu_power != SCHED_LOAD_SCALE) { |
7416 | printk(KERN_CONT " (__cpu_power = %d)", | |
7417 | group->__cpu_power); | |
7418 | } | |
1da177e4 | 7419 | |
4dcf6aff IM |
7420 | group = group->next; |
7421 | } while (group != sd->groups); | |
7422 | printk(KERN_CONT "\n"); | |
1da177e4 | 7423 | |
758b2cdc | 7424 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7425 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7426 | |
758b2cdc RR |
7427 | if (sd->parent && |
7428 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7429 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7430 | "of domain->span\n"); | |
7431 | return 0; | |
7432 | } | |
1da177e4 | 7433 | |
4dcf6aff IM |
7434 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7435 | { | |
d5dd3db1 | 7436 | cpumask_var_t groupmask; |
4dcf6aff | 7437 | int level = 0; |
1da177e4 | 7438 | |
4dcf6aff IM |
7439 | if (!sd) { |
7440 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7441 | return; | |
7442 | } | |
1da177e4 | 7443 | |
4dcf6aff IM |
7444 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7445 | ||
d5dd3db1 | 7446 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7447 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7448 | return; | |
7449 | } | |
7450 | ||
4dcf6aff | 7451 | for (;;) { |
7c16ec58 | 7452 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7453 | break; |
1da177e4 LT |
7454 | level++; |
7455 | sd = sd->parent; | |
33859f7f | 7456 | if (!sd) |
4dcf6aff IM |
7457 | break; |
7458 | } | |
d5dd3db1 | 7459 | free_cpumask_var(groupmask); |
1da177e4 | 7460 | } |
6d6bc0ad | 7461 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7462 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7463 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7464 | |
1a20ff27 | 7465 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 7466 | { |
758b2cdc | 7467 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
7468 | return 1; |
7469 | ||
7470 | /* Following flags need at least 2 groups */ | |
7471 | if (sd->flags & (SD_LOAD_BALANCE | | |
7472 | SD_BALANCE_NEWIDLE | | |
7473 | SD_BALANCE_FORK | | |
89c4710e SS |
7474 | SD_BALANCE_EXEC | |
7475 | SD_SHARE_CPUPOWER | | |
7476 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
7477 | if (sd->groups != sd->groups->next) |
7478 | return 0; | |
7479 | } | |
7480 | ||
7481 | /* Following flags don't use groups */ | |
7482 | if (sd->flags & (SD_WAKE_IDLE | | |
7483 | SD_WAKE_AFFINE | | |
7484 | SD_WAKE_BALANCE)) | |
7485 | return 0; | |
7486 | ||
7487 | return 1; | |
7488 | } | |
7489 | ||
48f24c4d IM |
7490 | static int |
7491 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
7492 | { |
7493 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
7494 | ||
7495 | if (sd_degenerate(parent)) | |
7496 | return 1; | |
7497 | ||
758b2cdc | 7498 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
7499 | return 0; |
7500 | ||
7501 | /* Does parent contain flags not in child? */ | |
7502 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
7503 | if (cflags & SD_WAKE_AFFINE) | |
7504 | pflags &= ~SD_WAKE_BALANCE; | |
7505 | /* Flags needing groups don't count if only 1 group in parent */ | |
7506 | if (parent->groups == parent->groups->next) { | |
7507 | pflags &= ~(SD_LOAD_BALANCE | | |
7508 | SD_BALANCE_NEWIDLE | | |
7509 | SD_BALANCE_FORK | | |
89c4710e SS |
7510 | SD_BALANCE_EXEC | |
7511 | SD_SHARE_CPUPOWER | | |
7512 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
7513 | if (nr_node_ids == 1) |
7514 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
7515 | } |
7516 | if (~cflags & pflags) | |
7517 | return 0; | |
7518 | ||
7519 | return 1; | |
7520 | } | |
7521 | ||
c6c4927b RR |
7522 | static void free_rootdomain(struct root_domain *rd) |
7523 | { | |
68e74568 RR |
7524 | cpupri_cleanup(&rd->cpupri); |
7525 | ||
c6c4927b RR |
7526 | free_cpumask_var(rd->rto_mask); |
7527 | free_cpumask_var(rd->online); | |
7528 | free_cpumask_var(rd->span); | |
7529 | kfree(rd); | |
7530 | } | |
7531 | ||
57d885fe GH |
7532 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
7533 | { | |
a0490fa3 | 7534 | struct root_domain *old_rd = NULL; |
57d885fe | 7535 | unsigned long flags; |
57d885fe GH |
7536 | |
7537 | spin_lock_irqsave(&rq->lock, flags); | |
7538 | ||
7539 | if (rq->rd) { | |
a0490fa3 | 7540 | old_rd = rq->rd; |
57d885fe | 7541 | |
c6c4927b | 7542 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 7543 | set_rq_offline(rq); |
57d885fe | 7544 | |
c6c4927b | 7545 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 7546 | |
a0490fa3 IM |
7547 | /* |
7548 | * If we dont want to free the old_rt yet then | |
7549 | * set old_rd to NULL to skip the freeing later | |
7550 | * in this function: | |
7551 | */ | |
7552 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
7553 | old_rd = NULL; | |
57d885fe GH |
7554 | } |
7555 | ||
7556 | atomic_inc(&rd->refcount); | |
7557 | rq->rd = rd; | |
7558 | ||
c6c4927b RR |
7559 | cpumask_set_cpu(rq->cpu, rd->span); |
7560 | if (cpumask_test_cpu(rq->cpu, cpu_online_mask)) | |
1f11eb6a | 7561 | set_rq_online(rq); |
57d885fe GH |
7562 | |
7563 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
7564 | |
7565 | if (old_rd) | |
7566 | free_rootdomain(old_rd); | |
57d885fe GH |
7567 | } |
7568 | ||
db2f59c8 | 7569 | static int __init_refok init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe GH |
7570 | { |
7571 | memset(rd, 0, sizeof(*rd)); | |
7572 | ||
c6c4927b RR |
7573 | if (bootmem) { |
7574 | alloc_bootmem_cpumask_var(&def_root_domain.span); | |
7575 | alloc_bootmem_cpumask_var(&def_root_domain.online); | |
7576 | alloc_bootmem_cpumask_var(&def_root_domain.rto_mask); | |
68e74568 | 7577 | cpupri_init(&rd->cpupri, true); |
c6c4927b RR |
7578 | return 0; |
7579 | } | |
7580 | ||
7581 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) | |
0c910d28 | 7582 | goto out; |
c6c4927b RR |
7583 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
7584 | goto free_span; | |
7585 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) | |
7586 | goto free_online; | |
6e0534f2 | 7587 | |
68e74568 RR |
7588 | if (cpupri_init(&rd->cpupri, false) != 0) |
7589 | goto free_rto_mask; | |
c6c4927b | 7590 | return 0; |
6e0534f2 | 7591 | |
68e74568 RR |
7592 | free_rto_mask: |
7593 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
7594 | free_online: |
7595 | free_cpumask_var(rd->online); | |
7596 | free_span: | |
7597 | free_cpumask_var(rd->span); | |
0c910d28 | 7598 | out: |
c6c4927b | 7599 | return -ENOMEM; |
57d885fe GH |
7600 | } |
7601 | ||
7602 | static void init_defrootdomain(void) | |
7603 | { | |
c6c4927b RR |
7604 | init_rootdomain(&def_root_domain, true); |
7605 | ||
57d885fe GH |
7606 | atomic_set(&def_root_domain.refcount, 1); |
7607 | } | |
7608 | ||
dc938520 | 7609 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
7610 | { |
7611 | struct root_domain *rd; | |
7612 | ||
7613 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
7614 | if (!rd) | |
7615 | return NULL; | |
7616 | ||
c6c4927b RR |
7617 | if (init_rootdomain(rd, false) != 0) { |
7618 | kfree(rd); | |
7619 | return NULL; | |
7620 | } | |
57d885fe GH |
7621 | |
7622 | return rd; | |
7623 | } | |
7624 | ||
1da177e4 | 7625 | /* |
0eab9146 | 7626 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
7627 | * hold the hotplug lock. |
7628 | */ | |
0eab9146 IM |
7629 | static void |
7630 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 7631 | { |
70b97a7f | 7632 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
7633 | struct sched_domain *tmp; |
7634 | ||
7635 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 7636 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
7637 | struct sched_domain *parent = tmp->parent; |
7638 | if (!parent) | |
7639 | break; | |
f29c9b1c | 7640 | |
1a848870 | 7641 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 7642 | tmp->parent = parent->parent; |
1a848870 SS |
7643 | if (parent->parent) |
7644 | parent->parent->child = tmp; | |
f29c9b1c LZ |
7645 | } else |
7646 | tmp = tmp->parent; | |
245af2c7 SS |
7647 | } |
7648 | ||
1a848870 | 7649 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 7650 | sd = sd->parent; |
1a848870 SS |
7651 | if (sd) |
7652 | sd->child = NULL; | |
7653 | } | |
1da177e4 LT |
7654 | |
7655 | sched_domain_debug(sd, cpu); | |
7656 | ||
57d885fe | 7657 | rq_attach_root(rq, rd); |
674311d5 | 7658 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
7659 | } |
7660 | ||
7661 | /* cpus with isolated domains */ | |
dcc30a35 | 7662 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
7663 | |
7664 | /* Setup the mask of cpus configured for isolated domains */ | |
7665 | static int __init isolated_cpu_setup(char *str) | |
7666 | { | |
968ea6d8 | 7667 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
7668 | return 1; |
7669 | } | |
7670 | ||
8927f494 | 7671 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
7672 | |
7673 | /* | |
6711cab4 SS |
7674 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
7675 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
7676 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
7677 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
7678 | * |
7679 | * init_sched_build_groups will build a circular linked list of the groups | |
7680 | * covered by the given span, and will set each group's ->cpumask correctly, | |
7681 | * and ->cpu_power to 0. | |
7682 | */ | |
a616058b | 7683 | static void |
96f874e2 RR |
7684 | init_sched_build_groups(const struct cpumask *span, |
7685 | const struct cpumask *cpu_map, | |
7686 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 7687 | struct sched_group **sg, |
96f874e2 RR |
7688 | struct cpumask *tmpmask), |
7689 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
7690 | { |
7691 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
7692 | int i; |
7693 | ||
96f874e2 | 7694 | cpumask_clear(covered); |
7c16ec58 | 7695 | |
abcd083a | 7696 | for_each_cpu(i, span) { |
6711cab4 | 7697 | struct sched_group *sg; |
7c16ec58 | 7698 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
7699 | int j; |
7700 | ||
758b2cdc | 7701 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
7702 | continue; |
7703 | ||
758b2cdc | 7704 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 7705 | sg->__cpu_power = 0; |
1da177e4 | 7706 | |
abcd083a | 7707 | for_each_cpu(j, span) { |
7c16ec58 | 7708 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
7709 | continue; |
7710 | ||
96f874e2 | 7711 | cpumask_set_cpu(j, covered); |
758b2cdc | 7712 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
7713 | } |
7714 | if (!first) | |
7715 | first = sg; | |
7716 | if (last) | |
7717 | last->next = sg; | |
7718 | last = sg; | |
7719 | } | |
7720 | last->next = first; | |
7721 | } | |
7722 | ||
9c1cfda2 | 7723 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 7724 | |
9c1cfda2 | 7725 | #ifdef CONFIG_NUMA |
198e2f18 | 7726 | |
9c1cfda2 JH |
7727 | /** |
7728 | * find_next_best_node - find the next node to include in a sched_domain | |
7729 | * @node: node whose sched_domain we're building | |
7730 | * @used_nodes: nodes already in the sched_domain | |
7731 | * | |
41a2d6cf | 7732 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
7733 | * finds the closest node not already in the @used_nodes map. |
7734 | * | |
7735 | * Should use nodemask_t. | |
7736 | */ | |
c5f59f08 | 7737 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
7738 | { |
7739 | int i, n, val, min_val, best_node = 0; | |
7740 | ||
7741 | min_val = INT_MAX; | |
7742 | ||
076ac2af | 7743 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 7744 | /* Start at @node */ |
076ac2af | 7745 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
7746 | |
7747 | if (!nr_cpus_node(n)) | |
7748 | continue; | |
7749 | ||
7750 | /* Skip already used nodes */ | |
c5f59f08 | 7751 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
7752 | continue; |
7753 | ||
7754 | /* Simple min distance search */ | |
7755 | val = node_distance(node, n); | |
7756 | ||
7757 | if (val < min_val) { | |
7758 | min_val = val; | |
7759 | best_node = n; | |
7760 | } | |
7761 | } | |
7762 | ||
c5f59f08 | 7763 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
7764 | return best_node; |
7765 | } | |
7766 | ||
7767 | /** | |
7768 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
7769 | * @node: node whose cpumask we're constructing | |
73486722 | 7770 | * @span: resulting cpumask |
9c1cfda2 | 7771 | * |
41a2d6cf | 7772 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
7773 | * should be one that prevents unnecessary balancing, but also spreads tasks |
7774 | * out optimally. | |
7775 | */ | |
96f874e2 | 7776 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 7777 | { |
c5f59f08 | 7778 | nodemask_t used_nodes; |
48f24c4d | 7779 | int i; |
9c1cfda2 | 7780 | |
6ca09dfc | 7781 | cpumask_clear(span); |
c5f59f08 | 7782 | nodes_clear(used_nodes); |
9c1cfda2 | 7783 | |
6ca09dfc | 7784 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 7785 | node_set(node, used_nodes); |
9c1cfda2 JH |
7786 | |
7787 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 7788 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 7789 | |
6ca09dfc | 7790 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 7791 | } |
9c1cfda2 | 7792 | } |
6d6bc0ad | 7793 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 7794 | |
5c45bf27 | 7795 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 7796 | |
6c99e9ad RR |
7797 | /* |
7798 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
7799 | * FIXME: use cpumask_var_t or dynamic percpu alloc to avoid wasting space | |
7800 | * for nr_cpu_ids < CONFIG_NR_CPUS. | |
7801 | */ | |
7802 | struct static_sched_group { | |
7803 | struct sched_group sg; | |
7804 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
7805 | }; | |
7806 | ||
7807 | struct static_sched_domain { | |
7808 | struct sched_domain sd; | |
7809 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
7810 | }; | |
7811 | ||
9c1cfda2 | 7812 | /* |
48f24c4d | 7813 | * SMT sched-domains: |
9c1cfda2 | 7814 | */ |
1da177e4 | 7815 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
7816 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
7817 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 7818 | |
41a2d6cf | 7819 | static int |
96f874e2 RR |
7820 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
7821 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 7822 | { |
6711cab4 | 7823 | if (sg) |
6c99e9ad | 7824 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
7825 | return cpu; |
7826 | } | |
6d6bc0ad | 7827 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 7828 | |
48f24c4d IM |
7829 | /* |
7830 | * multi-core sched-domains: | |
7831 | */ | |
1e9f28fa | 7832 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
7833 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
7834 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 7835 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
7836 | |
7837 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 7838 | static int |
96f874e2 RR |
7839 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7840 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 7841 | { |
6711cab4 | 7842 | int group; |
7c16ec58 | 7843 | |
c69fc56d | 7844 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 7845 | group = cpumask_first(mask); |
6711cab4 | 7846 | if (sg) |
6c99e9ad | 7847 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 7848 | return group; |
1e9f28fa SS |
7849 | } |
7850 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 7851 | static int |
96f874e2 RR |
7852 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
7853 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 7854 | { |
6711cab4 | 7855 | if (sg) |
6c99e9ad | 7856 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
7857 | return cpu; |
7858 | } | |
7859 | #endif | |
7860 | ||
6c99e9ad RR |
7861 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
7862 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 7863 | |
41a2d6cf | 7864 | static int |
96f874e2 RR |
7865 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
7866 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 7867 | { |
6711cab4 | 7868 | int group; |
48f24c4d | 7869 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 7870 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 7871 | group = cpumask_first(mask); |
1e9f28fa | 7872 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 7873 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 7874 | group = cpumask_first(mask); |
1da177e4 | 7875 | #else |
6711cab4 | 7876 | group = cpu; |
1da177e4 | 7877 | #endif |
6711cab4 | 7878 | if (sg) |
6c99e9ad | 7879 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 7880 | return group; |
1da177e4 LT |
7881 | } |
7882 | ||
7883 | #ifdef CONFIG_NUMA | |
1da177e4 | 7884 | /* |
9c1cfda2 JH |
7885 | * The init_sched_build_groups can't handle what we want to do with node |
7886 | * groups, so roll our own. Now each node has its own list of groups which | |
7887 | * gets dynamically allocated. | |
1da177e4 | 7888 | */ |
62ea9ceb | 7889 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 7890 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 7891 | |
62ea9ceb | 7892 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 7893 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 7894 | |
96f874e2 RR |
7895 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
7896 | struct sched_group **sg, | |
7897 | struct cpumask *nodemask) | |
9c1cfda2 | 7898 | { |
6711cab4 SS |
7899 | int group; |
7900 | ||
6ca09dfc | 7901 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 7902 | group = cpumask_first(nodemask); |
6711cab4 SS |
7903 | |
7904 | if (sg) | |
6c99e9ad | 7905 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7906 | return group; |
1da177e4 | 7907 | } |
6711cab4 | 7908 | |
08069033 SS |
7909 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7910 | { | |
7911 | struct sched_group *sg = group_head; | |
7912 | int j; | |
7913 | ||
7914 | if (!sg) | |
7915 | return; | |
3a5c359a | 7916 | do { |
758b2cdc | 7917 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7918 | struct sched_domain *sd; |
08069033 | 7919 | |
6c99e9ad | 7920 | sd = &per_cpu(phys_domains, j).sd; |
758b2cdc | 7921 | if (j != cpumask_first(sched_group_cpus(sd->groups))) { |
3a5c359a AK |
7922 | /* |
7923 | * Only add "power" once for each | |
7924 | * physical package. | |
7925 | */ | |
7926 | continue; | |
7927 | } | |
08069033 | 7928 | |
3a5c359a AK |
7929 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
7930 | } | |
7931 | sg = sg->next; | |
7932 | } while (sg != group_head); | |
08069033 | 7933 | } |
6d6bc0ad | 7934 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7935 | |
a616058b | 7936 | #ifdef CONFIG_NUMA |
51888ca2 | 7937 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7938 | static void free_sched_groups(const struct cpumask *cpu_map, |
7939 | struct cpumask *nodemask) | |
51888ca2 | 7940 | { |
a616058b | 7941 | int cpu, i; |
51888ca2 | 7942 | |
abcd083a | 7943 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7944 | struct sched_group **sched_group_nodes |
7945 | = sched_group_nodes_bycpu[cpu]; | |
7946 | ||
51888ca2 SV |
7947 | if (!sched_group_nodes) |
7948 | continue; | |
7949 | ||
076ac2af | 7950 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7951 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7952 | ||
6ca09dfc | 7953 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7954 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7955 | continue; |
7956 | ||
7957 | if (sg == NULL) | |
7958 | continue; | |
7959 | sg = sg->next; | |
7960 | next_sg: | |
7961 | oldsg = sg; | |
7962 | sg = sg->next; | |
7963 | kfree(oldsg); | |
7964 | if (oldsg != sched_group_nodes[i]) | |
7965 | goto next_sg; | |
7966 | } | |
7967 | kfree(sched_group_nodes); | |
7968 | sched_group_nodes_bycpu[cpu] = NULL; | |
7969 | } | |
51888ca2 | 7970 | } |
6d6bc0ad | 7971 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7972 | static void free_sched_groups(const struct cpumask *cpu_map, |
7973 | struct cpumask *nodemask) | |
a616058b SS |
7974 | { |
7975 | } | |
6d6bc0ad | 7976 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7977 | |
89c4710e SS |
7978 | /* |
7979 | * Initialize sched groups cpu_power. | |
7980 | * | |
7981 | * cpu_power indicates the capacity of sched group, which is used while | |
7982 | * distributing the load between different sched groups in a sched domain. | |
7983 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7984 | * there are asymmetries in the topology. If there are asymmetries, group | |
7985 | * having more cpu_power will pickup more load compared to the group having | |
7986 | * less cpu_power. | |
7987 | * | |
7988 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
7989 | * the maximum number of tasks a group can handle in the presence of other idle | |
7990 | * or lightly loaded groups in the same sched domain. | |
7991 | */ | |
7992 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7993 | { | |
7994 | struct sched_domain *child; | |
7995 | struct sched_group *group; | |
7996 | ||
7997 | WARN_ON(!sd || !sd->groups); | |
7998 | ||
758b2cdc | 7999 | if (cpu != cpumask_first(sched_group_cpus(sd->groups))) |
89c4710e SS |
8000 | return; |
8001 | ||
8002 | child = sd->child; | |
8003 | ||
5517d86b ED |
8004 | sd->groups->__cpu_power = 0; |
8005 | ||
89c4710e SS |
8006 | /* |
8007 | * For perf policy, if the groups in child domain share resources | |
8008 | * (for example cores sharing some portions of the cache hierarchy | |
8009 | * or SMT), then set this domain groups cpu_power such that each group | |
8010 | * can handle only one task, when there are other idle groups in the | |
8011 | * same sched domain. | |
8012 | */ | |
8013 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
8014 | (child->flags & | |
8015 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 8016 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
8017 | return; |
8018 | } | |
8019 | ||
89c4710e SS |
8020 | /* |
8021 | * add cpu_power of each child group to this groups cpu_power | |
8022 | */ | |
8023 | group = child->groups; | |
8024 | do { | |
5517d86b | 8025 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8026 | group = group->next; |
8027 | } while (group != child->groups); | |
8028 | } | |
8029 | ||
7c16ec58 MT |
8030 | /* |
8031 | * Initializers for schedule domains | |
8032 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8033 | */ | |
8034 | ||
a5d8c348 IM |
8035 | #ifdef CONFIG_SCHED_DEBUG |
8036 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8037 | #else | |
8038 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8039 | #endif | |
8040 | ||
7c16ec58 | 8041 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8042 | |
7c16ec58 MT |
8043 | #define SD_INIT_FUNC(type) \ |
8044 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8045 | { \ | |
8046 | memset(sd, 0, sizeof(*sd)); \ | |
8047 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8048 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8049 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8050 | } |
8051 | ||
8052 | SD_INIT_FUNC(CPU) | |
8053 | #ifdef CONFIG_NUMA | |
8054 | SD_INIT_FUNC(ALLNODES) | |
8055 | SD_INIT_FUNC(NODE) | |
8056 | #endif | |
8057 | #ifdef CONFIG_SCHED_SMT | |
8058 | SD_INIT_FUNC(SIBLING) | |
8059 | #endif | |
8060 | #ifdef CONFIG_SCHED_MC | |
8061 | SD_INIT_FUNC(MC) | |
8062 | #endif | |
8063 | ||
1d3504fc HS |
8064 | static int default_relax_domain_level = -1; |
8065 | ||
8066 | static int __init setup_relax_domain_level(char *str) | |
8067 | { | |
30e0e178 LZ |
8068 | unsigned long val; |
8069 | ||
8070 | val = simple_strtoul(str, NULL, 0); | |
8071 | if (val < SD_LV_MAX) | |
8072 | default_relax_domain_level = val; | |
8073 | ||
1d3504fc HS |
8074 | return 1; |
8075 | } | |
8076 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8077 | ||
8078 | static void set_domain_attribute(struct sched_domain *sd, | |
8079 | struct sched_domain_attr *attr) | |
8080 | { | |
8081 | int request; | |
8082 | ||
8083 | if (!attr || attr->relax_domain_level < 0) { | |
8084 | if (default_relax_domain_level < 0) | |
8085 | return; | |
8086 | else | |
8087 | request = default_relax_domain_level; | |
8088 | } else | |
8089 | request = attr->relax_domain_level; | |
8090 | if (request < sd->level) { | |
8091 | /* turn off idle balance on this domain */ | |
8092 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8093 | } else { | |
8094 | /* turn on idle balance on this domain */ | |
8095 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8096 | } | |
8097 | } | |
8098 | ||
1da177e4 | 8099 | /* |
1a20ff27 DG |
8100 | * Build sched domains for a given set of cpus and attach the sched domains |
8101 | * to the individual cpus | |
1da177e4 | 8102 | */ |
96f874e2 | 8103 | static int __build_sched_domains(const struct cpumask *cpu_map, |
1d3504fc | 8104 | struct sched_domain_attr *attr) |
1da177e4 | 8105 | { |
3404c8d9 | 8106 | int i, err = -ENOMEM; |
57d885fe | 8107 | struct root_domain *rd; |
3404c8d9 RR |
8108 | cpumask_var_t nodemask, this_sibling_map, this_core_map, send_covered, |
8109 | tmpmask; | |
d1b55138 | 8110 | #ifdef CONFIG_NUMA |
3404c8d9 | 8111 | cpumask_var_t domainspan, covered, notcovered; |
d1b55138 | 8112 | struct sched_group **sched_group_nodes = NULL; |
6711cab4 | 8113 | int sd_allnodes = 0; |
d1b55138 | 8114 | |
3404c8d9 RR |
8115 | if (!alloc_cpumask_var(&domainspan, GFP_KERNEL)) |
8116 | goto out; | |
8117 | if (!alloc_cpumask_var(&covered, GFP_KERNEL)) | |
8118 | goto free_domainspan; | |
8119 | if (!alloc_cpumask_var(¬covered, GFP_KERNEL)) | |
8120 | goto free_covered; | |
8121 | #endif | |
8122 | ||
8123 | if (!alloc_cpumask_var(&nodemask, GFP_KERNEL)) | |
8124 | goto free_notcovered; | |
8125 | if (!alloc_cpumask_var(&this_sibling_map, GFP_KERNEL)) | |
8126 | goto free_nodemask; | |
8127 | if (!alloc_cpumask_var(&this_core_map, GFP_KERNEL)) | |
8128 | goto free_this_sibling_map; | |
8129 | if (!alloc_cpumask_var(&send_covered, GFP_KERNEL)) | |
8130 | goto free_this_core_map; | |
8131 | if (!alloc_cpumask_var(&tmpmask, GFP_KERNEL)) | |
8132 | goto free_send_covered; | |
8133 | ||
8134 | #ifdef CONFIG_NUMA | |
d1b55138 JH |
8135 | /* |
8136 | * Allocate the per-node list of sched groups | |
8137 | */ | |
076ac2af | 8138 | sched_group_nodes = kcalloc(nr_node_ids, sizeof(struct sched_group *), |
41a2d6cf | 8139 | GFP_KERNEL); |
d1b55138 JH |
8140 | if (!sched_group_nodes) { |
8141 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
3404c8d9 | 8142 | goto free_tmpmask; |
d1b55138 | 8143 | } |
d1b55138 | 8144 | #endif |
1da177e4 | 8145 | |
dc938520 | 8146 | rd = alloc_rootdomain(); |
57d885fe GH |
8147 | if (!rd) { |
8148 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
3404c8d9 | 8149 | goto free_sched_groups; |
7c16ec58 | 8150 | } |
6d21cd62 | 8151 | |
7c16ec58 | 8152 | #ifdef CONFIG_NUMA |
96f874e2 | 8153 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = sched_group_nodes; |
7c16ec58 MT |
8154 | #endif |
8155 | ||
1da177e4 | 8156 | /* |
1a20ff27 | 8157 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8158 | */ |
abcd083a | 8159 | for_each_cpu(i, cpu_map) { |
1da177e4 | 8160 | struct sched_domain *sd = NULL, *p; |
1da177e4 | 8161 | |
6ca09dfc | 8162 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(i)), cpu_map); |
1da177e4 LT |
8163 | |
8164 | #ifdef CONFIG_NUMA | |
96f874e2 RR |
8165 | if (cpumask_weight(cpu_map) > |
8166 | SD_NODES_PER_DOMAIN*cpumask_weight(nodemask)) { | |
62ea9ceb | 8167 | sd = &per_cpu(allnodes_domains, i).sd; |
7c16ec58 | 8168 | SD_INIT(sd, ALLNODES); |
1d3504fc | 8169 | set_domain_attribute(sd, attr); |
758b2cdc | 8170 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7c16ec58 | 8171 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 8172 | p = sd; |
6711cab4 | 8173 | sd_allnodes = 1; |
9c1cfda2 JH |
8174 | } else |
8175 | p = NULL; | |
8176 | ||
62ea9ceb | 8177 | sd = &per_cpu(node_domains, i).sd; |
7c16ec58 | 8178 | SD_INIT(sd, NODE); |
1d3504fc | 8179 | set_domain_attribute(sd, attr); |
758b2cdc | 8180 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); |
9c1cfda2 | 8181 | sd->parent = p; |
1a848870 SS |
8182 | if (p) |
8183 | p->child = sd; | |
758b2cdc RR |
8184 | cpumask_and(sched_domain_span(sd), |
8185 | sched_domain_span(sd), cpu_map); | |
1da177e4 LT |
8186 | #endif |
8187 | ||
8188 | p = sd; | |
6c99e9ad | 8189 | sd = &per_cpu(phys_domains, i).sd; |
7c16ec58 | 8190 | SD_INIT(sd, CPU); |
1d3504fc | 8191 | set_domain_attribute(sd, attr); |
758b2cdc | 8192 | cpumask_copy(sched_domain_span(sd), nodemask); |
1da177e4 | 8193 | sd->parent = p; |
1a848870 SS |
8194 | if (p) |
8195 | p->child = sd; | |
7c16ec58 | 8196 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 8197 | |
1e9f28fa SS |
8198 | #ifdef CONFIG_SCHED_MC |
8199 | p = sd; | |
6c99e9ad | 8200 | sd = &per_cpu(core_domains, i).sd; |
7c16ec58 | 8201 | SD_INIT(sd, MC); |
1d3504fc | 8202 | set_domain_attribute(sd, attr); |
6ca09dfc MT |
8203 | cpumask_and(sched_domain_span(sd), cpu_map, |
8204 | cpu_coregroup_mask(i)); | |
1e9f28fa | 8205 | sd->parent = p; |
1a848870 | 8206 | p->child = sd; |
7c16ec58 | 8207 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
8208 | #endif |
8209 | ||
1da177e4 LT |
8210 | #ifdef CONFIG_SCHED_SMT |
8211 | p = sd; | |
6c99e9ad | 8212 | sd = &per_cpu(cpu_domains, i).sd; |
7c16ec58 | 8213 | SD_INIT(sd, SIBLING); |
1d3504fc | 8214 | set_domain_attribute(sd, attr); |
758b2cdc | 8215 | cpumask_and(sched_domain_span(sd), |
c69fc56d | 8216 | topology_thread_cpumask(i), cpu_map); |
1da177e4 | 8217 | sd->parent = p; |
1a848870 | 8218 | p->child = sd; |
7c16ec58 | 8219 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
8220 | #endif |
8221 | } | |
8222 | ||
8223 | #ifdef CONFIG_SCHED_SMT | |
8224 | /* Set up CPU (sibling) groups */ | |
abcd083a | 8225 | for_each_cpu(i, cpu_map) { |
96f874e2 | 8226 | cpumask_and(this_sibling_map, |
c69fc56d | 8227 | topology_thread_cpumask(i), cpu_map); |
96f874e2 | 8228 | if (i != cpumask_first(this_sibling_map)) |
1da177e4 LT |
8229 | continue; |
8230 | ||
dd41f596 | 8231 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
8232 | &cpu_to_cpu_group, |
8233 | send_covered, tmpmask); | |
1da177e4 LT |
8234 | } |
8235 | #endif | |
8236 | ||
1e9f28fa SS |
8237 | #ifdef CONFIG_SCHED_MC |
8238 | /* Set up multi-core groups */ | |
abcd083a | 8239 | for_each_cpu(i, cpu_map) { |
6ca09dfc | 8240 | cpumask_and(this_core_map, cpu_coregroup_mask(i), cpu_map); |
96f874e2 | 8241 | if (i != cpumask_first(this_core_map)) |
1e9f28fa | 8242 | continue; |
7c16ec58 | 8243 | |
dd41f596 | 8244 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
8245 | &cpu_to_core_group, |
8246 | send_covered, tmpmask); | |
1e9f28fa SS |
8247 | } |
8248 | #endif | |
8249 | ||
1da177e4 | 8250 | /* Set up physical groups */ |
076ac2af | 8251 | for (i = 0; i < nr_node_ids; i++) { |
6ca09dfc | 8252 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8253 | if (cpumask_empty(nodemask)) |
1da177e4 LT |
8254 | continue; |
8255 | ||
7c16ec58 MT |
8256 | init_sched_build_groups(nodemask, cpu_map, |
8257 | &cpu_to_phys_group, | |
8258 | send_covered, tmpmask); | |
1da177e4 LT |
8259 | } |
8260 | ||
8261 | #ifdef CONFIG_NUMA | |
8262 | /* Set up node groups */ | |
7c16ec58 | 8263 | if (sd_allnodes) { |
7c16ec58 MT |
8264 | init_sched_build_groups(cpu_map, cpu_map, |
8265 | &cpu_to_allnodes_group, | |
8266 | send_covered, tmpmask); | |
8267 | } | |
9c1cfda2 | 8268 | |
076ac2af | 8269 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 JH |
8270 | /* Set up node groups */ |
8271 | struct sched_group *sg, *prev; | |
9c1cfda2 JH |
8272 | int j; |
8273 | ||
96f874e2 | 8274 | cpumask_clear(covered); |
6ca09dfc | 8275 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8276 | if (cpumask_empty(nodemask)) { |
d1b55138 | 8277 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 8278 | continue; |
d1b55138 | 8279 | } |
9c1cfda2 | 8280 | |
4bdbaad3 | 8281 | sched_domain_node_span(i, domainspan); |
96f874e2 | 8282 | cpumask_and(domainspan, domainspan, cpu_map); |
9c1cfda2 | 8283 | |
6c99e9ad RR |
8284 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), |
8285 | GFP_KERNEL, i); | |
51888ca2 SV |
8286 | if (!sg) { |
8287 | printk(KERN_WARNING "Can not alloc domain group for " | |
8288 | "node %d\n", i); | |
8289 | goto error; | |
8290 | } | |
9c1cfda2 | 8291 | sched_group_nodes[i] = sg; |
abcd083a | 8292 | for_each_cpu(j, nodemask) { |
9c1cfda2 | 8293 | struct sched_domain *sd; |
9761eea8 | 8294 | |
62ea9ceb | 8295 | sd = &per_cpu(node_domains, j).sd; |
9c1cfda2 | 8296 | sd->groups = sg; |
9c1cfda2 | 8297 | } |
5517d86b | 8298 | sg->__cpu_power = 0; |
758b2cdc | 8299 | cpumask_copy(sched_group_cpus(sg), nodemask); |
51888ca2 | 8300 | sg->next = sg; |
96f874e2 | 8301 | cpumask_or(covered, covered, nodemask); |
9c1cfda2 JH |
8302 | prev = sg; |
8303 | ||
076ac2af | 8304 | for (j = 0; j < nr_node_ids; j++) { |
076ac2af | 8305 | int n = (i + j) % nr_node_ids; |
9c1cfda2 | 8306 | |
96f874e2 RR |
8307 | cpumask_complement(notcovered, covered); |
8308 | cpumask_and(tmpmask, notcovered, cpu_map); | |
8309 | cpumask_and(tmpmask, tmpmask, domainspan); | |
8310 | if (cpumask_empty(tmpmask)) | |
9c1cfda2 JH |
8311 | break; |
8312 | ||
6ca09dfc | 8313 | cpumask_and(tmpmask, tmpmask, cpumask_of_node(n)); |
96f874e2 | 8314 | if (cpumask_empty(tmpmask)) |
9c1cfda2 JH |
8315 | continue; |
8316 | ||
6c99e9ad RR |
8317 | sg = kmalloc_node(sizeof(struct sched_group) + |
8318 | cpumask_size(), | |
15f0b676 | 8319 | GFP_KERNEL, i); |
9c1cfda2 JH |
8320 | if (!sg) { |
8321 | printk(KERN_WARNING | |
8322 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 8323 | goto error; |
9c1cfda2 | 8324 | } |
5517d86b | 8325 | sg->__cpu_power = 0; |
758b2cdc | 8326 | cpumask_copy(sched_group_cpus(sg), tmpmask); |
51888ca2 | 8327 | sg->next = prev->next; |
96f874e2 | 8328 | cpumask_or(covered, covered, tmpmask); |
9c1cfda2 JH |
8329 | prev->next = sg; |
8330 | prev = sg; | |
8331 | } | |
9c1cfda2 | 8332 | } |
1da177e4 LT |
8333 | #endif |
8334 | ||
8335 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8336 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8337 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8338 | struct sched_domain *sd = &per_cpu(cpu_domains, i).sd; |
dd41f596 | 8339 | |
89c4710e | 8340 | init_sched_groups_power(i, sd); |
5c45bf27 | 8341 | } |
1da177e4 | 8342 | #endif |
1e9f28fa | 8343 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8344 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8345 | struct sched_domain *sd = &per_cpu(core_domains, i).sd; |
dd41f596 | 8346 | |
89c4710e | 8347 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8348 | } |
8349 | #endif | |
1e9f28fa | 8350 | |
abcd083a | 8351 | for_each_cpu(i, cpu_map) { |
6c99e9ad | 8352 | struct sched_domain *sd = &per_cpu(phys_domains, i).sd; |
dd41f596 | 8353 | |
89c4710e | 8354 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8355 | } |
8356 | ||
9c1cfda2 | 8357 | #ifdef CONFIG_NUMA |
076ac2af | 8358 | for (i = 0; i < nr_node_ids; i++) |
08069033 | 8359 | init_numa_sched_groups_power(sched_group_nodes[i]); |
9c1cfda2 | 8360 | |
6711cab4 SS |
8361 | if (sd_allnodes) { |
8362 | struct sched_group *sg; | |
f712c0c7 | 8363 | |
96f874e2 | 8364 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
7c16ec58 | 8365 | tmpmask); |
f712c0c7 SS |
8366 | init_numa_sched_groups_power(sg); |
8367 | } | |
9c1cfda2 JH |
8368 | #endif |
8369 | ||
1da177e4 | 8370 | /* Attach the domains */ |
abcd083a | 8371 | for_each_cpu(i, cpu_map) { |
1da177e4 LT |
8372 | struct sched_domain *sd; |
8373 | #ifdef CONFIG_SCHED_SMT | |
6c99e9ad | 8374 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 8375 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 8376 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 8377 | #else |
6c99e9ad | 8378 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 8379 | #endif |
57d885fe | 8380 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 8381 | } |
51888ca2 | 8382 | |
3404c8d9 RR |
8383 | err = 0; |
8384 | ||
8385 | free_tmpmask: | |
8386 | free_cpumask_var(tmpmask); | |
8387 | free_send_covered: | |
8388 | free_cpumask_var(send_covered); | |
8389 | free_this_core_map: | |
8390 | free_cpumask_var(this_core_map); | |
8391 | free_this_sibling_map: | |
8392 | free_cpumask_var(this_sibling_map); | |
8393 | free_nodemask: | |
8394 | free_cpumask_var(nodemask); | |
8395 | free_notcovered: | |
8396 | #ifdef CONFIG_NUMA | |
8397 | free_cpumask_var(notcovered); | |
8398 | free_covered: | |
8399 | free_cpumask_var(covered); | |
8400 | free_domainspan: | |
8401 | free_cpumask_var(domainspan); | |
8402 | out: | |
8403 | #endif | |
8404 | return err; | |
8405 | ||
8406 | free_sched_groups: | |
8407 | #ifdef CONFIG_NUMA | |
8408 | kfree(sched_group_nodes); | |
8409 | #endif | |
8410 | goto free_tmpmask; | |
51888ca2 | 8411 | |
a616058b | 8412 | #ifdef CONFIG_NUMA |
51888ca2 | 8413 | error: |
7c16ec58 | 8414 | free_sched_groups(cpu_map, tmpmask); |
c6c4927b | 8415 | free_rootdomain(rd); |
3404c8d9 | 8416 | goto free_tmpmask; |
a616058b | 8417 | #endif |
1da177e4 | 8418 | } |
029190c5 | 8419 | |
96f874e2 | 8420 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
8421 | { |
8422 | return __build_sched_domains(cpu_map, NULL); | |
8423 | } | |
8424 | ||
96f874e2 | 8425 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 8426 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
8427 | static struct sched_domain_attr *dattr_cur; |
8428 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
8429 | |
8430 | /* | |
8431 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
8432 | * cpumask) fails, then fallback to a single sched domain, |
8433 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 8434 | */ |
4212823f | 8435 | static cpumask_var_t fallback_doms; |
029190c5 | 8436 | |
ee79d1bd HC |
8437 | /* |
8438 | * arch_update_cpu_topology lets virtualized architectures update the | |
8439 | * cpu core maps. It is supposed to return 1 if the topology changed | |
8440 | * or 0 if it stayed the same. | |
8441 | */ | |
8442 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 8443 | { |
ee79d1bd | 8444 | return 0; |
22e52b07 HC |
8445 | } |
8446 | ||
1a20ff27 | 8447 | /* |
41a2d6cf | 8448 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
8449 | * For now this just excludes isolated cpus, but could be used to |
8450 | * exclude other special cases in the future. | |
1a20ff27 | 8451 | */ |
96f874e2 | 8452 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8453 | { |
7378547f MM |
8454 | int err; |
8455 | ||
22e52b07 | 8456 | arch_update_cpu_topology(); |
029190c5 | 8457 | ndoms_cur = 1; |
96f874e2 | 8458 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 8459 | if (!doms_cur) |
4212823f | 8460 | doms_cur = fallback_doms; |
dcc30a35 | 8461 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 8462 | dattr_cur = NULL; |
7378547f | 8463 | err = build_sched_domains(doms_cur); |
6382bc90 | 8464 | register_sched_domain_sysctl(); |
7378547f MM |
8465 | |
8466 | return err; | |
1a20ff27 DG |
8467 | } |
8468 | ||
96f874e2 RR |
8469 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
8470 | struct cpumask *tmpmask) | |
1da177e4 | 8471 | { |
7c16ec58 | 8472 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 8473 | } |
1da177e4 | 8474 | |
1a20ff27 DG |
8475 | /* |
8476 | * Detach sched domains from a group of cpus specified in cpu_map | |
8477 | * These cpus will now be attached to the NULL domain | |
8478 | */ | |
96f874e2 | 8479 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 8480 | { |
96f874e2 RR |
8481 | /* Save because hotplug lock held. */ |
8482 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
8483 | int i; |
8484 | ||
abcd083a | 8485 | for_each_cpu(i, cpu_map) |
57d885fe | 8486 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 8487 | synchronize_sched(); |
96f874e2 | 8488 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
8489 | } |
8490 | ||
1d3504fc HS |
8491 | /* handle null as "default" */ |
8492 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
8493 | struct sched_domain_attr *new, int idx_new) | |
8494 | { | |
8495 | struct sched_domain_attr tmp; | |
8496 | ||
8497 | /* fast path */ | |
8498 | if (!new && !cur) | |
8499 | return 1; | |
8500 | ||
8501 | tmp = SD_ATTR_INIT; | |
8502 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
8503 | new ? (new + idx_new) : &tmp, | |
8504 | sizeof(struct sched_domain_attr)); | |
8505 | } | |
8506 | ||
029190c5 PJ |
8507 | /* |
8508 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 8509 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
8510 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
8511 | * It destroys each deleted domain and builds each new domain. | |
8512 | * | |
96f874e2 | 8513 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
8514 | * The masks don't intersect (don't overlap.) We should setup one |
8515 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
8516 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
8517 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
8518 | * it as it is. | |
8519 | * | |
41a2d6cf IM |
8520 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
8521 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
8522 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
8523 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
8524 | * the single partition 'fallback_doms', it also forces the domains | |
8525 | * to be rebuilt. | |
029190c5 | 8526 | * |
96f874e2 | 8527 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
8528 | * ndoms_new == 0 is a special case for destroying existing domains, |
8529 | * and it will not create the default domain. | |
dfb512ec | 8530 | * |
029190c5 PJ |
8531 | * Call with hotplug lock held |
8532 | */ | |
96f874e2 RR |
8533 | /* FIXME: Change to struct cpumask *doms_new[] */ |
8534 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 8535 | struct sched_domain_attr *dattr_new) |
029190c5 | 8536 | { |
dfb512ec | 8537 | int i, j, n; |
d65bd5ec | 8538 | int new_topology; |
029190c5 | 8539 | |
712555ee | 8540 | mutex_lock(&sched_domains_mutex); |
a1835615 | 8541 | |
7378547f MM |
8542 | /* always unregister in case we don't destroy any domains */ |
8543 | unregister_sched_domain_sysctl(); | |
8544 | ||
d65bd5ec HC |
8545 | /* Let architecture update cpu core mappings. */ |
8546 | new_topology = arch_update_cpu_topology(); | |
8547 | ||
dfb512ec | 8548 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
8549 | |
8550 | /* Destroy deleted domains */ | |
8551 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 8552 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 8553 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 8554 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
8555 | goto match1; |
8556 | } | |
8557 | /* no match - a current sched domain not in new doms_new[] */ | |
8558 | detach_destroy_domains(doms_cur + i); | |
8559 | match1: | |
8560 | ; | |
8561 | } | |
8562 | ||
e761b772 MK |
8563 | if (doms_new == NULL) { |
8564 | ndoms_cur = 0; | |
4212823f | 8565 | doms_new = fallback_doms; |
dcc30a35 | 8566 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 8567 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
8568 | } |
8569 | ||
029190c5 PJ |
8570 | /* Build new domains */ |
8571 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 8572 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 8573 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 8574 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
8575 | goto match2; |
8576 | } | |
8577 | /* no match - add a new doms_new */ | |
1d3504fc HS |
8578 | __build_sched_domains(doms_new + i, |
8579 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
8580 | match2: |
8581 | ; | |
8582 | } | |
8583 | ||
8584 | /* Remember the new sched domains */ | |
4212823f | 8585 | if (doms_cur != fallback_doms) |
029190c5 | 8586 | kfree(doms_cur); |
1d3504fc | 8587 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 8588 | doms_cur = doms_new; |
1d3504fc | 8589 | dattr_cur = dattr_new; |
029190c5 | 8590 | ndoms_cur = ndoms_new; |
7378547f MM |
8591 | |
8592 | register_sched_domain_sysctl(); | |
a1835615 | 8593 | |
712555ee | 8594 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
8595 | } |
8596 | ||
5c45bf27 | 8597 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 8598 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 8599 | { |
95402b38 | 8600 | get_online_cpus(); |
dfb512ec MK |
8601 | |
8602 | /* Destroy domains first to force the rebuild */ | |
8603 | partition_sched_domains(0, NULL, NULL); | |
8604 | ||
e761b772 | 8605 | rebuild_sched_domains(); |
95402b38 | 8606 | put_online_cpus(); |
5c45bf27 SS |
8607 | } |
8608 | ||
8609 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
8610 | { | |
afb8a9b7 | 8611 | unsigned int level = 0; |
5c45bf27 | 8612 | |
afb8a9b7 GS |
8613 | if (sscanf(buf, "%u", &level) != 1) |
8614 | return -EINVAL; | |
8615 | ||
8616 | /* | |
8617 | * level is always be positive so don't check for | |
8618 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
8619 | * What happens on 0 or 1 byte write, | |
8620 | * need to check for count as well? | |
8621 | */ | |
5c45bf27 | 8622 | |
afb8a9b7 | 8623 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) |
5c45bf27 SS |
8624 | return -EINVAL; |
8625 | ||
8626 | if (smt) | |
afb8a9b7 | 8627 | sched_smt_power_savings = level; |
5c45bf27 | 8628 | else |
afb8a9b7 | 8629 | sched_mc_power_savings = level; |
5c45bf27 | 8630 | |
c70f22d2 | 8631 | arch_reinit_sched_domains(); |
5c45bf27 | 8632 | |
c70f22d2 | 8633 | return count; |
5c45bf27 SS |
8634 | } |
8635 | ||
5c45bf27 | 8636 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
8637 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
8638 | char *page) | |
5c45bf27 SS |
8639 | { |
8640 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
8641 | } | |
f718cd4a | 8642 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 8643 | const char *buf, size_t count) |
5c45bf27 SS |
8644 | { |
8645 | return sched_power_savings_store(buf, count, 0); | |
8646 | } | |
f718cd4a AK |
8647 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
8648 | sched_mc_power_savings_show, | |
8649 | sched_mc_power_savings_store); | |
5c45bf27 SS |
8650 | #endif |
8651 | ||
8652 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
8653 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
8654 | char *page) | |
5c45bf27 SS |
8655 | { |
8656 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
8657 | } | |
f718cd4a | 8658 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 8659 | const char *buf, size_t count) |
5c45bf27 SS |
8660 | { |
8661 | return sched_power_savings_store(buf, count, 1); | |
8662 | } | |
f718cd4a AK |
8663 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
8664 | sched_smt_power_savings_show, | |
6707de00 AB |
8665 | sched_smt_power_savings_store); |
8666 | #endif | |
8667 | ||
39aac648 | 8668 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
8669 | { |
8670 | int err = 0; | |
8671 | ||
8672 | #ifdef CONFIG_SCHED_SMT | |
8673 | if (smt_capable()) | |
8674 | err = sysfs_create_file(&cls->kset.kobj, | |
8675 | &attr_sched_smt_power_savings.attr); | |
8676 | #endif | |
8677 | #ifdef CONFIG_SCHED_MC | |
8678 | if (!err && mc_capable()) | |
8679 | err = sysfs_create_file(&cls->kset.kobj, | |
8680 | &attr_sched_mc_power_savings.attr); | |
8681 | #endif | |
8682 | return err; | |
8683 | } | |
6d6bc0ad | 8684 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 8685 | |
e761b772 | 8686 | #ifndef CONFIG_CPUSETS |
1da177e4 | 8687 | /* |
e761b772 MK |
8688 | * Add online and remove offline CPUs from the scheduler domains. |
8689 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
8690 | */ |
8691 | static int update_sched_domains(struct notifier_block *nfb, | |
8692 | unsigned long action, void *hcpu) | |
e761b772 MK |
8693 | { |
8694 | switch (action) { | |
8695 | case CPU_ONLINE: | |
8696 | case CPU_ONLINE_FROZEN: | |
8697 | case CPU_DEAD: | |
8698 | case CPU_DEAD_FROZEN: | |
dfb512ec | 8699 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
8700 | return NOTIFY_OK; |
8701 | ||
8702 | default: | |
8703 | return NOTIFY_DONE; | |
8704 | } | |
8705 | } | |
8706 | #endif | |
8707 | ||
8708 | static int update_runtime(struct notifier_block *nfb, | |
8709 | unsigned long action, void *hcpu) | |
1da177e4 | 8710 | { |
7def2be1 PZ |
8711 | int cpu = (int)(long)hcpu; |
8712 | ||
1da177e4 | 8713 | switch (action) { |
1da177e4 | 8714 | case CPU_DOWN_PREPARE: |
8bb78442 | 8715 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 8716 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
8717 | return NOTIFY_OK; |
8718 | ||
1da177e4 | 8719 | case CPU_DOWN_FAILED: |
8bb78442 | 8720 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 8721 | case CPU_ONLINE: |
8bb78442 | 8722 | case CPU_ONLINE_FROZEN: |
7def2be1 | 8723 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
8724 | return NOTIFY_OK; |
8725 | ||
1da177e4 LT |
8726 | default: |
8727 | return NOTIFY_DONE; | |
8728 | } | |
1da177e4 | 8729 | } |
1da177e4 LT |
8730 | |
8731 | void __init sched_init_smp(void) | |
8732 | { | |
dcc30a35 RR |
8733 | cpumask_var_t non_isolated_cpus; |
8734 | ||
8735 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 8736 | |
434d53b0 MT |
8737 | #if defined(CONFIG_NUMA) |
8738 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
8739 | GFP_KERNEL); | |
8740 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
8741 | #endif | |
95402b38 | 8742 | get_online_cpus(); |
712555ee | 8743 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
8744 | arch_init_sched_domains(cpu_online_mask); |
8745 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
8746 | if (cpumask_empty(non_isolated_cpus)) | |
8747 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 8748 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 8749 | put_online_cpus(); |
e761b772 MK |
8750 | |
8751 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
8752 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
8753 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
8754 | #endif |
8755 | ||
8756 | /* RT runtime code needs to handle some hotplug events */ | |
8757 | hotcpu_notifier(update_runtime, 0); | |
8758 | ||
b328ca18 | 8759 | init_hrtick(); |
5c1e1767 NP |
8760 | |
8761 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 8762 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 8763 | BUG(); |
19978ca6 | 8764 | sched_init_granularity(); |
dcc30a35 | 8765 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
8766 | |
8767 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 8768 | init_sched_rt_class(); |
1da177e4 LT |
8769 | } |
8770 | #else | |
8771 | void __init sched_init_smp(void) | |
8772 | { | |
19978ca6 | 8773 | sched_init_granularity(); |
1da177e4 LT |
8774 | } |
8775 | #endif /* CONFIG_SMP */ | |
8776 | ||
8777 | int in_sched_functions(unsigned long addr) | |
8778 | { | |
1da177e4 LT |
8779 | return in_lock_functions(addr) || |
8780 | (addr >= (unsigned long)__sched_text_start | |
8781 | && addr < (unsigned long)__sched_text_end); | |
8782 | } | |
8783 | ||
a9957449 | 8784 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8785 | { |
8786 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8787 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8788 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8789 | cfs_rq->rq = rq; | |
8790 | #endif | |
67e9fb2a | 8791 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8792 | } |
8793 | ||
fa85ae24 PZ |
8794 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8795 | { | |
8796 | struct rt_prio_array *array; | |
8797 | int i; | |
8798 | ||
8799 | array = &rt_rq->active; | |
8800 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8801 | INIT_LIST_HEAD(array->queue + i); | |
8802 | __clear_bit(i, array->bitmap); | |
8803 | } | |
8804 | /* delimiter for bitsearch: */ | |
8805 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8806 | ||
052f1dc7 | 8807 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8808 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8809 | #ifdef CONFIG_SMP |
e864c499 | 8810 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8811 | #endif |
48d5e258 | 8812 | #endif |
fa85ae24 PZ |
8813 | #ifdef CONFIG_SMP |
8814 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8815 | rt_rq->overloaded = 0; |
917b627d | 8816 | plist_head_init(&rq->rt.pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8817 | #endif |
8818 | ||
8819 | rt_rq->rt_time = 0; | |
8820 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
8821 | rt_rq->rt_runtime = 0; |
8822 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 8823 | |
052f1dc7 | 8824 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8825 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8826 | rt_rq->rq = rq; |
8827 | #endif | |
fa85ae24 PZ |
8828 | } |
8829 | ||
6f505b16 | 8830 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8831 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8832 | struct sched_entity *se, int cpu, int add, | |
8833 | struct sched_entity *parent) | |
6f505b16 | 8834 | { |
ec7dc8ac | 8835 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8836 | tg->cfs_rq[cpu] = cfs_rq; |
8837 | init_cfs_rq(cfs_rq, rq); | |
8838 | cfs_rq->tg = tg; | |
8839 | if (add) | |
8840 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8841 | ||
8842 | tg->se[cpu] = se; | |
354d60c2 DG |
8843 | /* se could be NULL for init_task_group */ |
8844 | if (!se) | |
8845 | return; | |
8846 | ||
ec7dc8ac DG |
8847 | if (!parent) |
8848 | se->cfs_rq = &rq->cfs; | |
8849 | else | |
8850 | se->cfs_rq = parent->my_q; | |
8851 | ||
6f505b16 PZ |
8852 | se->my_q = cfs_rq; |
8853 | se->load.weight = tg->shares; | |
e05510d0 | 8854 | se->load.inv_weight = 0; |
ec7dc8ac | 8855 | se->parent = parent; |
6f505b16 | 8856 | } |
052f1dc7 | 8857 | #endif |
6f505b16 | 8858 | |
052f1dc7 | 8859 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8860 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8861 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8862 | struct sched_rt_entity *parent) | |
6f505b16 | 8863 | { |
ec7dc8ac DG |
8864 | struct rq *rq = cpu_rq(cpu); |
8865 | ||
6f505b16 PZ |
8866 | tg->rt_rq[cpu] = rt_rq; |
8867 | init_rt_rq(rt_rq, rq); | |
8868 | rt_rq->tg = tg; | |
8869 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 8870 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8871 | if (add) |
8872 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8873 | ||
8874 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8875 | if (!rt_se) |
8876 | return; | |
8877 | ||
ec7dc8ac DG |
8878 | if (!parent) |
8879 | rt_se->rt_rq = &rq->rt; | |
8880 | else | |
8881 | rt_se->rt_rq = parent->my_q; | |
8882 | ||
6f505b16 | 8883 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8884 | rt_se->parent = parent; |
6f505b16 PZ |
8885 | INIT_LIST_HEAD(&rt_se->run_list); |
8886 | } | |
8887 | #endif | |
8888 | ||
1da177e4 LT |
8889 | void __init sched_init(void) |
8890 | { | |
dd41f596 | 8891 | int i, j; |
434d53b0 MT |
8892 | unsigned long alloc_size = 0, ptr; |
8893 | ||
8894 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8895 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8896 | #endif | |
8897 | #ifdef CONFIG_RT_GROUP_SCHED | |
8898 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8899 | #endif |
8900 | #ifdef CONFIG_USER_SCHED | |
8901 | alloc_size *= 2; | |
df7c8e84 RR |
8902 | #endif |
8903 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 8904 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
8905 | #endif |
8906 | /* | |
8907 | * As sched_init() is called before page_alloc is setup, | |
8908 | * we use alloc_bootmem(). | |
8909 | */ | |
8910 | if (alloc_size) { | |
5a9d3225 | 8911 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
8912 | |
8913 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8914 | init_task_group.se = (struct sched_entity **)ptr; | |
8915 | ptr += nr_cpu_ids * sizeof(void **); | |
8916 | ||
8917 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8918 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
8919 | |
8920 | #ifdef CONFIG_USER_SCHED | |
8921 | root_task_group.se = (struct sched_entity **)ptr; | |
8922 | ptr += nr_cpu_ids * sizeof(void **); | |
8923 | ||
8924 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8925 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8926 | #endif /* CONFIG_USER_SCHED */ |
8927 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
8928 | #ifdef CONFIG_RT_GROUP_SCHED |
8929 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8930 | ptr += nr_cpu_ids * sizeof(void **); | |
8931 | ||
8932 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8933 | ptr += nr_cpu_ids * sizeof(void **); |
8934 | ||
8935 | #ifdef CONFIG_USER_SCHED | |
8936 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8937 | ptr += nr_cpu_ids * sizeof(void **); | |
8938 | ||
8939 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
8940 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
8941 | #endif /* CONFIG_USER_SCHED */ |
8942 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
8943 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8944 | for_each_possible_cpu(i) { | |
8945 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
8946 | ptr += cpumask_size(); | |
8947 | } | |
8948 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 8949 | } |
dd41f596 | 8950 | |
57d885fe GH |
8951 | #ifdef CONFIG_SMP |
8952 | init_defrootdomain(); | |
8953 | #endif | |
8954 | ||
d0b27fa7 PZ |
8955 | init_rt_bandwidth(&def_rt_bandwidth, |
8956 | global_rt_period(), global_rt_runtime()); | |
8957 | ||
8958 | #ifdef CONFIG_RT_GROUP_SCHED | |
8959 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8960 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
8961 | #ifdef CONFIG_USER_SCHED |
8962 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
8963 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
8964 | #endif /* CONFIG_USER_SCHED */ |
8965 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 8966 | |
052f1dc7 | 8967 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 8968 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8969 | INIT_LIST_HEAD(&init_task_group.children); |
8970 | ||
8971 | #ifdef CONFIG_USER_SCHED | |
8972 | INIT_LIST_HEAD(&root_task_group.children); | |
8973 | init_task_group.parent = &root_task_group; | |
8974 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
8975 | #endif /* CONFIG_USER_SCHED */ |
8976 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 8977 | |
0a945022 | 8978 | for_each_possible_cpu(i) { |
70b97a7f | 8979 | struct rq *rq; |
1da177e4 LT |
8980 | |
8981 | rq = cpu_rq(i); | |
8982 | spin_lock_init(&rq->lock); | |
7897986b | 8983 | rq->nr_running = 0; |
dd41f596 | 8984 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8985 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8986 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8987 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8988 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8989 | #ifdef CONFIG_CGROUP_SCHED |
8990 | /* | |
8991 | * How much cpu bandwidth does init_task_group get? | |
8992 | * | |
8993 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8994 | * gets 100% of the cpu resources in the system. This overall | |
8995 | * system cpu resource is divided among the tasks of | |
8996 | * init_task_group and its child task-groups in a fair manner, | |
8997 | * based on each entity's (task or task-group's) weight | |
8998 | * (se->load.weight). | |
8999 | * | |
9000 | * In other words, if init_task_group has 10 tasks of weight | |
9001 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9002 | * then A0's share of the cpu resource is: | |
9003 | * | |
0d905bca | 9004 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9005 | * |
9006 | * We achieve this by letting init_task_group's tasks sit | |
9007 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9008 | */ | |
ec7dc8ac | 9009 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9010 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9011 | root_task_group.shares = NICE_0_LOAD; |
9012 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9013 | /* |
9014 | * In case of task-groups formed thr' the user id of tasks, | |
9015 | * init_task_group represents tasks belonging to root user. | |
9016 | * Hence it forms a sibling of all subsequent groups formed. | |
9017 | * In this case, init_task_group gets only a fraction of overall | |
9018 | * system cpu resource, based on the weight assigned to root | |
9019 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9020 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
9021 | * (init_cfs_rq) and having one entity represent this group of | |
9022 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
9023 | */ | |
ec7dc8ac | 9024 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 9025 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
9026 | &per_cpu(init_sched_entity, i), i, 1, |
9027 | root_task_group.se[i]); | |
6f505b16 | 9028 | |
052f1dc7 | 9029 | #endif |
354d60c2 DG |
9030 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9031 | ||
9032 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9033 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9034 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9035 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9036 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9037 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9038 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9039 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9040 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9041 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9042 | root_task_group.rt_se[i]); | |
354d60c2 | 9043 | #endif |
dd41f596 | 9044 | #endif |
1da177e4 | 9045 | |
dd41f596 IM |
9046 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9047 | rq->cpu_load[j] = 0; | |
1da177e4 | 9048 | #ifdef CONFIG_SMP |
41c7ce9a | 9049 | rq->sd = NULL; |
57d885fe | 9050 | rq->rd = NULL; |
1da177e4 | 9051 | rq->active_balance = 0; |
dd41f596 | 9052 | rq->next_balance = jiffies; |
1da177e4 | 9053 | rq->push_cpu = 0; |
0a2966b4 | 9054 | rq->cpu = i; |
1f11eb6a | 9055 | rq->online = 0; |
1da177e4 LT |
9056 | rq->migration_thread = NULL; |
9057 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9058 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9059 | #endif |
8f4d37ec | 9060 | init_rq_hrtick(rq); |
1da177e4 | 9061 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9062 | } |
9063 | ||
2dd73a4f | 9064 | set_load_weight(&init_task); |
b50f60ce | 9065 | |
e107be36 AK |
9066 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9067 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9068 | #endif | |
9069 | ||
c9819f45 | 9070 | #ifdef CONFIG_SMP |
962cf36c | 9071 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9072 | #endif |
9073 | ||
b50f60ce HC |
9074 | #ifdef CONFIG_RT_MUTEXES |
9075 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9076 | #endif | |
9077 | ||
1da177e4 LT |
9078 | /* |
9079 | * The boot idle thread does lazy MMU switching as well: | |
9080 | */ | |
9081 | atomic_inc(&init_mm.mm_count); | |
9082 | enter_lazy_tlb(&init_mm, current); | |
9083 | ||
9084 | /* | |
9085 | * Make us the idle thread. Technically, schedule() should not be | |
9086 | * called from this thread, however somewhere below it might be, | |
9087 | * but because we are the idle thread, we just pick up running again | |
9088 | * when this runqueue becomes "idle". | |
9089 | */ | |
9090 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
9091 | /* |
9092 | * During early bootup we pretend to be a normal task: | |
9093 | */ | |
9094 | current->sched_class = &fair_sched_class; | |
6892b75e | 9095 | |
6a7b3dc3 RR |
9096 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
9097 | alloc_bootmem_cpumask_var(&nohz_cpu_mask); | |
bf4d83f6 | 9098 | #ifdef CONFIG_SMP |
7d1e6a9b RR |
9099 | #ifdef CONFIG_NO_HZ |
9100 | alloc_bootmem_cpumask_var(&nohz.cpu_mask); | |
9101 | #endif | |
dcc30a35 | 9102 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
bf4d83f6 | 9103 | #endif /* SMP */ |
6a7b3dc3 | 9104 | |
0d905bca IM |
9105 | perf_counter_init(); |
9106 | ||
6892b75e | 9107 | scheduler_running = 1; |
1da177e4 LT |
9108 | } |
9109 | ||
9110 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
9111 | void __might_sleep(char *file, int line) | |
9112 | { | |
48f24c4d | 9113 | #ifdef in_atomic |
1da177e4 LT |
9114 | static unsigned long prev_jiffy; /* ratelimiting */ |
9115 | ||
aef745fc IM |
9116 | if ((!in_atomic() && !irqs_disabled()) || |
9117 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
9118 | return; | |
9119 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9120 | return; | |
9121 | prev_jiffy = jiffies; | |
9122 | ||
9123 | printk(KERN_ERR | |
9124 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9125 | file, line); | |
9126 | printk(KERN_ERR | |
9127 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9128 | in_atomic(), irqs_disabled(), | |
9129 | current->pid, current->comm); | |
9130 | ||
9131 | debug_show_held_locks(current); | |
9132 | if (irqs_disabled()) | |
9133 | print_irqtrace_events(current); | |
9134 | dump_stack(); | |
1da177e4 LT |
9135 | #endif |
9136 | } | |
9137 | EXPORT_SYMBOL(__might_sleep); | |
9138 | #endif | |
9139 | ||
9140 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9141 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9142 | { | |
9143 | int on_rq; | |
3e51f33f | 9144 | |
3a5e4dc1 AK |
9145 | update_rq_clock(rq); |
9146 | on_rq = p->se.on_rq; | |
9147 | if (on_rq) | |
9148 | deactivate_task(rq, p, 0); | |
9149 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9150 | if (on_rq) { | |
9151 | activate_task(rq, p, 0); | |
9152 | resched_task(rq->curr); | |
9153 | } | |
9154 | } | |
9155 | ||
1da177e4 LT |
9156 | void normalize_rt_tasks(void) |
9157 | { | |
a0f98a1c | 9158 | struct task_struct *g, *p; |
1da177e4 | 9159 | unsigned long flags; |
70b97a7f | 9160 | struct rq *rq; |
1da177e4 | 9161 | |
4cf5d77a | 9162 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9163 | do_each_thread(g, p) { |
178be793 IM |
9164 | /* |
9165 | * Only normalize user tasks: | |
9166 | */ | |
9167 | if (!p->mm) | |
9168 | continue; | |
9169 | ||
6cfb0d5d | 9170 | p->se.exec_start = 0; |
6cfb0d5d | 9171 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9172 | p->se.wait_start = 0; |
dd41f596 | 9173 | p->se.sleep_start = 0; |
dd41f596 | 9174 | p->se.block_start = 0; |
6cfb0d5d | 9175 | #endif |
dd41f596 IM |
9176 | |
9177 | if (!rt_task(p)) { | |
9178 | /* | |
9179 | * Renice negative nice level userspace | |
9180 | * tasks back to 0: | |
9181 | */ | |
9182 | if (TASK_NICE(p) < 0 && p->mm) | |
9183 | set_user_nice(p, 0); | |
1da177e4 | 9184 | continue; |
dd41f596 | 9185 | } |
1da177e4 | 9186 | |
4cf5d77a | 9187 | spin_lock(&p->pi_lock); |
b29739f9 | 9188 | rq = __task_rq_lock(p); |
1da177e4 | 9189 | |
178be793 | 9190 | normalize_task(rq, p); |
3a5e4dc1 | 9191 | |
b29739f9 | 9192 | __task_rq_unlock(rq); |
4cf5d77a | 9193 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9194 | } while_each_thread(g, p); |
9195 | ||
4cf5d77a | 9196 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9197 | } |
9198 | ||
9199 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9200 | |
9201 | #ifdef CONFIG_IA64 | |
9202 | /* | |
9203 | * These functions are only useful for the IA64 MCA handling. | |
9204 | * | |
9205 | * They can only be called when the whole system has been | |
9206 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9207 | * activity can take place. Using them for anything else would | |
9208 | * be a serious bug, and as a result, they aren't even visible | |
9209 | * under any other configuration. | |
9210 | */ | |
9211 | ||
9212 | /** | |
9213 | * curr_task - return the current task for a given cpu. | |
9214 | * @cpu: the processor in question. | |
9215 | * | |
9216 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9217 | */ | |
36c8b586 | 9218 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9219 | { |
9220 | return cpu_curr(cpu); | |
9221 | } | |
9222 | ||
9223 | /** | |
9224 | * set_curr_task - set the current task for a given cpu. | |
9225 | * @cpu: the processor in question. | |
9226 | * @p: the task pointer to set. | |
9227 | * | |
9228 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9229 | * are serviced on a separate stack. It allows the architecture to switch the |
9230 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9231 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9232 | * and caller must save the original value of the current task (see | |
9233 | * curr_task() above) and restore that value before reenabling interrupts and | |
9234 | * re-starting the system. | |
9235 | * | |
9236 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9237 | */ | |
36c8b586 | 9238 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9239 | { |
9240 | cpu_curr(cpu) = p; | |
9241 | } | |
9242 | ||
9243 | #endif | |
29f59db3 | 9244 | |
bccbe08a PZ |
9245 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9246 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9247 | { |
9248 | int i; | |
9249 | ||
9250 | for_each_possible_cpu(i) { | |
9251 | if (tg->cfs_rq) | |
9252 | kfree(tg->cfs_rq[i]); | |
9253 | if (tg->se) | |
9254 | kfree(tg->se[i]); | |
6f505b16 PZ |
9255 | } |
9256 | ||
9257 | kfree(tg->cfs_rq); | |
9258 | kfree(tg->se); | |
6f505b16 PZ |
9259 | } |
9260 | ||
ec7dc8ac DG |
9261 | static |
9262 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9263 | { |
29f59db3 | 9264 | struct cfs_rq *cfs_rq; |
eab17229 | 9265 | struct sched_entity *se; |
9b5b7751 | 9266 | struct rq *rq; |
29f59db3 SV |
9267 | int i; |
9268 | ||
434d53b0 | 9269 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9270 | if (!tg->cfs_rq) |
9271 | goto err; | |
434d53b0 | 9272 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9273 | if (!tg->se) |
9274 | goto err; | |
052f1dc7 PZ |
9275 | |
9276 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9277 | |
9278 | for_each_possible_cpu(i) { | |
9b5b7751 | 9279 | rq = cpu_rq(i); |
29f59db3 | 9280 | |
eab17229 LZ |
9281 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9282 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9283 | if (!cfs_rq) |
9284 | goto err; | |
9285 | ||
eab17229 LZ |
9286 | se = kzalloc_node(sizeof(struct sched_entity), |
9287 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9288 | if (!se) |
9289 | goto err; | |
9290 | ||
eab17229 | 9291 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9292 | } |
9293 | ||
9294 | return 1; | |
9295 | ||
9296 | err: | |
9297 | return 0; | |
9298 | } | |
9299 | ||
9300 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9301 | { | |
9302 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9303 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9304 | } | |
9305 | ||
9306 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9307 | { | |
9308 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9309 | } | |
6d6bc0ad | 9310 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9311 | static inline void free_fair_sched_group(struct task_group *tg) |
9312 | { | |
9313 | } | |
9314 | ||
ec7dc8ac DG |
9315 | static inline |
9316 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9317 | { |
9318 | return 1; | |
9319 | } | |
9320 | ||
9321 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9322 | { | |
9323 | } | |
9324 | ||
9325 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9326 | { | |
9327 | } | |
6d6bc0ad | 9328 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9329 | |
9330 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9331 | static void free_rt_sched_group(struct task_group *tg) |
9332 | { | |
9333 | int i; | |
9334 | ||
d0b27fa7 PZ |
9335 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9336 | ||
bccbe08a PZ |
9337 | for_each_possible_cpu(i) { |
9338 | if (tg->rt_rq) | |
9339 | kfree(tg->rt_rq[i]); | |
9340 | if (tg->rt_se) | |
9341 | kfree(tg->rt_se[i]); | |
9342 | } | |
9343 | ||
9344 | kfree(tg->rt_rq); | |
9345 | kfree(tg->rt_se); | |
9346 | } | |
9347 | ||
ec7dc8ac DG |
9348 | static |
9349 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9350 | { |
9351 | struct rt_rq *rt_rq; | |
eab17229 | 9352 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9353 | struct rq *rq; |
9354 | int i; | |
9355 | ||
434d53b0 | 9356 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9357 | if (!tg->rt_rq) |
9358 | goto err; | |
434d53b0 | 9359 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9360 | if (!tg->rt_se) |
9361 | goto err; | |
9362 | ||
d0b27fa7 PZ |
9363 | init_rt_bandwidth(&tg->rt_bandwidth, |
9364 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9365 | |
9366 | for_each_possible_cpu(i) { | |
9367 | rq = cpu_rq(i); | |
9368 | ||
eab17229 LZ |
9369 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9370 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9371 | if (!rt_rq) |
9372 | goto err; | |
29f59db3 | 9373 | |
eab17229 LZ |
9374 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
9375 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9376 | if (!rt_se) |
9377 | goto err; | |
29f59db3 | 9378 | |
eab17229 | 9379 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
9380 | } |
9381 | ||
bccbe08a PZ |
9382 | return 1; |
9383 | ||
9384 | err: | |
9385 | return 0; | |
9386 | } | |
9387 | ||
9388 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9389 | { | |
9390 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
9391 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
9392 | } | |
9393 | ||
9394 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9395 | { | |
9396 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
9397 | } | |
6d6bc0ad | 9398 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
9399 | static inline void free_rt_sched_group(struct task_group *tg) |
9400 | { | |
9401 | } | |
9402 | ||
ec7dc8ac DG |
9403 | static inline |
9404 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9405 | { |
9406 | return 1; | |
9407 | } | |
9408 | ||
9409 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
9410 | { | |
9411 | } | |
9412 | ||
9413 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
9414 | { | |
9415 | } | |
6d6bc0ad | 9416 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 9417 | |
d0b27fa7 | 9418 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
9419 | static void free_sched_group(struct task_group *tg) |
9420 | { | |
9421 | free_fair_sched_group(tg); | |
9422 | free_rt_sched_group(tg); | |
9423 | kfree(tg); | |
9424 | } | |
9425 | ||
9426 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 9427 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
9428 | { |
9429 | struct task_group *tg; | |
9430 | unsigned long flags; | |
9431 | int i; | |
9432 | ||
9433 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
9434 | if (!tg) | |
9435 | return ERR_PTR(-ENOMEM); | |
9436 | ||
ec7dc8ac | 9437 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
9438 | goto err; |
9439 | ||
ec7dc8ac | 9440 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
9441 | goto err; |
9442 | ||
8ed36996 | 9443 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9444 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9445 | register_fair_sched_group(tg, i); |
9446 | register_rt_sched_group(tg, i); | |
9b5b7751 | 9447 | } |
6f505b16 | 9448 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
9449 | |
9450 | WARN_ON(!parent); /* root should already exist */ | |
9451 | ||
9452 | tg->parent = parent; | |
f473aa5e | 9453 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 9454 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 9455 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 9456 | |
9b5b7751 | 9457 | return tg; |
29f59db3 SV |
9458 | |
9459 | err: | |
6f505b16 | 9460 | free_sched_group(tg); |
29f59db3 SV |
9461 | return ERR_PTR(-ENOMEM); |
9462 | } | |
9463 | ||
9b5b7751 | 9464 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 9465 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 9466 | { |
29f59db3 | 9467 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 9468 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
9469 | } |
9470 | ||
9b5b7751 | 9471 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 9472 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 9473 | { |
8ed36996 | 9474 | unsigned long flags; |
9b5b7751 | 9475 | int i; |
29f59db3 | 9476 | |
8ed36996 | 9477 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 9478 | for_each_possible_cpu(i) { |
bccbe08a PZ |
9479 | unregister_fair_sched_group(tg, i); |
9480 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 9481 | } |
6f505b16 | 9482 | list_del_rcu(&tg->list); |
f473aa5e | 9483 | list_del_rcu(&tg->siblings); |
8ed36996 | 9484 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 9485 | |
9b5b7751 | 9486 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 9487 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
9488 | } |
9489 | ||
9b5b7751 | 9490 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
9491 | * The caller of this function should have put the task in its new group |
9492 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
9493 | * reflect its new group. | |
9b5b7751 SV |
9494 | */ |
9495 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
9496 | { |
9497 | int on_rq, running; | |
9498 | unsigned long flags; | |
9499 | struct rq *rq; | |
9500 | ||
9501 | rq = task_rq_lock(tsk, &flags); | |
9502 | ||
29f59db3 SV |
9503 | update_rq_clock(rq); |
9504 | ||
051a1d1a | 9505 | running = task_current(rq, tsk); |
29f59db3 SV |
9506 | on_rq = tsk->se.on_rq; |
9507 | ||
0e1f3483 | 9508 | if (on_rq) |
29f59db3 | 9509 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
9510 | if (unlikely(running)) |
9511 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 9512 | |
6f505b16 | 9513 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 9514 | |
810b3817 PZ |
9515 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9516 | if (tsk->sched_class->moved_group) | |
9517 | tsk->sched_class->moved_group(tsk); | |
9518 | #endif | |
9519 | ||
0e1f3483 HS |
9520 | if (unlikely(running)) |
9521 | tsk->sched_class->set_curr_task(rq); | |
9522 | if (on_rq) | |
7074badb | 9523 | enqueue_task(rq, tsk, 0); |
29f59db3 | 9524 | |
29f59db3 SV |
9525 | task_rq_unlock(rq, &flags); |
9526 | } | |
6d6bc0ad | 9527 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 9528 | |
052f1dc7 | 9529 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 9530 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
9531 | { |
9532 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
9533 | int on_rq; |
9534 | ||
29f59db3 | 9535 | on_rq = se->on_rq; |
62fb1851 | 9536 | if (on_rq) |
29f59db3 SV |
9537 | dequeue_entity(cfs_rq, se, 0); |
9538 | ||
9539 | se->load.weight = shares; | |
e05510d0 | 9540 | se->load.inv_weight = 0; |
29f59db3 | 9541 | |
62fb1851 | 9542 | if (on_rq) |
29f59db3 | 9543 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 9544 | } |
62fb1851 | 9545 | |
c09595f6 PZ |
9546 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
9547 | { | |
9548 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
9549 | struct rq *rq = cfs_rq->rq; | |
9550 | unsigned long flags; | |
9551 | ||
9552 | spin_lock_irqsave(&rq->lock, flags); | |
9553 | __set_se_shares(se, shares); | |
9554 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
9555 | } |
9556 | ||
8ed36996 PZ |
9557 | static DEFINE_MUTEX(shares_mutex); |
9558 | ||
4cf86d77 | 9559 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
9560 | { |
9561 | int i; | |
8ed36996 | 9562 | unsigned long flags; |
c61935fd | 9563 | |
ec7dc8ac DG |
9564 | /* |
9565 | * We can't change the weight of the root cgroup. | |
9566 | */ | |
9567 | if (!tg->se[0]) | |
9568 | return -EINVAL; | |
9569 | ||
18d95a28 PZ |
9570 | if (shares < MIN_SHARES) |
9571 | shares = MIN_SHARES; | |
cb4ad1ff MX |
9572 | else if (shares > MAX_SHARES) |
9573 | shares = MAX_SHARES; | |
62fb1851 | 9574 | |
8ed36996 | 9575 | mutex_lock(&shares_mutex); |
9b5b7751 | 9576 | if (tg->shares == shares) |
5cb350ba | 9577 | goto done; |
29f59db3 | 9578 | |
8ed36996 | 9579 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9580 | for_each_possible_cpu(i) |
9581 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 9582 | list_del_rcu(&tg->siblings); |
8ed36996 | 9583 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
9584 | |
9585 | /* wait for any ongoing reference to this group to finish */ | |
9586 | synchronize_sched(); | |
9587 | ||
9588 | /* | |
9589 | * Now we are free to modify the group's share on each cpu | |
9590 | * w/o tripping rebalance_share or load_balance_fair. | |
9591 | */ | |
9b5b7751 | 9592 | tg->shares = shares; |
c09595f6 PZ |
9593 | for_each_possible_cpu(i) { |
9594 | /* | |
9595 | * force a rebalance | |
9596 | */ | |
9597 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 9598 | set_se_shares(tg->se[i], shares); |
c09595f6 | 9599 | } |
29f59db3 | 9600 | |
6b2d7700 SV |
9601 | /* |
9602 | * Enable load balance activity on this group, by inserting it back on | |
9603 | * each cpu's rq->leaf_cfs_rq_list. | |
9604 | */ | |
8ed36996 | 9605 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
9606 | for_each_possible_cpu(i) |
9607 | register_fair_sched_group(tg, i); | |
f473aa5e | 9608 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 9609 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 9610 | done: |
8ed36996 | 9611 | mutex_unlock(&shares_mutex); |
9b5b7751 | 9612 | return 0; |
29f59db3 SV |
9613 | } |
9614 | ||
5cb350ba DG |
9615 | unsigned long sched_group_shares(struct task_group *tg) |
9616 | { | |
9617 | return tg->shares; | |
9618 | } | |
052f1dc7 | 9619 | #endif |
5cb350ba | 9620 | |
052f1dc7 | 9621 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9622 | /* |
9f0c1e56 | 9623 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 9624 | */ |
9f0c1e56 PZ |
9625 | static DEFINE_MUTEX(rt_constraints_mutex); |
9626 | ||
9627 | static unsigned long to_ratio(u64 period, u64 runtime) | |
9628 | { | |
9629 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 9630 | return 1ULL << 20; |
9f0c1e56 | 9631 | |
9a7e0b18 | 9632 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
9633 | } |
9634 | ||
9a7e0b18 PZ |
9635 | /* Must be called with tasklist_lock held */ |
9636 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 9637 | { |
9a7e0b18 | 9638 | struct task_struct *g, *p; |
b40b2e8e | 9639 | |
9a7e0b18 PZ |
9640 | do_each_thread(g, p) { |
9641 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
9642 | return 1; | |
9643 | } while_each_thread(g, p); | |
b40b2e8e | 9644 | |
9a7e0b18 PZ |
9645 | return 0; |
9646 | } | |
b40b2e8e | 9647 | |
9a7e0b18 PZ |
9648 | struct rt_schedulable_data { |
9649 | struct task_group *tg; | |
9650 | u64 rt_period; | |
9651 | u64 rt_runtime; | |
9652 | }; | |
b40b2e8e | 9653 | |
9a7e0b18 PZ |
9654 | static int tg_schedulable(struct task_group *tg, void *data) |
9655 | { | |
9656 | struct rt_schedulable_data *d = data; | |
9657 | struct task_group *child; | |
9658 | unsigned long total, sum = 0; | |
9659 | u64 period, runtime; | |
b40b2e8e | 9660 | |
9a7e0b18 PZ |
9661 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
9662 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 9663 | |
9a7e0b18 PZ |
9664 | if (tg == d->tg) { |
9665 | period = d->rt_period; | |
9666 | runtime = d->rt_runtime; | |
b40b2e8e | 9667 | } |
b40b2e8e | 9668 | |
98a4826b PZ |
9669 | #ifdef CONFIG_USER_SCHED |
9670 | if (tg == &root_task_group) { | |
9671 | period = global_rt_period(); | |
9672 | runtime = global_rt_runtime(); | |
9673 | } | |
9674 | #endif | |
9675 | ||
4653f803 PZ |
9676 | /* |
9677 | * Cannot have more runtime than the period. | |
9678 | */ | |
9679 | if (runtime > period && runtime != RUNTIME_INF) | |
9680 | return -EINVAL; | |
6f505b16 | 9681 | |
4653f803 PZ |
9682 | /* |
9683 | * Ensure we don't starve existing RT tasks. | |
9684 | */ | |
9a7e0b18 PZ |
9685 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
9686 | return -EBUSY; | |
6f505b16 | 9687 | |
9a7e0b18 | 9688 | total = to_ratio(period, runtime); |
6f505b16 | 9689 | |
4653f803 PZ |
9690 | /* |
9691 | * Nobody can have more than the global setting allows. | |
9692 | */ | |
9693 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
9694 | return -EINVAL; | |
6f505b16 | 9695 | |
4653f803 PZ |
9696 | /* |
9697 | * The sum of our children's runtime should not exceed our own. | |
9698 | */ | |
9a7e0b18 PZ |
9699 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
9700 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
9701 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 9702 | |
9a7e0b18 PZ |
9703 | if (child == d->tg) { |
9704 | period = d->rt_period; | |
9705 | runtime = d->rt_runtime; | |
9706 | } | |
6f505b16 | 9707 | |
9a7e0b18 | 9708 | sum += to_ratio(period, runtime); |
9f0c1e56 | 9709 | } |
6f505b16 | 9710 | |
9a7e0b18 PZ |
9711 | if (sum > total) |
9712 | return -EINVAL; | |
9713 | ||
9714 | return 0; | |
6f505b16 PZ |
9715 | } |
9716 | ||
9a7e0b18 | 9717 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 9718 | { |
9a7e0b18 PZ |
9719 | struct rt_schedulable_data data = { |
9720 | .tg = tg, | |
9721 | .rt_period = period, | |
9722 | .rt_runtime = runtime, | |
9723 | }; | |
9724 | ||
9725 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
9726 | } |
9727 | ||
d0b27fa7 PZ |
9728 | static int tg_set_bandwidth(struct task_group *tg, |
9729 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 9730 | { |
ac086bc2 | 9731 | int i, err = 0; |
9f0c1e56 | 9732 | |
9f0c1e56 | 9733 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 9734 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
9735 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
9736 | if (err) | |
9f0c1e56 | 9737 | goto unlock; |
ac086bc2 PZ |
9738 | |
9739 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
9740 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
9741 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
9742 | |
9743 | for_each_possible_cpu(i) { | |
9744 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
9745 | ||
9746 | spin_lock(&rt_rq->rt_runtime_lock); | |
9747 | rt_rq->rt_runtime = rt_runtime; | |
9748 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9749 | } | |
9750 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 9751 | unlock: |
521f1a24 | 9752 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
9753 | mutex_unlock(&rt_constraints_mutex); |
9754 | ||
9755 | return err; | |
6f505b16 PZ |
9756 | } |
9757 | ||
d0b27fa7 PZ |
9758 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
9759 | { | |
9760 | u64 rt_runtime, rt_period; | |
9761 | ||
9762 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9763 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
9764 | if (rt_runtime_us < 0) | |
9765 | rt_runtime = RUNTIME_INF; | |
9766 | ||
9767 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
9768 | } | |
9769 | ||
9f0c1e56 PZ |
9770 | long sched_group_rt_runtime(struct task_group *tg) |
9771 | { | |
9772 | u64 rt_runtime_us; | |
9773 | ||
d0b27fa7 | 9774 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
9775 | return -1; |
9776 | ||
d0b27fa7 | 9777 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
9778 | do_div(rt_runtime_us, NSEC_PER_USEC); |
9779 | return rt_runtime_us; | |
9780 | } | |
d0b27fa7 PZ |
9781 | |
9782 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
9783 | { | |
9784 | u64 rt_runtime, rt_period; | |
9785 | ||
9786 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
9787 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
9788 | ||
619b0488 R |
9789 | if (rt_period == 0) |
9790 | return -EINVAL; | |
9791 | ||
d0b27fa7 PZ |
9792 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
9793 | } | |
9794 | ||
9795 | long sched_group_rt_period(struct task_group *tg) | |
9796 | { | |
9797 | u64 rt_period_us; | |
9798 | ||
9799 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
9800 | do_div(rt_period_us, NSEC_PER_USEC); | |
9801 | return rt_period_us; | |
9802 | } | |
9803 | ||
9804 | static int sched_rt_global_constraints(void) | |
9805 | { | |
4653f803 | 9806 | u64 runtime, period; |
d0b27fa7 PZ |
9807 | int ret = 0; |
9808 | ||
ec5d4989 HS |
9809 | if (sysctl_sched_rt_period <= 0) |
9810 | return -EINVAL; | |
9811 | ||
4653f803 PZ |
9812 | runtime = global_rt_runtime(); |
9813 | period = global_rt_period(); | |
9814 | ||
9815 | /* | |
9816 | * Sanity check on the sysctl variables. | |
9817 | */ | |
9818 | if (runtime > period && runtime != RUNTIME_INF) | |
9819 | return -EINVAL; | |
10b612f4 | 9820 | |
d0b27fa7 | 9821 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9822 | read_lock(&tasklist_lock); |
4653f803 | 9823 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9824 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9825 | mutex_unlock(&rt_constraints_mutex); |
9826 | ||
9827 | return ret; | |
9828 | } | |
54e99124 DG |
9829 | |
9830 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
9831 | { | |
9832 | /* Don't accept realtime tasks when there is no way for them to run */ | |
9833 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
9834 | return 0; | |
9835 | ||
9836 | return 1; | |
9837 | } | |
9838 | ||
6d6bc0ad | 9839 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9840 | static int sched_rt_global_constraints(void) |
9841 | { | |
ac086bc2 PZ |
9842 | unsigned long flags; |
9843 | int i; | |
9844 | ||
ec5d4989 HS |
9845 | if (sysctl_sched_rt_period <= 0) |
9846 | return -EINVAL; | |
9847 | ||
ac086bc2 PZ |
9848 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
9849 | for_each_possible_cpu(i) { | |
9850 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9851 | ||
9852 | spin_lock(&rt_rq->rt_runtime_lock); | |
9853 | rt_rq->rt_runtime = global_rt_runtime(); | |
9854 | spin_unlock(&rt_rq->rt_runtime_lock); | |
9855 | } | |
9856 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
9857 | ||
d0b27fa7 PZ |
9858 | return 0; |
9859 | } | |
6d6bc0ad | 9860 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9861 | |
9862 | int sched_rt_handler(struct ctl_table *table, int write, | |
9863 | struct file *filp, void __user *buffer, size_t *lenp, | |
9864 | loff_t *ppos) | |
9865 | { | |
9866 | int ret; | |
9867 | int old_period, old_runtime; | |
9868 | static DEFINE_MUTEX(mutex); | |
9869 | ||
9870 | mutex_lock(&mutex); | |
9871 | old_period = sysctl_sched_rt_period; | |
9872 | old_runtime = sysctl_sched_rt_runtime; | |
9873 | ||
9874 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
9875 | ||
9876 | if (!ret && write) { | |
9877 | ret = sched_rt_global_constraints(); | |
9878 | if (ret) { | |
9879 | sysctl_sched_rt_period = old_period; | |
9880 | sysctl_sched_rt_runtime = old_runtime; | |
9881 | } else { | |
9882 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9883 | def_rt_bandwidth.rt_period = | |
9884 | ns_to_ktime(global_rt_period()); | |
9885 | } | |
9886 | } | |
9887 | mutex_unlock(&mutex); | |
9888 | ||
9889 | return ret; | |
9890 | } | |
68318b8e | 9891 | |
052f1dc7 | 9892 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9893 | |
9894 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9895 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9896 | { |
2b01dfe3 PM |
9897 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9898 | struct task_group, css); | |
68318b8e SV |
9899 | } |
9900 | ||
9901 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9902 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9903 | { |
ec7dc8ac | 9904 | struct task_group *tg, *parent; |
68318b8e | 9905 | |
2b01dfe3 | 9906 | if (!cgrp->parent) { |
68318b8e | 9907 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9908 | return &init_task_group.css; |
9909 | } | |
9910 | ||
ec7dc8ac DG |
9911 | parent = cgroup_tg(cgrp->parent); |
9912 | tg = sched_create_group(parent); | |
68318b8e SV |
9913 | if (IS_ERR(tg)) |
9914 | return ERR_PTR(-ENOMEM); | |
9915 | ||
68318b8e SV |
9916 | return &tg->css; |
9917 | } | |
9918 | ||
41a2d6cf IM |
9919 | static void |
9920 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9921 | { |
2b01dfe3 | 9922 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9923 | |
9924 | sched_destroy_group(tg); | |
9925 | } | |
9926 | ||
41a2d6cf IM |
9927 | static int |
9928 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9929 | struct task_struct *tsk) | |
68318b8e | 9930 | { |
b68aa230 | 9931 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9932 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9933 | return -EINVAL; |
9934 | #else | |
68318b8e SV |
9935 | /* We don't support RT-tasks being in separate groups */ |
9936 | if (tsk->sched_class != &fair_sched_class) | |
9937 | return -EINVAL; | |
b68aa230 | 9938 | #endif |
68318b8e SV |
9939 | |
9940 | return 0; | |
9941 | } | |
9942 | ||
9943 | static void | |
2b01dfe3 | 9944 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
9945 | struct cgroup *old_cont, struct task_struct *tsk) |
9946 | { | |
9947 | sched_move_task(tsk); | |
9948 | } | |
9949 | ||
052f1dc7 | 9950 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9951 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9952 | u64 shareval) |
68318b8e | 9953 | { |
2b01dfe3 | 9954 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9955 | } |
9956 | ||
f4c753b7 | 9957 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9958 | { |
2b01dfe3 | 9959 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9960 | |
9961 | return (u64) tg->shares; | |
9962 | } | |
6d6bc0ad | 9963 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9964 | |
052f1dc7 | 9965 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9966 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9967 | s64 val) |
6f505b16 | 9968 | { |
06ecb27c | 9969 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9970 | } |
9971 | ||
06ecb27c | 9972 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9973 | { |
06ecb27c | 9974 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9975 | } |
d0b27fa7 PZ |
9976 | |
9977 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9978 | u64 rt_period_us) | |
9979 | { | |
9980 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9981 | } | |
9982 | ||
9983 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9984 | { | |
9985 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9986 | } | |
6d6bc0ad | 9987 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9988 | |
fe5c7cc2 | 9989 | static struct cftype cpu_files[] = { |
052f1dc7 | 9990 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9991 | { |
9992 | .name = "shares", | |
f4c753b7 PM |
9993 | .read_u64 = cpu_shares_read_u64, |
9994 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9995 | }, |
052f1dc7 PZ |
9996 | #endif |
9997 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9998 | { |
9f0c1e56 | 9999 | .name = "rt_runtime_us", |
06ecb27c PM |
10000 | .read_s64 = cpu_rt_runtime_read, |
10001 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10002 | }, |
d0b27fa7 PZ |
10003 | { |
10004 | .name = "rt_period_us", | |
f4c753b7 PM |
10005 | .read_u64 = cpu_rt_period_read_uint, |
10006 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10007 | }, |
052f1dc7 | 10008 | #endif |
68318b8e SV |
10009 | }; |
10010 | ||
10011 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10012 | { | |
fe5c7cc2 | 10013 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10014 | } |
10015 | ||
10016 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10017 | .name = "cpu", |
10018 | .create = cpu_cgroup_create, | |
10019 | .destroy = cpu_cgroup_destroy, | |
10020 | .can_attach = cpu_cgroup_can_attach, | |
10021 | .attach = cpu_cgroup_attach, | |
10022 | .populate = cpu_cgroup_populate, | |
10023 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10024 | .early_init = 1, |
10025 | }; | |
10026 | ||
052f1dc7 | 10027 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10028 | |
10029 | #ifdef CONFIG_CGROUP_CPUACCT | |
10030 | ||
10031 | /* | |
10032 | * CPU accounting code for task groups. | |
10033 | * | |
10034 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10035 | * (balbir@in.ibm.com). | |
10036 | */ | |
10037 | ||
934352f2 | 10038 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10039 | struct cpuacct { |
10040 | struct cgroup_subsys_state css; | |
10041 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10042 | u64 *cpuusage; | |
ef12fefa | 10043 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10044 | struct cpuacct *parent; |
d842de87 SV |
10045 | }; |
10046 | ||
10047 | struct cgroup_subsys cpuacct_subsys; | |
10048 | ||
10049 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10050 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10051 | { |
32cd756a | 10052 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10053 | struct cpuacct, css); |
10054 | } | |
10055 | ||
10056 | /* return cpu accounting group to which this task belongs */ | |
10057 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10058 | { | |
10059 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10060 | struct cpuacct, css); | |
10061 | } | |
10062 | ||
10063 | /* create a new cpu accounting group */ | |
10064 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10065 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10066 | { |
10067 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10068 | int i; |
d842de87 SV |
10069 | |
10070 | if (!ca) | |
ef12fefa | 10071 | goto out; |
d842de87 SV |
10072 | |
10073 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10074 | if (!ca->cpuusage) |
10075 | goto out_free_ca; | |
10076 | ||
10077 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10078 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10079 | goto out_free_counters; | |
d842de87 | 10080 | |
934352f2 BR |
10081 | if (cgrp->parent) |
10082 | ca->parent = cgroup_ca(cgrp->parent); | |
10083 | ||
d842de87 | 10084 | return &ca->css; |
ef12fefa BR |
10085 | |
10086 | out_free_counters: | |
10087 | while (--i >= 0) | |
10088 | percpu_counter_destroy(&ca->cpustat[i]); | |
10089 | free_percpu(ca->cpuusage); | |
10090 | out_free_ca: | |
10091 | kfree(ca); | |
10092 | out: | |
10093 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10094 | } |
10095 | ||
10096 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10097 | static void |
32cd756a | 10098 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10099 | { |
32cd756a | 10100 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10101 | int i; |
d842de87 | 10102 | |
ef12fefa BR |
10103 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10104 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10105 | free_percpu(ca->cpuusage); |
10106 | kfree(ca); | |
10107 | } | |
10108 | ||
720f5498 KC |
10109 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10110 | { | |
b36128c8 | 10111 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10112 | u64 data; |
10113 | ||
10114 | #ifndef CONFIG_64BIT | |
10115 | /* | |
10116 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10117 | */ | |
10118 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10119 | data = *cpuusage; | |
10120 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10121 | #else | |
10122 | data = *cpuusage; | |
10123 | #endif | |
10124 | ||
10125 | return data; | |
10126 | } | |
10127 | ||
10128 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10129 | { | |
b36128c8 | 10130 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10131 | |
10132 | #ifndef CONFIG_64BIT | |
10133 | /* | |
10134 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10135 | */ | |
10136 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10137 | *cpuusage = val; | |
10138 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10139 | #else | |
10140 | *cpuusage = val; | |
10141 | #endif | |
10142 | } | |
10143 | ||
d842de87 | 10144 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10145 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10146 | { |
32cd756a | 10147 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10148 | u64 totalcpuusage = 0; |
10149 | int i; | |
10150 | ||
720f5498 KC |
10151 | for_each_present_cpu(i) |
10152 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10153 | |
10154 | return totalcpuusage; | |
10155 | } | |
10156 | ||
0297b803 DG |
10157 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10158 | u64 reset) | |
10159 | { | |
10160 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10161 | int err = 0; | |
10162 | int i; | |
10163 | ||
10164 | if (reset) { | |
10165 | err = -EINVAL; | |
10166 | goto out; | |
10167 | } | |
10168 | ||
720f5498 KC |
10169 | for_each_present_cpu(i) |
10170 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10171 | |
0297b803 DG |
10172 | out: |
10173 | return err; | |
10174 | } | |
10175 | ||
e9515c3c KC |
10176 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10177 | struct seq_file *m) | |
10178 | { | |
10179 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10180 | u64 percpu; | |
10181 | int i; | |
10182 | ||
10183 | for_each_present_cpu(i) { | |
10184 | percpu = cpuacct_cpuusage_read(ca, i); | |
10185 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10186 | } | |
10187 | seq_printf(m, "\n"); | |
10188 | return 0; | |
10189 | } | |
10190 | ||
ef12fefa BR |
10191 | static const char *cpuacct_stat_desc[] = { |
10192 | [CPUACCT_STAT_USER] = "user", | |
10193 | [CPUACCT_STAT_SYSTEM] = "system", | |
10194 | }; | |
10195 | ||
10196 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10197 | struct cgroup_map_cb *cb) | |
10198 | { | |
10199 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10200 | int i; | |
10201 | ||
10202 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10203 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10204 | val = cputime64_to_clock_t(val); | |
10205 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10206 | } | |
10207 | return 0; | |
10208 | } | |
10209 | ||
d842de87 SV |
10210 | static struct cftype files[] = { |
10211 | { | |
10212 | .name = "usage", | |
f4c753b7 PM |
10213 | .read_u64 = cpuusage_read, |
10214 | .write_u64 = cpuusage_write, | |
d842de87 | 10215 | }, |
e9515c3c KC |
10216 | { |
10217 | .name = "usage_percpu", | |
10218 | .read_seq_string = cpuacct_percpu_seq_read, | |
10219 | }, | |
ef12fefa BR |
10220 | { |
10221 | .name = "stat", | |
10222 | .read_map = cpuacct_stats_show, | |
10223 | }, | |
d842de87 SV |
10224 | }; |
10225 | ||
32cd756a | 10226 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10227 | { |
32cd756a | 10228 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10229 | } |
10230 | ||
10231 | /* | |
10232 | * charge this task's execution time to its accounting group. | |
10233 | * | |
10234 | * called with rq->lock held. | |
10235 | */ | |
10236 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10237 | { | |
10238 | struct cpuacct *ca; | |
934352f2 | 10239 | int cpu; |
d842de87 | 10240 | |
c40c6f85 | 10241 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10242 | return; |
10243 | ||
934352f2 | 10244 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10245 | |
10246 | rcu_read_lock(); | |
10247 | ||
d842de87 | 10248 | ca = task_ca(tsk); |
d842de87 | 10249 | |
934352f2 | 10250 | for (; ca; ca = ca->parent) { |
b36128c8 | 10251 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10252 | *cpuusage += cputime; |
10253 | } | |
a18b83b7 BR |
10254 | |
10255 | rcu_read_unlock(); | |
d842de87 SV |
10256 | } |
10257 | ||
ef12fefa BR |
10258 | /* |
10259 | * Charge the system/user time to the task's accounting group. | |
10260 | */ | |
10261 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10262 | enum cpuacct_stat_index idx, cputime_t val) | |
10263 | { | |
10264 | struct cpuacct *ca; | |
10265 | ||
10266 | if (unlikely(!cpuacct_subsys.active)) | |
10267 | return; | |
10268 | ||
10269 | rcu_read_lock(); | |
10270 | ca = task_ca(tsk); | |
10271 | ||
10272 | do { | |
10273 | percpu_counter_add(&ca->cpustat[idx], val); | |
10274 | ca = ca->parent; | |
10275 | } while (ca); | |
10276 | rcu_read_unlock(); | |
d842de87 SV |
10277 | } |
10278 | ||
10279 | struct cgroup_subsys cpuacct_subsys = { | |
10280 | .name = "cpuacct", | |
10281 | .create = cpuacct_create, | |
10282 | .destroy = cpuacct_destroy, | |
10283 | .populate = cpuacct_populate, | |
10284 | .subsys_id = cpuacct_subsys_id, | |
10285 | }; | |
10286 | #endif /* CONFIG_CGROUP_CPUACCT */ |