]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
0d905bca | 42 | #include <linux/perf_counter.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
58 | #include <linux/kthread.h> | |
b5aadf7f | 59 | #include <linux/proc_fs.h> |
1da177e4 | 60 | #include <linux/seq_file.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
5517d86b | 67 | #include <linux/reciprocal_div.h> |
dff06c15 | 68 | #include <linux/unistd.h> |
f5ff8422 | 69 | #include <linux/pagemap.h> |
8f4d37ec | 70 | #include <linux/hrtimer.h> |
30914a58 | 71 | #include <linux/tick.h> |
f00b45c1 PZ |
72 | #include <linux/debugfs.h> |
73 | #include <linux/ctype.h> | |
6cd8a4bb | 74 | #include <linux/ftrace.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 GH |
79 | #include "sched_cpupri.h" |
80 | ||
a8d154b0 | 81 | #define CREATE_TRACE_POINTS |
ad8d75ff | 82 | #include <trace/events/sched.h> |
a8d154b0 | 83 | |
1da177e4 LT |
84 | /* |
85 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
86 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
87 | * and back. | |
88 | */ | |
89 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
90 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
91 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
92 | ||
93 | /* | |
94 | * 'User priority' is the nice value converted to something we | |
95 | * can work with better when scaling various scheduler parameters, | |
96 | * it's a [ 0 ... 39 ] range. | |
97 | */ | |
98 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
99 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
100 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
101 | ||
102 | /* | |
d7876a08 | 103 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 104 | */ |
d6322faf | 105 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 106 | |
6aa645ea IM |
107 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
108 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
109 | ||
1da177e4 LT |
110 | /* |
111 | * These are the 'tuning knobs' of the scheduler: | |
112 | * | |
a4ec24b4 | 113 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
114 | * Timeslices get refilled after they expire. |
115 | */ | |
1da177e4 | 116 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 117 | |
d0b27fa7 PZ |
118 | /* |
119 | * single value that denotes runtime == period, ie unlimited time. | |
120 | */ | |
121 | #define RUNTIME_INF ((u64)~0ULL) | |
122 | ||
5517d86b | 123 | #ifdef CONFIG_SMP |
fd2ab30b SN |
124 | |
125 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); | |
126 | ||
5517d86b ED |
127 | /* |
128 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
129 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
130 | */ | |
131 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
132 | { | |
133 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
134 | } | |
135 | ||
136 | /* | |
137 | * Each time a sched group cpu_power is changed, | |
138 | * we must compute its reciprocal value | |
139 | */ | |
140 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
141 | { | |
142 | sg->__cpu_power += val; | |
143 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
144 | } | |
145 | #endif | |
146 | ||
e05606d3 IM |
147 | static inline int rt_policy(int policy) |
148 | { | |
3f33a7ce | 149 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
150 | return 1; |
151 | return 0; | |
152 | } | |
153 | ||
154 | static inline int task_has_rt_policy(struct task_struct *p) | |
155 | { | |
156 | return rt_policy(p->policy); | |
157 | } | |
158 | ||
1da177e4 | 159 | /* |
6aa645ea | 160 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 161 | */ |
6aa645ea IM |
162 | struct rt_prio_array { |
163 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
164 | struct list_head queue[MAX_RT_PRIO]; | |
165 | }; | |
166 | ||
d0b27fa7 | 167 | struct rt_bandwidth { |
ea736ed5 IM |
168 | /* nests inside the rq lock: */ |
169 | spinlock_t rt_runtime_lock; | |
170 | ktime_t rt_period; | |
171 | u64 rt_runtime; | |
172 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
173 | }; |
174 | ||
175 | static struct rt_bandwidth def_rt_bandwidth; | |
176 | ||
177 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
178 | ||
179 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
180 | { | |
181 | struct rt_bandwidth *rt_b = | |
182 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
183 | ktime_t now; | |
184 | int overrun; | |
185 | int idle = 0; | |
186 | ||
187 | for (;;) { | |
188 | now = hrtimer_cb_get_time(timer); | |
189 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
190 | ||
191 | if (!overrun) | |
192 | break; | |
193 | ||
194 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
195 | } | |
196 | ||
197 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
198 | } | |
199 | ||
200 | static | |
201 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
202 | { | |
203 | rt_b->rt_period = ns_to_ktime(period); | |
204 | rt_b->rt_runtime = runtime; | |
205 | ||
ac086bc2 PZ |
206 | spin_lock_init(&rt_b->rt_runtime_lock); |
207 | ||
d0b27fa7 PZ |
208 | hrtimer_init(&rt_b->rt_period_timer, |
209 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
210 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
211 | } |
212 | ||
c8bfff6d KH |
213 | static inline int rt_bandwidth_enabled(void) |
214 | { | |
215 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
216 | } |
217 | ||
218 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
219 | { | |
220 | ktime_t now; | |
221 | ||
cac64d00 | 222 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
223 | return; |
224 | ||
225 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
226 | return; | |
227 | ||
228 | spin_lock(&rt_b->rt_runtime_lock); | |
229 | for (;;) { | |
7f1e2ca9 PZ |
230 | unsigned long delta; |
231 | ktime_t soft, hard; | |
232 | ||
d0b27fa7 PZ |
233 | if (hrtimer_active(&rt_b->rt_period_timer)) |
234 | break; | |
235 | ||
236 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
237 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
238 | |
239 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
240 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
241 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
242 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 243 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 PZ |
244 | } |
245 | spin_unlock(&rt_b->rt_runtime_lock); | |
246 | } | |
247 | ||
248 | #ifdef CONFIG_RT_GROUP_SCHED | |
249 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
250 | { | |
251 | hrtimer_cancel(&rt_b->rt_period_timer); | |
252 | } | |
253 | #endif | |
254 | ||
712555ee HC |
255 | /* |
256 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
257 | * detach_destroy_domains and partition_sched_domains. | |
258 | */ | |
259 | static DEFINE_MUTEX(sched_domains_mutex); | |
260 | ||
052f1dc7 | 261 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 262 | |
68318b8e SV |
263 | #include <linux/cgroup.h> |
264 | ||
29f59db3 SV |
265 | struct cfs_rq; |
266 | ||
6f505b16 PZ |
267 | static LIST_HEAD(task_groups); |
268 | ||
29f59db3 | 269 | /* task group related information */ |
4cf86d77 | 270 | struct task_group { |
052f1dc7 | 271 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
272 | struct cgroup_subsys_state css; |
273 | #endif | |
052f1dc7 | 274 | |
6c415b92 AB |
275 | #ifdef CONFIG_USER_SCHED |
276 | uid_t uid; | |
277 | #endif | |
278 | ||
052f1dc7 | 279 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
280 | /* schedulable entities of this group on each cpu */ |
281 | struct sched_entity **se; | |
282 | /* runqueue "owned" by this group on each cpu */ | |
283 | struct cfs_rq **cfs_rq; | |
284 | unsigned long shares; | |
052f1dc7 PZ |
285 | #endif |
286 | ||
287 | #ifdef CONFIG_RT_GROUP_SCHED | |
288 | struct sched_rt_entity **rt_se; | |
289 | struct rt_rq **rt_rq; | |
290 | ||
d0b27fa7 | 291 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 292 | #endif |
6b2d7700 | 293 | |
ae8393e5 | 294 | struct rcu_head rcu; |
6f505b16 | 295 | struct list_head list; |
f473aa5e PZ |
296 | |
297 | struct task_group *parent; | |
298 | struct list_head siblings; | |
299 | struct list_head children; | |
29f59db3 SV |
300 | }; |
301 | ||
354d60c2 | 302 | #ifdef CONFIG_USER_SCHED |
eff766a6 | 303 | |
6c415b92 AB |
304 | /* Helper function to pass uid information to create_sched_user() */ |
305 | void set_tg_uid(struct user_struct *user) | |
306 | { | |
307 | user->tg->uid = user->uid; | |
308 | } | |
309 | ||
eff766a6 PZ |
310 | /* |
311 | * Root task group. | |
84e9dabf AS |
312 | * Every UID task group (including init_task_group aka UID-0) will |
313 | * be a child to this group. | |
eff766a6 PZ |
314 | */ |
315 | struct task_group root_task_group; | |
316 | ||
052f1dc7 | 317 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
318 | /* Default task group's sched entity on each cpu */ |
319 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
320 | /* Default task group's cfs_rq on each cpu */ | |
84e9dabf | 321 | static DEFINE_PER_CPU(struct cfs_rq, init_tg_cfs_rq) ____cacheline_aligned_in_smp; |
6d6bc0ad | 322 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
323 | |
324 | #ifdef CONFIG_RT_GROUP_SCHED | |
325 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
326 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
6d6bc0ad | 327 | #endif /* CONFIG_RT_GROUP_SCHED */ |
9a7e0b18 | 328 | #else /* !CONFIG_USER_SCHED */ |
eff766a6 | 329 | #define root_task_group init_task_group |
9a7e0b18 | 330 | #endif /* CONFIG_USER_SCHED */ |
6f505b16 | 331 | |
8ed36996 | 332 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
333 | * a task group's cpu shares. |
334 | */ | |
8ed36996 | 335 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 336 | |
57310a98 PZ |
337 | #ifdef CONFIG_SMP |
338 | static int root_task_group_empty(void) | |
339 | { | |
340 | return list_empty(&root_task_group.children); | |
341 | } | |
342 | #endif | |
343 | ||
052f1dc7 | 344 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
345 | #ifdef CONFIG_USER_SCHED |
346 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
6d6bc0ad | 347 | #else /* !CONFIG_USER_SCHED */ |
052f1dc7 | 348 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
6d6bc0ad | 349 | #endif /* CONFIG_USER_SCHED */ |
052f1dc7 | 350 | |
cb4ad1ff | 351 | /* |
2e084786 LJ |
352 | * A weight of 0 or 1 can cause arithmetics problems. |
353 | * A weight of a cfs_rq is the sum of weights of which entities | |
354 | * are queued on this cfs_rq, so a weight of a entity should not be | |
355 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
356 | * (The default weight is 1024 - so there's no practical |
357 | * limitation from this.) | |
358 | */ | |
18d95a28 | 359 | #define MIN_SHARES 2 |
2e084786 | 360 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 361 | |
052f1dc7 PZ |
362 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
363 | #endif | |
364 | ||
29f59db3 | 365 | /* Default task group. |
3a252015 | 366 | * Every task in system belong to this group at bootup. |
29f59db3 | 367 | */ |
434d53b0 | 368 | struct task_group init_task_group; |
29f59db3 SV |
369 | |
370 | /* return group to which a task belongs */ | |
4cf86d77 | 371 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 372 | { |
4cf86d77 | 373 | struct task_group *tg; |
9b5b7751 | 374 | |
052f1dc7 | 375 | #ifdef CONFIG_USER_SCHED |
c69e8d9c DH |
376 | rcu_read_lock(); |
377 | tg = __task_cred(p)->user->tg; | |
378 | rcu_read_unlock(); | |
052f1dc7 | 379 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
380 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
381 | struct task_group, css); | |
24e377a8 | 382 | #else |
41a2d6cf | 383 | tg = &init_task_group; |
24e377a8 | 384 | #endif |
9b5b7751 | 385 | return tg; |
29f59db3 SV |
386 | } |
387 | ||
388 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 389 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 390 | { |
052f1dc7 | 391 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
392 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
393 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 394 | #endif |
6f505b16 | 395 | |
052f1dc7 | 396 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
397 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
398 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 399 | #endif |
29f59db3 SV |
400 | } |
401 | ||
402 | #else | |
403 | ||
57310a98 PZ |
404 | #ifdef CONFIG_SMP |
405 | static int root_task_group_empty(void) | |
406 | { | |
407 | return 1; | |
408 | } | |
409 | #endif | |
410 | ||
6f505b16 | 411 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
412 | static inline struct task_group *task_group(struct task_struct *p) |
413 | { | |
414 | return NULL; | |
415 | } | |
29f59db3 | 416 | |
052f1dc7 | 417 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 418 | |
6aa645ea IM |
419 | /* CFS-related fields in a runqueue */ |
420 | struct cfs_rq { | |
421 | struct load_weight load; | |
422 | unsigned long nr_running; | |
423 | ||
6aa645ea | 424 | u64 exec_clock; |
e9acbff6 | 425 | u64 min_vruntime; |
6aa645ea IM |
426 | |
427 | struct rb_root tasks_timeline; | |
428 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
429 | |
430 | struct list_head tasks; | |
431 | struct list_head *balance_iterator; | |
432 | ||
433 | /* | |
434 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
435 | * It is set to NULL otherwise (i.e when none are currently running). |
436 | */ | |
4793241b | 437 | struct sched_entity *curr, *next, *last; |
ddc97297 | 438 | |
5ac5c4d6 | 439 | unsigned int nr_spread_over; |
ddc97297 | 440 | |
62160e3f | 441 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
442 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
443 | ||
41a2d6cf IM |
444 | /* |
445 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
446 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
447 | * (like users, containers etc.) | |
448 | * | |
449 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
450 | * list is used during load balance. | |
451 | */ | |
41a2d6cf IM |
452 | struct list_head leaf_cfs_rq_list; |
453 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
454 | |
455 | #ifdef CONFIG_SMP | |
c09595f6 | 456 | /* |
c8cba857 | 457 | * the part of load.weight contributed by tasks |
c09595f6 | 458 | */ |
c8cba857 | 459 | unsigned long task_weight; |
c09595f6 | 460 | |
c8cba857 PZ |
461 | /* |
462 | * h_load = weight * f(tg) | |
463 | * | |
464 | * Where f(tg) is the recursive weight fraction assigned to | |
465 | * this group. | |
466 | */ | |
467 | unsigned long h_load; | |
c09595f6 | 468 | |
c8cba857 PZ |
469 | /* |
470 | * this cpu's part of tg->shares | |
471 | */ | |
472 | unsigned long shares; | |
f1d239f7 PZ |
473 | |
474 | /* | |
475 | * load.weight at the time we set shares | |
476 | */ | |
477 | unsigned long rq_weight; | |
c09595f6 | 478 | #endif |
6aa645ea IM |
479 | #endif |
480 | }; | |
1da177e4 | 481 | |
6aa645ea IM |
482 | /* Real-Time classes' related field in a runqueue: */ |
483 | struct rt_rq { | |
484 | struct rt_prio_array active; | |
63489e45 | 485 | unsigned long rt_nr_running; |
052f1dc7 | 486 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
487 | struct { |
488 | int curr; /* highest queued rt task prio */ | |
398a153b | 489 | #ifdef CONFIG_SMP |
e864c499 | 490 | int next; /* next highest */ |
398a153b | 491 | #endif |
e864c499 | 492 | } highest_prio; |
6f505b16 | 493 | #endif |
fa85ae24 | 494 | #ifdef CONFIG_SMP |
73fe6aae | 495 | unsigned long rt_nr_migratory; |
a1ba4d8b | 496 | unsigned long rt_nr_total; |
a22d7fc1 | 497 | int overloaded; |
917b627d | 498 | struct plist_head pushable_tasks; |
fa85ae24 | 499 | #endif |
6f505b16 | 500 | int rt_throttled; |
fa85ae24 | 501 | u64 rt_time; |
ac086bc2 | 502 | u64 rt_runtime; |
ea736ed5 | 503 | /* Nests inside the rq lock: */ |
ac086bc2 | 504 | spinlock_t rt_runtime_lock; |
6f505b16 | 505 | |
052f1dc7 | 506 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
507 | unsigned long rt_nr_boosted; |
508 | ||
6f505b16 PZ |
509 | struct rq *rq; |
510 | struct list_head leaf_rt_rq_list; | |
511 | struct task_group *tg; | |
512 | struct sched_rt_entity *rt_se; | |
513 | #endif | |
6aa645ea IM |
514 | }; |
515 | ||
57d885fe GH |
516 | #ifdef CONFIG_SMP |
517 | ||
518 | /* | |
519 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
520 | * variables. Each exclusive cpuset essentially defines an island domain by |
521 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
522 | * exclusive cpuset is created, we also create and attach a new root-domain |
523 | * object. | |
524 | * | |
57d885fe GH |
525 | */ |
526 | struct root_domain { | |
527 | atomic_t refcount; | |
c6c4927b RR |
528 | cpumask_var_t span; |
529 | cpumask_var_t online; | |
637f5085 | 530 | |
0eab9146 | 531 | /* |
637f5085 GH |
532 | * The "RT overload" flag: it gets set if a CPU has more than |
533 | * one runnable RT task. | |
534 | */ | |
c6c4927b | 535 | cpumask_var_t rto_mask; |
0eab9146 | 536 | atomic_t rto_count; |
6e0534f2 GH |
537 | #ifdef CONFIG_SMP |
538 | struct cpupri cpupri; | |
539 | #endif | |
7a09b1a2 VS |
540 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
541 | /* | |
542 | * Preferred wake up cpu nominated by sched_mc balance that will be | |
543 | * used when most cpus are idle in the system indicating overall very | |
544 | * low system utilisation. Triggered at POWERSAVINGS_BALANCE_WAKEUP(2) | |
545 | */ | |
546 | unsigned int sched_mc_preferred_wakeup_cpu; | |
547 | #endif | |
57d885fe GH |
548 | }; |
549 | ||
dc938520 GH |
550 | /* |
551 | * By default the system creates a single root-domain with all cpus as | |
552 | * members (mimicking the global state we have today). | |
553 | */ | |
57d885fe GH |
554 | static struct root_domain def_root_domain; |
555 | ||
556 | #endif | |
557 | ||
1da177e4 LT |
558 | /* |
559 | * This is the main, per-CPU runqueue data structure. | |
560 | * | |
561 | * Locking rule: those places that want to lock multiple runqueues | |
562 | * (such as the load balancing or the thread migration code), lock | |
563 | * acquire operations must be ordered by ascending &runqueue. | |
564 | */ | |
70b97a7f | 565 | struct rq { |
d8016491 IM |
566 | /* runqueue lock: */ |
567 | spinlock_t lock; | |
1da177e4 LT |
568 | |
569 | /* | |
570 | * nr_running and cpu_load should be in the same cacheline because | |
571 | * remote CPUs use both these fields when doing load calculation. | |
572 | */ | |
573 | unsigned long nr_running; | |
6aa645ea IM |
574 | #define CPU_LOAD_IDX_MAX 5 |
575 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 576 | #ifdef CONFIG_NO_HZ |
15934a37 | 577 | unsigned long last_tick_seen; |
46cb4b7c SS |
578 | unsigned char in_nohz_recently; |
579 | #endif | |
d8016491 IM |
580 | /* capture load from *all* tasks on this cpu: */ |
581 | struct load_weight load; | |
6aa645ea IM |
582 | unsigned long nr_load_updates; |
583 | u64 nr_switches; | |
23a185ca | 584 | u64 nr_migrations_in; |
6aa645ea IM |
585 | |
586 | struct cfs_rq cfs; | |
6f505b16 | 587 | struct rt_rq rt; |
6f505b16 | 588 | |
6aa645ea | 589 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
590 | /* list of leaf cfs_rq on this cpu: */ |
591 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
592 | #endif |
593 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 594 | struct list_head leaf_rt_rq_list; |
1da177e4 | 595 | #endif |
1da177e4 LT |
596 | |
597 | /* | |
598 | * This is part of a global counter where only the total sum | |
599 | * over all CPUs matters. A task can increase this counter on | |
600 | * one CPU and if it got migrated afterwards it may decrease | |
601 | * it on another CPU. Always updated under the runqueue lock: | |
602 | */ | |
603 | unsigned long nr_uninterruptible; | |
604 | ||
36c8b586 | 605 | struct task_struct *curr, *idle; |
c9819f45 | 606 | unsigned long next_balance; |
1da177e4 | 607 | struct mm_struct *prev_mm; |
6aa645ea | 608 | |
3e51f33f | 609 | u64 clock; |
6aa645ea | 610 | |
1da177e4 LT |
611 | atomic_t nr_iowait; |
612 | ||
613 | #ifdef CONFIG_SMP | |
0eab9146 | 614 | struct root_domain *rd; |
1da177e4 LT |
615 | struct sched_domain *sd; |
616 | ||
a0a522ce | 617 | unsigned char idle_at_tick; |
1da177e4 | 618 | /* For active balancing */ |
3f029d3c | 619 | int post_schedule; |
1da177e4 LT |
620 | int active_balance; |
621 | int push_cpu; | |
d8016491 IM |
622 | /* cpu of this runqueue: */ |
623 | int cpu; | |
1f11eb6a | 624 | int online; |
1da177e4 | 625 | |
a8a51d5e | 626 | unsigned long avg_load_per_task; |
1da177e4 | 627 | |
36c8b586 | 628 | struct task_struct *migration_thread; |
1da177e4 | 629 | struct list_head migration_queue; |
e9e9250b PZ |
630 | |
631 | u64 rt_avg; | |
632 | u64 age_stamp; | |
1da177e4 LT |
633 | #endif |
634 | ||
dce48a84 TG |
635 | /* calc_load related fields */ |
636 | unsigned long calc_load_update; | |
637 | long calc_load_active; | |
638 | ||
8f4d37ec | 639 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
640 | #ifdef CONFIG_SMP |
641 | int hrtick_csd_pending; | |
642 | struct call_single_data hrtick_csd; | |
643 | #endif | |
8f4d37ec PZ |
644 | struct hrtimer hrtick_timer; |
645 | #endif | |
646 | ||
1da177e4 LT |
647 | #ifdef CONFIG_SCHEDSTATS |
648 | /* latency stats */ | |
649 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
650 | unsigned long long rq_cpu_time; |
651 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
652 | |
653 | /* sys_sched_yield() stats */ | |
480b9434 | 654 | unsigned int yld_count; |
1da177e4 LT |
655 | |
656 | /* schedule() stats */ | |
480b9434 KC |
657 | unsigned int sched_switch; |
658 | unsigned int sched_count; | |
659 | unsigned int sched_goidle; | |
1da177e4 LT |
660 | |
661 | /* try_to_wake_up() stats */ | |
480b9434 KC |
662 | unsigned int ttwu_count; |
663 | unsigned int ttwu_local; | |
b8efb561 IM |
664 | |
665 | /* BKL stats */ | |
480b9434 | 666 | unsigned int bkl_count; |
1da177e4 LT |
667 | #endif |
668 | }; | |
669 | ||
f34e3b61 | 670 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 671 | |
15afe09b | 672 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p, int sync) |
dd41f596 | 673 | { |
15afe09b | 674 | rq->curr->sched_class->check_preempt_curr(rq, p, sync); |
dd41f596 IM |
675 | } |
676 | ||
0a2966b4 CL |
677 | static inline int cpu_of(struct rq *rq) |
678 | { | |
679 | #ifdef CONFIG_SMP | |
680 | return rq->cpu; | |
681 | #else | |
682 | return 0; | |
683 | #endif | |
684 | } | |
685 | ||
674311d5 NP |
686 | /* |
687 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 688 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
689 | * |
690 | * The domain tree of any CPU may only be accessed from within | |
691 | * preempt-disabled sections. | |
692 | */ | |
48f24c4d IM |
693 | #define for_each_domain(cpu, __sd) \ |
694 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
695 | |
696 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
697 | #define this_rq() (&__get_cpu_var(runqueues)) | |
698 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
699 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 700 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 701 | |
aa9c4c0f | 702 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
703 | { |
704 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
705 | } | |
706 | ||
bf5c91ba IM |
707 | /* |
708 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
709 | */ | |
710 | #ifdef CONFIG_SCHED_DEBUG | |
711 | # define const_debug __read_mostly | |
712 | #else | |
713 | # define const_debug static const | |
714 | #endif | |
715 | ||
017730c1 IM |
716 | /** |
717 | * runqueue_is_locked | |
718 | * | |
719 | * Returns true if the current cpu runqueue is locked. | |
720 | * This interface allows printk to be called with the runqueue lock | |
721 | * held and know whether or not it is OK to wake up the klogd. | |
722 | */ | |
723 | int runqueue_is_locked(void) | |
724 | { | |
725 | int cpu = get_cpu(); | |
726 | struct rq *rq = cpu_rq(cpu); | |
727 | int ret; | |
728 | ||
729 | ret = spin_is_locked(&rq->lock); | |
730 | put_cpu(); | |
731 | return ret; | |
732 | } | |
733 | ||
bf5c91ba IM |
734 | /* |
735 | * Debugging: various feature bits | |
736 | */ | |
f00b45c1 PZ |
737 | |
738 | #define SCHED_FEAT(name, enabled) \ | |
739 | __SCHED_FEAT_##name , | |
740 | ||
bf5c91ba | 741 | enum { |
f00b45c1 | 742 | #include "sched_features.h" |
bf5c91ba IM |
743 | }; |
744 | ||
f00b45c1 PZ |
745 | #undef SCHED_FEAT |
746 | ||
747 | #define SCHED_FEAT(name, enabled) \ | |
748 | (1UL << __SCHED_FEAT_##name) * enabled | | |
749 | ||
bf5c91ba | 750 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
751 | #include "sched_features.h" |
752 | 0; | |
753 | ||
754 | #undef SCHED_FEAT | |
755 | ||
756 | #ifdef CONFIG_SCHED_DEBUG | |
757 | #define SCHED_FEAT(name, enabled) \ | |
758 | #name , | |
759 | ||
983ed7a6 | 760 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
761 | #include "sched_features.h" |
762 | NULL | |
763 | }; | |
764 | ||
765 | #undef SCHED_FEAT | |
766 | ||
34f3a814 | 767 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 768 | { |
f00b45c1 PZ |
769 | int i; |
770 | ||
771 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
772 | if (!(sysctl_sched_features & (1UL << i))) |
773 | seq_puts(m, "NO_"); | |
774 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 775 | } |
34f3a814 | 776 | seq_puts(m, "\n"); |
f00b45c1 | 777 | |
34f3a814 | 778 | return 0; |
f00b45c1 PZ |
779 | } |
780 | ||
781 | static ssize_t | |
782 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
783 | size_t cnt, loff_t *ppos) | |
784 | { | |
785 | char buf[64]; | |
786 | char *cmp = buf; | |
787 | int neg = 0; | |
788 | int i; | |
789 | ||
790 | if (cnt > 63) | |
791 | cnt = 63; | |
792 | ||
793 | if (copy_from_user(&buf, ubuf, cnt)) | |
794 | return -EFAULT; | |
795 | ||
796 | buf[cnt] = 0; | |
797 | ||
c24b7c52 | 798 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
799 | neg = 1; |
800 | cmp += 3; | |
801 | } | |
802 | ||
803 | for (i = 0; sched_feat_names[i]; i++) { | |
804 | int len = strlen(sched_feat_names[i]); | |
805 | ||
806 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
807 | if (neg) | |
808 | sysctl_sched_features &= ~(1UL << i); | |
809 | else | |
810 | sysctl_sched_features |= (1UL << i); | |
811 | break; | |
812 | } | |
813 | } | |
814 | ||
815 | if (!sched_feat_names[i]) | |
816 | return -EINVAL; | |
817 | ||
818 | filp->f_pos += cnt; | |
819 | ||
820 | return cnt; | |
821 | } | |
822 | ||
34f3a814 LZ |
823 | static int sched_feat_open(struct inode *inode, struct file *filp) |
824 | { | |
825 | return single_open(filp, sched_feat_show, NULL); | |
826 | } | |
827 | ||
f00b45c1 | 828 | static struct file_operations sched_feat_fops = { |
34f3a814 LZ |
829 | .open = sched_feat_open, |
830 | .write = sched_feat_write, | |
831 | .read = seq_read, | |
832 | .llseek = seq_lseek, | |
833 | .release = single_release, | |
f00b45c1 PZ |
834 | }; |
835 | ||
836 | static __init int sched_init_debug(void) | |
837 | { | |
f00b45c1 PZ |
838 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
839 | &sched_feat_fops); | |
840 | ||
841 | return 0; | |
842 | } | |
843 | late_initcall(sched_init_debug); | |
844 | ||
845 | #endif | |
846 | ||
847 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 848 | |
b82d9fdd PZ |
849 | /* |
850 | * Number of tasks to iterate in a single balance run. | |
851 | * Limited because this is done with IRQs disabled. | |
852 | */ | |
853 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
854 | ||
2398f2c6 PZ |
855 | /* |
856 | * ratelimit for updating the group shares. | |
55cd5340 | 857 | * default: 0.25ms |
2398f2c6 | 858 | */ |
55cd5340 | 859 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 860 | |
ffda12a1 PZ |
861 | /* |
862 | * Inject some fuzzyness into changing the per-cpu group shares | |
863 | * this avoids remote rq-locks at the expense of fairness. | |
864 | * default: 4 | |
865 | */ | |
866 | unsigned int sysctl_sched_shares_thresh = 4; | |
867 | ||
e9e9250b PZ |
868 | /* |
869 | * period over which we average the RT time consumption, measured | |
870 | * in ms. | |
871 | * | |
872 | * default: 1s | |
873 | */ | |
874 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
875 | ||
fa85ae24 | 876 | /* |
9f0c1e56 | 877 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
878 | * default: 1s |
879 | */ | |
9f0c1e56 | 880 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 881 | |
6892b75e IM |
882 | static __read_mostly int scheduler_running; |
883 | ||
9f0c1e56 PZ |
884 | /* |
885 | * part of the period that we allow rt tasks to run in us. | |
886 | * default: 0.95s | |
887 | */ | |
888 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 889 | |
d0b27fa7 PZ |
890 | static inline u64 global_rt_period(void) |
891 | { | |
892 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
893 | } | |
894 | ||
895 | static inline u64 global_rt_runtime(void) | |
896 | { | |
e26873bb | 897 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
898 | return RUNTIME_INF; |
899 | ||
900 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
901 | } | |
fa85ae24 | 902 | |
1da177e4 | 903 | #ifndef prepare_arch_switch |
4866cde0 NP |
904 | # define prepare_arch_switch(next) do { } while (0) |
905 | #endif | |
906 | #ifndef finish_arch_switch | |
907 | # define finish_arch_switch(prev) do { } while (0) | |
908 | #endif | |
909 | ||
051a1d1a DA |
910 | static inline int task_current(struct rq *rq, struct task_struct *p) |
911 | { | |
912 | return rq->curr == p; | |
913 | } | |
914 | ||
4866cde0 | 915 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 916 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 917 | { |
051a1d1a | 918 | return task_current(rq, p); |
4866cde0 NP |
919 | } |
920 | ||
70b97a7f | 921 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
922 | { |
923 | } | |
924 | ||
70b97a7f | 925 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 926 | { |
da04c035 IM |
927 | #ifdef CONFIG_DEBUG_SPINLOCK |
928 | /* this is a valid case when another task releases the spinlock */ | |
929 | rq->lock.owner = current; | |
930 | #endif | |
8a25d5de IM |
931 | /* |
932 | * If we are tracking spinlock dependencies then we have to | |
933 | * fix up the runqueue lock - which gets 'carried over' from | |
934 | * prev into current: | |
935 | */ | |
936 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
937 | ||
4866cde0 NP |
938 | spin_unlock_irq(&rq->lock); |
939 | } | |
940 | ||
941 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 942 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
943 | { |
944 | #ifdef CONFIG_SMP | |
945 | return p->oncpu; | |
946 | #else | |
051a1d1a | 947 | return task_current(rq, p); |
4866cde0 NP |
948 | #endif |
949 | } | |
950 | ||
70b97a7f | 951 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
952 | { |
953 | #ifdef CONFIG_SMP | |
954 | /* | |
955 | * We can optimise this out completely for !SMP, because the | |
956 | * SMP rebalancing from interrupt is the only thing that cares | |
957 | * here. | |
958 | */ | |
959 | next->oncpu = 1; | |
960 | #endif | |
961 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
962 | spin_unlock_irq(&rq->lock); | |
963 | #else | |
964 | spin_unlock(&rq->lock); | |
965 | #endif | |
966 | } | |
967 | ||
70b97a7f | 968 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
969 | { |
970 | #ifdef CONFIG_SMP | |
971 | /* | |
972 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
973 | * We must ensure this doesn't happen until the switch is completely | |
974 | * finished. | |
975 | */ | |
976 | smp_wmb(); | |
977 | prev->oncpu = 0; | |
978 | #endif | |
979 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
980 | local_irq_enable(); | |
1da177e4 | 981 | #endif |
4866cde0 NP |
982 | } |
983 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 984 | |
b29739f9 IM |
985 | /* |
986 | * __task_rq_lock - lock the runqueue a given task resides on. | |
987 | * Must be called interrupts disabled. | |
988 | */ | |
70b97a7f | 989 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
990 | __acquires(rq->lock) |
991 | { | |
3a5c359a AK |
992 | for (;;) { |
993 | struct rq *rq = task_rq(p); | |
994 | spin_lock(&rq->lock); | |
995 | if (likely(rq == task_rq(p))) | |
996 | return rq; | |
b29739f9 | 997 | spin_unlock(&rq->lock); |
b29739f9 | 998 | } |
b29739f9 IM |
999 | } |
1000 | ||
1da177e4 LT |
1001 | /* |
1002 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 1003 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
1004 | * explicitly disabling preemption. |
1005 | */ | |
70b97a7f | 1006 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
1007 | __acquires(rq->lock) |
1008 | { | |
70b97a7f | 1009 | struct rq *rq; |
1da177e4 | 1010 | |
3a5c359a AK |
1011 | for (;;) { |
1012 | local_irq_save(*flags); | |
1013 | rq = task_rq(p); | |
1014 | spin_lock(&rq->lock); | |
1015 | if (likely(rq == task_rq(p))) | |
1016 | return rq; | |
1da177e4 | 1017 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 1018 | } |
1da177e4 LT |
1019 | } |
1020 | ||
ad474cac ON |
1021 | void task_rq_unlock_wait(struct task_struct *p) |
1022 | { | |
1023 | struct rq *rq = task_rq(p); | |
1024 | ||
1025 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
1026 | spin_unlock_wait(&rq->lock); | |
1027 | } | |
1028 | ||
a9957449 | 1029 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
1030 | __releases(rq->lock) |
1031 | { | |
1032 | spin_unlock(&rq->lock); | |
1033 | } | |
1034 | ||
70b97a7f | 1035 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
1036 | __releases(rq->lock) |
1037 | { | |
1038 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1039 | } | |
1040 | ||
1da177e4 | 1041 | /* |
cc2a73b5 | 1042 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1043 | */ |
a9957449 | 1044 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1045 | __acquires(rq->lock) |
1046 | { | |
70b97a7f | 1047 | struct rq *rq; |
1da177e4 LT |
1048 | |
1049 | local_irq_disable(); | |
1050 | rq = this_rq(); | |
1051 | spin_lock(&rq->lock); | |
1052 | ||
1053 | return rq; | |
1054 | } | |
1055 | ||
8f4d37ec PZ |
1056 | #ifdef CONFIG_SCHED_HRTICK |
1057 | /* | |
1058 | * Use HR-timers to deliver accurate preemption points. | |
1059 | * | |
1060 | * Its all a bit involved since we cannot program an hrt while holding the | |
1061 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1062 | * reschedule event. | |
1063 | * | |
1064 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1065 | * rq->lock. | |
1066 | */ | |
8f4d37ec PZ |
1067 | |
1068 | /* | |
1069 | * Use hrtick when: | |
1070 | * - enabled by features | |
1071 | * - hrtimer is actually high res | |
1072 | */ | |
1073 | static inline int hrtick_enabled(struct rq *rq) | |
1074 | { | |
1075 | if (!sched_feat(HRTICK)) | |
1076 | return 0; | |
ba42059f | 1077 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1078 | return 0; |
8f4d37ec PZ |
1079 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1080 | } | |
1081 | ||
8f4d37ec PZ |
1082 | static void hrtick_clear(struct rq *rq) |
1083 | { | |
1084 | if (hrtimer_active(&rq->hrtick_timer)) | |
1085 | hrtimer_cancel(&rq->hrtick_timer); | |
1086 | } | |
1087 | ||
8f4d37ec PZ |
1088 | /* |
1089 | * High-resolution timer tick. | |
1090 | * Runs from hardirq context with interrupts disabled. | |
1091 | */ | |
1092 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1093 | { | |
1094 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1095 | ||
1096 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1097 | ||
1098 | spin_lock(&rq->lock); | |
3e51f33f | 1099 | update_rq_clock(rq); |
8f4d37ec PZ |
1100 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1101 | spin_unlock(&rq->lock); | |
1102 | ||
1103 | return HRTIMER_NORESTART; | |
1104 | } | |
1105 | ||
95e904c7 | 1106 | #ifdef CONFIG_SMP |
31656519 PZ |
1107 | /* |
1108 | * called from hardirq (IPI) context | |
1109 | */ | |
1110 | static void __hrtick_start(void *arg) | |
b328ca18 | 1111 | { |
31656519 | 1112 | struct rq *rq = arg; |
b328ca18 | 1113 | |
31656519 PZ |
1114 | spin_lock(&rq->lock); |
1115 | hrtimer_restart(&rq->hrtick_timer); | |
1116 | rq->hrtick_csd_pending = 0; | |
1117 | spin_unlock(&rq->lock); | |
b328ca18 PZ |
1118 | } |
1119 | ||
31656519 PZ |
1120 | /* |
1121 | * Called to set the hrtick timer state. | |
1122 | * | |
1123 | * called with rq->lock held and irqs disabled | |
1124 | */ | |
1125 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1126 | { |
31656519 PZ |
1127 | struct hrtimer *timer = &rq->hrtick_timer; |
1128 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1129 | |
cc584b21 | 1130 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1131 | |
1132 | if (rq == this_rq()) { | |
1133 | hrtimer_restart(timer); | |
1134 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1135 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1136 | rq->hrtick_csd_pending = 1; |
1137 | } | |
b328ca18 PZ |
1138 | } |
1139 | ||
1140 | static int | |
1141 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1142 | { | |
1143 | int cpu = (int)(long)hcpu; | |
1144 | ||
1145 | switch (action) { | |
1146 | case CPU_UP_CANCELED: | |
1147 | case CPU_UP_CANCELED_FROZEN: | |
1148 | case CPU_DOWN_PREPARE: | |
1149 | case CPU_DOWN_PREPARE_FROZEN: | |
1150 | case CPU_DEAD: | |
1151 | case CPU_DEAD_FROZEN: | |
31656519 | 1152 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1153 | return NOTIFY_OK; |
1154 | } | |
1155 | ||
1156 | return NOTIFY_DONE; | |
1157 | } | |
1158 | ||
fa748203 | 1159 | static __init void init_hrtick(void) |
b328ca18 PZ |
1160 | { |
1161 | hotcpu_notifier(hotplug_hrtick, 0); | |
1162 | } | |
31656519 PZ |
1163 | #else |
1164 | /* | |
1165 | * Called to set the hrtick timer state. | |
1166 | * | |
1167 | * called with rq->lock held and irqs disabled | |
1168 | */ | |
1169 | static void hrtick_start(struct rq *rq, u64 delay) | |
1170 | { | |
7f1e2ca9 | 1171 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1172 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1173 | } |
b328ca18 | 1174 | |
006c75f1 | 1175 | static inline void init_hrtick(void) |
8f4d37ec | 1176 | { |
8f4d37ec | 1177 | } |
31656519 | 1178 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1179 | |
31656519 | 1180 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1181 | { |
31656519 PZ |
1182 | #ifdef CONFIG_SMP |
1183 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1184 | |
31656519 PZ |
1185 | rq->hrtick_csd.flags = 0; |
1186 | rq->hrtick_csd.func = __hrtick_start; | |
1187 | rq->hrtick_csd.info = rq; | |
1188 | #endif | |
8f4d37ec | 1189 | |
31656519 PZ |
1190 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1191 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1192 | } |
006c75f1 | 1193 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1194 | static inline void hrtick_clear(struct rq *rq) |
1195 | { | |
1196 | } | |
1197 | ||
8f4d37ec PZ |
1198 | static inline void init_rq_hrtick(struct rq *rq) |
1199 | { | |
1200 | } | |
1201 | ||
b328ca18 PZ |
1202 | static inline void init_hrtick(void) |
1203 | { | |
1204 | } | |
006c75f1 | 1205 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1206 | |
c24d20db IM |
1207 | /* |
1208 | * resched_task - mark a task 'to be rescheduled now'. | |
1209 | * | |
1210 | * On UP this means the setting of the need_resched flag, on SMP it | |
1211 | * might also involve a cross-CPU call to trigger the scheduler on | |
1212 | * the target CPU. | |
1213 | */ | |
1214 | #ifdef CONFIG_SMP | |
1215 | ||
1216 | #ifndef tsk_is_polling | |
1217 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1218 | #endif | |
1219 | ||
31656519 | 1220 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1221 | { |
1222 | int cpu; | |
1223 | ||
1224 | assert_spin_locked(&task_rq(p)->lock); | |
1225 | ||
5ed0cec0 | 1226 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1227 | return; |
1228 | ||
5ed0cec0 | 1229 | set_tsk_need_resched(p); |
c24d20db IM |
1230 | |
1231 | cpu = task_cpu(p); | |
1232 | if (cpu == smp_processor_id()) | |
1233 | return; | |
1234 | ||
1235 | /* NEED_RESCHED must be visible before we test polling */ | |
1236 | smp_mb(); | |
1237 | if (!tsk_is_polling(p)) | |
1238 | smp_send_reschedule(cpu); | |
1239 | } | |
1240 | ||
1241 | static void resched_cpu(int cpu) | |
1242 | { | |
1243 | struct rq *rq = cpu_rq(cpu); | |
1244 | unsigned long flags; | |
1245 | ||
1246 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1247 | return; | |
1248 | resched_task(cpu_curr(cpu)); | |
1249 | spin_unlock_irqrestore(&rq->lock, flags); | |
1250 | } | |
06d8308c TG |
1251 | |
1252 | #ifdef CONFIG_NO_HZ | |
1253 | /* | |
1254 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1255 | * idle CPU then this timer might expire before the next timer event | |
1256 | * which is scheduled to wake up that CPU. In case of a completely | |
1257 | * idle system the next event might even be infinite time into the | |
1258 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1259 | * leaves the inner idle loop so the newly added timer is taken into | |
1260 | * account when the CPU goes back to idle and evaluates the timer | |
1261 | * wheel for the next timer event. | |
1262 | */ | |
1263 | void wake_up_idle_cpu(int cpu) | |
1264 | { | |
1265 | struct rq *rq = cpu_rq(cpu); | |
1266 | ||
1267 | if (cpu == smp_processor_id()) | |
1268 | return; | |
1269 | ||
1270 | /* | |
1271 | * This is safe, as this function is called with the timer | |
1272 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1273 | * to idle and has not yet set rq->curr to idle then it will | |
1274 | * be serialized on the timer wheel base lock and take the new | |
1275 | * timer into account automatically. | |
1276 | */ | |
1277 | if (rq->curr != rq->idle) | |
1278 | return; | |
1279 | ||
1280 | /* | |
1281 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1282 | * lockless. The worst case is that the other CPU runs the | |
1283 | * idle task through an additional NOOP schedule() | |
1284 | */ | |
5ed0cec0 | 1285 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1286 | |
1287 | /* NEED_RESCHED must be visible before we test polling */ | |
1288 | smp_mb(); | |
1289 | if (!tsk_is_polling(rq->idle)) | |
1290 | smp_send_reschedule(cpu); | |
1291 | } | |
6d6bc0ad | 1292 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1293 | |
e9e9250b PZ |
1294 | static u64 sched_avg_period(void) |
1295 | { | |
1296 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1297 | } | |
1298 | ||
1299 | static void sched_avg_update(struct rq *rq) | |
1300 | { | |
1301 | s64 period = sched_avg_period(); | |
1302 | ||
1303 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1304 | rq->age_stamp += period; | |
1305 | rq->rt_avg /= 2; | |
1306 | } | |
1307 | } | |
1308 | ||
1309 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1310 | { | |
1311 | rq->rt_avg += rt_delta; | |
1312 | sched_avg_update(rq); | |
1313 | } | |
1314 | ||
6d6bc0ad | 1315 | #else /* !CONFIG_SMP */ |
31656519 | 1316 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1317 | { |
1318 | assert_spin_locked(&task_rq(p)->lock); | |
31656519 | 1319 | set_tsk_need_resched(p); |
c24d20db | 1320 | } |
e9e9250b PZ |
1321 | |
1322 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1323 | { | |
1324 | } | |
6d6bc0ad | 1325 | #endif /* CONFIG_SMP */ |
c24d20db | 1326 | |
45bf76df IM |
1327 | #if BITS_PER_LONG == 32 |
1328 | # define WMULT_CONST (~0UL) | |
1329 | #else | |
1330 | # define WMULT_CONST (1UL << 32) | |
1331 | #endif | |
1332 | ||
1333 | #define WMULT_SHIFT 32 | |
1334 | ||
194081eb IM |
1335 | /* |
1336 | * Shift right and round: | |
1337 | */ | |
cf2ab469 | 1338 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1339 | |
a7be37ac PZ |
1340 | /* |
1341 | * delta *= weight / lw | |
1342 | */ | |
cb1c4fc9 | 1343 | static unsigned long |
45bf76df IM |
1344 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1345 | struct load_weight *lw) | |
1346 | { | |
1347 | u64 tmp; | |
1348 | ||
7a232e03 LJ |
1349 | if (!lw->inv_weight) { |
1350 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1351 | lw->inv_weight = 1; | |
1352 | else | |
1353 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1354 | / (lw->weight+1); | |
1355 | } | |
45bf76df IM |
1356 | |
1357 | tmp = (u64)delta_exec * weight; | |
1358 | /* | |
1359 | * Check whether we'd overflow the 64-bit multiplication: | |
1360 | */ | |
194081eb | 1361 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1362 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1363 | WMULT_SHIFT/2); |
1364 | else | |
cf2ab469 | 1365 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1366 | |
ecf691da | 1367 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1368 | } |
1369 | ||
1091985b | 1370 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1371 | { |
1372 | lw->weight += inc; | |
e89996ae | 1373 | lw->inv_weight = 0; |
45bf76df IM |
1374 | } |
1375 | ||
1091985b | 1376 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1377 | { |
1378 | lw->weight -= dec; | |
e89996ae | 1379 | lw->inv_weight = 0; |
45bf76df IM |
1380 | } |
1381 | ||
2dd73a4f PW |
1382 | /* |
1383 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1384 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1385 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1386 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1387 | * scaled version of the new time slice allocation that they receive on time |
1388 | * slice expiry etc. | |
1389 | */ | |
1390 | ||
cce7ade8 PZ |
1391 | #define WEIGHT_IDLEPRIO 3 |
1392 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1393 | |
1394 | /* | |
1395 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1396 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1397 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1398 | * that remained on nice 0. | |
1399 | * | |
1400 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1401 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1402 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1403 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1404 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1405 | */ |
1406 | static const int prio_to_weight[40] = { | |
254753dc IM |
1407 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1408 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1409 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1410 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1411 | /* 0 */ 1024, 820, 655, 526, 423, | |
1412 | /* 5 */ 335, 272, 215, 172, 137, | |
1413 | /* 10 */ 110, 87, 70, 56, 45, | |
1414 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1415 | }; |
1416 | ||
5714d2de IM |
1417 | /* |
1418 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1419 | * | |
1420 | * In cases where the weight does not change often, we can use the | |
1421 | * precalculated inverse to speed up arithmetics by turning divisions | |
1422 | * into multiplications: | |
1423 | */ | |
dd41f596 | 1424 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1425 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1426 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1427 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1428 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1429 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1430 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1431 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1432 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1433 | }; |
2dd73a4f | 1434 | |
dd41f596 IM |
1435 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1436 | ||
1437 | /* | |
1438 | * runqueue iterator, to support SMP load-balancing between different | |
1439 | * scheduling classes, without having to expose their internal data | |
1440 | * structures to the load-balancing proper: | |
1441 | */ | |
1442 | struct rq_iterator { | |
1443 | void *arg; | |
1444 | struct task_struct *(*start)(void *); | |
1445 | struct task_struct *(*next)(void *); | |
1446 | }; | |
1447 | ||
e1d1484f PW |
1448 | #ifdef CONFIG_SMP |
1449 | static unsigned long | |
1450 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1451 | unsigned long max_load_move, struct sched_domain *sd, | |
1452 | enum cpu_idle_type idle, int *all_pinned, | |
1453 | int *this_best_prio, struct rq_iterator *iterator); | |
1454 | ||
1455 | static int | |
1456 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1457 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1458 | struct rq_iterator *iterator); | |
e1d1484f | 1459 | #endif |
dd41f596 | 1460 | |
ef12fefa BR |
1461 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1462 | enum cpuacct_stat_index { | |
1463 | CPUACCT_STAT_USER, /* ... user mode */ | |
1464 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1465 | ||
1466 | CPUACCT_STAT_NSTATS, | |
1467 | }; | |
1468 | ||
d842de87 SV |
1469 | #ifdef CONFIG_CGROUP_CPUACCT |
1470 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1471 | static void cpuacct_update_stats(struct task_struct *tsk, |
1472 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1473 | #else |
1474 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1475 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1476 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1477 | #endif |
1478 | ||
18d95a28 PZ |
1479 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1480 | { | |
1481 | update_load_add(&rq->load, load); | |
1482 | } | |
1483 | ||
1484 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1485 | { | |
1486 | update_load_sub(&rq->load, load); | |
1487 | } | |
1488 | ||
7940ca36 | 1489 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1490 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1491 | |
1492 | /* | |
1493 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1494 | * leaving it for the final time. | |
1495 | */ | |
eb755805 | 1496 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1497 | { |
1498 | struct task_group *parent, *child; | |
eb755805 | 1499 | int ret; |
c09595f6 PZ |
1500 | |
1501 | rcu_read_lock(); | |
1502 | parent = &root_task_group; | |
1503 | down: | |
eb755805 PZ |
1504 | ret = (*down)(parent, data); |
1505 | if (ret) | |
1506 | goto out_unlock; | |
c09595f6 PZ |
1507 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1508 | parent = child; | |
1509 | goto down; | |
1510 | ||
1511 | up: | |
1512 | continue; | |
1513 | } | |
eb755805 PZ |
1514 | ret = (*up)(parent, data); |
1515 | if (ret) | |
1516 | goto out_unlock; | |
c09595f6 PZ |
1517 | |
1518 | child = parent; | |
1519 | parent = parent->parent; | |
1520 | if (parent) | |
1521 | goto up; | |
eb755805 | 1522 | out_unlock: |
c09595f6 | 1523 | rcu_read_unlock(); |
eb755805 PZ |
1524 | |
1525 | return ret; | |
c09595f6 PZ |
1526 | } |
1527 | ||
eb755805 PZ |
1528 | static int tg_nop(struct task_group *tg, void *data) |
1529 | { | |
1530 | return 0; | |
c09595f6 | 1531 | } |
eb755805 PZ |
1532 | #endif |
1533 | ||
1534 | #ifdef CONFIG_SMP | |
1535 | static unsigned long source_load(int cpu, int type); | |
1536 | static unsigned long target_load(int cpu, int type); | |
1537 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
1538 | ||
1539 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1540 | { | |
1541 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1542 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1543 | |
4cd42620 SR |
1544 | if (nr_running) |
1545 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1546 | else |
1547 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1548 | |
1549 | return rq->avg_load_per_task; | |
1550 | } | |
1551 | ||
1552 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1553 | |
34d76c41 PZ |
1554 | struct update_shares_data { |
1555 | unsigned long rq_weight[NR_CPUS]; | |
1556 | }; | |
1557 | ||
1558 | static DEFINE_PER_CPU(struct update_shares_data, update_shares_data); | |
1559 | ||
c09595f6 PZ |
1560 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1561 | ||
1562 | /* | |
1563 | * Calculate and set the cpu's group shares. | |
1564 | */ | |
34d76c41 PZ |
1565 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1566 | unsigned long sd_shares, | |
1567 | unsigned long sd_rq_weight, | |
1568 | struct update_shares_data *usd) | |
18d95a28 | 1569 | { |
34d76c41 | 1570 | unsigned long shares, rq_weight; |
a5004278 | 1571 | int boost = 0; |
c09595f6 | 1572 | |
34d76c41 | 1573 | rq_weight = usd->rq_weight[cpu]; |
a5004278 PZ |
1574 | if (!rq_weight) { |
1575 | boost = 1; | |
1576 | rq_weight = NICE_0_LOAD; | |
1577 | } | |
c8cba857 | 1578 | |
c09595f6 | 1579 | /* |
a8af7246 PZ |
1580 | * \Sum_j shares_j * rq_weight_i |
1581 | * shares_i = ----------------------------- | |
1582 | * \Sum_j rq_weight_j | |
c09595f6 | 1583 | */ |
ec4e0e2f | 1584 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1585 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1586 | |
ffda12a1 PZ |
1587 | if (abs(shares - tg->se[cpu]->load.weight) > |
1588 | sysctl_sched_shares_thresh) { | |
1589 | struct rq *rq = cpu_rq(cpu); | |
1590 | unsigned long flags; | |
c09595f6 | 1591 | |
ffda12a1 | 1592 | spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1593 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1594 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 PZ |
1595 | __set_se_shares(tg->se[cpu], shares); |
1596 | spin_unlock_irqrestore(&rq->lock, flags); | |
1597 | } | |
18d95a28 | 1598 | } |
c09595f6 PZ |
1599 | |
1600 | /* | |
c8cba857 PZ |
1601 | * Re-compute the task group their per cpu shares over the given domain. |
1602 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1603 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1604 | */ |
eb755805 | 1605 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1606 | { |
34d76c41 PZ |
1607 | unsigned long weight, rq_weight = 0, shares = 0; |
1608 | struct update_shares_data *usd; | |
eb755805 | 1609 | struct sched_domain *sd = data; |
34d76c41 | 1610 | unsigned long flags; |
c8cba857 | 1611 | int i; |
c09595f6 | 1612 | |
34d76c41 PZ |
1613 | if (!tg->se[0]) |
1614 | return 0; | |
1615 | ||
1616 | local_irq_save(flags); | |
1617 | usd = &__get_cpu_var(update_shares_data); | |
1618 | ||
758b2cdc | 1619 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 PZ |
1620 | weight = tg->cfs_rq[i]->load.weight; |
1621 | usd->rq_weight[i] = weight; | |
1622 | ||
ec4e0e2f KC |
1623 | /* |
1624 | * If there are currently no tasks on the cpu pretend there | |
1625 | * is one of average load so that when a new task gets to | |
1626 | * run here it will not get delayed by group starvation. | |
1627 | */ | |
ec4e0e2f KC |
1628 | if (!weight) |
1629 | weight = NICE_0_LOAD; | |
1630 | ||
34d76c41 | 1631 | rq_weight += weight; |
c8cba857 | 1632 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1633 | } |
c09595f6 | 1634 | |
c8cba857 PZ |
1635 | if ((!shares && rq_weight) || shares > tg->shares) |
1636 | shares = tg->shares; | |
1637 | ||
1638 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1639 | shares = tg->shares; | |
c09595f6 | 1640 | |
a8af7246 | 1641 | for_each_cpu(i, sched_domain_span(sd)) |
34d76c41 PZ |
1642 | update_group_shares_cpu(tg, i, shares, rq_weight, usd); |
1643 | ||
1644 | local_irq_restore(flags); | |
eb755805 PZ |
1645 | |
1646 | return 0; | |
c09595f6 PZ |
1647 | } |
1648 | ||
1649 | /* | |
c8cba857 PZ |
1650 | * Compute the cpu's hierarchical load factor for each task group. |
1651 | * This needs to be done in a top-down fashion because the load of a child | |
1652 | * group is a fraction of its parents load. | |
c09595f6 | 1653 | */ |
eb755805 | 1654 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1655 | { |
c8cba857 | 1656 | unsigned long load; |
eb755805 | 1657 | long cpu = (long)data; |
c09595f6 | 1658 | |
c8cba857 PZ |
1659 | if (!tg->parent) { |
1660 | load = cpu_rq(cpu)->load.weight; | |
1661 | } else { | |
1662 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1663 | load *= tg->cfs_rq[cpu]->shares; | |
1664 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1665 | } | |
c09595f6 | 1666 | |
c8cba857 | 1667 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1668 | |
eb755805 | 1669 | return 0; |
c09595f6 PZ |
1670 | } |
1671 | ||
c8cba857 | 1672 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1673 | { |
e7097159 PZ |
1674 | s64 elapsed; |
1675 | u64 now; | |
1676 | ||
1677 | if (root_task_group_empty()) | |
1678 | return; | |
1679 | ||
1680 | now = cpu_clock(raw_smp_processor_id()); | |
1681 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1682 | |
1683 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1684 | sd->last_update = now; | |
eb755805 | 1685 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1686 | } |
4d8d595d PZ |
1687 | } |
1688 | ||
3e5459b4 PZ |
1689 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1690 | { | |
e7097159 PZ |
1691 | if (root_task_group_empty()) |
1692 | return; | |
1693 | ||
3e5459b4 PZ |
1694 | spin_unlock(&rq->lock); |
1695 | update_shares(sd); | |
1696 | spin_lock(&rq->lock); | |
1697 | } | |
1698 | ||
eb755805 | 1699 | static void update_h_load(long cpu) |
c09595f6 | 1700 | { |
e7097159 PZ |
1701 | if (root_task_group_empty()) |
1702 | return; | |
1703 | ||
eb755805 | 1704 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1705 | } |
1706 | ||
c09595f6 PZ |
1707 | #else |
1708 | ||
c8cba857 | 1709 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1710 | { |
1711 | } | |
1712 | ||
3e5459b4 PZ |
1713 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1714 | { | |
1715 | } | |
1716 | ||
18d95a28 PZ |
1717 | #endif |
1718 | ||
8f45e2b5 GH |
1719 | #ifdef CONFIG_PREEMPT |
1720 | ||
70574a99 | 1721 | /* |
8f45e2b5 GH |
1722 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1723 | * way at the expense of forcing extra atomic operations in all | |
1724 | * invocations. This assures that the double_lock is acquired using the | |
1725 | * same underlying policy as the spinlock_t on this architecture, which | |
1726 | * reduces latency compared to the unfair variant below. However, it | |
1727 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1728 | */ |
8f45e2b5 GH |
1729 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1730 | __releases(this_rq->lock) | |
1731 | __acquires(busiest->lock) | |
1732 | __acquires(this_rq->lock) | |
1733 | { | |
1734 | spin_unlock(&this_rq->lock); | |
1735 | double_rq_lock(this_rq, busiest); | |
1736 | ||
1737 | return 1; | |
1738 | } | |
1739 | ||
1740 | #else | |
1741 | /* | |
1742 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1743 | * latency by eliminating extra atomic operations when the locks are | |
1744 | * already in proper order on entry. This favors lower cpu-ids and will | |
1745 | * grant the double lock to lower cpus over higher ids under contention, | |
1746 | * regardless of entry order into the function. | |
1747 | */ | |
1748 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1749 | __releases(this_rq->lock) |
1750 | __acquires(busiest->lock) | |
1751 | __acquires(this_rq->lock) | |
1752 | { | |
1753 | int ret = 0; | |
1754 | ||
70574a99 AD |
1755 | if (unlikely(!spin_trylock(&busiest->lock))) { |
1756 | if (busiest < this_rq) { | |
1757 | spin_unlock(&this_rq->lock); | |
1758 | spin_lock(&busiest->lock); | |
1759 | spin_lock_nested(&this_rq->lock, SINGLE_DEPTH_NESTING); | |
1760 | ret = 1; | |
1761 | } else | |
1762 | spin_lock_nested(&busiest->lock, SINGLE_DEPTH_NESTING); | |
1763 | } | |
1764 | return ret; | |
1765 | } | |
1766 | ||
8f45e2b5 GH |
1767 | #endif /* CONFIG_PREEMPT */ |
1768 | ||
1769 | /* | |
1770 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1771 | */ | |
1772 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1773 | { | |
1774 | if (unlikely(!irqs_disabled())) { | |
1775 | /* printk() doesn't work good under rq->lock */ | |
1776 | spin_unlock(&this_rq->lock); | |
1777 | BUG_ON(1); | |
1778 | } | |
1779 | ||
1780 | return _double_lock_balance(this_rq, busiest); | |
1781 | } | |
1782 | ||
70574a99 AD |
1783 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1784 | __releases(busiest->lock) | |
1785 | { | |
1786 | spin_unlock(&busiest->lock); | |
1787 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); | |
1788 | } | |
18d95a28 PZ |
1789 | #endif |
1790 | ||
30432094 | 1791 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1792 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1793 | { | |
30432094 | 1794 | #ifdef CONFIG_SMP |
34e83e85 IM |
1795 | cfs_rq->shares = shares; |
1796 | #endif | |
1797 | } | |
30432094 | 1798 | #endif |
e7693a36 | 1799 | |
dce48a84 TG |
1800 | static void calc_load_account_active(struct rq *this_rq); |
1801 | ||
dd41f596 | 1802 | #include "sched_stats.h" |
dd41f596 | 1803 | #include "sched_idletask.c" |
5522d5d5 IM |
1804 | #include "sched_fair.c" |
1805 | #include "sched_rt.c" | |
dd41f596 IM |
1806 | #ifdef CONFIG_SCHED_DEBUG |
1807 | # include "sched_debug.c" | |
1808 | #endif | |
1809 | ||
1810 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1811 | #define for_each_class(class) \ |
1812 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1813 | |
c09595f6 | 1814 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1815 | { |
1816 | rq->nr_running++; | |
9c217245 IM |
1817 | } |
1818 | ||
c09595f6 | 1819 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1820 | { |
1821 | rq->nr_running--; | |
9c217245 IM |
1822 | } |
1823 | ||
45bf76df IM |
1824 | static void set_load_weight(struct task_struct *p) |
1825 | { | |
1826 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1827 | p->se.load.weight = prio_to_weight[0] * 2; |
1828 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1829 | return; | |
1830 | } | |
45bf76df | 1831 | |
dd41f596 IM |
1832 | /* |
1833 | * SCHED_IDLE tasks get minimal weight: | |
1834 | */ | |
1835 | if (p->policy == SCHED_IDLE) { | |
1836 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1837 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1838 | return; | |
1839 | } | |
71f8bd46 | 1840 | |
dd41f596 IM |
1841 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1842 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1843 | } |
1844 | ||
2087a1ad GH |
1845 | static void update_avg(u64 *avg, u64 sample) |
1846 | { | |
1847 | s64 diff = sample - *avg; | |
1848 | *avg += diff >> 3; | |
1849 | } | |
1850 | ||
8159f87e | 1851 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1852 | { |
831451ac PZ |
1853 | if (wakeup) |
1854 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1855 | ||
dd41f596 | 1856 | sched_info_queued(p); |
fd390f6a | 1857 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1858 | p->se.on_rq = 1; |
71f8bd46 IM |
1859 | } |
1860 | ||
69be72c1 | 1861 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1862 | { |
831451ac PZ |
1863 | if (sleep) { |
1864 | if (p->se.last_wakeup) { | |
1865 | update_avg(&p->se.avg_overlap, | |
1866 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1867 | p->se.last_wakeup = 0; | |
1868 | } else { | |
1869 | update_avg(&p->se.avg_wakeup, | |
1870 | sysctl_sched_wakeup_granularity); | |
1871 | } | |
2087a1ad GH |
1872 | } |
1873 | ||
46ac22ba | 1874 | sched_info_dequeued(p); |
f02231e5 | 1875 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1876 | p->se.on_rq = 0; |
71f8bd46 IM |
1877 | } |
1878 | ||
14531189 | 1879 | /* |
dd41f596 | 1880 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1881 | */ |
14531189 IM |
1882 | static inline int __normal_prio(struct task_struct *p) |
1883 | { | |
dd41f596 | 1884 | return p->static_prio; |
14531189 IM |
1885 | } |
1886 | ||
b29739f9 IM |
1887 | /* |
1888 | * Calculate the expected normal priority: i.e. priority | |
1889 | * without taking RT-inheritance into account. Might be | |
1890 | * boosted by interactivity modifiers. Changes upon fork, | |
1891 | * setprio syscalls, and whenever the interactivity | |
1892 | * estimator recalculates. | |
1893 | */ | |
36c8b586 | 1894 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1895 | { |
1896 | int prio; | |
1897 | ||
e05606d3 | 1898 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1899 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1900 | else | |
1901 | prio = __normal_prio(p); | |
1902 | return prio; | |
1903 | } | |
1904 | ||
1905 | /* | |
1906 | * Calculate the current priority, i.e. the priority | |
1907 | * taken into account by the scheduler. This value might | |
1908 | * be boosted by RT tasks, or might be boosted by | |
1909 | * interactivity modifiers. Will be RT if the task got | |
1910 | * RT-boosted. If not then it returns p->normal_prio. | |
1911 | */ | |
36c8b586 | 1912 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1913 | { |
1914 | p->normal_prio = normal_prio(p); | |
1915 | /* | |
1916 | * If we are RT tasks or we were boosted to RT priority, | |
1917 | * keep the priority unchanged. Otherwise, update priority | |
1918 | * to the normal priority: | |
1919 | */ | |
1920 | if (!rt_prio(p->prio)) | |
1921 | return p->normal_prio; | |
1922 | return p->prio; | |
1923 | } | |
1924 | ||
1da177e4 | 1925 | /* |
dd41f596 | 1926 | * activate_task - move a task to the runqueue. |
1da177e4 | 1927 | */ |
dd41f596 | 1928 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1929 | { |
d9514f6c | 1930 | if (task_contributes_to_load(p)) |
dd41f596 | 1931 | rq->nr_uninterruptible--; |
1da177e4 | 1932 | |
8159f87e | 1933 | enqueue_task(rq, p, wakeup); |
c09595f6 | 1934 | inc_nr_running(rq); |
1da177e4 LT |
1935 | } |
1936 | ||
1da177e4 LT |
1937 | /* |
1938 | * deactivate_task - remove a task from the runqueue. | |
1939 | */ | |
2e1cb74a | 1940 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1941 | { |
d9514f6c | 1942 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1943 | rq->nr_uninterruptible++; |
1944 | ||
69be72c1 | 1945 | dequeue_task(rq, p, sleep); |
c09595f6 | 1946 | dec_nr_running(rq); |
1da177e4 LT |
1947 | } |
1948 | ||
1da177e4 LT |
1949 | /** |
1950 | * task_curr - is this task currently executing on a CPU? | |
1951 | * @p: the task in question. | |
1952 | */ | |
36c8b586 | 1953 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1954 | { |
1955 | return cpu_curr(task_cpu(p)) == p; | |
1956 | } | |
1957 | ||
dd41f596 IM |
1958 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1959 | { | |
6f505b16 | 1960 | set_task_rq(p, cpu); |
dd41f596 | 1961 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1962 | /* |
1963 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1964 | * successfuly executed on another CPU. We must ensure that updates of | |
1965 | * per-task data have been completed by this moment. | |
1966 | */ | |
1967 | smp_wmb(); | |
dd41f596 | 1968 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1969 | #endif |
2dd73a4f PW |
1970 | } |
1971 | ||
cb469845 SR |
1972 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1973 | const struct sched_class *prev_class, | |
1974 | int oldprio, int running) | |
1975 | { | |
1976 | if (prev_class != p->sched_class) { | |
1977 | if (prev_class->switched_from) | |
1978 | prev_class->switched_from(rq, p, running); | |
1979 | p->sched_class->switched_to(rq, p, running); | |
1980 | } else | |
1981 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1982 | } | |
1983 | ||
1da177e4 | 1984 | #ifdef CONFIG_SMP |
c65cc870 | 1985 | |
e958b360 TG |
1986 | /* Used instead of source_load when we know the type == 0 */ |
1987 | static unsigned long weighted_cpuload(const int cpu) | |
1988 | { | |
1989 | return cpu_rq(cpu)->load.weight; | |
1990 | } | |
1991 | ||
cc367732 IM |
1992 | /* |
1993 | * Is this task likely cache-hot: | |
1994 | */ | |
e7693a36 | 1995 | static int |
cc367732 IM |
1996 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1997 | { | |
1998 | s64 delta; | |
1999 | ||
f540a608 IM |
2000 | /* |
2001 | * Buddy candidates are cache hot: | |
2002 | */ | |
4793241b PZ |
2003 | if (sched_feat(CACHE_HOT_BUDDY) && |
2004 | (&p->se == cfs_rq_of(&p->se)->next || | |
2005 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2006 | return 1; |
2007 | ||
cc367732 IM |
2008 | if (p->sched_class != &fair_sched_class) |
2009 | return 0; | |
2010 | ||
6bc1665b IM |
2011 | if (sysctl_sched_migration_cost == -1) |
2012 | return 1; | |
2013 | if (sysctl_sched_migration_cost == 0) | |
2014 | return 0; | |
2015 | ||
cc367732 IM |
2016 | delta = now - p->se.exec_start; |
2017 | ||
2018 | return delta < (s64)sysctl_sched_migration_cost; | |
2019 | } | |
2020 | ||
2021 | ||
dd41f596 | 2022 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2023 | { |
dd41f596 IM |
2024 | int old_cpu = task_cpu(p); |
2025 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
2026 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
2027 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 2028 | u64 clock_offset; |
dd41f596 IM |
2029 | |
2030 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d | 2031 | |
de1d7286 | 2032 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2033 | |
6cfb0d5d IM |
2034 | #ifdef CONFIG_SCHEDSTATS |
2035 | if (p->se.wait_start) | |
2036 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
2037 | if (p->se.sleep_start) |
2038 | p->se.sleep_start -= clock_offset; | |
2039 | if (p->se.block_start) | |
2040 | p->se.block_start -= clock_offset; | |
6c594c21 | 2041 | #endif |
cc367732 | 2042 | if (old_cpu != new_cpu) { |
6c594c21 | 2043 | p->se.nr_migrations++; |
23a185ca | 2044 | new_rq->nr_migrations_in++; |
6c594c21 | 2045 | #ifdef CONFIG_SCHEDSTATS |
cc367732 IM |
2046 | if (task_hot(p, old_rq->clock, NULL)) |
2047 | schedstat_inc(p, se.nr_forced2_migrations); | |
6cfb0d5d | 2048 | #endif |
e5289d4a PZ |
2049 | perf_swcounter_event(PERF_COUNT_SW_CPU_MIGRATIONS, |
2050 | 1, 1, NULL, 0); | |
6c594c21 | 2051 | } |
2830cf8c SV |
2052 | p->se.vruntime -= old_cfsrq->min_vruntime - |
2053 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
2054 | |
2055 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2056 | } |
2057 | ||
70b97a7f | 2058 | struct migration_req { |
1da177e4 | 2059 | struct list_head list; |
1da177e4 | 2060 | |
36c8b586 | 2061 | struct task_struct *task; |
1da177e4 LT |
2062 | int dest_cpu; |
2063 | ||
1da177e4 | 2064 | struct completion done; |
70b97a7f | 2065 | }; |
1da177e4 LT |
2066 | |
2067 | /* | |
2068 | * The task's runqueue lock must be held. | |
2069 | * Returns true if you have to wait for migration thread. | |
2070 | */ | |
36c8b586 | 2071 | static int |
70b97a7f | 2072 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2073 | { |
70b97a7f | 2074 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2075 | |
2076 | /* | |
2077 | * If the task is not on a runqueue (and not running), then | |
2078 | * it is sufficient to simply update the task's cpu field. | |
2079 | */ | |
dd41f596 | 2080 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
2081 | set_task_cpu(p, dest_cpu); |
2082 | return 0; | |
2083 | } | |
2084 | ||
2085 | init_completion(&req->done); | |
1da177e4 LT |
2086 | req->task = p; |
2087 | req->dest_cpu = dest_cpu; | |
2088 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2089 | |
1da177e4 LT |
2090 | return 1; |
2091 | } | |
2092 | ||
a26b89f0 MM |
2093 | /* |
2094 | * wait_task_context_switch - wait for a thread to complete at least one | |
2095 | * context switch. | |
2096 | * | |
2097 | * @p must not be current. | |
2098 | */ | |
2099 | void wait_task_context_switch(struct task_struct *p) | |
2100 | { | |
2101 | unsigned long nvcsw, nivcsw, flags; | |
2102 | int running; | |
2103 | struct rq *rq; | |
2104 | ||
2105 | nvcsw = p->nvcsw; | |
2106 | nivcsw = p->nivcsw; | |
2107 | for (;;) { | |
2108 | /* | |
2109 | * The runqueue is assigned before the actual context | |
2110 | * switch. We need to take the runqueue lock. | |
2111 | * | |
2112 | * We could check initially without the lock but it is | |
2113 | * very likely that we need to take the lock in every | |
2114 | * iteration. | |
2115 | */ | |
2116 | rq = task_rq_lock(p, &flags); | |
2117 | running = task_running(rq, p); | |
2118 | task_rq_unlock(rq, &flags); | |
2119 | ||
2120 | if (likely(!running)) | |
2121 | break; | |
2122 | /* | |
2123 | * The switch count is incremented before the actual | |
2124 | * context switch. We thus wait for two switches to be | |
2125 | * sure at least one completed. | |
2126 | */ | |
2127 | if ((p->nvcsw - nvcsw) > 1) | |
2128 | break; | |
2129 | if ((p->nivcsw - nivcsw) > 1) | |
2130 | break; | |
2131 | ||
2132 | cpu_relax(); | |
2133 | } | |
2134 | } | |
2135 | ||
1da177e4 LT |
2136 | /* |
2137 | * wait_task_inactive - wait for a thread to unschedule. | |
2138 | * | |
85ba2d86 RM |
2139 | * If @match_state is nonzero, it's the @p->state value just checked and |
2140 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2141 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2142 | * we return a positive number (its total switch count). If a second call | |
2143 | * a short while later returns the same number, the caller can be sure that | |
2144 | * @p has remained unscheduled the whole time. | |
2145 | * | |
1da177e4 LT |
2146 | * The caller must ensure that the task *will* unschedule sometime soon, |
2147 | * else this function might spin for a *long* time. This function can't | |
2148 | * be called with interrupts off, or it may introduce deadlock with | |
2149 | * smp_call_function() if an IPI is sent by the same process we are | |
2150 | * waiting to become inactive. | |
2151 | */ | |
85ba2d86 | 2152 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2153 | { |
2154 | unsigned long flags; | |
dd41f596 | 2155 | int running, on_rq; |
85ba2d86 | 2156 | unsigned long ncsw; |
70b97a7f | 2157 | struct rq *rq; |
1da177e4 | 2158 | |
3a5c359a AK |
2159 | for (;;) { |
2160 | /* | |
2161 | * We do the initial early heuristics without holding | |
2162 | * any task-queue locks at all. We'll only try to get | |
2163 | * the runqueue lock when things look like they will | |
2164 | * work out! | |
2165 | */ | |
2166 | rq = task_rq(p); | |
fa490cfd | 2167 | |
3a5c359a AK |
2168 | /* |
2169 | * If the task is actively running on another CPU | |
2170 | * still, just relax and busy-wait without holding | |
2171 | * any locks. | |
2172 | * | |
2173 | * NOTE! Since we don't hold any locks, it's not | |
2174 | * even sure that "rq" stays as the right runqueue! | |
2175 | * But we don't care, since "task_running()" will | |
2176 | * return false if the runqueue has changed and p | |
2177 | * is actually now running somewhere else! | |
2178 | */ | |
85ba2d86 RM |
2179 | while (task_running(rq, p)) { |
2180 | if (match_state && unlikely(p->state != match_state)) | |
2181 | return 0; | |
3a5c359a | 2182 | cpu_relax(); |
85ba2d86 | 2183 | } |
fa490cfd | 2184 | |
3a5c359a AK |
2185 | /* |
2186 | * Ok, time to look more closely! We need the rq | |
2187 | * lock now, to be *sure*. If we're wrong, we'll | |
2188 | * just go back and repeat. | |
2189 | */ | |
2190 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2191 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2192 | running = task_running(rq, p); |
2193 | on_rq = p->se.on_rq; | |
85ba2d86 | 2194 | ncsw = 0; |
f31e11d8 | 2195 | if (!match_state || p->state == match_state) |
93dcf55f | 2196 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2197 | task_rq_unlock(rq, &flags); |
fa490cfd | 2198 | |
85ba2d86 RM |
2199 | /* |
2200 | * If it changed from the expected state, bail out now. | |
2201 | */ | |
2202 | if (unlikely(!ncsw)) | |
2203 | break; | |
2204 | ||
3a5c359a AK |
2205 | /* |
2206 | * Was it really running after all now that we | |
2207 | * checked with the proper locks actually held? | |
2208 | * | |
2209 | * Oops. Go back and try again.. | |
2210 | */ | |
2211 | if (unlikely(running)) { | |
2212 | cpu_relax(); | |
2213 | continue; | |
2214 | } | |
fa490cfd | 2215 | |
3a5c359a AK |
2216 | /* |
2217 | * It's not enough that it's not actively running, | |
2218 | * it must be off the runqueue _entirely_, and not | |
2219 | * preempted! | |
2220 | * | |
80dd99b3 | 2221 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2222 | * running right now), it's preempted, and we should |
2223 | * yield - it could be a while. | |
2224 | */ | |
2225 | if (unlikely(on_rq)) { | |
2226 | schedule_timeout_uninterruptible(1); | |
2227 | continue; | |
2228 | } | |
fa490cfd | 2229 | |
3a5c359a AK |
2230 | /* |
2231 | * Ahh, all good. It wasn't running, and it wasn't | |
2232 | * runnable, which means that it will never become | |
2233 | * running in the future either. We're all done! | |
2234 | */ | |
2235 | break; | |
2236 | } | |
85ba2d86 RM |
2237 | |
2238 | return ncsw; | |
1da177e4 LT |
2239 | } |
2240 | ||
2241 | /*** | |
2242 | * kick_process - kick a running thread to enter/exit the kernel | |
2243 | * @p: the to-be-kicked thread | |
2244 | * | |
2245 | * Cause a process which is running on another CPU to enter | |
2246 | * kernel-mode, without any delay. (to get signals handled.) | |
2247 | * | |
2248 | * NOTE: this function doesnt have to take the runqueue lock, | |
2249 | * because all it wants to ensure is that the remote task enters | |
2250 | * the kernel. If the IPI races and the task has been migrated | |
2251 | * to another CPU then no harm is done and the purpose has been | |
2252 | * achieved as well. | |
2253 | */ | |
36c8b586 | 2254 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2255 | { |
2256 | int cpu; | |
2257 | ||
2258 | preempt_disable(); | |
2259 | cpu = task_cpu(p); | |
2260 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2261 | smp_send_reschedule(cpu); | |
2262 | preempt_enable(); | |
2263 | } | |
b43e3521 | 2264 | EXPORT_SYMBOL_GPL(kick_process); |
1da177e4 LT |
2265 | |
2266 | /* | |
2dd73a4f PW |
2267 | * Return a low guess at the load of a migration-source cpu weighted |
2268 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
2269 | * |
2270 | * We want to under-estimate the load of migration sources, to | |
2271 | * balance conservatively. | |
2272 | */ | |
a9957449 | 2273 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 2274 | { |
70b97a7f | 2275 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2276 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2277 | |
93b75217 | 2278 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2279 | return total; |
b910472d | 2280 | |
dd41f596 | 2281 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
2282 | } |
2283 | ||
2284 | /* | |
2dd73a4f PW |
2285 | * Return a high guess at the load of a migration-target cpu weighted |
2286 | * according to the scheduling class and "nice" value. | |
1da177e4 | 2287 | */ |
a9957449 | 2288 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 2289 | { |
70b97a7f | 2290 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 2291 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 2292 | |
93b75217 | 2293 | if (type == 0 || !sched_feat(LB_BIAS)) |
dd41f596 | 2294 | return total; |
3b0bd9bc | 2295 | |
dd41f596 | 2296 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
2297 | } |
2298 | ||
147cbb4b NP |
2299 | /* |
2300 | * find_idlest_group finds and returns the least busy CPU group within the | |
2301 | * domain. | |
2302 | */ | |
2303 | static struct sched_group * | |
2304 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
2305 | { | |
2306 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
2307 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
2308 | int load_idx = sd->forkexec_idx; | |
2309 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
2310 | ||
2311 | do { | |
2312 | unsigned long load, avg_load; | |
2313 | int local_group; | |
2314 | int i; | |
2315 | ||
da5a5522 | 2316 | /* Skip over this group if it has no CPUs allowed */ |
758b2cdc RR |
2317 | if (!cpumask_intersects(sched_group_cpus(group), |
2318 | &p->cpus_allowed)) | |
3a5c359a | 2319 | continue; |
da5a5522 | 2320 | |
758b2cdc RR |
2321 | local_group = cpumask_test_cpu(this_cpu, |
2322 | sched_group_cpus(group)); | |
147cbb4b NP |
2323 | |
2324 | /* Tally up the load of all CPUs in the group */ | |
2325 | avg_load = 0; | |
2326 | ||
758b2cdc | 2327 | for_each_cpu(i, sched_group_cpus(group)) { |
147cbb4b NP |
2328 | /* Bias balancing toward cpus of our domain */ |
2329 | if (local_group) | |
2330 | load = source_load(i, load_idx); | |
2331 | else | |
2332 | load = target_load(i, load_idx); | |
2333 | ||
2334 | avg_load += load; | |
2335 | } | |
2336 | ||
2337 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
2338 | avg_load = sg_div_cpu_power(group, |
2339 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
2340 | |
2341 | if (local_group) { | |
2342 | this_load = avg_load; | |
2343 | this = group; | |
2344 | } else if (avg_load < min_load) { | |
2345 | min_load = avg_load; | |
2346 | idlest = group; | |
2347 | } | |
3a5c359a | 2348 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
2349 | |
2350 | if (!idlest || 100*this_load < imbalance*min_load) | |
2351 | return NULL; | |
2352 | return idlest; | |
2353 | } | |
2354 | ||
2355 | /* | |
0feaece9 | 2356 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 2357 | */ |
95cdf3b7 | 2358 | static int |
758b2cdc | 2359 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu) |
147cbb4b NP |
2360 | { |
2361 | unsigned long load, min_load = ULONG_MAX; | |
2362 | int idlest = -1; | |
2363 | int i; | |
2364 | ||
da5a5522 | 2365 | /* Traverse only the allowed CPUs */ |
758b2cdc | 2366 | for_each_cpu_and(i, sched_group_cpus(group), &p->cpus_allowed) { |
2dd73a4f | 2367 | load = weighted_cpuload(i); |
147cbb4b NP |
2368 | |
2369 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2370 | min_load = load; | |
2371 | idlest = i; | |
2372 | } | |
2373 | } | |
2374 | ||
2375 | return idlest; | |
2376 | } | |
2377 | ||
476d139c NP |
2378 | /* |
2379 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2380 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2381 | * SD_BALANCE_EXEC. | |
2382 | * | |
2383 | * Balance, ie. select the least loaded group. | |
2384 | * | |
2385 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2386 | * | |
2387 | * preempt must be disabled. | |
2388 | */ | |
2389 | static int sched_balance_self(int cpu, int flag) | |
2390 | { | |
2391 | struct task_struct *t = current; | |
2392 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2393 | |
c96d145e | 2394 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2395 | /* |
2396 | * If power savings logic is enabled for a domain, stop there. | |
2397 | */ | |
5c45bf27 SS |
2398 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2399 | break; | |
476d139c NP |
2400 | if (tmp->flags & flag) |
2401 | sd = tmp; | |
c96d145e | 2402 | } |
476d139c | 2403 | |
039a1c41 PZ |
2404 | if (sd) |
2405 | update_shares(sd); | |
2406 | ||
476d139c | 2407 | while (sd) { |
476d139c | 2408 | struct sched_group *group; |
1a848870 SS |
2409 | int new_cpu, weight; |
2410 | ||
2411 | if (!(sd->flags & flag)) { | |
2412 | sd = sd->child; | |
2413 | continue; | |
2414 | } | |
476d139c | 2415 | |
476d139c | 2416 | group = find_idlest_group(sd, t, cpu); |
1a848870 SS |
2417 | if (!group) { |
2418 | sd = sd->child; | |
2419 | continue; | |
2420 | } | |
476d139c | 2421 | |
758b2cdc | 2422 | new_cpu = find_idlest_cpu(group, t, cpu); |
1a848870 SS |
2423 | if (new_cpu == -1 || new_cpu == cpu) { |
2424 | /* Now try balancing at a lower domain level of cpu */ | |
2425 | sd = sd->child; | |
2426 | continue; | |
2427 | } | |
476d139c | 2428 | |
1a848870 | 2429 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2430 | cpu = new_cpu; |
758b2cdc | 2431 | weight = cpumask_weight(sched_domain_span(sd)); |
476d139c | 2432 | sd = NULL; |
476d139c | 2433 | for_each_domain(cpu, tmp) { |
758b2cdc | 2434 | if (weight <= cpumask_weight(sched_domain_span(tmp))) |
476d139c NP |
2435 | break; |
2436 | if (tmp->flags & flag) | |
2437 | sd = tmp; | |
2438 | } | |
2439 | /* while loop will break here if sd == NULL */ | |
2440 | } | |
2441 | ||
2442 | return cpu; | |
2443 | } | |
2444 | ||
2445 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2446 | |
0793a61d TG |
2447 | /** |
2448 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2449 | * @p: the task to evaluate | |
2450 | * @func: the function to be called | |
2451 | * @info: the function call argument | |
2452 | * | |
2453 | * Calls the function @func when the task is currently running. This might | |
2454 | * be on the current CPU, which just calls the function directly | |
2455 | */ | |
2456 | void task_oncpu_function_call(struct task_struct *p, | |
2457 | void (*func) (void *info), void *info) | |
2458 | { | |
2459 | int cpu; | |
2460 | ||
2461 | preempt_disable(); | |
2462 | cpu = task_cpu(p); | |
2463 | if (task_curr(p)) | |
2464 | smp_call_function_single(cpu, func, info, 1); | |
2465 | preempt_enable(); | |
2466 | } | |
2467 | ||
1da177e4 LT |
2468 | /*** |
2469 | * try_to_wake_up - wake up a thread | |
2470 | * @p: the to-be-woken-up thread | |
2471 | * @state: the mask of task states that can be woken | |
2472 | * @sync: do a synchronous wakeup? | |
2473 | * | |
2474 | * Put it on the run-queue if it's not already there. The "current" | |
2475 | * thread is always on the run-queue (except when the actual | |
2476 | * re-schedule is in progress), and as such you're allowed to do | |
2477 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2478 | * runnable without the overhead of this. | |
2479 | * | |
2480 | * returns failure only if the task is already active. | |
2481 | */ | |
36c8b586 | 2482 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2483 | { |
cc367732 | 2484 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2485 | unsigned long flags; |
2486 | long old_state; | |
70b97a7f | 2487 | struct rq *rq; |
1da177e4 | 2488 | |
b85d0667 IM |
2489 | if (!sched_feat(SYNC_WAKEUPS)) |
2490 | sync = 0; | |
2491 | ||
2398f2c6 | 2492 | #ifdef CONFIG_SMP |
57310a98 | 2493 | if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) { |
2398f2c6 PZ |
2494 | struct sched_domain *sd; |
2495 | ||
2496 | this_cpu = raw_smp_processor_id(); | |
2497 | cpu = task_cpu(p); | |
2498 | ||
2499 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2500 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
2398f2c6 PZ |
2501 | update_shares(sd); |
2502 | break; | |
2503 | } | |
2504 | } | |
2505 | } | |
2506 | #endif | |
2507 | ||
04e2f174 | 2508 | smp_wmb(); |
1da177e4 | 2509 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2510 | update_rq_clock(rq); |
1da177e4 LT |
2511 | old_state = p->state; |
2512 | if (!(old_state & state)) | |
2513 | goto out; | |
2514 | ||
dd41f596 | 2515 | if (p->se.on_rq) |
1da177e4 LT |
2516 | goto out_running; |
2517 | ||
2518 | cpu = task_cpu(p); | |
cc367732 | 2519 | orig_cpu = cpu; |
1da177e4 LT |
2520 | this_cpu = smp_processor_id(); |
2521 | ||
2522 | #ifdef CONFIG_SMP | |
2523 | if (unlikely(task_running(rq, p))) | |
2524 | goto out_activate; | |
2525 | ||
5d2f5a61 DA |
2526 | cpu = p->sched_class->select_task_rq(p, sync); |
2527 | if (cpu != orig_cpu) { | |
2528 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2529 | task_rq_unlock(rq, &flags); |
2530 | /* might preempt at this point */ | |
2531 | rq = task_rq_lock(p, &flags); | |
2532 | old_state = p->state; | |
2533 | if (!(old_state & state)) | |
2534 | goto out; | |
dd41f596 | 2535 | if (p->se.on_rq) |
1da177e4 LT |
2536 | goto out_running; |
2537 | ||
2538 | this_cpu = smp_processor_id(); | |
2539 | cpu = task_cpu(p); | |
2540 | } | |
2541 | ||
e7693a36 GH |
2542 | #ifdef CONFIG_SCHEDSTATS |
2543 | schedstat_inc(rq, ttwu_count); | |
2544 | if (cpu == this_cpu) | |
2545 | schedstat_inc(rq, ttwu_local); | |
2546 | else { | |
2547 | struct sched_domain *sd; | |
2548 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2549 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2550 | schedstat_inc(sd, ttwu_wake_remote); |
2551 | break; | |
2552 | } | |
2553 | } | |
2554 | } | |
6d6bc0ad | 2555 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2556 | |
1da177e4 LT |
2557 | out_activate: |
2558 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2559 | schedstat_inc(p, se.nr_wakeups); |
2560 | if (sync) | |
2561 | schedstat_inc(p, se.nr_wakeups_sync); | |
2562 | if (orig_cpu != cpu) | |
2563 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2564 | if (cpu == this_cpu) | |
2565 | schedstat_inc(p, se.nr_wakeups_local); | |
2566 | else | |
2567 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2568 | activate_task(rq, p, 1); |
1da177e4 LT |
2569 | success = 1; |
2570 | ||
831451ac PZ |
2571 | /* |
2572 | * Only attribute actual wakeups done by this task. | |
2573 | */ | |
2574 | if (!in_interrupt()) { | |
2575 | struct sched_entity *se = ¤t->se; | |
2576 | u64 sample = se->sum_exec_runtime; | |
2577 | ||
2578 | if (se->last_wakeup) | |
2579 | sample -= se->last_wakeup; | |
2580 | else | |
2581 | sample -= se->start_runtime; | |
2582 | update_avg(&se->avg_wakeup, sample); | |
2583 | ||
2584 | se->last_wakeup = se->sum_exec_runtime; | |
2585 | } | |
2586 | ||
1da177e4 | 2587 | out_running: |
468a15bb | 2588 | trace_sched_wakeup(rq, p, success); |
15afe09b | 2589 | check_preempt_curr(rq, p, sync); |
4ae7d5ce | 2590 | |
1da177e4 | 2591 | p->state = TASK_RUNNING; |
9a897c5a SR |
2592 | #ifdef CONFIG_SMP |
2593 | if (p->sched_class->task_wake_up) | |
2594 | p->sched_class->task_wake_up(rq, p); | |
2595 | #endif | |
1da177e4 LT |
2596 | out: |
2597 | task_rq_unlock(rq, &flags); | |
2598 | ||
2599 | return success; | |
2600 | } | |
2601 | ||
50fa610a DH |
2602 | /** |
2603 | * wake_up_process - Wake up a specific process | |
2604 | * @p: The process to be woken up. | |
2605 | * | |
2606 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2607 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2608 | * running. | |
2609 | * | |
2610 | * It may be assumed that this function implies a write memory barrier before | |
2611 | * changing the task state if and only if any tasks are woken up. | |
2612 | */ | |
7ad5b3a5 | 2613 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2614 | { |
d9514f6c | 2615 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2616 | } |
1da177e4 LT |
2617 | EXPORT_SYMBOL(wake_up_process); |
2618 | ||
7ad5b3a5 | 2619 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2620 | { |
2621 | return try_to_wake_up(p, state, 0); | |
2622 | } | |
2623 | ||
1da177e4 LT |
2624 | /* |
2625 | * Perform scheduler related setup for a newly forked process p. | |
2626 | * p is forked by current. | |
dd41f596 IM |
2627 | * |
2628 | * __sched_fork() is basic setup used by init_idle() too: | |
2629 | */ | |
2630 | static void __sched_fork(struct task_struct *p) | |
2631 | { | |
dd41f596 IM |
2632 | p->se.exec_start = 0; |
2633 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2634 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2635 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2636 | p->se.last_wakeup = 0; |
2637 | p->se.avg_overlap = 0; | |
831451ac PZ |
2638 | p->se.start_runtime = 0; |
2639 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2640 | |
2641 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2642 | p->se.wait_start = 0; |
2643 | p->se.wait_max = 0; | |
2644 | p->se.wait_count = 0; | |
2645 | p->se.wait_sum = 0; | |
2646 | ||
2647 | p->se.sleep_start = 0; | |
2648 | p->se.sleep_max = 0; | |
2649 | p->se.sum_sleep_runtime = 0; | |
2650 | ||
2651 | p->se.block_start = 0; | |
2652 | p->se.block_max = 0; | |
2653 | p->se.exec_max = 0; | |
2654 | p->se.slice_max = 0; | |
2655 | ||
2656 | p->se.nr_migrations_cold = 0; | |
2657 | p->se.nr_failed_migrations_affine = 0; | |
2658 | p->se.nr_failed_migrations_running = 0; | |
2659 | p->se.nr_failed_migrations_hot = 0; | |
2660 | p->se.nr_forced_migrations = 0; | |
2661 | p->se.nr_forced2_migrations = 0; | |
2662 | ||
2663 | p->se.nr_wakeups = 0; | |
2664 | p->se.nr_wakeups_sync = 0; | |
2665 | p->se.nr_wakeups_migrate = 0; | |
2666 | p->se.nr_wakeups_local = 0; | |
2667 | p->se.nr_wakeups_remote = 0; | |
2668 | p->se.nr_wakeups_affine = 0; | |
2669 | p->se.nr_wakeups_affine_attempts = 0; | |
2670 | p->se.nr_wakeups_passive = 0; | |
2671 | p->se.nr_wakeups_idle = 0; | |
2672 | ||
6cfb0d5d | 2673 | #endif |
476d139c | 2674 | |
fa717060 | 2675 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2676 | p->se.on_rq = 0; |
4a55bd5e | 2677 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2678 | |
e107be36 AK |
2679 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2680 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2681 | #endif | |
2682 | ||
1da177e4 LT |
2683 | /* |
2684 | * We mark the process as running here, but have not actually | |
2685 | * inserted it onto the runqueue yet. This guarantees that | |
2686 | * nobody will actually run it, and a signal or other external | |
2687 | * event cannot wake it up and insert it on the runqueue either. | |
2688 | */ | |
2689 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2690 | } |
2691 | ||
2692 | /* | |
2693 | * fork()/clone()-time setup: | |
2694 | */ | |
2695 | void sched_fork(struct task_struct *p, int clone_flags) | |
2696 | { | |
2697 | int cpu = get_cpu(); | |
2698 | ||
2699 | __sched_fork(p); | |
2700 | ||
2701 | #ifdef CONFIG_SMP | |
2702 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2703 | #endif | |
02e4bac2 | 2704 | set_task_cpu(p, cpu); |
b29739f9 IM |
2705 | |
2706 | /* | |
b9dc29e7 | 2707 | * Make sure we do not leak PI boosting priority to the child. |
b29739f9 | 2708 | */ |
b9dc29e7 | 2709 | p->prio = current->normal_prio; |
ca94c442 | 2710 | |
b9dc29e7 MG |
2711 | /* |
2712 | * Revert to default priority/policy on fork if requested. | |
2713 | */ | |
2714 | if (unlikely(p->sched_reset_on_fork)) { | |
2715 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) | |
2716 | p->policy = SCHED_NORMAL; | |
2717 | ||
2718 | if (p->normal_prio < DEFAULT_PRIO) | |
2719 | p->prio = DEFAULT_PRIO; | |
2720 | ||
6c697bdf MG |
2721 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2722 | p->static_prio = NICE_TO_PRIO(0); | |
2723 | set_load_weight(p); | |
2724 | } | |
2725 | ||
b9dc29e7 MG |
2726 | /* |
2727 | * We don't need the reset flag anymore after the fork. It has | |
2728 | * fulfilled its duty: | |
2729 | */ | |
2730 | p->sched_reset_on_fork = 0; | |
2731 | } | |
ca94c442 | 2732 | |
2ddbf952 HS |
2733 | if (!rt_prio(p->prio)) |
2734 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2735 | |
52f17b6c | 2736 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2737 | if (likely(sched_info_on())) |
52f17b6c | 2738 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2739 | #endif |
d6077cb8 | 2740 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2741 | p->oncpu = 0; |
2742 | #endif | |
1da177e4 | 2743 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2744 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2745 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2746 | #endif |
917b627d GH |
2747 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2748 | ||
476d139c | 2749 | put_cpu(); |
1da177e4 LT |
2750 | } |
2751 | ||
2752 | /* | |
2753 | * wake_up_new_task - wake up a newly created task for the first time. | |
2754 | * | |
2755 | * This function will do some initial scheduler statistics housekeeping | |
2756 | * that must be done for every newly created context, then puts the task | |
2757 | * on the runqueue and wakes it. | |
2758 | */ | |
7ad5b3a5 | 2759 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2760 | { |
2761 | unsigned long flags; | |
dd41f596 | 2762 | struct rq *rq; |
1da177e4 LT |
2763 | |
2764 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2765 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2766 | update_rq_clock(rq); |
1da177e4 LT |
2767 | |
2768 | p->prio = effective_prio(p); | |
2769 | ||
b9dca1e0 | 2770 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2771 | activate_task(rq, p, 0); |
1da177e4 | 2772 | } else { |
1da177e4 | 2773 | /* |
dd41f596 IM |
2774 | * Let the scheduling class do new task startup |
2775 | * management (if any): | |
1da177e4 | 2776 | */ |
ee0827d8 | 2777 | p->sched_class->task_new(rq, p); |
c09595f6 | 2778 | inc_nr_running(rq); |
1da177e4 | 2779 | } |
c71dd42d | 2780 | trace_sched_wakeup_new(rq, p, 1); |
15afe09b | 2781 | check_preempt_curr(rq, p, 0); |
9a897c5a SR |
2782 | #ifdef CONFIG_SMP |
2783 | if (p->sched_class->task_wake_up) | |
2784 | p->sched_class->task_wake_up(rq, p); | |
2785 | #endif | |
dd41f596 | 2786 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2787 | } |
2788 | ||
e107be36 AK |
2789 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2790 | ||
2791 | /** | |
80dd99b3 | 2792 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2793 | * @notifier: notifier struct to register |
e107be36 AK |
2794 | */ |
2795 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2796 | { | |
2797 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2798 | } | |
2799 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2800 | ||
2801 | /** | |
2802 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2803 | * @notifier: notifier struct to unregister |
e107be36 AK |
2804 | * |
2805 | * This is safe to call from within a preemption notifier. | |
2806 | */ | |
2807 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2808 | { | |
2809 | hlist_del(¬ifier->link); | |
2810 | } | |
2811 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2812 | ||
2813 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2814 | { | |
2815 | struct preempt_notifier *notifier; | |
2816 | struct hlist_node *node; | |
2817 | ||
2818 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2819 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2820 | } | |
2821 | ||
2822 | static void | |
2823 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2824 | struct task_struct *next) | |
2825 | { | |
2826 | struct preempt_notifier *notifier; | |
2827 | struct hlist_node *node; | |
2828 | ||
2829 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2830 | notifier->ops->sched_out(notifier, next); | |
2831 | } | |
2832 | ||
6d6bc0ad | 2833 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2834 | |
2835 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2836 | { | |
2837 | } | |
2838 | ||
2839 | static void | |
2840 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2841 | struct task_struct *next) | |
2842 | { | |
2843 | } | |
2844 | ||
6d6bc0ad | 2845 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2846 | |
4866cde0 NP |
2847 | /** |
2848 | * prepare_task_switch - prepare to switch tasks | |
2849 | * @rq: the runqueue preparing to switch | |
421cee29 | 2850 | * @prev: the current task that is being switched out |
4866cde0 NP |
2851 | * @next: the task we are going to switch to. |
2852 | * | |
2853 | * This is called with the rq lock held and interrupts off. It must | |
2854 | * be paired with a subsequent finish_task_switch after the context | |
2855 | * switch. | |
2856 | * | |
2857 | * prepare_task_switch sets up locking and calls architecture specific | |
2858 | * hooks. | |
2859 | */ | |
e107be36 AK |
2860 | static inline void |
2861 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2862 | struct task_struct *next) | |
4866cde0 | 2863 | { |
e107be36 | 2864 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2865 | prepare_lock_switch(rq, next); |
2866 | prepare_arch_switch(next); | |
2867 | } | |
2868 | ||
1da177e4 LT |
2869 | /** |
2870 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2871 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2872 | * @prev: the thread we just switched away from. |
2873 | * | |
4866cde0 NP |
2874 | * finish_task_switch must be called after the context switch, paired |
2875 | * with a prepare_task_switch call before the context switch. | |
2876 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2877 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2878 | * |
2879 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2880 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2881 | * with the lock held can cause deadlocks; see schedule() for |
2882 | * details.) | |
2883 | */ | |
3f029d3c | 2884 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2885 | __releases(rq->lock) |
2886 | { | |
1da177e4 | 2887 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2888 | long prev_state; |
1da177e4 LT |
2889 | |
2890 | rq->prev_mm = NULL; | |
2891 | ||
2892 | /* | |
2893 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2894 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2895 | * schedule one last time. The schedule call will never return, and |
2896 | * the scheduled task must drop that reference. | |
c394cc9f | 2897 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2898 | * still held, otherwise prev could be scheduled on another cpu, die |
2899 | * there before we look at prev->state, and then the reference would | |
2900 | * be dropped twice. | |
2901 | * Manfred Spraul <manfred@colorfullife.com> | |
2902 | */ | |
55a101f8 | 2903 | prev_state = prev->state; |
4866cde0 | 2904 | finish_arch_switch(prev); |
0793a61d | 2905 | perf_counter_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2906 | finish_lock_switch(rq, prev); |
e8fa1362 | 2907 | |
e107be36 | 2908 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2909 | if (mm) |
2910 | mmdrop(mm); | |
c394cc9f | 2911 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2912 | /* |
2913 | * Remove function-return probe instances associated with this | |
2914 | * task and put them back on the free list. | |
9761eea8 | 2915 | */ |
c6fd91f0 | 2916 | kprobe_flush_task(prev); |
1da177e4 | 2917 | put_task_struct(prev); |
c6fd91f0 | 2918 | } |
3f029d3c GH |
2919 | } |
2920 | ||
2921 | #ifdef CONFIG_SMP | |
2922 | ||
2923 | /* assumes rq->lock is held */ | |
2924 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2925 | { | |
2926 | if (prev->sched_class->pre_schedule) | |
2927 | prev->sched_class->pre_schedule(rq, prev); | |
2928 | } | |
2929 | ||
2930 | /* rq->lock is NOT held, but preemption is disabled */ | |
2931 | static inline void post_schedule(struct rq *rq) | |
2932 | { | |
2933 | if (rq->post_schedule) { | |
2934 | unsigned long flags; | |
2935 | ||
2936 | spin_lock_irqsave(&rq->lock, flags); | |
2937 | if (rq->curr->sched_class->post_schedule) | |
2938 | rq->curr->sched_class->post_schedule(rq); | |
2939 | spin_unlock_irqrestore(&rq->lock, flags); | |
2940 | ||
2941 | rq->post_schedule = 0; | |
2942 | } | |
2943 | } | |
2944 | ||
2945 | #else | |
da19ab51 | 2946 | |
3f029d3c GH |
2947 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2948 | { | |
2949 | } | |
2950 | ||
2951 | static inline void post_schedule(struct rq *rq) | |
2952 | { | |
1da177e4 LT |
2953 | } |
2954 | ||
3f029d3c GH |
2955 | #endif |
2956 | ||
1da177e4 LT |
2957 | /** |
2958 | * schedule_tail - first thing a freshly forked thread must call. | |
2959 | * @prev: the thread we just switched away from. | |
2960 | */ | |
36c8b586 | 2961 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2962 | __releases(rq->lock) |
2963 | { | |
70b97a7f | 2964 | struct rq *rq = this_rq(); |
da19ab51 | 2965 | |
3f029d3c | 2966 | finish_task_switch(rq, prev); |
da19ab51 | 2967 | |
3f029d3c GH |
2968 | /* |
2969 | * FIXME: do we need to worry about rq being invalidated by the | |
2970 | * task_switch? | |
2971 | */ | |
2972 | post_schedule(rq); | |
70b97a7f | 2973 | |
4866cde0 NP |
2974 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2975 | /* In this case, finish_task_switch does not reenable preemption */ | |
2976 | preempt_enable(); | |
2977 | #endif | |
1da177e4 | 2978 | if (current->set_child_tid) |
b488893a | 2979 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2980 | } |
2981 | ||
2982 | /* | |
2983 | * context_switch - switch to the new MM and the new | |
2984 | * thread's register state. | |
2985 | */ | |
3f029d3c | 2986 | static inline void |
70b97a7f | 2987 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2988 | struct task_struct *next) |
1da177e4 | 2989 | { |
dd41f596 | 2990 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2991 | |
e107be36 | 2992 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2993 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2994 | mm = next->mm; |
2995 | oldmm = prev->active_mm; | |
9226d125 ZA |
2996 | /* |
2997 | * For paravirt, this is coupled with an exit in switch_to to | |
2998 | * combine the page table reload and the switch backend into | |
2999 | * one hypercall. | |
3000 | */ | |
224101ed | 3001 | arch_start_context_switch(prev); |
9226d125 | 3002 | |
dd41f596 | 3003 | if (unlikely(!mm)) { |
1da177e4 LT |
3004 | next->active_mm = oldmm; |
3005 | atomic_inc(&oldmm->mm_count); | |
3006 | enter_lazy_tlb(oldmm, next); | |
3007 | } else | |
3008 | switch_mm(oldmm, mm, next); | |
3009 | ||
dd41f596 | 3010 | if (unlikely(!prev->mm)) { |
1da177e4 | 3011 | prev->active_mm = NULL; |
1da177e4 LT |
3012 | rq->prev_mm = oldmm; |
3013 | } | |
3a5f5e48 IM |
3014 | /* |
3015 | * Since the runqueue lock will be released by the next | |
3016 | * task (which is an invalid locking op but in the case | |
3017 | * of the scheduler it's an obvious special-case), so we | |
3018 | * do an early lockdep release here: | |
3019 | */ | |
3020 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3021 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3022 | #endif |
1da177e4 LT |
3023 | |
3024 | /* Here we just switch the register state and the stack. */ | |
3025 | switch_to(prev, next, prev); | |
3026 | ||
dd41f596 IM |
3027 | barrier(); |
3028 | /* | |
3029 | * this_rq must be evaluated again because prev may have moved | |
3030 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3031 | * frame will be invalid. | |
3032 | */ | |
3f029d3c | 3033 | finish_task_switch(this_rq(), prev); |
1da177e4 LT |
3034 | } |
3035 | ||
3036 | /* | |
3037 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3038 | * | |
3039 | * externally visible scheduler statistics: current number of runnable | |
3040 | * threads, current number of uninterruptible-sleeping threads, total | |
3041 | * number of context switches performed since bootup. | |
3042 | */ | |
3043 | unsigned long nr_running(void) | |
3044 | { | |
3045 | unsigned long i, sum = 0; | |
3046 | ||
3047 | for_each_online_cpu(i) | |
3048 | sum += cpu_rq(i)->nr_running; | |
3049 | ||
3050 | return sum; | |
3051 | } | |
3052 | ||
3053 | unsigned long nr_uninterruptible(void) | |
3054 | { | |
3055 | unsigned long i, sum = 0; | |
3056 | ||
0a945022 | 3057 | for_each_possible_cpu(i) |
1da177e4 LT |
3058 | sum += cpu_rq(i)->nr_uninterruptible; |
3059 | ||
3060 | /* | |
3061 | * Since we read the counters lockless, it might be slightly | |
3062 | * inaccurate. Do not allow it to go below zero though: | |
3063 | */ | |
3064 | if (unlikely((long)sum < 0)) | |
3065 | sum = 0; | |
3066 | ||
3067 | return sum; | |
3068 | } | |
3069 | ||
3070 | unsigned long long nr_context_switches(void) | |
3071 | { | |
cc94abfc SR |
3072 | int i; |
3073 | unsigned long long sum = 0; | |
1da177e4 | 3074 | |
0a945022 | 3075 | for_each_possible_cpu(i) |
1da177e4 LT |
3076 | sum += cpu_rq(i)->nr_switches; |
3077 | ||
3078 | return sum; | |
3079 | } | |
3080 | ||
3081 | unsigned long nr_iowait(void) | |
3082 | { | |
3083 | unsigned long i, sum = 0; | |
3084 | ||
0a945022 | 3085 | for_each_possible_cpu(i) |
1da177e4 LT |
3086 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
3087 | ||
3088 | return sum; | |
3089 | } | |
3090 | ||
dce48a84 TG |
3091 | /* Variables and functions for calc_load */ |
3092 | static atomic_long_t calc_load_tasks; | |
3093 | static unsigned long calc_load_update; | |
3094 | unsigned long avenrun[3]; | |
3095 | EXPORT_SYMBOL(avenrun); | |
3096 | ||
2d02494f TG |
3097 | /** |
3098 | * get_avenrun - get the load average array | |
3099 | * @loads: pointer to dest load array | |
3100 | * @offset: offset to add | |
3101 | * @shift: shift count to shift the result left | |
3102 | * | |
3103 | * These values are estimates at best, so no need for locking. | |
3104 | */ | |
3105 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3106 | { | |
3107 | loads[0] = (avenrun[0] + offset) << shift; | |
3108 | loads[1] = (avenrun[1] + offset) << shift; | |
3109 | loads[2] = (avenrun[2] + offset) << shift; | |
3110 | } | |
3111 | ||
dce48a84 TG |
3112 | static unsigned long |
3113 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3114 | { |
dce48a84 TG |
3115 | load *= exp; |
3116 | load += active * (FIXED_1 - exp); | |
3117 | return load >> FSHIFT; | |
3118 | } | |
db1b1fef | 3119 | |
dce48a84 TG |
3120 | /* |
3121 | * calc_load - update the avenrun load estimates 10 ticks after the | |
3122 | * CPUs have updated calc_load_tasks. | |
3123 | */ | |
3124 | void calc_global_load(void) | |
3125 | { | |
3126 | unsigned long upd = calc_load_update + 10; | |
3127 | long active; | |
3128 | ||
3129 | if (time_before(jiffies, upd)) | |
3130 | return; | |
db1b1fef | 3131 | |
dce48a84 TG |
3132 | active = atomic_long_read(&calc_load_tasks); |
3133 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 3134 | |
dce48a84 TG |
3135 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3136 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3137 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
3138 | ||
3139 | calc_load_update += LOAD_FREQ; | |
3140 | } | |
3141 | ||
3142 | /* | |
3143 | * Either called from update_cpu_load() or from a cpu going idle | |
3144 | */ | |
3145 | static void calc_load_account_active(struct rq *this_rq) | |
3146 | { | |
3147 | long nr_active, delta; | |
3148 | ||
3149 | nr_active = this_rq->nr_running; | |
3150 | nr_active += (long) this_rq->nr_uninterruptible; | |
3151 | ||
3152 | if (nr_active != this_rq->calc_load_active) { | |
3153 | delta = nr_active - this_rq->calc_load_active; | |
3154 | this_rq->calc_load_active = nr_active; | |
3155 | atomic_long_add(delta, &calc_load_tasks); | |
3156 | } | |
db1b1fef JS |
3157 | } |
3158 | ||
23a185ca PM |
3159 | /* |
3160 | * Externally visible per-cpu scheduler statistics: | |
23a185ca PM |
3161 | * cpu_nr_migrations(cpu) - number of migrations into that cpu |
3162 | */ | |
23a185ca PM |
3163 | u64 cpu_nr_migrations(int cpu) |
3164 | { | |
3165 | return cpu_rq(cpu)->nr_migrations_in; | |
3166 | } | |
3167 | ||
48f24c4d | 3168 | /* |
dd41f596 IM |
3169 | * Update rq->cpu_load[] statistics. This function is usually called every |
3170 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3171 | */ |
dd41f596 | 3172 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3173 | { |
495eca49 | 3174 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3175 | int i, scale; |
3176 | ||
3177 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3178 | |
3179 | /* Update our load: */ | |
3180 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3181 | unsigned long old_load, new_load; | |
3182 | ||
3183 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3184 | ||
3185 | old_load = this_rq->cpu_load[i]; | |
3186 | new_load = this_load; | |
a25707f3 IM |
3187 | /* |
3188 | * Round up the averaging division if load is increasing. This | |
3189 | * prevents us from getting stuck on 9 if the load is 10, for | |
3190 | * example. | |
3191 | */ | |
3192 | if (new_load > old_load) | |
3193 | new_load += scale-1; | |
dd41f596 IM |
3194 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3195 | } | |
dce48a84 TG |
3196 | |
3197 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3198 | this_rq->calc_load_update += LOAD_FREQ; | |
3199 | calc_load_account_active(this_rq); | |
3200 | } | |
48f24c4d IM |
3201 | } |
3202 | ||
dd41f596 IM |
3203 | #ifdef CONFIG_SMP |
3204 | ||
1da177e4 LT |
3205 | /* |
3206 | * double_rq_lock - safely lock two runqueues | |
3207 | * | |
3208 | * Note this does not disable interrupts like task_rq_lock, | |
3209 | * you need to do so manually before calling. | |
3210 | */ | |
70b97a7f | 3211 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3212 | __acquires(rq1->lock) |
3213 | __acquires(rq2->lock) | |
3214 | { | |
054b9108 | 3215 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
3216 | if (rq1 == rq2) { |
3217 | spin_lock(&rq1->lock); | |
3218 | __acquire(rq2->lock); /* Fake it out ;) */ | |
3219 | } else { | |
c96d145e | 3220 | if (rq1 < rq2) { |
1da177e4 | 3221 | spin_lock(&rq1->lock); |
5e710e37 | 3222 | spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3223 | } else { |
3224 | spin_lock(&rq2->lock); | |
5e710e37 | 3225 | spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); |
1da177e4 LT |
3226 | } |
3227 | } | |
6e82a3be IM |
3228 | update_rq_clock(rq1); |
3229 | update_rq_clock(rq2); | |
1da177e4 LT |
3230 | } |
3231 | ||
3232 | /* | |
3233 | * double_rq_unlock - safely unlock two runqueues | |
3234 | * | |
3235 | * Note this does not restore interrupts like task_rq_unlock, | |
3236 | * you need to do so manually after calling. | |
3237 | */ | |
70b97a7f | 3238 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
3239 | __releases(rq1->lock) |
3240 | __releases(rq2->lock) | |
3241 | { | |
3242 | spin_unlock(&rq1->lock); | |
3243 | if (rq1 != rq2) | |
3244 | spin_unlock(&rq2->lock); | |
3245 | else | |
3246 | __release(rq2->lock); | |
3247 | } | |
3248 | ||
1da177e4 LT |
3249 | /* |
3250 | * If dest_cpu is allowed for this process, migrate the task to it. | |
3251 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 3252 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
3253 | * the cpu_allowed mask is restored. |
3254 | */ | |
36c8b586 | 3255 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 3256 | { |
70b97a7f | 3257 | struct migration_req req; |
1da177e4 | 3258 | unsigned long flags; |
70b97a7f | 3259 | struct rq *rq; |
1da177e4 LT |
3260 | |
3261 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 3262 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
e761b772 | 3263 | || unlikely(!cpu_active(dest_cpu))) |
1da177e4 LT |
3264 | goto out; |
3265 | ||
3266 | /* force the process onto the specified CPU */ | |
3267 | if (migrate_task(p, dest_cpu, &req)) { | |
3268 | /* Need to wait for migration thread (might exit: take ref). */ | |
3269 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3270 | |
1da177e4 LT |
3271 | get_task_struct(mt); |
3272 | task_rq_unlock(rq, &flags); | |
3273 | wake_up_process(mt); | |
3274 | put_task_struct(mt); | |
3275 | wait_for_completion(&req.done); | |
36c8b586 | 3276 | |
1da177e4 LT |
3277 | return; |
3278 | } | |
3279 | out: | |
3280 | task_rq_unlock(rq, &flags); | |
3281 | } | |
3282 | ||
3283 | /* | |
476d139c NP |
3284 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3285 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
3286 | */ |
3287 | void sched_exec(void) | |
3288 | { | |
1da177e4 | 3289 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 3290 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 3291 | put_cpu(); |
476d139c NP |
3292 | if (new_cpu != this_cpu) |
3293 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
3294 | } |
3295 | ||
3296 | /* | |
3297 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
3298 | * Both runqueues must be locked. | |
3299 | */ | |
dd41f596 IM |
3300 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
3301 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 3302 | { |
2e1cb74a | 3303 | deactivate_task(src_rq, p, 0); |
1da177e4 | 3304 | set_task_cpu(p, this_cpu); |
dd41f596 | 3305 | activate_task(this_rq, p, 0); |
1da177e4 LT |
3306 | /* |
3307 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
3308 | * to be always true for them. | |
3309 | */ | |
15afe09b | 3310 | check_preempt_curr(this_rq, p, 0); |
1da177e4 LT |
3311 | } |
3312 | ||
3313 | /* | |
3314 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
3315 | */ | |
858119e1 | 3316 | static |
70b97a7f | 3317 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 3318 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 3319 | int *all_pinned) |
1da177e4 | 3320 | { |
708dc512 | 3321 | int tsk_cache_hot = 0; |
1da177e4 LT |
3322 | /* |
3323 | * We do not migrate tasks that are: | |
3324 | * 1) running (obviously), or | |
3325 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
3326 | * 3) are cache-hot on their current CPU. | |
3327 | */ | |
96f874e2 | 3328 | if (!cpumask_test_cpu(this_cpu, &p->cpus_allowed)) { |
cc367732 | 3329 | schedstat_inc(p, se.nr_failed_migrations_affine); |
1da177e4 | 3330 | return 0; |
cc367732 | 3331 | } |
81026794 NP |
3332 | *all_pinned = 0; |
3333 | ||
cc367732 IM |
3334 | if (task_running(rq, p)) { |
3335 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 3336 | return 0; |
cc367732 | 3337 | } |
1da177e4 | 3338 | |
da84d961 IM |
3339 | /* |
3340 | * Aggressive migration if: | |
3341 | * 1) task is cache cold, or | |
3342 | * 2) too many balance attempts have failed. | |
3343 | */ | |
3344 | ||
708dc512 LH |
3345 | tsk_cache_hot = task_hot(p, rq->clock, sd); |
3346 | if (!tsk_cache_hot || | |
3347 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 3348 | #ifdef CONFIG_SCHEDSTATS |
708dc512 | 3349 | if (tsk_cache_hot) { |
da84d961 | 3350 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
3351 | schedstat_inc(p, se.nr_forced_migrations); |
3352 | } | |
da84d961 IM |
3353 | #endif |
3354 | return 1; | |
3355 | } | |
3356 | ||
708dc512 | 3357 | if (tsk_cache_hot) { |
cc367732 | 3358 | schedstat_inc(p, se.nr_failed_migrations_hot); |
da84d961 | 3359 | return 0; |
cc367732 | 3360 | } |
1da177e4 LT |
3361 | return 1; |
3362 | } | |
3363 | ||
e1d1484f PW |
3364 | static unsigned long |
3365 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3366 | unsigned long max_load_move, struct sched_domain *sd, | |
3367 | enum cpu_idle_type idle, int *all_pinned, | |
3368 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 3369 | { |
051c6764 | 3370 | int loops = 0, pulled = 0, pinned = 0; |
dd41f596 IM |
3371 | struct task_struct *p; |
3372 | long rem_load_move = max_load_move; | |
1da177e4 | 3373 | |
e1d1484f | 3374 | if (max_load_move == 0) |
1da177e4 LT |
3375 | goto out; |
3376 | ||
81026794 NP |
3377 | pinned = 1; |
3378 | ||
1da177e4 | 3379 | /* |
dd41f596 | 3380 | * Start the load-balancing iterator: |
1da177e4 | 3381 | */ |
dd41f596 IM |
3382 | p = iterator->start(iterator->arg); |
3383 | next: | |
b82d9fdd | 3384 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 3385 | goto out; |
051c6764 PZ |
3386 | |
3387 | if ((p->se.load.weight >> 1) > rem_load_move || | |
dd41f596 | 3388 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
3389 | p = iterator->next(iterator->arg); |
3390 | goto next; | |
1da177e4 LT |
3391 | } |
3392 | ||
dd41f596 | 3393 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 3394 | pulled++; |
dd41f596 | 3395 | rem_load_move -= p->se.load.weight; |
1da177e4 | 3396 | |
7e96fa58 GH |
3397 | #ifdef CONFIG_PREEMPT |
3398 | /* | |
3399 | * NEWIDLE balancing is a source of latency, so preemptible kernels | |
3400 | * will stop after the first task is pulled to minimize the critical | |
3401 | * section. | |
3402 | */ | |
3403 | if (idle == CPU_NEWLY_IDLE) | |
3404 | goto out; | |
3405 | #endif | |
3406 | ||
2dd73a4f | 3407 | /* |
b82d9fdd | 3408 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 3409 | */ |
e1d1484f | 3410 | if (rem_load_move > 0) { |
a4ac01c3 PW |
3411 | if (p->prio < *this_best_prio) |
3412 | *this_best_prio = p->prio; | |
dd41f596 IM |
3413 | p = iterator->next(iterator->arg); |
3414 | goto next; | |
1da177e4 LT |
3415 | } |
3416 | out: | |
3417 | /* | |
e1d1484f | 3418 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
3419 | * so we can safely collect pull_task() stats here rather than |
3420 | * inside pull_task(). | |
3421 | */ | |
3422 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
3423 | |
3424 | if (all_pinned) | |
3425 | *all_pinned = pinned; | |
e1d1484f PW |
3426 | |
3427 | return max_load_move - rem_load_move; | |
1da177e4 LT |
3428 | } |
3429 | ||
dd41f596 | 3430 | /* |
43010659 PW |
3431 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
3432 | * this_rq, as part of a balancing operation within domain "sd". | |
3433 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
3434 | * |
3435 | * Called with both runqueues locked. | |
3436 | */ | |
3437 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 3438 | unsigned long max_load_move, |
dd41f596 IM |
3439 | struct sched_domain *sd, enum cpu_idle_type idle, |
3440 | int *all_pinned) | |
3441 | { | |
5522d5d5 | 3442 | const struct sched_class *class = sched_class_highest; |
43010659 | 3443 | unsigned long total_load_moved = 0; |
a4ac01c3 | 3444 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
3445 | |
3446 | do { | |
43010659 PW |
3447 | total_load_moved += |
3448 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 3449 | max_load_move - total_load_moved, |
a4ac01c3 | 3450 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 3451 | class = class->next; |
c4acb2c0 | 3452 | |
7e96fa58 GH |
3453 | #ifdef CONFIG_PREEMPT |
3454 | /* | |
3455 | * NEWIDLE balancing is a source of latency, so preemptible | |
3456 | * kernels will stop after the first task is pulled to minimize | |
3457 | * the critical section. | |
3458 | */ | |
c4acb2c0 GH |
3459 | if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) |
3460 | break; | |
7e96fa58 | 3461 | #endif |
43010659 | 3462 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 3463 | |
43010659 PW |
3464 | return total_load_moved > 0; |
3465 | } | |
3466 | ||
e1d1484f PW |
3467 | static int |
3468 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3469 | struct sched_domain *sd, enum cpu_idle_type idle, | |
3470 | struct rq_iterator *iterator) | |
3471 | { | |
3472 | struct task_struct *p = iterator->start(iterator->arg); | |
3473 | int pinned = 0; | |
3474 | ||
3475 | while (p) { | |
3476 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
3477 | pull_task(busiest, p, this_rq, this_cpu); | |
3478 | /* | |
3479 | * Right now, this is only the second place pull_task() | |
3480 | * is called, so we can safely collect pull_task() | |
3481 | * stats here rather than inside pull_task(). | |
3482 | */ | |
3483 | schedstat_inc(sd, lb_gained[idle]); | |
3484 | ||
3485 | return 1; | |
3486 | } | |
3487 | p = iterator->next(iterator->arg); | |
3488 | } | |
3489 | ||
3490 | return 0; | |
3491 | } | |
3492 | ||
43010659 PW |
3493 | /* |
3494 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
3495 | * part of active balancing operations within "domain". | |
3496 | * Returns 1 if successful and 0 otherwise. | |
3497 | * | |
3498 | * Called with both runqueues locked. | |
3499 | */ | |
3500 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
3501 | struct sched_domain *sd, enum cpu_idle_type idle) | |
3502 | { | |
5522d5d5 | 3503 | const struct sched_class *class; |
43010659 | 3504 | |
cde7e5ca | 3505 | for_each_class(class) { |
e1d1484f | 3506 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 | 3507 | return 1; |
cde7e5ca | 3508 | } |
43010659 PW |
3509 | |
3510 | return 0; | |
dd41f596 | 3511 | } |
67bb6c03 | 3512 | /********** Helpers for find_busiest_group ************************/ |
1da177e4 | 3513 | /* |
222d656d GS |
3514 | * sd_lb_stats - Structure to store the statistics of a sched_domain |
3515 | * during load balancing. | |
1da177e4 | 3516 | */ |
222d656d GS |
3517 | struct sd_lb_stats { |
3518 | struct sched_group *busiest; /* Busiest group in this sd */ | |
3519 | struct sched_group *this; /* Local group in this sd */ | |
3520 | unsigned long total_load; /* Total load of all groups in sd */ | |
3521 | unsigned long total_pwr; /* Total power of all groups in sd */ | |
3522 | unsigned long avg_load; /* Average load across all groups in sd */ | |
3523 | ||
3524 | /** Statistics of this group */ | |
3525 | unsigned long this_load; | |
3526 | unsigned long this_load_per_task; | |
3527 | unsigned long this_nr_running; | |
3528 | ||
3529 | /* Statistics of the busiest group */ | |
3530 | unsigned long max_load; | |
3531 | unsigned long busiest_load_per_task; | |
3532 | unsigned long busiest_nr_running; | |
3533 | ||
3534 | int group_imb; /* Is there imbalance in this sd */ | |
5c45bf27 | 3535 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
222d656d GS |
3536 | int power_savings_balance; /* Is powersave balance needed for this sd */ |
3537 | struct sched_group *group_min; /* Least loaded group in sd */ | |
3538 | struct sched_group *group_leader; /* Group which relieves group_min */ | |
3539 | unsigned long min_load_per_task; /* load_per_task in group_min */ | |
3540 | unsigned long leader_nr_running; /* Nr running of group_leader */ | |
3541 | unsigned long min_nr_running; /* Nr running of group_min */ | |
5c45bf27 | 3542 | #endif |
222d656d | 3543 | }; |
1da177e4 | 3544 | |
d5ac537e | 3545 | /* |
381be78f GS |
3546 | * sg_lb_stats - stats of a sched_group required for load_balancing |
3547 | */ | |
3548 | struct sg_lb_stats { | |
3549 | unsigned long avg_load; /*Avg load across the CPUs of the group */ | |
3550 | unsigned long group_load; /* Total load over the CPUs of the group */ | |
3551 | unsigned long sum_nr_running; /* Nr tasks running in the group */ | |
3552 | unsigned long sum_weighted_load; /* Weighted load of group's tasks */ | |
3553 | unsigned long group_capacity; | |
3554 | int group_imb; /* Is there an imbalance in the group ? */ | |
3555 | }; | |
408ed066 | 3556 | |
67bb6c03 GS |
3557 | /** |
3558 | * group_first_cpu - Returns the first cpu in the cpumask of a sched_group. | |
3559 | * @group: The group whose first cpu is to be returned. | |
3560 | */ | |
3561 | static inline unsigned int group_first_cpu(struct sched_group *group) | |
3562 | { | |
3563 | return cpumask_first(sched_group_cpus(group)); | |
3564 | } | |
3565 | ||
3566 | /** | |
3567 | * get_sd_load_idx - Obtain the load index for a given sched domain. | |
3568 | * @sd: The sched_domain whose load_idx is to be obtained. | |
3569 | * @idle: The Idle status of the CPU for whose sd load_icx is obtained. | |
3570 | */ | |
3571 | static inline int get_sd_load_idx(struct sched_domain *sd, | |
3572 | enum cpu_idle_type idle) | |
3573 | { | |
3574 | int load_idx; | |
3575 | ||
3576 | switch (idle) { | |
3577 | case CPU_NOT_IDLE: | |
7897986b | 3578 | load_idx = sd->busy_idx; |
67bb6c03 GS |
3579 | break; |
3580 | ||
3581 | case CPU_NEWLY_IDLE: | |
7897986b | 3582 | load_idx = sd->newidle_idx; |
67bb6c03 GS |
3583 | break; |
3584 | default: | |
7897986b | 3585 | load_idx = sd->idle_idx; |
67bb6c03 GS |
3586 | break; |
3587 | } | |
1da177e4 | 3588 | |
67bb6c03 GS |
3589 | return load_idx; |
3590 | } | |
1da177e4 | 3591 | |
1da177e4 | 3592 | |
c071df18 GS |
3593 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
3594 | /** | |
3595 | * init_sd_power_savings_stats - Initialize power savings statistics for | |
3596 | * the given sched_domain, during load balancing. | |
3597 | * | |
3598 | * @sd: Sched domain whose power-savings statistics are to be initialized. | |
3599 | * @sds: Variable containing the statistics for sd. | |
3600 | * @idle: Idle status of the CPU at which we're performing load-balancing. | |
3601 | */ | |
3602 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3603 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3604 | { | |
3605 | /* | |
3606 | * Busy processors will not participate in power savings | |
3607 | * balance. | |
3608 | */ | |
3609 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3610 | sds->power_savings_balance = 0; | |
3611 | else { | |
3612 | sds->power_savings_balance = 1; | |
3613 | sds->min_nr_running = ULONG_MAX; | |
3614 | sds->leader_nr_running = 0; | |
3615 | } | |
3616 | } | |
783609c6 | 3617 | |
c071df18 GS |
3618 | /** |
3619 | * update_sd_power_savings_stats - Update the power saving stats for a | |
3620 | * sched_domain while performing load balancing. | |
3621 | * | |
3622 | * @group: sched_group belonging to the sched_domain under consideration. | |
3623 | * @sds: Variable containing the statistics of the sched_domain | |
3624 | * @local_group: Does group contain the CPU for which we're performing | |
3625 | * load balancing ? | |
3626 | * @sgs: Variable containing the statistics of the group. | |
3627 | */ | |
3628 | static inline void update_sd_power_savings_stats(struct sched_group *group, | |
3629 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3630 | { | |
408ed066 | 3631 | |
c071df18 GS |
3632 | if (!sds->power_savings_balance) |
3633 | return; | |
1da177e4 | 3634 | |
c071df18 GS |
3635 | /* |
3636 | * If the local group is idle or completely loaded | |
3637 | * no need to do power savings balance at this domain | |
3638 | */ | |
3639 | if (local_group && (sds->this_nr_running >= sgs->group_capacity || | |
3640 | !sds->this_nr_running)) | |
3641 | sds->power_savings_balance = 0; | |
2dd73a4f | 3642 | |
c071df18 GS |
3643 | /* |
3644 | * If a group is already running at full capacity or idle, | |
3645 | * don't include that group in power savings calculations | |
3646 | */ | |
3647 | if (!sds->power_savings_balance || | |
3648 | sgs->sum_nr_running >= sgs->group_capacity || | |
3649 | !sgs->sum_nr_running) | |
3650 | return; | |
5969fe06 | 3651 | |
c071df18 GS |
3652 | /* |
3653 | * Calculate the group which has the least non-idle load. | |
3654 | * This is the group from where we need to pick up the load | |
3655 | * for saving power | |
3656 | */ | |
3657 | if ((sgs->sum_nr_running < sds->min_nr_running) || | |
3658 | (sgs->sum_nr_running == sds->min_nr_running && | |
3659 | group_first_cpu(group) > group_first_cpu(sds->group_min))) { | |
3660 | sds->group_min = group; | |
3661 | sds->min_nr_running = sgs->sum_nr_running; | |
3662 | sds->min_load_per_task = sgs->sum_weighted_load / | |
3663 | sgs->sum_nr_running; | |
3664 | } | |
783609c6 | 3665 | |
c071df18 GS |
3666 | /* |
3667 | * Calculate the group which is almost near its | |
3668 | * capacity but still has some space to pick up some load | |
3669 | * from other group and save more power | |
3670 | */ | |
3671 | if (sgs->sum_nr_running > sgs->group_capacity - 1) | |
3672 | return; | |
1da177e4 | 3673 | |
c071df18 GS |
3674 | if (sgs->sum_nr_running > sds->leader_nr_running || |
3675 | (sgs->sum_nr_running == sds->leader_nr_running && | |
3676 | group_first_cpu(group) < group_first_cpu(sds->group_leader))) { | |
3677 | sds->group_leader = group; | |
3678 | sds->leader_nr_running = sgs->sum_nr_running; | |
3679 | } | |
3680 | } | |
408ed066 | 3681 | |
c071df18 | 3682 | /** |
d5ac537e | 3683 | * check_power_save_busiest_group - see if there is potential for some power-savings balance |
c071df18 GS |
3684 | * @sds: Variable containing the statistics of the sched_domain |
3685 | * under consideration. | |
3686 | * @this_cpu: Cpu at which we're currently performing load-balancing. | |
3687 | * @imbalance: Variable to store the imbalance. | |
3688 | * | |
d5ac537e RD |
3689 | * Description: |
3690 | * Check if we have potential to perform some power-savings balance. | |
3691 | * If yes, set the busiest group to be the least loaded group in the | |
3692 | * sched_domain, so that it's CPUs can be put to idle. | |
3693 | * | |
c071df18 GS |
3694 | * Returns 1 if there is potential to perform power-savings balance. |
3695 | * Else returns 0. | |
3696 | */ | |
3697 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3698 | int this_cpu, unsigned long *imbalance) | |
3699 | { | |
3700 | if (!sds->power_savings_balance) | |
3701 | return 0; | |
1da177e4 | 3702 | |
c071df18 GS |
3703 | if (sds->this != sds->group_leader || |
3704 | sds->group_leader == sds->group_min) | |
3705 | return 0; | |
783609c6 | 3706 | |
c071df18 GS |
3707 | *imbalance = sds->min_load_per_task; |
3708 | sds->busiest = sds->group_min; | |
1da177e4 | 3709 | |
c071df18 GS |
3710 | if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { |
3711 | cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = | |
3712 | group_first_cpu(sds->group_leader); | |
3713 | } | |
3714 | ||
3715 | return 1; | |
1da177e4 | 3716 | |
c071df18 GS |
3717 | } |
3718 | #else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3719 | static inline void init_sd_power_savings_stats(struct sched_domain *sd, | |
3720 | struct sd_lb_stats *sds, enum cpu_idle_type idle) | |
3721 | { | |
3722 | return; | |
3723 | } | |
408ed066 | 3724 | |
c071df18 GS |
3725 | static inline void update_sd_power_savings_stats(struct sched_group *group, |
3726 | struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs) | |
3727 | { | |
3728 | return; | |
3729 | } | |
3730 | ||
3731 | static inline int check_power_save_busiest_group(struct sd_lb_stats *sds, | |
3732 | int this_cpu, unsigned long *imbalance) | |
3733 | { | |
3734 | return 0; | |
3735 | } | |
3736 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ | |
3737 | ||
e9e9250b | 3738 | unsigned long __weak arch_scale_smt_power(struct sched_domain *sd, int cpu) |
ab29230e PZ |
3739 | { |
3740 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3741 | unsigned long smt_gain = sd->smt_gain; | |
3742 | ||
3743 | smt_gain /= weight; | |
3744 | ||
3745 | return smt_gain; | |
3746 | } | |
3747 | ||
e9e9250b PZ |
3748 | unsigned long scale_rt_power(int cpu) |
3749 | { | |
3750 | struct rq *rq = cpu_rq(cpu); | |
3751 | u64 total, available; | |
3752 | ||
3753 | sched_avg_update(rq); | |
3754 | ||
3755 | total = sched_avg_period() + (rq->clock - rq->age_stamp); | |
3756 | available = total - rq->rt_avg; | |
3757 | ||
3758 | if (unlikely((s64)total < SCHED_LOAD_SCALE)) | |
3759 | total = SCHED_LOAD_SCALE; | |
3760 | ||
3761 | total >>= SCHED_LOAD_SHIFT; | |
3762 | ||
3763 | return div_u64(available, total); | |
3764 | } | |
3765 | ||
ab29230e PZ |
3766 | static void update_cpu_power(struct sched_domain *sd, int cpu) |
3767 | { | |
3768 | unsigned long weight = cpumask_weight(sched_domain_span(sd)); | |
3769 | unsigned long power = SCHED_LOAD_SCALE; | |
3770 | struct sched_group *sdg = sd->groups; | |
3771 | unsigned long old = sdg->__cpu_power; | |
3772 | ||
3773 | /* here we could scale based on cpufreq */ | |
3774 | ||
3775 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { | |
e9e9250b | 3776 | power *= arch_scale_smt_power(sd, cpu); |
ab29230e PZ |
3777 | power >>= SCHED_LOAD_SHIFT; |
3778 | } | |
3779 | ||
e9e9250b PZ |
3780 | power *= scale_rt_power(cpu); |
3781 | power >>= SCHED_LOAD_SHIFT; | |
3782 | ||
3783 | if (!power) | |
3784 | power = 1; | |
ab29230e PZ |
3785 | |
3786 | if (power != old) { | |
3787 | sdg->__cpu_power = power; | |
3788 | sdg->reciprocal_cpu_power = reciprocal_value(power); | |
3789 | } | |
3790 | } | |
3791 | ||
3792 | static void update_group_power(struct sched_domain *sd, int cpu) | |
cc9fba7d PZ |
3793 | { |
3794 | struct sched_domain *child = sd->child; | |
3795 | struct sched_group *group, *sdg = sd->groups; | |
3796 | unsigned long power = sdg->__cpu_power; | |
3797 | ||
3798 | if (!child) { | |
ab29230e | 3799 | update_cpu_power(sd, cpu); |
cc9fba7d PZ |
3800 | return; |
3801 | } | |
3802 | ||
3803 | sdg->__cpu_power = 0; | |
3804 | ||
3805 | group = child->groups; | |
3806 | do { | |
3807 | sdg->__cpu_power += group->__cpu_power; | |
3808 | group = group->next; | |
3809 | } while (group != child->groups); | |
3810 | ||
3811 | if (power != sdg->__cpu_power) | |
3812 | sdg->reciprocal_cpu_power = reciprocal_value(sdg->__cpu_power); | |
3813 | } | |
c071df18 | 3814 | |
1f8c553d GS |
3815 | /** |
3816 | * update_sg_lb_stats - Update sched_group's statistics for load balancing. | |
3817 | * @group: sched_group whose statistics are to be updated. | |
3818 | * @this_cpu: Cpu for which load balance is currently performed. | |
3819 | * @idle: Idle status of this_cpu | |
3820 | * @load_idx: Load index of sched_domain of this_cpu for load calc. | |
3821 | * @sd_idle: Idle status of the sched_domain containing group. | |
3822 | * @local_group: Does group contain this_cpu. | |
3823 | * @cpus: Set of cpus considered for load balancing. | |
3824 | * @balance: Should we balance. | |
3825 | * @sgs: variable to hold the statistics for this group. | |
3826 | */ | |
cc9fba7d PZ |
3827 | static inline void update_sg_lb_stats(struct sched_domain *sd, |
3828 | struct sched_group *group, int this_cpu, | |
1f8c553d GS |
3829 | enum cpu_idle_type idle, int load_idx, int *sd_idle, |
3830 | int local_group, const struct cpumask *cpus, | |
3831 | int *balance, struct sg_lb_stats *sgs) | |
3832 | { | |
3833 | unsigned long load, max_cpu_load, min_cpu_load; | |
3834 | int i; | |
3835 | unsigned int balance_cpu = -1, first_idle_cpu = 0; | |
3836 | unsigned long sum_avg_load_per_task; | |
3837 | unsigned long avg_load_per_task; | |
3838 | ||
cc9fba7d | 3839 | if (local_group) { |
1f8c553d | 3840 | balance_cpu = group_first_cpu(group); |
cc9fba7d | 3841 | if (balance_cpu == this_cpu) |
ab29230e | 3842 | update_group_power(sd, this_cpu); |
cc9fba7d | 3843 | } |
1f8c553d GS |
3844 | |
3845 | /* Tally up the load of all CPUs in the group */ | |
3846 | sum_avg_load_per_task = avg_load_per_task = 0; | |
3847 | max_cpu_load = 0; | |
3848 | min_cpu_load = ~0UL; | |
408ed066 | 3849 | |
1f8c553d GS |
3850 | for_each_cpu_and(i, sched_group_cpus(group), cpus) { |
3851 | struct rq *rq = cpu_rq(i); | |
908a7c1b | 3852 | |
1f8c553d GS |
3853 | if (*sd_idle && rq->nr_running) |
3854 | *sd_idle = 0; | |
5c45bf27 | 3855 | |
1f8c553d | 3856 | /* Bias balancing toward cpus of our domain */ |
1da177e4 | 3857 | if (local_group) { |
1f8c553d GS |
3858 | if (idle_cpu(i) && !first_idle_cpu) { |
3859 | first_idle_cpu = 1; | |
3860 | balance_cpu = i; | |
3861 | } | |
3862 | ||
3863 | load = target_load(i, load_idx); | |
3864 | } else { | |
3865 | load = source_load(i, load_idx); | |
3866 | if (load > max_cpu_load) | |
3867 | max_cpu_load = load; | |
3868 | if (min_cpu_load > load) | |
3869 | min_cpu_load = load; | |
1da177e4 | 3870 | } |
5c45bf27 | 3871 | |
1f8c553d GS |
3872 | sgs->group_load += load; |
3873 | sgs->sum_nr_running += rq->nr_running; | |
3874 | sgs->sum_weighted_load += weighted_cpuload(i); | |
5c45bf27 | 3875 | |
1f8c553d GS |
3876 | sum_avg_load_per_task += cpu_avg_load_per_task(i); |
3877 | } | |
5c45bf27 | 3878 | |
1f8c553d GS |
3879 | /* |
3880 | * First idle cpu or the first cpu(busiest) in this sched group | |
3881 | * is eligible for doing load balancing at this and above | |
3882 | * domains. In the newly idle case, we will allow all the cpu's | |
3883 | * to do the newly idle load balance. | |
3884 | */ | |
3885 | if (idle != CPU_NEWLY_IDLE && local_group && | |
3886 | balance_cpu != this_cpu && balance) { | |
3887 | *balance = 0; | |
3888 | return; | |
3889 | } | |
5c45bf27 | 3890 | |
1f8c553d GS |
3891 | /* Adjust by relative CPU power of the group */ |
3892 | sgs->avg_load = sg_div_cpu_power(group, | |
3893 | sgs->group_load * SCHED_LOAD_SCALE); | |
5c45bf27 | 3894 | |
1f8c553d GS |
3895 | |
3896 | /* | |
3897 | * Consider the group unbalanced when the imbalance is larger | |
3898 | * than the average weight of two tasks. | |
3899 | * | |
3900 | * APZ: with cgroup the avg task weight can vary wildly and | |
3901 | * might not be a suitable number - should we keep a | |
3902 | * normalized nr_running number somewhere that negates | |
3903 | * the hierarchy? | |
3904 | */ | |
3905 | avg_load_per_task = sg_div_cpu_power(group, | |
3906 | sum_avg_load_per_task * SCHED_LOAD_SCALE); | |
3907 | ||
3908 | if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) | |
3909 | sgs->group_imb = 1; | |
3910 | ||
bdb94aa5 PZ |
3911 | sgs->group_capacity = |
3912 | DIV_ROUND_CLOSEST(group->__cpu_power, SCHED_LOAD_SCALE); | |
1f8c553d | 3913 | } |
dd41f596 | 3914 | |
37abe198 GS |
3915 | /** |
3916 | * update_sd_lb_stats - Update sched_group's statistics for load balancing. | |
3917 | * @sd: sched_domain whose statistics are to be updated. | |
3918 | * @this_cpu: Cpu for which load balance is currently performed. | |
3919 | * @idle: Idle status of this_cpu | |
3920 | * @sd_idle: Idle status of the sched_domain containing group. | |
3921 | * @cpus: Set of cpus considered for load balancing. | |
3922 | * @balance: Should we balance. | |
3923 | * @sds: variable to hold the statistics for this sched_domain. | |
1da177e4 | 3924 | */ |
37abe198 GS |
3925 | static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu, |
3926 | enum cpu_idle_type idle, int *sd_idle, | |
3927 | const struct cpumask *cpus, int *balance, | |
3928 | struct sd_lb_stats *sds) | |
1da177e4 | 3929 | { |
b5d978e0 | 3930 | struct sched_domain *child = sd->child; |
222d656d | 3931 | struct sched_group *group = sd->groups; |
37abe198 | 3932 | struct sg_lb_stats sgs; |
b5d978e0 PZ |
3933 | int load_idx, prefer_sibling = 0; |
3934 | ||
3935 | if (child && child->flags & SD_PREFER_SIBLING) | |
3936 | prefer_sibling = 1; | |
222d656d | 3937 | |
c071df18 | 3938 | init_sd_power_savings_stats(sd, sds, idle); |
67bb6c03 | 3939 | load_idx = get_sd_load_idx(sd, idle); |
1da177e4 LT |
3940 | |
3941 | do { | |
1da177e4 | 3942 | int local_group; |
1da177e4 | 3943 | |
758b2cdc RR |
3944 | local_group = cpumask_test_cpu(this_cpu, |
3945 | sched_group_cpus(group)); | |
381be78f | 3946 | memset(&sgs, 0, sizeof(sgs)); |
cc9fba7d | 3947 | update_sg_lb_stats(sd, group, this_cpu, idle, load_idx, sd_idle, |
1f8c553d | 3948 | local_group, cpus, balance, &sgs); |
1da177e4 | 3949 | |
37abe198 GS |
3950 | if (local_group && balance && !(*balance)) |
3951 | return; | |
783609c6 | 3952 | |
37abe198 GS |
3953 | sds->total_load += sgs.group_load; |
3954 | sds->total_pwr += group->__cpu_power; | |
1da177e4 | 3955 | |
b5d978e0 PZ |
3956 | /* |
3957 | * In case the child domain prefers tasks go to siblings | |
3958 | * first, lower the group capacity to one so that we'll try | |
3959 | * and move all the excess tasks away. | |
3960 | */ | |
3961 | if (prefer_sibling) | |
bdb94aa5 | 3962 | sgs.group_capacity = min(sgs.group_capacity, 1UL); |
b5d978e0 | 3963 | |
1da177e4 | 3964 | if (local_group) { |
37abe198 GS |
3965 | sds->this_load = sgs.avg_load; |
3966 | sds->this = group; | |
3967 | sds->this_nr_running = sgs.sum_nr_running; | |
3968 | sds->this_load_per_task = sgs.sum_weighted_load; | |
3969 | } else if (sgs.avg_load > sds->max_load && | |
381be78f GS |
3970 | (sgs.sum_nr_running > sgs.group_capacity || |
3971 | sgs.group_imb)) { | |
37abe198 GS |
3972 | sds->max_load = sgs.avg_load; |
3973 | sds->busiest = group; | |
3974 | sds->busiest_nr_running = sgs.sum_nr_running; | |
3975 | sds->busiest_load_per_task = sgs.sum_weighted_load; | |
3976 | sds->group_imb = sgs.group_imb; | |
48f24c4d | 3977 | } |
5c45bf27 | 3978 | |
c071df18 | 3979 | update_sd_power_savings_stats(group, sds, local_group, &sgs); |
1da177e4 LT |
3980 | group = group->next; |
3981 | } while (group != sd->groups); | |
37abe198 | 3982 | } |
1da177e4 | 3983 | |
2e6f44ae GS |
3984 | /** |
3985 | * fix_small_imbalance - Calculate the minor imbalance that exists | |
dbc523a3 GS |
3986 | * amongst the groups of a sched_domain, during |
3987 | * load balancing. | |
2e6f44ae GS |
3988 | * @sds: Statistics of the sched_domain whose imbalance is to be calculated. |
3989 | * @this_cpu: The cpu at whose sched_domain we're performing load-balance. | |
3990 | * @imbalance: Variable to store the imbalance. | |
3991 | */ | |
3992 | static inline void fix_small_imbalance(struct sd_lb_stats *sds, | |
3993 | int this_cpu, unsigned long *imbalance) | |
3994 | { | |
3995 | unsigned long tmp, pwr_now = 0, pwr_move = 0; | |
3996 | unsigned int imbn = 2; | |
3997 | ||
3998 | if (sds->this_nr_running) { | |
3999 | sds->this_load_per_task /= sds->this_nr_running; | |
4000 | if (sds->busiest_load_per_task > | |
4001 | sds->this_load_per_task) | |
4002 | imbn = 1; | |
4003 | } else | |
4004 | sds->this_load_per_task = | |
4005 | cpu_avg_load_per_task(this_cpu); | |
1da177e4 | 4006 | |
2e6f44ae GS |
4007 | if (sds->max_load - sds->this_load + sds->busiest_load_per_task >= |
4008 | sds->busiest_load_per_task * imbn) { | |
4009 | *imbalance = sds->busiest_load_per_task; | |
4010 | return; | |
4011 | } | |
908a7c1b | 4012 | |
1da177e4 | 4013 | /* |
2e6f44ae GS |
4014 | * OK, we don't have enough imbalance to justify moving tasks, |
4015 | * however we may be able to increase total CPU power used by | |
4016 | * moving them. | |
1da177e4 | 4017 | */ |
2dd73a4f | 4018 | |
2e6f44ae GS |
4019 | pwr_now += sds->busiest->__cpu_power * |
4020 | min(sds->busiest_load_per_task, sds->max_load); | |
4021 | pwr_now += sds->this->__cpu_power * | |
4022 | min(sds->this_load_per_task, sds->this_load); | |
4023 | pwr_now /= SCHED_LOAD_SCALE; | |
4024 | ||
4025 | /* Amount of load we'd subtract */ | |
4026 | tmp = sg_div_cpu_power(sds->busiest, | |
4027 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
4028 | if (sds->max_load > tmp) | |
4029 | pwr_move += sds->busiest->__cpu_power * | |
4030 | min(sds->busiest_load_per_task, sds->max_load - tmp); | |
4031 | ||
4032 | /* Amount of load we'd add */ | |
4033 | if (sds->max_load * sds->busiest->__cpu_power < | |
4034 | sds->busiest_load_per_task * SCHED_LOAD_SCALE) | |
4035 | tmp = sg_div_cpu_power(sds->this, | |
4036 | sds->max_load * sds->busiest->__cpu_power); | |
4037 | else | |
4038 | tmp = sg_div_cpu_power(sds->this, | |
4039 | sds->busiest_load_per_task * SCHED_LOAD_SCALE); | |
4040 | pwr_move += sds->this->__cpu_power * | |
4041 | min(sds->this_load_per_task, sds->this_load + tmp); | |
4042 | pwr_move /= SCHED_LOAD_SCALE; | |
4043 | ||
4044 | /* Move if we gain throughput */ | |
4045 | if (pwr_move > pwr_now) | |
4046 | *imbalance = sds->busiest_load_per_task; | |
4047 | } | |
dbc523a3 GS |
4048 | |
4049 | /** | |
4050 | * calculate_imbalance - Calculate the amount of imbalance present within the | |
4051 | * groups of a given sched_domain during load balance. | |
4052 | * @sds: statistics of the sched_domain whose imbalance is to be calculated. | |
4053 | * @this_cpu: Cpu for which currently load balance is being performed. | |
4054 | * @imbalance: The variable to store the imbalance. | |
4055 | */ | |
4056 | static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu, | |
4057 | unsigned long *imbalance) | |
4058 | { | |
4059 | unsigned long max_pull; | |
2dd73a4f PW |
4060 | /* |
4061 | * In the presence of smp nice balancing, certain scenarios can have | |
4062 | * max load less than avg load(as we skip the groups at or below | |
4063 | * its cpu_power, while calculating max_load..) | |
4064 | */ | |
dbc523a3 | 4065 | if (sds->max_load < sds->avg_load) { |
2dd73a4f | 4066 | *imbalance = 0; |
dbc523a3 | 4067 | return fix_small_imbalance(sds, this_cpu, imbalance); |
2dd73a4f | 4068 | } |
0c117f1b SS |
4069 | |
4070 | /* Don't want to pull so many tasks that a group would go idle */ | |
dbc523a3 GS |
4071 | max_pull = min(sds->max_load - sds->avg_load, |
4072 | sds->max_load - sds->busiest_load_per_task); | |
0c117f1b | 4073 | |
1da177e4 | 4074 | /* How much load to actually move to equalise the imbalance */ |
dbc523a3 GS |
4075 | *imbalance = min(max_pull * sds->busiest->__cpu_power, |
4076 | (sds->avg_load - sds->this_load) * sds->this->__cpu_power) | |
1da177e4 LT |
4077 | / SCHED_LOAD_SCALE; |
4078 | ||
2dd73a4f PW |
4079 | /* |
4080 | * if *imbalance is less than the average load per runnable task | |
4081 | * there is no gaurantee that any tasks will be moved so we'll have | |
4082 | * a think about bumping its value to force at least one task to be | |
4083 | * moved | |
4084 | */ | |
dbc523a3 GS |
4085 | if (*imbalance < sds->busiest_load_per_task) |
4086 | return fix_small_imbalance(sds, this_cpu, imbalance); | |
1da177e4 | 4087 | |
dbc523a3 | 4088 | } |
37abe198 | 4089 | /******* find_busiest_group() helpers end here *********************/ |
1da177e4 | 4090 | |
b7bb4c9b GS |
4091 | /** |
4092 | * find_busiest_group - Returns the busiest group within the sched_domain | |
4093 | * if there is an imbalance. If there isn't an imbalance, and | |
4094 | * the user has opted for power-savings, it returns a group whose | |
4095 | * CPUs can be put to idle by rebalancing those tasks elsewhere, if | |
4096 | * such a group exists. | |
4097 | * | |
4098 | * Also calculates the amount of weighted load which should be moved | |
4099 | * to restore balance. | |
4100 | * | |
4101 | * @sd: The sched_domain whose busiest group is to be returned. | |
4102 | * @this_cpu: The cpu for which load balancing is currently being performed. | |
4103 | * @imbalance: Variable which stores amount of weighted load which should | |
4104 | * be moved to restore balance/put a group to idle. | |
4105 | * @idle: The idle status of this_cpu. | |
4106 | * @sd_idle: The idleness of sd | |
4107 | * @cpus: The set of CPUs under consideration for load-balancing. | |
4108 | * @balance: Pointer to a variable indicating if this_cpu | |
4109 | * is the appropriate cpu to perform load balancing at this_level. | |
4110 | * | |
4111 | * Returns: - the busiest group if imbalance exists. | |
4112 | * - If no imbalance and user has opted for power-savings balance, | |
4113 | * return the least loaded group whose CPUs can be | |
4114 | * put to idle by rebalancing its tasks onto our group. | |
37abe198 GS |
4115 | */ |
4116 | static struct sched_group * | |
4117 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
4118 | unsigned long *imbalance, enum cpu_idle_type idle, | |
4119 | int *sd_idle, const struct cpumask *cpus, int *balance) | |
4120 | { | |
4121 | struct sd_lb_stats sds; | |
1da177e4 | 4122 | |
37abe198 | 4123 | memset(&sds, 0, sizeof(sds)); |
1da177e4 | 4124 | |
37abe198 GS |
4125 | /* |
4126 | * Compute the various statistics relavent for load balancing at | |
4127 | * this level. | |
4128 | */ | |
4129 | update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus, | |
4130 | balance, &sds); | |
4131 | ||
b7bb4c9b GS |
4132 | /* Cases where imbalance does not exist from POV of this_cpu */ |
4133 | /* 1) this_cpu is not the appropriate cpu to perform load balancing | |
4134 | * at this level. | |
4135 | * 2) There is no busy sibling group to pull from. | |
4136 | * 3) This group is the busiest group. | |
4137 | * 4) This group is more busy than the avg busieness at this | |
4138 | * sched_domain. | |
4139 | * 5) The imbalance is within the specified limit. | |
4140 | * 6) Any rebalance would lead to ping-pong | |
4141 | */ | |
37abe198 GS |
4142 | if (balance && !(*balance)) |
4143 | goto ret; | |
1da177e4 | 4144 | |
b7bb4c9b GS |
4145 | if (!sds.busiest || sds.busiest_nr_running == 0) |
4146 | goto out_balanced; | |
1da177e4 | 4147 | |
b7bb4c9b | 4148 | if (sds.this_load >= sds.max_load) |
1da177e4 | 4149 | goto out_balanced; |
1da177e4 | 4150 | |
222d656d | 4151 | sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr; |
1da177e4 | 4152 | |
b7bb4c9b GS |
4153 | if (sds.this_load >= sds.avg_load) |
4154 | goto out_balanced; | |
4155 | ||
4156 | if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) | |
1da177e4 LT |
4157 | goto out_balanced; |
4158 | ||
222d656d GS |
4159 | sds.busiest_load_per_task /= sds.busiest_nr_running; |
4160 | if (sds.group_imb) | |
4161 | sds.busiest_load_per_task = | |
4162 | min(sds.busiest_load_per_task, sds.avg_load); | |
908a7c1b | 4163 | |
1da177e4 LT |
4164 | /* |
4165 | * We're trying to get all the cpus to the average_load, so we don't | |
4166 | * want to push ourselves above the average load, nor do we wish to | |
4167 | * reduce the max loaded cpu below the average load, as either of these | |
4168 | * actions would just result in more rebalancing later, and ping-pong | |
4169 | * tasks around. Thus we look for the minimum possible imbalance. | |
4170 | * Negative imbalances (*we* are more loaded than anyone else) will | |
4171 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 4172 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
4173 | * appear as very large values with unsigned longs. |
4174 | */ | |
222d656d | 4175 | if (sds.max_load <= sds.busiest_load_per_task) |
2dd73a4f PW |
4176 | goto out_balanced; |
4177 | ||
dbc523a3 GS |
4178 | /* Looks like there is an imbalance. Compute it */ |
4179 | calculate_imbalance(&sds, this_cpu, imbalance); | |
222d656d | 4180 | return sds.busiest; |
1da177e4 LT |
4181 | |
4182 | out_balanced: | |
c071df18 GS |
4183 | /* |
4184 | * There is no obvious imbalance. But check if we can do some balancing | |
4185 | * to save power. | |
4186 | */ | |
4187 | if (check_power_save_busiest_group(&sds, this_cpu, imbalance)) | |
4188 | return sds.busiest; | |
783609c6 | 4189 | ret: |
1da177e4 LT |
4190 | *imbalance = 0; |
4191 | return NULL; | |
4192 | } | |
4193 | ||
bdb94aa5 PZ |
4194 | static struct sched_group *group_of(int cpu) |
4195 | { | |
4196 | struct sched_domain *sd = rcu_dereference(cpu_rq(cpu)->sd); | |
4197 | ||
4198 | if (!sd) | |
4199 | return NULL; | |
4200 | ||
4201 | return sd->groups; | |
4202 | } | |
4203 | ||
4204 | static unsigned long power_of(int cpu) | |
4205 | { | |
4206 | struct sched_group *group = group_of(cpu); | |
4207 | ||
4208 | if (!group) | |
4209 | return SCHED_LOAD_SCALE; | |
4210 | ||
4211 | return group->__cpu_power; | |
4212 | } | |
4213 | ||
1da177e4 LT |
4214 | /* |
4215 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
4216 | */ | |
70b97a7f | 4217 | static struct rq * |
d15bcfdb | 4218 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
96f874e2 | 4219 | unsigned long imbalance, const struct cpumask *cpus) |
1da177e4 | 4220 | { |
70b97a7f | 4221 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 4222 | unsigned long max_load = 0; |
1da177e4 LT |
4223 | int i; |
4224 | ||
758b2cdc | 4225 | for_each_cpu(i, sched_group_cpus(group)) { |
bdb94aa5 PZ |
4226 | unsigned long power = power_of(i); |
4227 | unsigned long capacity = DIV_ROUND_CLOSEST(power, SCHED_LOAD_SCALE); | |
dd41f596 | 4228 | unsigned long wl; |
0a2966b4 | 4229 | |
96f874e2 | 4230 | if (!cpumask_test_cpu(i, cpus)) |
0a2966b4 CL |
4231 | continue; |
4232 | ||
48f24c4d | 4233 | rq = cpu_rq(i); |
bdb94aa5 PZ |
4234 | wl = weighted_cpuload(i) * SCHED_LOAD_SCALE; |
4235 | wl /= power; | |
2dd73a4f | 4236 | |
bdb94aa5 | 4237 | if (capacity && rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 4238 | continue; |
1da177e4 | 4239 | |
dd41f596 IM |
4240 | if (wl > max_load) { |
4241 | max_load = wl; | |
48f24c4d | 4242 | busiest = rq; |
1da177e4 LT |
4243 | } |
4244 | } | |
4245 | ||
4246 | return busiest; | |
4247 | } | |
4248 | ||
77391d71 NP |
4249 | /* |
4250 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
4251 | * so long as it is large enough. | |
4252 | */ | |
4253 | #define MAX_PINNED_INTERVAL 512 | |
4254 | ||
df7c8e84 RR |
4255 | /* Working cpumask for load_balance and load_balance_newidle. */ |
4256 | static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask); | |
4257 | ||
1da177e4 LT |
4258 | /* |
4259 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4260 | * tasks if there is an imbalance. | |
1da177e4 | 4261 | */ |
70b97a7f | 4262 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 4263 | struct sched_domain *sd, enum cpu_idle_type idle, |
df7c8e84 | 4264 | int *balance) |
1da177e4 | 4265 | { |
43010659 | 4266 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 4267 | struct sched_group *group; |
1da177e4 | 4268 | unsigned long imbalance; |
70b97a7f | 4269 | struct rq *busiest; |
fe2eea3f | 4270 | unsigned long flags; |
df7c8e84 | 4271 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
5969fe06 | 4272 | |
96f874e2 | 4273 | cpumask_setall(cpus); |
7c16ec58 | 4274 | |
89c4710e SS |
4275 | /* |
4276 | * When power savings policy is enabled for the parent domain, idle | |
4277 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 4278 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 4279 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 4280 | */ |
d15bcfdb | 4281 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4282 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4283 | sd_idle = 1; |
1da177e4 | 4284 | |
2d72376b | 4285 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 4286 | |
0a2966b4 | 4287 | redo: |
c8cba857 | 4288 | update_shares(sd); |
0a2966b4 | 4289 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, |
7c16ec58 | 4290 | cpus, balance); |
783609c6 | 4291 | |
06066714 | 4292 | if (*balance == 0) |
783609c6 | 4293 | goto out_balanced; |
783609c6 | 4294 | |
1da177e4 LT |
4295 | if (!group) { |
4296 | schedstat_inc(sd, lb_nobusyg[idle]); | |
4297 | goto out_balanced; | |
4298 | } | |
4299 | ||
7c16ec58 | 4300 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
4301 | if (!busiest) { |
4302 | schedstat_inc(sd, lb_nobusyq[idle]); | |
4303 | goto out_balanced; | |
4304 | } | |
4305 | ||
db935dbd | 4306 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
4307 | |
4308 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
4309 | ||
43010659 | 4310 | ld_moved = 0; |
1da177e4 LT |
4311 | if (busiest->nr_running > 1) { |
4312 | /* | |
4313 | * Attempt to move tasks. If find_busiest_group has found | |
4314 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 4315 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
4316 | * correctly treated as an imbalance. |
4317 | */ | |
fe2eea3f | 4318 | local_irq_save(flags); |
e17224bf | 4319 | double_rq_lock(this_rq, busiest); |
43010659 | 4320 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 4321 | imbalance, sd, idle, &all_pinned); |
e17224bf | 4322 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 4323 | local_irq_restore(flags); |
81026794 | 4324 | |
46cb4b7c SS |
4325 | /* |
4326 | * some other cpu did the load balance for us. | |
4327 | */ | |
43010659 | 4328 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
4329 | resched_cpu(this_cpu); |
4330 | ||
81026794 | 4331 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 4332 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4333 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4334 | if (!cpumask_empty(cpus)) | |
0a2966b4 | 4335 | goto redo; |
81026794 | 4336 | goto out_balanced; |
0a2966b4 | 4337 | } |
1da177e4 | 4338 | } |
81026794 | 4339 | |
43010659 | 4340 | if (!ld_moved) { |
1da177e4 LT |
4341 | schedstat_inc(sd, lb_failed[idle]); |
4342 | sd->nr_balance_failed++; | |
4343 | ||
4344 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 4345 | |
fe2eea3f | 4346 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
4347 | |
4348 | /* don't kick the migration_thread, if the curr | |
4349 | * task on busiest cpu can't be moved to this_cpu | |
4350 | */ | |
96f874e2 RR |
4351 | if (!cpumask_test_cpu(this_cpu, |
4352 | &busiest->curr->cpus_allowed)) { | |
fe2eea3f | 4353 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
4354 | all_pinned = 1; |
4355 | goto out_one_pinned; | |
4356 | } | |
4357 | ||
1da177e4 LT |
4358 | if (!busiest->active_balance) { |
4359 | busiest->active_balance = 1; | |
4360 | busiest->push_cpu = this_cpu; | |
81026794 | 4361 | active_balance = 1; |
1da177e4 | 4362 | } |
fe2eea3f | 4363 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 4364 | if (active_balance) |
1da177e4 LT |
4365 | wake_up_process(busiest->migration_thread); |
4366 | ||
4367 | /* | |
4368 | * We've kicked active balancing, reset the failure | |
4369 | * counter. | |
4370 | */ | |
39507451 | 4371 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 4372 | } |
81026794 | 4373 | } else |
1da177e4 LT |
4374 | sd->nr_balance_failed = 0; |
4375 | ||
81026794 | 4376 | if (likely(!active_balance)) { |
1da177e4 LT |
4377 | /* We were unbalanced, so reset the balancing interval */ |
4378 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
4379 | } else { |
4380 | /* | |
4381 | * If we've begun active balancing, start to back off. This | |
4382 | * case may not be covered by the all_pinned logic if there | |
4383 | * is only 1 task on the busy runqueue (because we don't call | |
4384 | * move_tasks). | |
4385 | */ | |
4386 | if (sd->balance_interval < sd->max_interval) | |
4387 | sd->balance_interval *= 2; | |
1da177e4 LT |
4388 | } |
4389 | ||
43010659 | 4390 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4391 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4392 | ld_moved = -1; |
4393 | ||
4394 | goto out; | |
1da177e4 LT |
4395 | |
4396 | out_balanced: | |
1da177e4 LT |
4397 | schedstat_inc(sd, lb_balanced[idle]); |
4398 | ||
16cfb1c0 | 4399 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
4400 | |
4401 | out_one_pinned: | |
1da177e4 | 4402 | /* tune up the balancing interval */ |
77391d71 NP |
4403 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
4404 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
4405 | sd->balance_interval *= 2; |
4406 | ||
48f24c4d | 4407 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4408 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
c09595f6 PZ |
4409 | ld_moved = -1; |
4410 | else | |
4411 | ld_moved = 0; | |
4412 | out: | |
c8cba857 PZ |
4413 | if (ld_moved) |
4414 | update_shares(sd); | |
c09595f6 | 4415 | return ld_moved; |
1da177e4 LT |
4416 | } |
4417 | ||
4418 | /* | |
4419 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
4420 | * tasks if there is an imbalance. | |
4421 | * | |
d15bcfdb | 4422 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
4423 | * this_rq is locked. |
4424 | */ | |
48f24c4d | 4425 | static int |
df7c8e84 | 4426 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd) |
1da177e4 LT |
4427 | { |
4428 | struct sched_group *group; | |
70b97a7f | 4429 | struct rq *busiest = NULL; |
1da177e4 | 4430 | unsigned long imbalance; |
43010659 | 4431 | int ld_moved = 0; |
5969fe06 | 4432 | int sd_idle = 0; |
969bb4e4 | 4433 | int all_pinned = 0; |
df7c8e84 | 4434 | struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask); |
7c16ec58 | 4435 | |
96f874e2 | 4436 | cpumask_setall(cpus); |
5969fe06 | 4437 | |
89c4710e SS |
4438 | /* |
4439 | * When power savings policy is enabled for the parent domain, idle | |
4440 | * sibling can pick up load irrespective of busy siblings. In this case, | |
4441 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 4442 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
4443 | */ |
4444 | if (sd->flags & SD_SHARE_CPUPOWER && | |
4445 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4446 | sd_idle = 1; |
1da177e4 | 4447 | |
2d72376b | 4448 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 4449 | redo: |
3e5459b4 | 4450 | update_shares_locked(this_rq, sd); |
d15bcfdb | 4451 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 4452 | &sd_idle, cpus, NULL); |
1da177e4 | 4453 | if (!group) { |
d15bcfdb | 4454 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4455 | goto out_balanced; |
1da177e4 LT |
4456 | } |
4457 | ||
7c16ec58 | 4458 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 4459 | if (!busiest) { |
d15bcfdb | 4460 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 4461 | goto out_balanced; |
1da177e4 LT |
4462 | } |
4463 | ||
db935dbd NP |
4464 | BUG_ON(busiest == this_rq); |
4465 | ||
d15bcfdb | 4466 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 4467 | |
43010659 | 4468 | ld_moved = 0; |
d6d5cfaf NP |
4469 | if (busiest->nr_running > 1) { |
4470 | /* Attempt to move tasks */ | |
4471 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
4472 | /* this_rq->clock is already updated */ |
4473 | update_rq_clock(busiest); | |
43010659 | 4474 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
4475 | imbalance, sd, CPU_NEWLY_IDLE, |
4476 | &all_pinned); | |
1b12bbc7 | 4477 | double_unlock_balance(this_rq, busiest); |
0a2966b4 | 4478 | |
969bb4e4 | 4479 | if (unlikely(all_pinned)) { |
96f874e2 RR |
4480 | cpumask_clear_cpu(cpu_of(busiest), cpus); |
4481 | if (!cpumask_empty(cpus)) | |
0a2966b4 CL |
4482 | goto redo; |
4483 | } | |
d6d5cfaf NP |
4484 | } |
4485 | ||
43010659 | 4486 | if (!ld_moved) { |
36dffab6 | 4487 | int active_balance = 0; |
ad273b32 | 4488 | |
d15bcfdb | 4489 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
4490 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
4491 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 4492 | return -1; |
ad273b32 VS |
4493 | |
4494 | if (sched_mc_power_savings < POWERSAVINGS_BALANCE_WAKEUP) | |
4495 | return -1; | |
4496 | ||
4497 | if (sd->nr_balance_failed++ < 2) | |
4498 | return -1; | |
4499 | ||
4500 | /* | |
4501 | * The only task running in a non-idle cpu can be moved to this | |
4502 | * cpu in an attempt to completely freeup the other CPU | |
4503 | * package. The same method used to move task in load_balance() | |
4504 | * have been extended for load_balance_newidle() to speedup | |
4505 | * consolidation at sched_mc=POWERSAVINGS_BALANCE_WAKEUP (2) | |
4506 | * | |
4507 | * The package power saving logic comes from | |
4508 | * find_busiest_group(). If there are no imbalance, then | |
4509 | * f_b_g() will return NULL. However when sched_mc={1,2} then | |
4510 | * f_b_g() will select a group from which a running task may be | |
4511 | * pulled to this cpu in order to make the other package idle. | |
4512 | * If there is no opportunity to make a package idle and if | |
4513 | * there are no imbalance, then f_b_g() will return NULL and no | |
4514 | * action will be taken in load_balance_newidle(). | |
4515 | * | |
4516 | * Under normal task pull operation due to imbalance, there | |
4517 | * will be more than one task in the source run queue and | |
4518 | * move_tasks() will succeed. ld_moved will be true and this | |
4519 | * active balance code will not be triggered. | |
4520 | */ | |
4521 | ||
4522 | /* Lock busiest in correct order while this_rq is held */ | |
4523 | double_lock_balance(this_rq, busiest); | |
4524 | ||
4525 | /* | |
4526 | * don't kick the migration_thread, if the curr | |
4527 | * task on busiest cpu can't be moved to this_cpu | |
4528 | */ | |
6ca09dfc | 4529 | if (!cpumask_test_cpu(this_cpu, &busiest->curr->cpus_allowed)) { |
ad273b32 VS |
4530 | double_unlock_balance(this_rq, busiest); |
4531 | all_pinned = 1; | |
4532 | return ld_moved; | |
4533 | } | |
4534 | ||
4535 | if (!busiest->active_balance) { | |
4536 | busiest->active_balance = 1; | |
4537 | busiest->push_cpu = this_cpu; | |
4538 | active_balance = 1; | |
4539 | } | |
4540 | ||
4541 | double_unlock_balance(this_rq, busiest); | |
da8d5089 PZ |
4542 | /* |
4543 | * Should not call ttwu while holding a rq->lock | |
4544 | */ | |
4545 | spin_unlock(&this_rq->lock); | |
ad273b32 VS |
4546 | if (active_balance) |
4547 | wake_up_process(busiest->migration_thread); | |
da8d5089 | 4548 | spin_lock(&this_rq->lock); |
ad273b32 | 4549 | |
5969fe06 | 4550 | } else |
16cfb1c0 | 4551 | sd->nr_balance_failed = 0; |
1da177e4 | 4552 | |
3e5459b4 | 4553 | update_shares_locked(this_rq, sd); |
43010659 | 4554 | return ld_moved; |
16cfb1c0 NP |
4555 | |
4556 | out_balanced: | |
d15bcfdb | 4557 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 4558 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 4559 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 4560 | return -1; |
16cfb1c0 | 4561 | sd->nr_balance_failed = 0; |
48f24c4d | 4562 | |
16cfb1c0 | 4563 | return 0; |
1da177e4 LT |
4564 | } |
4565 | ||
4566 | /* | |
4567 | * idle_balance is called by schedule() if this_cpu is about to become | |
4568 | * idle. Attempts to pull tasks from other CPUs. | |
4569 | */ | |
70b97a7f | 4570 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
4571 | { |
4572 | struct sched_domain *sd; | |
efbe027e | 4573 | int pulled_task = 0; |
dd41f596 | 4574 | unsigned long next_balance = jiffies + HZ; |
1da177e4 LT |
4575 | |
4576 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
4577 | unsigned long interval; |
4578 | ||
4579 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
4580 | continue; | |
4581 | ||
4582 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 4583 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 | 4584 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
df7c8e84 | 4585 | sd); |
92c4ca5c CL |
4586 | |
4587 | interval = msecs_to_jiffies(sd->balance_interval); | |
4588 | if (time_after(next_balance, sd->last_balance + interval)) | |
4589 | next_balance = sd->last_balance + interval; | |
4590 | if (pulled_task) | |
4591 | break; | |
1da177e4 | 4592 | } |
dd41f596 | 4593 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
4594 | /* |
4595 | * We are going idle. next_balance may be set based on | |
4596 | * a busy processor. So reset next_balance. | |
4597 | */ | |
4598 | this_rq->next_balance = next_balance; | |
dd41f596 | 4599 | } |
1da177e4 LT |
4600 | } |
4601 | ||
4602 | /* | |
4603 | * active_load_balance is run by migration threads. It pushes running tasks | |
4604 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
4605 | * running on each physical CPU where possible, and avoids physical / | |
4606 | * logical imbalances. | |
4607 | * | |
4608 | * Called with busiest_rq locked. | |
4609 | */ | |
70b97a7f | 4610 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 4611 | { |
39507451 | 4612 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
4613 | struct sched_domain *sd; |
4614 | struct rq *target_rq; | |
39507451 | 4615 | |
48f24c4d | 4616 | /* Is there any task to move? */ |
39507451 | 4617 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
4618 | return; |
4619 | ||
4620 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
4621 | |
4622 | /* | |
39507451 | 4623 | * This condition is "impossible", if it occurs |
41a2d6cf | 4624 | * we need to fix it. Originally reported by |
39507451 | 4625 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 4626 | */ |
39507451 | 4627 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 4628 | |
39507451 NP |
4629 | /* move a task from busiest_rq to target_rq */ |
4630 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
4631 | update_rq_clock(busiest_rq); |
4632 | update_rq_clock(target_rq); | |
39507451 NP |
4633 | |
4634 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 4635 | for_each_domain(target_cpu, sd) { |
39507451 | 4636 | if ((sd->flags & SD_LOAD_BALANCE) && |
758b2cdc | 4637 | cpumask_test_cpu(busiest_cpu, sched_domain_span(sd))) |
39507451 | 4638 | break; |
c96d145e | 4639 | } |
39507451 | 4640 | |
48f24c4d | 4641 | if (likely(sd)) { |
2d72376b | 4642 | schedstat_inc(sd, alb_count); |
39507451 | 4643 | |
43010659 PW |
4644 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
4645 | sd, CPU_IDLE)) | |
48f24c4d IM |
4646 | schedstat_inc(sd, alb_pushed); |
4647 | else | |
4648 | schedstat_inc(sd, alb_failed); | |
4649 | } | |
1b12bbc7 | 4650 | double_unlock_balance(busiest_rq, target_rq); |
1da177e4 LT |
4651 | } |
4652 | ||
46cb4b7c SS |
4653 | #ifdef CONFIG_NO_HZ |
4654 | static struct { | |
4655 | atomic_t load_balancer; | |
7d1e6a9b | 4656 | cpumask_var_t cpu_mask; |
f711f609 | 4657 | cpumask_var_t ilb_grp_nohz_mask; |
46cb4b7c SS |
4658 | } nohz ____cacheline_aligned = { |
4659 | .load_balancer = ATOMIC_INIT(-1), | |
46cb4b7c SS |
4660 | }; |
4661 | ||
eea08f32 AB |
4662 | int get_nohz_load_balancer(void) |
4663 | { | |
4664 | return atomic_read(&nohz.load_balancer); | |
4665 | } | |
4666 | ||
f711f609 GS |
4667 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
4668 | /** | |
4669 | * lowest_flag_domain - Return lowest sched_domain containing flag. | |
4670 | * @cpu: The cpu whose lowest level of sched domain is to | |
4671 | * be returned. | |
4672 | * @flag: The flag to check for the lowest sched_domain | |
4673 | * for the given cpu. | |
4674 | * | |
4675 | * Returns the lowest sched_domain of a cpu which contains the given flag. | |
4676 | */ | |
4677 | static inline struct sched_domain *lowest_flag_domain(int cpu, int flag) | |
4678 | { | |
4679 | struct sched_domain *sd; | |
4680 | ||
4681 | for_each_domain(cpu, sd) | |
4682 | if (sd && (sd->flags & flag)) | |
4683 | break; | |
4684 | ||
4685 | return sd; | |
4686 | } | |
4687 | ||
4688 | /** | |
4689 | * for_each_flag_domain - Iterates over sched_domains containing the flag. | |
4690 | * @cpu: The cpu whose domains we're iterating over. | |
4691 | * @sd: variable holding the value of the power_savings_sd | |
4692 | * for cpu. | |
4693 | * @flag: The flag to filter the sched_domains to be iterated. | |
4694 | * | |
4695 | * Iterates over all the scheduler domains for a given cpu that has the 'flag' | |
4696 | * set, starting from the lowest sched_domain to the highest. | |
4697 | */ | |
4698 | #define for_each_flag_domain(cpu, sd, flag) \ | |
4699 | for (sd = lowest_flag_domain(cpu, flag); \ | |
4700 | (sd && (sd->flags & flag)); sd = sd->parent) | |
4701 | ||
4702 | /** | |
4703 | * is_semi_idle_group - Checks if the given sched_group is semi-idle. | |
4704 | * @ilb_group: group to be checked for semi-idleness | |
4705 | * | |
4706 | * Returns: 1 if the group is semi-idle. 0 otherwise. | |
4707 | * | |
4708 | * We define a sched_group to be semi idle if it has atleast one idle-CPU | |
4709 | * and atleast one non-idle CPU. This helper function checks if the given | |
4710 | * sched_group is semi-idle or not. | |
4711 | */ | |
4712 | static inline int is_semi_idle_group(struct sched_group *ilb_group) | |
4713 | { | |
4714 | cpumask_and(nohz.ilb_grp_nohz_mask, nohz.cpu_mask, | |
4715 | sched_group_cpus(ilb_group)); | |
4716 | ||
4717 | /* | |
4718 | * A sched_group is semi-idle when it has atleast one busy cpu | |
4719 | * and atleast one idle cpu. | |
4720 | */ | |
4721 | if (cpumask_empty(nohz.ilb_grp_nohz_mask)) | |
4722 | return 0; | |
4723 | ||
4724 | if (cpumask_equal(nohz.ilb_grp_nohz_mask, sched_group_cpus(ilb_group))) | |
4725 | return 0; | |
4726 | ||
4727 | return 1; | |
4728 | } | |
4729 | /** | |
4730 | * find_new_ilb - Finds the optimum idle load balancer for nomination. | |
4731 | * @cpu: The cpu which is nominating a new idle_load_balancer. | |
4732 | * | |
4733 | * Returns: Returns the id of the idle load balancer if it exists, | |
4734 | * Else, returns >= nr_cpu_ids. | |
4735 | * | |
4736 | * This algorithm picks the idle load balancer such that it belongs to a | |
4737 | * semi-idle powersavings sched_domain. The idea is to try and avoid | |
4738 | * completely idle packages/cores just for the purpose of idle load balancing | |
4739 | * when there are other idle cpu's which are better suited for that job. | |
4740 | */ | |
4741 | static int find_new_ilb(int cpu) | |
4742 | { | |
4743 | struct sched_domain *sd; | |
4744 | struct sched_group *ilb_group; | |
4745 | ||
4746 | /* | |
4747 | * Have idle load balancer selection from semi-idle packages only | |
4748 | * when power-aware load balancing is enabled | |
4749 | */ | |
4750 | if (!(sched_smt_power_savings || sched_mc_power_savings)) | |
4751 | goto out_done; | |
4752 | ||
4753 | /* | |
4754 | * Optimize for the case when we have no idle CPUs or only one | |
4755 | * idle CPU. Don't walk the sched_domain hierarchy in such cases | |
4756 | */ | |
4757 | if (cpumask_weight(nohz.cpu_mask) < 2) | |
4758 | goto out_done; | |
4759 | ||
4760 | for_each_flag_domain(cpu, sd, SD_POWERSAVINGS_BALANCE) { | |
4761 | ilb_group = sd->groups; | |
4762 | ||
4763 | do { | |
4764 | if (is_semi_idle_group(ilb_group)) | |
4765 | return cpumask_first(nohz.ilb_grp_nohz_mask); | |
4766 | ||
4767 | ilb_group = ilb_group->next; | |
4768 | ||
4769 | } while (ilb_group != sd->groups); | |
4770 | } | |
4771 | ||
4772 | out_done: | |
4773 | return cpumask_first(nohz.cpu_mask); | |
4774 | } | |
4775 | #else /* (CONFIG_SCHED_MC || CONFIG_SCHED_SMT) */ | |
4776 | static inline int find_new_ilb(int call_cpu) | |
4777 | { | |
6e29ec57 | 4778 | return cpumask_first(nohz.cpu_mask); |
f711f609 GS |
4779 | } |
4780 | #endif | |
4781 | ||
7835b98b | 4782 | /* |
46cb4b7c SS |
4783 | * This routine will try to nominate the ilb (idle load balancing) |
4784 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
4785 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
4786 | * go into this tickless mode, then there will be no ilb owner (as there is | |
4787 | * no need for one) and all the cpus will sleep till the next wakeup event | |
4788 | * arrives... | |
4789 | * | |
4790 | * For the ilb owner, tick is not stopped. And this tick will be used | |
4791 | * for idle load balancing. ilb owner will still be part of | |
4792 | * nohz.cpu_mask.. | |
7835b98b | 4793 | * |
46cb4b7c SS |
4794 | * While stopping the tick, this cpu will become the ilb owner if there |
4795 | * is no other owner. And will be the owner till that cpu becomes busy | |
4796 | * or if all cpus in the system stop their ticks at which point | |
4797 | * there is no need for ilb owner. | |
4798 | * | |
4799 | * When the ilb owner becomes busy, it nominates another owner, during the | |
4800 | * next busy scheduler_tick() | |
4801 | */ | |
4802 | int select_nohz_load_balancer(int stop_tick) | |
4803 | { | |
4804 | int cpu = smp_processor_id(); | |
4805 | ||
4806 | if (stop_tick) { | |
46cb4b7c SS |
4807 | cpu_rq(cpu)->in_nohz_recently = 1; |
4808 | ||
483b4ee6 SS |
4809 | if (!cpu_active(cpu)) { |
4810 | if (atomic_read(&nohz.load_balancer) != cpu) | |
4811 | return 0; | |
4812 | ||
4813 | /* | |
4814 | * If we are going offline and still the leader, | |
4815 | * give up! | |
4816 | */ | |
46cb4b7c SS |
4817 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) |
4818 | BUG(); | |
483b4ee6 | 4819 | |
46cb4b7c SS |
4820 | return 0; |
4821 | } | |
4822 | ||
483b4ee6 SS |
4823 | cpumask_set_cpu(cpu, nohz.cpu_mask); |
4824 | ||
46cb4b7c | 4825 | /* time for ilb owner also to sleep */ |
7d1e6a9b | 4826 | if (cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
4827 | if (atomic_read(&nohz.load_balancer) == cpu) |
4828 | atomic_set(&nohz.load_balancer, -1); | |
4829 | return 0; | |
4830 | } | |
4831 | ||
4832 | if (atomic_read(&nohz.load_balancer) == -1) { | |
4833 | /* make me the ilb owner */ | |
4834 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
4835 | return 1; | |
e790fb0b GS |
4836 | } else if (atomic_read(&nohz.load_balancer) == cpu) { |
4837 | int new_ilb; | |
4838 | ||
4839 | if (!(sched_smt_power_savings || | |
4840 | sched_mc_power_savings)) | |
4841 | return 1; | |
4842 | /* | |
4843 | * Check to see if there is a more power-efficient | |
4844 | * ilb. | |
4845 | */ | |
4846 | new_ilb = find_new_ilb(cpu); | |
4847 | if (new_ilb < nr_cpu_ids && new_ilb != cpu) { | |
4848 | atomic_set(&nohz.load_balancer, -1); | |
4849 | resched_cpu(new_ilb); | |
4850 | return 0; | |
4851 | } | |
46cb4b7c | 4852 | return 1; |
e790fb0b | 4853 | } |
46cb4b7c | 4854 | } else { |
7d1e6a9b | 4855 | if (!cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
4856 | return 0; |
4857 | ||
7d1e6a9b | 4858 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
4859 | |
4860 | if (atomic_read(&nohz.load_balancer) == cpu) | |
4861 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
4862 | BUG(); | |
4863 | } | |
4864 | return 0; | |
4865 | } | |
4866 | #endif | |
4867 | ||
4868 | static DEFINE_SPINLOCK(balancing); | |
4869 | ||
4870 | /* | |
7835b98b CL |
4871 | * It checks each scheduling domain to see if it is due to be balanced, |
4872 | * and initiates a balancing operation if so. | |
4873 | * | |
4874 | * Balancing parameters are set up in arch_init_sched_domains. | |
4875 | */ | |
a9957449 | 4876 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 4877 | { |
46cb4b7c SS |
4878 | int balance = 1; |
4879 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
4880 | unsigned long interval; |
4881 | struct sched_domain *sd; | |
46cb4b7c | 4882 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 4883 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 4884 | int update_next_balance = 0; |
d07355f5 | 4885 | int need_serialize; |
1da177e4 | 4886 | |
46cb4b7c | 4887 | for_each_domain(cpu, sd) { |
1da177e4 LT |
4888 | if (!(sd->flags & SD_LOAD_BALANCE)) |
4889 | continue; | |
4890 | ||
4891 | interval = sd->balance_interval; | |
d15bcfdb | 4892 | if (idle != CPU_IDLE) |
1da177e4 LT |
4893 | interval *= sd->busy_factor; |
4894 | ||
4895 | /* scale ms to jiffies */ | |
4896 | interval = msecs_to_jiffies(interval); | |
4897 | if (unlikely(!interval)) | |
4898 | interval = 1; | |
dd41f596 IM |
4899 | if (interval > HZ*NR_CPUS/10) |
4900 | interval = HZ*NR_CPUS/10; | |
4901 | ||
d07355f5 | 4902 | need_serialize = sd->flags & SD_SERIALIZE; |
1da177e4 | 4903 | |
d07355f5 | 4904 | if (need_serialize) { |
08c183f3 CL |
4905 | if (!spin_trylock(&balancing)) |
4906 | goto out; | |
4907 | } | |
4908 | ||
c9819f45 | 4909 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
df7c8e84 | 4910 | if (load_balance(cpu, rq, sd, idle, &balance)) { |
fa3b6ddc SS |
4911 | /* |
4912 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
4913 | * longer idle, or one of our SMT siblings is |
4914 | * not idle. | |
4915 | */ | |
d15bcfdb | 4916 | idle = CPU_NOT_IDLE; |
1da177e4 | 4917 | } |
1bd77f2d | 4918 | sd->last_balance = jiffies; |
1da177e4 | 4919 | } |
d07355f5 | 4920 | if (need_serialize) |
08c183f3 CL |
4921 | spin_unlock(&balancing); |
4922 | out: | |
f549da84 | 4923 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 4924 | next_balance = sd->last_balance + interval; |
f549da84 SS |
4925 | update_next_balance = 1; |
4926 | } | |
783609c6 SS |
4927 | |
4928 | /* | |
4929 | * Stop the load balance at this level. There is another | |
4930 | * CPU in our sched group which is doing load balancing more | |
4931 | * actively. | |
4932 | */ | |
4933 | if (!balance) | |
4934 | break; | |
1da177e4 | 4935 | } |
f549da84 SS |
4936 | |
4937 | /* | |
4938 | * next_balance will be updated only when there is a need. | |
4939 | * When the cpu is attached to null domain for ex, it will not be | |
4940 | * updated. | |
4941 | */ | |
4942 | if (likely(update_next_balance)) | |
4943 | rq->next_balance = next_balance; | |
46cb4b7c SS |
4944 | } |
4945 | ||
4946 | /* | |
4947 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
4948 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
4949 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
4950 | */ | |
4951 | static void run_rebalance_domains(struct softirq_action *h) | |
4952 | { | |
dd41f596 IM |
4953 | int this_cpu = smp_processor_id(); |
4954 | struct rq *this_rq = cpu_rq(this_cpu); | |
4955 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
4956 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 4957 | |
dd41f596 | 4958 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
4959 | |
4960 | #ifdef CONFIG_NO_HZ | |
4961 | /* | |
4962 | * If this cpu is the owner for idle load balancing, then do the | |
4963 | * balancing on behalf of the other idle cpus whose ticks are | |
4964 | * stopped. | |
4965 | */ | |
dd41f596 IM |
4966 | if (this_rq->idle_at_tick && |
4967 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
4968 | struct rq *rq; |
4969 | int balance_cpu; | |
4970 | ||
7d1e6a9b RR |
4971 | for_each_cpu(balance_cpu, nohz.cpu_mask) { |
4972 | if (balance_cpu == this_cpu) | |
4973 | continue; | |
4974 | ||
46cb4b7c SS |
4975 | /* |
4976 | * If this cpu gets work to do, stop the load balancing | |
4977 | * work being done for other cpus. Next load | |
4978 | * balancing owner will pick it up. | |
4979 | */ | |
4980 | if (need_resched()) | |
4981 | break; | |
4982 | ||
de0cf899 | 4983 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
4984 | |
4985 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
4986 | if (time_after(this_rq->next_balance, rq->next_balance)) |
4987 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
4988 | } |
4989 | } | |
4990 | #endif | |
4991 | } | |
4992 | ||
8a0be9ef FW |
4993 | static inline int on_null_domain(int cpu) |
4994 | { | |
4995 | return !rcu_dereference(cpu_rq(cpu)->sd); | |
4996 | } | |
4997 | ||
46cb4b7c SS |
4998 | /* |
4999 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
5000 | * | |
5001 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
5002 | * idle load balancing owner or decide to stop the periodic load balancing, | |
5003 | * if the whole system is idle. | |
5004 | */ | |
dd41f596 | 5005 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 5006 | { |
46cb4b7c SS |
5007 | #ifdef CONFIG_NO_HZ |
5008 | /* | |
5009 | * If we were in the nohz mode recently and busy at the current | |
5010 | * scheduler tick, then check if we need to nominate new idle | |
5011 | * load balancer. | |
5012 | */ | |
5013 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
5014 | rq->in_nohz_recently = 0; | |
5015 | ||
5016 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
7d1e6a9b | 5017 | cpumask_clear_cpu(cpu, nohz.cpu_mask); |
46cb4b7c SS |
5018 | atomic_set(&nohz.load_balancer, -1); |
5019 | } | |
5020 | ||
5021 | if (atomic_read(&nohz.load_balancer) == -1) { | |
f711f609 | 5022 | int ilb = find_new_ilb(cpu); |
46cb4b7c | 5023 | |
434d53b0 | 5024 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
5025 | resched_cpu(ilb); |
5026 | } | |
5027 | } | |
5028 | ||
5029 | /* | |
5030 | * If this cpu is idle and doing idle load balancing for all the | |
5031 | * cpus with ticks stopped, is it time for that to stop? | |
5032 | */ | |
5033 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
7d1e6a9b | 5034 | cpumask_weight(nohz.cpu_mask) == num_online_cpus()) { |
46cb4b7c SS |
5035 | resched_cpu(cpu); |
5036 | return; | |
5037 | } | |
5038 | ||
5039 | /* | |
5040 | * If this cpu is idle and the idle load balancing is done by | |
5041 | * someone else, then no need raise the SCHED_SOFTIRQ | |
5042 | */ | |
5043 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
7d1e6a9b | 5044 | cpumask_test_cpu(cpu, nohz.cpu_mask)) |
46cb4b7c SS |
5045 | return; |
5046 | #endif | |
8a0be9ef FW |
5047 | /* Don't need to rebalance while attached to NULL domain */ |
5048 | if (time_after_eq(jiffies, rq->next_balance) && | |
5049 | likely(!on_null_domain(cpu))) | |
46cb4b7c | 5050 | raise_softirq(SCHED_SOFTIRQ); |
1da177e4 | 5051 | } |
dd41f596 IM |
5052 | |
5053 | #else /* CONFIG_SMP */ | |
5054 | ||
1da177e4 LT |
5055 | /* |
5056 | * on UP we do not need to balance between CPUs: | |
5057 | */ | |
70b97a7f | 5058 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
5059 | { |
5060 | } | |
dd41f596 | 5061 | |
1da177e4 LT |
5062 | #endif |
5063 | ||
1da177e4 LT |
5064 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
5065 | ||
5066 | EXPORT_PER_CPU_SYMBOL(kstat); | |
5067 | ||
5068 | /* | |
c5f8d995 | 5069 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 5070 | * @p in case that task is currently running. |
c5f8d995 HS |
5071 | * |
5072 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 5073 | */ |
c5f8d995 HS |
5074 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
5075 | { | |
5076 | u64 ns = 0; | |
5077 | ||
5078 | if (task_current(rq, p)) { | |
5079 | update_rq_clock(rq); | |
5080 | ns = rq->clock - p->se.exec_start; | |
5081 | if ((s64)ns < 0) | |
5082 | ns = 0; | |
5083 | } | |
5084 | ||
5085 | return ns; | |
5086 | } | |
5087 | ||
bb34d92f | 5088 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 5089 | { |
1da177e4 | 5090 | unsigned long flags; |
41b86e9c | 5091 | struct rq *rq; |
bb34d92f | 5092 | u64 ns = 0; |
48f24c4d | 5093 | |
41b86e9c | 5094 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
5095 | ns = do_task_delta_exec(p, rq); |
5096 | task_rq_unlock(rq, &flags); | |
1508487e | 5097 | |
c5f8d995 HS |
5098 | return ns; |
5099 | } | |
f06febc9 | 5100 | |
c5f8d995 HS |
5101 | /* |
5102 | * Return accounted runtime for the task. | |
5103 | * In case the task is currently running, return the runtime plus current's | |
5104 | * pending runtime that have not been accounted yet. | |
5105 | */ | |
5106 | unsigned long long task_sched_runtime(struct task_struct *p) | |
5107 | { | |
5108 | unsigned long flags; | |
5109 | struct rq *rq; | |
5110 | u64 ns = 0; | |
5111 | ||
5112 | rq = task_rq_lock(p, &flags); | |
5113 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
5114 | task_rq_unlock(rq, &flags); | |
5115 | ||
5116 | return ns; | |
5117 | } | |
48f24c4d | 5118 | |
c5f8d995 HS |
5119 | /* |
5120 | * Return sum_exec_runtime for the thread group. | |
5121 | * In case the task is currently running, return the sum plus current's | |
5122 | * pending runtime that have not been accounted yet. | |
5123 | * | |
5124 | * Note that the thread group might have other running tasks as well, | |
5125 | * so the return value not includes other pending runtime that other | |
5126 | * running tasks might have. | |
5127 | */ | |
5128 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
5129 | { | |
5130 | struct task_cputime totals; | |
5131 | unsigned long flags; | |
5132 | struct rq *rq; | |
5133 | u64 ns; | |
5134 | ||
5135 | rq = task_rq_lock(p, &flags); | |
5136 | thread_group_cputime(p, &totals); | |
5137 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 5138 | task_rq_unlock(rq, &flags); |
48f24c4d | 5139 | |
1da177e4 LT |
5140 | return ns; |
5141 | } | |
5142 | ||
1da177e4 LT |
5143 | /* |
5144 | * Account user cpu time to a process. | |
5145 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 5146 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 5147 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 5148 | */ |
457533a7 MS |
5149 | void account_user_time(struct task_struct *p, cputime_t cputime, |
5150 | cputime_t cputime_scaled) | |
1da177e4 LT |
5151 | { |
5152 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5153 | cputime64_t tmp; | |
5154 | ||
457533a7 | 5155 | /* Add user time to process. */ |
1da177e4 | 5156 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5157 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5158 | account_group_user_time(p, cputime); |
1da177e4 LT |
5159 | |
5160 | /* Add user time to cpustat. */ | |
5161 | tmp = cputime_to_cputime64(cputime); | |
5162 | if (TASK_NICE(p) > 0) | |
5163 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
5164 | else | |
5165 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
5166 | |
5167 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
5168 | /* Account for user time used */ |
5169 | acct_update_integrals(p); | |
1da177e4 LT |
5170 | } |
5171 | ||
94886b84 LV |
5172 | /* |
5173 | * Account guest cpu time to a process. | |
5174 | * @p: the process that the cpu time gets accounted to | |
5175 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 5176 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 5177 | */ |
457533a7 MS |
5178 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
5179 | cputime_t cputime_scaled) | |
94886b84 LV |
5180 | { |
5181 | cputime64_t tmp; | |
5182 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
5183 | ||
5184 | tmp = cputime_to_cputime64(cputime); | |
5185 | ||
457533a7 | 5186 | /* Add guest time to process. */ |
94886b84 | 5187 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 5188 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 5189 | account_group_user_time(p, cputime); |
94886b84 LV |
5190 | p->gtime = cputime_add(p->gtime, cputime); |
5191 | ||
457533a7 | 5192 | /* Add guest time to cpustat. */ |
94886b84 LV |
5193 | cpustat->user = cputime64_add(cpustat->user, tmp); |
5194 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
5195 | } | |
5196 | ||
1da177e4 LT |
5197 | /* |
5198 | * Account system cpu time to a process. | |
5199 | * @p: the process that the cpu time gets accounted to | |
5200 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
5201 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 5202 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
5203 | */ |
5204 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 5205 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
5206 | { |
5207 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
5208 | cputime64_t tmp; |
5209 | ||
983ed7a6 | 5210 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 5211 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
5212 | return; |
5213 | } | |
94886b84 | 5214 | |
457533a7 | 5215 | /* Add system time to process. */ |
1da177e4 | 5216 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 5217 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 5218 | account_group_system_time(p, cputime); |
1da177e4 LT |
5219 | |
5220 | /* Add system time to cpustat. */ | |
5221 | tmp = cputime_to_cputime64(cputime); | |
5222 | if (hardirq_count() - hardirq_offset) | |
5223 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
5224 | else if (softirq_count()) | |
5225 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 5226 | else |
79741dd3 MS |
5227 | cpustat->system = cputime64_add(cpustat->system, tmp); |
5228 | ||
ef12fefa BR |
5229 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
5230 | ||
1da177e4 LT |
5231 | /* Account for system time used */ |
5232 | acct_update_integrals(p); | |
1da177e4 LT |
5233 | } |
5234 | ||
c66f08be | 5235 | /* |
1da177e4 | 5236 | * Account for involuntary wait time. |
1da177e4 | 5237 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 5238 | */ |
79741dd3 | 5239 | void account_steal_time(cputime_t cputime) |
c66f08be | 5240 | { |
79741dd3 MS |
5241 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
5242 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
5243 | ||
5244 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
5245 | } |
5246 | ||
1da177e4 | 5247 | /* |
79741dd3 MS |
5248 | * Account for idle time. |
5249 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 5250 | */ |
79741dd3 | 5251 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
5252 | { |
5253 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 5254 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 5255 | struct rq *rq = this_rq(); |
1da177e4 | 5256 | |
79741dd3 MS |
5257 | if (atomic_read(&rq->nr_iowait) > 0) |
5258 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
5259 | else | |
5260 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
5261 | } |
5262 | ||
79741dd3 MS |
5263 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
5264 | ||
5265 | /* | |
5266 | * Account a single tick of cpu time. | |
5267 | * @p: the process that the cpu time gets accounted to | |
5268 | * @user_tick: indicates if the tick is a user or a system tick | |
5269 | */ | |
5270 | void account_process_tick(struct task_struct *p, int user_tick) | |
5271 | { | |
5272 | cputime_t one_jiffy = jiffies_to_cputime(1); | |
5273 | cputime_t one_jiffy_scaled = cputime_to_scaled(one_jiffy); | |
5274 | struct rq *rq = this_rq(); | |
5275 | ||
5276 | if (user_tick) | |
5277 | account_user_time(p, one_jiffy, one_jiffy_scaled); | |
f5f293a4 | 5278 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
79741dd3 MS |
5279 | account_system_time(p, HARDIRQ_OFFSET, one_jiffy, |
5280 | one_jiffy_scaled); | |
5281 | else | |
5282 | account_idle_time(one_jiffy); | |
5283 | } | |
5284 | ||
5285 | /* | |
5286 | * Account multiple ticks of steal time. | |
5287 | * @p: the process from which the cpu time has been stolen | |
5288 | * @ticks: number of stolen ticks | |
5289 | */ | |
5290 | void account_steal_ticks(unsigned long ticks) | |
5291 | { | |
5292 | account_steal_time(jiffies_to_cputime(ticks)); | |
5293 | } | |
5294 | ||
5295 | /* | |
5296 | * Account multiple ticks of idle time. | |
5297 | * @ticks: number of stolen ticks | |
5298 | */ | |
5299 | void account_idle_ticks(unsigned long ticks) | |
5300 | { | |
5301 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
5302 | } |
5303 | ||
79741dd3 MS |
5304 | #endif |
5305 | ||
49048622 BS |
5306 | /* |
5307 | * Use precise platform statistics if available: | |
5308 | */ | |
5309 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
5310 | cputime_t task_utime(struct task_struct *p) | |
5311 | { | |
5312 | return p->utime; | |
5313 | } | |
5314 | ||
5315 | cputime_t task_stime(struct task_struct *p) | |
5316 | { | |
5317 | return p->stime; | |
5318 | } | |
5319 | #else | |
5320 | cputime_t task_utime(struct task_struct *p) | |
5321 | { | |
5322 | clock_t utime = cputime_to_clock_t(p->utime), | |
5323 | total = utime + cputime_to_clock_t(p->stime); | |
5324 | u64 temp; | |
5325 | ||
5326 | /* | |
5327 | * Use CFS's precise accounting: | |
5328 | */ | |
5329 | temp = (u64)nsec_to_clock_t(p->se.sum_exec_runtime); | |
5330 | ||
5331 | if (total) { | |
5332 | temp *= utime; | |
5333 | do_div(temp, total); | |
5334 | } | |
5335 | utime = (clock_t)temp; | |
5336 | ||
5337 | p->prev_utime = max(p->prev_utime, clock_t_to_cputime(utime)); | |
5338 | return p->prev_utime; | |
5339 | } | |
5340 | ||
5341 | cputime_t task_stime(struct task_struct *p) | |
5342 | { | |
5343 | clock_t stime; | |
5344 | ||
5345 | /* | |
5346 | * Use CFS's precise accounting. (we subtract utime from | |
5347 | * the total, to make sure the total observed by userspace | |
5348 | * grows monotonically - apps rely on that): | |
5349 | */ | |
5350 | stime = nsec_to_clock_t(p->se.sum_exec_runtime) - | |
5351 | cputime_to_clock_t(task_utime(p)); | |
5352 | ||
5353 | if (stime >= 0) | |
5354 | p->prev_stime = max(p->prev_stime, clock_t_to_cputime(stime)); | |
5355 | ||
5356 | return p->prev_stime; | |
5357 | } | |
5358 | #endif | |
5359 | ||
5360 | inline cputime_t task_gtime(struct task_struct *p) | |
5361 | { | |
5362 | return p->gtime; | |
5363 | } | |
5364 | ||
7835b98b CL |
5365 | /* |
5366 | * This function gets called by the timer code, with HZ frequency. | |
5367 | * We call it with interrupts disabled. | |
5368 | * | |
5369 | * It also gets called by the fork code, when changing the parent's | |
5370 | * timeslices. | |
5371 | */ | |
5372 | void scheduler_tick(void) | |
5373 | { | |
7835b98b CL |
5374 | int cpu = smp_processor_id(); |
5375 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 5376 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
5377 | |
5378 | sched_clock_tick(); | |
dd41f596 IM |
5379 | |
5380 | spin_lock(&rq->lock); | |
3e51f33f | 5381 | update_rq_clock(rq); |
f1a438d8 | 5382 | update_cpu_load(rq); |
fa85ae24 | 5383 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 5384 | spin_unlock(&rq->lock); |
7835b98b | 5385 | |
e220d2dc PZ |
5386 | perf_counter_task_tick(curr, cpu); |
5387 | ||
e418e1c2 | 5388 | #ifdef CONFIG_SMP |
dd41f596 IM |
5389 | rq->idle_at_tick = idle_cpu(cpu); |
5390 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 5391 | #endif |
1da177e4 LT |
5392 | } |
5393 | ||
132380a0 | 5394 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
5395 | { |
5396 | if (in_lock_functions(addr)) { | |
5397 | addr = CALLER_ADDR2; | |
5398 | if (in_lock_functions(addr)) | |
5399 | addr = CALLER_ADDR3; | |
5400 | } | |
5401 | return addr; | |
5402 | } | |
1da177e4 | 5403 | |
7e49fcce SR |
5404 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
5405 | defined(CONFIG_PREEMPT_TRACER)) | |
5406 | ||
43627582 | 5407 | void __kprobes add_preempt_count(int val) |
1da177e4 | 5408 | { |
6cd8a4bb | 5409 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5410 | /* |
5411 | * Underflow? | |
5412 | */ | |
9a11b49a IM |
5413 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
5414 | return; | |
6cd8a4bb | 5415 | #endif |
1da177e4 | 5416 | preempt_count() += val; |
6cd8a4bb | 5417 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5418 | /* |
5419 | * Spinlock count overflowing soon? | |
5420 | */ | |
33859f7f MOS |
5421 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
5422 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
5423 | #endif |
5424 | if (preempt_count() == val) | |
5425 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5426 | } |
5427 | EXPORT_SYMBOL(add_preempt_count); | |
5428 | ||
43627582 | 5429 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 5430 | { |
6cd8a4bb | 5431 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
5432 | /* |
5433 | * Underflow? | |
5434 | */ | |
01e3eb82 | 5435 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 5436 | return; |
1da177e4 LT |
5437 | /* |
5438 | * Is the spinlock portion underflowing? | |
5439 | */ | |
9a11b49a IM |
5440 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
5441 | !(preempt_count() & PREEMPT_MASK))) | |
5442 | return; | |
6cd8a4bb | 5443 | #endif |
9a11b49a | 5444 | |
6cd8a4bb SR |
5445 | if (preempt_count() == val) |
5446 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
5447 | preempt_count() -= val; |
5448 | } | |
5449 | EXPORT_SYMBOL(sub_preempt_count); | |
5450 | ||
5451 | #endif | |
5452 | ||
5453 | /* | |
dd41f596 | 5454 | * Print scheduling while atomic bug: |
1da177e4 | 5455 | */ |
dd41f596 | 5456 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 5457 | { |
838225b4 SS |
5458 | struct pt_regs *regs = get_irq_regs(); |
5459 | ||
5460 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
5461 | prev->comm, prev->pid, preempt_count()); | |
5462 | ||
dd41f596 | 5463 | debug_show_held_locks(prev); |
e21f5b15 | 5464 | print_modules(); |
dd41f596 IM |
5465 | if (irqs_disabled()) |
5466 | print_irqtrace_events(prev); | |
838225b4 SS |
5467 | |
5468 | if (regs) | |
5469 | show_regs(regs); | |
5470 | else | |
5471 | dump_stack(); | |
dd41f596 | 5472 | } |
1da177e4 | 5473 | |
dd41f596 IM |
5474 | /* |
5475 | * Various schedule()-time debugging checks and statistics: | |
5476 | */ | |
5477 | static inline void schedule_debug(struct task_struct *prev) | |
5478 | { | |
1da177e4 | 5479 | /* |
41a2d6cf | 5480 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
5481 | * schedule() atomically, we ignore that path for now. |
5482 | * Otherwise, whine if we are scheduling when we should not be. | |
5483 | */ | |
3f33a7ce | 5484 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
5485 | __schedule_bug(prev); |
5486 | ||
1da177e4 LT |
5487 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
5488 | ||
2d72376b | 5489 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
5490 | #ifdef CONFIG_SCHEDSTATS |
5491 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
5492 | schedstat_inc(this_rq(), bkl_count); |
5493 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
5494 | } |
5495 | #endif | |
dd41f596 IM |
5496 | } |
5497 | ||
df1c99d4 MG |
5498 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
5499 | { | |
5500 | if (prev->state == TASK_RUNNING) { | |
5501 | u64 runtime = prev->se.sum_exec_runtime; | |
5502 | ||
5503 | runtime -= prev->se.prev_sum_exec_runtime; | |
5504 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
5505 | ||
5506 | /* | |
5507 | * In order to avoid avg_overlap growing stale when we are | |
5508 | * indeed overlapping and hence not getting put to sleep, grow | |
5509 | * the avg_overlap on preemption. | |
5510 | * | |
5511 | * We use the average preemption runtime because that | |
5512 | * correlates to the amount of cache footprint a task can | |
5513 | * build up. | |
5514 | */ | |
5515 | update_avg(&prev->se.avg_overlap, runtime); | |
5516 | } | |
5517 | prev->sched_class->put_prev_task(rq, prev); | |
5518 | } | |
5519 | ||
dd41f596 IM |
5520 | /* |
5521 | * Pick up the highest-prio task: | |
5522 | */ | |
5523 | static inline struct task_struct * | |
b67802ea | 5524 | pick_next_task(struct rq *rq) |
dd41f596 | 5525 | { |
5522d5d5 | 5526 | const struct sched_class *class; |
dd41f596 | 5527 | struct task_struct *p; |
1da177e4 LT |
5528 | |
5529 | /* | |
dd41f596 IM |
5530 | * Optimization: we know that if all tasks are in |
5531 | * the fair class we can call that function directly: | |
1da177e4 | 5532 | */ |
dd41f596 | 5533 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 5534 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
5535 | if (likely(p)) |
5536 | return p; | |
1da177e4 LT |
5537 | } |
5538 | ||
dd41f596 IM |
5539 | class = sched_class_highest; |
5540 | for ( ; ; ) { | |
fb8d4724 | 5541 | p = class->pick_next_task(rq); |
dd41f596 IM |
5542 | if (p) |
5543 | return p; | |
5544 | /* | |
5545 | * Will never be NULL as the idle class always | |
5546 | * returns a non-NULL p: | |
5547 | */ | |
5548 | class = class->next; | |
5549 | } | |
5550 | } | |
1da177e4 | 5551 | |
dd41f596 IM |
5552 | /* |
5553 | * schedule() is the main scheduler function. | |
5554 | */ | |
ff743345 | 5555 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
5556 | { |
5557 | struct task_struct *prev, *next; | |
67ca7bde | 5558 | unsigned long *switch_count; |
dd41f596 | 5559 | struct rq *rq; |
31656519 | 5560 | int cpu; |
dd41f596 | 5561 | |
ff743345 PZ |
5562 | need_resched: |
5563 | preempt_disable(); | |
dd41f596 IM |
5564 | cpu = smp_processor_id(); |
5565 | rq = cpu_rq(cpu); | |
5566 | rcu_qsctr_inc(cpu); | |
5567 | prev = rq->curr; | |
5568 | switch_count = &prev->nivcsw; | |
5569 | ||
5570 | release_kernel_lock(prev); | |
5571 | need_resched_nonpreemptible: | |
5572 | ||
5573 | schedule_debug(prev); | |
1da177e4 | 5574 | |
31656519 | 5575 | if (sched_feat(HRTICK)) |
f333fdc9 | 5576 | hrtick_clear(rq); |
8f4d37ec | 5577 | |
8cd162ce | 5578 | spin_lock_irq(&rq->lock); |
3e51f33f | 5579 | update_rq_clock(rq); |
1e819950 | 5580 | clear_tsk_need_resched(prev); |
1da177e4 | 5581 | |
1da177e4 | 5582 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 5583 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 5584 | prev->state = TASK_RUNNING; |
16882c1e | 5585 | else |
2e1cb74a | 5586 | deactivate_task(rq, prev, 1); |
dd41f596 | 5587 | switch_count = &prev->nvcsw; |
1da177e4 LT |
5588 | } |
5589 | ||
3f029d3c | 5590 | pre_schedule(rq, prev); |
f65eda4f | 5591 | |
dd41f596 | 5592 | if (unlikely(!rq->nr_running)) |
1da177e4 | 5593 | idle_balance(cpu, rq); |
1da177e4 | 5594 | |
df1c99d4 | 5595 | put_prev_task(rq, prev); |
b67802ea | 5596 | next = pick_next_task(rq); |
1da177e4 | 5597 | |
1da177e4 | 5598 | if (likely(prev != next)) { |
673a90a1 | 5599 | sched_info_switch(prev, next); |
564c2b21 | 5600 | perf_counter_task_sched_out(prev, next, cpu); |
673a90a1 | 5601 | |
1da177e4 LT |
5602 | rq->nr_switches++; |
5603 | rq->curr = next; | |
5604 | ++*switch_count; | |
5605 | ||
3f029d3c | 5606 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
5607 | /* |
5608 | * the context switch might have flipped the stack from under | |
5609 | * us, hence refresh the local variables. | |
5610 | */ | |
5611 | cpu = smp_processor_id(); | |
5612 | rq = cpu_rq(cpu); | |
3f029d3c | 5613 | } else |
1da177e4 | 5614 | spin_unlock_irq(&rq->lock); |
da19ab51 | 5615 | |
3f029d3c | 5616 | post_schedule(rq); |
1da177e4 | 5617 | |
8f4d37ec | 5618 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 5619 | goto need_resched_nonpreemptible; |
8f4d37ec | 5620 | |
1da177e4 | 5621 | preempt_enable_no_resched(); |
ff743345 | 5622 | if (need_resched()) |
1da177e4 LT |
5623 | goto need_resched; |
5624 | } | |
1da177e4 LT |
5625 | EXPORT_SYMBOL(schedule); |
5626 | ||
0d66bf6d PZ |
5627 | #ifdef CONFIG_SMP |
5628 | /* | |
5629 | * Look out! "owner" is an entirely speculative pointer | |
5630 | * access and not reliable. | |
5631 | */ | |
5632 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
5633 | { | |
5634 | unsigned int cpu; | |
5635 | struct rq *rq; | |
5636 | ||
5637 | if (!sched_feat(OWNER_SPIN)) | |
5638 | return 0; | |
5639 | ||
5640 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
5641 | /* | |
5642 | * Need to access the cpu field knowing that | |
5643 | * DEBUG_PAGEALLOC could have unmapped it if | |
5644 | * the mutex owner just released it and exited. | |
5645 | */ | |
5646 | if (probe_kernel_address(&owner->cpu, cpu)) | |
5647 | goto out; | |
5648 | #else | |
5649 | cpu = owner->cpu; | |
5650 | #endif | |
5651 | ||
5652 | /* | |
5653 | * Even if the access succeeded (likely case), | |
5654 | * the cpu field may no longer be valid. | |
5655 | */ | |
5656 | if (cpu >= nr_cpumask_bits) | |
5657 | goto out; | |
5658 | ||
5659 | /* | |
5660 | * We need to validate that we can do a | |
5661 | * get_cpu() and that we have the percpu area. | |
5662 | */ | |
5663 | if (!cpu_online(cpu)) | |
5664 | goto out; | |
5665 | ||
5666 | rq = cpu_rq(cpu); | |
5667 | ||
5668 | for (;;) { | |
5669 | /* | |
5670 | * Owner changed, break to re-assess state. | |
5671 | */ | |
5672 | if (lock->owner != owner) | |
5673 | break; | |
5674 | ||
5675 | /* | |
5676 | * Is that owner really running on that cpu? | |
5677 | */ | |
5678 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
5679 | return 0; | |
5680 | ||
5681 | cpu_relax(); | |
5682 | } | |
5683 | out: | |
5684 | return 1; | |
5685 | } | |
5686 | #endif | |
5687 | ||
1da177e4 LT |
5688 | #ifdef CONFIG_PREEMPT |
5689 | /* | |
2ed6e34f | 5690 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 5691 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
5692 | * occur there and call schedule directly. |
5693 | */ | |
5694 | asmlinkage void __sched preempt_schedule(void) | |
5695 | { | |
5696 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5697 | |
1da177e4 LT |
5698 | /* |
5699 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 5700 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 5701 | */ |
beed33a8 | 5702 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
5703 | return; |
5704 | ||
3a5c359a AK |
5705 | do { |
5706 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 5707 | schedule(); |
3a5c359a | 5708 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5709 | |
3a5c359a AK |
5710 | /* |
5711 | * Check again in case we missed a preemption opportunity | |
5712 | * between schedule and now. | |
5713 | */ | |
5714 | barrier(); | |
5ed0cec0 | 5715 | } while (need_resched()); |
1da177e4 | 5716 | } |
1da177e4 LT |
5717 | EXPORT_SYMBOL(preempt_schedule); |
5718 | ||
5719 | /* | |
2ed6e34f | 5720 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
5721 | * off of irq context. |
5722 | * Note, that this is called and return with irqs disabled. This will | |
5723 | * protect us against recursive calling from irq. | |
5724 | */ | |
5725 | asmlinkage void __sched preempt_schedule_irq(void) | |
5726 | { | |
5727 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 5728 | |
2ed6e34f | 5729 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
5730 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
5731 | ||
3a5c359a AK |
5732 | do { |
5733 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
5734 | local_irq_enable(); |
5735 | schedule(); | |
5736 | local_irq_disable(); | |
3a5c359a | 5737 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 5738 | |
3a5c359a AK |
5739 | /* |
5740 | * Check again in case we missed a preemption opportunity | |
5741 | * between schedule and now. | |
5742 | */ | |
5743 | barrier(); | |
5ed0cec0 | 5744 | } while (need_resched()); |
1da177e4 LT |
5745 | } |
5746 | ||
5747 | #endif /* CONFIG_PREEMPT */ | |
5748 | ||
95cdf3b7 IM |
5749 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
5750 | void *key) | |
1da177e4 | 5751 | { |
48f24c4d | 5752 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 5753 | } |
1da177e4 LT |
5754 | EXPORT_SYMBOL(default_wake_function); |
5755 | ||
5756 | /* | |
41a2d6cf IM |
5757 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
5758 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
5759 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
5760 | * | |
5761 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 5762 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
5763 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
5764 | */ | |
78ddb08f | 5765 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
777c6c5f | 5766 | int nr_exclusive, int sync, void *key) |
1da177e4 | 5767 | { |
2e45874c | 5768 | wait_queue_t *curr, *next; |
1da177e4 | 5769 | |
2e45874c | 5770 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
5771 | unsigned flags = curr->flags; |
5772 | ||
1da177e4 | 5773 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 5774 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
5775 | break; |
5776 | } | |
5777 | } | |
5778 | ||
5779 | /** | |
5780 | * __wake_up - wake up threads blocked on a waitqueue. | |
5781 | * @q: the waitqueue | |
5782 | * @mode: which threads | |
5783 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 5784 | * @key: is directly passed to the wakeup function |
50fa610a DH |
5785 | * |
5786 | * It may be assumed that this function implies a write memory barrier before | |
5787 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5788 | */ |
7ad5b3a5 | 5789 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 5790 | int nr_exclusive, void *key) |
1da177e4 LT |
5791 | { |
5792 | unsigned long flags; | |
5793 | ||
5794 | spin_lock_irqsave(&q->lock, flags); | |
5795 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
5796 | spin_unlock_irqrestore(&q->lock, flags); | |
5797 | } | |
1da177e4 LT |
5798 | EXPORT_SYMBOL(__wake_up); |
5799 | ||
5800 | /* | |
5801 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
5802 | */ | |
7ad5b3a5 | 5803 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
5804 | { |
5805 | __wake_up_common(q, mode, 1, 0, NULL); | |
5806 | } | |
5807 | ||
4ede816a DL |
5808 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
5809 | { | |
5810 | __wake_up_common(q, mode, 1, 0, key); | |
5811 | } | |
5812 | ||
1da177e4 | 5813 | /** |
4ede816a | 5814 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
5815 | * @q: the waitqueue |
5816 | * @mode: which threads | |
5817 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 5818 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
5819 | * |
5820 | * The sync wakeup differs that the waker knows that it will schedule | |
5821 | * away soon, so while the target thread will be woken up, it will not | |
5822 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
5823 | * with each other. This can prevent needless bouncing between CPUs. | |
5824 | * | |
5825 | * On UP it can prevent extra preemption. | |
50fa610a DH |
5826 | * |
5827 | * It may be assumed that this function implies a write memory barrier before | |
5828 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 5829 | */ |
4ede816a DL |
5830 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
5831 | int nr_exclusive, void *key) | |
1da177e4 LT |
5832 | { |
5833 | unsigned long flags; | |
5834 | int sync = 1; | |
5835 | ||
5836 | if (unlikely(!q)) | |
5837 | return; | |
5838 | ||
5839 | if (unlikely(!nr_exclusive)) | |
5840 | sync = 0; | |
5841 | ||
5842 | spin_lock_irqsave(&q->lock, flags); | |
4ede816a | 5843 | __wake_up_common(q, mode, nr_exclusive, sync, key); |
1da177e4 LT |
5844 | spin_unlock_irqrestore(&q->lock, flags); |
5845 | } | |
4ede816a DL |
5846 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
5847 | ||
5848 | /* | |
5849 | * __wake_up_sync - see __wake_up_sync_key() | |
5850 | */ | |
5851 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
5852 | { | |
5853 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
5854 | } | |
1da177e4 LT |
5855 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
5856 | ||
65eb3dc6 KD |
5857 | /** |
5858 | * complete: - signals a single thread waiting on this completion | |
5859 | * @x: holds the state of this particular completion | |
5860 | * | |
5861 | * This will wake up a single thread waiting on this completion. Threads will be | |
5862 | * awakened in the same order in which they were queued. | |
5863 | * | |
5864 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
5865 | * |
5866 | * It may be assumed that this function implies a write memory barrier before | |
5867 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5868 | */ |
b15136e9 | 5869 | void complete(struct completion *x) |
1da177e4 LT |
5870 | { |
5871 | unsigned long flags; | |
5872 | ||
5873 | spin_lock_irqsave(&x->wait.lock, flags); | |
5874 | x->done++; | |
d9514f6c | 5875 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
5876 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5877 | } | |
5878 | EXPORT_SYMBOL(complete); | |
5879 | ||
65eb3dc6 KD |
5880 | /** |
5881 | * complete_all: - signals all threads waiting on this completion | |
5882 | * @x: holds the state of this particular completion | |
5883 | * | |
5884 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
5885 | * |
5886 | * It may be assumed that this function implies a write memory barrier before | |
5887 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 5888 | */ |
b15136e9 | 5889 | void complete_all(struct completion *x) |
1da177e4 LT |
5890 | { |
5891 | unsigned long flags; | |
5892 | ||
5893 | spin_lock_irqsave(&x->wait.lock, flags); | |
5894 | x->done += UINT_MAX/2; | |
d9514f6c | 5895 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
5896 | spin_unlock_irqrestore(&x->wait.lock, flags); |
5897 | } | |
5898 | EXPORT_SYMBOL(complete_all); | |
5899 | ||
8cbbe86d AK |
5900 | static inline long __sched |
5901 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5902 | { |
1da177e4 LT |
5903 | if (!x->done) { |
5904 | DECLARE_WAITQUEUE(wait, current); | |
5905 | ||
5906 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
5907 | __add_wait_queue_tail(&x->wait, &wait); | |
5908 | do { | |
94d3d824 | 5909 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
5910 | timeout = -ERESTARTSYS; |
5911 | break; | |
8cbbe86d AK |
5912 | } |
5913 | __set_current_state(state); | |
1da177e4 LT |
5914 | spin_unlock_irq(&x->wait.lock); |
5915 | timeout = schedule_timeout(timeout); | |
5916 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 5917 | } while (!x->done && timeout); |
1da177e4 | 5918 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
5919 | if (!x->done) |
5920 | return timeout; | |
1da177e4 LT |
5921 | } |
5922 | x->done--; | |
ea71a546 | 5923 | return timeout ?: 1; |
1da177e4 | 5924 | } |
1da177e4 | 5925 | |
8cbbe86d AK |
5926 | static long __sched |
5927 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 5928 | { |
1da177e4 LT |
5929 | might_sleep(); |
5930 | ||
5931 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 5932 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 5933 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
5934 | return timeout; |
5935 | } | |
1da177e4 | 5936 | |
65eb3dc6 KD |
5937 | /** |
5938 | * wait_for_completion: - waits for completion of a task | |
5939 | * @x: holds the state of this particular completion | |
5940 | * | |
5941 | * This waits to be signaled for completion of a specific task. It is NOT | |
5942 | * interruptible and there is no timeout. | |
5943 | * | |
5944 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
5945 | * and interrupt capability. Also see complete(). | |
5946 | */ | |
b15136e9 | 5947 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
5948 | { |
5949 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 5950 | } |
8cbbe86d | 5951 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 5952 | |
65eb3dc6 KD |
5953 | /** |
5954 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
5955 | * @x: holds the state of this particular completion | |
5956 | * @timeout: timeout value in jiffies | |
5957 | * | |
5958 | * This waits for either a completion of a specific task to be signaled or for a | |
5959 | * specified timeout to expire. The timeout is in jiffies. It is not | |
5960 | * interruptible. | |
5961 | */ | |
b15136e9 | 5962 | unsigned long __sched |
8cbbe86d | 5963 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 5964 | { |
8cbbe86d | 5965 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 5966 | } |
8cbbe86d | 5967 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 5968 | |
65eb3dc6 KD |
5969 | /** |
5970 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
5971 | * @x: holds the state of this particular completion | |
5972 | * | |
5973 | * This waits for completion of a specific task to be signaled. It is | |
5974 | * interruptible. | |
5975 | */ | |
8cbbe86d | 5976 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 5977 | { |
51e97990 AK |
5978 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
5979 | if (t == -ERESTARTSYS) | |
5980 | return t; | |
5981 | return 0; | |
0fec171c | 5982 | } |
8cbbe86d | 5983 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 5984 | |
65eb3dc6 KD |
5985 | /** |
5986 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
5987 | * @x: holds the state of this particular completion | |
5988 | * @timeout: timeout value in jiffies | |
5989 | * | |
5990 | * This waits for either a completion of a specific task to be signaled or for a | |
5991 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
5992 | */ | |
b15136e9 | 5993 | unsigned long __sched |
8cbbe86d AK |
5994 | wait_for_completion_interruptible_timeout(struct completion *x, |
5995 | unsigned long timeout) | |
0fec171c | 5996 | { |
8cbbe86d | 5997 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 5998 | } |
8cbbe86d | 5999 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 6000 | |
65eb3dc6 KD |
6001 | /** |
6002 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
6003 | * @x: holds the state of this particular completion | |
6004 | * | |
6005 | * This waits to be signaled for completion of a specific task. It can be | |
6006 | * interrupted by a kill signal. | |
6007 | */ | |
009e577e MW |
6008 | int __sched wait_for_completion_killable(struct completion *x) |
6009 | { | |
6010 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
6011 | if (t == -ERESTARTSYS) | |
6012 | return t; | |
6013 | return 0; | |
6014 | } | |
6015 | EXPORT_SYMBOL(wait_for_completion_killable); | |
6016 | ||
be4de352 DC |
6017 | /** |
6018 | * try_wait_for_completion - try to decrement a completion without blocking | |
6019 | * @x: completion structure | |
6020 | * | |
6021 | * Returns: 0 if a decrement cannot be done without blocking | |
6022 | * 1 if a decrement succeeded. | |
6023 | * | |
6024 | * If a completion is being used as a counting completion, | |
6025 | * attempt to decrement the counter without blocking. This | |
6026 | * enables us to avoid waiting if the resource the completion | |
6027 | * is protecting is not available. | |
6028 | */ | |
6029 | bool try_wait_for_completion(struct completion *x) | |
6030 | { | |
6031 | int ret = 1; | |
6032 | ||
6033 | spin_lock_irq(&x->wait.lock); | |
6034 | if (!x->done) | |
6035 | ret = 0; | |
6036 | else | |
6037 | x->done--; | |
6038 | spin_unlock_irq(&x->wait.lock); | |
6039 | return ret; | |
6040 | } | |
6041 | EXPORT_SYMBOL(try_wait_for_completion); | |
6042 | ||
6043 | /** | |
6044 | * completion_done - Test to see if a completion has any waiters | |
6045 | * @x: completion structure | |
6046 | * | |
6047 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
6048 | * 1 if there are no waiters. | |
6049 | * | |
6050 | */ | |
6051 | bool completion_done(struct completion *x) | |
6052 | { | |
6053 | int ret = 1; | |
6054 | ||
6055 | spin_lock_irq(&x->wait.lock); | |
6056 | if (!x->done) | |
6057 | ret = 0; | |
6058 | spin_unlock_irq(&x->wait.lock); | |
6059 | return ret; | |
6060 | } | |
6061 | EXPORT_SYMBOL(completion_done); | |
6062 | ||
8cbbe86d AK |
6063 | static long __sched |
6064 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 6065 | { |
0fec171c IM |
6066 | unsigned long flags; |
6067 | wait_queue_t wait; | |
6068 | ||
6069 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 6070 | |
8cbbe86d | 6071 | __set_current_state(state); |
1da177e4 | 6072 | |
8cbbe86d AK |
6073 | spin_lock_irqsave(&q->lock, flags); |
6074 | __add_wait_queue(q, &wait); | |
6075 | spin_unlock(&q->lock); | |
6076 | timeout = schedule_timeout(timeout); | |
6077 | spin_lock_irq(&q->lock); | |
6078 | __remove_wait_queue(q, &wait); | |
6079 | spin_unlock_irqrestore(&q->lock, flags); | |
6080 | ||
6081 | return timeout; | |
6082 | } | |
6083 | ||
6084 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
6085 | { | |
6086 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 6087 | } |
1da177e4 LT |
6088 | EXPORT_SYMBOL(interruptible_sleep_on); |
6089 | ||
0fec171c | 6090 | long __sched |
95cdf3b7 | 6091 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 6092 | { |
8cbbe86d | 6093 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 6094 | } |
1da177e4 LT |
6095 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
6096 | ||
0fec171c | 6097 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 6098 | { |
8cbbe86d | 6099 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 6100 | } |
1da177e4 LT |
6101 | EXPORT_SYMBOL(sleep_on); |
6102 | ||
0fec171c | 6103 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 6104 | { |
8cbbe86d | 6105 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 6106 | } |
1da177e4 LT |
6107 | EXPORT_SYMBOL(sleep_on_timeout); |
6108 | ||
b29739f9 IM |
6109 | #ifdef CONFIG_RT_MUTEXES |
6110 | ||
6111 | /* | |
6112 | * rt_mutex_setprio - set the current priority of a task | |
6113 | * @p: task | |
6114 | * @prio: prio value (kernel-internal form) | |
6115 | * | |
6116 | * This function changes the 'effective' priority of a task. It does | |
6117 | * not touch ->normal_prio like __setscheduler(). | |
6118 | * | |
6119 | * Used by the rt_mutex code to implement priority inheritance logic. | |
6120 | */ | |
36c8b586 | 6121 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
6122 | { |
6123 | unsigned long flags; | |
83b699ed | 6124 | int oldprio, on_rq, running; |
70b97a7f | 6125 | struct rq *rq; |
cb469845 | 6126 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
6127 | |
6128 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
6129 | ||
6130 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6131 | update_rq_clock(rq); |
b29739f9 | 6132 | |
d5f9f942 | 6133 | oldprio = p->prio; |
dd41f596 | 6134 | on_rq = p->se.on_rq; |
051a1d1a | 6135 | running = task_current(rq, p); |
0e1f3483 | 6136 | if (on_rq) |
69be72c1 | 6137 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
6138 | if (running) |
6139 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
6140 | |
6141 | if (rt_prio(prio)) | |
6142 | p->sched_class = &rt_sched_class; | |
6143 | else | |
6144 | p->sched_class = &fair_sched_class; | |
6145 | ||
b29739f9 IM |
6146 | p->prio = prio; |
6147 | ||
0e1f3483 HS |
6148 | if (running) |
6149 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 6150 | if (on_rq) { |
8159f87e | 6151 | enqueue_task(rq, p, 0); |
cb469845 SR |
6152 | |
6153 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
6154 | } |
6155 | task_rq_unlock(rq, &flags); | |
6156 | } | |
6157 | ||
6158 | #endif | |
6159 | ||
36c8b586 | 6160 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 6161 | { |
dd41f596 | 6162 | int old_prio, delta, on_rq; |
1da177e4 | 6163 | unsigned long flags; |
70b97a7f | 6164 | struct rq *rq; |
1da177e4 LT |
6165 | |
6166 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
6167 | return; | |
6168 | /* | |
6169 | * We have to be careful, if called from sys_setpriority(), | |
6170 | * the task might be in the middle of scheduling on another CPU. | |
6171 | */ | |
6172 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 6173 | update_rq_clock(rq); |
1da177e4 LT |
6174 | /* |
6175 | * The RT priorities are set via sched_setscheduler(), but we still | |
6176 | * allow the 'normal' nice value to be set - but as expected | |
6177 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 6178 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 6179 | */ |
e05606d3 | 6180 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
6181 | p->static_prio = NICE_TO_PRIO(nice); |
6182 | goto out_unlock; | |
6183 | } | |
dd41f596 | 6184 | on_rq = p->se.on_rq; |
c09595f6 | 6185 | if (on_rq) |
69be72c1 | 6186 | dequeue_task(rq, p, 0); |
1da177e4 | 6187 | |
1da177e4 | 6188 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 6189 | set_load_weight(p); |
b29739f9 IM |
6190 | old_prio = p->prio; |
6191 | p->prio = effective_prio(p); | |
6192 | delta = p->prio - old_prio; | |
1da177e4 | 6193 | |
dd41f596 | 6194 | if (on_rq) { |
8159f87e | 6195 | enqueue_task(rq, p, 0); |
1da177e4 | 6196 | /* |
d5f9f942 AM |
6197 | * If the task increased its priority or is running and |
6198 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 6199 | */ |
d5f9f942 | 6200 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
6201 | resched_task(rq->curr); |
6202 | } | |
6203 | out_unlock: | |
6204 | task_rq_unlock(rq, &flags); | |
6205 | } | |
1da177e4 LT |
6206 | EXPORT_SYMBOL(set_user_nice); |
6207 | ||
e43379f1 MM |
6208 | /* |
6209 | * can_nice - check if a task can reduce its nice value | |
6210 | * @p: task | |
6211 | * @nice: nice value | |
6212 | */ | |
36c8b586 | 6213 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 6214 | { |
024f4747 MM |
6215 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
6216 | int nice_rlim = 20 - nice; | |
48f24c4d | 6217 | |
e43379f1 MM |
6218 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
6219 | capable(CAP_SYS_NICE)); | |
6220 | } | |
6221 | ||
1da177e4 LT |
6222 | #ifdef __ARCH_WANT_SYS_NICE |
6223 | ||
6224 | /* | |
6225 | * sys_nice - change the priority of the current process. | |
6226 | * @increment: priority increment | |
6227 | * | |
6228 | * sys_setpriority is a more generic, but much slower function that | |
6229 | * does similar things. | |
6230 | */ | |
5add95d4 | 6231 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 6232 | { |
48f24c4d | 6233 | long nice, retval; |
1da177e4 LT |
6234 | |
6235 | /* | |
6236 | * Setpriority might change our priority at the same moment. | |
6237 | * We don't have to worry. Conceptually one call occurs first | |
6238 | * and we have a single winner. | |
6239 | */ | |
e43379f1 MM |
6240 | if (increment < -40) |
6241 | increment = -40; | |
1da177e4 LT |
6242 | if (increment > 40) |
6243 | increment = 40; | |
6244 | ||
2b8f836f | 6245 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
6246 | if (nice < -20) |
6247 | nice = -20; | |
6248 | if (nice > 19) | |
6249 | nice = 19; | |
6250 | ||
e43379f1 MM |
6251 | if (increment < 0 && !can_nice(current, nice)) |
6252 | return -EPERM; | |
6253 | ||
1da177e4 LT |
6254 | retval = security_task_setnice(current, nice); |
6255 | if (retval) | |
6256 | return retval; | |
6257 | ||
6258 | set_user_nice(current, nice); | |
6259 | return 0; | |
6260 | } | |
6261 | ||
6262 | #endif | |
6263 | ||
6264 | /** | |
6265 | * task_prio - return the priority value of a given task. | |
6266 | * @p: the task in question. | |
6267 | * | |
6268 | * This is the priority value as seen by users in /proc. | |
6269 | * RT tasks are offset by -200. Normal tasks are centered | |
6270 | * around 0, value goes from -16 to +15. | |
6271 | */ | |
36c8b586 | 6272 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
6273 | { |
6274 | return p->prio - MAX_RT_PRIO; | |
6275 | } | |
6276 | ||
6277 | /** | |
6278 | * task_nice - return the nice value of a given task. | |
6279 | * @p: the task in question. | |
6280 | */ | |
36c8b586 | 6281 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
6282 | { |
6283 | return TASK_NICE(p); | |
6284 | } | |
150d8bed | 6285 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
6286 | |
6287 | /** | |
6288 | * idle_cpu - is a given cpu idle currently? | |
6289 | * @cpu: the processor in question. | |
6290 | */ | |
6291 | int idle_cpu(int cpu) | |
6292 | { | |
6293 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
6294 | } | |
6295 | ||
1da177e4 LT |
6296 | /** |
6297 | * idle_task - return the idle task for a given cpu. | |
6298 | * @cpu: the processor in question. | |
6299 | */ | |
36c8b586 | 6300 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
6301 | { |
6302 | return cpu_rq(cpu)->idle; | |
6303 | } | |
6304 | ||
6305 | /** | |
6306 | * find_process_by_pid - find a process with a matching PID value. | |
6307 | * @pid: the pid in question. | |
6308 | */ | |
a9957449 | 6309 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 6310 | { |
228ebcbe | 6311 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
6312 | } |
6313 | ||
6314 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
6315 | static void |
6316 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 6317 | { |
dd41f596 | 6318 | BUG_ON(p->se.on_rq); |
48f24c4d | 6319 | |
1da177e4 | 6320 | p->policy = policy; |
dd41f596 IM |
6321 | switch (p->policy) { |
6322 | case SCHED_NORMAL: | |
6323 | case SCHED_BATCH: | |
6324 | case SCHED_IDLE: | |
6325 | p->sched_class = &fair_sched_class; | |
6326 | break; | |
6327 | case SCHED_FIFO: | |
6328 | case SCHED_RR: | |
6329 | p->sched_class = &rt_sched_class; | |
6330 | break; | |
6331 | } | |
6332 | ||
1da177e4 | 6333 | p->rt_priority = prio; |
b29739f9 IM |
6334 | p->normal_prio = normal_prio(p); |
6335 | /* we are holding p->pi_lock already */ | |
6336 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 6337 | set_load_weight(p); |
1da177e4 LT |
6338 | } |
6339 | ||
c69e8d9c DH |
6340 | /* |
6341 | * check the target process has a UID that matches the current process's | |
6342 | */ | |
6343 | static bool check_same_owner(struct task_struct *p) | |
6344 | { | |
6345 | const struct cred *cred = current_cred(), *pcred; | |
6346 | bool match; | |
6347 | ||
6348 | rcu_read_lock(); | |
6349 | pcred = __task_cred(p); | |
6350 | match = (cred->euid == pcred->euid || | |
6351 | cred->euid == pcred->uid); | |
6352 | rcu_read_unlock(); | |
6353 | return match; | |
6354 | } | |
6355 | ||
961ccddd RR |
6356 | static int __sched_setscheduler(struct task_struct *p, int policy, |
6357 | struct sched_param *param, bool user) | |
1da177e4 | 6358 | { |
83b699ed | 6359 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 6360 | unsigned long flags; |
cb469845 | 6361 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 6362 | struct rq *rq; |
ca94c442 | 6363 | int reset_on_fork; |
1da177e4 | 6364 | |
66e5393a SR |
6365 | /* may grab non-irq protected spin_locks */ |
6366 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
6367 | recheck: |
6368 | /* double check policy once rq lock held */ | |
ca94c442 LP |
6369 | if (policy < 0) { |
6370 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 6371 | policy = oldpolicy = p->policy; |
ca94c442 LP |
6372 | } else { |
6373 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
6374 | policy &= ~SCHED_RESET_ON_FORK; | |
6375 | ||
6376 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
6377 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
6378 | policy != SCHED_IDLE) | |
6379 | return -EINVAL; | |
6380 | } | |
6381 | ||
1da177e4 LT |
6382 | /* |
6383 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
6384 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
6385 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
6386 | */ |
6387 | if (param->sched_priority < 0 || | |
95cdf3b7 | 6388 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 6389 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 6390 | return -EINVAL; |
e05606d3 | 6391 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
6392 | return -EINVAL; |
6393 | ||
37e4ab3f OC |
6394 | /* |
6395 | * Allow unprivileged RT tasks to decrease priority: | |
6396 | */ | |
961ccddd | 6397 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 6398 | if (rt_policy(policy)) { |
8dc3e909 | 6399 | unsigned long rlim_rtprio; |
8dc3e909 ON |
6400 | |
6401 | if (!lock_task_sighand(p, &flags)) | |
6402 | return -ESRCH; | |
6403 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
6404 | unlock_task_sighand(p, &flags); | |
6405 | ||
6406 | /* can't set/change the rt policy */ | |
6407 | if (policy != p->policy && !rlim_rtprio) | |
6408 | return -EPERM; | |
6409 | ||
6410 | /* can't increase priority */ | |
6411 | if (param->sched_priority > p->rt_priority && | |
6412 | param->sched_priority > rlim_rtprio) | |
6413 | return -EPERM; | |
6414 | } | |
dd41f596 IM |
6415 | /* |
6416 | * Like positive nice levels, dont allow tasks to | |
6417 | * move out of SCHED_IDLE either: | |
6418 | */ | |
6419 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
6420 | return -EPERM; | |
5fe1d75f | 6421 | |
37e4ab3f | 6422 | /* can't change other user's priorities */ |
c69e8d9c | 6423 | if (!check_same_owner(p)) |
37e4ab3f | 6424 | return -EPERM; |
ca94c442 LP |
6425 | |
6426 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
6427 | if (p->sched_reset_on_fork && !reset_on_fork) | |
6428 | return -EPERM; | |
37e4ab3f | 6429 | } |
1da177e4 | 6430 | |
725aad24 | 6431 | if (user) { |
b68aa230 | 6432 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
6433 | /* |
6434 | * Do not allow realtime tasks into groups that have no runtime | |
6435 | * assigned. | |
6436 | */ | |
9a7e0b18 PZ |
6437 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
6438 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 6439 | return -EPERM; |
b68aa230 PZ |
6440 | #endif |
6441 | ||
725aad24 JF |
6442 | retval = security_task_setscheduler(p, policy, param); |
6443 | if (retval) | |
6444 | return retval; | |
6445 | } | |
6446 | ||
b29739f9 IM |
6447 | /* |
6448 | * make sure no PI-waiters arrive (or leave) while we are | |
6449 | * changing the priority of the task: | |
6450 | */ | |
6451 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
6452 | /* |
6453 | * To be able to change p->policy safely, the apropriate | |
6454 | * runqueue lock must be held. | |
6455 | */ | |
b29739f9 | 6456 | rq = __task_rq_lock(p); |
1da177e4 LT |
6457 | /* recheck policy now with rq lock held */ |
6458 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
6459 | policy = oldpolicy = -1; | |
b29739f9 IM |
6460 | __task_rq_unlock(rq); |
6461 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
6462 | goto recheck; |
6463 | } | |
2daa3577 | 6464 | update_rq_clock(rq); |
dd41f596 | 6465 | on_rq = p->se.on_rq; |
051a1d1a | 6466 | running = task_current(rq, p); |
0e1f3483 | 6467 | if (on_rq) |
2e1cb74a | 6468 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
6469 | if (running) |
6470 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 6471 | |
ca94c442 LP |
6472 | p->sched_reset_on_fork = reset_on_fork; |
6473 | ||
1da177e4 | 6474 | oldprio = p->prio; |
dd41f596 | 6475 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 6476 | |
0e1f3483 HS |
6477 | if (running) |
6478 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
6479 | if (on_rq) { |
6480 | activate_task(rq, p, 0); | |
cb469845 SR |
6481 | |
6482 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 6483 | } |
b29739f9 IM |
6484 | __task_rq_unlock(rq); |
6485 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
6486 | ||
95e02ca9 TG |
6487 | rt_mutex_adjust_pi(p); |
6488 | ||
1da177e4 LT |
6489 | return 0; |
6490 | } | |
961ccddd RR |
6491 | |
6492 | /** | |
6493 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
6494 | * @p: the task in question. | |
6495 | * @policy: new policy. | |
6496 | * @param: structure containing the new RT priority. | |
6497 | * | |
6498 | * NOTE that the task may be already dead. | |
6499 | */ | |
6500 | int sched_setscheduler(struct task_struct *p, int policy, | |
6501 | struct sched_param *param) | |
6502 | { | |
6503 | return __sched_setscheduler(p, policy, param, true); | |
6504 | } | |
1da177e4 LT |
6505 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
6506 | ||
961ccddd RR |
6507 | /** |
6508 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
6509 | * @p: the task in question. | |
6510 | * @policy: new policy. | |
6511 | * @param: structure containing the new RT priority. | |
6512 | * | |
6513 | * Just like sched_setscheduler, only don't bother checking if the | |
6514 | * current context has permission. For example, this is needed in | |
6515 | * stop_machine(): we create temporary high priority worker threads, | |
6516 | * but our caller might not have that capability. | |
6517 | */ | |
6518 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
6519 | struct sched_param *param) | |
6520 | { | |
6521 | return __sched_setscheduler(p, policy, param, false); | |
6522 | } | |
6523 | ||
95cdf3b7 IM |
6524 | static int |
6525 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 6526 | { |
1da177e4 LT |
6527 | struct sched_param lparam; |
6528 | struct task_struct *p; | |
36c8b586 | 6529 | int retval; |
1da177e4 LT |
6530 | |
6531 | if (!param || pid < 0) | |
6532 | return -EINVAL; | |
6533 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
6534 | return -EFAULT; | |
5fe1d75f ON |
6535 | |
6536 | rcu_read_lock(); | |
6537 | retval = -ESRCH; | |
1da177e4 | 6538 | p = find_process_by_pid(pid); |
5fe1d75f ON |
6539 | if (p != NULL) |
6540 | retval = sched_setscheduler(p, policy, &lparam); | |
6541 | rcu_read_unlock(); | |
36c8b586 | 6542 | |
1da177e4 LT |
6543 | return retval; |
6544 | } | |
6545 | ||
6546 | /** | |
6547 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
6548 | * @pid: the pid in question. | |
6549 | * @policy: new policy. | |
6550 | * @param: structure containing the new RT priority. | |
6551 | */ | |
5add95d4 HC |
6552 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
6553 | struct sched_param __user *, param) | |
1da177e4 | 6554 | { |
c21761f1 JB |
6555 | /* negative values for policy are not valid */ |
6556 | if (policy < 0) | |
6557 | return -EINVAL; | |
6558 | ||
1da177e4 LT |
6559 | return do_sched_setscheduler(pid, policy, param); |
6560 | } | |
6561 | ||
6562 | /** | |
6563 | * sys_sched_setparam - set/change the RT priority of a thread | |
6564 | * @pid: the pid in question. | |
6565 | * @param: structure containing the new RT priority. | |
6566 | */ | |
5add95d4 | 6567 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6568 | { |
6569 | return do_sched_setscheduler(pid, -1, param); | |
6570 | } | |
6571 | ||
6572 | /** | |
6573 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
6574 | * @pid: the pid in question. | |
6575 | */ | |
5add95d4 | 6576 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 6577 | { |
36c8b586 | 6578 | struct task_struct *p; |
3a5c359a | 6579 | int retval; |
1da177e4 LT |
6580 | |
6581 | if (pid < 0) | |
3a5c359a | 6582 | return -EINVAL; |
1da177e4 LT |
6583 | |
6584 | retval = -ESRCH; | |
6585 | read_lock(&tasklist_lock); | |
6586 | p = find_process_by_pid(pid); | |
6587 | if (p) { | |
6588 | retval = security_task_getscheduler(p); | |
6589 | if (!retval) | |
ca94c442 LP |
6590 | retval = p->policy |
6591 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 LT |
6592 | } |
6593 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
6594 | return retval; |
6595 | } | |
6596 | ||
6597 | /** | |
ca94c442 | 6598 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
6599 | * @pid: the pid in question. |
6600 | * @param: structure containing the RT priority. | |
6601 | */ | |
5add95d4 | 6602 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
6603 | { |
6604 | struct sched_param lp; | |
36c8b586 | 6605 | struct task_struct *p; |
3a5c359a | 6606 | int retval; |
1da177e4 LT |
6607 | |
6608 | if (!param || pid < 0) | |
3a5c359a | 6609 | return -EINVAL; |
1da177e4 LT |
6610 | |
6611 | read_lock(&tasklist_lock); | |
6612 | p = find_process_by_pid(pid); | |
6613 | retval = -ESRCH; | |
6614 | if (!p) | |
6615 | goto out_unlock; | |
6616 | ||
6617 | retval = security_task_getscheduler(p); | |
6618 | if (retval) | |
6619 | goto out_unlock; | |
6620 | ||
6621 | lp.sched_priority = p->rt_priority; | |
6622 | read_unlock(&tasklist_lock); | |
6623 | ||
6624 | /* | |
6625 | * This one might sleep, we cannot do it with a spinlock held ... | |
6626 | */ | |
6627 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
6628 | ||
1da177e4 LT |
6629 | return retval; |
6630 | ||
6631 | out_unlock: | |
6632 | read_unlock(&tasklist_lock); | |
6633 | return retval; | |
6634 | } | |
6635 | ||
96f874e2 | 6636 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 6637 | { |
5a16f3d3 | 6638 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
6639 | struct task_struct *p; |
6640 | int retval; | |
1da177e4 | 6641 | |
95402b38 | 6642 | get_online_cpus(); |
1da177e4 LT |
6643 | read_lock(&tasklist_lock); |
6644 | ||
6645 | p = find_process_by_pid(pid); | |
6646 | if (!p) { | |
6647 | read_unlock(&tasklist_lock); | |
95402b38 | 6648 | put_online_cpus(); |
1da177e4 LT |
6649 | return -ESRCH; |
6650 | } | |
6651 | ||
6652 | /* | |
6653 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 6654 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
6655 | * usage count and then drop tasklist_lock. |
6656 | */ | |
6657 | get_task_struct(p); | |
6658 | read_unlock(&tasklist_lock); | |
6659 | ||
5a16f3d3 RR |
6660 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
6661 | retval = -ENOMEM; | |
6662 | goto out_put_task; | |
6663 | } | |
6664 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
6665 | retval = -ENOMEM; | |
6666 | goto out_free_cpus_allowed; | |
6667 | } | |
1da177e4 | 6668 | retval = -EPERM; |
c69e8d9c | 6669 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
6670 | goto out_unlock; |
6671 | ||
e7834f8f DQ |
6672 | retval = security_task_setscheduler(p, 0, NULL); |
6673 | if (retval) | |
6674 | goto out_unlock; | |
6675 | ||
5a16f3d3 RR |
6676 | cpuset_cpus_allowed(p, cpus_allowed); |
6677 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 6678 | again: |
5a16f3d3 | 6679 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 6680 | |
8707d8b8 | 6681 | if (!retval) { |
5a16f3d3 RR |
6682 | cpuset_cpus_allowed(p, cpus_allowed); |
6683 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
6684 | /* |
6685 | * We must have raced with a concurrent cpuset | |
6686 | * update. Just reset the cpus_allowed to the | |
6687 | * cpuset's cpus_allowed | |
6688 | */ | |
5a16f3d3 | 6689 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
6690 | goto again; |
6691 | } | |
6692 | } | |
1da177e4 | 6693 | out_unlock: |
5a16f3d3 RR |
6694 | free_cpumask_var(new_mask); |
6695 | out_free_cpus_allowed: | |
6696 | free_cpumask_var(cpus_allowed); | |
6697 | out_put_task: | |
1da177e4 | 6698 | put_task_struct(p); |
95402b38 | 6699 | put_online_cpus(); |
1da177e4 LT |
6700 | return retval; |
6701 | } | |
6702 | ||
6703 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 6704 | struct cpumask *new_mask) |
1da177e4 | 6705 | { |
96f874e2 RR |
6706 | if (len < cpumask_size()) |
6707 | cpumask_clear(new_mask); | |
6708 | else if (len > cpumask_size()) | |
6709 | len = cpumask_size(); | |
6710 | ||
1da177e4 LT |
6711 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
6712 | } | |
6713 | ||
6714 | /** | |
6715 | * sys_sched_setaffinity - set the cpu affinity of a process | |
6716 | * @pid: pid of the process | |
6717 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6718 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
6719 | */ | |
5add95d4 HC |
6720 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
6721 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 6722 | { |
5a16f3d3 | 6723 | cpumask_var_t new_mask; |
1da177e4 LT |
6724 | int retval; |
6725 | ||
5a16f3d3 RR |
6726 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
6727 | return -ENOMEM; | |
1da177e4 | 6728 | |
5a16f3d3 RR |
6729 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
6730 | if (retval == 0) | |
6731 | retval = sched_setaffinity(pid, new_mask); | |
6732 | free_cpumask_var(new_mask); | |
6733 | return retval; | |
1da177e4 LT |
6734 | } |
6735 | ||
96f874e2 | 6736 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 6737 | { |
36c8b586 | 6738 | struct task_struct *p; |
1da177e4 | 6739 | int retval; |
1da177e4 | 6740 | |
95402b38 | 6741 | get_online_cpus(); |
1da177e4 LT |
6742 | read_lock(&tasklist_lock); |
6743 | ||
6744 | retval = -ESRCH; | |
6745 | p = find_process_by_pid(pid); | |
6746 | if (!p) | |
6747 | goto out_unlock; | |
6748 | ||
e7834f8f DQ |
6749 | retval = security_task_getscheduler(p); |
6750 | if (retval) | |
6751 | goto out_unlock; | |
6752 | ||
96f874e2 | 6753 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
1da177e4 LT |
6754 | |
6755 | out_unlock: | |
6756 | read_unlock(&tasklist_lock); | |
95402b38 | 6757 | put_online_cpus(); |
1da177e4 | 6758 | |
9531b62f | 6759 | return retval; |
1da177e4 LT |
6760 | } |
6761 | ||
6762 | /** | |
6763 | * sys_sched_getaffinity - get the cpu affinity of a process | |
6764 | * @pid: pid of the process | |
6765 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
6766 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
6767 | */ | |
5add95d4 HC |
6768 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
6769 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
6770 | { |
6771 | int ret; | |
f17c8607 | 6772 | cpumask_var_t mask; |
1da177e4 | 6773 | |
f17c8607 | 6774 | if (len < cpumask_size()) |
1da177e4 LT |
6775 | return -EINVAL; |
6776 | ||
f17c8607 RR |
6777 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
6778 | return -ENOMEM; | |
1da177e4 | 6779 | |
f17c8607 RR |
6780 | ret = sched_getaffinity(pid, mask); |
6781 | if (ret == 0) { | |
6782 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
6783 | ret = -EFAULT; | |
6784 | else | |
6785 | ret = cpumask_size(); | |
6786 | } | |
6787 | free_cpumask_var(mask); | |
1da177e4 | 6788 | |
f17c8607 | 6789 | return ret; |
1da177e4 LT |
6790 | } |
6791 | ||
6792 | /** | |
6793 | * sys_sched_yield - yield the current processor to other threads. | |
6794 | * | |
dd41f596 IM |
6795 | * This function yields the current CPU to other tasks. If there are no |
6796 | * other threads running on this CPU then this function will return. | |
1da177e4 | 6797 | */ |
5add95d4 | 6798 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 6799 | { |
70b97a7f | 6800 | struct rq *rq = this_rq_lock(); |
1da177e4 | 6801 | |
2d72376b | 6802 | schedstat_inc(rq, yld_count); |
4530d7ab | 6803 | current->sched_class->yield_task(rq); |
1da177e4 LT |
6804 | |
6805 | /* | |
6806 | * Since we are going to call schedule() anyway, there's | |
6807 | * no need to preempt or enable interrupts: | |
6808 | */ | |
6809 | __release(rq->lock); | |
8a25d5de | 6810 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
6811 | _raw_spin_unlock(&rq->lock); |
6812 | preempt_enable_no_resched(); | |
6813 | ||
6814 | schedule(); | |
6815 | ||
6816 | return 0; | |
6817 | } | |
6818 | ||
d86ee480 PZ |
6819 | static inline int should_resched(void) |
6820 | { | |
6821 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
6822 | } | |
6823 | ||
e7b38404 | 6824 | static void __cond_resched(void) |
1da177e4 | 6825 | { |
e7aaaa69 FW |
6826 | add_preempt_count(PREEMPT_ACTIVE); |
6827 | schedule(); | |
6828 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
6829 | } |
6830 | ||
02b67cc3 | 6831 | int __sched _cond_resched(void) |
1da177e4 | 6832 | { |
d86ee480 | 6833 | if (should_resched()) { |
1da177e4 LT |
6834 | __cond_resched(); |
6835 | return 1; | |
6836 | } | |
6837 | return 0; | |
6838 | } | |
02b67cc3 | 6839 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
6840 | |
6841 | /* | |
613afbf8 | 6842 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
6843 | * call schedule, and on return reacquire the lock. |
6844 | * | |
41a2d6cf | 6845 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
6846 | * operations here to prevent schedule() from being called twice (once via |
6847 | * spin_unlock(), once by hand). | |
6848 | */ | |
613afbf8 | 6849 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 6850 | { |
d86ee480 | 6851 | int resched = should_resched(); |
6df3cecb JK |
6852 | int ret = 0; |
6853 | ||
95c354fe | 6854 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 6855 | spin_unlock(lock); |
d86ee480 | 6856 | if (resched) |
95c354fe NP |
6857 | __cond_resched(); |
6858 | else | |
6859 | cpu_relax(); | |
6df3cecb | 6860 | ret = 1; |
1da177e4 | 6861 | spin_lock(lock); |
1da177e4 | 6862 | } |
6df3cecb | 6863 | return ret; |
1da177e4 | 6864 | } |
613afbf8 | 6865 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 6866 | |
613afbf8 | 6867 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
6868 | { |
6869 | BUG_ON(!in_softirq()); | |
6870 | ||
d86ee480 | 6871 | if (should_resched()) { |
98d82567 | 6872 | local_bh_enable(); |
1da177e4 LT |
6873 | __cond_resched(); |
6874 | local_bh_disable(); | |
6875 | return 1; | |
6876 | } | |
6877 | return 0; | |
6878 | } | |
613afbf8 | 6879 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 6880 | |
1da177e4 LT |
6881 | /** |
6882 | * yield - yield the current processor to other threads. | |
6883 | * | |
72fd4a35 | 6884 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
6885 | * thread runnable and calls sys_sched_yield(). |
6886 | */ | |
6887 | void __sched yield(void) | |
6888 | { | |
6889 | set_current_state(TASK_RUNNING); | |
6890 | sys_sched_yield(); | |
6891 | } | |
1da177e4 LT |
6892 | EXPORT_SYMBOL(yield); |
6893 | ||
6894 | /* | |
41a2d6cf | 6895 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
6896 | * that process accounting knows that this is a task in IO wait state. |
6897 | * | |
6898 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
6899 | * has set its backing_dev_info: the queue against which it should throttle) | |
6900 | */ | |
6901 | void __sched io_schedule(void) | |
6902 | { | |
54d35f29 | 6903 | struct rq *rq = raw_rq(); |
1da177e4 | 6904 | |
0ff92245 | 6905 | delayacct_blkio_start(); |
1da177e4 | 6906 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6907 | current->in_iowait = 1; |
1da177e4 | 6908 | schedule(); |
8f0dfc34 | 6909 | current->in_iowait = 0; |
1da177e4 | 6910 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6911 | delayacct_blkio_end(); |
1da177e4 | 6912 | } |
1da177e4 LT |
6913 | EXPORT_SYMBOL(io_schedule); |
6914 | ||
6915 | long __sched io_schedule_timeout(long timeout) | |
6916 | { | |
54d35f29 | 6917 | struct rq *rq = raw_rq(); |
1da177e4 LT |
6918 | long ret; |
6919 | ||
0ff92245 | 6920 | delayacct_blkio_start(); |
1da177e4 | 6921 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 6922 | current->in_iowait = 1; |
1da177e4 | 6923 | ret = schedule_timeout(timeout); |
8f0dfc34 | 6924 | current->in_iowait = 0; |
1da177e4 | 6925 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 6926 | delayacct_blkio_end(); |
1da177e4 LT |
6927 | return ret; |
6928 | } | |
6929 | ||
6930 | /** | |
6931 | * sys_sched_get_priority_max - return maximum RT priority. | |
6932 | * @policy: scheduling class. | |
6933 | * | |
6934 | * this syscall returns the maximum rt_priority that can be used | |
6935 | * by a given scheduling class. | |
6936 | */ | |
5add95d4 | 6937 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
6938 | { |
6939 | int ret = -EINVAL; | |
6940 | ||
6941 | switch (policy) { | |
6942 | case SCHED_FIFO: | |
6943 | case SCHED_RR: | |
6944 | ret = MAX_USER_RT_PRIO-1; | |
6945 | break; | |
6946 | case SCHED_NORMAL: | |
b0a9499c | 6947 | case SCHED_BATCH: |
dd41f596 | 6948 | case SCHED_IDLE: |
1da177e4 LT |
6949 | ret = 0; |
6950 | break; | |
6951 | } | |
6952 | return ret; | |
6953 | } | |
6954 | ||
6955 | /** | |
6956 | * sys_sched_get_priority_min - return minimum RT priority. | |
6957 | * @policy: scheduling class. | |
6958 | * | |
6959 | * this syscall returns the minimum rt_priority that can be used | |
6960 | * by a given scheduling class. | |
6961 | */ | |
5add95d4 | 6962 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
6963 | { |
6964 | int ret = -EINVAL; | |
6965 | ||
6966 | switch (policy) { | |
6967 | case SCHED_FIFO: | |
6968 | case SCHED_RR: | |
6969 | ret = 1; | |
6970 | break; | |
6971 | case SCHED_NORMAL: | |
b0a9499c | 6972 | case SCHED_BATCH: |
dd41f596 | 6973 | case SCHED_IDLE: |
1da177e4 LT |
6974 | ret = 0; |
6975 | } | |
6976 | return ret; | |
6977 | } | |
6978 | ||
6979 | /** | |
6980 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
6981 | * @pid: pid of the process. | |
6982 | * @interval: userspace pointer to the timeslice value. | |
6983 | * | |
6984 | * this syscall writes the default timeslice value of a given process | |
6985 | * into the user-space timespec buffer. A value of '0' means infinity. | |
6986 | */ | |
17da2bd9 | 6987 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 6988 | struct timespec __user *, interval) |
1da177e4 | 6989 | { |
36c8b586 | 6990 | struct task_struct *p; |
a4ec24b4 | 6991 | unsigned int time_slice; |
3a5c359a | 6992 | int retval; |
1da177e4 | 6993 | struct timespec t; |
1da177e4 LT |
6994 | |
6995 | if (pid < 0) | |
3a5c359a | 6996 | return -EINVAL; |
1da177e4 LT |
6997 | |
6998 | retval = -ESRCH; | |
6999 | read_lock(&tasklist_lock); | |
7000 | p = find_process_by_pid(pid); | |
7001 | if (!p) | |
7002 | goto out_unlock; | |
7003 | ||
7004 | retval = security_task_getscheduler(p); | |
7005 | if (retval) | |
7006 | goto out_unlock; | |
7007 | ||
77034937 IM |
7008 | /* |
7009 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
7010 | * tasks that are on an otherwise idle runqueue: | |
7011 | */ | |
7012 | time_slice = 0; | |
7013 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 7014 | time_slice = DEF_TIMESLICE; |
1868f958 | 7015 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
7016 | struct sched_entity *se = &p->se; |
7017 | unsigned long flags; | |
7018 | struct rq *rq; | |
7019 | ||
7020 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
7021 | if (rq->cfs.load.weight) |
7022 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
7023 | task_rq_unlock(rq, &flags); |
7024 | } | |
1da177e4 | 7025 | read_unlock(&tasklist_lock); |
a4ec24b4 | 7026 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 7027 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 7028 | return retval; |
3a5c359a | 7029 | |
1da177e4 LT |
7030 | out_unlock: |
7031 | read_unlock(&tasklist_lock); | |
7032 | return retval; | |
7033 | } | |
7034 | ||
7c731e0a | 7035 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 7036 | |
82a1fcb9 | 7037 | void sched_show_task(struct task_struct *p) |
1da177e4 | 7038 | { |
1da177e4 | 7039 | unsigned long free = 0; |
36c8b586 | 7040 | unsigned state; |
1da177e4 | 7041 | |
1da177e4 | 7042 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 7043 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 7044 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 7045 | #if BITS_PER_LONG == 32 |
1da177e4 | 7046 | if (state == TASK_RUNNING) |
cc4ea795 | 7047 | printk(KERN_CONT " running "); |
1da177e4 | 7048 | else |
cc4ea795 | 7049 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
7050 | #else |
7051 | if (state == TASK_RUNNING) | |
cc4ea795 | 7052 | printk(KERN_CONT " running task "); |
1da177e4 | 7053 | else |
cc4ea795 | 7054 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
7055 | #endif |
7056 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 7057 | free = stack_not_used(p); |
1da177e4 | 7058 | #endif |
aa47b7e0 DR |
7059 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
7060 | task_pid_nr(p), task_pid_nr(p->real_parent), | |
7061 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 7062 | |
5fb5e6de | 7063 | show_stack(p, NULL); |
1da177e4 LT |
7064 | } |
7065 | ||
e59e2ae2 | 7066 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 7067 | { |
36c8b586 | 7068 | struct task_struct *g, *p; |
1da177e4 | 7069 | |
4bd77321 IM |
7070 | #if BITS_PER_LONG == 32 |
7071 | printk(KERN_INFO | |
7072 | " task PC stack pid father\n"); | |
1da177e4 | 7073 | #else |
4bd77321 IM |
7074 | printk(KERN_INFO |
7075 | " task PC stack pid father\n"); | |
1da177e4 LT |
7076 | #endif |
7077 | read_lock(&tasklist_lock); | |
7078 | do_each_thread(g, p) { | |
7079 | /* | |
7080 | * reset the NMI-timeout, listing all files on a slow | |
7081 | * console might take alot of time: | |
7082 | */ | |
7083 | touch_nmi_watchdog(); | |
39bc89fd | 7084 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 7085 | sched_show_task(p); |
1da177e4 LT |
7086 | } while_each_thread(g, p); |
7087 | ||
04c9167f JF |
7088 | touch_all_softlockup_watchdogs(); |
7089 | ||
dd41f596 IM |
7090 | #ifdef CONFIG_SCHED_DEBUG |
7091 | sysrq_sched_debug_show(); | |
7092 | #endif | |
1da177e4 | 7093 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
7094 | /* |
7095 | * Only show locks if all tasks are dumped: | |
7096 | */ | |
7097 | if (state_filter == -1) | |
7098 | debug_show_all_locks(); | |
1da177e4 LT |
7099 | } |
7100 | ||
1df21055 IM |
7101 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
7102 | { | |
dd41f596 | 7103 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
7104 | } |
7105 | ||
f340c0d1 IM |
7106 | /** |
7107 | * init_idle - set up an idle thread for a given CPU | |
7108 | * @idle: task in question | |
7109 | * @cpu: cpu the idle task belongs to | |
7110 | * | |
7111 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
7112 | * flag, to make booting more robust. | |
7113 | */ | |
5c1e1767 | 7114 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 7115 | { |
70b97a7f | 7116 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
7117 | unsigned long flags; |
7118 | ||
5cbd54ef IM |
7119 | spin_lock_irqsave(&rq->lock, flags); |
7120 | ||
dd41f596 IM |
7121 | __sched_fork(idle); |
7122 | idle->se.exec_start = sched_clock(); | |
7123 | ||
b29739f9 | 7124 | idle->prio = idle->normal_prio = MAX_PRIO; |
96f874e2 | 7125 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 7126 | __set_task_cpu(idle, cpu); |
1da177e4 | 7127 | |
1da177e4 | 7128 | rq->curr = rq->idle = idle; |
4866cde0 NP |
7129 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
7130 | idle->oncpu = 1; | |
7131 | #endif | |
1da177e4 LT |
7132 | spin_unlock_irqrestore(&rq->lock, flags); |
7133 | ||
7134 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
7135 | #if defined(CONFIG_PREEMPT) |
7136 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
7137 | #else | |
a1261f54 | 7138 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 7139 | #endif |
dd41f596 IM |
7140 | /* |
7141 | * The idle tasks have their own, simple scheduling class: | |
7142 | */ | |
7143 | idle->sched_class = &idle_sched_class; | |
fb52607a | 7144 | ftrace_graph_init_task(idle); |
1da177e4 LT |
7145 | } |
7146 | ||
7147 | /* | |
7148 | * In a system that switches off the HZ timer nohz_cpu_mask | |
7149 | * indicates which cpus entered this state. This is used | |
7150 | * in the rcu update to wait only for active cpus. For system | |
7151 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 7152 | * always be CPU_BITS_NONE. |
1da177e4 | 7153 | */ |
6a7b3dc3 | 7154 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 7155 | |
19978ca6 IM |
7156 | /* |
7157 | * Increase the granularity value when there are more CPUs, | |
7158 | * because with more CPUs the 'effective latency' as visible | |
7159 | * to users decreases. But the relationship is not linear, | |
7160 | * so pick a second-best guess by going with the log2 of the | |
7161 | * number of CPUs. | |
7162 | * | |
7163 | * This idea comes from the SD scheduler of Con Kolivas: | |
7164 | */ | |
7165 | static inline void sched_init_granularity(void) | |
7166 | { | |
7167 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
7168 | const unsigned long limit = 200000000; | |
7169 | ||
7170 | sysctl_sched_min_granularity *= factor; | |
7171 | if (sysctl_sched_min_granularity > limit) | |
7172 | sysctl_sched_min_granularity = limit; | |
7173 | ||
7174 | sysctl_sched_latency *= factor; | |
7175 | if (sysctl_sched_latency > limit) | |
7176 | sysctl_sched_latency = limit; | |
7177 | ||
7178 | sysctl_sched_wakeup_granularity *= factor; | |
55cd5340 PZ |
7179 | |
7180 | sysctl_sched_shares_ratelimit *= factor; | |
19978ca6 IM |
7181 | } |
7182 | ||
1da177e4 LT |
7183 | #ifdef CONFIG_SMP |
7184 | /* | |
7185 | * This is how migration works: | |
7186 | * | |
70b97a7f | 7187 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
7188 | * runqueue and wake up that CPU's migration thread. |
7189 | * 2) we down() the locked semaphore => thread blocks. | |
7190 | * 3) migration thread wakes up (implicitly it forces the migrated | |
7191 | * thread off the CPU) | |
7192 | * 4) it gets the migration request and checks whether the migrated | |
7193 | * task is still in the wrong runqueue. | |
7194 | * 5) if it's in the wrong runqueue then the migration thread removes | |
7195 | * it and puts it into the right queue. | |
7196 | * 6) migration thread up()s the semaphore. | |
7197 | * 7) we wake up and the migration is done. | |
7198 | */ | |
7199 | ||
7200 | /* | |
7201 | * Change a given task's CPU affinity. Migrate the thread to a | |
7202 | * proper CPU and schedule it away if the CPU it's executing on | |
7203 | * is removed from the allowed bitmask. | |
7204 | * | |
7205 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 7206 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
7207 | * call is not atomic; no spinlocks may be held. |
7208 | */ | |
96f874e2 | 7209 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 7210 | { |
70b97a7f | 7211 | struct migration_req req; |
1da177e4 | 7212 | unsigned long flags; |
70b97a7f | 7213 | struct rq *rq; |
48f24c4d | 7214 | int ret = 0; |
1da177e4 LT |
7215 | |
7216 | rq = task_rq_lock(p, &flags); | |
96f874e2 | 7217 | if (!cpumask_intersects(new_mask, cpu_online_mask)) { |
1da177e4 LT |
7218 | ret = -EINVAL; |
7219 | goto out; | |
7220 | } | |
7221 | ||
9985b0ba | 7222 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 7223 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
7224 | ret = -EINVAL; |
7225 | goto out; | |
7226 | } | |
7227 | ||
73fe6aae | 7228 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 7229 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 7230 | else { |
96f874e2 RR |
7231 | cpumask_copy(&p->cpus_allowed, new_mask); |
7232 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
7233 | } |
7234 | ||
1da177e4 | 7235 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 7236 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
7237 | goto out; |
7238 | ||
1e5ce4f4 | 7239 | if (migrate_task(p, cpumask_any_and(cpu_online_mask, new_mask), &req)) { |
1da177e4 | 7240 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
7241 | struct task_struct *mt = rq->migration_thread; |
7242 | ||
7243 | get_task_struct(mt); | |
1da177e4 LT |
7244 | task_rq_unlock(rq, &flags); |
7245 | wake_up_process(rq->migration_thread); | |
693525e3 | 7246 | put_task_struct(mt); |
1da177e4 LT |
7247 | wait_for_completion(&req.done); |
7248 | tlb_migrate_finish(p->mm); | |
7249 | return 0; | |
7250 | } | |
7251 | out: | |
7252 | task_rq_unlock(rq, &flags); | |
48f24c4d | 7253 | |
1da177e4 LT |
7254 | return ret; |
7255 | } | |
cd8ba7cd | 7256 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
7257 | |
7258 | /* | |
41a2d6cf | 7259 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
7260 | * this because either it can't run here any more (set_cpus_allowed() |
7261 | * away from this CPU, or CPU going down), or because we're | |
7262 | * attempting to rebalance this task on exec (sched_exec). | |
7263 | * | |
7264 | * So we race with normal scheduler movements, but that's OK, as long | |
7265 | * as the task is no longer on this CPU. | |
efc30814 KK |
7266 | * |
7267 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 7268 | */ |
efc30814 | 7269 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 7270 | { |
70b97a7f | 7271 | struct rq *rq_dest, *rq_src; |
dd41f596 | 7272 | int ret = 0, on_rq; |
1da177e4 | 7273 | |
e761b772 | 7274 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 7275 | return ret; |
1da177e4 LT |
7276 | |
7277 | rq_src = cpu_rq(src_cpu); | |
7278 | rq_dest = cpu_rq(dest_cpu); | |
7279 | ||
7280 | double_rq_lock(rq_src, rq_dest); | |
7281 | /* Already moved. */ | |
7282 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 7283 | goto done; |
1da177e4 | 7284 | /* Affinity changed (again). */ |
96f874e2 | 7285 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 7286 | goto fail; |
1da177e4 | 7287 | |
dd41f596 | 7288 | on_rq = p->se.on_rq; |
6e82a3be | 7289 | if (on_rq) |
2e1cb74a | 7290 | deactivate_task(rq_src, p, 0); |
6e82a3be | 7291 | |
1da177e4 | 7292 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
7293 | if (on_rq) { |
7294 | activate_task(rq_dest, p, 0); | |
15afe09b | 7295 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 7296 | } |
b1e38734 | 7297 | done: |
efc30814 | 7298 | ret = 1; |
b1e38734 | 7299 | fail: |
1da177e4 | 7300 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 7301 | return ret; |
1da177e4 LT |
7302 | } |
7303 | ||
7304 | /* | |
7305 | * migration_thread - this is a highprio system thread that performs | |
7306 | * thread migration by bumping thread off CPU then 'pushing' onto | |
7307 | * another runqueue. | |
7308 | */ | |
95cdf3b7 | 7309 | static int migration_thread(void *data) |
1da177e4 | 7310 | { |
1da177e4 | 7311 | int cpu = (long)data; |
70b97a7f | 7312 | struct rq *rq; |
1da177e4 LT |
7313 | |
7314 | rq = cpu_rq(cpu); | |
7315 | BUG_ON(rq->migration_thread != current); | |
7316 | ||
7317 | set_current_state(TASK_INTERRUPTIBLE); | |
7318 | while (!kthread_should_stop()) { | |
70b97a7f | 7319 | struct migration_req *req; |
1da177e4 | 7320 | struct list_head *head; |
1da177e4 | 7321 | |
1da177e4 LT |
7322 | spin_lock_irq(&rq->lock); |
7323 | ||
7324 | if (cpu_is_offline(cpu)) { | |
7325 | spin_unlock_irq(&rq->lock); | |
371cbb38 | 7326 | break; |
1da177e4 LT |
7327 | } |
7328 | ||
7329 | if (rq->active_balance) { | |
7330 | active_load_balance(rq, cpu); | |
7331 | rq->active_balance = 0; | |
7332 | } | |
7333 | ||
7334 | head = &rq->migration_queue; | |
7335 | ||
7336 | if (list_empty(head)) { | |
7337 | spin_unlock_irq(&rq->lock); | |
7338 | schedule(); | |
7339 | set_current_state(TASK_INTERRUPTIBLE); | |
7340 | continue; | |
7341 | } | |
70b97a7f | 7342 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
7343 | list_del_init(head->next); |
7344 | ||
674311d5 NP |
7345 | spin_unlock(&rq->lock); |
7346 | __migrate_task(req->task, cpu, req->dest_cpu); | |
7347 | local_irq_enable(); | |
1da177e4 LT |
7348 | |
7349 | complete(&req->done); | |
7350 | } | |
7351 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 7352 | |
1da177e4 LT |
7353 | return 0; |
7354 | } | |
7355 | ||
7356 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
7357 | |
7358 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
7359 | { | |
7360 | int ret; | |
7361 | ||
7362 | local_irq_disable(); | |
7363 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
7364 | local_irq_enable(); | |
7365 | return ret; | |
7366 | } | |
7367 | ||
054b9108 | 7368 | /* |
3a4fa0a2 | 7369 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 7370 | */ |
48f24c4d | 7371 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 7372 | { |
70b97a7f | 7373 | int dest_cpu; |
6ca09dfc | 7374 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(dead_cpu)); |
e76bd8d9 RR |
7375 | |
7376 | again: | |
7377 | /* Look for allowed, online CPU in same node. */ | |
7378 | for_each_cpu_and(dest_cpu, nodemask, cpu_online_mask) | |
7379 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
7380 | goto move; | |
7381 | ||
7382 | /* Any allowed, online CPU? */ | |
7383 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_online_mask); | |
7384 | if (dest_cpu < nr_cpu_ids) | |
7385 | goto move; | |
7386 | ||
7387 | /* No more Mr. Nice Guy. */ | |
7388 | if (dest_cpu >= nr_cpu_ids) { | |
e76bd8d9 RR |
7389 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); |
7390 | dest_cpu = cpumask_any_and(cpu_online_mask, &p->cpus_allowed); | |
1da177e4 | 7391 | |
e76bd8d9 RR |
7392 | /* |
7393 | * Don't tell them about moving exiting tasks or | |
7394 | * kernel threads (both mm NULL), since they never | |
7395 | * leave kernel. | |
7396 | */ | |
7397 | if (p->mm && printk_ratelimit()) { | |
7398 | printk(KERN_INFO "process %d (%s) no " | |
7399 | "longer affine to cpu%d\n", | |
7400 | task_pid_nr(p), p->comm, dead_cpu); | |
3a5c359a | 7401 | } |
e76bd8d9 RR |
7402 | } |
7403 | ||
7404 | move: | |
7405 | /* It can have affinity changed while we were choosing. */ | |
7406 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
7407 | goto again; | |
1da177e4 LT |
7408 | } |
7409 | ||
7410 | /* | |
7411 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
7412 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
7413 | * for performance reasons the counter is not stricly tracking tasks to | |
7414 | * their home CPUs. So we just add the counter to another CPU's counter, | |
7415 | * to keep the global sum constant after CPU-down: | |
7416 | */ | |
70b97a7f | 7417 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 7418 | { |
1e5ce4f4 | 7419 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_online_mask)); |
1da177e4 LT |
7420 | unsigned long flags; |
7421 | ||
7422 | local_irq_save(flags); | |
7423 | double_rq_lock(rq_src, rq_dest); | |
7424 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
7425 | rq_src->nr_uninterruptible = 0; | |
7426 | double_rq_unlock(rq_src, rq_dest); | |
7427 | local_irq_restore(flags); | |
7428 | } | |
7429 | ||
7430 | /* Run through task list and migrate tasks from the dead cpu. */ | |
7431 | static void migrate_live_tasks(int src_cpu) | |
7432 | { | |
48f24c4d | 7433 | struct task_struct *p, *t; |
1da177e4 | 7434 | |
f7b4cddc | 7435 | read_lock(&tasklist_lock); |
1da177e4 | 7436 | |
48f24c4d IM |
7437 | do_each_thread(t, p) { |
7438 | if (p == current) | |
1da177e4 LT |
7439 | continue; |
7440 | ||
48f24c4d IM |
7441 | if (task_cpu(p) == src_cpu) |
7442 | move_task_off_dead_cpu(src_cpu, p); | |
7443 | } while_each_thread(t, p); | |
1da177e4 | 7444 | |
f7b4cddc | 7445 | read_unlock(&tasklist_lock); |
1da177e4 LT |
7446 | } |
7447 | ||
dd41f596 IM |
7448 | /* |
7449 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
7450 | * It does so by boosting its priority to highest possible. |
7451 | * Used by CPU offline code. | |
1da177e4 LT |
7452 | */ |
7453 | void sched_idle_next(void) | |
7454 | { | |
48f24c4d | 7455 | int this_cpu = smp_processor_id(); |
70b97a7f | 7456 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
7457 | struct task_struct *p = rq->idle; |
7458 | unsigned long flags; | |
7459 | ||
7460 | /* cpu has to be offline */ | |
48f24c4d | 7461 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 7462 | |
48f24c4d IM |
7463 | /* |
7464 | * Strictly not necessary since rest of the CPUs are stopped by now | |
7465 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
7466 | */ |
7467 | spin_lock_irqsave(&rq->lock, flags); | |
7468 | ||
dd41f596 | 7469 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 7470 | |
94bc9a7b DA |
7471 | update_rq_clock(rq); |
7472 | activate_task(rq, p, 0); | |
1da177e4 LT |
7473 | |
7474 | spin_unlock_irqrestore(&rq->lock, flags); | |
7475 | } | |
7476 | ||
48f24c4d IM |
7477 | /* |
7478 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
7479 | * offline. |
7480 | */ | |
7481 | void idle_task_exit(void) | |
7482 | { | |
7483 | struct mm_struct *mm = current->active_mm; | |
7484 | ||
7485 | BUG_ON(cpu_online(smp_processor_id())); | |
7486 | ||
7487 | if (mm != &init_mm) | |
7488 | switch_mm(mm, &init_mm, current); | |
7489 | mmdrop(mm); | |
7490 | } | |
7491 | ||
054b9108 | 7492 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 7493 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 7494 | { |
70b97a7f | 7495 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
7496 | |
7497 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 7498 | BUG_ON(!p->exit_state); |
1da177e4 LT |
7499 | |
7500 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 7501 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 7502 | |
48f24c4d | 7503 | get_task_struct(p); |
1da177e4 LT |
7504 | |
7505 | /* | |
7506 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 7507 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
7508 | * fine. |
7509 | */ | |
f7b4cddc | 7510 | spin_unlock_irq(&rq->lock); |
48f24c4d | 7511 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 7512 | spin_lock_irq(&rq->lock); |
1da177e4 | 7513 | |
48f24c4d | 7514 | put_task_struct(p); |
1da177e4 LT |
7515 | } |
7516 | ||
7517 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
7518 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
7519 | { | |
70b97a7f | 7520 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 7521 | struct task_struct *next; |
48f24c4d | 7522 | |
dd41f596 IM |
7523 | for ( ; ; ) { |
7524 | if (!rq->nr_running) | |
7525 | break; | |
a8e504d2 | 7526 | update_rq_clock(rq); |
b67802ea | 7527 | next = pick_next_task(rq); |
dd41f596 IM |
7528 | if (!next) |
7529 | break; | |
79c53799 | 7530 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 7531 | migrate_dead(dead_cpu, next); |
e692ab53 | 7532 | |
1da177e4 LT |
7533 | } |
7534 | } | |
dce48a84 TG |
7535 | |
7536 | /* | |
7537 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
7538 | */ | |
7539 | static void calc_global_load_remove(struct rq *rq) | |
7540 | { | |
7541 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 7542 | rq->calc_load_active = 0; |
dce48a84 | 7543 | } |
1da177e4 LT |
7544 | #endif /* CONFIG_HOTPLUG_CPU */ |
7545 | ||
e692ab53 NP |
7546 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
7547 | ||
7548 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
7549 | { |
7550 | .procname = "sched_domain", | |
c57baf1e | 7551 | .mode = 0555, |
e0361851 | 7552 | }, |
38605cae | 7553 | {0, }, |
e692ab53 NP |
7554 | }; |
7555 | ||
7556 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 7557 | { |
c57baf1e | 7558 | .ctl_name = CTL_KERN, |
e0361851 | 7559 | .procname = "kernel", |
c57baf1e | 7560 | .mode = 0555, |
e0361851 AD |
7561 | .child = sd_ctl_dir, |
7562 | }, | |
38605cae | 7563 | {0, }, |
e692ab53 NP |
7564 | }; |
7565 | ||
7566 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
7567 | { | |
7568 | struct ctl_table *entry = | |
5cf9f062 | 7569 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 7570 | |
e692ab53 NP |
7571 | return entry; |
7572 | } | |
7573 | ||
6382bc90 MM |
7574 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
7575 | { | |
cd790076 | 7576 | struct ctl_table *entry; |
6382bc90 | 7577 | |
cd790076 MM |
7578 | /* |
7579 | * In the intermediate directories, both the child directory and | |
7580 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 7581 | * will always be set. In the lowest directory the names are |
cd790076 MM |
7582 | * static strings and all have proc handlers. |
7583 | */ | |
7584 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
7585 | if (entry->child) |
7586 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
7587 | if (entry->proc_handler == NULL) |
7588 | kfree(entry->procname); | |
7589 | } | |
6382bc90 MM |
7590 | |
7591 | kfree(*tablep); | |
7592 | *tablep = NULL; | |
7593 | } | |
7594 | ||
e692ab53 | 7595 | static void |
e0361851 | 7596 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
7597 | const char *procname, void *data, int maxlen, |
7598 | mode_t mode, proc_handler *proc_handler) | |
7599 | { | |
e692ab53 NP |
7600 | entry->procname = procname; |
7601 | entry->data = data; | |
7602 | entry->maxlen = maxlen; | |
7603 | entry->mode = mode; | |
7604 | entry->proc_handler = proc_handler; | |
7605 | } | |
7606 | ||
7607 | static struct ctl_table * | |
7608 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
7609 | { | |
a5d8c348 | 7610 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 7611 | |
ad1cdc1d MM |
7612 | if (table == NULL) |
7613 | return NULL; | |
7614 | ||
e0361851 | 7615 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 7616 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7617 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 7618 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 7619 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 7620 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7621 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 7622 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7623 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 7624 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7625 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 7626 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7627 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 7628 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7629 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 7630 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 7631 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 7632 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 7633 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
7634 | &sd->cache_nice_tries, |
7635 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 7636 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 7637 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
7638 | set_table_entry(&table[11], "name", sd->name, |
7639 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
7640 | /* &table[12] is terminator */ | |
e692ab53 NP |
7641 | |
7642 | return table; | |
7643 | } | |
7644 | ||
9a4e7159 | 7645 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
7646 | { |
7647 | struct ctl_table *entry, *table; | |
7648 | struct sched_domain *sd; | |
7649 | int domain_num = 0, i; | |
7650 | char buf[32]; | |
7651 | ||
7652 | for_each_domain(cpu, sd) | |
7653 | domain_num++; | |
7654 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
7655 | if (table == NULL) |
7656 | return NULL; | |
e692ab53 NP |
7657 | |
7658 | i = 0; | |
7659 | for_each_domain(cpu, sd) { | |
7660 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 7661 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7662 | entry->mode = 0555; |
e692ab53 NP |
7663 | entry->child = sd_alloc_ctl_domain_table(sd); |
7664 | entry++; | |
7665 | i++; | |
7666 | } | |
7667 | return table; | |
7668 | } | |
7669 | ||
7670 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 7671 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
7672 | { |
7673 | int i, cpu_num = num_online_cpus(); | |
7674 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
7675 | char buf[32]; | |
7676 | ||
7378547f MM |
7677 | WARN_ON(sd_ctl_dir[0].child); |
7678 | sd_ctl_dir[0].child = entry; | |
7679 | ||
ad1cdc1d MM |
7680 | if (entry == NULL) |
7681 | return; | |
7682 | ||
97b6ea7b | 7683 | for_each_online_cpu(i) { |
e692ab53 | 7684 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 7685 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 7686 | entry->mode = 0555; |
e692ab53 | 7687 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 7688 | entry++; |
e692ab53 | 7689 | } |
7378547f MM |
7690 | |
7691 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
7692 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
7693 | } | |
6382bc90 | 7694 | |
7378547f | 7695 | /* may be called multiple times per register */ |
6382bc90 MM |
7696 | static void unregister_sched_domain_sysctl(void) |
7697 | { | |
7378547f MM |
7698 | if (sd_sysctl_header) |
7699 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 7700 | sd_sysctl_header = NULL; |
7378547f MM |
7701 | if (sd_ctl_dir[0].child) |
7702 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 7703 | } |
e692ab53 | 7704 | #else |
6382bc90 MM |
7705 | static void register_sched_domain_sysctl(void) |
7706 | { | |
7707 | } | |
7708 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
7709 | { |
7710 | } | |
7711 | #endif | |
7712 | ||
1f11eb6a GH |
7713 | static void set_rq_online(struct rq *rq) |
7714 | { | |
7715 | if (!rq->online) { | |
7716 | const struct sched_class *class; | |
7717 | ||
c6c4927b | 7718 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7719 | rq->online = 1; |
7720 | ||
7721 | for_each_class(class) { | |
7722 | if (class->rq_online) | |
7723 | class->rq_online(rq); | |
7724 | } | |
7725 | } | |
7726 | } | |
7727 | ||
7728 | static void set_rq_offline(struct rq *rq) | |
7729 | { | |
7730 | if (rq->online) { | |
7731 | const struct sched_class *class; | |
7732 | ||
7733 | for_each_class(class) { | |
7734 | if (class->rq_offline) | |
7735 | class->rq_offline(rq); | |
7736 | } | |
7737 | ||
c6c4927b | 7738 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
7739 | rq->online = 0; |
7740 | } | |
7741 | } | |
7742 | ||
1da177e4 LT |
7743 | /* |
7744 | * migration_call - callback that gets triggered when a CPU is added. | |
7745 | * Here we can start up the necessary migration thread for the new CPU. | |
7746 | */ | |
48f24c4d IM |
7747 | static int __cpuinit |
7748 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 7749 | { |
1da177e4 | 7750 | struct task_struct *p; |
48f24c4d | 7751 | int cpu = (long)hcpu; |
1da177e4 | 7752 | unsigned long flags; |
70b97a7f | 7753 | struct rq *rq; |
1da177e4 LT |
7754 | |
7755 | switch (action) { | |
5be9361c | 7756 | |
1da177e4 | 7757 | case CPU_UP_PREPARE: |
8bb78442 | 7758 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 7759 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
7760 | if (IS_ERR(p)) |
7761 | return NOTIFY_BAD; | |
1da177e4 LT |
7762 | kthread_bind(p, cpu); |
7763 | /* Must be high prio: stop_machine expects to yield to it. */ | |
7764 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 7765 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 7766 | task_rq_unlock(rq, &flags); |
371cbb38 | 7767 | get_task_struct(p); |
1da177e4 | 7768 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 7769 | rq->calc_load_update = calc_load_update; |
1da177e4 | 7770 | break; |
48f24c4d | 7771 | |
1da177e4 | 7772 | case CPU_ONLINE: |
8bb78442 | 7773 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 7774 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 7775 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
7776 | |
7777 | /* Update our root-domain */ | |
7778 | rq = cpu_rq(cpu); | |
7779 | spin_lock_irqsave(&rq->lock, flags); | |
7780 | if (rq->rd) { | |
c6c4927b | 7781 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
7782 | |
7783 | set_rq_online(rq); | |
1f94ef59 GH |
7784 | } |
7785 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 7786 | break; |
48f24c4d | 7787 | |
1da177e4 LT |
7788 | #ifdef CONFIG_HOTPLUG_CPU |
7789 | case CPU_UP_CANCELED: | |
8bb78442 | 7790 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
7791 | if (!cpu_rq(cpu)->migration_thread) |
7792 | break; | |
41a2d6cf | 7793 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 7794 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 7795 | cpumask_any(cpu_online_mask)); |
1da177e4 | 7796 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 7797 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
7798 | cpu_rq(cpu)->migration_thread = NULL; |
7799 | break; | |
48f24c4d | 7800 | |
1da177e4 | 7801 | case CPU_DEAD: |
8bb78442 | 7802 | case CPU_DEAD_FROZEN: |
470fd646 | 7803 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
7804 | migrate_live_tasks(cpu); |
7805 | rq = cpu_rq(cpu); | |
7806 | kthread_stop(rq->migration_thread); | |
371cbb38 | 7807 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
7808 | rq->migration_thread = NULL; |
7809 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 7810 | spin_lock_irq(&rq->lock); |
a8e504d2 | 7811 | update_rq_clock(rq); |
2e1cb74a | 7812 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 7813 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
7814 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
7815 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 7816 | migrate_dead_tasks(cpu); |
d2da272a | 7817 | spin_unlock_irq(&rq->lock); |
470fd646 | 7818 | cpuset_unlock(); |
1da177e4 LT |
7819 | migrate_nr_uninterruptible(rq); |
7820 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 7821 | calc_global_load_remove(rq); |
41a2d6cf IM |
7822 | /* |
7823 | * No need to migrate the tasks: it was best-effort if | |
7824 | * they didn't take sched_hotcpu_mutex. Just wake up | |
7825 | * the requestors. | |
7826 | */ | |
1da177e4 LT |
7827 | spin_lock_irq(&rq->lock); |
7828 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
7829 | struct migration_req *req; |
7830 | ||
1da177e4 | 7831 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 7832 | struct migration_req, list); |
1da177e4 | 7833 | list_del_init(&req->list); |
9a2bd244 | 7834 | spin_unlock_irq(&rq->lock); |
1da177e4 | 7835 | complete(&req->done); |
9a2bd244 | 7836 | spin_lock_irq(&rq->lock); |
1da177e4 LT |
7837 | } |
7838 | spin_unlock_irq(&rq->lock); | |
7839 | break; | |
57d885fe | 7840 | |
08f503b0 GH |
7841 | case CPU_DYING: |
7842 | case CPU_DYING_FROZEN: | |
57d885fe GH |
7843 | /* Update our root-domain */ |
7844 | rq = cpu_rq(cpu); | |
7845 | spin_lock_irqsave(&rq->lock, flags); | |
7846 | if (rq->rd) { | |
c6c4927b | 7847 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 7848 | set_rq_offline(rq); |
57d885fe GH |
7849 | } |
7850 | spin_unlock_irqrestore(&rq->lock, flags); | |
7851 | break; | |
1da177e4 LT |
7852 | #endif |
7853 | } | |
7854 | return NOTIFY_OK; | |
7855 | } | |
7856 | ||
f38b0820 PM |
7857 | /* |
7858 | * Register at high priority so that task migration (migrate_all_tasks) | |
7859 | * happens before everything else. This has to be lower priority than | |
7860 | * the notifier in the perf_counter subsystem, though. | |
1da177e4 | 7861 | */ |
26c2143b | 7862 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
7863 | .notifier_call = migration_call, |
7864 | .priority = 10 | |
7865 | }; | |
7866 | ||
7babe8db | 7867 | static int __init migration_init(void) |
1da177e4 LT |
7868 | { |
7869 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 7870 | int err; |
48f24c4d IM |
7871 | |
7872 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
7873 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
7874 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
7875 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
7876 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 7877 | |
a004cd42 | 7878 | return 0; |
1da177e4 | 7879 | } |
7babe8db | 7880 | early_initcall(migration_init); |
1da177e4 LT |
7881 | #endif |
7882 | ||
7883 | #ifdef CONFIG_SMP | |
476f3534 | 7884 | |
3e9830dc | 7885 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 7886 | |
7c16ec58 | 7887 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 7888 | struct cpumask *groupmask) |
1da177e4 | 7889 | { |
4dcf6aff | 7890 | struct sched_group *group = sd->groups; |
434d53b0 | 7891 | char str[256]; |
1da177e4 | 7892 | |
968ea6d8 | 7893 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 7894 | cpumask_clear(groupmask); |
4dcf6aff IM |
7895 | |
7896 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
7897 | ||
7898 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
7899 | printk("does not load-balance\n"); | |
7900 | if (sd->parent) | |
7901 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
7902 | " has parent"); | |
7903 | return -1; | |
41c7ce9a NP |
7904 | } |
7905 | ||
eefd796a | 7906 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 7907 | |
758b2cdc | 7908 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
4dcf6aff IM |
7909 | printk(KERN_ERR "ERROR: domain->span does not contain " |
7910 | "CPU%d\n", cpu); | |
7911 | } | |
758b2cdc | 7912 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
4dcf6aff IM |
7913 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
7914 | " CPU%d\n", cpu); | |
7915 | } | |
1da177e4 | 7916 | |
4dcf6aff | 7917 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 7918 | do { |
4dcf6aff IM |
7919 | if (!group) { |
7920 | printk("\n"); | |
7921 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
7922 | break; |
7923 | } | |
7924 | ||
4dcf6aff IM |
7925 | if (!group->__cpu_power) { |
7926 | printk(KERN_CONT "\n"); | |
7927 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
7928 | "set\n"); | |
7929 | break; | |
7930 | } | |
1da177e4 | 7931 | |
758b2cdc | 7932 | if (!cpumask_weight(sched_group_cpus(group))) { |
4dcf6aff IM |
7933 | printk(KERN_CONT "\n"); |
7934 | printk(KERN_ERR "ERROR: empty group\n"); | |
7935 | break; | |
7936 | } | |
1da177e4 | 7937 | |
758b2cdc | 7938 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
4dcf6aff IM |
7939 | printk(KERN_CONT "\n"); |
7940 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
7941 | break; | |
7942 | } | |
1da177e4 | 7943 | |
758b2cdc | 7944 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 7945 | |
968ea6d8 | 7946 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf GS |
7947 | |
7948 | printk(KERN_CONT " %s", str); | |
7949 | if (group->__cpu_power != SCHED_LOAD_SCALE) { | |
7950 | printk(KERN_CONT " (__cpu_power = %d)", | |
7951 | group->__cpu_power); | |
7952 | } | |
1da177e4 | 7953 | |
4dcf6aff IM |
7954 | group = group->next; |
7955 | } while (group != sd->groups); | |
7956 | printk(KERN_CONT "\n"); | |
1da177e4 | 7957 | |
758b2cdc | 7958 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
4dcf6aff | 7959 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 7960 | |
758b2cdc RR |
7961 | if (sd->parent && |
7962 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
4dcf6aff IM |
7963 | printk(KERN_ERR "ERROR: parent span is not a superset " |
7964 | "of domain->span\n"); | |
7965 | return 0; | |
7966 | } | |
1da177e4 | 7967 | |
4dcf6aff IM |
7968 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
7969 | { | |
d5dd3db1 | 7970 | cpumask_var_t groupmask; |
4dcf6aff | 7971 | int level = 0; |
1da177e4 | 7972 | |
4dcf6aff IM |
7973 | if (!sd) { |
7974 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
7975 | return; | |
7976 | } | |
1da177e4 | 7977 | |
4dcf6aff IM |
7978 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
7979 | ||
d5dd3db1 | 7980 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
7981 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
7982 | return; | |
7983 | } | |
7984 | ||
4dcf6aff | 7985 | for (;;) { |
7c16ec58 | 7986 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 7987 | break; |
1da177e4 LT |
7988 | level++; |
7989 | sd = sd->parent; | |
33859f7f | 7990 | if (!sd) |
4dcf6aff IM |
7991 | break; |
7992 | } | |
d5dd3db1 | 7993 | free_cpumask_var(groupmask); |
1da177e4 | 7994 | } |
6d6bc0ad | 7995 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 7996 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 7997 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 7998 | |
1a20ff27 | 7999 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 8000 | { |
758b2cdc | 8001 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
8002 | return 1; |
8003 | ||
8004 | /* Following flags need at least 2 groups */ | |
8005 | if (sd->flags & (SD_LOAD_BALANCE | | |
8006 | SD_BALANCE_NEWIDLE | | |
8007 | SD_BALANCE_FORK | | |
89c4710e SS |
8008 | SD_BALANCE_EXEC | |
8009 | SD_SHARE_CPUPOWER | | |
8010 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
8011 | if (sd->groups != sd->groups->next) |
8012 | return 0; | |
8013 | } | |
8014 | ||
8015 | /* Following flags don't use groups */ | |
8016 | if (sd->flags & (SD_WAKE_IDLE | | |
8017 | SD_WAKE_AFFINE | | |
8018 | SD_WAKE_BALANCE)) | |
8019 | return 0; | |
8020 | ||
8021 | return 1; | |
8022 | } | |
8023 | ||
48f24c4d IM |
8024 | static int |
8025 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
8026 | { |
8027 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
8028 | ||
8029 | if (sd_degenerate(parent)) | |
8030 | return 1; | |
8031 | ||
758b2cdc | 8032 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
8033 | return 0; |
8034 | ||
8035 | /* Does parent contain flags not in child? */ | |
8036 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
8037 | if (cflags & SD_WAKE_AFFINE) | |
8038 | pflags &= ~SD_WAKE_BALANCE; | |
8039 | /* Flags needing groups don't count if only 1 group in parent */ | |
8040 | if (parent->groups == parent->groups->next) { | |
8041 | pflags &= ~(SD_LOAD_BALANCE | | |
8042 | SD_BALANCE_NEWIDLE | | |
8043 | SD_BALANCE_FORK | | |
89c4710e SS |
8044 | SD_BALANCE_EXEC | |
8045 | SD_SHARE_CPUPOWER | | |
8046 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
8047 | if (nr_node_ids == 1) |
8048 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
8049 | } |
8050 | if (~cflags & pflags) | |
8051 | return 0; | |
8052 | ||
8053 | return 1; | |
8054 | } | |
8055 | ||
c6c4927b RR |
8056 | static void free_rootdomain(struct root_domain *rd) |
8057 | { | |
68e74568 RR |
8058 | cpupri_cleanup(&rd->cpupri); |
8059 | ||
c6c4927b RR |
8060 | free_cpumask_var(rd->rto_mask); |
8061 | free_cpumask_var(rd->online); | |
8062 | free_cpumask_var(rd->span); | |
8063 | kfree(rd); | |
8064 | } | |
8065 | ||
57d885fe GH |
8066 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
8067 | { | |
a0490fa3 | 8068 | struct root_domain *old_rd = NULL; |
57d885fe | 8069 | unsigned long flags; |
57d885fe GH |
8070 | |
8071 | spin_lock_irqsave(&rq->lock, flags); | |
8072 | ||
8073 | if (rq->rd) { | |
a0490fa3 | 8074 | old_rd = rq->rd; |
57d885fe | 8075 | |
c6c4927b | 8076 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 8077 | set_rq_offline(rq); |
57d885fe | 8078 | |
c6c4927b | 8079 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 8080 | |
a0490fa3 IM |
8081 | /* |
8082 | * If we dont want to free the old_rt yet then | |
8083 | * set old_rd to NULL to skip the freeing later | |
8084 | * in this function: | |
8085 | */ | |
8086 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
8087 | old_rd = NULL; | |
57d885fe GH |
8088 | } |
8089 | ||
8090 | atomic_inc(&rd->refcount); | |
8091 | rq->rd = rd; | |
8092 | ||
c6c4927b | 8093 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 8094 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 8095 | set_rq_online(rq); |
57d885fe GH |
8096 | |
8097 | spin_unlock_irqrestore(&rq->lock, flags); | |
a0490fa3 IM |
8098 | |
8099 | if (old_rd) | |
8100 | free_rootdomain(old_rd); | |
57d885fe GH |
8101 | } |
8102 | ||
fd5e1b5d | 8103 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 8104 | { |
36b7b6d4 PE |
8105 | gfp_t gfp = GFP_KERNEL; |
8106 | ||
57d885fe GH |
8107 | memset(rd, 0, sizeof(*rd)); |
8108 | ||
36b7b6d4 PE |
8109 | if (bootmem) |
8110 | gfp = GFP_NOWAIT; | |
c6c4927b | 8111 | |
36b7b6d4 | 8112 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 8113 | goto out; |
36b7b6d4 | 8114 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 8115 | goto free_span; |
36b7b6d4 | 8116 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 8117 | goto free_online; |
6e0534f2 | 8118 | |
0fb53029 | 8119 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 8120 | goto free_rto_mask; |
c6c4927b | 8121 | return 0; |
6e0534f2 | 8122 | |
68e74568 RR |
8123 | free_rto_mask: |
8124 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
8125 | free_online: |
8126 | free_cpumask_var(rd->online); | |
8127 | free_span: | |
8128 | free_cpumask_var(rd->span); | |
0c910d28 | 8129 | out: |
c6c4927b | 8130 | return -ENOMEM; |
57d885fe GH |
8131 | } |
8132 | ||
8133 | static void init_defrootdomain(void) | |
8134 | { | |
c6c4927b RR |
8135 | init_rootdomain(&def_root_domain, true); |
8136 | ||
57d885fe GH |
8137 | atomic_set(&def_root_domain.refcount, 1); |
8138 | } | |
8139 | ||
dc938520 | 8140 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
8141 | { |
8142 | struct root_domain *rd; | |
8143 | ||
8144 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
8145 | if (!rd) | |
8146 | return NULL; | |
8147 | ||
c6c4927b RR |
8148 | if (init_rootdomain(rd, false) != 0) { |
8149 | kfree(rd); | |
8150 | return NULL; | |
8151 | } | |
57d885fe GH |
8152 | |
8153 | return rd; | |
8154 | } | |
8155 | ||
1da177e4 | 8156 | /* |
0eab9146 | 8157 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
8158 | * hold the hotplug lock. |
8159 | */ | |
0eab9146 IM |
8160 | static void |
8161 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 8162 | { |
70b97a7f | 8163 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
8164 | struct sched_domain *tmp; |
8165 | ||
8166 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 8167 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
8168 | struct sched_domain *parent = tmp->parent; |
8169 | if (!parent) | |
8170 | break; | |
f29c9b1c | 8171 | |
1a848870 | 8172 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 8173 | tmp->parent = parent->parent; |
1a848870 SS |
8174 | if (parent->parent) |
8175 | parent->parent->child = tmp; | |
f29c9b1c LZ |
8176 | } else |
8177 | tmp = tmp->parent; | |
245af2c7 SS |
8178 | } |
8179 | ||
1a848870 | 8180 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 8181 | sd = sd->parent; |
1a848870 SS |
8182 | if (sd) |
8183 | sd->child = NULL; | |
8184 | } | |
1da177e4 LT |
8185 | |
8186 | sched_domain_debug(sd, cpu); | |
8187 | ||
57d885fe | 8188 | rq_attach_root(rq, rd); |
674311d5 | 8189 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
8190 | } |
8191 | ||
8192 | /* cpus with isolated domains */ | |
dcc30a35 | 8193 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
8194 | |
8195 | /* Setup the mask of cpus configured for isolated domains */ | |
8196 | static int __init isolated_cpu_setup(char *str) | |
8197 | { | |
968ea6d8 | 8198 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
8199 | return 1; |
8200 | } | |
8201 | ||
8927f494 | 8202 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
8203 | |
8204 | /* | |
6711cab4 SS |
8205 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
8206 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
8207 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
8208 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
8209 | * |
8210 | * init_sched_build_groups will build a circular linked list of the groups | |
8211 | * covered by the given span, and will set each group's ->cpumask correctly, | |
8212 | * and ->cpu_power to 0. | |
8213 | */ | |
a616058b | 8214 | static void |
96f874e2 RR |
8215 | init_sched_build_groups(const struct cpumask *span, |
8216 | const struct cpumask *cpu_map, | |
8217 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 8218 | struct sched_group **sg, |
96f874e2 RR |
8219 | struct cpumask *tmpmask), |
8220 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
8221 | { |
8222 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
8223 | int i; |
8224 | ||
96f874e2 | 8225 | cpumask_clear(covered); |
7c16ec58 | 8226 | |
abcd083a | 8227 | for_each_cpu(i, span) { |
6711cab4 | 8228 | struct sched_group *sg; |
7c16ec58 | 8229 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
8230 | int j; |
8231 | ||
758b2cdc | 8232 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
8233 | continue; |
8234 | ||
758b2cdc | 8235 | cpumask_clear(sched_group_cpus(sg)); |
5517d86b | 8236 | sg->__cpu_power = 0; |
1da177e4 | 8237 | |
abcd083a | 8238 | for_each_cpu(j, span) { |
7c16ec58 | 8239 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
8240 | continue; |
8241 | ||
96f874e2 | 8242 | cpumask_set_cpu(j, covered); |
758b2cdc | 8243 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
8244 | } |
8245 | if (!first) | |
8246 | first = sg; | |
8247 | if (last) | |
8248 | last->next = sg; | |
8249 | last = sg; | |
8250 | } | |
8251 | last->next = first; | |
8252 | } | |
8253 | ||
9c1cfda2 | 8254 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 8255 | |
9c1cfda2 | 8256 | #ifdef CONFIG_NUMA |
198e2f18 | 8257 | |
9c1cfda2 JH |
8258 | /** |
8259 | * find_next_best_node - find the next node to include in a sched_domain | |
8260 | * @node: node whose sched_domain we're building | |
8261 | * @used_nodes: nodes already in the sched_domain | |
8262 | * | |
41a2d6cf | 8263 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
8264 | * finds the closest node not already in the @used_nodes map. |
8265 | * | |
8266 | * Should use nodemask_t. | |
8267 | */ | |
c5f59f08 | 8268 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
8269 | { |
8270 | int i, n, val, min_val, best_node = 0; | |
8271 | ||
8272 | min_val = INT_MAX; | |
8273 | ||
076ac2af | 8274 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 8275 | /* Start at @node */ |
076ac2af | 8276 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
8277 | |
8278 | if (!nr_cpus_node(n)) | |
8279 | continue; | |
8280 | ||
8281 | /* Skip already used nodes */ | |
c5f59f08 | 8282 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
8283 | continue; |
8284 | ||
8285 | /* Simple min distance search */ | |
8286 | val = node_distance(node, n); | |
8287 | ||
8288 | if (val < min_val) { | |
8289 | min_val = val; | |
8290 | best_node = n; | |
8291 | } | |
8292 | } | |
8293 | ||
c5f59f08 | 8294 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
8295 | return best_node; |
8296 | } | |
8297 | ||
8298 | /** | |
8299 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
8300 | * @node: node whose cpumask we're constructing | |
73486722 | 8301 | * @span: resulting cpumask |
9c1cfda2 | 8302 | * |
41a2d6cf | 8303 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
8304 | * should be one that prevents unnecessary balancing, but also spreads tasks |
8305 | * out optimally. | |
8306 | */ | |
96f874e2 | 8307 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 8308 | { |
c5f59f08 | 8309 | nodemask_t used_nodes; |
48f24c4d | 8310 | int i; |
9c1cfda2 | 8311 | |
6ca09dfc | 8312 | cpumask_clear(span); |
c5f59f08 | 8313 | nodes_clear(used_nodes); |
9c1cfda2 | 8314 | |
6ca09dfc | 8315 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 8316 | node_set(node, used_nodes); |
9c1cfda2 JH |
8317 | |
8318 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 8319 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 8320 | |
6ca09dfc | 8321 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 8322 | } |
9c1cfda2 | 8323 | } |
6d6bc0ad | 8324 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 8325 | |
5c45bf27 | 8326 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 8327 | |
6c99e9ad RR |
8328 | /* |
8329 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
8330 | * |
8331 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
8332 | * and struct sched_domain. ) | |
6c99e9ad RR |
8333 | */ |
8334 | struct static_sched_group { | |
8335 | struct sched_group sg; | |
8336 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
8337 | }; | |
8338 | ||
8339 | struct static_sched_domain { | |
8340 | struct sched_domain sd; | |
8341 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
8342 | }; | |
8343 | ||
49a02c51 AH |
8344 | struct s_data { |
8345 | #ifdef CONFIG_NUMA | |
8346 | int sd_allnodes; | |
8347 | cpumask_var_t domainspan; | |
8348 | cpumask_var_t covered; | |
8349 | cpumask_var_t notcovered; | |
8350 | #endif | |
8351 | cpumask_var_t nodemask; | |
8352 | cpumask_var_t this_sibling_map; | |
8353 | cpumask_var_t this_core_map; | |
8354 | cpumask_var_t send_covered; | |
8355 | cpumask_var_t tmpmask; | |
8356 | struct sched_group **sched_group_nodes; | |
8357 | struct root_domain *rd; | |
8358 | }; | |
8359 | ||
2109b99e AH |
8360 | enum s_alloc { |
8361 | sa_sched_groups = 0, | |
8362 | sa_rootdomain, | |
8363 | sa_tmpmask, | |
8364 | sa_send_covered, | |
8365 | sa_this_core_map, | |
8366 | sa_this_sibling_map, | |
8367 | sa_nodemask, | |
8368 | sa_sched_group_nodes, | |
8369 | #ifdef CONFIG_NUMA | |
8370 | sa_notcovered, | |
8371 | sa_covered, | |
8372 | sa_domainspan, | |
8373 | #endif | |
8374 | sa_none, | |
8375 | }; | |
8376 | ||
9c1cfda2 | 8377 | /* |
48f24c4d | 8378 | * SMT sched-domains: |
9c1cfda2 | 8379 | */ |
1da177e4 | 8380 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad RR |
8381 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
8382 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_cpus); | |
48f24c4d | 8383 | |
41a2d6cf | 8384 | static int |
96f874e2 RR |
8385 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
8386 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 8387 | { |
6711cab4 | 8388 | if (sg) |
6c99e9ad | 8389 | *sg = &per_cpu(sched_group_cpus, cpu).sg; |
1da177e4 LT |
8390 | return cpu; |
8391 | } | |
6d6bc0ad | 8392 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 8393 | |
48f24c4d IM |
8394 | /* |
8395 | * multi-core sched-domains: | |
8396 | */ | |
1e9f28fa | 8397 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
8398 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
8399 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 8400 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
8401 | |
8402 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 8403 | static int |
96f874e2 RR |
8404 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8405 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 8406 | { |
6711cab4 | 8407 | int group; |
7c16ec58 | 8408 | |
c69fc56d | 8409 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8410 | group = cpumask_first(mask); |
6711cab4 | 8411 | if (sg) |
6c99e9ad | 8412 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 8413 | return group; |
1e9f28fa SS |
8414 | } |
8415 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 8416 | static int |
96f874e2 RR |
8417 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
8418 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 8419 | { |
6711cab4 | 8420 | if (sg) |
6c99e9ad | 8421 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
8422 | return cpu; |
8423 | } | |
8424 | #endif | |
8425 | ||
6c99e9ad RR |
8426 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
8427 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 8428 | |
41a2d6cf | 8429 | static int |
96f874e2 RR |
8430 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
8431 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 8432 | { |
6711cab4 | 8433 | int group; |
48f24c4d | 8434 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 8435 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 8436 | group = cpumask_first(mask); |
1e9f28fa | 8437 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 8438 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 8439 | group = cpumask_first(mask); |
1da177e4 | 8440 | #else |
6711cab4 | 8441 | group = cpu; |
1da177e4 | 8442 | #endif |
6711cab4 | 8443 | if (sg) |
6c99e9ad | 8444 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 8445 | return group; |
1da177e4 LT |
8446 | } |
8447 | ||
8448 | #ifdef CONFIG_NUMA | |
1da177e4 | 8449 | /* |
9c1cfda2 JH |
8450 | * The init_sched_build_groups can't handle what we want to do with node |
8451 | * groups, so roll our own. Now each node has its own list of groups which | |
8452 | * gets dynamically allocated. | |
1da177e4 | 8453 | */ |
62ea9ceb | 8454 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 8455 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 8456 | |
62ea9ceb | 8457 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 8458 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 8459 | |
96f874e2 RR |
8460 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
8461 | struct sched_group **sg, | |
8462 | struct cpumask *nodemask) | |
9c1cfda2 | 8463 | { |
6711cab4 SS |
8464 | int group; |
8465 | ||
6ca09dfc | 8466 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 8467 | group = cpumask_first(nodemask); |
6711cab4 SS |
8468 | |
8469 | if (sg) | |
6c99e9ad | 8470 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 8471 | return group; |
1da177e4 | 8472 | } |
6711cab4 | 8473 | |
08069033 SS |
8474 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
8475 | { | |
8476 | struct sched_group *sg = group_head; | |
8477 | int j; | |
8478 | ||
8479 | if (!sg) | |
8480 | return; | |
3a5c359a | 8481 | do { |
758b2cdc | 8482 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 8483 | struct sched_domain *sd; |
08069033 | 8484 | |
6c99e9ad | 8485 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 8486 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
8487 | /* |
8488 | * Only add "power" once for each | |
8489 | * physical package. | |
8490 | */ | |
8491 | continue; | |
8492 | } | |
08069033 | 8493 | |
3a5c359a AK |
8494 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
8495 | } | |
8496 | sg = sg->next; | |
8497 | } while (sg != group_head); | |
08069033 | 8498 | } |
0601a88d AH |
8499 | |
8500 | static int build_numa_sched_groups(struct s_data *d, | |
8501 | const struct cpumask *cpu_map, int num) | |
8502 | { | |
8503 | struct sched_domain *sd; | |
8504 | struct sched_group *sg, *prev; | |
8505 | int n, j; | |
8506 | ||
8507 | cpumask_clear(d->covered); | |
8508 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
8509 | if (cpumask_empty(d->nodemask)) { | |
8510 | d->sched_group_nodes[num] = NULL; | |
8511 | goto out; | |
8512 | } | |
8513 | ||
8514 | sched_domain_node_span(num, d->domainspan); | |
8515 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
8516 | ||
8517 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8518 | GFP_KERNEL, num); | |
8519 | if (!sg) { | |
8520 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", | |
8521 | num); | |
8522 | return -ENOMEM; | |
8523 | } | |
8524 | d->sched_group_nodes[num] = sg; | |
8525 | ||
8526 | for_each_cpu(j, d->nodemask) { | |
8527 | sd = &per_cpu(node_domains, j).sd; | |
8528 | sd->groups = sg; | |
8529 | } | |
8530 | ||
8531 | sg->__cpu_power = 0; | |
8532 | cpumask_copy(sched_group_cpus(sg), d->nodemask); | |
8533 | sg->next = sg; | |
8534 | cpumask_or(d->covered, d->covered, d->nodemask); | |
8535 | ||
8536 | prev = sg; | |
8537 | for (j = 0; j < nr_node_ids; j++) { | |
8538 | n = (num + j) % nr_node_ids; | |
8539 | cpumask_complement(d->notcovered, d->covered); | |
8540 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
8541 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
8542 | if (cpumask_empty(d->tmpmask)) | |
8543 | break; | |
8544 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
8545 | if (cpumask_empty(d->tmpmask)) | |
8546 | continue; | |
8547 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
8548 | GFP_KERNEL, num); | |
8549 | if (!sg) { | |
8550 | printk(KERN_WARNING | |
8551 | "Can not alloc domain group for node %d\n", j); | |
8552 | return -ENOMEM; | |
8553 | } | |
8554 | sg->__cpu_power = 0; | |
8555 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); | |
8556 | sg->next = prev->next; | |
8557 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
8558 | prev->next = sg; | |
8559 | prev = sg; | |
8560 | } | |
8561 | out: | |
8562 | return 0; | |
8563 | } | |
6d6bc0ad | 8564 | #endif /* CONFIG_NUMA */ |
1da177e4 | 8565 | |
a616058b | 8566 | #ifdef CONFIG_NUMA |
51888ca2 | 8567 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
8568 | static void free_sched_groups(const struct cpumask *cpu_map, |
8569 | struct cpumask *nodemask) | |
51888ca2 | 8570 | { |
a616058b | 8571 | int cpu, i; |
51888ca2 | 8572 | |
abcd083a | 8573 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
8574 | struct sched_group **sched_group_nodes |
8575 | = sched_group_nodes_bycpu[cpu]; | |
8576 | ||
51888ca2 SV |
8577 | if (!sched_group_nodes) |
8578 | continue; | |
8579 | ||
076ac2af | 8580 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
8581 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
8582 | ||
6ca09dfc | 8583 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 8584 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
8585 | continue; |
8586 | ||
8587 | if (sg == NULL) | |
8588 | continue; | |
8589 | sg = sg->next; | |
8590 | next_sg: | |
8591 | oldsg = sg; | |
8592 | sg = sg->next; | |
8593 | kfree(oldsg); | |
8594 | if (oldsg != sched_group_nodes[i]) | |
8595 | goto next_sg; | |
8596 | } | |
8597 | kfree(sched_group_nodes); | |
8598 | sched_group_nodes_bycpu[cpu] = NULL; | |
8599 | } | |
51888ca2 | 8600 | } |
6d6bc0ad | 8601 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
8602 | static void free_sched_groups(const struct cpumask *cpu_map, |
8603 | struct cpumask *nodemask) | |
a616058b SS |
8604 | { |
8605 | } | |
6d6bc0ad | 8606 | #endif /* CONFIG_NUMA */ |
51888ca2 | 8607 | |
89c4710e SS |
8608 | /* |
8609 | * Initialize sched groups cpu_power. | |
8610 | * | |
8611 | * cpu_power indicates the capacity of sched group, which is used while | |
8612 | * distributing the load between different sched groups in a sched domain. | |
8613 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
8614 | * there are asymmetries in the topology. If there are asymmetries, group | |
8615 | * having more cpu_power will pickup more load compared to the group having | |
8616 | * less cpu_power. | |
89c4710e SS |
8617 | */ |
8618 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
8619 | { | |
8620 | struct sched_domain *child; | |
8621 | struct sched_group *group; | |
f93e65c1 PZ |
8622 | long power; |
8623 | int weight; | |
89c4710e SS |
8624 | |
8625 | WARN_ON(!sd || !sd->groups); | |
8626 | ||
13318a71 | 8627 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
8628 | return; |
8629 | ||
8630 | child = sd->child; | |
8631 | ||
5517d86b ED |
8632 | sd->groups->__cpu_power = 0; |
8633 | ||
f93e65c1 PZ |
8634 | if (!child) { |
8635 | power = SCHED_LOAD_SCALE; | |
8636 | weight = cpumask_weight(sched_domain_span(sd)); | |
8637 | /* | |
8638 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
8639 | * Usually multiple threads get a better yield out of |
8640 | * that one core than a single thread would have, | |
8641 | * reflect that in sd->smt_gain. | |
f93e65c1 | 8642 | */ |
a52bfd73 PZ |
8643 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
8644 | power *= sd->smt_gain; | |
f93e65c1 | 8645 | power /= weight; |
a52bfd73 PZ |
8646 | power >>= SCHED_LOAD_SHIFT; |
8647 | } | |
f93e65c1 | 8648 | sg_inc_cpu_power(sd->groups, power); |
89c4710e SS |
8649 | return; |
8650 | } | |
8651 | ||
89c4710e | 8652 | /* |
f93e65c1 | 8653 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
8654 | */ |
8655 | group = child->groups; | |
8656 | do { | |
5517d86b | 8657 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
8658 | group = group->next; |
8659 | } while (group != child->groups); | |
8660 | } | |
8661 | ||
7c16ec58 MT |
8662 | /* |
8663 | * Initializers for schedule domains | |
8664 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
8665 | */ | |
8666 | ||
a5d8c348 IM |
8667 | #ifdef CONFIG_SCHED_DEBUG |
8668 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
8669 | #else | |
8670 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
8671 | #endif | |
8672 | ||
7c16ec58 | 8673 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 8674 | |
7c16ec58 MT |
8675 | #define SD_INIT_FUNC(type) \ |
8676 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
8677 | { \ | |
8678 | memset(sd, 0, sizeof(*sd)); \ | |
8679 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 8680 | sd->level = SD_LV_##type; \ |
a5d8c348 | 8681 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
8682 | } |
8683 | ||
8684 | SD_INIT_FUNC(CPU) | |
8685 | #ifdef CONFIG_NUMA | |
8686 | SD_INIT_FUNC(ALLNODES) | |
8687 | SD_INIT_FUNC(NODE) | |
8688 | #endif | |
8689 | #ifdef CONFIG_SCHED_SMT | |
8690 | SD_INIT_FUNC(SIBLING) | |
8691 | #endif | |
8692 | #ifdef CONFIG_SCHED_MC | |
8693 | SD_INIT_FUNC(MC) | |
8694 | #endif | |
8695 | ||
1d3504fc HS |
8696 | static int default_relax_domain_level = -1; |
8697 | ||
8698 | static int __init setup_relax_domain_level(char *str) | |
8699 | { | |
30e0e178 LZ |
8700 | unsigned long val; |
8701 | ||
8702 | val = simple_strtoul(str, NULL, 0); | |
8703 | if (val < SD_LV_MAX) | |
8704 | default_relax_domain_level = val; | |
8705 | ||
1d3504fc HS |
8706 | return 1; |
8707 | } | |
8708 | __setup("relax_domain_level=", setup_relax_domain_level); | |
8709 | ||
8710 | static void set_domain_attribute(struct sched_domain *sd, | |
8711 | struct sched_domain_attr *attr) | |
8712 | { | |
8713 | int request; | |
8714 | ||
8715 | if (!attr || attr->relax_domain_level < 0) { | |
8716 | if (default_relax_domain_level < 0) | |
8717 | return; | |
8718 | else | |
8719 | request = default_relax_domain_level; | |
8720 | } else | |
8721 | request = attr->relax_domain_level; | |
8722 | if (request < sd->level) { | |
8723 | /* turn off idle balance on this domain */ | |
8724 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
8725 | } else { | |
8726 | /* turn on idle balance on this domain */ | |
8727 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
8728 | } | |
8729 | } | |
8730 | ||
2109b99e AH |
8731 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
8732 | const struct cpumask *cpu_map) | |
8733 | { | |
8734 | switch (what) { | |
8735 | case sa_sched_groups: | |
8736 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
8737 | d->sched_group_nodes = NULL; | |
8738 | case sa_rootdomain: | |
8739 | free_rootdomain(d->rd); /* fall through */ | |
8740 | case sa_tmpmask: | |
8741 | free_cpumask_var(d->tmpmask); /* fall through */ | |
8742 | case sa_send_covered: | |
8743 | free_cpumask_var(d->send_covered); /* fall through */ | |
8744 | case sa_this_core_map: | |
8745 | free_cpumask_var(d->this_core_map); /* fall through */ | |
8746 | case sa_this_sibling_map: | |
8747 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
8748 | case sa_nodemask: | |
8749 | free_cpumask_var(d->nodemask); /* fall through */ | |
8750 | case sa_sched_group_nodes: | |
d1b55138 | 8751 | #ifdef CONFIG_NUMA |
2109b99e AH |
8752 | kfree(d->sched_group_nodes); /* fall through */ |
8753 | case sa_notcovered: | |
8754 | free_cpumask_var(d->notcovered); /* fall through */ | |
8755 | case sa_covered: | |
8756 | free_cpumask_var(d->covered); /* fall through */ | |
8757 | case sa_domainspan: | |
8758 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 8759 | #endif |
2109b99e AH |
8760 | case sa_none: |
8761 | break; | |
8762 | } | |
8763 | } | |
3404c8d9 | 8764 | |
2109b99e AH |
8765 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
8766 | const struct cpumask *cpu_map) | |
8767 | { | |
3404c8d9 | 8768 | #ifdef CONFIG_NUMA |
2109b99e AH |
8769 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
8770 | return sa_none; | |
8771 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
8772 | return sa_domainspan; | |
8773 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
8774 | return sa_covered; | |
8775 | /* Allocate the per-node list of sched groups */ | |
8776 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
8777 | sizeof(struct sched_group *), GFP_KERNEL); | |
8778 | if (!d->sched_group_nodes) { | |
d1b55138 | 8779 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 8780 | return sa_notcovered; |
d1b55138 | 8781 | } |
2109b99e | 8782 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 8783 | #endif |
2109b99e AH |
8784 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
8785 | return sa_sched_group_nodes; | |
8786 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
8787 | return sa_nodemask; | |
8788 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
8789 | return sa_this_sibling_map; | |
8790 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
8791 | return sa_this_core_map; | |
8792 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
8793 | return sa_send_covered; | |
8794 | d->rd = alloc_rootdomain(); | |
8795 | if (!d->rd) { | |
57d885fe | 8796 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 8797 | return sa_tmpmask; |
57d885fe | 8798 | } |
2109b99e AH |
8799 | return sa_rootdomain; |
8800 | } | |
57d885fe | 8801 | |
7f4588f3 AH |
8802 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
8803 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
8804 | { | |
8805 | struct sched_domain *sd = NULL; | |
7c16ec58 | 8806 | #ifdef CONFIG_NUMA |
7f4588f3 | 8807 | struct sched_domain *parent; |
9c1cfda2 | 8808 | |
7f4588f3 AH |
8809 | d->sd_allnodes = 0; |
8810 | if (cpumask_weight(cpu_map) > | |
8811 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
8812 | sd = &per_cpu(allnodes_domains, i).sd; | |
8813 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 8814 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
8815 | cpumask_copy(sched_domain_span(sd), cpu_map); |
8816 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8817 | d->sd_allnodes = 1; | |
8818 | } | |
8819 | parent = sd; | |
8820 | ||
8821 | sd = &per_cpu(node_domains, i).sd; | |
8822 | SD_INIT(sd, NODE); | |
8823 | set_domain_attribute(sd, attr); | |
8824 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
8825 | sd->parent = parent; | |
8826 | if (parent) | |
8827 | parent->child = sd; | |
8828 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 8829 | #endif |
7f4588f3 AH |
8830 | return sd; |
8831 | } | |
1da177e4 | 8832 | |
87cce662 AH |
8833 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
8834 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8835 | struct sched_domain *parent, int i) | |
8836 | { | |
8837 | struct sched_domain *sd; | |
8838 | sd = &per_cpu(phys_domains, i).sd; | |
8839 | SD_INIT(sd, CPU); | |
8840 | set_domain_attribute(sd, attr); | |
8841 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
8842 | sd->parent = parent; | |
8843 | if (parent) | |
8844 | parent->child = sd; | |
8845 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
8846 | return sd; | |
8847 | } | |
1da177e4 | 8848 | |
410c4081 AH |
8849 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
8850 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8851 | struct sched_domain *parent, int i) | |
8852 | { | |
8853 | struct sched_domain *sd = parent; | |
1e9f28fa | 8854 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
8855 | sd = &per_cpu(core_domains, i).sd; |
8856 | SD_INIT(sd, MC); | |
8857 | set_domain_attribute(sd, attr); | |
8858 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
8859 | sd->parent = parent; | |
8860 | parent->child = sd; | |
8861 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 8862 | #endif |
410c4081 AH |
8863 | return sd; |
8864 | } | |
1e9f28fa | 8865 | |
d8173535 AH |
8866 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
8867 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
8868 | struct sched_domain *parent, int i) | |
8869 | { | |
8870 | struct sched_domain *sd = parent; | |
1da177e4 | 8871 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
8872 | sd = &per_cpu(cpu_domains, i).sd; |
8873 | SD_INIT(sd, SIBLING); | |
8874 | set_domain_attribute(sd, attr); | |
8875 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
8876 | sd->parent = parent; | |
8877 | parent->child = sd; | |
8878 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 8879 | #endif |
d8173535 AH |
8880 | return sd; |
8881 | } | |
1da177e4 | 8882 | |
0e8e85c9 AH |
8883 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
8884 | const struct cpumask *cpu_map, int cpu) | |
8885 | { | |
8886 | switch (l) { | |
1da177e4 | 8887 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
8888 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
8889 | cpumask_and(d->this_sibling_map, cpu_map, | |
8890 | topology_thread_cpumask(cpu)); | |
8891 | if (cpu == cpumask_first(d->this_sibling_map)) | |
8892 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
8893 | &cpu_to_cpu_group, | |
8894 | d->send_covered, d->tmpmask); | |
8895 | break; | |
1da177e4 | 8896 | #endif |
1e9f28fa | 8897 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
8898 | case SD_LV_MC: /* set up multi-core groups */ |
8899 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
8900 | if (cpu == cpumask_first(d->this_core_map)) | |
8901 | init_sched_build_groups(d->this_core_map, cpu_map, | |
8902 | &cpu_to_core_group, | |
8903 | d->send_covered, d->tmpmask); | |
8904 | break; | |
1e9f28fa | 8905 | #endif |
86548096 AH |
8906 | case SD_LV_CPU: /* set up physical groups */ |
8907 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
8908 | if (!cpumask_empty(d->nodemask)) | |
8909 | init_sched_build_groups(d->nodemask, cpu_map, | |
8910 | &cpu_to_phys_group, | |
8911 | d->send_covered, d->tmpmask); | |
8912 | break; | |
1da177e4 | 8913 | #ifdef CONFIG_NUMA |
de616e36 AH |
8914 | case SD_LV_ALLNODES: |
8915 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
8916 | d->send_covered, d->tmpmask); | |
8917 | break; | |
8918 | #endif | |
0e8e85c9 AH |
8919 | default: |
8920 | break; | |
7c16ec58 | 8921 | } |
0e8e85c9 | 8922 | } |
9c1cfda2 | 8923 | |
2109b99e AH |
8924 | /* |
8925 | * Build sched domains for a given set of cpus and attach the sched domains | |
8926 | * to the individual cpus | |
8927 | */ | |
8928 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
8929 | struct sched_domain_attr *attr) | |
8930 | { | |
8931 | enum s_alloc alloc_state = sa_none; | |
8932 | struct s_data d; | |
294b0c96 | 8933 | struct sched_domain *sd; |
2109b99e | 8934 | int i; |
7c16ec58 | 8935 | #ifdef CONFIG_NUMA |
2109b99e | 8936 | d.sd_allnodes = 0; |
7c16ec58 | 8937 | #endif |
9c1cfda2 | 8938 | |
2109b99e AH |
8939 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
8940 | if (alloc_state != sa_rootdomain) | |
8941 | goto error; | |
8942 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 8943 | |
1da177e4 | 8944 | /* |
1a20ff27 | 8945 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 8946 | */ |
abcd083a | 8947 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
8948 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
8949 | cpu_map); | |
9761eea8 | 8950 | |
7f4588f3 | 8951 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 8952 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 8953 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 8954 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 8955 | } |
9c1cfda2 | 8956 | |
abcd083a | 8957 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 8958 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 8959 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 8960 | } |
9c1cfda2 | 8961 | |
1da177e4 | 8962 | /* Set up physical groups */ |
86548096 AH |
8963 | for (i = 0; i < nr_node_ids; i++) |
8964 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 8965 | |
1da177e4 LT |
8966 | #ifdef CONFIG_NUMA |
8967 | /* Set up node groups */ | |
de616e36 AH |
8968 | if (d.sd_allnodes) |
8969 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 8970 | |
0601a88d AH |
8971 | for (i = 0; i < nr_node_ids; i++) |
8972 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 8973 | goto error; |
1da177e4 LT |
8974 | #endif |
8975 | ||
8976 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 8977 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 8978 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8979 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 8980 | init_sched_groups_power(i, sd); |
5c45bf27 | 8981 | } |
1da177e4 | 8982 | #endif |
1e9f28fa | 8983 | #ifdef CONFIG_SCHED_MC |
abcd083a | 8984 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8985 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 8986 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
8987 | } |
8988 | #endif | |
1e9f28fa | 8989 | |
abcd083a | 8990 | for_each_cpu(i, cpu_map) { |
294b0c96 | 8991 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 8992 | init_sched_groups_power(i, sd); |
1da177e4 LT |
8993 | } |
8994 | ||
9c1cfda2 | 8995 | #ifdef CONFIG_NUMA |
076ac2af | 8996 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 8997 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 8998 | |
49a02c51 | 8999 | if (d.sd_allnodes) { |
6711cab4 | 9000 | struct sched_group *sg; |
f712c0c7 | 9001 | |
96f874e2 | 9002 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 9003 | d.tmpmask); |
f712c0c7 SS |
9004 | init_numa_sched_groups_power(sg); |
9005 | } | |
9c1cfda2 JH |
9006 | #endif |
9007 | ||
1da177e4 | 9008 | /* Attach the domains */ |
abcd083a | 9009 | for_each_cpu(i, cpu_map) { |
1da177e4 | 9010 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 9011 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 9012 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 9013 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 9014 | #else |
6c99e9ad | 9015 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 9016 | #endif |
49a02c51 | 9017 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 9018 | } |
3404c8d9 | 9019 | |
2109b99e AH |
9020 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
9021 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
9022 | return 0; | |
51888ca2 | 9023 | |
51888ca2 | 9024 | error: |
2109b99e AH |
9025 | __free_domain_allocs(&d, alloc_state, cpu_map); |
9026 | return -ENOMEM; | |
1da177e4 | 9027 | } |
029190c5 | 9028 | |
96f874e2 | 9029 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
9030 | { |
9031 | return __build_sched_domains(cpu_map, NULL); | |
9032 | } | |
9033 | ||
96f874e2 | 9034 | static struct cpumask *doms_cur; /* current sched domains */ |
029190c5 | 9035 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
9036 | static struct sched_domain_attr *dattr_cur; |
9037 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
9038 | |
9039 | /* | |
9040 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
9041 | * cpumask) fails, then fallback to a single sched domain, |
9042 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 9043 | */ |
4212823f | 9044 | static cpumask_var_t fallback_doms; |
029190c5 | 9045 | |
ee79d1bd HC |
9046 | /* |
9047 | * arch_update_cpu_topology lets virtualized architectures update the | |
9048 | * cpu core maps. It is supposed to return 1 if the topology changed | |
9049 | * or 0 if it stayed the same. | |
9050 | */ | |
9051 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 9052 | { |
ee79d1bd | 9053 | return 0; |
22e52b07 HC |
9054 | } |
9055 | ||
1a20ff27 | 9056 | /* |
41a2d6cf | 9057 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
9058 | * For now this just excludes isolated cpus, but could be used to |
9059 | * exclude other special cases in the future. | |
1a20ff27 | 9060 | */ |
96f874e2 | 9061 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 9062 | { |
7378547f MM |
9063 | int err; |
9064 | ||
22e52b07 | 9065 | arch_update_cpu_topology(); |
029190c5 | 9066 | ndoms_cur = 1; |
96f874e2 | 9067 | doms_cur = kmalloc(cpumask_size(), GFP_KERNEL); |
029190c5 | 9068 | if (!doms_cur) |
4212823f | 9069 | doms_cur = fallback_doms; |
dcc30a35 | 9070 | cpumask_andnot(doms_cur, cpu_map, cpu_isolated_map); |
1d3504fc | 9071 | dattr_cur = NULL; |
7378547f | 9072 | err = build_sched_domains(doms_cur); |
6382bc90 | 9073 | register_sched_domain_sysctl(); |
7378547f MM |
9074 | |
9075 | return err; | |
1a20ff27 DG |
9076 | } |
9077 | ||
96f874e2 RR |
9078 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
9079 | struct cpumask *tmpmask) | |
1da177e4 | 9080 | { |
7c16ec58 | 9081 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 9082 | } |
1da177e4 | 9083 | |
1a20ff27 DG |
9084 | /* |
9085 | * Detach sched domains from a group of cpus specified in cpu_map | |
9086 | * These cpus will now be attached to the NULL domain | |
9087 | */ | |
96f874e2 | 9088 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 9089 | { |
96f874e2 RR |
9090 | /* Save because hotplug lock held. */ |
9091 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
9092 | int i; |
9093 | ||
abcd083a | 9094 | for_each_cpu(i, cpu_map) |
57d885fe | 9095 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 9096 | synchronize_sched(); |
96f874e2 | 9097 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
9098 | } |
9099 | ||
1d3504fc HS |
9100 | /* handle null as "default" */ |
9101 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
9102 | struct sched_domain_attr *new, int idx_new) | |
9103 | { | |
9104 | struct sched_domain_attr tmp; | |
9105 | ||
9106 | /* fast path */ | |
9107 | if (!new && !cur) | |
9108 | return 1; | |
9109 | ||
9110 | tmp = SD_ATTR_INIT; | |
9111 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
9112 | new ? (new + idx_new) : &tmp, | |
9113 | sizeof(struct sched_domain_attr)); | |
9114 | } | |
9115 | ||
029190c5 PJ |
9116 | /* |
9117 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 9118 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
9119 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
9120 | * It destroys each deleted domain and builds each new domain. | |
9121 | * | |
96f874e2 | 9122 | * 'doms_new' is an array of cpumask's of length 'ndoms_new'. |
41a2d6cf IM |
9123 | * The masks don't intersect (don't overlap.) We should setup one |
9124 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
9125 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
9126 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
9127 | * it as it is. | |
9128 | * | |
41a2d6cf IM |
9129 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
9130 | * ownership of it and will kfree it when done with it. If the caller | |
700018e0 LZ |
9131 | * failed the kmalloc call, then it can pass in doms_new == NULL && |
9132 | * ndoms_new == 1, and partition_sched_domains() will fallback to | |
9133 | * the single partition 'fallback_doms', it also forces the domains | |
9134 | * to be rebuilt. | |
029190c5 | 9135 | * |
96f874e2 | 9136 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
9137 | * ndoms_new == 0 is a special case for destroying existing domains, |
9138 | * and it will not create the default domain. | |
dfb512ec | 9139 | * |
029190c5 PJ |
9140 | * Call with hotplug lock held |
9141 | */ | |
96f874e2 RR |
9142 | /* FIXME: Change to struct cpumask *doms_new[] */ |
9143 | void partition_sched_domains(int ndoms_new, struct cpumask *doms_new, | |
1d3504fc | 9144 | struct sched_domain_attr *dattr_new) |
029190c5 | 9145 | { |
dfb512ec | 9146 | int i, j, n; |
d65bd5ec | 9147 | int new_topology; |
029190c5 | 9148 | |
712555ee | 9149 | mutex_lock(&sched_domains_mutex); |
a1835615 | 9150 | |
7378547f MM |
9151 | /* always unregister in case we don't destroy any domains */ |
9152 | unregister_sched_domain_sysctl(); | |
9153 | ||
d65bd5ec HC |
9154 | /* Let architecture update cpu core mappings. */ |
9155 | new_topology = arch_update_cpu_topology(); | |
9156 | ||
dfb512ec | 9157 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
9158 | |
9159 | /* Destroy deleted domains */ | |
9160 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 9161 | for (j = 0; j < n && !new_topology; j++) { |
96f874e2 | 9162 | if (cpumask_equal(&doms_cur[i], &doms_new[j]) |
1d3504fc | 9163 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
9164 | goto match1; |
9165 | } | |
9166 | /* no match - a current sched domain not in new doms_new[] */ | |
9167 | detach_destroy_domains(doms_cur + i); | |
9168 | match1: | |
9169 | ; | |
9170 | } | |
9171 | ||
e761b772 MK |
9172 | if (doms_new == NULL) { |
9173 | ndoms_cur = 0; | |
4212823f | 9174 | doms_new = fallback_doms; |
dcc30a35 | 9175 | cpumask_andnot(&doms_new[0], cpu_online_mask, cpu_isolated_map); |
faa2f98f | 9176 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
9177 | } |
9178 | ||
029190c5 PJ |
9179 | /* Build new domains */ |
9180 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 9181 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
96f874e2 | 9182 | if (cpumask_equal(&doms_new[i], &doms_cur[j]) |
1d3504fc | 9183 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
9184 | goto match2; |
9185 | } | |
9186 | /* no match - add a new doms_new */ | |
1d3504fc HS |
9187 | __build_sched_domains(doms_new + i, |
9188 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
9189 | match2: |
9190 | ; | |
9191 | } | |
9192 | ||
9193 | /* Remember the new sched domains */ | |
4212823f | 9194 | if (doms_cur != fallback_doms) |
029190c5 | 9195 | kfree(doms_cur); |
1d3504fc | 9196 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 9197 | doms_cur = doms_new; |
1d3504fc | 9198 | dattr_cur = dattr_new; |
029190c5 | 9199 | ndoms_cur = ndoms_new; |
7378547f MM |
9200 | |
9201 | register_sched_domain_sysctl(); | |
a1835615 | 9202 | |
712555ee | 9203 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
9204 | } |
9205 | ||
5c45bf27 | 9206 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 9207 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 9208 | { |
95402b38 | 9209 | get_online_cpus(); |
dfb512ec MK |
9210 | |
9211 | /* Destroy domains first to force the rebuild */ | |
9212 | partition_sched_domains(0, NULL, NULL); | |
9213 | ||
e761b772 | 9214 | rebuild_sched_domains(); |
95402b38 | 9215 | put_online_cpus(); |
5c45bf27 SS |
9216 | } |
9217 | ||
9218 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
9219 | { | |
afb8a9b7 | 9220 | unsigned int level = 0; |
5c45bf27 | 9221 | |
afb8a9b7 GS |
9222 | if (sscanf(buf, "%u", &level) != 1) |
9223 | return -EINVAL; | |
9224 | ||
9225 | /* | |
9226 | * level is always be positive so don't check for | |
9227 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
9228 | * What happens on 0 or 1 byte write, | |
9229 | * need to check for count as well? | |
9230 | */ | |
9231 | ||
9232 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
9233 | return -EINVAL; |
9234 | ||
9235 | if (smt) | |
afb8a9b7 | 9236 | sched_smt_power_savings = level; |
5c45bf27 | 9237 | else |
afb8a9b7 | 9238 | sched_mc_power_savings = level; |
5c45bf27 | 9239 | |
c70f22d2 | 9240 | arch_reinit_sched_domains(); |
5c45bf27 | 9241 | |
c70f22d2 | 9242 | return count; |
5c45bf27 SS |
9243 | } |
9244 | ||
5c45bf27 | 9245 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
9246 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
9247 | char *page) | |
5c45bf27 SS |
9248 | { |
9249 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
9250 | } | |
f718cd4a | 9251 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 9252 | const char *buf, size_t count) |
5c45bf27 SS |
9253 | { |
9254 | return sched_power_savings_store(buf, count, 0); | |
9255 | } | |
f718cd4a AK |
9256 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
9257 | sched_mc_power_savings_show, | |
9258 | sched_mc_power_savings_store); | |
5c45bf27 SS |
9259 | #endif |
9260 | ||
9261 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
9262 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
9263 | char *page) | |
5c45bf27 SS |
9264 | { |
9265 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
9266 | } | |
f718cd4a | 9267 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 9268 | const char *buf, size_t count) |
5c45bf27 SS |
9269 | { |
9270 | return sched_power_savings_store(buf, count, 1); | |
9271 | } | |
f718cd4a AK |
9272 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
9273 | sched_smt_power_savings_show, | |
6707de00 AB |
9274 | sched_smt_power_savings_store); |
9275 | #endif | |
9276 | ||
39aac648 | 9277 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
9278 | { |
9279 | int err = 0; | |
9280 | ||
9281 | #ifdef CONFIG_SCHED_SMT | |
9282 | if (smt_capable()) | |
9283 | err = sysfs_create_file(&cls->kset.kobj, | |
9284 | &attr_sched_smt_power_savings.attr); | |
9285 | #endif | |
9286 | #ifdef CONFIG_SCHED_MC | |
9287 | if (!err && mc_capable()) | |
9288 | err = sysfs_create_file(&cls->kset.kobj, | |
9289 | &attr_sched_mc_power_savings.attr); | |
9290 | #endif | |
9291 | return err; | |
9292 | } | |
6d6bc0ad | 9293 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 9294 | |
e761b772 | 9295 | #ifndef CONFIG_CPUSETS |
1da177e4 | 9296 | /* |
e761b772 MK |
9297 | * Add online and remove offline CPUs from the scheduler domains. |
9298 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
9299 | */ |
9300 | static int update_sched_domains(struct notifier_block *nfb, | |
9301 | unsigned long action, void *hcpu) | |
e761b772 MK |
9302 | { |
9303 | switch (action) { | |
9304 | case CPU_ONLINE: | |
9305 | case CPU_ONLINE_FROZEN: | |
9306 | case CPU_DEAD: | |
9307 | case CPU_DEAD_FROZEN: | |
dfb512ec | 9308 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
9309 | return NOTIFY_OK; |
9310 | ||
9311 | default: | |
9312 | return NOTIFY_DONE; | |
9313 | } | |
9314 | } | |
9315 | #endif | |
9316 | ||
9317 | static int update_runtime(struct notifier_block *nfb, | |
9318 | unsigned long action, void *hcpu) | |
1da177e4 | 9319 | { |
7def2be1 PZ |
9320 | int cpu = (int)(long)hcpu; |
9321 | ||
1da177e4 | 9322 | switch (action) { |
1da177e4 | 9323 | case CPU_DOWN_PREPARE: |
8bb78442 | 9324 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 9325 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
9326 | return NOTIFY_OK; |
9327 | ||
1da177e4 | 9328 | case CPU_DOWN_FAILED: |
8bb78442 | 9329 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 9330 | case CPU_ONLINE: |
8bb78442 | 9331 | case CPU_ONLINE_FROZEN: |
7def2be1 | 9332 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
9333 | return NOTIFY_OK; |
9334 | ||
1da177e4 LT |
9335 | default: |
9336 | return NOTIFY_DONE; | |
9337 | } | |
1da177e4 | 9338 | } |
1da177e4 LT |
9339 | |
9340 | void __init sched_init_smp(void) | |
9341 | { | |
dcc30a35 RR |
9342 | cpumask_var_t non_isolated_cpus; |
9343 | ||
9344 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
5c1e1767 | 9345 | |
434d53b0 MT |
9346 | #if defined(CONFIG_NUMA) |
9347 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
9348 | GFP_KERNEL); | |
9349 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
9350 | #endif | |
95402b38 | 9351 | get_online_cpus(); |
712555ee | 9352 | mutex_lock(&sched_domains_mutex); |
dcc30a35 RR |
9353 | arch_init_sched_domains(cpu_online_mask); |
9354 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); | |
9355 | if (cpumask_empty(non_isolated_cpus)) | |
9356 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 9357 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 9358 | put_online_cpus(); |
e761b772 MK |
9359 | |
9360 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
9361 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
9362 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
9363 | #endif |
9364 | ||
9365 | /* RT runtime code needs to handle some hotplug events */ | |
9366 | hotcpu_notifier(update_runtime, 0); | |
9367 | ||
b328ca18 | 9368 | init_hrtick(); |
5c1e1767 NP |
9369 | |
9370 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 9371 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 9372 | BUG(); |
19978ca6 | 9373 | sched_init_granularity(); |
dcc30a35 | 9374 | free_cpumask_var(non_isolated_cpus); |
4212823f RR |
9375 | |
9376 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); | |
0e3900e6 | 9377 | init_sched_rt_class(); |
1da177e4 LT |
9378 | } |
9379 | #else | |
9380 | void __init sched_init_smp(void) | |
9381 | { | |
19978ca6 | 9382 | sched_init_granularity(); |
1da177e4 LT |
9383 | } |
9384 | #endif /* CONFIG_SMP */ | |
9385 | ||
cd1bb94b AB |
9386 | const_debug unsigned int sysctl_timer_migration = 1; |
9387 | ||
1da177e4 LT |
9388 | int in_sched_functions(unsigned long addr) |
9389 | { | |
1da177e4 LT |
9390 | return in_lock_functions(addr) || |
9391 | (addr >= (unsigned long)__sched_text_start | |
9392 | && addr < (unsigned long)__sched_text_end); | |
9393 | } | |
9394 | ||
a9957449 | 9395 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
9396 | { |
9397 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 9398 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
9399 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9400 | cfs_rq->rq = rq; | |
9401 | #endif | |
67e9fb2a | 9402 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
9403 | } |
9404 | ||
fa85ae24 PZ |
9405 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
9406 | { | |
9407 | struct rt_prio_array *array; | |
9408 | int i; | |
9409 | ||
9410 | array = &rt_rq->active; | |
9411 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
9412 | INIT_LIST_HEAD(array->queue + i); | |
9413 | __clear_bit(i, array->bitmap); | |
9414 | } | |
9415 | /* delimiter for bitsearch: */ | |
9416 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
9417 | ||
052f1dc7 | 9418 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 9419 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 9420 | #ifdef CONFIG_SMP |
e864c499 | 9421 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 9422 | #endif |
48d5e258 | 9423 | #endif |
fa85ae24 PZ |
9424 | #ifdef CONFIG_SMP |
9425 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 9426 | rt_rq->overloaded = 0; |
c20b08e3 | 9427 | plist_head_init(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
9428 | #endif |
9429 | ||
9430 | rt_rq->rt_time = 0; | |
9431 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
9432 | rt_rq->rt_runtime = 0; |
9433 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 9434 | |
052f1dc7 | 9435 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 9436 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
9437 | rt_rq->rq = rq; |
9438 | #endif | |
fa85ae24 PZ |
9439 | } |
9440 | ||
6f505b16 | 9441 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
9442 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
9443 | struct sched_entity *se, int cpu, int add, | |
9444 | struct sched_entity *parent) | |
6f505b16 | 9445 | { |
ec7dc8ac | 9446 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
9447 | tg->cfs_rq[cpu] = cfs_rq; |
9448 | init_cfs_rq(cfs_rq, rq); | |
9449 | cfs_rq->tg = tg; | |
9450 | if (add) | |
9451 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
9452 | ||
9453 | tg->se[cpu] = se; | |
354d60c2 DG |
9454 | /* se could be NULL for init_task_group */ |
9455 | if (!se) | |
9456 | return; | |
9457 | ||
ec7dc8ac DG |
9458 | if (!parent) |
9459 | se->cfs_rq = &rq->cfs; | |
9460 | else | |
9461 | se->cfs_rq = parent->my_q; | |
9462 | ||
6f505b16 PZ |
9463 | se->my_q = cfs_rq; |
9464 | se->load.weight = tg->shares; | |
e05510d0 | 9465 | se->load.inv_weight = 0; |
ec7dc8ac | 9466 | se->parent = parent; |
6f505b16 | 9467 | } |
052f1dc7 | 9468 | #endif |
6f505b16 | 9469 | |
052f1dc7 | 9470 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
9471 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
9472 | struct sched_rt_entity *rt_se, int cpu, int add, | |
9473 | struct sched_rt_entity *parent) | |
6f505b16 | 9474 | { |
ec7dc8ac DG |
9475 | struct rq *rq = cpu_rq(cpu); |
9476 | ||
6f505b16 PZ |
9477 | tg->rt_rq[cpu] = rt_rq; |
9478 | init_rt_rq(rt_rq, rq); | |
9479 | rt_rq->tg = tg; | |
9480 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 9481 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
9482 | if (add) |
9483 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
9484 | ||
9485 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
9486 | if (!rt_se) |
9487 | return; | |
9488 | ||
ec7dc8ac DG |
9489 | if (!parent) |
9490 | rt_se->rt_rq = &rq->rt; | |
9491 | else | |
9492 | rt_se->rt_rq = parent->my_q; | |
9493 | ||
6f505b16 | 9494 | rt_se->my_q = rt_rq; |
ec7dc8ac | 9495 | rt_se->parent = parent; |
6f505b16 PZ |
9496 | INIT_LIST_HEAD(&rt_se->run_list); |
9497 | } | |
9498 | #endif | |
9499 | ||
1da177e4 LT |
9500 | void __init sched_init(void) |
9501 | { | |
dd41f596 | 9502 | int i, j; |
434d53b0 MT |
9503 | unsigned long alloc_size = 0, ptr; |
9504 | ||
9505 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9506 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
9507 | #endif | |
9508 | #ifdef CONFIG_RT_GROUP_SCHED | |
9509 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9510 | #endif |
9511 | #ifdef CONFIG_USER_SCHED | |
9512 | alloc_size *= 2; | |
df7c8e84 RR |
9513 | #endif |
9514 | #ifdef CONFIG_CPUMASK_OFFSTACK | |
8c083f08 | 9515 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 MT |
9516 | #endif |
9517 | /* | |
9518 | * As sched_init() is called before page_alloc is setup, | |
9519 | * we use alloc_bootmem(). | |
9520 | */ | |
9521 | if (alloc_size) { | |
36b7b6d4 | 9522 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
9523 | |
9524 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
9525 | init_task_group.se = (struct sched_entity **)ptr; | |
9526 | ptr += nr_cpu_ids * sizeof(void **); | |
9527 | ||
9528 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9529 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
9530 | |
9531 | #ifdef CONFIG_USER_SCHED | |
9532 | root_task_group.se = (struct sched_entity **)ptr; | |
9533 | ptr += nr_cpu_ids * sizeof(void **); | |
9534 | ||
9535 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
9536 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9537 | #endif /* CONFIG_USER_SCHED */ |
9538 | #endif /* CONFIG_FAIR_GROUP_SCHED */ | |
434d53b0 MT |
9539 | #ifdef CONFIG_RT_GROUP_SCHED |
9540 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9541 | ptr += nr_cpu_ids * sizeof(void **); | |
9542 | ||
9543 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
9544 | ptr += nr_cpu_ids * sizeof(void **); |
9545 | ||
9546 | #ifdef CONFIG_USER_SCHED | |
9547 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
9548 | ptr += nr_cpu_ids * sizeof(void **); | |
9549 | ||
9550 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
9551 | ptr += nr_cpu_ids * sizeof(void **); | |
6d6bc0ad DG |
9552 | #endif /* CONFIG_USER_SCHED */ |
9553 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
df7c8e84 RR |
9554 | #ifdef CONFIG_CPUMASK_OFFSTACK |
9555 | for_each_possible_cpu(i) { | |
9556 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
9557 | ptr += cpumask_size(); | |
9558 | } | |
9559 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 9560 | } |
dd41f596 | 9561 | |
57d885fe GH |
9562 | #ifdef CONFIG_SMP |
9563 | init_defrootdomain(); | |
9564 | #endif | |
9565 | ||
d0b27fa7 PZ |
9566 | init_rt_bandwidth(&def_rt_bandwidth, |
9567 | global_rt_period(), global_rt_runtime()); | |
9568 | ||
9569 | #ifdef CONFIG_RT_GROUP_SCHED | |
9570 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
9571 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
9572 | #ifdef CONFIG_USER_SCHED |
9573 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
9574 | global_rt_period(), RUNTIME_INF); | |
6d6bc0ad DG |
9575 | #endif /* CONFIG_USER_SCHED */ |
9576 | #endif /* CONFIG_RT_GROUP_SCHED */ | |
d0b27fa7 | 9577 | |
052f1dc7 | 9578 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 9579 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
9580 | INIT_LIST_HEAD(&init_task_group.children); |
9581 | ||
9582 | #ifdef CONFIG_USER_SCHED | |
9583 | INIT_LIST_HEAD(&root_task_group.children); | |
9584 | init_task_group.parent = &root_task_group; | |
9585 | list_add(&init_task_group.siblings, &root_task_group.children); | |
6d6bc0ad DG |
9586 | #endif /* CONFIG_USER_SCHED */ |
9587 | #endif /* CONFIG_GROUP_SCHED */ | |
6f505b16 | 9588 | |
0a945022 | 9589 | for_each_possible_cpu(i) { |
70b97a7f | 9590 | struct rq *rq; |
1da177e4 LT |
9591 | |
9592 | rq = cpu_rq(i); | |
9593 | spin_lock_init(&rq->lock); | |
7897986b | 9594 | rq->nr_running = 0; |
dce48a84 TG |
9595 | rq->calc_load_active = 0; |
9596 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 9597 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 9598 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 9599 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 9600 | init_task_group.shares = init_task_group_load; |
6f505b16 | 9601 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
9602 | #ifdef CONFIG_CGROUP_SCHED |
9603 | /* | |
9604 | * How much cpu bandwidth does init_task_group get? | |
9605 | * | |
9606 | * In case of task-groups formed thr' the cgroup filesystem, it | |
9607 | * gets 100% of the cpu resources in the system. This overall | |
9608 | * system cpu resource is divided among the tasks of | |
9609 | * init_task_group and its child task-groups in a fair manner, | |
9610 | * based on each entity's (task or task-group's) weight | |
9611 | * (se->load.weight). | |
9612 | * | |
9613 | * In other words, if init_task_group has 10 tasks of weight | |
9614 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
9615 | * then A0's share of the cpu resource is: | |
9616 | * | |
0d905bca | 9617 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
9618 | * |
9619 | * We achieve this by letting init_task_group's tasks sit | |
9620 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
9621 | */ | |
ec7dc8ac | 9622 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 9623 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
9624 | root_task_group.shares = NICE_0_LOAD; |
9625 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
9626 | /* |
9627 | * In case of task-groups formed thr' the user id of tasks, | |
9628 | * init_task_group represents tasks belonging to root user. | |
9629 | * Hence it forms a sibling of all subsequent groups formed. | |
9630 | * In this case, init_task_group gets only a fraction of overall | |
9631 | * system cpu resource, based on the weight assigned to root | |
9632 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
9633 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
84e9dabf | 9634 | * (init_tg_cfs_rq) and having one entity represent this group of |
354d60c2 DG |
9635 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). |
9636 | */ | |
ec7dc8ac | 9637 | init_tg_cfs_entry(&init_task_group, |
84e9dabf | 9638 | &per_cpu(init_tg_cfs_rq, i), |
eff766a6 PZ |
9639 | &per_cpu(init_sched_entity, i), i, 1, |
9640 | root_task_group.se[i]); | |
6f505b16 | 9641 | |
052f1dc7 | 9642 | #endif |
354d60c2 DG |
9643 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
9644 | ||
9645 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 9646 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 9647 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 9648 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 9649 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 9650 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 9651 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 9652 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 9653 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
9654 | &per_cpu(init_sched_rt_entity, i), i, 1, |
9655 | root_task_group.rt_se[i]); | |
354d60c2 | 9656 | #endif |
dd41f596 | 9657 | #endif |
1da177e4 | 9658 | |
dd41f596 IM |
9659 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
9660 | rq->cpu_load[j] = 0; | |
1da177e4 | 9661 | #ifdef CONFIG_SMP |
41c7ce9a | 9662 | rq->sd = NULL; |
57d885fe | 9663 | rq->rd = NULL; |
3f029d3c | 9664 | rq->post_schedule = 0; |
1da177e4 | 9665 | rq->active_balance = 0; |
dd41f596 | 9666 | rq->next_balance = jiffies; |
1da177e4 | 9667 | rq->push_cpu = 0; |
0a2966b4 | 9668 | rq->cpu = i; |
1f11eb6a | 9669 | rq->online = 0; |
1da177e4 LT |
9670 | rq->migration_thread = NULL; |
9671 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 9672 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 9673 | #endif |
8f4d37ec | 9674 | init_rq_hrtick(rq); |
1da177e4 | 9675 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
9676 | } |
9677 | ||
2dd73a4f | 9678 | set_load_weight(&init_task); |
b50f60ce | 9679 | |
e107be36 AK |
9680 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
9681 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
9682 | #endif | |
9683 | ||
c9819f45 | 9684 | #ifdef CONFIG_SMP |
962cf36c | 9685 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
9686 | #endif |
9687 | ||
b50f60ce HC |
9688 | #ifdef CONFIG_RT_MUTEXES |
9689 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
9690 | #endif | |
9691 | ||
1da177e4 LT |
9692 | /* |
9693 | * The boot idle thread does lazy MMU switching as well: | |
9694 | */ | |
9695 | atomic_inc(&init_mm.mm_count); | |
9696 | enter_lazy_tlb(&init_mm, current); | |
9697 | ||
9698 | /* | |
9699 | * Make us the idle thread. Technically, schedule() should not be | |
9700 | * called from this thread, however somewhere below it might be, | |
9701 | * but because we are the idle thread, we just pick up running again | |
9702 | * when this runqueue becomes "idle". | |
9703 | */ | |
9704 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
9705 | |
9706 | calc_load_update = jiffies + LOAD_FREQ; | |
9707 | ||
dd41f596 IM |
9708 | /* |
9709 | * During early bootup we pretend to be a normal task: | |
9710 | */ | |
9711 | current->sched_class = &fair_sched_class; | |
6892b75e | 9712 | |
6a7b3dc3 | 9713 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
4bdddf8f | 9714 | alloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 9715 | #ifdef CONFIG_SMP |
7d1e6a9b | 9716 | #ifdef CONFIG_NO_HZ |
4bdddf8f PE |
9717 | alloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
9718 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); | |
7d1e6a9b | 9719 | #endif |
4bdddf8f | 9720 | alloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); |
bf4d83f6 | 9721 | #endif /* SMP */ |
6a7b3dc3 | 9722 | |
0d905bca IM |
9723 | perf_counter_init(); |
9724 | ||
6892b75e | 9725 | scheduler_running = 1; |
1da177e4 LT |
9726 | } |
9727 | ||
9728 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
9729 | static inline int preempt_count_equals(int preempt_offset) |
9730 | { | |
9731 | int nested = preempt_count() & ~PREEMPT_ACTIVE; | |
9732 | ||
9733 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
9734 | } | |
9735 | ||
9736 | void __might_sleep(char *file, int line, int preempt_offset) | |
1da177e4 | 9737 | { |
48f24c4d | 9738 | #ifdef in_atomic |
1da177e4 LT |
9739 | static unsigned long prev_jiffy; /* ratelimiting */ |
9740 | ||
e4aafea2 FW |
9741 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
9742 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
9743 | return; |
9744 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
9745 | return; | |
9746 | prev_jiffy = jiffies; | |
9747 | ||
9748 | printk(KERN_ERR | |
9749 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
9750 | file, line); | |
9751 | printk(KERN_ERR | |
9752 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
9753 | in_atomic(), irqs_disabled(), | |
9754 | current->pid, current->comm); | |
9755 | ||
9756 | debug_show_held_locks(current); | |
9757 | if (irqs_disabled()) | |
9758 | print_irqtrace_events(current); | |
9759 | dump_stack(); | |
1da177e4 LT |
9760 | #endif |
9761 | } | |
9762 | EXPORT_SYMBOL(__might_sleep); | |
9763 | #endif | |
9764 | ||
9765 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
9766 | static void normalize_task(struct rq *rq, struct task_struct *p) |
9767 | { | |
9768 | int on_rq; | |
3e51f33f | 9769 | |
3a5e4dc1 AK |
9770 | update_rq_clock(rq); |
9771 | on_rq = p->se.on_rq; | |
9772 | if (on_rq) | |
9773 | deactivate_task(rq, p, 0); | |
9774 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
9775 | if (on_rq) { | |
9776 | activate_task(rq, p, 0); | |
9777 | resched_task(rq->curr); | |
9778 | } | |
9779 | } | |
9780 | ||
1da177e4 LT |
9781 | void normalize_rt_tasks(void) |
9782 | { | |
a0f98a1c | 9783 | struct task_struct *g, *p; |
1da177e4 | 9784 | unsigned long flags; |
70b97a7f | 9785 | struct rq *rq; |
1da177e4 | 9786 | |
4cf5d77a | 9787 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 9788 | do_each_thread(g, p) { |
178be793 IM |
9789 | /* |
9790 | * Only normalize user tasks: | |
9791 | */ | |
9792 | if (!p->mm) | |
9793 | continue; | |
9794 | ||
6cfb0d5d | 9795 | p->se.exec_start = 0; |
6cfb0d5d | 9796 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 9797 | p->se.wait_start = 0; |
dd41f596 | 9798 | p->se.sleep_start = 0; |
dd41f596 | 9799 | p->se.block_start = 0; |
6cfb0d5d | 9800 | #endif |
dd41f596 IM |
9801 | |
9802 | if (!rt_task(p)) { | |
9803 | /* | |
9804 | * Renice negative nice level userspace | |
9805 | * tasks back to 0: | |
9806 | */ | |
9807 | if (TASK_NICE(p) < 0 && p->mm) | |
9808 | set_user_nice(p, 0); | |
1da177e4 | 9809 | continue; |
dd41f596 | 9810 | } |
1da177e4 | 9811 | |
4cf5d77a | 9812 | spin_lock(&p->pi_lock); |
b29739f9 | 9813 | rq = __task_rq_lock(p); |
1da177e4 | 9814 | |
178be793 | 9815 | normalize_task(rq, p); |
3a5e4dc1 | 9816 | |
b29739f9 | 9817 | __task_rq_unlock(rq); |
4cf5d77a | 9818 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
9819 | } while_each_thread(g, p); |
9820 | ||
4cf5d77a | 9821 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
9822 | } |
9823 | ||
9824 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
9825 | |
9826 | #ifdef CONFIG_IA64 | |
9827 | /* | |
9828 | * These functions are only useful for the IA64 MCA handling. | |
9829 | * | |
9830 | * They can only be called when the whole system has been | |
9831 | * stopped - every CPU needs to be quiescent, and no scheduling | |
9832 | * activity can take place. Using them for anything else would | |
9833 | * be a serious bug, and as a result, they aren't even visible | |
9834 | * under any other configuration. | |
9835 | */ | |
9836 | ||
9837 | /** | |
9838 | * curr_task - return the current task for a given cpu. | |
9839 | * @cpu: the processor in question. | |
9840 | * | |
9841 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9842 | */ | |
36c8b586 | 9843 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
9844 | { |
9845 | return cpu_curr(cpu); | |
9846 | } | |
9847 | ||
9848 | /** | |
9849 | * set_curr_task - set the current task for a given cpu. | |
9850 | * @cpu: the processor in question. | |
9851 | * @p: the task pointer to set. | |
9852 | * | |
9853 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
9854 | * are serviced on a separate stack. It allows the architecture to switch the |
9855 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
9856 | * must be called with all CPU's synchronized, and interrupts disabled, the |
9857 | * and caller must save the original value of the current task (see | |
9858 | * curr_task() above) and restore that value before reenabling interrupts and | |
9859 | * re-starting the system. | |
9860 | * | |
9861 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
9862 | */ | |
36c8b586 | 9863 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
9864 | { |
9865 | cpu_curr(cpu) = p; | |
9866 | } | |
9867 | ||
9868 | #endif | |
29f59db3 | 9869 | |
bccbe08a PZ |
9870 | #ifdef CONFIG_FAIR_GROUP_SCHED |
9871 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
9872 | { |
9873 | int i; | |
9874 | ||
9875 | for_each_possible_cpu(i) { | |
9876 | if (tg->cfs_rq) | |
9877 | kfree(tg->cfs_rq[i]); | |
9878 | if (tg->se) | |
9879 | kfree(tg->se[i]); | |
6f505b16 PZ |
9880 | } |
9881 | ||
9882 | kfree(tg->cfs_rq); | |
9883 | kfree(tg->se); | |
6f505b16 PZ |
9884 | } |
9885 | ||
ec7dc8ac DG |
9886 | static |
9887 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 9888 | { |
29f59db3 | 9889 | struct cfs_rq *cfs_rq; |
eab17229 | 9890 | struct sched_entity *se; |
9b5b7751 | 9891 | struct rq *rq; |
29f59db3 SV |
9892 | int i; |
9893 | ||
434d53b0 | 9894 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9895 | if (!tg->cfs_rq) |
9896 | goto err; | |
434d53b0 | 9897 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
9898 | if (!tg->se) |
9899 | goto err; | |
052f1dc7 PZ |
9900 | |
9901 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
9902 | |
9903 | for_each_possible_cpu(i) { | |
9b5b7751 | 9904 | rq = cpu_rq(i); |
29f59db3 | 9905 | |
eab17229 LZ |
9906 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
9907 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9908 | if (!cfs_rq) |
9909 | goto err; | |
9910 | ||
eab17229 LZ |
9911 | se = kzalloc_node(sizeof(struct sched_entity), |
9912 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
9913 | if (!se) |
9914 | goto err; | |
9915 | ||
eab17229 | 9916 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
9917 | } |
9918 | ||
9919 | return 1; | |
9920 | ||
9921 | err: | |
9922 | return 0; | |
9923 | } | |
9924 | ||
9925 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9926 | { | |
9927 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
9928 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
9929 | } | |
9930 | ||
9931 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9932 | { | |
9933 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
9934 | } | |
6d6bc0ad | 9935 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
9936 | static inline void free_fair_sched_group(struct task_group *tg) |
9937 | { | |
9938 | } | |
9939 | ||
ec7dc8ac DG |
9940 | static inline |
9941 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9942 | { |
9943 | return 1; | |
9944 | } | |
9945 | ||
9946 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
9947 | { | |
9948 | } | |
9949 | ||
9950 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
9951 | { | |
9952 | } | |
6d6bc0ad | 9953 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
9954 | |
9955 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
9956 | static void free_rt_sched_group(struct task_group *tg) |
9957 | { | |
9958 | int i; | |
9959 | ||
d0b27fa7 PZ |
9960 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
9961 | ||
bccbe08a PZ |
9962 | for_each_possible_cpu(i) { |
9963 | if (tg->rt_rq) | |
9964 | kfree(tg->rt_rq[i]); | |
9965 | if (tg->rt_se) | |
9966 | kfree(tg->rt_se[i]); | |
9967 | } | |
9968 | ||
9969 | kfree(tg->rt_rq); | |
9970 | kfree(tg->rt_se); | |
9971 | } | |
9972 | ||
ec7dc8ac DG |
9973 | static |
9974 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
9975 | { |
9976 | struct rt_rq *rt_rq; | |
eab17229 | 9977 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
9978 | struct rq *rq; |
9979 | int i; | |
9980 | ||
434d53b0 | 9981 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9982 | if (!tg->rt_rq) |
9983 | goto err; | |
434d53b0 | 9984 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
9985 | if (!tg->rt_se) |
9986 | goto err; | |
9987 | ||
d0b27fa7 PZ |
9988 | init_rt_bandwidth(&tg->rt_bandwidth, |
9989 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
9990 | |
9991 | for_each_possible_cpu(i) { | |
9992 | rq = cpu_rq(i); | |
9993 | ||
eab17229 LZ |
9994 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
9995 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
9996 | if (!rt_rq) |
9997 | goto err; | |
29f59db3 | 9998 | |
eab17229 LZ |
9999 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
10000 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
10001 | if (!rt_se) |
10002 | goto err; | |
29f59db3 | 10003 | |
eab17229 | 10004 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
10005 | } |
10006 | ||
bccbe08a PZ |
10007 | return 1; |
10008 | ||
10009 | err: | |
10010 | return 0; | |
10011 | } | |
10012 | ||
10013 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
10014 | { | |
10015 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
10016 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
10017 | } | |
10018 | ||
10019 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
10020 | { | |
10021 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
10022 | } | |
6d6bc0ad | 10023 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
10024 | static inline void free_rt_sched_group(struct task_group *tg) |
10025 | { | |
10026 | } | |
10027 | ||
ec7dc8ac DG |
10028 | static inline |
10029 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
10030 | { |
10031 | return 1; | |
10032 | } | |
10033 | ||
10034 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
10035 | { | |
10036 | } | |
10037 | ||
10038 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
10039 | { | |
10040 | } | |
6d6bc0ad | 10041 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 10042 | |
d0b27fa7 | 10043 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
10044 | static void free_sched_group(struct task_group *tg) |
10045 | { | |
10046 | free_fair_sched_group(tg); | |
10047 | free_rt_sched_group(tg); | |
10048 | kfree(tg); | |
10049 | } | |
10050 | ||
10051 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 10052 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
10053 | { |
10054 | struct task_group *tg; | |
10055 | unsigned long flags; | |
10056 | int i; | |
10057 | ||
10058 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
10059 | if (!tg) | |
10060 | return ERR_PTR(-ENOMEM); | |
10061 | ||
ec7dc8ac | 10062 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
10063 | goto err; |
10064 | ||
ec7dc8ac | 10065 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
10066 | goto err; |
10067 | ||
8ed36996 | 10068 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 10069 | for_each_possible_cpu(i) { |
bccbe08a PZ |
10070 | register_fair_sched_group(tg, i); |
10071 | register_rt_sched_group(tg, i); | |
9b5b7751 | 10072 | } |
6f505b16 | 10073 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
10074 | |
10075 | WARN_ON(!parent); /* root should already exist */ | |
10076 | ||
10077 | tg->parent = parent; | |
f473aa5e | 10078 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 10079 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 10080 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 10081 | |
9b5b7751 | 10082 | return tg; |
29f59db3 SV |
10083 | |
10084 | err: | |
6f505b16 | 10085 | free_sched_group(tg); |
29f59db3 SV |
10086 | return ERR_PTR(-ENOMEM); |
10087 | } | |
10088 | ||
9b5b7751 | 10089 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 10090 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 10091 | { |
29f59db3 | 10092 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 10093 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
10094 | } |
10095 | ||
9b5b7751 | 10096 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 10097 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 10098 | { |
8ed36996 | 10099 | unsigned long flags; |
9b5b7751 | 10100 | int i; |
29f59db3 | 10101 | |
8ed36996 | 10102 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 10103 | for_each_possible_cpu(i) { |
bccbe08a PZ |
10104 | unregister_fair_sched_group(tg, i); |
10105 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 10106 | } |
6f505b16 | 10107 | list_del_rcu(&tg->list); |
f473aa5e | 10108 | list_del_rcu(&tg->siblings); |
8ed36996 | 10109 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 10110 | |
9b5b7751 | 10111 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 10112 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
10113 | } |
10114 | ||
9b5b7751 | 10115 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
10116 | * The caller of this function should have put the task in its new group |
10117 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
10118 | * reflect its new group. | |
9b5b7751 SV |
10119 | */ |
10120 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
10121 | { |
10122 | int on_rq, running; | |
10123 | unsigned long flags; | |
10124 | struct rq *rq; | |
10125 | ||
10126 | rq = task_rq_lock(tsk, &flags); | |
10127 | ||
29f59db3 SV |
10128 | update_rq_clock(rq); |
10129 | ||
051a1d1a | 10130 | running = task_current(rq, tsk); |
29f59db3 SV |
10131 | on_rq = tsk->se.on_rq; |
10132 | ||
0e1f3483 | 10133 | if (on_rq) |
29f59db3 | 10134 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
10135 | if (unlikely(running)) |
10136 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 10137 | |
6f505b16 | 10138 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 10139 | |
810b3817 PZ |
10140 | #ifdef CONFIG_FAIR_GROUP_SCHED |
10141 | if (tsk->sched_class->moved_group) | |
10142 | tsk->sched_class->moved_group(tsk); | |
10143 | #endif | |
10144 | ||
0e1f3483 HS |
10145 | if (unlikely(running)) |
10146 | tsk->sched_class->set_curr_task(rq); | |
10147 | if (on_rq) | |
7074badb | 10148 | enqueue_task(rq, tsk, 0); |
29f59db3 | 10149 | |
29f59db3 SV |
10150 | task_rq_unlock(rq, &flags); |
10151 | } | |
6d6bc0ad | 10152 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 10153 | |
052f1dc7 | 10154 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 10155 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
10156 | { |
10157 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
10158 | int on_rq; |
10159 | ||
29f59db3 | 10160 | on_rq = se->on_rq; |
62fb1851 | 10161 | if (on_rq) |
29f59db3 SV |
10162 | dequeue_entity(cfs_rq, se, 0); |
10163 | ||
10164 | se->load.weight = shares; | |
e05510d0 | 10165 | se->load.inv_weight = 0; |
29f59db3 | 10166 | |
62fb1851 | 10167 | if (on_rq) |
29f59db3 | 10168 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 10169 | } |
62fb1851 | 10170 | |
c09595f6 PZ |
10171 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
10172 | { | |
10173 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
10174 | struct rq *rq = cfs_rq->rq; | |
10175 | unsigned long flags; | |
10176 | ||
10177 | spin_lock_irqsave(&rq->lock, flags); | |
10178 | __set_se_shares(se, shares); | |
10179 | spin_unlock_irqrestore(&rq->lock, flags); | |
29f59db3 SV |
10180 | } |
10181 | ||
8ed36996 PZ |
10182 | static DEFINE_MUTEX(shares_mutex); |
10183 | ||
4cf86d77 | 10184 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
10185 | { |
10186 | int i; | |
8ed36996 | 10187 | unsigned long flags; |
c61935fd | 10188 | |
ec7dc8ac DG |
10189 | /* |
10190 | * We can't change the weight of the root cgroup. | |
10191 | */ | |
10192 | if (!tg->se[0]) | |
10193 | return -EINVAL; | |
10194 | ||
18d95a28 PZ |
10195 | if (shares < MIN_SHARES) |
10196 | shares = MIN_SHARES; | |
cb4ad1ff MX |
10197 | else if (shares > MAX_SHARES) |
10198 | shares = MAX_SHARES; | |
62fb1851 | 10199 | |
8ed36996 | 10200 | mutex_lock(&shares_mutex); |
9b5b7751 | 10201 | if (tg->shares == shares) |
5cb350ba | 10202 | goto done; |
29f59db3 | 10203 | |
8ed36996 | 10204 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10205 | for_each_possible_cpu(i) |
10206 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 10207 | list_del_rcu(&tg->siblings); |
8ed36996 | 10208 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
10209 | |
10210 | /* wait for any ongoing reference to this group to finish */ | |
10211 | synchronize_sched(); | |
10212 | ||
10213 | /* | |
10214 | * Now we are free to modify the group's share on each cpu | |
10215 | * w/o tripping rebalance_share or load_balance_fair. | |
10216 | */ | |
9b5b7751 | 10217 | tg->shares = shares; |
c09595f6 PZ |
10218 | for_each_possible_cpu(i) { |
10219 | /* | |
10220 | * force a rebalance | |
10221 | */ | |
10222 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 10223 | set_se_shares(tg->se[i], shares); |
c09595f6 | 10224 | } |
29f59db3 | 10225 | |
6b2d7700 SV |
10226 | /* |
10227 | * Enable load balance activity on this group, by inserting it back on | |
10228 | * each cpu's rq->leaf_cfs_rq_list. | |
10229 | */ | |
8ed36996 | 10230 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
10231 | for_each_possible_cpu(i) |
10232 | register_fair_sched_group(tg, i); | |
f473aa5e | 10233 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 10234 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 10235 | done: |
8ed36996 | 10236 | mutex_unlock(&shares_mutex); |
9b5b7751 | 10237 | return 0; |
29f59db3 SV |
10238 | } |
10239 | ||
5cb350ba DG |
10240 | unsigned long sched_group_shares(struct task_group *tg) |
10241 | { | |
10242 | return tg->shares; | |
10243 | } | |
052f1dc7 | 10244 | #endif |
5cb350ba | 10245 | |
052f1dc7 | 10246 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 10247 | /* |
9f0c1e56 | 10248 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 10249 | */ |
9f0c1e56 PZ |
10250 | static DEFINE_MUTEX(rt_constraints_mutex); |
10251 | ||
10252 | static unsigned long to_ratio(u64 period, u64 runtime) | |
10253 | { | |
10254 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 10255 | return 1ULL << 20; |
9f0c1e56 | 10256 | |
9a7e0b18 | 10257 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
10258 | } |
10259 | ||
9a7e0b18 PZ |
10260 | /* Must be called with tasklist_lock held */ |
10261 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 10262 | { |
9a7e0b18 | 10263 | struct task_struct *g, *p; |
b40b2e8e | 10264 | |
9a7e0b18 PZ |
10265 | do_each_thread(g, p) { |
10266 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
10267 | return 1; | |
10268 | } while_each_thread(g, p); | |
b40b2e8e | 10269 | |
9a7e0b18 PZ |
10270 | return 0; |
10271 | } | |
b40b2e8e | 10272 | |
9a7e0b18 PZ |
10273 | struct rt_schedulable_data { |
10274 | struct task_group *tg; | |
10275 | u64 rt_period; | |
10276 | u64 rt_runtime; | |
10277 | }; | |
b40b2e8e | 10278 | |
9a7e0b18 PZ |
10279 | static int tg_schedulable(struct task_group *tg, void *data) |
10280 | { | |
10281 | struct rt_schedulable_data *d = data; | |
10282 | struct task_group *child; | |
10283 | unsigned long total, sum = 0; | |
10284 | u64 period, runtime; | |
b40b2e8e | 10285 | |
9a7e0b18 PZ |
10286 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
10287 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 10288 | |
9a7e0b18 PZ |
10289 | if (tg == d->tg) { |
10290 | period = d->rt_period; | |
10291 | runtime = d->rt_runtime; | |
b40b2e8e | 10292 | } |
b40b2e8e | 10293 | |
98a4826b PZ |
10294 | #ifdef CONFIG_USER_SCHED |
10295 | if (tg == &root_task_group) { | |
10296 | period = global_rt_period(); | |
10297 | runtime = global_rt_runtime(); | |
10298 | } | |
10299 | #endif | |
10300 | ||
4653f803 PZ |
10301 | /* |
10302 | * Cannot have more runtime than the period. | |
10303 | */ | |
10304 | if (runtime > period && runtime != RUNTIME_INF) | |
10305 | return -EINVAL; | |
6f505b16 | 10306 | |
4653f803 PZ |
10307 | /* |
10308 | * Ensure we don't starve existing RT tasks. | |
10309 | */ | |
9a7e0b18 PZ |
10310 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
10311 | return -EBUSY; | |
6f505b16 | 10312 | |
9a7e0b18 | 10313 | total = to_ratio(period, runtime); |
6f505b16 | 10314 | |
4653f803 PZ |
10315 | /* |
10316 | * Nobody can have more than the global setting allows. | |
10317 | */ | |
10318 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
10319 | return -EINVAL; | |
6f505b16 | 10320 | |
4653f803 PZ |
10321 | /* |
10322 | * The sum of our children's runtime should not exceed our own. | |
10323 | */ | |
9a7e0b18 PZ |
10324 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
10325 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
10326 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 10327 | |
9a7e0b18 PZ |
10328 | if (child == d->tg) { |
10329 | period = d->rt_period; | |
10330 | runtime = d->rt_runtime; | |
10331 | } | |
6f505b16 | 10332 | |
9a7e0b18 | 10333 | sum += to_ratio(period, runtime); |
9f0c1e56 | 10334 | } |
6f505b16 | 10335 | |
9a7e0b18 PZ |
10336 | if (sum > total) |
10337 | return -EINVAL; | |
10338 | ||
10339 | return 0; | |
6f505b16 PZ |
10340 | } |
10341 | ||
9a7e0b18 | 10342 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 10343 | { |
9a7e0b18 PZ |
10344 | struct rt_schedulable_data data = { |
10345 | .tg = tg, | |
10346 | .rt_period = period, | |
10347 | .rt_runtime = runtime, | |
10348 | }; | |
10349 | ||
10350 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
10351 | } |
10352 | ||
d0b27fa7 PZ |
10353 | static int tg_set_bandwidth(struct task_group *tg, |
10354 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 10355 | { |
ac086bc2 | 10356 | int i, err = 0; |
9f0c1e56 | 10357 | |
9f0c1e56 | 10358 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 10359 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
10360 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
10361 | if (err) | |
9f0c1e56 | 10362 | goto unlock; |
ac086bc2 PZ |
10363 | |
10364 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
10365 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
10366 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
10367 | |
10368 | for_each_possible_cpu(i) { | |
10369 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
10370 | ||
10371 | spin_lock(&rt_rq->rt_runtime_lock); | |
10372 | rt_rq->rt_runtime = rt_runtime; | |
10373 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10374 | } | |
10375 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 10376 | unlock: |
521f1a24 | 10377 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
10378 | mutex_unlock(&rt_constraints_mutex); |
10379 | ||
10380 | return err; | |
6f505b16 PZ |
10381 | } |
10382 | ||
d0b27fa7 PZ |
10383 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
10384 | { | |
10385 | u64 rt_runtime, rt_period; | |
10386 | ||
10387 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10388 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
10389 | if (rt_runtime_us < 0) | |
10390 | rt_runtime = RUNTIME_INF; | |
10391 | ||
10392 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
10393 | } | |
10394 | ||
9f0c1e56 PZ |
10395 | long sched_group_rt_runtime(struct task_group *tg) |
10396 | { | |
10397 | u64 rt_runtime_us; | |
10398 | ||
d0b27fa7 | 10399 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
10400 | return -1; |
10401 | ||
d0b27fa7 | 10402 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
10403 | do_div(rt_runtime_us, NSEC_PER_USEC); |
10404 | return rt_runtime_us; | |
10405 | } | |
d0b27fa7 PZ |
10406 | |
10407 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
10408 | { | |
10409 | u64 rt_runtime, rt_period; | |
10410 | ||
10411 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
10412 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
10413 | ||
619b0488 R |
10414 | if (rt_period == 0) |
10415 | return -EINVAL; | |
10416 | ||
d0b27fa7 PZ |
10417 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
10418 | } | |
10419 | ||
10420 | long sched_group_rt_period(struct task_group *tg) | |
10421 | { | |
10422 | u64 rt_period_us; | |
10423 | ||
10424 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
10425 | do_div(rt_period_us, NSEC_PER_USEC); | |
10426 | return rt_period_us; | |
10427 | } | |
10428 | ||
10429 | static int sched_rt_global_constraints(void) | |
10430 | { | |
4653f803 | 10431 | u64 runtime, period; |
d0b27fa7 PZ |
10432 | int ret = 0; |
10433 | ||
ec5d4989 HS |
10434 | if (sysctl_sched_rt_period <= 0) |
10435 | return -EINVAL; | |
10436 | ||
4653f803 PZ |
10437 | runtime = global_rt_runtime(); |
10438 | period = global_rt_period(); | |
10439 | ||
10440 | /* | |
10441 | * Sanity check on the sysctl variables. | |
10442 | */ | |
10443 | if (runtime > period && runtime != RUNTIME_INF) | |
10444 | return -EINVAL; | |
10b612f4 | 10445 | |
d0b27fa7 | 10446 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 10447 | read_lock(&tasklist_lock); |
4653f803 | 10448 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 10449 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
10450 | mutex_unlock(&rt_constraints_mutex); |
10451 | ||
10452 | return ret; | |
10453 | } | |
54e99124 DG |
10454 | |
10455 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
10456 | { | |
10457 | /* Don't accept realtime tasks when there is no way for them to run */ | |
10458 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
10459 | return 0; | |
10460 | ||
10461 | return 1; | |
10462 | } | |
10463 | ||
6d6bc0ad | 10464 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10465 | static int sched_rt_global_constraints(void) |
10466 | { | |
ac086bc2 PZ |
10467 | unsigned long flags; |
10468 | int i; | |
10469 | ||
ec5d4989 HS |
10470 | if (sysctl_sched_rt_period <= 0) |
10471 | return -EINVAL; | |
10472 | ||
60aa605d PZ |
10473 | /* |
10474 | * There's always some RT tasks in the root group | |
10475 | * -- migration, kstopmachine etc.. | |
10476 | */ | |
10477 | if (sysctl_sched_rt_runtime == 0) | |
10478 | return -EBUSY; | |
10479 | ||
ac086bc2 PZ |
10480 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
10481 | for_each_possible_cpu(i) { | |
10482 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
10483 | ||
10484 | spin_lock(&rt_rq->rt_runtime_lock); | |
10485 | rt_rq->rt_runtime = global_rt_runtime(); | |
10486 | spin_unlock(&rt_rq->rt_runtime_lock); | |
10487 | } | |
10488 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
10489 | ||
d0b27fa7 PZ |
10490 | return 0; |
10491 | } | |
6d6bc0ad | 10492 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
10493 | |
10494 | int sched_rt_handler(struct ctl_table *table, int write, | |
10495 | struct file *filp, void __user *buffer, size_t *lenp, | |
10496 | loff_t *ppos) | |
10497 | { | |
10498 | int ret; | |
10499 | int old_period, old_runtime; | |
10500 | static DEFINE_MUTEX(mutex); | |
10501 | ||
10502 | mutex_lock(&mutex); | |
10503 | old_period = sysctl_sched_rt_period; | |
10504 | old_runtime = sysctl_sched_rt_runtime; | |
10505 | ||
10506 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
10507 | ||
10508 | if (!ret && write) { | |
10509 | ret = sched_rt_global_constraints(); | |
10510 | if (ret) { | |
10511 | sysctl_sched_rt_period = old_period; | |
10512 | sysctl_sched_rt_runtime = old_runtime; | |
10513 | } else { | |
10514 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
10515 | def_rt_bandwidth.rt_period = | |
10516 | ns_to_ktime(global_rt_period()); | |
10517 | } | |
10518 | } | |
10519 | mutex_unlock(&mutex); | |
10520 | ||
10521 | return ret; | |
10522 | } | |
68318b8e | 10523 | |
052f1dc7 | 10524 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
10525 | |
10526 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 10527 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 10528 | { |
2b01dfe3 PM |
10529 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
10530 | struct task_group, css); | |
68318b8e SV |
10531 | } |
10532 | ||
10533 | static struct cgroup_subsys_state * | |
2b01dfe3 | 10534 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 10535 | { |
ec7dc8ac | 10536 | struct task_group *tg, *parent; |
68318b8e | 10537 | |
2b01dfe3 | 10538 | if (!cgrp->parent) { |
68318b8e | 10539 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
10540 | return &init_task_group.css; |
10541 | } | |
10542 | ||
ec7dc8ac DG |
10543 | parent = cgroup_tg(cgrp->parent); |
10544 | tg = sched_create_group(parent); | |
68318b8e SV |
10545 | if (IS_ERR(tg)) |
10546 | return ERR_PTR(-ENOMEM); | |
10547 | ||
68318b8e SV |
10548 | return &tg->css; |
10549 | } | |
10550 | ||
41a2d6cf IM |
10551 | static void |
10552 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 10553 | { |
2b01dfe3 | 10554 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10555 | |
10556 | sched_destroy_group(tg); | |
10557 | } | |
10558 | ||
41a2d6cf IM |
10559 | static int |
10560 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
10561 | struct task_struct *tsk) | |
68318b8e | 10562 | { |
b68aa230 | 10563 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 10564 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
10565 | return -EINVAL; |
10566 | #else | |
68318b8e SV |
10567 | /* We don't support RT-tasks being in separate groups */ |
10568 | if (tsk->sched_class != &fair_sched_class) | |
10569 | return -EINVAL; | |
b68aa230 | 10570 | #endif |
68318b8e SV |
10571 | |
10572 | return 0; | |
10573 | } | |
10574 | ||
10575 | static void | |
2b01dfe3 | 10576 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
10577 | struct cgroup *old_cont, struct task_struct *tsk) |
10578 | { | |
10579 | sched_move_task(tsk); | |
10580 | } | |
10581 | ||
052f1dc7 | 10582 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 10583 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 10584 | u64 shareval) |
68318b8e | 10585 | { |
2b01dfe3 | 10586 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
10587 | } |
10588 | ||
f4c753b7 | 10589 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 10590 | { |
2b01dfe3 | 10591 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
10592 | |
10593 | return (u64) tg->shares; | |
10594 | } | |
6d6bc0ad | 10595 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 10596 | |
052f1dc7 | 10597 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 10598 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 10599 | s64 val) |
6f505b16 | 10600 | { |
06ecb27c | 10601 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
10602 | } |
10603 | ||
06ecb27c | 10604 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 10605 | { |
06ecb27c | 10606 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 10607 | } |
d0b27fa7 PZ |
10608 | |
10609 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
10610 | u64 rt_period_us) | |
10611 | { | |
10612 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
10613 | } | |
10614 | ||
10615 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
10616 | { | |
10617 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
10618 | } | |
6d6bc0ad | 10619 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 10620 | |
fe5c7cc2 | 10621 | static struct cftype cpu_files[] = { |
052f1dc7 | 10622 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
10623 | { |
10624 | .name = "shares", | |
f4c753b7 PM |
10625 | .read_u64 = cpu_shares_read_u64, |
10626 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 10627 | }, |
052f1dc7 PZ |
10628 | #endif |
10629 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 10630 | { |
9f0c1e56 | 10631 | .name = "rt_runtime_us", |
06ecb27c PM |
10632 | .read_s64 = cpu_rt_runtime_read, |
10633 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 10634 | }, |
d0b27fa7 PZ |
10635 | { |
10636 | .name = "rt_period_us", | |
f4c753b7 PM |
10637 | .read_u64 = cpu_rt_period_read_uint, |
10638 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 10639 | }, |
052f1dc7 | 10640 | #endif |
68318b8e SV |
10641 | }; |
10642 | ||
10643 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
10644 | { | |
fe5c7cc2 | 10645 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
10646 | } |
10647 | ||
10648 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
10649 | .name = "cpu", |
10650 | .create = cpu_cgroup_create, | |
10651 | .destroy = cpu_cgroup_destroy, | |
10652 | .can_attach = cpu_cgroup_can_attach, | |
10653 | .attach = cpu_cgroup_attach, | |
10654 | .populate = cpu_cgroup_populate, | |
10655 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
10656 | .early_init = 1, |
10657 | }; | |
10658 | ||
052f1dc7 | 10659 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
10660 | |
10661 | #ifdef CONFIG_CGROUP_CPUACCT | |
10662 | ||
10663 | /* | |
10664 | * CPU accounting code for task groups. | |
10665 | * | |
10666 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
10667 | * (balbir@in.ibm.com). | |
10668 | */ | |
10669 | ||
934352f2 | 10670 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
10671 | struct cpuacct { |
10672 | struct cgroup_subsys_state css; | |
10673 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
10674 | u64 *cpuusage; | |
ef12fefa | 10675 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 10676 | struct cpuacct *parent; |
d842de87 SV |
10677 | }; |
10678 | ||
10679 | struct cgroup_subsys cpuacct_subsys; | |
10680 | ||
10681 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 10682 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 10683 | { |
32cd756a | 10684 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
10685 | struct cpuacct, css); |
10686 | } | |
10687 | ||
10688 | /* return cpu accounting group to which this task belongs */ | |
10689 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
10690 | { | |
10691 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
10692 | struct cpuacct, css); | |
10693 | } | |
10694 | ||
10695 | /* create a new cpu accounting group */ | |
10696 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 10697 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
10698 | { |
10699 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 10700 | int i; |
d842de87 SV |
10701 | |
10702 | if (!ca) | |
ef12fefa | 10703 | goto out; |
d842de87 SV |
10704 | |
10705 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
10706 | if (!ca->cpuusage) |
10707 | goto out_free_ca; | |
10708 | ||
10709 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
10710 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
10711 | goto out_free_counters; | |
d842de87 | 10712 | |
934352f2 BR |
10713 | if (cgrp->parent) |
10714 | ca->parent = cgroup_ca(cgrp->parent); | |
10715 | ||
d842de87 | 10716 | return &ca->css; |
ef12fefa BR |
10717 | |
10718 | out_free_counters: | |
10719 | while (--i >= 0) | |
10720 | percpu_counter_destroy(&ca->cpustat[i]); | |
10721 | free_percpu(ca->cpuusage); | |
10722 | out_free_ca: | |
10723 | kfree(ca); | |
10724 | out: | |
10725 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
10726 | } |
10727 | ||
10728 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 10729 | static void |
32cd756a | 10730 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10731 | { |
32cd756a | 10732 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 10733 | int i; |
d842de87 | 10734 | |
ef12fefa BR |
10735 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
10736 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
10737 | free_percpu(ca->cpuusage); |
10738 | kfree(ca); | |
10739 | } | |
10740 | ||
720f5498 KC |
10741 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
10742 | { | |
b36128c8 | 10743 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10744 | u64 data; |
10745 | ||
10746 | #ifndef CONFIG_64BIT | |
10747 | /* | |
10748 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
10749 | */ | |
10750 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10751 | data = *cpuusage; | |
10752 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10753 | #else | |
10754 | data = *cpuusage; | |
10755 | #endif | |
10756 | ||
10757 | return data; | |
10758 | } | |
10759 | ||
10760 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
10761 | { | |
b36128c8 | 10762 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
10763 | |
10764 | #ifndef CONFIG_64BIT | |
10765 | /* | |
10766 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
10767 | */ | |
10768 | spin_lock_irq(&cpu_rq(cpu)->lock); | |
10769 | *cpuusage = val; | |
10770 | spin_unlock_irq(&cpu_rq(cpu)->lock); | |
10771 | #else | |
10772 | *cpuusage = val; | |
10773 | #endif | |
10774 | } | |
10775 | ||
d842de87 | 10776 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 10777 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 10778 | { |
32cd756a | 10779 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
10780 | u64 totalcpuusage = 0; |
10781 | int i; | |
10782 | ||
720f5498 KC |
10783 | for_each_present_cpu(i) |
10784 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
10785 | |
10786 | return totalcpuusage; | |
10787 | } | |
10788 | ||
0297b803 DG |
10789 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
10790 | u64 reset) | |
10791 | { | |
10792 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10793 | int err = 0; | |
10794 | int i; | |
10795 | ||
10796 | if (reset) { | |
10797 | err = -EINVAL; | |
10798 | goto out; | |
10799 | } | |
10800 | ||
720f5498 KC |
10801 | for_each_present_cpu(i) |
10802 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 10803 | |
0297b803 DG |
10804 | out: |
10805 | return err; | |
10806 | } | |
10807 | ||
e9515c3c KC |
10808 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
10809 | struct seq_file *m) | |
10810 | { | |
10811 | struct cpuacct *ca = cgroup_ca(cgroup); | |
10812 | u64 percpu; | |
10813 | int i; | |
10814 | ||
10815 | for_each_present_cpu(i) { | |
10816 | percpu = cpuacct_cpuusage_read(ca, i); | |
10817 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
10818 | } | |
10819 | seq_printf(m, "\n"); | |
10820 | return 0; | |
10821 | } | |
10822 | ||
ef12fefa BR |
10823 | static const char *cpuacct_stat_desc[] = { |
10824 | [CPUACCT_STAT_USER] = "user", | |
10825 | [CPUACCT_STAT_SYSTEM] = "system", | |
10826 | }; | |
10827 | ||
10828 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
10829 | struct cgroup_map_cb *cb) | |
10830 | { | |
10831 | struct cpuacct *ca = cgroup_ca(cgrp); | |
10832 | int i; | |
10833 | ||
10834 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
10835 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
10836 | val = cputime64_to_clock_t(val); | |
10837 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
10838 | } | |
10839 | return 0; | |
10840 | } | |
10841 | ||
d842de87 SV |
10842 | static struct cftype files[] = { |
10843 | { | |
10844 | .name = "usage", | |
f4c753b7 PM |
10845 | .read_u64 = cpuusage_read, |
10846 | .write_u64 = cpuusage_write, | |
d842de87 | 10847 | }, |
e9515c3c KC |
10848 | { |
10849 | .name = "usage_percpu", | |
10850 | .read_seq_string = cpuacct_percpu_seq_read, | |
10851 | }, | |
ef12fefa BR |
10852 | { |
10853 | .name = "stat", | |
10854 | .read_map = cpuacct_stats_show, | |
10855 | }, | |
d842de87 SV |
10856 | }; |
10857 | ||
32cd756a | 10858 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 10859 | { |
32cd756a | 10860 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
10861 | } |
10862 | ||
10863 | /* | |
10864 | * charge this task's execution time to its accounting group. | |
10865 | * | |
10866 | * called with rq->lock held. | |
10867 | */ | |
10868 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
10869 | { | |
10870 | struct cpuacct *ca; | |
934352f2 | 10871 | int cpu; |
d842de87 | 10872 | |
c40c6f85 | 10873 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
10874 | return; |
10875 | ||
934352f2 | 10876 | cpu = task_cpu(tsk); |
a18b83b7 BR |
10877 | |
10878 | rcu_read_lock(); | |
10879 | ||
d842de87 | 10880 | ca = task_ca(tsk); |
d842de87 | 10881 | |
934352f2 | 10882 | for (; ca; ca = ca->parent) { |
b36128c8 | 10883 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
10884 | *cpuusage += cputime; |
10885 | } | |
a18b83b7 BR |
10886 | |
10887 | rcu_read_unlock(); | |
d842de87 SV |
10888 | } |
10889 | ||
ef12fefa BR |
10890 | /* |
10891 | * Charge the system/user time to the task's accounting group. | |
10892 | */ | |
10893 | static void cpuacct_update_stats(struct task_struct *tsk, | |
10894 | enum cpuacct_stat_index idx, cputime_t val) | |
10895 | { | |
10896 | struct cpuacct *ca; | |
10897 | ||
10898 | if (unlikely(!cpuacct_subsys.active)) | |
10899 | return; | |
10900 | ||
10901 | rcu_read_lock(); | |
10902 | ca = task_ca(tsk); | |
10903 | ||
10904 | do { | |
10905 | percpu_counter_add(&ca->cpustat[idx], val); | |
10906 | ca = ca->parent; | |
10907 | } while (ca); | |
10908 | rcu_read_unlock(); | |
10909 | } | |
10910 | ||
d842de87 SV |
10911 | struct cgroup_subsys cpuacct_subsys = { |
10912 | .name = "cpuacct", | |
10913 | .create = cpuacct_create, | |
10914 | .destroy = cpuacct_destroy, | |
10915 | .populate = cpuacct_populate, | |
10916 | .subsys_id = cpuacct_subsys_id, | |
10917 | }; | |
10918 | #endif /* CONFIG_CGROUP_CPUACCT */ |