Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
57 | #include <linux/kthread.h> | |
58 | #include <linux/seq_file.h> | |
e692ab53 | 59 | #include <linux/sysctl.h> |
1da177e4 LT |
60 | #include <linux/syscalls.h> |
61 | #include <linux/times.h> | |
8f0ab514 | 62 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 63 | #include <linux/kprobes.h> |
0ff92245 | 64 | #include <linux/delayacct.h> |
5517d86b | 65 | #include <linux/reciprocal_div.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
434d53b0 | 70 | #include <linux/bootmem.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
1da177e4 LT |
77 | |
78 | /* | |
79 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
80 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
81 | * and back. | |
82 | */ | |
83 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
84 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
85 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
86 | ||
87 | /* | |
88 | * 'User priority' is the nice value converted to something we | |
89 | * can work with better when scaling various scheduler parameters, | |
90 | * it's a [ 0 ... 39 ] range. | |
91 | */ | |
92 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
93 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
94 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
95 | ||
96 | /* | |
d7876a08 | 97 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 98 | */ |
d6322faf | 99 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 100 | |
6aa645ea IM |
101 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
102 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
103 | ||
1da177e4 LT |
104 | /* |
105 | * These are the 'tuning knobs' of the scheduler: | |
106 | * | |
a4ec24b4 | 107 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
108 | * Timeslices get refilled after they expire. |
109 | */ | |
1da177e4 | 110 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 111 | |
d0b27fa7 PZ |
112 | /* |
113 | * single value that denotes runtime == period, ie unlimited time. | |
114 | */ | |
115 | #define RUNTIME_INF ((u64)~0ULL) | |
116 | ||
5517d86b ED |
117 | #ifdef CONFIG_SMP |
118 | /* | |
119 | * Divide a load by a sched group cpu_power : (load / sg->__cpu_power) | |
120 | * Since cpu_power is a 'constant', we can use a reciprocal divide. | |
121 | */ | |
122 | static inline u32 sg_div_cpu_power(const struct sched_group *sg, u32 load) | |
123 | { | |
124 | return reciprocal_divide(load, sg->reciprocal_cpu_power); | |
125 | } | |
126 | ||
127 | /* | |
128 | * Each time a sched group cpu_power is changed, | |
129 | * we must compute its reciprocal value | |
130 | */ | |
131 | static inline void sg_inc_cpu_power(struct sched_group *sg, u32 val) | |
132 | { | |
133 | sg->__cpu_power += val; | |
134 | sg->reciprocal_cpu_power = reciprocal_value(sg->__cpu_power); | |
135 | } | |
136 | #endif | |
137 | ||
e05606d3 IM |
138 | static inline int rt_policy(int policy) |
139 | { | |
3f33a7ce | 140 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
141 | return 1; |
142 | return 0; | |
143 | } | |
144 | ||
145 | static inline int task_has_rt_policy(struct task_struct *p) | |
146 | { | |
147 | return rt_policy(p->policy); | |
148 | } | |
149 | ||
1da177e4 | 150 | /* |
6aa645ea | 151 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 152 | */ |
6aa645ea IM |
153 | struct rt_prio_array { |
154 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
155 | struct list_head queue[MAX_RT_PRIO]; | |
156 | }; | |
157 | ||
d0b27fa7 | 158 | struct rt_bandwidth { |
ea736ed5 IM |
159 | /* nests inside the rq lock: */ |
160 | spinlock_t rt_runtime_lock; | |
161 | ktime_t rt_period; | |
162 | u64 rt_runtime; | |
163 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
164 | }; |
165 | ||
166 | static struct rt_bandwidth def_rt_bandwidth; | |
167 | ||
168 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
169 | ||
170 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
171 | { | |
172 | struct rt_bandwidth *rt_b = | |
173 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
174 | ktime_t now; | |
175 | int overrun; | |
176 | int idle = 0; | |
177 | ||
178 | for (;;) { | |
179 | now = hrtimer_cb_get_time(timer); | |
180 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
181 | ||
182 | if (!overrun) | |
183 | break; | |
184 | ||
185 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
186 | } | |
187 | ||
188 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
189 | } | |
190 | ||
191 | static | |
192 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
193 | { | |
194 | rt_b->rt_period = ns_to_ktime(period); | |
195 | rt_b->rt_runtime = runtime; | |
196 | ||
ac086bc2 PZ |
197 | spin_lock_init(&rt_b->rt_runtime_lock); |
198 | ||
d0b27fa7 PZ |
199 | hrtimer_init(&rt_b->rt_period_timer, |
200 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
201 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
202 | rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
203 | } | |
204 | ||
205 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
206 | { | |
207 | ktime_t now; | |
208 | ||
209 | if (rt_b->rt_runtime == RUNTIME_INF) | |
210 | return; | |
211 | ||
212 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
213 | return; | |
214 | ||
215 | spin_lock(&rt_b->rt_runtime_lock); | |
216 | for (;;) { | |
217 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
218 | break; | |
219 | ||
220 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
221 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
222 | hrtimer_start(&rt_b->rt_period_timer, | |
223 | rt_b->rt_period_timer.expires, | |
224 | HRTIMER_MODE_ABS); | |
225 | } | |
226 | spin_unlock(&rt_b->rt_runtime_lock); | |
227 | } | |
228 | ||
229 | #ifdef CONFIG_RT_GROUP_SCHED | |
230 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
231 | { | |
232 | hrtimer_cancel(&rt_b->rt_period_timer); | |
233 | } | |
234 | #endif | |
235 | ||
712555ee HC |
236 | /* |
237 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
238 | * detach_destroy_domains and partition_sched_domains. | |
239 | */ | |
240 | static DEFINE_MUTEX(sched_domains_mutex); | |
241 | ||
052f1dc7 | 242 | #ifdef CONFIG_GROUP_SCHED |
29f59db3 | 243 | |
68318b8e SV |
244 | #include <linux/cgroup.h> |
245 | ||
29f59db3 SV |
246 | struct cfs_rq; |
247 | ||
6f505b16 PZ |
248 | static LIST_HEAD(task_groups); |
249 | ||
29f59db3 | 250 | /* task group related information */ |
4cf86d77 | 251 | struct task_group { |
052f1dc7 | 252 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
253 | struct cgroup_subsys_state css; |
254 | #endif | |
052f1dc7 PZ |
255 | |
256 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
29f59db3 SV |
257 | /* schedulable entities of this group on each cpu */ |
258 | struct sched_entity **se; | |
259 | /* runqueue "owned" by this group on each cpu */ | |
260 | struct cfs_rq **cfs_rq; | |
261 | unsigned long shares; | |
052f1dc7 PZ |
262 | #endif |
263 | ||
264 | #ifdef CONFIG_RT_GROUP_SCHED | |
265 | struct sched_rt_entity **rt_se; | |
266 | struct rt_rq **rt_rq; | |
267 | ||
d0b27fa7 | 268 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 269 | #endif |
6b2d7700 | 270 | |
ae8393e5 | 271 | struct rcu_head rcu; |
6f505b16 | 272 | struct list_head list; |
f473aa5e PZ |
273 | |
274 | struct task_group *parent; | |
275 | struct list_head siblings; | |
276 | struct list_head children; | |
29f59db3 SV |
277 | }; |
278 | ||
354d60c2 | 279 | #ifdef CONFIG_USER_SCHED |
eff766a6 PZ |
280 | |
281 | /* | |
282 | * Root task group. | |
283 | * Every UID task group (including init_task_group aka UID-0) will | |
284 | * be a child to this group. | |
285 | */ | |
286 | struct task_group root_task_group; | |
287 | ||
052f1dc7 | 288 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
289 | /* Default task group's sched entity on each cpu */ |
290 | static DEFINE_PER_CPU(struct sched_entity, init_sched_entity); | |
291 | /* Default task group's cfs_rq on each cpu */ | |
292 | static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp; | |
052f1dc7 PZ |
293 | #endif |
294 | ||
295 | #ifdef CONFIG_RT_GROUP_SCHED | |
296 | static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity); | |
297 | static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp; | |
052f1dc7 | 298 | #endif |
eff766a6 PZ |
299 | #else |
300 | #define root_task_group init_task_group | |
354d60c2 | 301 | #endif |
6f505b16 | 302 | |
8ed36996 | 303 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
304 | * a task group's cpu shares. |
305 | */ | |
8ed36996 | 306 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 307 | |
052f1dc7 | 308 | #ifdef CONFIG_FAIR_GROUP_SCHED |
052f1dc7 PZ |
309 | #ifdef CONFIG_USER_SCHED |
310 | # define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) | |
311 | #else | |
312 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD | |
313 | #endif | |
314 | ||
cb4ad1ff | 315 | /* |
2e084786 LJ |
316 | * A weight of 0 or 1 can cause arithmetics problems. |
317 | * A weight of a cfs_rq is the sum of weights of which entities | |
318 | * are queued on this cfs_rq, so a weight of a entity should not be | |
319 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
320 | * (The default weight is 1024 - so there's no practical |
321 | * limitation from this.) | |
322 | */ | |
18d95a28 | 323 | #define MIN_SHARES 2 |
2e084786 | 324 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 325 | |
052f1dc7 PZ |
326 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
327 | #endif | |
328 | ||
29f59db3 | 329 | /* Default task group. |
3a252015 | 330 | * Every task in system belong to this group at bootup. |
29f59db3 | 331 | */ |
434d53b0 | 332 | struct task_group init_task_group; |
29f59db3 SV |
333 | |
334 | /* return group to which a task belongs */ | |
4cf86d77 | 335 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 336 | { |
4cf86d77 | 337 | struct task_group *tg; |
9b5b7751 | 338 | |
052f1dc7 | 339 | #ifdef CONFIG_USER_SCHED |
24e377a8 | 340 | tg = p->user->tg; |
052f1dc7 | 341 | #elif defined(CONFIG_CGROUP_SCHED) |
68318b8e SV |
342 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
343 | struct task_group, css); | |
24e377a8 | 344 | #else |
41a2d6cf | 345 | tg = &init_task_group; |
24e377a8 | 346 | #endif |
9b5b7751 | 347 | return tg; |
29f59db3 SV |
348 | } |
349 | ||
350 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 351 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 352 | { |
052f1dc7 | 353 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
354 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
355 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 356 | #endif |
6f505b16 | 357 | |
052f1dc7 | 358 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
359 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
360 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 361 | #endif |
29f59db3 SV |
362 | } |
363 | ||
364 | #else | |
365 | ||
6f505b16 | 366 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
29f59db3 | 367 | |
052f1dc7 | 368 | #endif /* CONFIG_GROUP_SCHED */ |
29f59db3 | 369 | |
6aa645ea IM |
370 | /* CFS-related fields in a runqueue */ |
371 | struct cfs_rq { | |
372 | struct load_weight load; | |
373 | unsigned long nr_running; | |
374 | ||
6aa645ea | 375 | u64 exec_clock; |
e9acbff6 | 376 | u64 min_vruntime; |
6aa645ea IM |
377 | |
378 | struct rb_root tasks_timeline; | |
379 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
380 | |
381 | struct list_head tasks; | |
382 | struct list_head *balance_iterator; | |
383 | ||
384 | /* | |
385 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
386 | * It is set to NULL otherwise (i.e when none are currently running). |
387 | */ | |
aa2ac252 | 388 | struct sched_entity *curr, *next; |
ddc97297 PZ |
389 | |
390 | unsigned long nr_spread_over; | |
391 | ||
62160e3f | 392 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
393 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
394 | ||
41a2d6cf IM |
395 | /* |
396 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
397 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
398 | * (like users, containers etc.) | |
399 | * | |
400 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
401 | * list is used during load balance. | |
402 | */ | |
41a2d6cf IM |
403 | struct list_head leaf_cfs_rq_list; |
404 | struct task_group *tg; /* group that "owns" this runqueue */ | |
6aa645ea IM |
405 | #endif |
406 | }; | |
1da177e4 | 407 | |
6aa645ea IM |
408 | /* Real-Time classes' related field in a runqueue: */ |
409 | struct rt_rq { | |
410 | struct rt_prio_array active; | |
63489e45 | 411 | unsigned long rt_nr_running; |
052f1dc7 | 412 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
413 | int highest_prio; /* highest queued rt task prio */ |
414 | #endif | |
fa85ae24 | 415 | #ifdef CONFIG_SMP |
73fe6aae | 416 | unsigned long rt_nr_migratory; |
a22d7fc1 | 417 | int overloaded; |
fa85ae24 | 418 | #endif |
6f505b16 | 419 | int rt_throttled; |
fa85ae24 | 420 | u64 rt_time; |
ac086bc2 | 421 | u64 rt_runtime; |
ea736ed5 | 422 | /* Nests inside the rq lock: */ |
ac086bc2 | 423 | spinlock_t rt_runtime_lock; |
6f505b16 | 424 | |
052f1dc7 | 425 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
426 | unsigned long rt_nr_boosted; |
427 | ||
6f505b16 PZ |
428 | struct rq *rq; |
429 | struct list_head leaf_rt_rq_list; | |
430 | struct task_group *tg; | |
431 | struct sched_rt_entity *rt_se; | |
432 | #endif | |
6aa645ea IM |
433 | }; |
434 | ||
57d885fe GH |
435 | #ifdef CONFIG_SMP |
436 | ||
437 | /* | |
438 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
439 | * variables. Each exclusive cpuset essentially defines an island domain by |
440 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
441 | * exclusive cpuset is created, we also create and attach a new root-domain |
442 | * object. | |
443 | * | |
57d885fe GH |
444 | */ |
445 | struct root_domain { | |
446 | atomic_t refcount; | |
447 | cpumask_t span; | |
448 | cpumask_t online; | |
637f5085 | 449 | |
0eab9146 | 450 | /* |
637f5085 GH |
451 | * The "RT overload" flag: it gets set if a CPU has more than |
452 | * one runnable RT task. | |
453 | */ | |
454 | cpumask_t rto_mask; | |
0eab9146 | 455 | atomic_t rto_count; |
57d885fe GH |
456 | }; |
457 | ||
dc938520 GH |
458 | /* |
459 | * By default the system creates a single root-domain with all cpus as | |
460 | * members (mimicking the global state we have today). | |
461 | */ | |
57d885fe GH |
462 | static struct root_domain def_root_domain; |
463 | ||
464 | #endif | |
465 | ||
1da177e4 LT |
466 | /* |
467 | * This is the main, per-CPU runqueue data structure. | |
468 | * | |
469 | * Locking rule: those places that want to lock multiple runqueues | |
470 | * (such as the load balancing or the thread migration code), lock | |
471 | * acquire operations must be ordered by ascending &runqueue. | |
472 | */ | |
70b97a7f | 473 | struct rq { |
d8016491 IM |
474 | /* runqueue lock: */ |
475 | spinlock_t lock; | |
1da177e4 LT |
476 | |
477 | /* | |
478 | * nr_running and cpu_load should be in the same cacheline because | |
479 | * remote CPUs use both these fields when doing load calculation. | |
480 | */ | |
481 | unsigned long nr_running; | |
6aa645ea IM |
482 | #define CPU_LOAD_IDX_MAX 5 |
483 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
bdecea3a | 484 | unsigned char idle_at_tick; |
46cb4b7c | 485 | #ifdef CONFIG_NO_HZ |
15934a37 | 486 | unsigned long last_tick_seen; |
46cb4b7c SS |
487 | unsigned char in_nohz_recently; |
488 | #endif | |
d8016491 IM |
489 | /* capture load from *all* tasks on this cpu: */ |
490 | struct load_weight load; | |
6aa645ea IM |
491 | unsigned long nr_load_updates; |
492 | u64 nr_switches; | |
493 | ||
494 | struct cfs_rq cfs; | |
6f505b16 | 495 | struct rt_rq rt; |
6f505b16 | 496 | |
6aa645ea | 497 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
498 | /* list of leaf cfs_rq on this cpu: */ |
499 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
500 | #endif |
501 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 502 | struct list_head leaf_rt_rq_list; |
1da177e4 | 503 | #endif |
1da177e4 LT |
504 | |
505 | /* | |
506 | * This is part of a global counter where only the total sum | |
507 | * over all CPUs matters. A task can increase this counter on | |
508 | * one CPU and if it got migrated afterwards it may decrease | |
509 | * it on another CPU. Always updated under the runqueue lock: | |
510 | */ | |
511 | unsigned long nr_uninterruptible; | |
512 | ||
36c8b586 | 513 | struct task_struct *curr, *idle; |
c9819f45 | 514 | unsigned long next_balance; |
1da177e4 | 515 | struct mm_struct *prev_mm; |
6aa645ea | 516 | |
3e51f33f | 517 | u64 clock; |
6aa645ea | 518 | |
1da177e4 LT |
519 | atomic_t nr_iowait; |
520 | ||
521 | #ifdef CONFIG_SMP | |
0eab9146 | 522 | struct root_domain *rd; |
1da177e4 LT |
523 | struct sched_domain *sd; |
524 | ||
525 | /* For active balancing */ | |
526 | int active_balance; | |
527 | int push_cpu; | |
d8016491 IM |
528 | /* cpu of this runqueue: */ |
529 | int cpu; | |
1da177e4 | 530 | |
36c8b586 | 531 | struct task_struct *migration_thread; |
1da177e4 LT |
532 | struct list_head migration_queue; |
533 | #endif | |
534 | ||
8f4d37ec PZ |
535 | #ifdef CONFIG_SCHED_HRTICK |
536 | unsigned long hrtick_flags; | |
537 | ktime_t hrtick_expire; | |
538 | struct hrtimer hrtick_timer; | |
539 | #endif | |
540 | ||
1da177e4 LT |
541 | #ifdef CONFIG_SCHEDSTATS |
542 | /* latency stats */ | |
543 | struct sched_info rq_sched_info; | |
544 | ||
545 | /* sys_sched_yield() stats */ | |
480b9434 KC |
546 | unsigned int yld_exp_empty; |
547 | unsigned int yld_act_empty; | |
548 | unsigned int yld_both_empty; | |
549 | unsigned int yld_count; | |
1da177e4 LT |
550 | |
551 | /* schedule() stats */ | |
480b9434 KC |
552 | unsigned int sched_switch; |
553 | unsigned int sched_count; | |
554 | unsigned int sched_goidle; | |
1da177e4 LT |
555 | |
556 | /* try_to_wake_up() stats */ | |
480b9434 KC |
557 | unsigned int ttwu_count; |
558 | unsigned int ttwu_local; | |
b8efb561 IM |
559 | |
560 | /* BKL stats */ | |
480b9434 | 561 | unsigned int bkl_count; |
1da177e4 | 562 | #endif |
fcb99371 | 563 | struct lock_class_key rq_lock_key; |
1da177e4 LT |
564 | }; |
565 | ||
f34e3b61 | 566 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 567 | |
dd41f596 IM |
568 | static inline void check_preempt_curr(struct rq *rq, struct task_struct *p) |
569 | { | |
570 | rq->curr->sched_class->check_preempt_curr(rq, p); | |
571 | } | |
572 | ||
0a2966b4 CL |
573 | static inline int cpu_of(struct rq *rq) |
574 | { | |
575 | #ifdef CONFIG_SMP | |
576 | return rq->cpu; | |
577 | #else | |
578 | return 0; | |
579 | #endif | |
580 | } | |
581 | ||
674311d5 NP |
582 | /* |
583 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 584 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
585 | * |
586 | * The domain tree of any CPU may only be accessed from within | |
587 | * preempt-disabled sections. | |
588 | */ | |
48f24c4d IM |
589 | #define for_each_domain(cpu, __sd) \ |
590 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
591 | |
592 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
593 | #define this_rq() (&__get_cpu_var(runqueues)) | |
594 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
595 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
596 | ||
3e51f33f PZ |
597 | static inline void update_rq_clock(struct rq *rq) |
598 | { | |
599 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
600 | } | |
601 | ||
bf5c91ba IM |
602 | /* |
603 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
604 | */ | |
605 | #ifdef CONFIG_SCHED_DEBUG | |
606 | # define const_debug __read_mostly | |
607 | #else | |
608 | # define const_debug static const | |
609 | #endif | |
610 | ||
017730c1 IM |
611 | /** |
612 | * runqueue_is_locked | |
613 | * | |
614 | * Returns true if the current cpu runqueue is locked. | |
615 | * This interface allows printk to be called with the runqueue lock | |
616 | * held and know whether or not it is OK to wake up the klogd. | |
617 | */ | |
618 | int runqueue_is_locked(void) | |
619 | { | |
620 | int cpu = get_cpu(); | |
621 | struct rq *rq = cpu_rq(cpu); | |
622 | int ret; | |
623 | ||
624 | ret = spin_is_locked(&rq->lock); | |
625 | put_cpu(); | |
626 | return ret; | |
627 | } | |
628 | ||
bf5c91ba IM |
629 | /* |
630 | * Debugging: various feature bits | |
631 | */ | |
f00b45c1 PZ |
632 | |
633 | #define SCHED_FEAT(name, enabled) \ | |
634 | __SCHED_FEAT_##name , | |
635 | ||
bf5c91ba | 636 | enum { |
f00b45c1 | 637 | #include "sched_features.h" |
bf5c91ba IM |
638 | }; |
639 | ||
f00b45c1 PZ |
640 | #undef SCHED_FEAT |
641 | ||
642 | #define SCHED_FEAT(name, enabled) \ | |
643 | (1UL << __SCHED_FEAT_##name) * enabled | | |
644 | ||
bf5c91ba | 645 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
646 | #include "sched_features.h" |
647 | 0; | |
648 | ||
649 | #undef SCHED_FEAT | |
650 | ||
651 | #ifdef CONFIG_SCHED_DEBUG | |
652 | #define SCHED_FEAT(name, enabled) \ | |
653 | #name , | |
654 | ||
983ed7a6 | 655 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
656 | #include "sched_features.h" |
657 | NULL | |
658 | }; | |
659 | ||
660 | #undef SCHED_FEAT | |
661 | ||
983ed7a6 | 662 | static int sched_feat_open(struct inode *inode, struct file *filp) |
f00b45c1 PZ |
663 | { |
664 | filp->private_data = inode->i_private; | |
665 | return 0; | |
666 | } | |
667 | ||
668 | static ssize_t | |
669 | sched_feat_read(struct file *filp, char __user *ubuf, | |
670 | size_t cnt, loff_t *ppos) | |
671 | { | |
672 | char *buf; | |
673 | int r = 0; | |
674 | int len = 0; | |
675 | int i; | |
676 | ||
677 | for (i = 0; sched_feat_names[i]; i++) { | |
678 | len += strlen(sched_feat_names[i]); | |
679 | len += 4; | |
680 | } | |
681 | ||
682 | buf = kmalloc(len + 2, GFP_KERNEL); | |
683 | if (!buf) | |
684 | return -ENOMEM; | |
685 | ||
686 | for (i = 0; sched_feat_names[i]; i++) { | |
687 | if (sysctl_sched_features & (1UL << i)) | |
688 | r += sprintf(buf + r, "%s ", sched_feat_names[i]); | |
689 | else | |
c24b7c52 | 690 | r += sprintf(buf + r, "NO_%s ", sched_feat_names[i]); |
f00b45c1 PZ |
691 | } |
692 | ||
693 | r += sprintf(buf + r, "\n"); | |
694 | WARN_ON(r >= len + 2); | |
695 | ||
696 | r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); | |
697 | ||
698 | kfree(buf); | |
699 | ||
700 | return r; | |
701 | } | |
702 | ||
703 | static ssize_t | |
704 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
705 | size_t cnt, loff_t *ppos) | |
706 | { | |
707 | char buf[64]; | |
708 | char *cmp = buf; | |
709 | int neg = 0; | |
710 | int i; | |
711 | ||
712 | if (cnt > 63) | |
713 | cnt = 63; | |
714 | ||
715 | if (copy_from_user(&buf, ubuf, cnt)) | |
716 | return -EFAULT; | |
717 | ||
718 | buf[cnt] = 0; | |
719 | ||
c24b7c52 | 720 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
721 | neg = 1; |
722 | cmp += 3; | |
723 | } | |
724 | ||
725 | for (i = 0; sched_feat_names[i]; i++) { | |
726 | int len = strlen(sched_feat_names[i]); | |
727 | ||
728 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
729 | if (neg) | |
730 | sysctl_sched_features &= ~(1UL << i); | |
731 | else | |
732 | sysctl_sched_features |= (1UL << i); | |
733 | break; | |
734 | } | |
735 | } | |
736 | ||
737 | if (!sched_feat_names[i]) | |
738 | return -EINVAL; | |
739 | ||
740 | filp->f_pos += cnt; | |
741 | ||
742 | return cnt; | |
743 | } | |
744 | ||
745 | static struct file_operations sched_feat_fops = { | |
746 | .open = sched_feat_open, | |
747 | .read = sched_feat_read, | |
748 | .write = sched_feat_write, | |
749 | }; | |
750 | ||
751 | static __init int sched_init_debug(void) | |
752 | { | |
f00b45c1 PZ |
753 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
754 | &sched_feat_fops); | |
755 | ||
756 | return 0; | |
757 | } | |
758 | late_initcall(sched_init_debug); | |
759 | ||
760 | #endif | |
761 | ||
762 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 763 | |
b82d9fdd PZ |
764 | /* |
765 | * Number of tasks to iterate in a single balance run. | |
766 | * Limited because this is done with IRQs disabled. | |
767 | */ | |
768 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
769 | ||
fa85ae24 | 770 | /* |
9f0c1e56 | 771 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
772 | * default: 1s |
773 | */ | |
9f0c1e56 | 774 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 775 | |
6892b75e IM |
776 | static __read_mostly int scheduler_running; |
777 | ||
9f0c1e56 PZ |
778 | /* |
779 | * part of the period that we allow rt tasks to run in us. | |
780 | * default: 0.95s | |
781 | */ | |
782 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 783 | |
d0b27fa7 PZ |
784 | static inline u64 global_rt_period(void) |
785 | { | |
786 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
787 | } | |
788 | ||
789 | static inline u64 global_rt_runtime(void) | |
790 | { | |
791 | if (sysctl_sched_rt_period < 0) | |
792 | return RUNTIME_INF; | |
793 | ||
794 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
795 | } | |
fa85ae24 | 796 | |
690229a0 | 797 | unsigned long long time_sync_thresh = 100000; |
27ec4407 IM |
798 | |
799 | static DEFINE_PER_CPU(unsigned long long, time_offset); | |
800 | static DEFINE_PER_CPU(unsigned long long, prev_cpu_time); | |
801 | ||
e436d800 | 802 | /* |
27ec4407 IM |
803 | * Global lock which we take every now and then to synchronize |
804 | * the CPUs time. This method is not warp-safe, but it's good | |
805 | * enough to synchronize slowly diverging time sources and thus | |
806 | * it's good enough for tracing: | |
e436d800 | 807 | */ |
27ec4407 IM |
808 | static DEFINE_SPINLOCK(time_sync_lock); |
809 | static unsigned long long prev_global_time; | |
810 | ||
dfbf4a1b | 811 | static unsigned long long __sync_cpu_clock(unsigned long long time, int cpu) |
27ec4407 | 812 | { |
dfbf4a1b IM |
813 | /* |
814 | * We want this inlined, to not get tracer function calls | |
815 | * in this critical section: | |
816 | */ | |
817 | spin_acquire(&time_sync_lock.dep_map, 0, 0, _THIS_IP_); | |
818 | __raw_spin_lock(&time_sync_lock.raw_lock); | |
27ec4407 IM |
819 | |
820 | if (time < prev_global_time) { | |
821 | per_cpu(time_offset, cpu) += prev_global_time - time; | |
822 | time = prev_global_time; | |
823 | } else { | |
824 | prev_global_time = time; | |
825 | } | |
826 | ||
dfbf4a1b IM |
827 | __raw_spin_unlock(&time_sync_lock.raw_lock); |
828 | spin_release(&time_sync_lock.dep_map, 1, _THIS_IP_); | |
27ec4407 IM |
829 | |
830 | return time; | |
831 | } | |
832 | ||
833 | static unsigned long long __cpu_clock(int cpu) | |
e436d800 | 834 | { |
e436d800 | 835 | unsigned long long now; |
e436d800 | 836 | |
8ced5f69 IM |
837 | /* |
838 | * Only call sched_clock() if the scheduler has already been | |
839 | * initialized (some code might call cpu_clock() very early): | |
840 | */ | |
6892b75e IM |
841 | if (unlikely(!scheduler_running)) |
842 | return 0; | |
843 | ||
3e51f33f | 844 | now = sched_clock_cpu(cpu); |
e436d800 IM |
845 | |
846 | return now; | |
847 | } | |
27ec4407 IM |
848 | |
849 | /* | |
850 | * For kernel-internal use: high-speed (but slightly incorrect) per-cpu | |
851 | * clock constructed from sched_clock(): | |
852 | */ | |
c6531cce | 853 | unsigned long long notrace cpu_clock(int cpu) |
27ec4407 IM |
854 | { |
855 | unsigned long long prev_cpu_time, time, delta_time; | |
dfbf4a1b | 856 | unsigned long flags; |
27ec4407 | 857 | |
040ec23d | 858 | local_irq_save(flags); |
27ec4407 IM |
859 | prev_cpu_time = per_cpu(prev_cpu_time, cpu); |
860 | time = __cpu_clock(cpu) + per_cpu(time_offset, cpu); | |
861 | delta_time = time-prev_cpu_time; | |
862 | ||
dfbf4a1b | 863 | if (unlikely(delta_time > time_sync_thresh)) { |
27ec4407 | 864 | time = __sync_cpu_clock(time, cpu); |
dfbf4a1b IM |
865 | per_cpu(prev_cpu_time, cpu) = time; |
866 | } | |
040ec23d | 867 | local_irq_restore(flags); |
27ec4407 IM |
868 | |
869 | return time; | |
870 | } | |
a58f6f25 | 871 | EXPORT_SYMBOL_GPL(cpu_clock); |
e436d800 | 872 | |
1da177e4 | 873 | #ifndef prepare_arch_switch |
4866cde0 NP |
874 | # define prepare_arch_switch(next) do { } while (0) |
875 | #endif | |
876 | #ifndef finish_arch_switch | |
877 | # define finish_arch_switch(prev) do { } while (0) | |
878 | #endif | |
879 | ||
051a1d1a DA |
880 | static inline int task_current(struct rq *rq, struct task_struct *p) |
881 | { | |
882 | return rq->curr == p; | |
883 | } | |
884 | ||
4866cde0 | 885 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 886 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 887 | { |
051a1d1a | 888 | return task_current(rq, p); |
4866cde0 NP |
889 | } |
890 | ||
70b97a7f | 891 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
892 | { |
893 | } | |
894 | ||
70b97a7f | 895 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 896 | { |
da04c035 IM |
897 | #ifdef CONFIG_DEBUG_SPINLOCK |
898 | /* this is a valid case when another task releases the spinlock */ | |
899 | rq->lock.owner = current; | |
900 | #endif | |
8a25d5de IM |
901 | /* |
902 | * If we are tracking spinlock dependencies then we have to | |
903 | * fix up the runqueue lock - which gets 'carried over' from | |
904 | * prev into current: | |
905 | */ | |
906 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
907 | ||
4866cde0 NP |
908 | spin_unlock_irq(&rq->lock); |
909 | } | |
910 | ||
911 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 912 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
913 | { |
914 | #ifdef CONFIG_SMP | |
915 | return p->oncpu; | |
916 | #else | |
051a1d1a | 917 | return task_current(rq, p); |
4866cde0 NP |
918 | #endif |
919 | } | |
920 | ||
70b97a7f | 921 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
922 | { |
923 | #ifdef CONFIG_SMP | |
924 | /* | |
925 | * We can optimise this out completely for !SMP, because the | |
926 | * SMP rebalancing from interrupt is the only thing that cares | |
927 | * here. | |
928 | */ | |
929 | next->oncpu = 1; | |
930 | #endif | |
931 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
932 | spin_unlock_irq(&rq->lock); | |
933 | #else | |
934 | spin_unlock(&rq->lock); | |
935 | #endif | |
936 | } | |
937 | ||
70b97a7f | 938 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
939 | { |
940 | #ifdef CONFIG_SMP | |
941 | /* | |
942 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
943 | * We must ensure this doesn't happen until the switch is completely | |
944 | * finished. | |
945 | */ | |
946 | smp_wmb(); | |
947 | prev->oncpu = 0; | |
948 | #endif | |
949 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
950 | local_irq_enable(); | |
1da177e4 | 951 | #endif |
4866cde0 NP |
952 | } |
953 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 954 | |
b29739f9 IM |
955 | /* |
956 | * __task_rq_lock - lock the runqueue a given task resides on. | |
957 | * Must be called interrupts disabled. | |
958 | */ | |
70b97a7f | 959 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
960 | __acquires(rq->lock) |
961 | { | |
3a5c359a AK |
962 | for (;;) { |
963 | struct rq *rq = task_rq(p); | |
964 | spin_lock(&rq->lock); | |
965 | if (likely(rq == task_rq(p))) | |
966 | return rq; | |
b29739f9 | 967 | spin_unlock(&rq->lock); |
b29739f9 | 968 | } |
b29739f9 IM |
969 | } |
970 | ||
1da177e4 LT |
971 | /* |
972 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 973 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
974 | * explicitly disabling preemption. |
975 | */ | |
70b97a7f | 976 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
977 | __acquires(rq->lock) |
978 | { | |
70b97a7f | 979 | struct rq *rq; |
1da177e4 | 980 | |
3a5c359a AK |
981 | for (;;) { |
982 | local_irq_save(*flags); | |
983 | rq = task_rq(p); | |
984 | spin_lock(&rq->lock); | |
985 | if (likely(rq == task_rq(p))) | |
986 | return rq; | |
1da177e4 | 987 | spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 988 | } |
1da177e4 LT |
989 | } |
990 | ||
a9957449 | 991 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
992 | __releases(rq->lock) |
993 | { | |
994 | spin_unlock(&rq->lock); | |
995 | } | |
996 | ||
70b97a7f | 997 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
998 | __releases(rq->lock) |
999 | { | |
1000 | spin_unlock_irqrestore(&rq->lock, *flags); | |
1001 | } | |
1002 | ||
1da177e4 | 1003 | /* |
cc2a73b5 | 1004 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 1005 | */ |
a9957449 | 1006 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
1007 | __acquires(rq->lock) |
1008 | { | |
70b97a7f | 1009 | struct rq *rq; |
1da177e4 LT |
1010 | |
1011 | local_irq_disable(); | |
1012 | rq = this_rq(); | |
1013 | spin_lock(&rq->lock); | |
1014 | ||
1015 | return rq; | |
1016 | } | |
1017 | ||
8f4d37ec PZ |
1018 | static void __resched_task(struct task_struct *p, int tif_bit); |
1019 | ||
1020 | static inline void resched_task(struct task_struct *p) | |
1021 | { | |
1022 | __resched_task(p, TIF_NEED_RESCHED); | |
1023 | } | |
1024 | ||
1025 | #ifdef CONFIG_SCHED_HRTICK | |
1026 | /* | |
1027 | * Use HR-timers to deliver accurate preemption points. | |
1028 | * | |
1029 | * Its all a bit involved since we cannot program an hrt while holding the | |
1030 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1031 | * reschedule event. | |
1032 | * | |
1033 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1034 | * rq->lock. | |
1035 | */ | |
1036 | static inline void resched_hrt(struct task_struct *p) | |
1037 | { | |
1038 | __resched_task(p, TIF_HRTICK_RESCHED); | |
1039 | } | |
1040 | ||
1041 | static inline void resched_rq(struct rq *rq) | |
1042 | { | |
1043 | unsigned long flags; | |
1044 | ||
1045 | spin_lock_irqsave(&rq->lock, flags); | |
1046 | resched_task(rq->curr); | |
1047 | spin_unlock_irqrestore(&rq->lock, flags); | |
1048 | } | |
1049 | ||
1050 | enum { | |
1051 | HRTICK_SET, /* re-programm hrtick_timer */ | |
1052 | HRTICK_RESET, /* not a new slice */ | |
b328ca18 | 1053 | HRTICK_BLOCK, /* stop hrtick operations */ |
8f4d37ec PZ |
1054 | }; |
1055 | ||
1056 | /* | |
1057 | * Use hrtick when: | |
1058 | * - enabled by features | |
1059 | * - hrtimer is actually high res | |
1060 | */ | |
1061 | static inline int hrtick_enabled(struct rq *rq) | |
1062 | { | |
1063 | if (!sched_feat(HRTICK)) | |
1064 | return 0; | |
b328ca18 PZ |
1065 | if (unlikely(test_bit(HRTICK_BLOCK, &rq->hrtick_flags))) |
1066 | return 0; | |
8f4d37ec PZ |
1067 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1068 | } | |
1069 | ||
1070 | /* | |
1071 | * Called to set the hrtick timer state. | |
1072 | * | |
1073 | * called with rq->lock held and irqs disabled | |
1074 | */ | |
1075 | static void hrtick_start(struct rq *rq, u64 delay, int reset) | |
1076 | { | |
1077 | assert_spin_locked(&rq->lock); | |
1078 | ||
1079 | /* | |
1080 | * preempt at: now + delay | |
1081 | */ | |
1082 | rq->hrtick_expire = | |
1083 | ktime_add_ns(rq->hrtick_timer.base->get_time(), delay); | |
1084 | /* | |
1085 | * indicate we need to program the timer | |
1086 | */ | |
1087 | __set_bit(HRTICK_SET, &rq->hrtick_flags); | |
1088 | if (reset) | |
1089 | __set_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1090 | ||
1091 | /* | |
1092 | * New slices are called from the schedule path and don't need a | |
1093 | * forced reschedule. | |
1094 | */ | |
1095 | if (reset) | |
1096 | resched_hrt(rq->curr); | |
1097 | } | |
1098 | ||
1099 | static void hrtick_clear(struct rq *rq) | |
1100 | { | |
1101 | if (hrtimer_active(&rq->hrtick_timer)) | |
1102 | hrtimer_cancel(&rq->hrtick_timer); | |
1103 | } | |
1104 | ||
1105 | /* | |
1106 | * Update the timer from the possible pending state. | |
1107 | */ | |
1108 | static void hrtick_set(struct rq *rq) | |
1109 | { | |
1110 | ktime_t time; | |
1111 | int set, reset; | |
1112 | unsigned long flags; | |
1113 | ||
1114 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1115 | ||
1116 | spin_lock_irqsave(&rq->lock, flags); | |
1117 | set = __test_and_clear_bit(HRTICK_SET, &rq->hrtick_flags); | |
1118 | reset = __test_and_clear_bit(HRTICK_RESET, &rq->hrtick_flags); | |
1119 | time = rq->hrtick_expire; | |
1120 | clear_thread_flag(TIF_HRTICK_RESCHED); | |
1121 | spin_unlock_irqrestore(&rq->lock, flags); | |
1122 | ||
1123 | if (set) { | |
1124 | hrtimer_start(&rq->hrtick_timer, time, HRTIMER_MODE_ABS); | |
1125 | if (reset && !hrtimer_active(&rq->hrtick_timer)) | |
1126 | resched_rq(rq); | |
1127 | } else | |
1128 | hrtick_clear(rq); | |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * High-resolution timer tick. | |
1133 | * Runs from hardirq context with interrupts disabled. | |
1134 | */ | |
1135 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1136 | { | |
1137 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1138 | ||
1139 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1140 | ||
1141 | spin_lock(&rq->lock); | |
3e51f33f | 1142 | update_rq_clock(rq); |
8f4d37ec PZ |
1143 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
1144 | spin_unlock(&rq->lock); | |
1145 | ||
1146 | return HRTIMER_NORESTART; | |
1147 | } | |
1148 | ||
95e904c7 | 1149 | #ifdef CONFIG_SMP |
b328ca18 PZ |
1150 | static void hotplug_hrtick_disable(int cpu) |
1151 | { | |
1152 | struct rq *rq = cpu_rq(cpu); | |
1153 | unsigned long flags; | |
1154 | ||
1155 | spin_lock_irqsave(&rq->lock, flags); | |
1156 | rq->hrtick_flags = 0; | |
1157 | __set_bit(HRTICK_BLOCK, &rq->hrtick_flags); | |
1158 | spin_unlock_irqrestore(&rq->lock, flags); | |
1159 | ||
1160 | hrtick_clear(rq); | |
1161 | } | |
1162 | ||
1163 | static void hotplug_hrtick_enable(int cpu) | |
1164 | { | |
1165 | struct rq *rq = cpu_rq(cpu); | |
1166 | unsigned long flags; | |
1167 | ||
1168 | spin_lock_irqsave(&rq->lock, flags); | |
1169 | __clear_bit(HRTICK_BLOCK, &rq->hrtick_flags); | |
1170 | spin_unlock_irqrestore(&rq->lock, flags); | |
1171 | } | |
1172 | ||
1173 | static int | |
1174 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1175 | { | |
1176 | int cpu = (int)(long)hcpu; | |
1177 | ||
1178 | switch (action) { | |
1179 | case CPU_UP_CANCELED: | |
1180 | case CPU_UP_CANCELED_FROZEN: | |
1181 | case CPU_DOWN_PREPARE: | |
1182 | case CPU_DOWN_PREPARE_FROZEN: | |
1183 | case CPU_DEAD: | |
1184 | case CPU_DEAD_FROZEN: | |
1185 | hotplug_hrtick_disable(cpu); | |
1186 | return NOTIFY_OK; | |
1187 | ||
1188 | case CPU_UP_PREPARE: | |
1189 | case CPU_UP_PREPARE_FROZEN: | |
1190 | case CPU_DOWN_FAILED: | |
1191 | case CPU_DOWN_FAILED_FROZEN: | |
1192 | case CPU_ONLINE: | |
1193 | case CPU_ONLINE_FROZEN: | |
1194 | hotplug_hrtick_enable(cpu); | |
1195 | return NOTIFY_OK; | |
1196 | } | |
1197 | ||
1198 | return NOTIFY_DONE; | |
1199 | } | |
1200 | ||
1201 | static void init_hrtick(void) | |
1202 | { | |
1203 | hotcpu_notifier(hotplug_hrtick, 0); | |
1204 | } | |
95e904c7 | 1205 | #endif /* CONFIG_SMP */ |
b328ca18 PZ |
1206 | |
1207 | static void init_rq_hrtick(struct rq *rq) | |
8f4d37ec PZ |
1208 | { |
1209 | rq->hrtick_flags = 0; | |
1210 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
1211 | rq->hrtick_timer.function = hrtick; | |
1212 | rq->hrtick_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ; | |
1213 | } | |
1214 | ||
1215 | void hrtick_resched(void) | |
1216 | { | |
1217 | struct rq *rq; | |
1218 | unsigned long flags; | |
1219 | ||
1220 | if (!test_thread_flag(TIF_HRTICK_RESCHED)) | |
1221 | return; | |
1222 | ||
1223 | local_irq_save(flags); | |
1224 | rq = cpu_rq(smp_processor_id()); | |
1225 | hrtick_set(rq); | |
1226 | local_irq_restore(flags); | |
1227 | } | |
1228 | #else | |
1229 | static inline void hrtick_clear(struct rq *rq) | |
1230 | { | |
1231 | } | |
1232 | ||
1233 | static inline void hrtick_set(struct rq *rq) | |
1234 | { | |
1235 | } | |
1236 | ||
1237 | static inline void init_rq_hrtick(struct rq *rq) | |
1238 | { | |
1239 | } | |
1240 | ||
1241 | void hrtick_resched(void) | |
1242 | { | |
1243 | } | |
b328ca18 PZ |
1244 | |
1245 | static inline void init_hrtick(void) | |
1246 | { | |
1247 | } | |
8f4d37ec PZ |
1248 | #endif |
1249 | ||
c24d20db IM |
1250 | /* |
1251 | * resched_task - mark a task 'to be rescheduled now'. | |
1252 | * | |
1253 | * On UP this means the setting of the need_resched flag, on SMP it | |
1254 | * might also involve a cross-CPU call to trigger the scheduler on | |
1255 | * the target CPU. | |
1256 | */ | |
1257 | #ifdef CONFIG_SMP | |
1258 | ||
1259 | #ifndef tsk_is_polling | |
1260 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1261 | #endif | |
1262 | ||
8f4d37ec | 1263 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1264 | { |
1265 | int cpu; | |
1266 | ||
1267 | assert_spin_locked(&task_rq(p)->lock); | |
1268 | ||
8f4d37ec | 1269 | if (unlikely(test_tsk_thread_flag(p, tif_bit))) |
c24d20db IM |
1270 | return; |
1271 | ||
8f4d37ec | 1272 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1273 | |
1274 | cpu = task_cpu(p); | |
1275 | if (cpu == smp_processor_id()) | |
1276 | return; | |
1277 | ||
1278 | /* NEED_RESCHED must be visible before we test polling */ | |
1279 | smp_mb(); | |
1280 | if (!tsk_is_polling(p)) | |
1281 | smp_send_reschedule(cpu); | |
1282 | } | |
1283 | ||
1284 | static void resched_cpu(int cpu) | |
1285 | { | |
1286 | struct rq *rq = cpu_rq(cpu); | |
1287 | unsigned long flags; | |
1288 | ||
1289 | if (!spin_trylock_irqsave(&rq->lock, flags)) | |
1290 | return; | |
1291 | resched_task(cpu_curr(cpu)); | |
1292 | spin_unlock_irqrestore(&rq->lock, flags); | |
1293 | } | |
06d8308c TG |
1294 | |
1295 | #ifdef CONFIG_NO_HZ | |
1296 | /* | |
1297 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1298 | * idle CPU then this timer might expire before the next timer event | |
1299 | * which is scheduled to wake up that CPU. In case of a completely | |
1300 | * idle system the next event might even be infinite time into the | |
1301 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1302 | * leaves the inner idle loop so the newly added timer is taken into | |
1303 | * account when the CPU goes back to idle and evaluates the timer | |
1304 | * wheel for the next timer event. | |
1305 | */ | |
1306 | void wake_up_idle_cpu(int cpu) | |
1307 | { | |
1308 | struct rq *rq = cpu_rq(cpu); | |
1309 | ||
1310 | if (cpu == smp_processor_id()) | |
1311 | return; | |
1312 | ||
1313 | /* | |
1314 | * This is safe, as this function is called with the timer | |
1315 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1316 | * to idle and has not yet set rq->curr to idle then it will | |
1317 | * be serialized on the timer wheel base lock and take the new | |
1318 | * timer into account automatically. | |
1319 | */ | |
1320 | if (rq->curr != rq->idle) | |
1321 | return; | |
1322 | ||
1323 | /* | |
1324 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1325 | * lockless. The worst case is that the other CPU runs the | |
1326 | * idle task through an additional NOOP schedule() | |
1327 | */ | |
1328 | set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); | |
1329 | ||
1330 | /* NEED_RESCHED must be visible before we test polling */ | |
1331 | smp_mb(); | |
1332 | if (!tsk_is_polling(rq->idle)) | |
1333 | smp_send_reschedule(cpu); | |
1334 | } | |
1335 | #endif | |
1336 | ||
c24d20db | 1337 | #else |
8f4d37ec | 1338 | static void __resched_task(struct task_struct *p, int tif_bit) |
c24d20db IM |
1339 | { |
1340 | assert_spin_locked(&task_rq(p)->lock); | |
8f4d37ec | 1341 | set_tsk_thread_flag(p, tif_bit); |
c24d20db IM |
1342 | } |
1343 | #endif | |
1344 | ||
45bf76df IM |
1345 | #if BITS_PER_LONG == 32 |
1346 | # define WMULT_CONST (~0UL) | |
1347 | #else | |
1348 | # define WMULT_CONST (1UL << 32) | |
1349 | #endif | |
1350 | ||
1351 | #define WMULT_SHIFT 32 | |
1352 | ||
194081eb IM |
1353 | /* |
1354 | * Shift right and round: | |
1355 | */ | |
cf2ab469 | 1356 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1357 | |
cb1c4fc9 | 1358 | static unsigned long |
45bf76df IM |
1359 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1360 | struct load_weight *lw) | |
1361 | { | |
1362 | u64 tmp; | |
1363 | ||
7a232e03 LJ |
1364 | if (!lw->inv_weight) { |
1365 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1366 | lw->inv_weight = 1; | |
1367 | else | |
1368 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1369 | / (lw->weight+1); | |
1370 | } | |
45bf76df IM |
1371 | |
1372 | tmp = (u64)delta_exec * weight; | |
1373 | /* | |
1374 | * Check whether we'd overflow the 64-bit multiplication: | |
1375 | */ | |
194081eb | 1376 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1377 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1378 | WMULT_SHIFT/2); |
1379 | else | |
cf2ab469 | 1380 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1381 | |
ecf691da | 1382 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1383 | } |
1384 | ||
f9305d4a IM |
1385 | static inline unsigned long |
1386 | calc_delta_fair(unsigned long delta_exec, struct load_weight *lw) | |
1387 | { | |
1388 | return calc_delta_mine(delta_exec, NICE_0_LOAD, lw); | |
1389 | } | |
1390 | ||
1091985b | 1391 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1392 | { |
1393 | lw->weight += inc; | |
e89996ae | 1394 | lw->inv_weight = 0; |
45bf76df IM |
1395 | } |
1396 | ||
1091985b | 1397 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1398 | { |
1399 | lw->weight -= dec; | |
e89996ae | 1400 | lw->inv_weight = 0; |
45bf76df IM |
1401 | } |
1402 | ||
2dd73a4f PW |
1403 | /* |
1404 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1405 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1406 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1407 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1408 | * scaled version of the new time slice allocation that they receive on time |
1409 | * slice expiry etc. | |
1410 | */ | |
1411 | ||
dd41f596 IM |
1412 | #define WEIGHT_IDLEPRIO 2 |
1413 | #define WMULT_IDLEPRIO (1 << 31) | |
1414 | ||
1415 | /* | |
1416 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1417 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1418 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1419 | * that remained on nice 0. | |
1420 | * | |
1421 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1422 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1423 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1424 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1425 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1426 | */ |
1427 | static const int prio_to_weight[40] = { | |
254753dc IM |
1428 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1429 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1430 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1431 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1432 | /* 0 */ 1024, 820, 655, 526, 423, | |
1433 | /* 5 */ 335, 272, 215, 172, 137, | |
1434 | /* 10 */ 110, 87, 70, 56, 45, | |
1435 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1436 | }; |
1437 | ||
5714d2de IM |
1438 | /* |
1439 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1440 | * | |
1441 | * In cases where the weight does not change often, we can use the | |
1442 | * precalculated inverse to speed up arithmetics by turning divisions | |
1443 | * into multiplications: | |
1444 | */ | |
dd41f596 | 1445 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1446 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1447 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1448 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1449 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1450 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1451 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1452 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1453 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1454 | }; |
2dd73a4f | 1455 | |
dd41f596 IM |
1456 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup); |
1457 | ||
1458 | /* | |
1459 | * runqueue iterator, to support SMP load-balancing between different | |
1460 | * scheduling classes, without having to expose their internal data | |
1461 | * structures to the load-balancing proper: | |
1462 | */ | |
1463 | struct rq_iterator { | |
1464 | void *arg; | |
1465 | struct task_struct *(*start)(void *); | |
1466 | struct task_struct *(*next)(void *); | |
1467 | }; | |
1468 | ||
e1d1484f PW |
1469 | #ifdef CONFIG_SMP |
1470 | static unsigned long | |
1471 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1472 | unsigned long max_load_move, struct sched_domain *sd, | |
1473 | enum cpu_idle_type idle, int *all_pinned, | |
1474 | int *this_best_prio, struct rq_iterator *iterator); | |
1475 | ||
1476 | static int | |
1477 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
1478 | struct sched_domain *sd, enum cpu_idle_type idle, | |
1479 | struct rq_iterator *iterator); | |
e1d1484f | 1480 | #endif |
dd41f596 | 1481 | |
d842de87 SV |
1482 | #ifdef CONFIG_CGROUP_CPUACCT |
1483 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
1484 | #else | |
1485 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
1486 | #endif | |
1487 | ||
18d95a28 PZ |
1488 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1489 | { | |
1490 | update_load_add(&rq->load, load); | |
1491 | } | |
1492 | ||
1493 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1494 | { | |
1495 | update_load_sub(&rq->load, load); | |
1496 | } | |
1497 | ||
e7693a36 GH |
1498 | #ifdef CONFIG_SMP |
1499 | static unsigned long source_load(int cpu, int type); | |
1500 | static unsigned long target_load(int cpu, int type); | |
1501 | static unsigned long cpu_avg_load_per_task(int cpu); | |
1502 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); | |
18d95a28 PZ |
1503 | #else /* CONFIG_SMP */ |
1504 | ||
1505 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
1506 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) | |
1507 | { | |
1508 | } | |
1509 | #endif | |
1510 | ||
e7693a36 GH |
1511 | #endif /* CONFIG_SMP */ |
1512 | ||
dd41f596 | 1513 | #include "sched_stats.h" |
dd41f596 | 1514 | #include "sched_idletask.c" |
5522d5d5 IM |
1515 | #include "sched_fair.c" |
1516 | #include "sched_rt.c" | |
dd41f596 IM |
1517 | #ifdef CONFIG_SCHED_DEBUG |
1518 | # include "sched_debug.c" | |
1519 | #endif | |
1520 | ||
1521 | #define sched_class_highest (&rt_sched_class) | |
1522 | ||
6363ca57 IM |
1523 | static inline void inc_load(struct rq *rq, const struct task_struct *p) |
1524 | { | |
1525 | update_load_add(&rq->load, p->se.load.weight); | |
1526 | } | |
1527 | ||
1528 | static inline void dec_load(struct rq *rq, const struct task_struct *p) | |
1529 | { | |
1530 | update_load_sub(&rq->load, p->se.load.weight); | |
1531 | } | |
1532 | ||
1533 | static void inc_nr_running(struct task_struct *p, struct rq *rq) | |
9c217245 IM |
1534 | { |
1535 | rq->nr_running++; | |
6363ca57 | 1536 | inc_load(rq, p); |
9c217245 IM |
1537 | } |
1538 | ||
6363ca57 | 1539 | static void dec_nr_running(struct task_struct *p, struct rq *rq) |
9c217245 IM |
1540 | { |
1541 | rq->nr_running--; | |
6363ca57 | 1542 | dec_load(rq, p); |
9c217245 IM |
1543 | } |
1544 | ||
45bf76df IM |
1545 | static void set_load_weight(struct task_struct *p) |
1546 | { | |
1547 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1548 | p->se.load.weight = prio_to_weight[0] * 2; |
1549 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1550 | return; | |
1551 | } | |
45bf76df | 1552 | |
dd41f596 IM |
1553 | /* |
1554 | * SCHED_IDLE tasks get minimal weight: | |
1555 | */ | |
1556 | if (p->policy == SCHED_IDLE) { | |
1557 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1558 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1559 | return; | |
1560 | } | |
71f8bd46 | 1561 | |
dd41f596 IM |
1562 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1563 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1564 | } |
1565 | ||
8159f87e | 1566 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1567 | { |
dd41f596 | 1568 | sched_info_queued(p); |
fd390f6a | 1569 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1570 | p->se.on_rq = 1; |
71f8bd46 IM |
1571 | } |
1572 | ||
69be72c1 | 1573 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1574 | { |
f02231e5 | 1575 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1576 | p->se.on_rq = 0; |
71f8bd46 IM |
1577 | } |
1578 | ||
14531189 | 1579 | /* |
dd41f596 | 1580 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1581 | */ |
14531189 IM |
1582 | static inline int __normal_prio(struct task_struct *p) |
1583 | { | |
dd41f596 | 1584 | return p->static_prio; |
14531189 IM |
1585 | } |
1586 | ||
b29739f9 IM |
1587 | /* |
1588 | * Calculate the expected normal priority: i.e. priority | |
1589 | * without taking RT-inheritance into account. Might be | |
1590 | * boosted by interactivity modifiers. Changes upon fork, | |
1591 | * setprio syscalls, and whenever the interactivity | |
1592 | * estimator recalculates. | |
1593 | */ | |
36c8b586 | 1594 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1595 | { |
1596 | int prio; | |
1597 | ||
e05606d3 | 1598 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1599 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1600 | else | |
1601 | prio = __normal_prio(p); | |
1602 | return prio; | |
1603 | } | |
1604 | ||
1605 | /* | |
1606 | * Calculate the current priority, i.e. the priority | |
1607 | * taken into account by the scheduler. This value might | |
1608 | * be boosted by RT tasks, or might be boosted by | |
1609 | * interactivity modifiers. Will be RT if the task got | |
1610 | * RT-boosted. If not then it returns p->normal_prio. | |
1611 | */ | |
36c8b586 | 1612 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1613 | { |
1614 | p->normal_prio = normal_prio(p); | |
1615 | /* | |
1616 | * If we are RT tasks or we were boosted to RT priority, | |
1617 | * keep the priority unchanged. Otherwise, update priority | |
1618 | * to the normal priority: | |
1619 | */ | |
1620 | if (!rt_prio(p->prio)) | |
1621 | return p->normal_prio; | |
1622 | return p->prio; | |
1623 | } | |
1624 | ||
1da177e4 | 1625 | /* |
dd41f596 | 1626 | * activate_task - move a task to the runqueue. |
1da177e4 | 1627 | */ |
dd41f596 | 1628 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) |
1da177e4 | 1629 | { |
d9514f6c | 1630 | if (task_contributes_to_load(p)) |
dd41f596 | 1631 | rq->nr_uninterruptible--; |
1da177e4 | 1632 | |
8159f87e | 1633 | enqueue_task(rq, p, wakeup); |
6363ca57 | 1634 | inc_nr_running(p, rq); |
1da177e4 LT |
1635 | } |
1636 | ||
1da177e4 LT |
1637 | /* |
1638 | * deactivate_task - remove a task from the runqueue. | |
1639 | */ | |
2e1cb74a | 1640 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) |
1da177e4 | 1641 | { |
d9514f6c | 1642 | if (task_contributes_to_load(p)) |
dd41f596 IM |
1643 | rq->nr_uninterruptible++; |
1644 | ||
69be72c1 | 1645 | dequeue_task(rq, p, sleep); |
6363ca57 | 1646 | dec_nr_running(p, rq); |
1da177e4 LT |
1647 | } |
1648 | ||
1da177e4 LT |
1649 | /** |
1650 | * task_curr - is this task currently executing on a CPU? | |
1651 | * @p: the task in question. | |
1652 | */ | |
36c8b586 | 1653 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1654 | { |
1655 | return cpu_curr(task_cpu(p)) == p; | |
1656 | } | |
1657 | ||
2dd73a4f PW |
1658 | /* Used instead of source_load when we know the type == 0 */ |
1659 | unsigned long weighted_cpuload(const int cpu) | |
1660 | { | |
495eca49 | 1661 | return cpu_rq(cpu)->load.weight; |
dd41f596 IM |
1662 | } |
1663 | ||
1664 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) | |
1665 | { | |
6f505b16 | 1666 | set_task_rq(p, cpu); |
dd41f596 | 1667 | #ifdef CONFIG_SMP |
ce96b5ac DA |
1668 | /* |
1669 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1670 | * successfuly executed on another CPU. We must ensure that updates of | |
1671 | * per-task data have been completed by this moment. | |
1672 | */ | |
1673 | smp_wmb(); | |
dd41f596 | 1674 | task_thread_info(p)->cpu = cpu; |
dd41f596 | 1675 | #endif |
2dd73a4f PW |
1676 | } |
1677 | ||
cb469845 SR |
1678 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1679 | const struct sched_class *prev_class, | |
1680 | int oldprio, int running) | |
1681 | { | |
1682 | if (prev_class != p->sched_class) { | |
1683 | if (prev_class->switched_from) | |
1684 | prev_class->switched_from(rq, p, running); | |
1685 | p->sched_class->switched_to(rq, p, running); | |
1686 | } else | |
1687 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1688 | } | |
1689 | ||
1da177e4 | 1690 | #ifdef CONFIG_SMP |
c65cc870 | 1691 | |
cc367732 IM |
1692 | /* |
1693 | * Is this task likely cache-hot: | |
1694 | */ | |
e7693a36 | 1695 | static int |
cc367732 IM |
1696 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1697 | { | |
1698 | s64 delta; | |
1699 | ||
f540a608 IM |
1700 | /* |
1701 | * Buddy candidates are cache hot: | |
1702 | */ | |
d25ce4cd | 1703 | if (sched_feat(CACHE_HOT_BUDDY) && (&p->se == cfs_rq_of(&p->se)->next)) |
f540a608 IM |
1704 | return 1; |
1705 | ||
cc367732 IM |
1706 | if (p->sched_class != &fair_sched_class) |
1707 | return 0; | |
1708 | ||
6bc1665b IM |
1709 | if (sysctl_sched_migration_cost == -1) |
1710 | return 1; | |
1711 | if (sysctl_sched_migration_cost == 0) | |
1712 | return 0; | |
1713 | ||
cc367732 IM |
1714 | delta = now - p->se.exec_start; |
1715 | ||
1716 | return delta < (s64)sysctl_sched_migration_cost; | |
1717 | } | |
1718 | ||
1719 | ||
dd41f596 | 1720 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 1721 | { |
dd41f596 IM |
1722 | int old_cpu = task_cpu(p); |
1723 | struct rq *old_rq = cpu_rq(old_cpu), *new_rq = cpu_rq(new_cpu); | |
2830cf8c SV |
1724 | struct cfs_rq *old_cfsrq = task_cfs_rq(p), |
1725 | *new_cfsrq = cpu_cfs_rq(old_cfsrq, new_cpu); | |
bbdba7c0 | 1726 | u64 clock_offset; |
dd41f596 IM |
1727 | |
1728 | clock_offset = old_rq->clock - new_rq->clock; | |
6cfb0d5d IM |
1729 | |
1730 | #ifdef CONFIG_SCHEDSTATS | |
1731 | if (p->se.wait_start) | |
1732 | p->se.wait_start -= clock_offset; | |
dd41f596 IM |
1733 | if (p->se.sleep_start) |
1734 | p->se.sleep_start -= clock_offset; | |
1735 | if (p->se.block_start) | |
1736 | p->se.block_start -= clock_offset; | |
cc367732 IM |
1737 | if (old_cpu != new_cpu) { |
1738 | schedstat_inc(p, se.nr_migrations); | |
1739 | if (task_hot(p, old_rq->clock, NULL)) | |
1740 | schedstat_inc(p, se.nr_forced2_migrations); | |
1741 | } | |
6cfb0d5d | 1742 | #endif |
2830cf8c SV |
1743 | p->se.vruntime -= old_cfsrq->min_vruntime - |
1744 | new_cfsrq->min_vruntime; | |
dd41f596 IM |
1745 | |
1746 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
1747 | } |
1748 | ||
70b97a7f | 1749 | struct migration_req { |
1da177e4 | 1750 | struct list_head list; |
1da177e4 | 1751 | |
36c8b586 | 1752 | struct task_struct *task; |
1da177e4 LT |
1753 | int dest_cpu; |
1754 | ||
1da177e4 | 1755 | struct completion done; |
70b97a7f | 1756 | }; |
1da177e4 LT |
1757 | |
1758 | /* | |
1759 | * The task's runqueue lock must be held. | |
1760 | * Returns true if you have to wait for migration thread. | |
1761 | */ | |
36c8b586 | 1762 | static int |
70b97a7f | 1763 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 1764 | { |
70b97a7f | 1765 | struct rq *rq = task_rq(p); |
1da177e4 LT |
1766 | |
1767 | /* | |
1768 | * If the task is not on a runqueue (and not running), then | |
1769 | * it is sufficient to simply update the task's cpu field. | |
1770 | */ | |
dd41f596 | 1771 | if (!p->se.on_rq && !task_running(rq, p)) { |
1da177e4 LT |
1772 | set_task_cpu(p, dest_cpu); |
1773 | return 0; | |
1774 | } | |
1775 | ||
1776 | init_completion(&req->done); | |
1da177e4 LT |
1777 | req->task = p; |
1778 | req->dest_cpu = dest_cpu; | |
1779 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 1780 | |
1da177e4 LT |
1781 | return 1; |
1782 | } | |
1783 | ||
1784 | /* | |
1785 | * wait_task_inactive - wait for a thread to unschedule. | |
1786 | * | |
1787 | * The caller must ensure that the task *will* unschedule sometime soon, | |
1788 | * else this function might spin for a *long* time. This function can't | |
1789 | * be called with interrupts off, or it may introduce deadlock with | |
1790 | * smp_call_function() if an IPI is sent by the same process we are | |
1791 | * waiting to become inactive. | |
1792 | */ | |
36c8b586 | 1793 | void wait_task_inactive(struct task_struct *p) |
1da177e4 LT |
1794 | { |
1795 | unsigned long flags; | |
dd41f596 | 1796 | int running, on_rq; |
70b97a7f | 1797 | struct rq *rq; |
1da177e4 | 1798 | |
3a5c359a AK |
1799 | for (;;) { |
1800 | /* | |
1801 | * We do the initial early heuristics without holding | |
1802 | * any task-queue locks at all. We'll only try to get | |
1803 | * the runqueue lock when things look like they will | |
1804 | * work out! | |
1805 | */ | |
1806 | rq = task_rq(p); | |
fa490cfd | 1807 | |
3a5c359a AK |
1808 | /* |
1809 | * If the task is actively running on another CPU | |
1810 | * still, just relax and busy-wait without holding | |
1811 | * any locks. | |
1812 | * | |
1813 | * NOTE! Since we don't hold any locks, it's not | |
1814 | * even sure that "rq" stays as the right runqueue! | |
1815 | * But we don't care, since "task_running()" will | |
1816 | * return false if the runqueue has changed and p | |
1817 | * is actually now running somewhere else! | |
1818 | */ | |
1819 | while (task_running(rq, p)) | |
1820 | cpu_relax(); | |
fa490cfd | 1821 | |
3a5c359a AK |
1822 | /* |
1823 | * Ok, time to look more closely! We need the rq | |
1824 | * lock now, to be *sure*. If we're wrong, we'll | |
1825 | * just go back and repeat. | |
1826 | */ | |
1827 | rq = task_rq_lock(p, &flags); | |
1828 | running = task_running(rq, p); | |
1829 | on_rq = p->se.on_rq; | |
1830 | task_rq_unlock(rq, &flags); | |
fa490cfd | 1831 | |
3a5c359a AK |
1832 | /* |
1833 | * Was it really running after all now that we | |
1834 | * checked with the proper locks actually held? | |
1835 | * | |
1836 | * Oops. Go back and try again.. | |
1837 | */ | |
1838 | if (unlikely(running)) { | |
1839 | cpu_relax(); | |
1840 | continue; | |
1841 | } | |
fa490cfd | 1842 | |
3a5c359a AK |
1843 | /* |
1844 | * It's not enough that it's not actively running, | |
1845 | * it must be off the runqueue _entirely_, and not | |
1846 | * preempted! | |
1847 | * | |
1848 | * So if it wa still runnable (but just not actively | |
1849 | * running right now), it's preempted, and we should | |
1850 | * yield - it could be a while. | |
1851 | */ | |
1852 | if (unlikely(on_rq)) { | |
1853 | schedule_timeout_uninterruptible(1); | |
1854 | continue; | |
1855 | } | |
fa490cfd | 1856 | |
3a5c359a AK |
1857 | /* |
1858 | * Ahh, all good. It wasn't running, and it wasn't | |
1859 | * runnable, which means that it will never become | |
1860 | * running in the future either. We're all done! | |
1861 | */ | |
1862 | break; | |
1863 | } | |
1da177e4 LT |
1864 | } |
1865 | ||
1866 | /*** | |
1867 | * kick_process - kick a running thread to enter/exit the kernel | |
1868 | * @p: the to-be-kicked thread | |
1869 | * | |
1870 | * Cause a process which is running on another CPU to enter | |
1871 | * kernel-mode, without any delay. (to get signals handled.) | |
1872 | * | |
1873 | * NOTE: this function doesnt have to take the runqueue lock, | |
1874 | * because all it wants to ensure is that the remote task enters | |
1875 | * the kernel. If the IPI races and the task has been migrated | |
1876 | * to another CPU then no harm is done and the purpose has been | |
1877 | * achieved as well. | |
1878 | */ | |
36c8b586 | 1879 | void kick_process(struct task_struct *p) |
1da177e4 LT |
1880 | { |
1881 | int cpu; | |
1882 | ||
1883 | preempt_disable(); | |
1884 | cpu = task_cpu(p); | |
1885 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
1886 | smp_send_reschedule(cpu); | |
1887 | preempt_enable(); | |
1888 | } | |
1889 | ||
1890 | /* | |
2dd73a4f PW |
1891 | * Return a low guess at the load of a migration-source cpu weighted |
1892 | * according to the scheduling class and "nice" value. | |
1da177e4 LT |
1893 | * |
1894 | * We want to under-estimate the load of migration sources, to | |
1895 | * balance conservatively. | |
1896 | */ | |
a9957449 | 1897 | static unsigned long source_load(int cpu, int type) |
1da177e4 | 1898 | { |
70b97a7f | 1899 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1900 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1901 | |
3b0bd9bc | 1902 | if (type == 0) |
dd41f596 | 1903 | return total; |
b910472d | 1904 | |
dd41f596 | 1905 | return min(rq->cpu_load[type-1], total); |
1da177e4 LT |
1906 | } |
1907 | ||
1908 | /* | |
2dd73a4f PW |
1909 | * Return a high guess at the load of a migration-target cpu weighted |
1910 | * according to the scheduling class and "nice" value. | |
1da177e4 | 1911 | */ |
a9957449 | 1912 | static unsigned long target_load(int cpu, int type) |
1da177e4 | 1913 | { |
70b97a7f | 1914 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1915 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f | 1916 | |
7897986b | 1917 | if (type == 0) |
dd41f596 | 1918 | return total; |
3b0bd9bc | 1919 | |
dd41f596 | 1920 | return max(rq->cpu_load[type-1], total); |
2dd73a4f PW |
1921 | } |
1922 | ||
1923 | /* | |
1924 | * Return the average load per task on the cpu's run queue | |
1925 | */ | |
e7693a36 | 1926 | static unsigned long cpu_avg_load_per_task(int cpu) |
2dd73a4f | 1927 | { |
70b97a7f | 1928 | struct rq *rq = cpu_rq(cpu); |
dd41f596 | 1929 | unsigned long total = weighted_cpuload(cpu); |
2dd73a4f PW |
1930 | unsigned long n = rq->nr_running; |
1931 | ||
dd41f596 | 1932 | return n ? total / n : SCHED_LOAD_SCALE; |
1da177e4 LT |
1933 | } |
1934 | ||
147cbb4b NP |
1935 | /* |
1936 | * find_idlest_group finds and returns the least busy CPU group within the | |
1937 | * domain. | |
1938 | */ | |
1939 | static struct sched_group * | |
1940 | find_idlest_group(struct sched_domain *sd, struct task_struct *p, int this_cpu) | |
1941 | { | |
1942 | struct sched_group *idlest = NULL, *this = NULL, *group = sd->groups; | |
1943 | unsigned long min_load = ULONG_MAX, this_load = 0; | |
1944 | int load_idx = sd->forkexec_idx; | |
1945 | int imbalance = 100 + (sd->imbalance_pct-100)/2; | |
1946 | ||
1947 | do { | |
1948 | unsigned long load, avg_load; | |
1949 | int local_group; | |
1950 | int i; | |
1951 | ||
da5a5522 BD |
1952 | /* Skip over this group if it has no CPUs allowed */ |
1953 | if (!cpus_intersects(group->cpumask, p->cpus_allowed)) | |
3a5c359a | 1954 | continue; |
da5a5522 | 1955 | |
147cbb4b | 1956 | local_group = cpu_isset(this_cpu, group->cpumask); |
147cbb4b NP |
1957 | |
1958 | /* Tally up the load of all CPUs in the group */ | |
1959 | avg_load = 0; | |
1960 | ||
1961 | for_each_cpu_mask(i, group->cpumask) { | |
1962 | /* Bias balancing toward cpus of our domain */ | |
1963 | if (local_group) | |
1964 | load = source_load(i, load_idx); | |
1965 | else | |
1966 | load = target_load(i, load_idx); | |
1967 | ||
1968 | avg_load += load; | |
1969 | } | |
1970 | ||
1971 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
1972 | avg_load = sg_div_cpu_power(group, |
1973 | avg_load * SCHED_LOAD_SCALE); | |
147cbb4b NP |
1974 | |
1975 | if (local_group) { | |
1976 | this_load = avg_load; | |
1977 | this = group; | |
1978 | } else if (avg_load < min_load) { | |
1979 | min_load = avg_load; | |
1980 | idlest = group; | |
1981 | } | |
3a5c359a | 1982 | } while (group = group->next, group != sd->groups); |
147cbb4b NP |
1983 | |
1984 | if (!idlest || 100*this_load < imbalance*min_load) | |
1985 | return NULL; | |
1986 | return idlest; | |
1987 | } | |
1988 | ||
1989 | /* | |
0feaece9 | 1990 | * find_idlest_cpu - find the idlest cpu among the cpus in group. |
147cbb4b | 1991 | */ |
95cdf3b7 | 1992 | static int |
7c16ec58 MT |
1993 | find_idlest_cpu(struct sched_group *group, struct task_struct *p, int this_cpu, |
1994 | cpumask_t *tmp) | |
147cbb4b NP |
1995 | { |
1996 | unsigned long load, min_load = ULONG_MAX; | |
1997 | int idlest = -1; | |
1998 | int i; | |
1999 | ||
da5a5522 | 2000 | /* Traverse only the allowed CPUs */ |
7c16ec58 | 2001 | cpus_and(*tmp, group->cpumask, p->cpus_allowed); |
da5a5522 | 2002 | |
7c16ec58 | 2003 | for_each_cpu_mask(i, *tmp) { |
2dd73a4f | 2004 | load = weighted_cpuload(i); |
147cbb4b NP |
2005 | |
2006 | if (load < min_load || (load == min_load && i == this_cpu)) { | |
2007 | min_load = load; | |
2008 | idlest = i; | |
2009 | } | |
2010 | } | |
2011 | ||
2012 | return idlest; | |
2013 | } | |
2014 | ||
476d139c NP |
2015 | /* |
2016 | * sched_balance_self: balance the current task (running on cpu) in domains | |
2017 | * that have the 'flag' flag set. In practice, this is SD_BALANCE_FORK and | |
2018 | * SD_BALANCE_EXEC. | |
2019 | * | |
2020 | * Balance, ie. select the least loaded group. | |
2021 | * | |
2022 | * Returns the target CPU number, or the same CPU if no balancing is needed. | |
2023 | * | |
2024 | * preempt must be disabled. | |
2025 | */ | |
2026 | static int sched_balance_self(int cpu, int flag) | |
2027 | { | |
2028 | struct task_struct *t = current; | |
2029 | struct sched_domain *tmp, *sd = NULL; | |
147cbb4b | 2030 | |
c96d145e | 2031 | for_each_domain(cpu, tmp) { |
9761eea8 IM |
2032 | /* |
2033 | * If power savings logic is enabled for a domain, stop there. | |
2034 | */ | |
5c45bf27 SS |
2035 | if (tmp->flags & SD_POWERSAVINGS_BALANCE) |
2036 | break; | |
476d139c NP |
2037 | if (tmp->flags & flag) |
2038 | sd = tmp; | |
c96d145e | 2039 | } |
476d139c NP |
2040 | |
2041 | while (sd) { | |
7c16ec58 | 2042 | cpumask_t span, tmpmask; |
476d139c | 2043 | struct sched_group *group; |
1a848870 SS |
2044 | int new_cpu, weight; |
2045 | ||
2046 | if (!(sd->flags & flag)) { | |
2047 | sd = sd->child; | |
2048 | continue; | |
2049 | } | |
476d139c NP |
2050 | |
2051 | span = sd->span; | |
2052 | group = find_idlest_group(sd, t, cpu); | |
1a848870 SS |
2053 | if (!group) { |
2054 | sd = sd->child; | |
2055 | continue; | |
2056 | } | |
476d139c | 2057 | |
7c16ec58 | 2058 | new_cpu = find_idlest_cpu(group, t, cpu, &tmpmask); |
1a848870 SS |
2059 | if (new_cpu == -1 || new_cpu == cpu) { |
2060 | /* Now try balancing at a lower domain level of cpu */ | |
2061 | sd = sd->child; | |
2062 | continue; | |
2063 | } | |
476d139c | 2064 | |
1a848870 | 2065 | /* Now try balancing at a lower domain level of new_cpu */ |
476d139c | 2066 | cpu = new_cpu; |
476d139c NP |
2067 | sd = NULL; |
2068 | weight = cpus_weight(span); | |
2069 | for_each_domain(cpu, tmp) { | |
2070 | if (weight <= cpus_weight(tmp->span)) | |
2071 | break; | |
2072 | if (tmp->flags & flag) | |
2073 | sd = tmp; | |
2074 | } | |
2075 | /* while loop will break here if sd == NULL */ | |
2076 | } | |
2077 | ||
2078 | return cpu; | |
2079 | } | |
2080 | ||
2081 | #endif /* CONFIG_SMP */ | |
1da177e4 | 2082 | |
1da177e4 LT |
2083 | /*** |
2084 | * try_to_wake_up - wake up a thread | |
2085 | * @p: the to-be-woken-up thread | |
2086 | * @state: the mask of task states that can be woken | |
2087 | * @sync: do a synchronous wakeup? | |
2088 | * | |
2089 | * Put it on the run-queue if it's not already there. The "current" | |
2090 | * thread is always on the run-queue (except when the actual | |
2091 | * re-schedule is in progress), and as such you're allowed to do | |
2092 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2093 | * runnable without the overhead of this. | |
2094 | * | |
2095 | * returns failure only if the task is already active. | |
2096 | */ | |
36c8b586 | 2097 | static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync) |
1da177e4 | 2098 | { |
cc367732 | 2099 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 LT |
2100 | unsigned long flags; |
2101 | long old_state; | |
70b97a7f | 2102 | struct rq *rq; |
1da177e4 | 2103 | |
b85d0667 IM |
2104 | if (!sched_feat(SYNC_WAKEUPS)) |
2105 | sync = 0; | |
2106 | ||
04e2f174 | 2107 | smp_wmb(); |
1da177e4 LT |
2108 | rq = task_rq_lock(p, &flags); |
2109 | old_state = p->state; | |
2110 | if (!(old_state & state)) | |
2111 | goto out; | |
2112 | ||
dd41f596 | 2113 | if (p->se.on_rq) |
1da177e4 LT |
2114 | goto out_running; |
2115 | ||
2116 | cpu = task_cpu(p); | |
cc367732 | 2117 | orig_cpu = cpu; |
1da177e4 LT |
2118 | this_cpu = smp_processor_id(); |
2119 | ||
2120 | #ifdef CONFIG_SMP | |
2121 | if (unlikely(task_running(rq, p))) | |
2122 | goto out_activate; | |
2123 | ||
5d2f5a61 DA |
2124 | cpu = p->sched_class->select_task_rq(p, sync); |
2125 | if (cpu != orig_cpu) { | |
2126 | set_task_cpu(p, cpu); | |
1da177e4 LT |
2127 | task_rq_unlock(rq, &flags); |
2128 | /* might preempt at this point */ | |
2129 | rq = task_rq_lock(p, &flags); | |
2130 | old_state = p->state; | |
2131 | if (!(old_state & state)) | |
2132 | goto out; | |
dd41f596 | 2133 | if (p->se.on_rq) |
1da177e4 LT |
2134 | goto out_running; |
2135 | ||
2136 | this_cpu = smp_processor_id(); | |
2137 | cpu = task_cpu(p); | |
2138 | } | |
2139 | ||
e7693a36 GH |
2140 | #ifdef CONFIG_SCHEDSTATS |
2141 | schedstat_inc(rq, ttwu_count); | |
2142 | if (cpu == this_cpu) | |
2143 | schedstat_inc(rq, ttwu_local); | |
2144 | else { | |
2145 | struct sched_domain *sd; | |
2146 | for_each_domain(this_cpu, sd) { | |
2147 | if (cpu_isset(cpu, sd->span)) { | |
2148 | schedstat_inc(sd, ttwu_wake_remote); | |
2149 | break; | |
2150 | } | |
2151 | } | |
2152 | } | |
e7693a36 GH |
2153 | #endif |
2154 | ||
1da177e4 LT |
2155 | out_activate: |
2156 | #endif /* CONFIG_SMP */ | |
cc367732 IM |
2157 | schedstat_inc(p, se.nr_wakeups); |
2158 | if (sync) | |
2159 | schedstat_inc(p, se.nr_wakeups_sync); | |
2160 | if (orig_cpu != cpu) | |
2161 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2162 | if (cpu == this_cpu) | |
2163 | schedstat_inc(p, se.nr_wakeups_local); | |
2164 | else | |
2165 | schedstat_inc(p, se.nr_wakeups_remote); | |
2daa3577 | 2166 | update_rq_clock(rq); |
dd41f596 | 2167 | activate_task(rq, p, 1); |
1da177e4 LT |
2168 | success = 1; |
2169 | ||
2170 | out_running: | |
5b82a1b0 MD |
2171 | trace_mark(kernel_sched_wakeup, |
2172 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2173 | p->pid, p->state, rq, p, rq->curr); | |
4ae7d5ce IM |
2174 | check_preempt_curr(rq, p); |
2175 | ||
1da177e4 | 2176 | p->state = TASK_RUNNING; |
9a897c5a SR |
2177 | #ifdef CONFIG_SMP |
2178 | if (p->sched_class->task_wake_up) | |
2179 | p->sched_class->task_wake_up(rq, p); | |
2180 | #endif | |
1da177e4 LT |
2181 | out: |
2182 | task_rq_unlock(rq, &flags); | |
2183 | ||
2184 | return success; | |
2185 | } | |
2186 | ||
7ad5b3a5 | 2187 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2188 | { |
d9514f6c | 2189 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2190 | } |
1da177e4 LT |
2191 | EXPORT_SYMBOL(wake_up_process); |
2192 | ||
7ad5b3a5 | 2193 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2194 | { |
2195 | return try_to_wake_up(p, state, 0); | |
2196 | } | |
2197 | ||
1da177e4 LT |
2198 | /* |
2199 | * Perform scheduler related setup for a newly forked process p. | |
2200 | * p is forked by current. | |
dd41f596 IM |
2201 | * |
2202 | * __sched_fork() is basic setup used by init_idle() too: | |
2203 | */ | |
2204 | static void __sched_fork(struct task_struct *p) | |
2205 | { | |
dd41f596 IM |
2206 | p->se.exec_start = 0; |
2207 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2208 | p->se.prev_sum_exec_runtime = 0; |
4ae7d5ce IM |
2209 | p->se.last_wakeup = 0; |
2210 | p->se.avg_overlap = 0; | |
6cfb0d5d IM |
2211 | |
2212 | #ifdef CONFIG_SCHEDSTATS | |
2213 | p->se.wait_start = 0; | |
dd41f596 IM |
2214 | p->se.sum_sleep_runtime = 0; |
2215 | p->se.sleep_start = 0; | |
dd41f596 IM |
2216 | p->se.block_start = 0; |
2217 | p->se.sleep_max = 0; | |
2218 | p->se.block_max = 0; | |
2219 | p->se.exec_max = 0; | |
eba1ed4b | 2220 | p->se.slice_max = 0; |
dd41f596 | 2221 | p->se.wait_max = 0; |
6cfb0d5d | 2222 | #endif |
476d139c | 2223 | |
fa717060 | 2224 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2225 | p->se.on_rq = 0; |
4a55bd5e | 2226 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2227 | |
e107be36 AK |
2228 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2229 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2230 | #endif | |
2231 | ||
1da177e4 LT |
2232 | /* |
2233 | * We mark the process as running here, but have not actually | |
2234 | * inserted it onto the runqueue yet. This guarantees that | |
2235 | * nobody will actually run it, and a signal or other external | |
2236 | * event cannot wake it up and insert it on the runqueue either. | |
2237 | */ | |
2238 | p->state = TASK_RUNNING; | |
dd41f596 IM |
2239 | } |
2240 | ||
2241 | /* | |
2242 | * fork()/clone()-time setup: | |
2243 | */ | |
2244 | void sched_fork(struct task_struct *p, int clone_flags) | |
2245 | { | |
2246 | int cpu = get_cpu(); | |
2247 | ||
2248 | __sched_fork(p); | |
2249 | ||
2250 | #ifdef CONFIG_SMP | |
2251 | cpu = sched_balance_self(cpu, SD_BALANCE_FORK); | |
2252 | #endif | |
02e4bac2 | 2253 | set_task_cpu(p, cpu); |
b29739f9 IM |
2254 | |
2255 | /* | |
2256 | * Make sure we do not leak PI boosting priority to the child: | |
2257 | */ | |
2258 | p->prio = current->normal_prio; | |
2ddbf952 HS |
2259 | if (!rt_prio(p->prio)) |
2260 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2261 | |
52f17b6c | 2262 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2263 | if (likely(sched_info_on())) |
52f17b6c | 2264 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2265 | #endif |
d6077cb8 | 2266 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2267 | p->oncpu = 0; |
2268 | #endif | |
1da177e4 | 2269 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2270 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2271 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2272 | #endif |
476d139c | 2273 | put_cpu(); |
1da177e4 LT |
2274 | } |
2275 | ||
2276 | /* | |
2277 | * wake_up_new_task - wake up a newly created task for the first time. | |
2278 | * | |
2279 | * This function will do some initial scheduler statistics housekeeping | |
2280 | * that must be done for every newly created context, then puts the task | |
2281 | * on the runqueue and wakes it. | |
2282 | */ | |
7ad5b3a5 | 2283 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2284 | { |
2285 | unsigned long flags; | |
dd41f596 | 2286 | struct rq *rq; |
1da177e4 LT |
2287 | |
2288 | rq = task_rq_lock(p, &flags); | |
147cbb4b | 2289 | BUG_ON(p->state != TASK_RUNNING); |
a8e504d2 | 2290 | update_rq_clock(rq); |
1da177e4 LT |
2291 | |
2292 | p->prio = effective_prio(p); | |
2293 | ||
b9dca1e0 | 2294 | if (!p->sched_class->task_new || !current->se.on_rq) { |
dd41f596 | 2295 | activate_task(rq, p, 0); |
1da177e4 | 2296 | } else { |
1da177e4 | 2297 | /* |
dd41f596 IM |
2298 | * Let the scheduling class do new task startup |
2299 | * management (if any): | |
1da177e4 | 2300 | */ |
ee0827d8 | 2301 | p->sched_class->task_new(rq, p); |
6363ca57 | 2302 | inc_nr_running(p, rq); |
1da177e4 | 2303 | } |
5b82a1b0 MD |
2304 | trace_mark(kernel_sched_wakeup_new, |
2305 | "pid %d state %ld ## rq %p task %p rq->curr %p", | |
2306 | p->pid, p->state, rq, p, rq->curr); | |
dd41f596 | 2307 | check_preempt_curr(rq, p); |
9a897c5a SR |
2308 | #ifdef CONFIG_SMP |
2309 | if (p->sched_class->task_wake_up) | |
2310 | p->sched_class->task_wake_up(rq, p); | |
2311 | #endif | |
dd41f596 | 2312 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2313 | } |
2314 | ||
e107be36 AK |
2315 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2316 | ||
2317 | /** | |
421cee29 RD |
2318 | * preempt_notifier_register - tell me when current is being being preempted & rescheduled |
2319 | * @notifier: notifier struct to register | |
e107be36 AK |
2320 | */ |
2321 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2322 | { | |
2323 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2324 | } | |
2325 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2326 | ||
2327 | /** | |
2328 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2329 | * @notifier: notifier struct to unregister |
e107be36 AK |
2330 | * |
2331 | * This is safe to call from within a preemption notifier. | |
2332 | */ | |
2333 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2334 | { | |
2335 | hlist_del(¬ifier->link); | |
2336 | } | |
2337 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2338 | ||
2339 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2340 | { | |
2341 | struct preempt_notifier *notifier; | |
2342 | struct hlist_node *node; | |
2343 | ||
2344 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2345 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2346 | } | |
2347 | ||
2348 | static void | |
2349 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2350 | struct task_struct *next) | |
2351 | { | |
2352 | struct preempt_notifier *notifier; | |
2353 | struct hlist_node *node; | |
2354 | ||
2355 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2356 | notifier->ops->sched_out(notifier, next); | |
2357 | } | |
2358 | ||
2359 | #else | |
2360 | ||
2361 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2362 | { | |
2363 | } | |
2364 | ||
2365 | static void | |
2366 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2367 | struct task_struct *next) | |
2368 | { | |
2369 | } | |
2370 | ||
2371 | #endif | |
2372 | ||
4866cde0 NP |
2373 | /** |
2374 | * prepare_task_switch - prepare to switch tasks | |
2375 | * @rq: the runqueue preparing to switch | |
421cee29 | 2376 | * @prev: the current task that is being switched out |
4866cde0 NP |
2377 | * @next: the task we are going to switch to. |
2378 | * | |
2379 | * This is called with the rq lock held and interrupts off. It must | |
2380 | * be paired with a subsequent finish_task_switch after the context | |
2381 | * switch. | |
2382 | * | |
2383 | * prepare_task_switch sets up locking and calls architecture specific | |
2384 | * hooks. | |
2385 | */ | |
e107be36 AK |
2386 | static inline void |
2387 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2388 | struct task_struct *next) | |
4866cde0 | 2389 | { |
e107be36 | 2390 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2391 | prepare_lock_switch(rq, next); |
2392 | prepare_arch_switch(next); | |
2393 | } | |
2394 | ||
1da177e4 LT |
2395 | /** |
2396 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2397 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2398 | * @prev: the thread we just switched away from. |
2399 | * | |
4866cde0 NP |
2400 | * finish_task_switch must be called after the context switch, paired |
2401 | * with a prepare_task_switch call before the context switch. | |
2402 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2403 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2404 | * |
2405 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2406 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2407 | * with the lock held can cause deadlocks; see schedule() for |
2408 | * details.) | |
2409 | */ | |
a9957449 | 2410 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2411 | __releases(rq->lock) |
2412 | { | |
1da177e4 | 2413 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2414 | long prev_state; |
1da177e4 LT |
2415 | |
2416 | rq->prev_mm = NULL; | |
2417 | ||
2418 | /* | |
2419 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2420 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2421 | * schedule one last time. The schedule call will never return, and |
2422 | * the scheduled task must drop that reference. | |
c394cc9f | 2423 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2424 | * still held, otherwise prev could be scheduled on another cpu, die |
2425 | * there before we look at prev->state, and then the reference would | |
2426 | * be dropped twice. | |
2427 | * Manfred Spraul <manfred@colorfullife.com> | |
2428 | */ | |
55a101f8 | 2429 | prev_state = prev->state; |
4866cde0 NP |
2430 | finish_arch_switch(prev); |
2431 | finish_lock_switch(rq, prev); | |
9a897c5a SR |
2432 | #ifdef CONFIG_SMP |
2433 | if (current->sched_class->post_schedule) | |
2434 | current->sched_class->post_schedule(rq); | |
2435 | #endif | |
e8fa1362 | 2436 | |
e107be36 | 2437 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2438 | if (mm) |
2439 | mmdrop(mm); | |
c394cc9f | 2440 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2441 | /* |
2442 | * Remove function-return probe instances associated with this | |
2443 | * task and put them back on the free list. | |
9761eea8 | 2444 | */ |
c6fd91f0 | 2445 | kprobe_flush_task(prev); |
1da177e4 | 2446 | put_task_struct(prev); |
c6fd91f0 | 2447 | } |
1da177e4 LT |
2448 | } |
2449 | ||
2450 | /** | |
2451 | * schedule_tail - first thing a freshly forked thread must call. | |
2452 | * @prev: the thread we just switched away from. | |
2453 | */ | |
36c8b586 | 2454 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2455 | __releases(rq->lock) |
2456 | { | |
70b97a7f IM |
2457 | struct rq *rq = this_rq(); |
2458 | ||
4866cde0 NP |
2459 | finish_task_switch(rq, prev); |
2460 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW | |
2461 | /* In this case, finish_task_switch does not reenable preemption */ | |
2462 | preempt_enable(); | |
2463 | #endif | |
1da177e4 | 2464 | if (current->set_child_tid) |
b488893a | 2465 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2466 | } |
2467 | ||
2468 | /* | |
2469 | * context_switch - switch to the new MM and the new | |
2470 | * thread's register state. | |
2471 | */ | |
dd41f596 | 2472 | static inline void |
70b97a7f | 2473 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2474 | struct task_struct *next) |
1da177e4 | 2475 | { |
dd41f596 | 2476 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2477 | |
e107be36 | 2478 | prepare_task_switch(rq, prev, next); |
5b82a1b0 MD |
2479 | trace_mark(kernel_sched_schedule, |
2480 | "prev_pid %d next_pid %d prev_state %ld " | |
2481 | "## rq %p prev %p next %p", | |
2482 | prev->pid, next->pid, prev->state, | |
2483 | rq, prev, next); | |
dd41f596 IM |
2484 | mm = next->mm; |
2485 | oldmm = prev->active_mm; | |
9226d125 ZA |
2486 | /* |
2487 | * For paravirt, this is coupled with an exit in switch_to to | |
2488 | * combine the page table reload and the switch backend into | |
2489 | * one hypercall. | |
2490 | */ | |
2491 | arch_enter_lazy_cpu_mode(); | |
2492 | ||
dd41f596 | 2493 | if (unlikely(!mm)) { |
1da177e4 LT |
2494 | next->active_mm = oldmm; |
2495 | atomic_inc(&oldmm->mm_count); | |
2496 | enter_lazy_tlb(oldmm, next); | |
2497 | } else | |
2498 | switch_mm(oldmm, mm, next); | |
2499 | ||
dd41f596 | 2500 | if (unlikely(!prev->mm)) { |
1da177e4 | 2501 | prev->active_mm = NULL; |
1da177e4 LT |
2502 | rq->prev_mm = oldmm; |
2503 | } | |
3a5f5e48 IM |
2504 | /* |
2505 | * Since the runqueue lock will be released by the next | |
2506 | * task (which is an invalid locking op but in the case | |
2507 | * of the scheduler it's an obvious special-case), so we | |
2508 | * do an early lockdep release here: | |
2509 | */ | |
2510 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2511 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2512 | #endif |
1da177e4 LT |
2513 | |
2514 | /* Here we just switch the register state and the stack. */ | |
2515 | switch_to(prev, next, prev); | |
2516 | ||
dd41f596 IM |
2517 | barrier(); |
2518 | /* | |
2519 | * this_rq must be evaluated again because prev may have moved | |
2520 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2521 | * frame will be invalid. | |
2522 | */ | |
2523 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2524 | } |
2525 | ||
2526 | /* | |
2527 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2528 | * | |
2529 | * externally visible scheduler statistics: current number of runnable | |
2530 | * threads, current number of uninterruptible-sleeping threads, total | |
2531 | * number of context switches performed since bootup. | |
2532 | */ | |
2533 | unsigned long nr_running(void) | |
2534 | { | |
2535 | unsigned long i, sum = 0; | |
2536 | ||
2537 | for_each_online_cpu(i) | |
2538 | sum += cpu_rq(i)->nr_running; | |
2539 | ||
2540 | return sum; | |
2541 | } | |
2542 | ||
2543 | unsigned long nr_uninterruptible(void) | |
2544 | { | |
2545 | unsigned long i, sum = 0; | |
2546 | ||
0a945022 | 2547 | for_each_possible_cpu(i) |
1da177e4 LT |
2548 | sum += cpu_rq(i)->nr_uninterruptible; |
2549 | ||
2550 | /* | |
2551 | * Since we read the counters lockless, it might be slightly | |
2552 | * inaccurate. Do not allow it to go below zero though: | |
2553 | */ | |
2554 | if (unlikely((long)sum < 0)) | |
2555 | sum = 0; | |
2556 | ||
2557 | return sum; | |
2558 | } | |
2559 | ||
2560 | unsigned long long nr_context_switches(void) | |
2561 | { | |
cc94abfc SR |
2562 | int i; |
2563 | unsigned long long sum = 0; | |
1da177e4 | 2564 | |
0a945022 | 2565 | for_each_possible_cpu(i) |
1da177e4 LT |
2566 | sum += cpu_rq(i)->nr_switches; |
2567 | ||
2568 | return sum; | |
2569 | } | |
2570 | ||
2571 | unsigned long nr_iowait(void) | |
2572 | { | |
2573 | unsigned long i, sum = 0; | |
2574 | ||
0a945022 | 2575 | for_each_possible_cpu(i) |
1da177e4 LT |
2576 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2577 | ||
2578 | return sum; | |
2579 | } | |
2580 | ||
db1b1fef JS |
2581 | unsigned long nr_active(void) |
2582 | { | |
2583 | unsigned long i, running = 0, uninterruptible = 0; | |
2584 | ||
2585 | for_each_online_cpu(i) { | |
2586 | running += cpu_rq(i)->nr_running; | |
2587 | uninterruptible += cpu_rq(i)->nr_uninterruptible; | |
2588 | } | |
2589 | ||
2590 | if (unlikely((long)uninterruptible < 0)) | |
2591 | uninterruptible = 0; | |
2592 | ||
2593 | return running + uninterruptible; | |
2594 | } | |
2595 | ||
48f24c4d | 2596 | /* |
dd41f596 IM |
2597 | * Update rq->cpu_load[] statistics. This function is usually called every |
2598 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 2599 | */ |
dd41f596 | 2600 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 2601 | { |
495eca49 | 2602 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
2603 | int i, scale; |
2604 | ||
2605 | this_rq->nr_load_updates++; | |
dd41f596 IM |
2606 | |
2607 | /* Update our load: */ | |
2608 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
2609 | unsigned long old_load, new_load; | |
2610 | ||
2611 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
2612 | ||
2613 | old_load = this_rq->cpu_load[i]; | |
2614 | new_load = this_load; | |
a25707f3 IM |
2615 | /* |
2616 | * Round up the averaging division if load is increasing. This | |
2617 | * prevents us from getting stuck on 9 if the load is 10, for | |
2618 | * example. | |
2619 | */ | |
2620 | if (new_load > old_load) | |
2621 | new_load += scale-1; | |
dd41f596 IM |
2622 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
2623 | } | |
48f24c4d IM |
2624 | } |
2625 | ||
dd41f596 IM |
2626 | #ifdef CONFIG_SMP |
2627 | ||
1da177e4 LT |
2628 | /* |
2629 | * double_rq_lock - safely lock two runqueues | |
2630 | * | |
2631 | * Note this does not disable interrupts like task_rq_lock, | |
2632 | * you need to do so manually before calling. | |
2633 | */ | |
70b97a7f | 2634 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2635 | __acquires(rq1->lock) |
2636 | __acquires(rq2->lock) | |
2637 | { | |
054b9108 | 2638 | BUG_ON(!irqs_disabled()); |
1da177e4 LT |
2639 | if (rq1 == rq2) { |
2640 | spin_lock(&rq1->lock); | |
2641 | __acquire(rq2->lock); /* Fake it out ;) */ | |
2642 | } else { | |
c96d145e | 2643 | if (rq1 < rq2) { |
1da177e4 LT |
2644 | spin_lock(&rq1->lock); |
2645 | spin_lock(&rq2->lock); | |
2646 | } else { | |
2647 | spin_lock(&rq2->lock); | |
2648 | spin_lock(&rq1->lock); | |
2649 | } | |
2650 | } | |
6e82a3be IM |
2651 | update_rq_clock(rq1); |
2652 | update_rq_clock(rq2); | |
1da177e4 LT |
2653 | } |
2654 | ||
2655 | /* | |
2656 | * double_rq_unlock - safely unlock two runqueues | |
2657 | * | |
2658 | * Note this does not restore interrupts like task_rq_unlock, | |
2659 | * you need to do so manually after calling. | |
2660 | */ | |
70b97a7f | 2661 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) |
1da177e4 LT |
2662 | __releases(rq1->lock) |
2663 | __releases(rq2->lock) | |
2664 | { | |
2665 | spin_unlock(&rq1->lock); | |
2666 | if (rq1 != rq2) | |
2667 | spin_unlock(&rq2->lock); | |
2668 | else | |
2669 | __release(rq2->lock); | |
2670 | } | |
2671 | ||
2672 | /* | |
2673 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
2674 | */ | |
e8fa1362 | 2675 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1da177e4 LT |
2676 | __releases(this_rq->lock) |
2677 | __acquires(busiest->lock) | |
2678 | __acquires(this_rq->lock) | |
2679 | { | |
e8fa1362 SR |
2680 | int ret = 0; |
2681 | ||
054b9108 KK |
2682 | if (unlikely(!irqs_disabled())) { |
2683 | /* printk() doesn't work good under rq->lock */ | |
2684 | spin_unlock(&this_rq->lock); | |
2685 | BUG_ON(1); | |
2686 | } | |
1da177e4 | 2687 | if (unlikely(!spin_trylock(&busiest->lock))) { |
c96d145e | 2688 | if (busiest < this_rq) { |
1da177e4 LT |
2689 | spin_unlock(&this_rq->lock); |
2690 | spin_lock(&busiest->lock); | |
2691 | spin_lock(&this_rq->lock); | |
e8fa1362 | 2692 | ret = 1; |
1da177e4 LT |
2693 | } else |
2694 | spin_lock(&busiest->lock); | |
2695 | } | |
e8fa1362 | 2696 | return ret; |
1da177e4 LT |
2697 | } |
2698 | ||
1da177e4 LT |
2699 | /* |
2700 | * If dest_cpu is allowed for this process, migrate the task to it. | |
2701 | * This is accomplished by forcing the cpu_allowed mask to only | |
41a2d6cf | 2702 | * allow dest_cpu, which will force the cpu onto dest_cpu. Then |
1da177e4 LT |
2703 | * the cpu_allowed mask is restored. |
2704 | */ | |
36c8b586 | 2705 | static void sched_migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2706 | { |
70b97a7f | 2707 | struct migration_req req; |
1da177e4 | 2708 | unsigned long flags; |
70b97a7f | 2709 | struct rq *rq; |
1da177e4 LT |
2710 | |
2711 | rq = task_rq_lock(p, &flags); | |
2712 | if (!cpu_isset(dest_cpu, p->cpus_allowed) | |
2713 | || unlikely(cpu_is_offline(dest_cpu))) | |
2714 | goto out; | |
2715 | ||
2716 | /* force the process onto the specified CPU */ | |
2717 | if (migrate_task(p, dest_cpu, &req)) { | |
2718 | /* Need to wait for migration thread (might exit: take ref). */ | |
2719 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 2720 | |
1da177e4 LT |
2721 | get_task_struct(mt); |
2722 | task_rq_unlock(rq, &flags); | |
2723 | wake_up_process(mt); | |
2724 | put_task_struct(mt); | |
2725 | wait_for_completion(&req.done); | |
36c8b586 | 2726 | |
1da177e4 LT |
2727 | return; |
2728 | } | |
2729 | out: | |
2730 | task_rq_unlock(rq, &flags); | |
2731 | } | |
2732 | ||
2733 | /* | |
476d139c NP |
2734 | * sched_exec - execve() is a valuable balancing opportunity, because at |
2735 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 LT |
2736 | */ |
2737 | void sched_exec(void) | |
2738 | { | |
1da177e4 | 2739 | int new_cpu, this_cpu = get_cpu(); |
476d139c | 2740 | new_cpu = sched_balance_self(this_cpu, SD_BALANCE_EXEC); |
1da177e4 | 2741 | put_cpu(); |
476d139c NP |
2742 | if (new_cpu != this_cpu) |
2743 | sched_migrate_task(current, new_cpu); | |
1da177e4 LT |
2744 | } |
2745 | ||
2746 | /* | |
2747 | * pull_task - move a task from a remote runqueue to the local runqueue. | |
2748 | * Both runqueues must be locked. | |
2749 | */ | |
dd41f596 IM |
2750 | static void pull_task(struct rq *src_rq, struct task_struct *p, |
2751 | struct rq *this_rq, int this_cpu) | |
1da177e4 | 2752 | { |
2e1cb74a | 2753 | deactivate_task(src_rq, p, 0); |
1da177e4 | 2754 | set_task_cpu(p, this_cpu); |
dd41f596 | 2755 | activate_task(this_rq, p, 0); |
1da177e4 LT |
2756 | /* |
2757 | * Note that idle threads have a prio of MAX_PRIO, for this test | |
2758 | * to be always true for them. | |
2759 | */ | |
dd41f596 | 2760 | check_preempt_curr(this_rq, p); |
1da177e4 LT |
2761 | } |
2762 | ||
2763 | /* | |
2764 | * can_migrate_task - may task p from runqueue rq be migrated to this_cpu? | |
2765 | */ | |
858119e1 | 2766 | static |
70b97a7f | 2767 | int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu, |
d15bcfdb | 2768 | struct sched_domain *sd, enum cpu_idle_type idle, |
95cdf3b7 | 2769 | int *all_pinned) |
1da177e4 LT |
2770 | { |
2771 | /* | |
2772 | * We do not migrate tasks that are: | |
2773 | * 1) running (obviously), or | |
2774 | * 2) cannot be migrated to this CPU due to cpus_allowed, or | |
2775 | * 3) are cache-hot on their current CPU. | |
2776 | */ | |
cc367732 IM |
2777 | if (!cpu_isset(this_cpu, p->cpus_allowed)) { |
2778 | schedstat_inc(p, se.nr_failed_migrations_affine); | |
1da177e4 | 2779 | return 0; |
cc367732 | 2780 | } |
81026794 NP |
2781 | *all_pinned = 0; |
2782 | ||
cc367732 IM |
2783 | if (task_running(rq, p)) { |
2784 | schedstat_inc(p, se.nr_failed_migrations_running); | |
81026794 | 2785 | return 0; |
cc367732 | 2786 | } |
1da177e4 | 2787 | |
da84d961 IM |
2788 | /* |
2789 | * Aggressive migration if: | |
2790 | * 1) task is cache cold, or | |
2791 | * 2) too many balance attempts have failed. | |
2792 | */ | |
2793 | ||
6bc1665b IM |
2794 | if (!task_hot(p, rq->clock, sd) || |
2795 | sd->nr_balance_failed > sd->cache_nice_tries) { | |
da84d961 | 2796 | #ifdef CONFIG_SCHEDSTATS |
cc367732 | 2797 | if (task_hot(p, rq->clock, sd)) { |
da84d961 | 2798 | schedstat_inc(sd, lb_hot_gained[idle]); |
cc367732 IM |
2799 | schedstat_inc(p, se.nr_forced_migrations); |
2800 | } | |
da84d961 IM |
2801 | #endif |
2802 | return 1; | |
2803 | } | |
2804 | ||
cc367732 IM |
2805 | if (task_hot(p, rq->clock, sd)) { |
2806 | schedstat_inc(p, se.nr_failed_migrations_hot); | |
da84d961 | 2807 | return 0; |
cc367732 | 2808 | } |
1da177e4 LT |
2809 | return 1; |
2810 | } | |
2811 | ||
e1d1484f PW |
2812 | static unsigned long |
2813 | balance_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2814 | unsigned long max_load_move, struct sched_domain *sd, | |
2815 | enum cpu_idle_type idle, int *all_pinned, | |
2816 | int *this_best_prio, struct rq_iterator *iterator) | |
1da177e4 | 2817 | { |
b82d9fdd | 2818 | int loops = 0, pulled = 0, pinned = 0, skip_for_load; |
dd41f596 IM |
2819 | struct task_struct *p; |
2820 | long rem_load_move = max_load_move; | |
1da177e4 | 2821 | |
e1d1484f | 2822 | if (max_load_move == 0) |
1da177e4 LT |
2823 | goto out; |
2824 | ||
81026794 NP |
2825 | pinned = 1; |
2826 | ||
1da177e4 | 2827 | /* |
dd41f596 | 2828 | * Start the load-balancing iterator: |
1da177e4 | 2829 | */ |
dd41f596 IM |
2830 | p = iterator->start(iterator->arg); |
2831 | next: | |
b82d9fdd | 2832 | if (!p || loops++ > sysctl_sched_nr_migrate) |
1da177e4 | 2833 | goto out; |
50ddd969 | 2834 | /* |
b82d9fdd | 2835 | * To help distribute high priority tasks across CPUs we don't |
50ddd969 PW |
2836 | * skip a task if it will be the highest priority task (i.e. smallest |
2837 | * prio value) on its new queue regardless of its load weight | |
2838 | */ | |
dd41f596 IM |
2839 | skip_for_load = (p->se.load.weight >> 1) > rem_load_move + |
2840 | SCHED_LOAD_SCALE_FUZZ; | |
a4ac01c3 | 2841 | if ((skip_for_load && p->prio >= *this_best_prio) || |
dd41f596 | 2842 | !can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { |
dd41f596 IM |
2843 | p = iterator->next(iterator->arg); |
2844 | goto next; | |
1da177e4 LT |
2845 | } |
2846 | ||
dd41f596 | 2847 | pull_task(busiest, p, this_rq, this_cpu); |
1da177e4 | 2848 | pulled++; |
dd41f596 | 2849 | rem_load_move -= p->se.load.weight; |
1da177e4 | 2850 | |
2dd73a4f | 2851 | /* |
b82d9fdd | 2852 | * We only want to steal up to the prescribed amount of weighted load. |
2dd73a4f | 2853 | */ |
e1d1484f | 2854 | if (rem_load_move > 0) { |
a4ac01c3 PW |
2855 | if (p->prio < *this_best_prio) |
2856 | *this_best_prio = p->prio; | |
dd41f596 IM |
2857 | p = iterator->next(iterator->arg); |
2858 | goto next; | |
1da177e4 LT |
2859 | } |
2860 | out: | |
2861 | /* | |
e1d1484f | 2862 | * Right now, this is one of only two places pull_task() is called, |
1da177e4 LT |
2863 | * so we can safely collect pull_task() stats here rather than |
2864 | * inside pull_task(). | |
2865 | */ | |
2866 | schedstat_add(sd, lb_gained[idle], pulled); | |
81026794 NP |
2867 | |
2868 | if (all_pinned) | |
2869 | *all_pinned = pinned; | |
e1d1484f PW |
2870 | |
2871 | return max_load_move - rem_load_move; | |
1da177e4 LT |
2872 | } |
2873 | ||
dd41f596 | 2874 | /* |
43010659 PW |
2875 | * move_tasks tries to move up to max_load_move weighted load from busiest to |
2876 | * this_rq, as part of a balancing operation within domain "sd". | |
2877 | * Returns 1 if successful and 0 otherwise. | |
dd41f596 IM |
2878 | * |
2879 | * Called with both runqueues locked. | |
2880 | */ | |
2881 | static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
43010659 | 2882 | unsigned long max_load_move, |
dd41f596 IM |
2883 | struct sched_domain *sd, enum cpu_idle_type idle, |
2884 | int *all_pinned) | |
2885 | { | |
5522d5d5 | 2886 | const struct sched_class *class = sched_class_highest; |
43010659 | 2887 | unsigned long total_load_moved = 0; |
a4ac01c3 | 2888 | int this_best_prio = this_rq->curr->prio; |
dd41f596 IM |
2889 | |
2890 | do { | |
43010659 PW |
2891 | total_load_moved += |
2892 | class->load_balance(this_rq, this_cpu, busiest, | |
e1d1484f | 2893 | max_load_move - total_load_moved, |
a4ac01c3 | 2894 | sd, idle, all_pinned, &this_best_prio); |
dd41f596 | 2895 | class = class->next; |
43010659 | 2896 | } while (class && max_load_move > total_load_moved); |
dd41f596 | 2897 | |
43010659 PW |
2898 | return total_load_moved > 0; |
2899 | } | |
2900 | ||
e1d1484f PW |
2901 | static int |
2902 | iter_move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2903 | struct sched_domain *sd, enum cpu_idle_type idle, | |
2904 | struct rq_iterator *iterator) | |
2905 | { | |
2906 | struct task_struct *p = iterator->start(iterator->arg); | |
2907 | int pinned = 0; | |
2908 | ||
2909 | while (p) { | |
2910 | if (can_migrate_task(p, busiest, this_cpu, sd, idle, &pinned)) { | |
2911 | pull_task(busiest, p, this_rq, this_cpu); | |
2912 | /* | |
2913 | * Right now, this is only the second place pull_task() | |
2914 | * is called, so we can safely collect pull_task() | |
2915 | * stats here rather than inside pull_task(). | |
2916 | */ | |
2917 | schedstat_inc(sd, lb_gained[idle]); | |
2918 | ||
2919 | return 1; | |
2920 | } | |
2921 | p = iterator->next(iterator->arg); | |
2922 | } | |
2923 | ||
2924 | return 0; | |
2925 | } | |
2926 | ||
43010659 PW |
2927 | /* |
2928 | * move_one_task tries to move exactly one task from busiest to this_rq, as | |
2929 | * part of active balancing operations within "domain". | |
2930 | * Returns 1 if successful and 0 otherwise. | |
2931 | * | |
2932 | * Called with both runqueues locked. | |
2933 | */ | |
2934 | static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest, | |
2935 | struct sched_domain *sd, enum cpu_idle_type idle) | |
2936 | { | |
5522d5d5 | 2937 | const struct sched_class *class; |
43010659 PW |
2938 | |
2939 | for (class = sched_class_highest; class; class = class->next) | |
e1d1484f | 2940 | if (class->move_one_task(this_rq, this_cpu, busiest, sd, idle)) |
43010659 PW |
2941 | return 1; |
2942 | ||
2943 | return 0; | |
dd41f596 IM |
2944 | } |
2945 | ||
1da177e4 LT |
2946 | /* |
2947 | * find_busiest_group finds and returns the busiest CPU group within the | |
48f24c4d IM |
2948 | * domain. It calculates and returns the amount of weighted load which |
2949 | * should be moved to restore balance via the imbalance parameter. | |
1da177e4 LT |
2950 | */ |
2951 | static struct sched_group * | |
2952 | find_busiest_group(struct sched_domain *sd, int this_cpu, | |
dd41f596 | 2953 | unsigned long *imbalance, enum cpu_idle_type idle, |
7c16ec58 | 2954 | int *sd_idle, const cpumask_t *cpus, int *balance) |
1da177e4 LT |
2955 | { |
2956 | struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; | |
2957 | unsigned long max_load, avg_load, total_load, this_load, total_pwr; | |
0c117f1b | 2958 | unsigned long max_pull; |
2dd73a4f PW |
2959 | unsigned long busiest_load_per_task, busiest_nr_running; |
2960 | unsigned long this_load_per_task, this_nr_running; | |
908a7c1b | 2961 | int load_idx, group_imb = 0; |
5c45bf27 SS |
2962 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
2963 | int power_savings_balance = 1; | |
2964 | unsigned long leader_nr_running = 0, min_load_per_task = 0; | |
2965 | unsigned long min_nr_running = ULONG_MAX; | |
2966 | struct sched_group *group_min = NULL, *group_leader = NULL; | |
2967 | #endif | |
1da177e4 LT |
2968 | |
2969 | max_load = this_load = total_load = total_pwr = 0; | |
2dd73a4f PW |
2970 | busiest_load_per_task = busiest_nr_running = 0; |
2971 | this_load_per_task = this_nr_running = 0; | |
d15bcfdb | 2972 | if (idle == CPU_NOT_IDLE) |
7897986b | 2973 | load_idx = sd->busy_idx; |
d15bcfdb | 2974 | else if (idle == CPU_NEWLY_IDLE) |
7897986b NP |
2975 | load_idx = sd->newidle_idx; |
2976 | else | |
2977 | load_idx = sd->idle_idx; | |
1da177e4 LT |
2978 | |
2979 | do { | |
908a7c1b | 2980 | unsigned long load, group_capacity, max_cpu_load, min_cpu_load; |
1da177e4 LT |
2981 | int local_group; |
2982 | int i; | |
908a7c1b | 2983 | int __group_imb = 0; |
783609c6 | 2984 | unsigned int balance_cpu = -1, first_idle_cpu = 0; |
2dd73a4f | 2985 | unsigned long sum_nr_running, sum_weighted_load; |
1da177e4 LT |
2986 | |
2987 | local_group = cpu_isset(this_cpu, group->cpumask); | |
2988 | ||
783609c6 SS |
2989 | if (local_group) |
2990 | balance_cpu = first_cpu(group->cpumask); | |
2991 | ||
1da177e4 | 2992 | /* Tally up the load of all CPUs in the group */ |
2dd73a4f | 2993 | sum_weighted_load = sum_nr_running = avg_load = 0; |
908a7c1b KC |
2994 | max_cpu_load = 0; |
2995 | min_cpu_load = ~0UL; | |
1da177e4 LT |
2996 | |
2997 | for_each_cpu_mask(i, group->cpumask) { | |
0a2966b4 CL |
2998 | struct rq *rq; |
2999 | ||
3000 | if (!cpu_isset(i, *cpus)) | |
3001 | continue; | |
3002 | ||
3003 | rq = cpu_rq(i); | |
2dd73a4f | 3004 | |
9439aab8 | 3005 | if (*sd_idle && rq->nr_running) |
5969fe06 NP |
3006 | *sd_idle = 0; |
3007 | ||
1da177e4 | 3008 | /* Bias balancing toward cpus of our domain */ |
783609c6 SS |
3009 | if (local_group) { |
3010 | if (idle_cpu(i) && !first_idle_cpu) { | |
3011 | first_idle_cpu = 1; | |
3012 | balance_cpu = i; | |
3013 | } | |
3014 | ||
a2000572 | 3015 | load = target_load(i, load_idx); |
908a7c1b | 3016 | } else { |
a2000572 | 3017 | load = source_load(i, load_idx); |
908a7c1b KC |
3018 | if (load > max_cpu_load) |
3019 | max_cpu_load = load; | |
3020 | if (min_cpu_load > load) | |
3021 | min_cpu_load = load; | |
3022 | } | |
1da177e4 LT |
3023 | |
3024 | avg_load += load; | |
2dd73a4f | 3025 | sum_nr_running += rq->nr_running; |
dd41f596 | 3026 | sum_weighted_load += weighted_cpuload(i); |
1da177e4 LT |
3027 | } |
3028 | ||
783609c6 SS |
3029 | /* |
3030 | * First idle cpu or the first cpu(busiest) in this sched group | |
3031 | * is eligible for doing load balancing at this and above | |
9439aab8 SS |
3032 | * domains. In the newly idle case, we will allow all the cpu's |
3033 | * to do the newly idle load balance. | |
783609c6 | 3034 | */ |
9439aab8 SS |
3035 | if (idle != CPU_NEWLY_IDLE && local_group && |
3036 | balance_cpu != this_cpu && balance) { | |
783609c6 SS |
3037 | *balance = 0; |
3038 | goto ret; | |
3039 | } | |
3040 | ||
1da177e4 | 3041 | total_load += avg_load; |
5517d86b | 3042 | total_pwr += group->__cpu_power; |
1da177e4 LT |
3043 | |
3044 | /* Adjust by relative CPU power of the group */ | |
5517d86b ED |
3045 | avg_load = sg_div_cpu_power(group, |
3046 | avg_load * SCHED_LOAD_SCALE); | |
1da177e4 | 3047 | |
908a7c1b KC |
3048 | if ((max_cpu_load - min_cpu_load) > SCHED_LOAD_SCALE) |
3049 | __group_imb = 1; | |
3050 | ||
5517d86b | 3051 | group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; |
5c45bf27 | 3052 | |
1da177e4 LT |
3053 | if (local_group) { |
3054 | this_load = avg_load; | |
3055 | this = group; | |
2dd73a4f PW |
3056 | this_nr_running = sum_nr_running; |
3057 | this_load_per_task = sum_weighted_load; | |
3058 | } else if (avg_load > max_load && | |
908a7c1b | 3059 | (sum_nr_running > group_capacity || __group_imb)) { |
1da177e4 LT |
3060 | max_load = avg_load; |
3061 | busiest = group; | |
2dd73a4f PW |
3062 | busiest_nr_running = sum_nr_running; |
3063 | busiest_load_per_task = sum_weighted_load; | |
908a7c1b | 3064 | group_imb = __group_imb; |
1da177e4 | 3065 | } |
5c45bf27 SS |
3066 | |
3067 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) | |
3068 | /* | |
3069 | * Busy processors will not participate in power savings | |
3070 | * balance. | |
3071 | */ | |
dd41f596 IM |
3072 | if (idle == CPU_NOT_IDLE || |
3073 | !(sd->flags & SD_POWERSAVINGS_BALANCE)) | |
3074 | goto group_next; | |
5c45bf27 SS |
3075 | |
3076 | /* | |
3077 | * If the local group is idle or completely loaded | |
3078 | * no need to do power savings balance at this domain | |
3079 | */ | |
3080 | if (local_group && (this_nr_running >= group_capacity || | |
3081 | !this_nr_running)) | |
3082 | power_savings_balance = 0; | |
3083 | ||
dd41f596 | 3084 | /* |
5c45bf27 SS |
3085 | * If a group is already running at full capacity or idle, |
3086 | * don't include that group in power savings calculations | |
dd41f596 IM |
3087 | */ |
3088 | if (!power_savings_balance || sum_nr_running >= group_capacity | |
5c45bf27 | 3089 | || !sum_nr_running) |
dd41f596 | 3090 | goto group_next; |
5c45bf27 | 3091 | |
dd41f596 | 3092 | /* |
5c45bf27 | 3093 | * Calculate the group which has the least non-idle load. |
dd41f596 IM |
3094 | * This is the group from where we need to pick up the load |
3095 | * for saving power | |
3096 | */ | |
3097 | if ((sum_nr_running < min_nr_running) || | |
3098 | (sum_nr_running == min_nr_running && | |
5c45bf27 SS |
3099 | first_cpu(group->cpumask) < |
3100 | first_cpu(group_min->cpumask))) { | |
dd41f596 IM |
3101 | group_min = group; |
3102 | min_nr_running = sum_nr_running; | |
5c45bf27 SS |
3103 | min_load_per_task = sum_weighted_load / |
3104 | sum_nr_running; | |
dd41f596 | 3105 | } |
5c45bf27 | 3106 | |
dd41f596 | 3107 | /* |
5c45bf27 | 3108 | * Calculate the group which is almost near its |
dd41f596 IM |
3109 | * capacity but still has some space to pick up some load |
3110 | * from other group and save more power | |
3111 | */ | |
3112 | if (sum_nr_running <= group_capacity - 1) { | |
3113 | if (sum_nr_running > leader_nr_running || | |
3114 | (sum_nr_running == leader_nr_running && | |
3115 | first_cpu(group->cpumask) > | |
3116 | first_cpu(group_leader->cpumask))) { | |
3117 | group_leader = group; | |
3118 | leader_nr_running = sum_nr_running; | |
3119 | } | |
48f24c4d | 3120 | } |
5c45bf27 SS |
3121 | group_next: |
3122 | #endif | |
1da177e4 LT |
3123 | group = group->next; |
3124 | } while (group != sd->groups); | |
3125 | ||
2dd73a4f | 3126 | if (!busiest || this_load >= max_load || busiest_nr_running == 0) |
1da177e4 LT |
3127 | goto out_balanced; |
3128 | ||
3129 | avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr; | |
3130 | ||
3131 | if (this_load >= avg_load || | |
3132 | 100*max_load <= sd->imbalance_pct*this_load) | |
3133 | goto out_balanced; | |
3134 | ||
2dd73a4f | 3135 | busiest_load_per_task /= busiest_nr_running; |
908a7c1b KC |
3136 | if (group_imb) |
3137 | busiest_load_per_task = min(busiest_load_per_task, avg_load); | |
3138 | ||
1da177e4 LT |
3139 | /* |
3140 | * We're trying to get all the cpus to the average_load, so we don't | |
3141 | * want to push ourselves above the average load, nor do we wish to | |
3142 | * reduce the max loaded cpu below the average load, as either of these | |
3143 | * actions would just result in more rebalancing later, and ping-pong | |
3144 | * tasks around. Thus we look for the minimum possible imbalance. | |
3145 | * Negative imbalances (*we* are more loaded than anyone else) will | |
3146 | * be counted as no imbalance for these purposes -- we can't fix that | |
41a2d6cf | 3147 | * by pulling tasks to us. Be careful of negative numbers as they'll |
1da177e4 LT |
3148 | * appear as very large values with unsigned longs. |
3149 | */ | |
2dd73a4f PW |
3150 | if (max_load <= busiest_load_per_task) |
3151 | goto out_balanced; | |
3152 | ||
3153 | /* | |
3154 | * In the presence of smp nice balancing, certain scenarios can have | |
3155 | * max load less than avg load(as we skip the groups at or below | |
3156 | * its cpu_power, while calculating max_load..) | |
3157 | */ | |
3158 | if (max_load < avg_load) { | |
3159 | *imbalance = 0; | |
3160 | goto small_imbalance; | |
3161 | } | |
0c117f1b SS |
3162 | |
3163 | /* Don't want to pull so many tasks that a group would go idle */ | |
2dd73a4f | 3164 | max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); |
0c117f1b | 3165 | |
1da177e4 | 3166 | /* How much load to actually move to equalise the imbalance */ |
5517d86b ED |
3167 | *imbalance = min(max_pull * busiest->__cpu_power, |
3168 | (avg_load - this_load) * this->__cpu_power) | |
1da177e4 LT |
3169 | / SCHED_LOAD_SCALE; |
3170 | ||
2dd73a4f PW |
3171 | /* |
3172 | * if *imbalance is less than the average load per runnable task | |
3173 | * there is no gaurantee that any tasks will be moved so we'll have | |
3174 | * a think about bumping its value to force at least one task to be | |
3175 | * moved | |
3176 | */ | |
7fd0d2dd | 3177 | if (*imbalance < busiest_load_per_task) { |
48f24c4d | 3178 | unsigned long tmp, pwr_now, pwr_move; |
2dd73a4f PW |
3179 | unsigned int imbn; |
3180 | ||
3181 | small_imbalance: | |
3182 | pwr_move = pwr_now = 0; | |
3183 | imbn = 2; | |
3184 | if (this_nr_running) { | |
3185 | this_load_per_task /= this_nr_running; | |
3186 | if (busiest_load_per_task > this_load_per_task) | |
3187 | imbn = 1; | |
3188 | } else | |
3189 | this_load_per_task = SCHED_LOAD_SCALE; | |
1da177e4 | 3190 | |
dd41f596 IM |
3191 | if (max_load - this_load + SCHED_LOAD_SCALE_FUZZ >= |
3192 | busiest_load_per_task * imbn) { | |
2dd73a4f | 3193 | *imbalance = busiest_load_per_task; |
1da177e4 LT |
3194 | return busiest; |
3195 | } | |
3196 | ||
3197 | /* | |
3198 | * OK, we don't have enough imbalance to justify moving tasks, | |
3199 | * however we may be able to increase total CPU power used by | |
3200 | * moving them. | |
3201 | */ | |
3202 | ||
5517d86b ED |
3203 | pwr_now += busiest->__cpu_power * |
3204 | min(busiest_load_per_task, max_load); | |
3205 | pwr_now += this->__cpu_power * | |
3206 | min(this_load_per_task, this_load); | |
1da177e4 LT |
3207 | pwr_now /= SCHED_LOAD_SCALE; |
3208 | ||
3209 | /* Amount of load we'd subtract */ | |
5517d86b ED |
3210 | tmp = sg_div_cpu_power(busiest, |
3211 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
1da177e4 | 3212 | if (max_load > tmp) |
5517d86b | 3213 | pwr_move += busiest->__cpu_power * |
2dd73a4f | 3214 | min(busiest_load_per_task, max_load - tmp); |
1da177e4 LT |
3215 | |
3216 | /* Amount of load we'd add */ | |
5517d86b | 3217 | if (max_load * busiest->__cpu_power < |
33859f7f | 3218 | busiest_load_per_task * SCHED_LOAD_SCALE) |
5517d86b ED |
3219 | tmp = sg_div_cpu_power(this, |
3220 | max_load * busiest->__cpu_power); | |
1da177e4 | 3221 | else |
5517d86b ED |
3222 | tmp = sg_div_cpu_power(this, |
3223 | busiest_load_per_task * SCHED_LOAD_SCALE); | |
3224 | pwr_move += this->__cpu_power * | |
3225 | min(this_load_per_task, this_load + tmp); | |
1da177e4 LT |
3226 | pwr_move /= SCHED_LOAD_SCALE; |
3227 | ||
3228 | /* Move if we gain throughput */ | |
7fd0d2dd SS |
3229 | if (pwr_move > pwr_now) |
3230 | *imbalance = busiest_load_per_task; | |
1da177e4 LT |
3231 | } |
3232 | ||
1da177e4 LT |
3233 | return busiest; |
3234 | ||
3235 | out_balanced: | |
5c45bf27 | 3236 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
d15bcfdb | 3237 | if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE)) |
5c45bf27 | 3238 | goto ret; |
1da177e4 | 3239 | |
5c45bf27 SS |
3240 | if (this == group_leader && group_leader != group_min) { |
3241 | *imbalance = min_load_per_task; | |
3242 | return group_min; | |
3243 | } | |
5c45bf27 | 3244 | #endif |
783609c6 | 3245 | ret: |
1da177e4 LT |
3246 | *imbalance = 0; |
3247 | return NULL; | |
3248 | } | |
3249 | ||
3250 | /* | |
3251 | * find_busiest_queue - find the busiest runqueue among the cpus in group. | |
3252 | */ | |
70b97a7f | 3253 | static struct rq * |
d15bcfdb | 3254 | find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle, |
7c16ec58 | 3255 | unsigned long imbalance, const cpumask_t *cpus) |
1da177e4 | 3256 | { |
70b97a7f | 3257 | struct rq *busiest = NULL, *rq; |
2dd73a4f | 3258 | unsigned long max_load = 0; |
1da177e4 LT |
3259 | int i; |
3260 | ||
3261 | for_each_cpu_mask(i, group->cpumask) { | |
dd41f596 | 3262 | unsigned long wl; |
0a2966b4 CL |
3263 | |
3264 | if (!cpu_isset(i, *cpus)) | |
3265 | continue; | |
3266 | ||
48f24c4d | 3267 | rq = cpu_rq(i); |
dd41f596 | 3268 | wl = weighted_cpuload(i); |
2dd73a4f | 3269 | |
dd41f596 | 3270 | if (rq->nr_running == 1 && wl > imbalance) |
2dd73a4f | 3271 | continue; |
1da177e4 | 3272 | |
dd41f596 IM |
3273 | if (wl > max_load) { |
3274 | max_load = wl; | |
48f24c4d | 3275 | busiest = rq; |
1da177e4 LT |
3276 | } |
3277 | } | |
3278 | ||
3279 | return busiest; | |
3280 | } | |
3281 | ||
77391d71 NP |
3282 | /* |
3283 | * Max backoff if we encounter pinned tasks. Pretty arbitrary value, but | |
3284 | * so long as it is large enough. | |
3285 | */ | |
3286 | #define MAX_PINNED_INTERVAL 512 | |
3287 | ||
1da177e4 LT |
3288 | /* |
3289 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3290 | * tasks if there is an imbalance. | |
1da177e4 | 3291 | */ |
70b97a7f | 3292 | static int load_balance(int this_cpu, struct rq *this_rq, |
d15bcfdb | 3293 | struct sched_domain *sd, enum cpu_idle_type idle, |
7c16ec58 | 3294 | int *balance, cpumask_t *cpus) |
1da177e4 | 3295 | { |
43010659 | 3296 | int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; |
1da177e4 | 3297 | struct sched_group *group; |
1da177e4 | 3298 | unsigned long imbalance; |
70b97a7f | 3299 | struct rq *busiest; |
fe2eea3f | 3300 | unsigned long flags; |
5969fe06 | 3301 | |
7c16ec58 MT |
3302 | cpus_setall(*cpus); |
3303 | ||
89c4710e SS |
3304 | /* |
3305 | * When power savings policy is enabled for the parent domain, idle | |
3306 | * sibling can pick up load irrespective of busy siblings. In this case, | |
dd41f596 | 3307 | * let the state of idle sibling percolate up as CPU_IDLE, instead of |
d15bcfdb | 3308 | * portraying it as CPU_NOT_IDLE. |
89c4710e | 3309 | */ |
d15bcfdb | 3310 | if (idle != CPU_NOT_IDLE && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3311 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3312 | sd_idle = 1; |
1da177e4 | 3313 | |
2d72376b | 3314 | schedstat_inc(sd, lb_count[idle]); |
1da177e4 | 3315 | |
0a2966b4 CL |
3316 | redo: |
3317 | group = find_busiest_group(sd, this_cpu, &imbalance, idle, &sd_idle, | |
7c16ec58 | 3318 | cpus, balance); |
783609c6 | 3319 | |
06066714 | 3320 | if (*balance == 0) |
783609c6 | 3321 | goto out_balanced; |
783609c6 | 3322 | |
1da177e4 LT |
3323 | if (!group) { |
3324 | schedstat_inc(sd, lb_nobusyg[idle]); | |
3325 | goto out_balanced; | |
3326 | } | |
3327 | ||
7c16ec58 | 3328 | busiest = find_busiest_queue(group, idle, imbalance, cpus); |
1da177e4 LT |
3329 | if (!busiest) { |
3330 | schedstat_inc(sd, lb_nobusyq[idle]); | |
3331 | goto out_balanced; | |
3332 | } | |
3333 | ||
db935dbd | 3334 | BUG_ON(busiest == this_rq); |
1da177e4 LT |
3335 | |
3336 | schedstat_add(sd, lb_imbalance[idle], imbalance); | |
3337 | ||
43010659 | 3338 | ld_moved = 0; |
1da177e4 LT |
3339 | if (busiest->nr_running > 1) { |
3340 | /* | |
3341 | * Attempt to move tasks. If find_busiest_group has found | |
3342 | * an imbalance but busiest->nr_running <= 1, the group is | |
43010659 | 3343 | * still unbalanced. ld_moved simply stays zero, so it is |
1da177e4 LT |
3344 | * correctly treated as an imbalance. |
3345 | */ | |
fe2eea3f | 3346 | local_irq_save(flags); |
e17224bf | 3347 | double_rq_lock(this_rq, busiest); |
43010659 | 3348 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
48f24c4d | 3349 | imbalance, sd, idle, &all_pinned); |
e17224bf | 3350 | double_rq_unlock(this_rq, busiest); |
fe2eea3f | 3351 | local_irq_restore(flags); |
81026794 | 3352 | |
46cb4b7c SS |
3353 | /* |
3354 | * some other cpu did the load balance for us. | |
3355 | */ | |
43010659 | 3356 | if (ld_moved && this_cpu != smp_processor_id()) |
46cb4b7c SS |
3357 | resched_cpu(this_cpu); |
3358 | ||
81026794 | 3359 | /* All tasks on this runqueue were pinned by CPU affinity */ |
0a2966b4 | 3360 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3361 | cpu_clear(cpu_of(busiest), *cpus); |
3362 | if (!cpus_empty(*cpus)) | |
0a2966b4 | 3363 | goto redo; |
81026794 | 3364 | goto out_balanced; |
0a2966b4 | 3365 | } |
1da177e4 | 3366 | } |
81026794 | 3367 | |
43010659 | 3368 | if (!ld_moved) { |
1da177e4 LT |
3369 | schedstat_inc(sd, lb_failed[idle]); |
3370 | sd->nr_balance_failed++; | |
3371 | ||
3372 | if (unlikely(sd->nr_balance_failed > sd->cache_nice_tries+2)) { | |
1da177e4 | 3373 | |
fe2eea3f | 3374 | spin_lock_irqsave(&busiest->lock, flags); |
fa3b6ddc SS |
3375 | |
3376 | /* don't kick the migration_thread, if the curr | |
3377 | * task on busiest cpu can't be moved to this_cpu | |
3378 | */ | |
3379 | if (!cpu_isset(this_cpu, busiest->curr->cpus_allowed)) { | |
fe2eea3f | 3380 | spin_unlock_irqrestore(&busiest->lock, flags); |
fa3b6ddc SS |
3381 | all_pinned = 1; |
3382 | goto out_one_pinned; | |
3383 | } | |
3384 | ||
1da177e4 LT |
3385 | if (!busiest->active_balance) { |
3386 | busiest->active_balance = 1; | |
3387 | busiest->push_cpu = this_cpu; | |
81026794 | 3388 | active_balance = 1; |
1da177e4 | 3389 | } |
fe2eea3f | 3390 | spin_unlock_irqrestore(&busiest->lock, flags); |
81026794 | 3391 | if (active_balance) |
1da177e4 LT |
3392 | wake_up_process(busiest->migration_thread); |
3393 | ||
3394 | /* | |
3395 | * We've kicked active balancing, reset the failure | |
3396 | * counter. | |
3397 | */ | |
39507451 | 3398 | sd->nr_balance_failed = sd->cache_nice_tries+1; |
1da177e4 | 3399 | } |
81026794 | 3400 | } else |
1da177e4 LT |
3401 | sd->nr_balance_failed = 0; |
3402 | ||
81026794 | 3403 | if (likely(!active_balance)) { |
1da177e4 LT |
3404 | /* We were unbalanced, so reset the balancing interval */ |
3405 | sd->balance_interval = sd->min_interval; | |
81026794 NP |
3406 | } else { |
3407 | /* | |
3408 | * If we've begun active balancing, start to back off. This | |
3409 | * case may not be covered by the all_pinned logic if there | |
3410 | * is only 1 task on the busy runqueue (because we don't call | |
3411 | * move_tasks). | |
3412 | */ | |
3413 | if (sd->balance_interval < sd->max_interval) | |
3414 | sd->balance_interval *= 2; | |
1da177e4 LT |
3415 | } |
3416 | ||
43010659 | 3417 | if (!ld_moved && !sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3418 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
6363ca57 IM |
3419 | return -1; |
3420 | return ld_moved; | |
1da177e4 LT |
3421 | |
3422 | out_balanced: | |
1da177e4 LT |
3423 | schedstat_inc(sd, lb_balanced[idle]); |
3424 | ||
16cfb1c0 | 3425 | sd->nr_balance_failed = 0; |
fa3b6ddc SS |
3426 | |
3427 | out_one_pinned: | |
1da177e4 | 3428 | /* tune up the balancing interval */ |
77391d71 NP |
3429 | if ((all_pinned && sd->balance_interval < MAX_PINNED_INTERVAL) || |
3430 | (sd->balance_interval < sd->max_interval)) | |
1da177e4 LT |
3431 | sd->balance_interval *= 2; |
3432 | ||
48f24c4d | 3433 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3434 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
6363ca57 IM |
3435 | return -1; |
3436 | return 0; | |
1da177e4 LT |
3437 | } |
3438 | ||
3439 | /* | |
3440 | * Check this_cpu to ensure it is balanced within domain. Attempt to move | |
3441 | * tasks if there is an imbalance. | |
3442 | * | |
d15bcfdb | 3443 | * Called from schedule when this_rq is about to become idle (CPU_NEWLY_IDLE). |
1da177e4 LT |
3444 | * this_rq is locked. |
3445 | */ | |
48f24c4d | 3446 | static int |
7c16ec58 MT |
3447 | load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, |
3448 | cpumask_t *cpus) | |
1da177e4 LT |
3449 | { |
3450 | struct sched_group *group; | |
70b97a7f | 3451 | struct rq *busiest = NULL; |
1da177e4 | 3452 | unsigned long imbalance; |
43010659 | 3453 | int ld_moved = 0; |
5969fe06 | 3454 | int sd_idle = 0; |
969bb4e4 | 3455 | int all_pinned = 0; |
7c16ec58 MT |
3456 | |
3457 | cpus_setall(*cpus); | |
5969fe06 | 3458 | |
89c4710e SS |
3459 | /* |
3460 | * When power savings policy is enabled for the parent domain, idle | |
3461 | * sibling can pick up load irrespective of busy siblings. In this case, | |
3462 | * let the state of idle sibling percolate up as IDLE, instead of | |
d15bcfdb | 3463 | * portraying it as CPU_NOT_IDLE. |
89c4710e SS |
3464 | */ |
3465 | if (sd->flags & SD_SHARE_CPUPOWER && | |
3466 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 | 3467 | sd_idle = 1; |
1da177e4 | 3468 | |
2d72376b | 3469 | schedstat_inc(sd, lb_count[CPU_NEWLY_IDLE]); |
0a2966b4 | 3470 | redo: |
d15bcfdb | 3471 | group = find_busiest_group(sd, this_cpu, &imbalance, CPU_NEWLY_IDLE, |
7c16ec58 | 3472 | &sd_idle, cpus, NULL); |
1da177e4 | 3473 | if (!group) { |
d15bcfdb | 3474 | schedstat_inc(sd, lb_nobusyg[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3475 | goto out_balanced; |
1da177e4 LT |
3476 | } |
3477 | ||
7c16ec58 | 3478 | busiest = find_busiest_queue(group, CPU_NEWLY_IDLE, imbalance, cpus); |
db935dbd | 3479 | if (!busiest) { |
d15bcfdb | 3480 | schedstat_inc(sd, lb_nobusyq[CPU_NEWLY_IDLE]); |
16cfb1c0 | 3481 | goto out_balanced; |
1da177e4 LT |
3482 | } |
3483 | ||
db935dbd NP |
3484 | BUG_ON(busiest == this_rq); |
3485 | ||
d15bcfdb | 3486 | schedstat_add(sd, lb_imbalance[CPU_NEWLY_IDLE], imbalance); |
d6d5cfaf | 3487 | |
43010659 | 3488 | ld_moved = 0; |
d6d5cfaf NP |
3489 | if (busiest->nr_running > 1) { |
3490 | /* Attempt to move tasks */ | |
3491 | double_lock_balance(this_rq, busiest); | |
6e82a3be IM |
3492 | /* this_rq->clock is already updated */ |
3493 | update_rq_clock(busiest); | |
43010659 | 3494 | ld_moved = move_tasks(this_rq, this_cpu, busiest, |
969bb4e4 SS |
3495 | imbalance, sd, CPU_NEWLY_IDLE, |
3496 | &all_pinned); | |
d6d5cfaf | 3497 | spin_unlock(&busiest->lock); |
0a2966b4 | 3498 | |
969bb4e4 | 3499 | if (unlikely(all_pinned)) { |
7c16ec58 MT |
3500 | cpu_clear(cpu_of(busiest), *cpus); |
3501 | if (!cpus_empty(*cpus)) | |
0a2966b4 CL |
3502 | goto redo; |
3503 | } | |
d6d5cfaf NP |
3504 | } |
3505 | ||
43010659 | 3506 | if (!ld_moved) { |
d15bcfdb | 3507 | schedstat_inc(sd, lb_failed[CPU_NEWLY_IDLE]); |
89c4710e SS |
3508 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
3509 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) | |
5969fe06 NP |
3510 | return -1; |
3511 | } else | |
16cfb1c0 | 3512 | sd->nr_balance_failed = 0; |
1da177e4 | 3513 | |
43010659 | 3514 | return ld_moved; |
16cfb1c0 NP |
3515 | |
3516 | out_balanced: | |
d15bcfdb | 3517 | schedstat_inc(sd, lb_balanced[CPU_NEWLY_IDLE]); |
48f24c4d | 3518 | if (!sd_idle && sd->flags & SD_SHARE_CPUPOWER && |
89c4710e | 3519 | !test_sd_parent(sd, SD_POWERSAVINGS_BALANCE)) |
5969fe06 | 3520 | return -1; |
16cfb1c0 | 3521 | sd->nr_balance_failed = 0; |
48f24c4d | 3522 | |
16cfb1c0 | 3523 | return 0; |
1da177e4 LT |
3524 | } |
3525 | ||
3526 | /* | |
3527 | * idle_balance is called by schedule() if this_cpu is about to become | |
3528 | * idle. Attempts to pull tasks from other CPUs. | |
3529 | */ | |
70b97a7f | 3530 | static void idle_balance(int this_cpu, struct rq *this_rq) |
1da177e4 LT |
3531 | { |
3532 | struct sched_domain *sd; | |
dd41f596 IM |
3533 | int pulled_task = -1; |
3534 | unsigned long next_balance = jiffies + HZ; | |
7c16ec58 | 3535 | cpumask_t tmpmask; |
1da177e4 LT |
3536 | |
3537 | for_each_domain(this_cpu, sd) { | |
92c4ca5c CL |
3538 | unsigned long interval; |
3539 | ||
3540 | if (!(sd->flags & SD_LOAD_BALANCE)) | |
3541 | continue; | |
3542 | ||
3543 | if (sd->flags & SD_BALANCE_NEWIDLE) | |
48f24c4d | 3544 | /* If we've pulled tasks over stop searching: */ |
7c16ec58 MT |
3545 | pulled_task = load_balance_newidle(this_cpu, this_rq, |
3546 | sd, &tmpmask); | |
92c4ca5c CL |
3547 | |
3548 | interval = msecs_to_jiffies(sd->balance_interval); | |
3549 | if (time_after(next_balance, sd->last_balance + interval)) | |
3550 | next_balance = sd->last_balance + interval; | |
3551 | if (pulled_task) | |
3552 | break; | |
1da177e4 | 3553 | } |
dd41f596 | 3554 | if (pulled_task || time_after(jiffies, this_rq->next_balance)) { |
1bd77f2d CL |
3555 | /* |
3556 | * We are going idle. next_balance may be set based on | |
3557 | * a busy processor. So reset next_balance. | |
3558 | */ | |
3559 | this_rq->next_balance = next_balance; | |
dd41f596 | 3560 | } |
1da177e4 LT |
3561 | } |
3562 | ||
3563 | /* | |
3564 | * active_load_balance is run by migration threads. It pushes running tasks | |
3565 | * off the busiest CPU onto idle CPUs. It requires at least 1 task to be | |
3566 | * running on each physical CPU where possible, and avoids physical / | |
3567 | * logical imbalances. | |
3568 | * | |
3569 | * Called with busiest_rq locked. | |
3570 | */ | |
70b97a7f | 3571 | static void active_load_balance(struct rq *busiest_rq, int busiest_cpu) |
1da177e4 | 3572 | { |
39507451 | 3573 | int target_cpu = busiest_rq->push_cpu; |
70b97a7f IM |
3574 | struct sched_domain *sd; |
3575 | struct rq *target_rq; | |
39507451 | 3576 | |
48f24c4d | 3577 | /* Is there any task to move? */ |
39507451 | 3578 | if (busiest_rq->nr_running <= 1) |
39507451 NP |
3579 | return; |
3580 | ||
3581 | target_rq = cpu_rq(target_cpu); | |
1da177e4 LT |
3582 | |
3583 | /* | |
39507451 | 3584 | * This condition is "impossible", if it occurs |
41a2d6cf | 3585 | * we need to fix it. Originally reported by |
39507451 | 3586 | * Bjorn Helgaas on a 128-cpu setup. |
1da177e4 | 3587 | */ |
39507451 | 3588 | BUG_ON(busiest_rq == target_rq); |
1da177e4 | 3589 | |
39507451 NP |
3590 | /* move a task from busiest_rq to target_rq */ |
3591 | double_lock_balance(busiest_rq, target_rq); | |
6e82a3be IM |
3592 | update_rq_clock(busiest_rq); |
3593 | update_rq_clock(target_rq); | |
39507451 NP |
3594 | |
3595 | /* Search for an sd spanning us and the target CPU. */ | |
c96d145e | 3596 | for_each_domain(target_cpu, sd) { |
39507451 | 3597 | if ((sd->flags & SD_LOAD_BALANCE) && |
48f24c4d | 3598 | cpu_isset(busiest_cpu, sd->span)) |
39507451 | 3599 | break; |
c96d145e | 3600 | } |
39507451 | 3601 | |
48f24c4d | 3602 | if (likely(sd)) { |
2d72376b | 3603 | schedstat_inc(sd, alb_count); |
39507451 | 3604 | |
43010659 PW |
3605 | if (move_one_task(target_rq, target_cpu, busiest_rq, |
3606 | sd, CPU_IDLE)) | |
48f24c4d IM |
3607 | schedstat_inc(sd, alb_pushed); |
3608 | else | |
3609 | schedstat_inc(sd, alb_failed); | |
3610 | } | |
39507451 | 3611 | spin_unlock(&target_rq->lock); |
1da177e4 LT |
3612 | } |
3613 | ||
46cb4b7c SS |
3614 | #ifdef CONFIG_NO_HZ |
3615 | static struct { | |
3616 | atomic_t load_balancer; | |
41a2d6cf | 3617 | cpumask_t cpu_mask; |
46cb4b7c SS |
3618 | } nohz ____cacheline_aligned = { |
3619 | .load_balancer = ATOMIC_INIT(-1), | |
3620 | .cpu_mask = CPU_MASK_NONE, | |
3621 | }; | |
3622 | ||
7835b98b | 3623 | /* |
46cb4b7c SS |
3624 | * This routine will try to nominate the ilb (idle load balancing) |
3625 | * owner among the cpus whose ticks are stopped. ilb owner will do the idle | |
3626 | * load balancing on behalf of all those cpus. If all the cpus in the system | |
3627 | * go into this tickless mode, then there will be no ilb owner (as there is | |
3628 | * no need for one) and all the cpus will sleep till the next wakeup event | |
3629 | * arrives... | |
3630 | * | |
3631 | * For the ilb owner, tick is not stopped. And this tick will be used | |
3632 | * for idle load balancing. ilb owner will still be part of | |
3633 | * nohz.cpu_mask.. | |
7835b98b | 3634 | * |
46cb4b7c SS |
3635 | * While stopping the tick, this cpu will become the ilb owner if there |
3636 | * is no other owner. And will be the owner till that cpu becomes busy | |
3637 | * or if all cpus in the system stop their ticks at which point | |
3638 | * there is no need for ilb owner. | |
3639 | * | |
3640 | * When the ilb owner becomes busy, it nominates another owner, during the | |
3641 | * next busy scheduler_tick() | |
3642 | */ | |
3643 | int select_nohz_load_balancer(int stop_tick) | |
3644 | { | |
3645 | int cpu = smp_processor_id(); | |
3646 | ||
3647 | if (stop_tick) { | |
3648 | cpu_set(cpu, nohz.cpu_mask); | |
3649 | cpu_rq(cpu)->in_nohz_recently = 1; | |
3650 | ||
3651 | /* | |
3652 | * If we are going offline and still the leader, give up! | |
3653 | */ | |
3654 | if (cpu_is_offline(cpu) && | |
3655 | atomic_read(&nohz.load_balancer) == cpu) { | |
3656 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3657 | BUG(); | |
3658 | return 0; | |
3659 | } | |
3660 | ||
3661 | /* time for ilb owner also to sleep */ | |
3662 | if (cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3663 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3664 | atomic_set(&nohz.load_balancer, -1); | |
3665 | return 0; | |
3666 | } | |
3667 | ||
3668 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3669 | /* make me the ilb owner */ | |
3670 | if (atomic_cmpxchg(&nohz.load_balancer, -1, cpu) == -1) | |
3671 | return 1; | |
3672 | } else if (atomic_read(&nohz.load_balancer) == cpu) | |
3673 | return 1; | |
3674 | } else { | |
3675 | if (!cpu_isset(cpu, nohz.cpu_mask)) | |
3676 | return 0; | |
3677 | ||
3678 | cpu_clear(cpu, nohz.cpu_mask); | |
3679 | ||
3680 | if (atomic_read(&nohz.load_balancer) == cpu) | |
3681 | if (atomic_cmpxchg(&nohz.load_balancer, cpu, -1) != cpu) | |
3682 | BUG(); | |
3683 | } | |
3684 | return 0; | |
3685 | } | |
3686 | #endif | |
3687 | ||
3688 | static DEFINE_SPINLOCK(balancing); | |
3689 | ||
3690 | /* | |
7835b98b CL |
3691 | * It checks each scheduling domain to see if it is due to be balanced, |
3692 | * and initiates a balancing operation if so. | |
3693 | * | |
3694 | * Balancing parameters are set up in arch_init_sched_domains. | |
3695 | */ | |
a9957449 | 3696 | static void rebalance_domains(int cpu, enum cpu_idle_type idle) |
7835b98b | 3697 | { |
46cb4b7c SS |
3698 | int balance = 1; |
3699 | struct rq *rq = cpu_rq(cpu); | |
7835b98b CL |
3700 | unsigned long interval; |
3701 | struct sched_domain *sd; | |
46cb4b7c | 3702 | /* Earliest time when we have to do rebalance again */ |
c9819f45 | 3703 | unsigned long next_balance = jiffies + 60*HZ; |
f549da84 | 3704 | int update_next_balance = 0; |
7c16ec58 | 3705 | cpumask_t tmp; |
1da177e4 | 3706 | |
46cb4b7c | 3707 | for_each_domain(cpu, sd) { |
1da177e4 LT |
3708 | if (!(sd->flags & SD_LOAD_BALANCE)) |
3709 | continue; | |
3710 | ||
3711 | interval = sd->balance_interval; | |
d15bcfdb | 3712 | if (idle != CPU_IDLE) |
1da177e4 LT |
3713 | interval *= sd->busy_factor; |
3714 | ||
3715 | /* scale ms to jiffies */ | |
3716 | interval = msecs_to_jiffies(interval); | |
3717 | if (unlikely(!interval)) | |
3718 | interval = 1; | |
dd41f596 IM |
3719 | if (interval > HZ*NR_CPUS/10) |
3720 | interval = HZ*NR_CPUS/10; | |
3721 | ||
1da177e4 | 3722 | |
08c183f3 CL |
3723 | if (sd->flags & SD_SERIALIZE) { |
3724 | if (!spin_trylock(&balancing)) | |
3725 | goto out; | |
3726 | } | |
3727 | ||
c9819f45 | 3728 | if (time_after_eq(jiffies, sd->last_balance + interval)) { |
7c16ec58 | 3729 | if (load_balance(cpu, rq, sd, idle, &balance, &tmp)) { |
fa3b6ddc SS |
3730 | /* |
3731 | * We've pulled tasks over so either we're no | |
5969fe06 NP |
3732 | * longer idle, or one of our SMT siblings is |
3733 | * not idle. | |
3734 | */ | |
d15bcfdb | 3735 | idle = CPU_NOT_IDLE; |
1da177e4 | 3736 | } |
1bd77f2d | 3737 | sd->last_balance = jiffies; |
1da177e4 | 3738 | } |
08c183f3 CL |
3739 | if (sd->flags & SD_SERIALIZE) |
3740 | spin_unlock(&balancing); | |
3741 | out: | |
f549da84 | 3742 | if (time_after(next_balance, sd->last_balance + interval)) { |
c9819f45 | 3743 | next_balance = sd->last_balance + interval; |
f549da84 SS |
3744 | update_next_balance = 1; |
3745 | } | |
783609c6 SS |
3746 | |
3747 | /* | |
3748 | * Stop the load balance at this level. There is another | |
3749 | * CPU in our sched group which is doing load balancing more | |
3750 | * actively. | |
3751 | */ | |
3752 | if (!balance) | |
3753 | break; | |
1da177e4 | 3754 | } |
f549da84 SS |
3755 | |
3756 | /* | |
3757 | * next_balance will be updated only when there is a need. | |
3758 | * When the cpu is attached to null domain for ex, it will not be | |
3759 | * updated. | |
3760 | */ | |
3761 | if (likely(update_next_balance)) | |
3762 | rq->next_balance = next_balance; | |
46cb4b7c SS |
3763 | } |
3764 | ||
3765 | /* | |
3766 | * run_rebalance_domains is triggered when needed from the scheduler tick. | |
3767 | * In CONFIG_NO_HZ case, the idle load balance owner will do the | |
3768 | * rebalancing for all the cpus for whom scheduler ticks are stopped. | |
3769 | */ | |
3770 | static void run_rebalance_domains(struct softirq_action *h) | |
3771 | { | |
dd41f596 IM |
3772 | int this_cpu = smp_processor_id(); |
3773 | struct rq *this_rq = cpu_rq(this_cpu); | |
3774 | enum cpu_idle_type idle = this_rq->idle_at_tick ? | |
3775 | CPU_IDLE : CPU_NOT_IDLE; | |
46cb4b7c | 3776 | |
dd41f596 | 3777 | rebalance_domains(this_cpu, idle); |
46cb4b7c SS |
3778 | |
3779 | #ifdef CONFIG_NO_HZ | |
3780 | /* | |
3781 | * If this cpu is the owner for idle load balancing, then do the | |
3782 | * balancing on behalf of the other idle cpus whose ticks are | |
3783 | * stopped. | |
3784 | */ | |
dd41f596 IM |
3785 | if (this_rq->idle_at_tick && |
3786 | atomic_read(&nohz.load_balancer) == this_cpu) { | |
46cb4b7c SS |
3787 | cpumask_t cpus = nohz.cpu_mask; |
3788 | struct rq *rq; | |
3789 | int balance_cpu; | |
3790 | ||
dd41f596 | 3791 | cpu_clear(this_cpu, cpus); |
46cb4b7c SS |
3792 | for_each_cpu_mask(balance_cpu, cpus) { |
3793 | /* | |
3794 | * If this cpu gets work to do, stop the load balancing | |
3795 | * work being done for other cpus. Next load | |
3796 | * balancing owner will pick it up. | |
3797 | */ | |
3798 | if (need_resched()) | |
3799 | break; | |
3800 | ||
de0cf899 | 3801 | rebalance_domains(balance_cpu, CPU_IDLE); |
46cb4b7c SS |
3802 | |
3803 | rq = cpu_rq(balance_cpu); | |
dd41f596 IM |
3804 | if (time_after(this_rq->next_balance, rq->next_balance)) |
3805 | this_rq->next_balance = rq->next_balance; | |
46cb4b7c SS |
3806 | } |
3807 | } | |
3808 | #endif | |
3809 | } | |
3810 | ||
3811 | /* | |
3812 | * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. | |
3813 | * | |
3814 | * In case of CONFIG_NO_HZ, this is the place where we nominate a new | |
3815 | * idle load balancing owner or decide to stop the periodic load balancing, | |
3816 | * if the whole system is idle. | |
3817 | */ | |
dd41f596 | 3818 | static inline void trigger_load_balance(struct rq *rq, int cpu) |
46cb4b7c | 3819 | { |
46cb4b7c SS |
3820 | #ifdef CONFIG_NO_HZ |
3821 | /* | |
3822 | * If we were in the nohz mode recently and busy at the current | |
3823 | * scheduler tick, then check if we need to nominate new idle | |
3824 | * load balancer. | |
3825 | */ | |
3826 | if (rq->in_nohz_recently && !rq->idle_at_tick) { | |
3827 | rq->in_nohz_recently = 0; | |
3828 | ||
3829 | if (atomic_read(&nohz.load_balancer) == cpu) { | |
3830 | cpu_clear(cpu, nohz.cpu_mask); | |
3831 | atomic_set(&nohz.load_balancer, -1); | |
3832 | } | |
3833 | ||
3834 | if (atomic_read(&nohz.load_balancer) == -1) { | |
3835 | /* | |
3836 | * simple selection for now: Nominate the | |
3837 | * first cpu in the nohz list to be the next | |
3838 | * ilb owner. | |
3839 | * | |
3840 | * TBD: Traverse the sched domains and nominate | |
3841 | * the nearest cpu in the nohz.cpu_mask. | |
3842 | */ | |
3843 | int ilb = first_cpu(nohz.cpu_mask); | |
3844 | ||
434d53b0 | 3845 | if (ilb < nr_cpu_ids) |
46cb4b7c SS |
3846 | resched_cpu(ilb); |
3847 | } | |
3848 | } | |
3849 | ||
3850 | /* | |
3851 | * If this cpu is idle and doing idle load balancing for all the | |
3852 | * cpus with ticks stopped, is it time for that to stop? | |
3853 | */ | |
3854 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) == cpu && | |
3855 | cpus_weight(nohz.cpu_mask) == num_online_cpus()) { | |
3856 | resched_cpu(cpu); | |
3857 | return; | |
3858 | } | |
3859 | ||
3860 | /* | |
3861 | * If this cpu is idle and the idle load balancing is done by | |
3862 | * someone else, then no need raise the SCHED_SOFTIRQ | |
3863 | */ | |
3864 | if (rq->idle_at_tick && atomic_read(&nohz.load_balancer) != cpu && | |
3865 | cpu_isset(cpu, nohz.cpu_mask)) | |
3866 | return; | |
3867 | #endif | |
3868 | if (time_after_eq(jiffies, rq->next_balance)) | |
3869 | raise_softirq(SCHED_SOFTIRQ); | |
1da177e4 | 3870 | } |
dd41f596 IM |
3871 | |
3872 | #else /* CONFIG_SMP */ | |
3873 | ||
1da177e4 LT |
3874 | /* |
3875 | * on UP we do not need to balance between CPUs: | |
3876 | */ | |
70b97a7f | 3877 | static inline void idle_balance(int cpu, struct rq *rq) |
1da177e4 LT |
3878 | { |
3879 | } | |
dd41f596 | 3880 | |
1da177e4 LT |
3881 | #endif |
3882 | ||
1da177e4 LT |
3883 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3884 | ||
3885 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3886 | ||
3887 | /* | |
41b86e9c IM |
3888 | * Return p->sum_exec_runtime plus any more ns on the sched_clock |
3889 | * that have not yet been banked in case the task is currently running. | |
1da177e4 | 3890 | */ |
41b86e9c | 3891 | unsigned long long task_sched_runtime(struct task_struct *p) |
1da177e4 | 3892 | { |
1da177e4 | 3893 | unsigned long flags; |
41b86e9c IM |
3894 | u64 ns, delta_exec; |
3895 | struct rq *rq; | |
48f24c4d | 3896 | |
41b86e9c IM |
3897 | rq = task_rq_lock(p, &flags); |
3898 | ns = p->se.sum_exec_runtime; | |
051a1d1a | 3899 | if (task_current(rq, p)) { |
a8e504d2 IM |
3900 | update_rq_clock(rq); |
3901 | delta_exec = rq->clock - p->se.exec_start; | |
41b86e9c IM |
3902 | if ((s64)delta_exec > 0) |
3903 | ns += delta_exec; | |
3904 | } | |
3905 | task_rq_unlock(rq, &flags); | |
48f24c4d | 3906 | |
1da177e4 LT |
3907 | return ns; |
3908 | } | |
3909 | ||
1da177e4 LT |
3910 | /* |
3911 | * Account user cpu time to a process. | |
3912 | * @p: the process that the cpu time gets accounted to | |
1da177e4 LT |
3913 | * @cputime: the cpu time spent in user space since the last update |
3914 | */ | |
3915 | void account_user_time(struct task_struct *p, cputime_t cputime) | |
3916 | { | |
3917 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3918 | cputime64_t tmp; | |
3919 | ||
3920 | p->utime = cputime_add(p->utime, cputime); | |
3921 | ||
3922 | /* Add user time to cpustat. */ | |
3923 | tmp = cputime_to_cputime64(cputime); | |
3924 | if (TASK_NICE(p) > 0) | |
3925 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3926 | else | |
3927 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3928 | } | |
3929 | ||
94886b84 LV |
3930 | /* |
3931 | * Account guest cpu time to a process. | |
3932 | * @p: the process that the cpu time gets accounted to | |
3933 | * @cputime: the cpu time spent in virtual machine since the last update | |
3934 | */ | |
f7402e03 | 3935 | static void account_guest_time(struct task_struct *p, cputime_t cputime) |
94886b84 LV |
3936 | { |
3937 | cputime64_t tmp; | |
3938 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3939 | ||
3940 | tmp = cputime_to_cputime64(cputime); | |
3941 | ||
3942 | p->utime = cputime_add(p->utime, cputime); | |
3943 | p->gtime = cputime_add(p->gtime, cputime); | |
3944 | ||
3945 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3946 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3947 | } | |
3948 | ||
c66f08be MN |
3949 | /* |
3950 | * Account scaled user cpu time to a process. | |
3951 | * @p: the process that the cpu time gets accounted to | |
3952 | * @cputime: the cpu time spent in user space since the last update | |
3953 | */ | |
3954 | void account_user_time_scaled(struct task_struct *p, cputime_t cputime) | |
3955 | { | |
3956 | p->utimescaled = cputime_add(p->utimescaled, cputime); | |
3957 | } | |
3958 | ||
1da177e4 LT |
3959 | /* |
3960 | * Account system cpu time to a process. | |
3961 | * @p: the process that the cpu time gets accounted to | |
3962 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3963 | * @cputime: the cpu time spent in kernel space since the last update | |
3964 | */ | |
3965 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
3966 | cputime_t cputime) | |
3967 | { | |
3968 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70b97a7f | 3969 | struct rq *rq = this_rq(); |
1da177e4 LT |
3970 | cputime64_t tmp; |
3971 | ||
983ed7a6 HH |
3972 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
3973 | account_guest_time(p, cputime); | |
3974 | return; | |
3975 | } | |
94886b84 | 3976 | |
1da177e4 LT |
3977 | p->stime = cputime_add(p->stime, cputime); |
3978 | ||
3979 | /* Add system time to cpustat. */ | |
3980 | tmp = cputime_to_cputime64(cputime); | |
3981 | if (hardirq_count() - hardirq_offset) | |
3982 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3983 | else if (softirq_count()) | |
3984 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
cfb52856 | 3985 | else if (p != rq->idle) |
1da177e4 | 3986 | cpustat->system = cputime64_add(cpustat->system, tmp); |
cfb52856 | 3987 | else if (atomic_read(&rq->nr_iowait) > 0) |
1da177e4 LT |
3988 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); |
3989 | else | |
3990 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
3991 | /* Account for system time used */ | |
3992 | acct_update_integrals(p); | |
1da177e4 LT |
3993 | } |
3994 | ||
c66f08be MN |
3995 | /* |
3996 | * Account scaled system cpu time to a process. | |
3997 | * @p: the process that the cpu time gets accounted to | |
3998 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3999 | * @cputime: the cpu time spent in kernel space since the last update | |
4000 | */ | |
4001 | void account_system_time_scaled(struct task_struct *p, cputime_t cputime) | |
4002 | { | |
4003 | p->stimescaled = cputime_add(p->stimescaled, cputime); | |
4004 | } | |
4005 | ||
1da177e4 LT |
4006 | /* |
4007 | * Account for involuntary wait time. | |
4008 | * @p: the process from which the cpu time has been stolen | |
4009 | * @steal: the cpu time spent in involuntary wait | |
4010 | */ | |
4011 | void account_steal_time(struct task_struct *p, cputime_t steal) | |
4012 | { | |
4013 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
4014 | cputime64_t tmp = cputime_to_cputime64(steal); | |
70b97a7f | 4015 | struct rq *rq = this_rq(); |
1da177e4 LT |
4016 | |
4017 | if (p == rq->idle) { | |
4018 | p->stime = cputime_add(p->stime, steal); | |
4019 | if (atomic_read(&rq->nr_iowait) > 0) | |
4020 | cpustat->iowait = cputime64_add(cpustat->iowait, tmp); | |
4021 | else | |
4022 | cpustat->idle = cputime64_add(cpustat->idle, tmp); | |
cfb52856 | 4023 | } else |
1da177e4 LT |
4024 | cpustat->steal = cputime64_add(cpustat->steal, tmp); |
4025 | } | |
4026 | ||
7835b98b CL |
4027 | /* |
4028 | * This function gets called by the timer code, with HZ frequency. | |
4029 | * We call it with interrupts disabled. | |
4030 | * | |
4031 | * It also gets called by the fork code, when changing the parent's | |
4032 | * timeslices. | |
4033 | */ | |
4034 | void scheduler_tick(void) | |
4035 | { | |
7835b98b CL |
4036 | int cpu = smp_processor_id(); |
4037 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4038 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4039 | |
4040 | sched_clock_tick(); | |
dd41f596 IM |
4041 | |
4042 | spin_lock(&rq->lock); | |
3e51f33f | 4043 | update_rq_clock(rq); |
f1a438d8 | 4044 | update_cpu_load(rq); |
fa85ae24 | 4045 | curr->sched_class->task_tick(rq, curr, 0); |
dd41f596 | 4046 | spin_unlock(&rq->lock); |
7835b98b | 4047 | |
e418e1c2 | 4048 | #ifdef CONFIG_SMP |
dd41f596 IM |
4049 | rq->idle_at_tick = idle_cpu(cpu); |
4050 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4051 | #endif |
1da177e4 LT |
4052 | } |
4053 | ||
6cd8a4bb SR |
4054 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4055 | defined(CONFIG_PREEMPT_TRACER)) | |
4056 | ||
4057 | static inline unsigned long get_parent_ip(unsigned long addr) | |
4058 | { | |
4059 | if (in_lock_functions(addr)) { | |
4060 | addr = CALLER_ADDR2; | |
4061 | if (in_lock_functions(addr)) | |
4062 | addr = CALLER_ADDR3; | |
4063 | } | |
4064 | return addr; | |
4065 | } | |
1da177e4 | 4066 | |
43627582 | 4067 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4068 | { |
6cd8a4bb | 4069 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4070 | /* |
4071 | * Underflow? | |
4072 | */ | |
9a11b49a IM |
4073 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4074 | return; | |
6cd8a4bb | 4075 | #endif |
1da177e4 | 4076 | preempt_count() += val; |
6cd8a4bb | 4077 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4078 | /* |
4079 | * Spinlock count overflowing soon? | |
4080 | */ | |
33859f7f MOS |
4081 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4082 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4083 | #endif |
4084 | if (preempt_count() == val) | |
4085 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4086 | } |
4087 | EXPORT_SYMBOL(add_preempt_count); | |
4088 | ||
43627582 | 4089 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4090 | { |
6cd8a4bb | 4091 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4092 | /* |
4093 | * Underflow? | |
4094 | */ | |
9a11b49a IM |
4095 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
4096 | return; | |
1da177e4 LT |
4097 | /* |
4098 | * Is the spinlock portion underflowing? | |
4099 | */ | |
9a11b49a IM |
4100 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4101 | !(preempt_count() & PREEMPT_MASK))) | |
4102 | return; | |
6cd8a4bb | 4103 | #endif |
9a11b49a | 4104 | |
6cd8a4bb SR |
4105 | if (preempt_count() == val) |
4106 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4107 | preempt_count() -= val; |
4108 | } | |
4109 | EXPORT_SYMBOL(sub_preempt_count); | |
4110 | ||
4111 | #endif | |
4112 | ||
4113 | /* | |
dd41f596 | 4114 | * Print scheduling while atomic bug: |
1da177e4 | 4115 | */ |
dd41f596 | 4116 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4117 | { |
838225b4 SS |
4118 | struct pt_regs *regs = get_irq_regs(); |
4119 | ||
4120 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", | |
4121 | prev->comm, prev->pid, preempt_count()); | |
4122 | ||
dd41f596 IM |
4123 | debug_show_held_locks(prev); |
4124 | if (irqs_disabled()) | |
4125 | print_irqtrace_events(prev); | |
838225b4 SS |
4126 | |
4127 | if (regs) | |
4128 | show_regs(regs); | |
4129 | else | |
4130 | dump_stack(); | |
dd41f596 | 4131 | } |
1da177e4 | 4132 | |
dd41f596 IM |
4133 | /* |
4134 | * Various schedule()-time debugging checks and statistics: | |
4135 | */ | |
4136 | static inline void schedule_debug(struct task_struct *prev) | |
4137 | { | |
1da177e4 | 4138 | /* |
41a2d6cf | 4139 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4140 | * schedule() atomically, we ignore that path for now. |
4141 | * Otherwise, whine if we are scheduling when we should not be. | |
4142 | */ | |
3f33a7ce | 4143 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4144 | __schedule_bug(prev); |
4145 | ||
1da177e4 LT |
4146 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4147 | ||
2d72376b | 4148 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4149 | #ifdef CONFIG_SCHEDSTATS |
4150 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4151 | schedstat_inc(this_rq(), bkl_count); |
4152 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4153 | } |
4154 | #endif | |
dd41f596 IM |
4155 | } |
4156 | ||
4157 | /* | |
4158 | * Pick up the highest-prio task: | |
4159 | */ | |
4160 | static inline struct task_struct * | |
ff95f3df | 4161 | pick_next_task(struct rq *rq, struct task_struct *prev) |
dd41f596 | 4162 | { |
5522d5d5 | 4163 | const struct sched_class *class; |
dd41f596 | 4164 | struct task_struct *p; |
1da177e4 LT |
4165 | |
4166 | /* | |
dd41f596 IM |
4167 | * Optimization: we know that if all tasks are in |
4168 | * the fair class we can call that function directly: | |
1da177e4 | 4169 | */ |
dd41f596 | 4170 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4171 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4172 | if (likely(p)) |
4173 | return p; | |
1da177e4 LT |
4174 | } |
4175 | ||
dd41f596 IM |
4176 | class = sched_class_highest; |
4177 | for ( ; ; ) { | |
fb8d4724 | 4178 | p = class->pick_next_task(rq); |
dd41f596 IM |
4179 | if (p) |
4180 | return p; | |
4181 | /* | |
4182 | * Will never be NULL as the idle class always | |
4183 | * returns a non-NULL p: | |
4184 | */ | |
4185 | class = class->next; | |
4186 | } | |
4187 | } | |
1da177e4 | 4188 | |
dd41f596 IM |
4189 | /* |
4190 | * schedule() is the main scheduler function. | |
4191 | */ | |
4192 | asmlinkage void __sched schedule(void) | |
4193 | { | |
4194 | struct task_struct *prev, *next; | |
67ca7bde | 4195 | unsigned long *switch_count; |
dd41f596 | 4196 | struct rq *rq; |
dd41f596 IM |
4197 | int cpu; |
4198 | ||
4199 | need_resched: | |
4200 | preempt_disable(); | |
4201 | cpu = smp_processor_id(); | |
4202 | rq = cpu_rq(cpu); | |
4203 | rcu_qsctr_inc(cpu); | |
4204 | prev = rq->curr; | |
4205 | switch_count = &prev->nivcsw; | |
4206 | ||
4207 | release_kernel_lock(prev); | |
4208 | need_resched_nonpreemptible: | |
4209 | ||
4210 | schedule_debug(prev); | |
1da177e4 | 4211 | |
8f4d37ec PZ |
4212 | hrtick_clear(rq); |
4213 | ||
1e819950 IM |
4214 | /* |
4215 | * Do the rq-clock update outside the rq lock: | |
4216 | */ | |
4217 | local_irq_disable(); | |
3e51f33f | 4218 | update_rq_clock(rq); |
1e819950 IM |
4219 | spin_lock(&rq->lock); |
4220 | clear_tsk_need_resched(prev); | |
1da177e4 | 4221 | |
1da177e4 | 4222 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 4223 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 4224 | prev->state = TASK_RUNNING; |
16882c1e | 4225 | else |
2e1cb74a | 4226 | deactivate_task(rq, prev, 1); |
dd41f596 | 4227 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4228 | } |
4229 | ||
9a897c5a SR |
4230 | #ifdef CONFIG_SMP |
4231 | if (prev->sched_class->pre_schedule) | |
4232 | prev->sched_class->pre_schedule(rq, prev); | |
4233 | #endif | |
f65eda4f | 4234 | |
dd41f596 | 4235 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4236 | idle_balance(cpu, rq); |
1da177e4 | 4237 | |
31ee529c | 4238 | prev->sched_class->put_prev_task(rq, prev); |
ff95f3df | 4239 | next = pick_next_task(rq, prev); |
1da177e4 | 4240 | |
1da177e4 | 4241 | if (likely(prev != next)) { |
673a90a1 DS |
4242 | sched_info_switch(prev, next); |
4243 | ||
1da177e4 LT |
4244 | rq->nr_switches++; |
4245 | rq->curr = next; | |
4246 | ++*switch_count; | |
4247 | ||
dd41f596 | 4248 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
4249 | /* |
4250 | * the context switch might have flipped the stack from under | |
4251 | * us, hence refresh the local variables. | |
4252 | */ | |
4253 | cpu = smp_processor_id(); | |
4254 | rq = cpu_rq(cpu); | |
1da177e4 LT |
4255 | } else |
4256 | spin_unlock_irq(&rq->lock); | |
4257 | ||
8f4d37ec PZ |
4258 | hrtick_set(rq); |
4259 | ||
4260 | if (unlikely(reacquire_kernel_lock(current) < 0)) | |
1da177e4 | 4261 | goto need_resched_nonpreemptible; |
8f4d37ec | 4262 | |
1da177e4 LT |
4263 | preempt_enable_no_resched(); |
4264 | if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) | |
4265 | goto need_resched; | |
4266 | } | |
1da177e4 LT |
4267 | EXPORT_SYMBOL(schedule); |
4268 | ||
4269 | #ifdef CONFIG_PREEMPT | |
4270 | /* | |
2ed6e34f | 4271 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4272 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4273 | * occur there and call schedule directly. |
4274 | */ | |
4275 | asmlinkage void __sched preempt_schedule(void) | |
4276 | { | |
4277 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4278 | |
1da177e4 LT |
4279 | /* |
4280 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4281 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4282 | */ |
beed33a8 | 4283 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4284 | return; |
4285 | ||
3a5c359a AK |
4286 | do { |
4287 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 4288 | schedule(); |
3a5c359a | 4289 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4290 | |
3a5c359a AK |
4291 | /* |
4292 | * Check again in case we missed a preemption opportunity | |
4293 | * between schedule and now. | |
4294 | */ | |
4295 | barrier(); | |
4296 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 | 4297 | } |
1da177e4 LT |
4298 | EXPORT_SYMBOL(preempt_schedule); |
4299 | ||
4300 | /* | |
2ed6e34f | 4301 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4302 | * off of irq context. |
4303 | * Note, that this is called and return with irqs disabled. This will | |
4304 | * protect us against recursive calling from irq. | |
4305 | */ | |
4306 | asmlinkage void __sched preempt_schedule_irq(void) | |
4307 | { | |
4308 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4309 | |
2ed6e34f | 4310 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4311 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4312 | ||
3a5c359a AK |
4313 | do { |
4314 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4315 | local_irq_enable(); |
4316 | schedule(); | |
4317 | local_irq_disable(); | |
3a5c359a | 4318 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4319 | |
3a5c359a AK |
4320 | /* |
4321 | * Check again in case we missed a preemption opportunity | |
4322 | * between schedule and now. | |
4323 | */ | |
4324 | barrier(); | |
4325 | } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); | |
1da177e4 LT |
4326 | } |
4327 | ||
4328 | #endif /* CONFIG_PREEMPT */ | |
4329 | ||
95cdf3b7 IM |
4330 | int default_wake_function(wait_queue_t *curr, unsigned mode, int sync, |
4331 | void *key) | |
1da177e4 | 4332 | { |
48f24c4d | 4333 | return try_to_wake_up(curr->private, mode, sync); |
1da177e4 | 4334 | } |
1da177e4 LT |
4335 | EXPORT_SYMBOL(default_wake_function); |
4336 | ||
4337 | /* | |
41a2d6cf IM |
4338 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4339 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4340 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4341 | * | |
4342 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4343 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4344 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4345 | */ | |
4346 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, | |
4347 | int nr_exclusive, int sync, void *key) | |
4348 | { | |
2e45874c | 4349 | wait_queue_t *curr, *next; |
1da177e4 | 4350 | |
2e45874c | 4351 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4352 | unsigned flags = curr->flags; |
4353 | ||
1da177e4 | 4354 | if (curr->func(curr, mode, sync, key) && |
48f24c4d | 4355 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4356 | break; |
4357 | } | |
4358 | } | |
4359 | ||
4360 | /** | |
4361 | * __wake_up - wake up threads blocked on a waitqueue. | |
4362 | * @q: the waitqueue | |
4363 | * @mode: which threads | |
4364 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4365 | * @key: is directly passed to the wakeup function |
1da177e4 | 4366 | */ |
7ad5b3a5 | 4367 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4368 | int nr_exclusive, void *key) |
1da177e4 LT |
4369 | { |
4370 | unsigned long flags; | |
4371 | ||
4372 | spin_lock_irqsave(&q->lock, flags); | |
4373 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4374 | spin_unlock_irqrestore(&q->lock, flags); | |
4375 | } | |
1da177e4 LT |
4376 | EXPORT_SYMBOL(__wake_up); |
4377 | ||
4378 | /* | |
4379 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4380 | */ | |
7ad5b3a5 | 4381 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4382 | { |
4383 | __wake_up_common(q, mode, 1, 0, NULL); | |
4384 | } | |
4385 | ||
4386 | /** | |
67be2dd1 | 4387 | * __wake_up_sync - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4388 | * @q: the waitqueue |
4389 | * @mode: which threads | |
4390 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4391 | * | |
4392 | * The sync wakeup differs that the waker knows that it will schedule | |
4393 | * away soon, so while the target thread will be woken up, it will not | |
4394 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4395 | * with each other. This can prevent needless bouncing between CPUs. | |
4396 | * | |
4397 | * On UP it can prevent extra preemption. | |
4398 | */ | |
7ad5b3a5 | 4399 | void |
95cdf3b7 | 4400 | __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) |
1da177e4 LT |
4401 | { |
4402 | unsigned long flags; | |
4403 | int sync = 1; | |
4404 | ||
4405 | if (unlikely(!q)) | |
4406 | return; | |
4407 | ||
4408 | if (unlikely(!nr_exclusive)) | |
4409 | sync = 0; | |
4410 | ||
4411 | spin_lock_irqsave(&q->lock, flags); | |
4412 | __wake_up_common(q, mode, nr_exclusive, sync, NULL); | |
4413 | spin_unlock_irqrestore(&q->lock, flags); | |
4414 | } | |
4415 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ | |
4416 | ||
b15136e9 | 4417 | void complete(struct completion *x) |
1da177e4 LT |
4418 | { |
4419 | unsigned long flags; | |
4420 | ||
4421 | spin_lock_irqsave(&x->wait.lock, flags); | |
4422 | x->done++; | |
d9514f6c | 4423 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4424 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4425 | } | |
4426 | EXPORT_SYMBOL(complete); | |
4427 | ||
b15136e9 | 4428 | void complete_all(struct completion *x) |
1da177e4 LT |
4429 | { |
4430 | unsigned long flags; | |
4431 | ||
4432 | spin_lock_irqsave(&x->wait.lock, flags); | |
4433 | x->done += UINT_MAX/2; | |
d9514f6c | 4434 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4435 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4436 | } | |
4437 | EXPORT_SYMBOL(complete_all); | |
4438 | ||
8cbbe86d AK |
4439 | static inline long __sched |
4440 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4441 | { |
1da177e4 LT |
4442 | if (!x->done) { |
4443 | DECLARE_WAITQUEUE(wait, current); | |
4444 | ||
4445 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4446 | __add_wait_queue_tail(&x->wait, &wait); | |
4447 | do { | |
009e577e MW |
4448 | if ((state == TASK_INTERRUPTIBLE && |
4449 | signal_pending(current)) || | |
4450 | (state == TASK_KILLABLE && | |
4451 | fatal_signal_pending(current))) { | |
ea71a546 ON |
4452 | timeout = -ERESTARTSYS; |
4453 | break; | |
8cbbe86d AK |
4454 | } |
4455 | __set_current_state(state); | |
1da177e4 LT |
4456 | spin_unlock_irq(&x->wait.lock); |
4457 | timeout = schedule_timeout(timeout); | |
4458 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4459 | } while (!x->done && timeout); |
1da177e4 | 4460 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4461 | if (!x->done) |
4462 | return timeout; | |
1da177e4 LT |
4463 | } |
4464 | x->done--; | |
ea71a546 | 4465 | return timeout ?: 1; |
1da177e4 | 4466 | } |
1da177e4 | 4467 | |
8cbbe86d AK |
4468 | static long __sched |
4469 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4470 | { |
1da177e4 LT |
4471 | might_sleep(); |
4472 | ||
4473 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4474 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4475 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4476 | return timeout; |
4477 | } | |
1da177e4 | 4478 | |
b15136e9 | 4479 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4480 | { |
4481 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4482 | } |
8cbbe86d | 4483 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4484 | |
b15136e9 | 4485 | unsigned long __sched |
8cbbe86d | 4486 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4487 | { |
8cbbe86d | 4488 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4489 | } |
8cbbe86d | 4490 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4491 | |
8cbbe86d | 4492 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4493 | { |
51e97990 AK |
4494 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4495 | if (t == -ERESTARTSYS) | |
4496 | return t; | |
4497 | return 0; | |
0fec171c | 4498 | } |
8cbbe86d | 4499 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4500 | |
b15136e9 | 4501 | unsigned long __sched |
8cbbe86d AK |
4502 | wait_for_completion_interruptible_timeout(struct completion *x, |
4503 | unsigned long timeout) | |
0fec171c | 4504 | { |
8cbbe86d | 4505 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4506 | } |
8cbbe86d | 4507 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4508 | |
009e577e MW |
4509 | int __sched wait_for_completion_killable(struct completion *x) |
4510 | { | |
4511 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4512 | if (t == -ERESTARTSYS) | |
4513 | return t; | |
4514 | return 0; | |
4515 | } | |
4516 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4517 | ||
8cbbe86d AK |
4518 | static long __sched |
4519 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4520 | { |
0fec171c IM |
4521 | unsigned long flags; |
4522 | wait_queue_t wait; | |
4523 | ||
4524 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4525 | |
8cbbe86d | 4526 | __set_current_state(state); |
1da177e4 | 4527 | |
8cbbe86d AK |
4528 | spin_lock_irqsave(&q->lock, flags); |
4529 | __add_wait_queue(q, &wait); | |
4530 | spin_unlock(&q->lock); | |
4531 | timeout = schedule_timeout(timeout); | |
4532 | spin_lock_irq(&q->lock); | |
4533 | __remove_wait_queue(q, &wait); | |
4534 | spin_unlock_irqrestore(&q->lock, flags); | |
4535 | ||
4536 | return timeout; | |
4537 | } | |
4538 | ||
4539 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4540 | { | |
4541 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4542 | } |
1da177e4 LT |
4543 | EXPORT_SYMBOL(interruptible_sleep_on); |
4544 | ||
0fec171c | 4545 | long __sched |
95cdf3b7 | 4546 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4547 | { |
8cbbe86d | 4548 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4549 | } |
1da177e4 LT |
4550 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4551 | ||
0fec171c | 4552 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4553 | { |
8cbbe86d | 4554 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4555 | } |
1da177e4 LT |
4556 | EXPORT_SYMBOL(sleep_on); |
4557 | ||
0fec171c | 4558 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4559 | { |
8cbbe86d | 4560 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4561 | } |
1da177e4 LT |
4562 | EXPORT_SYMBOL(sleep_on_timeout); |
4563 | ||
b29739f9 IM |
4564 | #ifdef CONFIG_RT_MUTEXES |
4565 | ||
4566 | /* | |
4567 | * rt_mutex_setprio - set the current priority of a task | |
4568 | * @p: task | |
4569 | * @prio: prio value (kernel-internal form) | |
4570 | * | |
4571 | * This function changes the 'effective' priority of a task. It does | |
4572 | * not touch ->normal_prio like __setscheduler(). | |
4573 | * | |
4574 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4575 | */ | |
36c8b586 | 4576 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4577 | { |
4578 | unsigned long flags; | |
83b699ed | 4579 | int oldprio, on_rq, running; |
70b97a7f | 4580 | struct rq *rq; |
cb469845 | 4581 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4582 | |
4583 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4584 | ||
4585 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4586 | update_rq_clock(rq); |
b29739f9 | 4587 | |
d5f9f942 | 4588 | oldprio = p->prio; |
dd41f596 | 4589 | on_rq = p->se.on_rq; |
051a1d1a | 4590 | running = task_current(rq, p); |
0e1f3483 | 4591 | if (on_rq) |
69be72c1 | 4592 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4593 | if (running) |
4594 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4595 | |
4596 | if (rt_prio(prio)) | |
4597 | p->sched_class = &rt_sched_class; | |
4598 | else | |
4599 | p->sched_class = &fair_sched_class; | |
4600 | ||
b29739f9 IM |
4601 | p->prio = prio; |
4602 | ||
0e1f3483 HS |
4603 | if (running) |
4604 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4605 | if (on_rq) { |
8159f87e | 4606 | enqueue_task(rq, p, 0); |
cb469845 SR |
4607 | |
4608 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4609 | } |
4610 | task_rq_unlock(rq, &flags); | |
4611 | } | |
4612 | ||
4613 | #endif | |
4614 | ||
36c8b586 | 4615 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4616 | { |
dd41f596 | 4617 | int old_prio, delta, on_rq; |
1da177e4 | 4618 | unsigned long flags; |
70b97a7f | 4619 | struct rq *rq; |
1da177e4 LT |
4620 | |
4621 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4622 | return; | |
4623 | /* | |
4624 | * We have to be careful, if called from sys_setpriority(), | |
4625 | * the task might be in the middle of scheduling on another CPU. | |
4626 | */ | |
4627 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4628 | update_rq_clock(rq); |
1da177e4 LT |
4629 | /* |
4630 | * The RT priorities are set via sched_setscheduler(), but we still | |
4631 | * allow the 'normal' nice value to be set - but as expected | |
4632 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4633 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4634 | */ |
e05606d3 | 4635 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4636 | p->static_prio = NICE_TO_PRIO(nice); |
4637 | goto out_unlock; | |
4638 | } | |
dd41f596 | 4639 | on_rq = p->se.on_rq; |
6363ca57 | 4640 | if (on_rq) { |
69be72c1 | 4641 | dequeue_task(rq, p, 0); |
6363ca57 IM |
4642 | dec_load(rq, p); |
4643 | } | |
1da177e4 | 4644 | |
1da177e4 | 4645 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4646 | set_load_weight(p); |
b29739f9 IM |
4647 | old_prio = p->prio; |
4648 | p->prio = effective_prio(p); | |
4649 | delta = p->prio - old_prio; | |
1da177e4 | 4650 | |
dd41f596 | 4651 | if (on_rq) { |
8159f87e | 4652 | enqueue_task(rq, p, 0); |
6363ca57 | 4653 | inc_load(rq, p); |
1da177e4 | 4654 | /* |
d5f9f942 AM |
4655 | * If the task increased its priority or is running and |
4656 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4657 | */ |
d5f9f942 | 4658 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4659 | resched_task(rq->curr); |
4660 | } | |
4661 | out_unlock: | |
4662 | task_rq_unlock(rq, &flags); | |
4663 | } | |
1da177e4 LT |
4664 | EXPORT_SYMBOL(set_user_nice); |
4665 | ||
e43379f1 MM |
4666 | /* |
4667 | * can_nice - check if a task can reduce its nice value | |
4668 | * @p: task | |
4669 | * @nice: nice value | |
4670 | */ | |
36c8b586 | 4671 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4672 | { |
024f4747 MM |
4673 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4674 | int nice_rlim = 20 - nice; | |
48f24c4d | 4675 | |
e43379f1 MM |
4676 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4677 | capable(CAP_SYS_NICE)); | |
4678 | } | |
4679 | ||
1da177e4 LT |
4680 | #ifdef __ARCH_WANT_SYS_NICE |
4681 | ||
4682 | /* | |
4683 | * sys_nice - change the priority of the current process. | |
4684 | * @increment: priority increment | |
4685 | * | |
4686 | * sys_setpriority is a more generic, but much slower function that | |
4687 | * does similar things. | |
4688 | */ | |
4689 | asmlinkage long sys_nice(int increment) | |
4690 | { | |
48f24c4d | 4691 | long nice, retval; |
1da177e4 LT |
4692 | |
4693 | /* | |
4694 | * Setpriority might change our priority at the same moment. | |
4695 | * We don't have to worry. Conceptually one call occurs first | |
4696 | * and we have a single winner. | |
4697 | */ | |
e43379f1 MM |
4698 | if (increment < -40) |
4699 | increment = -40; | |
1da177e4 LT |
4700 | if (increment > 40) |
4701 | increment = 40; | |
4702 | ||
4703 | nice = PRIO_TO_NICE(current->static_prio) + increment; | |
4704 | if (nice < -20) | |
4705 | nice = -20; | |
4706 | if (nice > 19) | |
4707 | nice = 19; | |
4708 | ||
e43379f1 MM |
4709 | if (increment < 0 && !can_nice(current, nice)) |
4710 | return -EPERM; | |
4711 | ||
1da177e4 LT |
4712 | retval = security_task_setnice(current, nice); |
4713 | if (retval) | |
4714 | return retval; | |
4715 | ||
4716 | set_user_nice(current, nice); | |
4717 | return 0; | |
4718 | } | |
4719 | ||
4720 | #endif | |
4721 | ||
4722 | /** | |
4723 | * task_prio - return the priority value of a given task. | |
4724 | * @p: the task in question. | |
4725 | * | |
4726 | * This is the priority value as seen by users in /proc. | |
4727 | * RT tasks are offset by -200. Normal tasks are centered | |
4728 | * around 0, value goes from -16 to +15. | |
4729 | */ | |
36c8b586 | 4730 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4731 | { |
4732 | return p->prio - MAX_RT_PRIO; | |
4733 | } | |
4734 | ||
4735 | /** | |
4736 | * task_nice - return the nice value of a given task. | |
4737 | * @p: the task in question. | |
4738 | */ | |
36c8b586 | 4739 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4740 | { |
4741 | return TASK_NICE(p); | |
4742 | } | |
150d8bed | 4743 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4744 | |
4745 | /** | |
4746 | * idle_cpu - is a given cpu idle currently? | |
4747 | * @cpu: the processor in question. | |
4748 | */ | |
4749 | int idle_cpu(int cpu) | |
4750 | { | |
4751 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4752 | } | |
4753 | ||
1da177e4 LT |
4754 | /** |
4755 | * idle_task - return the idle task for a given cpu. | |
4756 | * @cpu: the processor in question. | |
4757 | */ | |
36c8b586 | 4758 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4759 | { |
4760 | return cpu_rq(cpu)->idle; | |
4761 | } | |
4762 | ||
4763 | /** | |
4764 | * find_process_by_pid - find a process with a matching PID value. | |
4765 | * @pid: the pid in question. | |
4766 | */ | |
a9957449 | 4767 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4768 | { |
228ebcbe | 4769 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4770 | } |
4771 | ||
4772 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4773 | static void |
4774 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4775 | { |
dd41f596 | 4776 | BUG_ON(p->se.on_rq); |
48f24c4d | 4777 | |
1da177e4 | 4778 | p->policy = policy; |
dd41f596 IM |
4779 | switch (p->policy) { |
4780 | case SCHED_NORMAL: | |
4781 | case SCHED_BATCH: | |
4782 | case SCHED_IDLE: | |
4783 | p->sched_class = &fair_sched_class; | |
4784 | break; | |
4785 | case SCHED_FIFO: | |
4786 | case SCHED_RR: | |
4787 | p->sched_class = &rt_sched_class; | |
4788 | break; | |
4789 | } | |
4790 | ||
1da177e4 | 4791 | p->rt_priority = prio; |
b29739f9 IM |
4792 | p->normal_prio = normal_prio(p); |
4793 | /* we are holding p->pi_lock already */ | |
4794 | p->prio = rt_mutex_getprio(p); | |
2dd73a4f | 4795 | set_load_weight(p); |
1da177e4 LT |
4796 | } |
4797 | ||
4798 | /** | |
72fd4a35 | 4799 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. |
1da177e4 LT |
4800 | * @p: the task in question. |
4801 | * @policy: new policy. | |
4802 | * @param: structure containing the new RT priority. | |
5fe1d75f | 4803 | * |
72fd4a35 | 4804 | * NOTE that the task may be already dead. |
1da177e4 | 4805 | */ |
95cdf3b7 IM |
4806 | int sched_setscheduler(struct task_struct *p, int policy, |
4807 | struct sched_param *param) | |
1da177e4 | 4808 | { |
83b699ed | 4809 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4810 | unsigned long flags; |
cb469845 | 4811 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4812 | struct rq *rq; |
1da177e4 | 4813 | |
66e5393a SR |
4814 | /* may grab non-irq protected spin_locks */ |
4815 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4816 | recheck: |
4817 | /* double check policy once rq lock held */ | |
4818 | if (policy < 0) | |
4819 | policy = oldpolicy = p->policy; | |
4820 | else if (policy != SCHED_FIFO && policy != SCHED_RR && | |
dd41f596 IM |
4821 | policy != SCHED_NORMAL && policy != SCHED_BATCH && |
4822 | policy != SCHED_IDLE) | |
b0a9499c | 4823 | return -EINVAL; |
1da177e4 LT |
4824 | /* |
4825 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4826 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4827 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4828 | */ |
4829 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4830 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4831 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4832 | return -EINVAL; |
e05606d3 | 4833 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4834 | return -EINVAL; |
4835 | ||
37e4ab3f OC |
4836 | /* |
4837 | * Allow unprivileged RT tasks to decrease priority: | |
4838 | */ | |
4839 | if (!capable(CAP_SYS_NICE)) { | |
e05606d3 | 4840 | if (rt_policy(policy)) { |
8dc3e909 | 4841 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4842 | |
4843 | if (!lock_task_sighand(p, &flags)) | |
4844 | return -ESRCH; | |
4845 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4846 | unlock_task_sighand(p, &flags); | |
4847 | ||
4848 | /* can't set/change the rt policy */ | |
4849 | if (policy != p->policy && !rlim_rtprio) | |
4850 | return -EPERM; | |
4851 | ||
4852 | /* can't increase priority */ | |
4853 | if (param->sched_priority > p->rt_priority && | |
4854 | param->sched_priority > rlim_rtprio) | |
4855 | return -EPERM; | |
4856 | } | |
dd41f596 IM |
4857 | /* |
4858 | * Like positive nice levels, dont allow tasks to | |
4859 | * move out of SCHED_IDLE either: | |
4860 | */ | |
4861 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4862 | return -EPERM; | |
5fe1d75f | 4863 | |
37e4ab3f OC |
4864 | /* can't change other user's priorities */ |
4865 | if ((current->euid != p->euid) && | |
4866 | (current->euid != p->uid)) | |
4867 | return -EPERM; | |
4868 | } | |
1da177e4 | 4869 | |
b68aa230 PZ |
4870 | #ifdef CONFIG_RT_GROUP_SCHED |
4871 | /* | |
4872 | * Do not allow realtime tasks into groups that have no runtime | |
4873 | * assigned. | |
4874 | */ | |
d0b27fa7 | 4875 | if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
4876 | return -EPERM; |
4877 | #endif | |
4878 | ||
1da177e4 LT |
4879 | retval = security_task_setscheduler(p, policy, param); |
4880 | if (retval) | |
4881 | return retval; | |
b29739f9 IM |
4882 | /* |
4883 | * make sure no PI-waiters arrive (or leave) while we are | |
4884 | * changing the priority of the task: | |
4885 | */ | |
4886 | spin_lock_irqsave(&p->pi_lock, flags); | |
1da177e4 LT |
4887 | /* |
4888 | * To be able to change p->policy safely, the apropriate | |
4889 | * runqueue lock must be held. | |
4890 | */ | |
b29739f9 | 4891 | rq = __task_rq_lock(p); |
1da177e4 LT |
4892 | /* recheck policy now with rq lock held */ |
4893 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4894 | policy = oldpolicy = -1; | |
b29739f9 IM |
4895 | __task_rq_unlock(rq); |
4896 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
1da177e4 LT |
4897 | goto recheck; |
4898 | } | |
2daa3577 | 4899 | update_rq_clock(rq); |
dd41f596 | 4900 | on_rq = p->se.on_rq; |
051a1d1a | 4901 | running = task_current(rq, p); |
0e1f3483 | 4902 | if (on_rq) |
2e1cb74a | 4903 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4904 | if (running) |
4905 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4906 | |
1da177e4 | 4907 | oldprio = p->prio; |
dd41f596 | 4908 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4909 | |
0e1f3483 HS |
4910 | if (running) |
4911 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4912 | if (on_rq) { |
4913 | activate_task(rq, p, 0); | |
cb469845 SR |
4914 | |
4915 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4916 | } |
b29739f9 IM |
4917 | __task_rq_unlock(rq); |
4918 | spin_unlock_irqrestore(&p->pi_lock, flags); | |
4919 | ||
95e02ca9 TG |
4920 | rt_mutex_adjust_pi(p); |
4921 | ||
1da177e4 LT |
4922 | return 0; |
4923 | } | |
4924 | EXPORT_SYMBOL_GPL(sched_setscheduler); | |
4925 | ||
95cdf3b7 IM |
4926 | static int |
4927 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4928 | { |
1da177e4 LT |
4929 | struct sched_param lparam; |
4930 | struct task_struct *p; | |
36c8b586 | 4931 | int retval; |
1da177e4 LT |
4932 | |
4933 | if (!param || pid < 0) | |
4934 | return -EINVAL; | |
4935 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4936 | return -EFAULT; | |
5fe1d75f ON |
4937 | |
4938 | rcu_read_lock(); | |
4939 | retval = -ESRCH; | |
1da177e4 | 4940 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4941 | if (p != NULL) |
4942 | retval = sched_setscheduler(p, policy, &lparam); | |
4943 | rcu_read_unlock(); | |
36c8b586 | 4944 | |
1da177e4 LT |
4945 | return retval; |
4946 | } | |
4947 | ||
4948 | /** | |
4949 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4950 | * @pid: the pid in question. | |
4951 | * @policy: new policy. | |
4952 | * @param: structure containing the new RT priority. | |
4953 | */ | |
41a2d6cf IM |
4954 | asmlinkage long |
4955 | sys_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4956 | { |
c21761f1 JB |
4957 | /* negative values for policy are not valid */ |
4958 | if (policy < 0) | |
4959 | return -EINVAL; | |
4960 | ||
1da177e4 LT |
4961 | return do_sched_setscheduler(pid, policy, param); |
4962 | } | |
4963 | ||
4964 | /** | |
4965 | * sys_sched_setparam - set/change the RT priority of a thread | |
4966 | * @pid: the pid in question. | |
4967 | * @param: structure containing the new RT priority. | |
4968 | */ | |
4969 | asmlinkage long sys_sched_setparam(pid_t pid, struct sched_param __user *param) | |
4970 | { | |
4971 | return do_sched_setscheduler(pid, -1, param); | |
4972 | } | |
4973 | ||
4974 | /** | |
4975 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4976 | * @pid: the pid in question. | |
4977 | */ | |
4978 | asmlinkage long sys_sched_getscheduler(pid_t pid) | |
4979 | { | |
36c8b586 | 4980 | struct task_struct *p; |
3a5c359a | 4981 | int retval; |
1da177e4 LT |
4982 | |
4983 | if (pid < 0) | |
3a5c359a | 4984 | return -EINVAL; |
1da177e4 LT |
4985 | |
4986 | retval = -ESRCH; | |
4987 | read_lock(&tasklist_lock); | |
4988 | p = find_process_by_pid(pid); | |
4989 | if (p) { | |
4990 | retval = security_task_getscheduler(p); | |
4991 | if (!retval) | |
4992 | retval = p->policy; | |
4993 | } | |
4994 | read_unlock(&tasklist_lock); | |
1da177e4 LT |
4995 | return retval; |
4996 | } | |
4997 | ||
4998 | /** | |
4999 | * sys_sched_getscheduler - get the RT priority of a thread | |
5000 | * @pid: the pid in question. | |
5001 | * @param: structure containing the RT priority. | |
5002 | */ | |
5003 | asmlinkage long sys_sched_getparam(pid_t pid, struct sched_param __user *param) | |
5004 | { | |
5005 | struct sched_param lp; | |
36c8b586 | 5006 | struct task_struct *p; |
3a5c359a | 5007 | int retval; |
1da177e4 LT |
5008 | |
5009 | if (!param || pid < 0) | |
3a5c359a | 5010 | return -EINVAL; |
1da177e4 LT |
5011 | |
5012 | read_lock(&tasklist_lock); | |
5013 | p = find_process_by_pid(pid); | |
5014 | retval = -ESRCH; | |
5015 | if (!p) | |
5016 | goto out_unlock; | |
5017 | ||
5018 | retval = security_task_getscheduler(p); | |
5019 | if (retval) | |
5020 | goto out_unlock; | |
5021 | ||
5022 | lp.sched_priority = p->rt_priority; | |
5023 | read_unlock(&tasklist_lock); | |
5024 | ||
5025 | /* | |
5026 | * This one might sleep, we cannot do it with a spinlock held ... | |
5027 | */ | |
5028 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5029 | ||
1da177e4 LT |
5030 | return retval; |
5031 | ||
5032 | out_unlock: | |
5033 | read_unlock(&tasklist_lock); | |
5034 | return retval; | |
5035 | } | |
5036 | ||
b53e921b | 5037 | long sched_setaffinity(pid_t pid, const cpumask_t *in_mask) |
1da177e4 | 5038 | { |
1da177e4 | 5039 | cpumask_t cpus_allowed; |
b53e921b | 5040 | cpumask_t new_mask = *in_mask; |
36c8b586 IM |
5041 | struct task_struct *p; |
5042 | int retval; | |
1da177e4 | 5043 | |
95402b38 | 5044 | get_online_cpus(); |
1da177e4 LT |
5045 | read_lock(&tasklist_lock); |
5046 | ||
5047 | p = find_process_by_pid(pid); | |
5048 | if (!p) { | |
5049 | read_unlock(&tasklist_lock); | |
95402b38 | 5050 | put_online_cpus(); |
1da177e4 LT |
5051 | return -ESRCH; |
5052 | } | |
5053 | ||
5054 | /* | |
5055 | * It is not safe to call set_cpus_allowed with the | |
41a2d6cf | 5056 | * tasklist_lock held. We will bump the task_struct's |
1da177e4 LT |
5057 | * usage count and then drop tasklist_lock. |
5058 | */ | |
5059 | get_task_struct(p); | |
5060 | read_unlock(&tasklist_lock); | |
5061 | ||
5062 | retval = -EPERM; | |
5063 | if ((current->euid != p->euid) && (current->euid != p->uid) && | |
5064 | !capable(CAP_SYS_NICE)) | |
5065 | goto out_unlock; | |
5066 | ||
e7834f8f DQ |
5067 | retval = security_task_setscheduler(p, 0, NULL); |
5068 | if (retval) | |
5069 | goto out_unlock; | |
5070 | ||
f9a86fcb | 5071 | cpuset_cpus_allowed(p, &cpus_allowed); |
1da177e4 | 5072 | cpus_and(new_mask, new_mask, cpus_allowed); |
8707d8b8 | 5073 | again: |
7c16ec58 | 5074 | retval = set_cpus_allowed_ptr(p, &new_mask); |
1da177e4 | 5075 | |
8707d8b8 | 5076 | if (!retval) { |
f9a86fcb | 5077 | cpuset_cpus_allowed(p, &cpus_allowed); |
8707d8b8 PM |
5078 | if (!cpus_subset(new_mask, cpus_allowed)) { |
5079 | /* | |
5080 | * We must have raced with a concurrent cpuset | |
5081 | * update. Just reset the cpus_allowed to the | |
5082 | * cpuset's cpus_allowed | |
5083 | */ | |
5084 | new_mask = cpus_allowed; | |
5085 | goto again; | |
5086 | } | |
5087 | } | |
1da177e4 LT |
5088 | out_unlock: |
5089 | put_task_struct(p); | |
95402b38 | 5090 | put_online_cpus(); |
1da177e4 LT |
5091 | return retval; |
5092 | } | |
5093 | ||
5094 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
5095 | cpumask_t *new_mask) | |
5096 | { | |
5097 | if (len < sizeof(cpumask_t)) { | |
5098 | memset(new_mask, 0, sizeof(cpumask_t)); | |
5099 | } else if (len > sizeof(cpumask_t)) { | |
5100 | len = sizeof(cpumask_t); | |
5101 | } | |
5102 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; | |
5103 | } | |
5104 | ||
5105 | /** | |
5106 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5107 | * @pid: pid of the process | |
5108 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5109 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5110 | */ | |
5111 | asmlinkage long sys_sched_setaffinity(pid_t pid, unsigned int len, | |
5112 | unsigned long __user *user_mask_ptr) | |
5113 | { | |
5114 | cpumask_t new_mask; | |
5115 | int retval; | |
5116 | ||
5117 | retval = get_user_cpu_mask(user_mask_ptr, len, &new_mask); | |
5118 | if (retval) | |
5119 | return retval; | |
5120 | ||
b53e921b | 5121 | return sched_setaffinity(pid, &new_mask); |
1da177e4 LT |
5122 | } |
5123 | ||
5124 | /* | |
5125 | * Represents all cpu's present in the system | |
5126 | * In systems capable of hotplug, this map could dynamically grow | |
5127 | * as new cpu's are detected in the system via any platform specific | |
5128 | * method, such as ACPI for e.g. | |
5129 | */ | |
5130 | ||
4cef0c61 | 5131 | cpumask_t cpu_present_map __read_mostly; |
1da177e4 LT |
5132 | EXPORT_SYMBOL(cpu_present_map); |
5133 | ||
5134 | #ifndef CONFIG_SMP | |
4cef0c61 | 5135 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 GB |
5136 | EXPORT_SYMBOL(cpu_online_map); |
5137 | ||
4cef0c61 | 5138 | cpumask_t cpu_possible_map __read_mostly = CPU_MASK_ALL; |
e16b38f7 | 5139 | EXPORT_SYMBOL(cpu_possible_map); |
1da177e4 LT |
5140 | #endif |
5141 | ||
5142 | long sched_getaffinity(pid_t pid, cpumask_t *mask) | |
5143 | { | |
36c8b586 | 5144 | struct task_struct *p; |
1da177e4 | 5145 | int retval; |
1da177e4 | 5146 | |
95402b38 | 5147 | get_online_cpus(); |
1da177e4 LT |
5148 | read_lock(&tasklist_lock); |
5149 | ||
5150 | retval = -ESRCH; | |
5151 | p = find_process_by_pid(pid); | |
5152 | if (!p) | |
5153 | goto out_unlock; | |
5154 | ||
e7834f8f DQ |
5155 | retval = security_task_getscheduler(p); |
5156 | if (retval) | |
5157 | goto out_unlock; | |
5158 | ||
2f7016d9 | 5159 | cpus_and(*mask, p->cpus_allowed, cpu_online_map); |
1da177e4 LT |
5160 | |
5161 | out_unlock: | |
5162 | read_unlock(&tasklist_lock); | |
95402b38 | 5163 | put_online_cpus(); |
1da177e4 | 5164 | |
9531b62f | 5165 | return retval; |
1da177e4 LT |
5166 | } |
5167 | ||
5168 | /** | |
5169 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5170 | * @pid: pid of the process | |
5171 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5172 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5173 | */ | |
5174 | asmlinkage long sys_sched_getaffinity(pid_t pid, unsigned int len, | |
5175 | unsigned long __user *user_mask_ptr) | |
5176 | { | |
5177 | int ret; | |
5178 | cpumask_t mask; | |
5179 | ||
5180 | if (len < sizeof(cpumask_t)) | |
5181 | return -EINVAL; | |
5182 | ||
5183 | ret = sched_getaffinity(pid, &mask); | |
5184 | if (ret < 0) | |
5185 | return ret; | |
5186 | ||
5187 | if (copy_to_user(user_mask_ptr, &mask, sizeof(cpumask_t))) | |
5188 | return -EFAULT; | |
5189 | ||
5190 | return sizeof(cpumask_t); | |
5191 | } | |
5192 | ||
5193 | /** | |
5194 | * sys_sched_yield - yield the current processor to other threads. | |
5195 | * | |
dd41f596 IM |
5196 | * This function yields the current CPU to other tasks. If there are no |
5197 | * other threads running on this CPU then this function will return. | |
1da177e4 LT |
5198 | */ |
5199 | asmlinkage long sys_sched_yield(void) | |
5200 | { | |
70b97a7f | 5201 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5202 | |
2d72376b | 5203 | schedstat_inc(rq, yld_count); |
4530d7ab | 5204 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5205 | |
5206 | /* | |
5207 | * Since we are going to call schedule() anyway, there's | |
5208 | * no need to preempt or enable interrupts: | |
5209 | */ | |
5210 | __release(rq->lock); | |
8a25d5de | 5211 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
1da177e4 LT |
5212 | _raw_spin_unlock(&rq->lock); |
5213 | preempt_enable_no_resched(); | |
5214 | ||
5215 | schedule(); | |
5216 | ||
5217 | return 0; | |
5218 | } | |
5219 | ||
e7b38404 | 5220 | static void __cond_resched(void) |
1da177e4 | 5221 | { |
8e0a43d8 IM |
5222 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP |
5223 | __might_sleep(__FILE__, __LINE__); | |
5224 | #endif | |
5bbcfd90 IM |
5225 | /* |
5226 | * The BKS might be reacquired before we have dropped | |
5227 | * PREEMPT_ACTIVE, which could trigger a second | |
5228 | * cond_resched() call. | |
5229 | */ | |
1da177e4 LT |
5230 | do { |
5231 | add_preempt_count(PREEMPT_ACTIVE); | |
5232 | schedule(); | |
5233 | sub_preempt_count(PREEMPT_ACTIVE); | |
5234 | } while (need_resched()); | |
5235 | } | |
5236 | ||
02b67cc3 | 5237 | int __sched _cond_resched(void) |
1da177e4 | 5238 | { |
9414232f IM |
5239 | if (need_resched() && !(preempt_count() & PREEMPT_ACTIVE) && |
5240 | system_state == SYSTEM_RUNNING) { | |
1da177e4 LT |
5241 | __cond_resched(); |
5242 | return 1; | |
5243 | } | |
5244 | return 0; | |
5245 | } | |
02b67cc3 | 5246 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5247 | |
5248 | /* | |
5249 | * cond_resched_lock() - if a reschedule is pending, drop the given lock, | |
5250 | * call schedule, and on return reacquire the lock. | |
5251 | * | |
41a2d6cf | 5252 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5253 | * operations here to prevent schedule() from being called twice (once via |
5254 | * spin_unlock(), once by hand). | |
5255 | */ | |
95cdf3b7 | 5256 | int cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5257 | { |
95c354fe | 5258 | int resched = need_resched() && system_state == SYSTEM_RUNNING; |
6df3cecb JK |
5259 | int ret = 0; |
5260 | ||
95c354fe | 5261 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5262 | spin_unlock(lock); |
95c354fe NP |
5263 | if (resched && need_resched()) |
5264 | __cond_resched(); | |
5265 | else | |
5266 | cpu_relax(); | |
6df3cecb | 5267 | ret = 1; |
1da177e4 | 5268 | spin_lock(lock); |
1da177e4 | 5269 | } |
6df3cecb | 5270 | return ret; |
1da177e4 | 5271 | } |
1da177e4 LT |
5272 | EXPORT_SYMBOL(cond_resched_lock); |
5273 | ||
5274 | int __sched cond_resched_softirq(void) | |
5275 | { | |
5276 | BUG_ON(!in_softirq()); | |
5277 | ||
9414232f | 5278 | if (need_resched() && system_state == SYSTEM_RUNNING) { |
98d82567 | 5279 | local_bh_enable(); |
1da177e4 LT |
5280 | __cond_resched(); |
5281 | local_bh_disable(); | |
5282 | return 1; | |
5283 | } | |
5284 | return 0; | |
5285 | } | |
1da177e4 LT |
5286 | EXPORT_SYMBOL(cond_resched_softirq); |
5287 | ||
1da177e4 LT |
5288 | /** |
5289 | * yield - yield the current processor to other threads. | |
5290 | * | |
72fd4a35 | 5291 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5292 | * thread runnable and calls sys_sched_yield(). |
5293 | */ | |
5294 | void __sched yield(void) | |
5295 | { | |
5296 | set_current_state(TASK_RUNNING); | |
5297 | sys_sched_yield(); | |
5298 | } | |
1da177e4 LT |
5299 | EXPORT_SYMBOL(yield); |
5300 | ||
5301 | /* | |
41a2d6cf | 5302 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 LT |
5303 | * that process accounting knows that this is a task in IO wait state. |
5304 | * | |
5305 | * But don't do that if it is a deliberate, throttling IO wait (this task | |
5306 | * has set its backing_dev_info: the queue against which it should throttle) | |
5307 | */ | |
5308 | void __sched io_schedule(void) | |
5309 | { | |
70b97a7f | 5310 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 | 5311 | |
0ff92245 | 5312 | delayacct_blkio_start(); |
1da177e4 LT |
5313 | atomic_inc(&rq->nr_iowait); |
5314 | schedule(); | |
5315 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5316 | delayacct_blkio_end(); |
1da177e4 | 5317 | } |
1da177e4 LT |
5318 | EXPORT_SYMBOL(io_schedule); |
5319 | ||
5320 | long __sched io_schedule_timeout(long timeout) | |
5321 | { | |
70b97a7f | 5322 | struct rq *rq = &__raw_get_cpu_var(runqueues); |
1da177e4 LT |
5323 | long ret; |
5324 | ||
0ff92245 | 5325 | delayacct_blkio_start(); |
1da177e4 LT |
5326 | atomic_inc(&rq->nr_iowait); |
5327 | ret = schedule_timeout(timeout); | |
5328 | atomic_dec(&rq->nr_iowait); | |
0ff92245 | 5329 | delayacct_blkio_end(); |
1da177e4 LT |
5330 | return ret; |
5331 | } | |
5332 | ||
5333 | /** | |
5334 | * sys_sched_get_priority_max - return maximum RT priority. | |
5335 | * @policy: scheduling class. | |
5336 | * | |
5337 | * this syscall returns the maximum rt_priority that can be used | |
5338 | * by a given scheduling class. | |
5339 | */ | |
5340 | asmlinkage long sys_sched_get_priority_max(int policy) | |
5341 | { | |
5342 | int ret = -EINVAL; | |
5343 | ||
5344 | switch (policy) { | |
5345 | case SCHED_FIFO: | |
5346 | case SCHED_RR: | |
5347 | ret = MAX_USER_RT_PRIO-1; | |
5348 | break; | |
5349 | case SCHED_NORMAL: | |
b0a9499c | 5350 | case SCHED_BATCH: |
dd41f596 | 5351 | case SCHED_IDLE: |
1da177e4 LT |
5352 | ret = 0; |
5353 | break; | |
5354 | } | |
5355 | return ret; | |
5356 | } | |
5357 | ||
5358 | /** | |
5359 | * sys_sched_get_priority_min - return minimum RT priority. | |
5360 | * @policy: scheduling class. | |
5361 | * | |
5362 | * this syscall returns the minimum rt_priority that can be used | |
5363 | * by a given scheduling class. | |
5364 | */ | |
5365 | asmlinkage long sys_sched_get_priority_min(int policy) | |
5366 | { | |
5367 | int ret = -EINVAL; | |
5368 | ||
5369 | switch (policy) { | |
5370 | case SCHED_FIFO: | |
5371 | case SCHED_RR: | |
5372 | ret = 1; | |
5373 | break; | |
5374 | case SCHED_NORMAL: | |
b0a9499c | 5375 | case SCHED_BATCH: |
dd41f596 | 5376 | case SCHED_IDLE: |
1da177e4 LT |
5377 | ret = 0; |
5378 | } | |
5379 | return ret; | |
5380 | } | |
5381 | ||
5382 | /** | |
5383 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5384 | * @pid: pid of the process. | |
5385 | * @interval: userspace pointer to the timeslice value. | |
5386 | * | |
5387 | * this syscall writes the default timeslice value of a given process | |
5388 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5389 | */ | |
5390 | asmlinkage | |
5391 | long sys_sched_rr_get_interval(pid_t pid, struct timespec __user *interval) | |
5392 | { | |
36c8b586 | 5393 | struct task_struct *p; |
a4ec24b4 | 5394 | unsigned int time_slice; |
3a5c359a | 5395 | int retval; |
1da177e4 | 5396 | struct timespec t; |
1da177e4 LT |
5397 | |
5398 | if (pid < 0) | |
3a5c359a | 5399 | return -EINVAL; |
1da177e4 LT |
5400 | |
5401 | retval = -ESRCH; | |
5402 | read_lock(&tasklist_lock); | |
5403 | p = find_process_by_pid(pid); | |
5404 | if (!p) | |
5405 | goto out_unlock; | |
5406 | ||
5407 | retval = security_task_getscheduler(p); | |
5408 | if (retval) | |
5409 | goto out_unlock; | |
5410 | ||
77034937 IM |
5411 | /* |
5412 | * Time slice is 0 for SCHED_FIFO tasks and for SCHED_OTHER | |
5413 | * tasks that are on an otherwise idle runqueue: | |
5414 | */ | |
5415 | time_slice = 0; | |
5416 | if (p->policy == SCHED_RR) { | |
a4ec24b4 | 5417 | time_slice = DEF_TIMESLICE; |
1868f958 | 5418 | } else if (p->policy != SCHED_FIFO) { |
a4ec24b4 DA |
5419 | struct sched_entity *se = &p->se; |
5420 | unsigned long flags; | |
5421 | struct rq *rq; | |
5422 | ||
5423 | rq = task_rq_lock(p, &flags); | |
77034937 IM |
5424 | if (rq->cfs.load.weight) |
5425 | time_slice = NS_TO_JIFFIES(sched_slice(&rq->cfs, se)); | |
a4ec24b4 DA |
5426 | task_rq_unlock(rq, &flags); |
5427 | } | |
1da177e4 | 5428 | read_unlock(&tasklist_lock); |
a4ec24b4 | 5429 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5430 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5431 | return retval; |
3a5c359a | 5432 | |
1da177e4 LT |
5433 | out_unlock: |
5434 | read_unlock(&tasklist_lock); | |
5435 | return retval; | |
5436 | } | |
5437 | ||
7c731e0a | 5438 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5439 | |
82a1fcb9 | 5440 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5441 | { |
1da177e4 | 5442 | unsigned long free = 0; |
36c8b586 | 5443 | unsigned state; |
1da177e4 | 5444 | |
1da177e4 | 5445 | state = p->state ? __ffs(p->state) + 1 : 0; |
cc4ea795 | 5446 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5447 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5448 | #if BITS_PER_LONG == 32 |
1da177e4 | 5449 | if (state == TASK_RUNNING) |
cc4ea795 | 5450 | printk(KERN_CONT " running "); |
1da177e4 | 5451 | else |
cc4ea795 | 5452 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5453 | #else |
5454 | if (state == TASK_RUNNING) | |
cc4ea795 | 5455 | printk(KERN_CONT " running task "); |
1da177e4 | 5456 | else |
cc4ea795 | 5457 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5458 | #endif |
5459 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
5460 | { | |
10ebffde | 5461 | unsigned long *n = end_of_stack(p); |
1da177e4 LT |
5462 | while (!*n) |
5463 | n++; | |
10ebffde | 5464 | free = (unsigned long)n - (unsigned long)end_of_stack(p); |
1da177e4 LT |
5465 | } |
5466 | #endif | |
ba25f9dc | 5467 | printk(KERN_CONT "%5lu %5d %6d\n", free, |
fcfd50af | 5468 | task_pid_nr(p), task_pid_nr(p->real_parent)); |
1da177e4 | 5469 | |
5fb5e6de | 5470 | show_stack(p, NULL); |
1da177e4 LT |
5471 | } |
5472 | ||
e59e2ae2 | 5473 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5474 | { |
36c8b586 | 5475 | struct task_struct *g, *p; |
1da177e4 | 5476 | |
4bd77321 IM |
5477 | #if BITS_PER_LONG == 32 |
5478 | printk(KERN_INFO | |
5479 | " task PC stack pid father\n"); | |
1da177e4 | 5480 | #else |
4bd77321 IM |
5481 | printk(KERN_INFO |
5482 | " task PC stack pid father\n"); | |
1da177e4 LT |
5483 | #endif |
5484 | read_lock(&tasklist_lock); | |
5485 | do_each_thread(g, p) { | |
5486 | /* | |
5487 | * reset the NMI-timeout, listing all files on a slow | |
5488 | * console might take alot of time: | |
5489 | */ | |
5490 | touch_nmi_watchdog(); | |
39bc89fd | 5491 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5492 | sched_show_task(p); |
1da177e4 LT |
5493 | } while_each_thread(g, p); |
5494 | ||
04c9167f JF |
5495 | touch_all_softlockup_watchdogs(); |
5496 | ||
dd41f596 IM |
5497 | #ifdef CONFIG_SCHED_DEBUG |
5498 | sysrq_sched_debug_show(); | |
5499 | #endif | |
1da177e4 | 5500 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5501 | /* |
5502 | * Only show locks if all tasks are dumped: | |
5503 | */ | |
5504 | if (state_filter == -1) | |
5505 | debug_show_all_locks(); | |
1da177e4 LT |
5506 | } |
5507 | ||
1df21055 IM |
5508 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5509 | { | |
dd41f596 | 5510 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5511 | } |
5512 | ||
f340c0d1 IM |
5513 | /** |
5514 | * init_idle - set up an idle thread for a given CPU | |
5515 | * @idle: task in question | |
5516 | * @cpu: cpu the idle task belongs to | |
5517 | * | |
5518 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5519 | * flag, to make booting more robust. | |
5520 | */ | |
5c1e1767 | 5521 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5522 | { |
70b97a7f | 5523 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5524 | unsigned long flags; |
5525 | ||
dd41f596 IM |
5526 | __sched_fork(idle); |
5527 | idle->se.exec_start = sched_clock(); | |
5528 | ||
b29739f9 | 5529 | idle->prio = idle->normal_prio = MAX_PRIO; |
1da177e4 | 5530 | idle->cpus_allowed = cpumask_of_cpu(cpu); |
dd41f596 | 5531 | __set_task_cpu(idle, cpu); |
1da177e4 LT |
5532 | |
5533 | spin_lock_irqsave(&rq->lock, flags); | |
5534 | rq->curr = rq->idle = idle; | |
4866cde0 NP |
5535 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5536 | idle->oncpu = 1; | |
5537 | #endif | |
1da177e4 LT |
5538 | spin_unlock_irqrestore(&rq->lock, flags); |
5539 | ||
5540 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5541 | #if defined(CONFIG_PREEMPT) |
5542 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5543 | #else | |
a1261f54 | 5544 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5545 | #endif |
dd41f596 IM |
5546 | /* |
5547 | * The idle tasks have their own, simple scheduling class: | |
5548 | */ | |
5549 | idle->sched_class = &idle_sched_class; | |
1da177e4 LT |
5550 | } |
5551 | ||
5552 | /* | |
5553 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5554 | * indicates which cpus entered this state. This is used | |
5555 | * in the rcu update to wait only for active cpus. For system | |
5556 | * which do not switch off the HZ timer nohz_cpu_mask should | |
5557 | * always be CPU_MASK_NONE. | |
5558 | */ | |
5559 | cpumask_t nohz_cpu_mask = CPU_MASK_NONE; | |
5560 | ||
19978ca6 IM |
5561 | /* |
5562 | * Increase the granularity value when there are more CPUs, | |
5563 | * because with more CPUs the 'effective latency' as visible | |
5564 | * to users decreases. But the relationship is not linear, | |
5565 | * so pick a second-best guess by going with the log2 of the | |
5566 | * number of CPUs. | |
5567 | * | |
5568 | * This idea comes from the SD scheduler of Con Kolivas: | |
5569 | */ | |
5570 | static inline void sched_init_granularity(void) | |
5571 | { | |
5572 | unsigned int factor = 1 + ilog2(num_online_cpus()); | |
5573 | const unsigned long limit = 200000000; | |
5574 | ||
5575 | sysctl_sched_min_granularity *= factor; | |
5576 | if (sysctl_sched_min_granularity > limit) | |
5577 | sysctl_sched_min_granularity = limit; | |
5578 | ||
5579 | sysctl_sched_latency *= factor; | |
5580 | if (sysctl_sched_latency > limit) | |
5581 | sysctl_sched_latency = limit; | |
5582 | ||
5583 | sysctl_sched_wakeup_granularity *= factor; | |
19978ca6 IM |
5584 | } |
5585 | ||
1da177e4 LT |
5586 | #ifdef CONFIG_SMP |
5587 | /* | |
5588 | * This is how migration works: | |
5589 | * | |
70b97a7f | 5590 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5591 | * runqueue and wake up that CPU's migration thread. |
5592 | * 2) we down() the locked semaphore => thread blocks. | |
5593 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5594 | * thread off the CPU) | |
5595 | * 4) it gets the migration request and checks whether the migrated | |
5596 | * task is still in the wrong runqueue. | |
5597 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5598 | * it and puts it into the right queue. | |
5599 | * 6) migration thread up()s the semaphore. | |
5600 | * 7) we wake up and the migration is done. | |
5601 | */ | |
5602 | ||
5603 | /* | |
5604 | * Change a given task's CPU affinity. Migrate the thread to a | |
5605 | * proper CPU and schedule it away if the CPU it's executing on | |
5606 | * is removed from the allowed bitmask. | |
5607 | * | |
5608 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5609 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5610 | * call is not atomic; no spinlocks may be held. |
5611 | */ | |
cd8ba7cd | 5612 | int set_cpus_allowed_ptr(struct task_struct *p, const cpumask_t *new_mask) |
1da177e4 | 5613 | { |
70b97a7f | 5614 | struct migration_req req; |
1da177e4 | 5615 | unsigned long flags; |
70b97a7f | 5616 | struct rq *rq; |
48f24c4d | 5617 | int ret = 0; |
1da177e4 LT |
5618 | |
5619 | rq = task_rq_lock(p, &flags); | |
cd8ba7cd | 5620 | if (!cpus_intersects(*new_mask, cpu_online_map)) { |
1da177e4 LT |
5621 | ret = -EINVAL; |
5622 | goto out; | |
5623 | } | |
5624 | ||
73fe6aae | 5625 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5626 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5627 | else { |
cd8ba7cd MT |
5628 | p->cpus_allowed = *new_mask; |
5629 | p->rt.nr_cpus_allowed = cpus_weight(*new_mask); | |
73fe6aae GH |
5630 | } |
5631 | ||
1da177e4 | 5632 | /* Can the task run on the task's current CPU? If so, we're done */ |
cd8ba7cd | 5633 | if (cpu_isset(task_cpu(p), *new_mask)) |
1da177e4 LT |
5634 | goto out; |
5635 | ||
cd8ba7cd | 5636 | if (migrate_task(p, any_online_cpu(*new_mask), &req)) { |
1da177e4 LT |
5637 | /* Need help from migration thread: drop lock and wait. */ |
5638 | task_rq_unlock(rq, &flags); | |
5639 | wake_up_process(rq->migration_thread); | |
5640 | wait_for_completion(&req.done); | |
5641 | tlb_migrate_finish(p->mm); | |
5642 | return 0; | |
5643 | } | |
5644 | out: | |
5645 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5646 | |
1da177e4 LT |
5647 | return ret; |
5648 | } | |
cd8ba7cd | 5649 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5650 | |
5651 | /* | |
41a2d6cf | 5652 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5653 | * this because either it can't run here any more (set_cpus_allowed() |
5654 | * away from this CPU, or CPU going down), or because we're | |
5655 | * attempting to rebalance this task on exec (sched_exec). | |
5656 | * | |
5657 | * So we race with normal scheduler movements, but that's OK, as long | |
5658 | * as the task is no longer on this CPU. | |
efc30814 KK |
5659 | * |
5660 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5661 | */ |
efc30814 | 5662 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5663 | { |
70b97a7f | 5664 | struct rq *rq_dest, *rq_src; |
dd41f596 | 5665 | int ret = 0, on_rq; |
1da177e4 LT |
5666 | |
5667 | if (unlikely(cpu_is_offline(dest_cpu))) | |
efc30814 | 5668 | return ret; |
1da177e4 LT |
5669 | |
5670 | rq_src = cpu_rq(src_cpu); | |
5671 | rq_dest = cpu_rq(dest_cpu); | |
5672 | ||
5673 | double_rq_lock(rq_src, rq_dest); | |
5674 | /* Already moved. */ | |
5675 | if (task_cpu(p) != src_cpu) | |
5676 | goto out; | |
5677 | /* Affinity changed (again). */ | |
5678 | if (!cpu_isset(dest_cpu, p->cpus_allowed)) | |
5679 | goto out; | |
5680 | ||
dd41f596 | 5681 | on_rq = p->se.on_rq; |
6e82a3be | 5682 | if (on_rq) |
2e1cb74a | 5683 | deactivate_task(rq_src, p, 0); |
6e82a3be | 5684 | |
1da177e4 | 5685 | set_task_cpu(p, dest_cpu); |
dd41f596 IM |
5686 | if (on_rq) { |
5687 | activate_task(rq_dest, p, 0); | |
5688 | check_preempt_curr(rq_dest, p); | |
1da177e4 | 5689 | } |
efc30814 | 5690 | ret = 1; |
1da177e4 LT |
5691 | out: |
5692 | double_rq_unlock(rq_src, rq_dest); | |
efc30814 | 5693 | return ret; |
1da177e4 LT |
5694 | } |
5695 | ||
5696 | /* | |
5697 | * migration_thread - this is a highprio system thread that performs | |
5698 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5699 | * another runqueue. | |
5700 | */ | |
95cdf3b7 | 5701 | static int migration_thread(void *data) |
1da177e4 | 5702 | { |
1da177e4 | 5703 | int cpu = (long)data; |
70b97a7f | 5704 | struct rq *rq; |
1da177e4 LT |
5705 | |
5706 | rq = cpu_rq(cpu); | |
5707 | BUG_ON(rq->migration_thread != current); | |
5708 | ||
5709 | set_current_state(TASK_INTERRUPTIBLE); | |
5710 | while (!kthread_should_stop()) { | |
70b97a7f | 5711 | struct migration_req *req; |
1da177e4 | 5712 | struct list_head *head; |
1da177e4 | 5713 | |
1da177e4 LT |
5714 | spin_lock_irq(&rq->lock); |
5715 | ||
5716 | if (cpu_is_offline(cpu)) { | |
5717 | spin_unlock_irq(&rq->lock); | |
5718 | goto wait_to_die; | |
5719 | } | |
5720 | ||
5721 | if (rq->active_balance) { | |
5722 | active_load_balance(rq, cpu); | |
5723 | rq->active_balance = 0; | |
5724 | } | |
5725 | ||
5726 | head = &rq->migration_queue; | |
5727 | ||
5728 | if (list_empty(head)) { | |
5729 | spin_unlock_irq(&rq->lock); | |
5730 | schedule(); | |
5731 | set_current_state(TASK_INTERRUPTIBLE); | |
5732 | continue; | |
5733 | } | |
70b97a7f | 5734 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5735 | list_del_init(head->next); |
5736 | ||
674311d5 NP |
5737 | spin_unlock(&rq->lock); |
5738 | __migrate_task(req->task, cpu, req->dest_cpu); | |
5739 | local_irq_enable(); | |
1da177e4 LT |
5740 | |
5741 | complete(&req->done); | |
5742 | } | |
5743 | __set_current_state(TASK_RUNNING); | |
5744 | return 0; | |
5745 | ||
5746 | wait_to_die: | |
5747 | /* Wait for kthread_stop */ | |
5748 | set_current_state(TASK_INTERRUPTIBLE); | |
5749 | while (!kthread_should_stop()) { | |
5750 | schedule(); | |
5751 | set_current_state(TASK_INTERRUPTIBLE); | |
5752 | } | |
5753 | __set_current_state(TASK_RUNNING); | |
5754 | return 0; | |
5755 | } | |
5756 | ||
5757 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5758 | |
5759 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5760 | { | |
5761 | int ret; | |
5762 | ||
5763 | local_irq_disable(); | |
5764 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5765 | local_irq_enable(); | |
5766 | return ret; | |
5767 | } | |
5768 | ||
054b9108 | 5769 | /* |
3a4fa0a2 | 5770 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 KK |
5771 | * NOTE: interrupts should be disabled by the caller |
5772 | */ | |
48f24c4d | 5773 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5774 | { |
efc30814 | 5775 | unsigned long flags; |
1da177e4 | 5776 | cpumask_t mask; |
70b97a7f IM |
5777 | struct rq *rq; |
5778 | int dest_cpu; | |
1da177e4 | 5779 | |
3a5c359a AK |
5780 | do { |
5781 | /* On same node? */ | |
5782 | mask = node_to_cpumask(cpu_to_node(dead_cpu)); | |
5783 | cpus_and(mask, mask, p->cpus_allowed); | |
5784 | dest_cpu = any_online_cpu(mask); | |
5785 | ||
5786 | /* On any allowed CPU? */ | |
434d53b0 | 5787 | if (dest_cpu >= nr_cpu_ids) |
3a5c359a AK |
5788 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5789 | ||
5790 | /* No more Mr. Nice Guy. */ | |
434d53b0 | 5791 | if (dest_cpu >= nr_cpu_ids) { |
f9a86fcb MT |
5792 | cpumask_t cpus_allowed; |
5793 | ||
5794 | cpuset_cpus_allowed_locked(p, &cpus_allowed); | |
470fd646 CW |
5795 | /* |
5796 | * Try to stay on the same cpuset, where the | |
5797 | * current cpuset may be a subset of all cpus. | |
5798 | * The cpuset_cpus_allowed_locked() variant of | |
41a2d6cf | 5799 | * cpuset_cpus_allowed() will not block. It must be |
470fd646 CW |
5800 | * called within calls to cpuset_lock/cpuset_unlock. |
5801 | */ | |
3a5c359a | 5802 | rq = task_rq_lock(p, &flags); |
470fd646 | 5803 | p->cpus_allowed = cpus_allowed; |
3a5c359a AK |
5804 | dest_cpu = any_online_cpu(p->cpus_allowed); |
5805 | task_rq_unlock(rq, &flags); | |
1da177e4 | 5806 | |
3a5c359a AK |
5807 | /* |
5808 | * Don't tell them about moving exiting tasks or | |
5809 | * kernel threads (both mm NULL), since they never | |
5810 | * leave kernel. | |
5811 | */ | |
41a2d6cf | 5812 | if (p->mm && printk_ratelimit()) { |
3a5c359a AK |
5813 | printk(KERN_INFO "process %d (%s) no " |
5814 | "longer affine to cpu%d\n", | |
41a2d6cf IM |
5815 | task_pid_nr(p), p->comm, dead_cpu); |
5816 | } | |
3a5c359a | 5817 | } |
f7b4cddc | 5818 | } while (!__migrate_task_irq(p, dead_cpu, dest_cpu)); |
1da177e4 LT |
5819 | } |
5820 | ||
5821 | /* | |
5822 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5823 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5824 | * for performance reasons the counter is not stricly tracking tasks to | |
5825 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5826 | * to keep the global sum constant after CPU-down: | |
5827 | */ | |
70b97a7f | 5828 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5829 | { |
7c16ec58 | 5830 | struct rq *rq_dest = cpu_rq(any_online_cpu(*CPU_MASK_ALL_PTR)); |
1da177e4 LT |
5831 | unsigned long flags; |
5832 | ||
5833 | local_irq_save(flags); | |
5834 | double_rq_lock(rq_src, rq_dest); | |
5835 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5836 | rq_src->nr_uninterruptible = 0; | |
5837 | double_rq_unlock(rq_src, rq_dest); | |
5838 | local_irq_restore(flags); | |
5839 | } | |
5840 | ||
5841 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5842 | static void migrate_live_tasks(int src_cpu) | |
5843 | { | |
48f24c4d | 5844 | struct task_struct *p, *t; |
1da177e4 | 5845 | |
f7b4cddc | 5846 | read_lock(&tasklist_lock); |
1da177e4 | 5847 | |
48f24c4d IM |
5848 | do_each_thread(t, p) { |
5849 | if (p == current) | |
1da177e4 LT |
5850 | continue; |
5851 | ||
48f24c4d IM |
5852 | if (task_cpu(p) == src_cpu) |
5853 | move_task_off_dead_cpu(src_cpu, p); | |
5854 | } while_each_thread(t, p); | |
1da177e4 | 5855 | |
f7b4cddc | 5856 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5857 | } |
5858 | ||
dd41f596 IM |
5859 | /* |
5860 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5861 | * It does so by boosting its priority to highest possible. |
5862 | * Used by CPU offline code. | |
1da177e4 LT |
5863 | */ |
5864 | void sched_idle_next(void) | |
5865 | { | |
48f24c4d | 5866 | int this_cpu = smp_processor_id(); |
70b97a7f | 5867 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5868 | struct task_struct *p = rq->idle; |
5869 | unsigned long flags; | |
5870 | ||
5871 | /* cpu has to be offline */ | |
48f24c4d | 5872 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5873 | |
48f24c4d IM |
5874 | /* |
5875 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5876 | * and interrupts disabled on the current cpu. | |
1da177e4 LT |
5877 | */ |
5878 | spin_lock_irqsave(&rq->lock, flags); | |
5879 | ||
dd41f596 | 5880 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5881 | |
94bc9a7b DA |
5882 | update_rq_clock(rq); |
5883 | activate_task(rq, p, 0); | |
1da177e4 LT |
5884 | |
5885 | spin_unlock_irqrestore(&rq->lock, flags); | |
5886 | } | |
5887 | ||
48f24c4d IM |
5888 | /* |
5889 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5890 | * offline. |
5891 | */ | |
5892 | void idle_task_exit(void) | |
5893 | { | |
5894 | struct mm_struct *mm = current->active_mm; | |
5895 | ||
5896 | BUG_ON(cpu_online(smp_processor_id())); | |
5897 | ||
5898 | if (mm != &init_mm) | |
5899 | switch_mm(mm, &init_mm, current); | |
5900 | mmdrop(mm); | |
5901 | } | |
5902 | ||
054b9108 | 5903 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5904 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5905 | { |
70b97a7f | 5906 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5907 | |
5908 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5909 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5910 | |
5911 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5912 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5913 | |
48f24c4d | 5914 | get_task_struct(p); |
1da177e4 LT |
5915 | |
5916 | /* | |
5917 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5918 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5919 | * fine. |
5920 | */ | |
f7b4cddc | 5921 | spin_unlock_irq(&rq->lock); |
48f24c4d | 5922 | move_task_off_dead_cpu(dead_cpu, p); |
f7b4cddc | 5923 | spin_lock_irq(&rq->lock); |
1da177e4 | 5924 | |
48f24c4d | 5925 | put_task_struct(p); |
1da177e4 LT |
5926 | } |
5927 | ||
5928 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5929 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5930 | { | |
70b97a7f | 5931 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5932 | struct task_struct *next; |
48f24c4d | 5933 | |
dd41f596 IM |
5934 | for ( ; ; ) { |
5935 | if (!rq->nr_running) | |
5936 | break; | |
a8e504d2 | 5937 | update_rq_clock(rq); |
ff95f3df | 5938 | next = pick_next_task(rq, rq->curr); |
dd41f596 IM |
5939 | if (!next) |
5940 | break; | |
5941 | migrate_dead(dead_cpu, next); | |
e692ab53 | 5942 | |
1da177e4 LT |
5943 | } |
5944 | } | |
5945 | #endif /* CONFIG_HOTPLUG_CPU */ | |
5946 | ||
e692ab53 NP |
5947 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5948 | ||
5949 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5950 | { |
5951 | .procname = "sched_domain", | |
c57baf1e | 5952 | .mode = 0555, |
e0361851 | 5953 | }, |
38605cae | 5954 | {0, }, |
e692ab53 NP |
5955 | }; |
5956 | ||
5957 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 | 5958 | { |
c57baf1e | 5959 | .ctl_name = CTL_KERN, |
e0361851 | 5960 | .procname = "kernel", |
c57baf1e | 5961 | .mode = 0555, |
e0361851 AD |
5962 | .child = sd_ctl_dir, |
5963 | }, | |
38605cae | 5964 | {0, }, |
e692ab53 NP |
5965 | }; |
5966 | ||
5967 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5968 | { | |
5969 | struct ctl_table *entry = | |
5cf9f062 | 5970 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5971 | |
e692ab53 NP |
5972 | return entry; |
5973 | } | |
5974 | ||
6382bc90 MM |
5975 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5976 | { | |
cd790076 | 5977 | struct ctl_table *entry; |
6382bc90 | 5978 | |
cd790076 MM |
5979 | /* |
5980 | * In the intermediate directories, both the child directory and | |
5981 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5982 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5983 | * static strings and all have proc handlers. |
5984 | */ | |
5985 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5986 | if (entry->child) |
5987 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5988 | if (entry->proc_handler == NULL) |
5989 | kfree(entry->procname); | |
5990 | } | |
6382bc90 MM |
5991 | |
5992 | kfree(*tablep); | |
5993 | *tablep = NULL; | |
5994 | } | |
5995 | ||
e692ab53 | 5996 | static void |
e0361851 | 5997 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5998 | const char *procname, void *data, int maxlen, |
5999 | mode_t mode, proc_handler *proc_handler) | |
6000 | { | |
e692ab53 NP |
6001 | entry->procname = procname; |
6002 | entry->data = data; | |
6003 | entry->maxlen = maxlen; | |
6004 | entry->mode = mode; | |
6005 | entry->proc_handler = proc_handler; | |
6006 | } | |
6007 | ||
6008 | static struct ctl_table * | |
6009 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6010 | { | |
ace8b3d6 | 6011 | struct ctl_table *table = sd_alloc_ctl_entry(12); |
e692ab53 | 6012 | |
ad1cdc1d MM |
6013 | if (table == NULL) |
6014 | return NULL; | |
6015 | ||
e0361851 | 6016 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6017 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6018 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6019 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6020 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6021 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6022 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6023 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6024 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6025 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6026 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6027 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6028 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6029 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6030 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6031 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6032 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6033 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6034 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6035 | &sd->cache_nice_tries, |
6036 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6037 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6038 | sizeof(int), 0644, proc_dointvec_minmax); |
6323469f | 6039 | /* &table[11] is terminator */ |
e692ab53 NP |
6040 | |
6041 | return table; | |
6042 | } | |
6043 | ||
9a4e7159 | 6044 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6045 | { |
6046 | struct ctl_table *entry, *table; | |
6047 | struct sched_domain *sd; | |
6048 | int domain_num = 0, i; | |
6049 | char buf[32]; | |
6050 | ||
6051 | for_each_domain(cpu, sd) | |
6052 | domain_num++; | |
6053 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6054 | if (table == NULL) |
6055 | return NULL; | |
e692ab53 NP |
6056 | |
6057 | i = 0; | |
6058 | for_each_domain(cpu, sd) { | |
6059 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6060 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6061 | entry->mode = 0555; |
e692ab53 NP |
6062 | entry->child = sd_alloc_ctl_domain_table(sd); |
6063 | entry++; | |
6064 | i++; | |
6065 | } | |
6066 | return table; | |
6067 | } | |
6068 | ||
6069 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6070 | static void register_sched_domain_sysctl(void) |
e692ab53 NP |
6071 | { |
6072 | int i, cpu_num = num_online_cpus(); | |
6073 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); | |
6074 | char buf[32]; | |
6075 | ||
7378547f MM |
6076 | WARN_ON(sd_ctl_dir[0].child); |
6077 | sd_ctl_dir[0].child = entry; | |
6078 | ||
ad1cdc1d MM |
6079 | if (entry == NULL) |
6080 | return; | |
6081 | ||
97b6ea7b | 6082 | for_each_online_cpu(i) { |
e692ab53 | 6083 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6084 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6085 | entry->mode = 0555; |
e692ab53 | 6086 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6087 | entry++; |
e692ab53 | 6088 | } |
7378547f MM |
6089 | |
6090 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6091 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6092 | } | |
6382bc90 | 6093 | |
7378547f | 6094 | /* may be called multiple times per register */ |
6382bc90 MM |
6095 | static void unregister_sched_domain_sysctl(void) |
6096 | { | |
7378547f MM |
6097 | if (sd_sysctl_header) |
6098 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6099 | sd_sysctl_header = NULL; |
7378547f MM |
6100 | if (sd_ctl_dir[0].child) |
6101 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6102 | } |
e692ab53 | 6103 | #else |
6382bc90 MM |
6104 | static void register_sched_domain_sysctl(void) |
6105 | { | |
6106 | } | |
6107 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6108 | { |
6109 | } | |
6110 | #endif | |
6111 | ||
1da177e4 LT |
6112 | /* |
6113 | * migration_call - callback that gets triggered when a CPU is added. | |
6114 | * Here we can start up the necessary migration thread for the new CPU. | |
6115 | */ | |
48f24c4d IM |
6116 | static int __cpuinit |
6117 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6118 | { |
1da177e4 | 6119 | struct task_struct *p; |
48f24c4d | 6120 | int cpu = (long)hcpu; |
1da177e4 | 6121 | unsigned long flags; |
70b97a7f | 6122 | struct rq *rq; |
1da177e4 LT |
6123 | |
6124 | switch (action) { | |
5be9361c | 6125 | |
1da177e4 | 6126 | case CPU_UP_PREPARE: |
8bb78442 | 6127 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 6128 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
6129 | if (IS_ERR(p)) |
6130 | return NOTIFY_BAD; | |
1da177e4 LT |
6131 | kthread_bind(p, cpu); |
6132 | /* Must be high prio: stop_machine expects to yield to it. */ | |
6133 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 6134 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 LT |
6135 | task_rq_unlock(rq, &flags); |
6136 | cpu_rq(cpu)->migration_thread = p; | |
6137 | break; | |
48f24c4d | 6138 | |
1da177e4 | 6139 | case CPU_ONLINE: |
8bb78442 | 6140 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 6141 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 6142 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
6143 | |
6144 | /* Update our root-domain */ | |
6145 | rq = cpu_rq(cpu); | |
6146 | spin_lock_irqsave(&rq->lock, flags); | |
6147 | if (rq->rd) { | |
6148 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
6149 | cpu_set(cpu, rq->rd->online); | |
6150 | } | |
6151 | spin_unlock_irqrestore(&rq->lock, flags); | |
1da177e4 | 6152 | break; |
48f24c4d | 6153 | |
1da177e4 LT |
6154 | #ifdef CONFIG_HOTPLUG_CPU |
6155 | case CPU_UP_CANCELED: | |
8bb78442 | 6156 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
6157 | if (!cpu_rq(cpu)->migration_thread) |
6158 | break; | |
41a2d6cf | 6159 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c HC |
6160 | kthread_bind(cpu_rq(cpu)->migration_thread, |
6161 | any_online_cpu(cpu_online_map)); | |
1da177e4 LT |
6162 | kthread_stop(cpu_rq(cpu)->migration_thread); |
6163 | cpu_rq(cpu)->migration_thread = NULL; | |
6164 | break; | |
48f24c4d | 6165 | |
1da177e4 | 6166 | case CPU_DEAD: |
8bb78442 | 6167 | case CPU_DEAD_FROZEN: |
470fd646 | 6168 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
6169 | migrate_live_tasks(cpu); |
6170 | rq = cpu_rq(cpu); | |
6171 | kthread_stop(rq->migration_thread); | |
6172 | rq->migration_thread = NULL; | |
6173 | /* Idle task back to normal (off runqueue, low prio) */ | |
d2da272a | 6174 | spin_lock_irq(&rq->lock); |
a8e504d2 | 6175 | update_rq_clock(rq); |
2e1cb74a | 6176 | deactivate_task(rq, rq->idle, 0); |
1da177e4 | 6177 | rq->idle->static_prio = MAX_PRIO; |
dd41f596 IM |
6178 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6179 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6180 | migrate_dead_tasks(cpu); |
d2da272a | 6181 | spin_unlock_irq(&rq->lock); |
470fd646 | 6182 | cpuset_unlock(); |
1da177e4 LT |
6183 | migrate_nr_uninterruptible(rq); |
6184 | BUG_ON(rq->nr_running != 0); | |
6185 | ||
41a2d6cf IM |
6186 | /* |
6187 | * No need to migrate the tasks: it was best-effort if | |
6188 | * they didn't take sched_hotcpu_mutex. Just wake up | |
6189 | * the requestors. | |
6190 | */ | |
1da177e4 LT |
6191 | spin_lock_irq(&rq->lock); |
6192 | while (!list_empty(&rq->migration_queue)) { | |
70b97a7f IM |
6193 | struct migration_req *req; |
6194 | ||
1da177e4 | 6195 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 6196 | struct migration_req, list); |
1da177e4 LT |
6197 | list_del_init(&req->list); |
6198 | complete(&req->done); | |
6199 | } | |
6200 | spin_unlock_irq(&rq->lock); | |
6201 | break; | |
57d885fe | 6202 | |
08f503b0 GH |
6203 | case CPU_DYING: |
6204 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6205 | /* Update our root-domain */ |
6206 | rq = cpu_rq(cpu); | |
6207 | spin_lock_irqsave(&rq->lock, flags); | |
6208 | if (rq->rd) { | |
6209 | BUG_ON(!cpu_isset(cpu, rq->rd->span)); | |
6210 | cpu_clear(cpu, rq->rd->online); | |
6211 | } | |
6212 | spin_unlock_irqrestore(&rq->lock, flags); | |
6213 | break; | |
1da177e4 LT |
6214 | #endif |
6215 | } | |
6216 | return NOTIFY_OK; | |
6217 | } | |
6218 | ||
6219 | /* Register at highest priority so that task migration (migrate_all_tasks) | |
6220 | * happens before everything else. | |
6221 | */ | |
26c2143b | 6222 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6223 | .notifier_call = migration_call, |
6224 | .priority = 10 | |
6225 | }; | |
6226 | ||
e6fe6649 | 6227 | void __init migration_init(void) |
1da177e4 LT |
6228 | { |
6229 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6230 | int err; |
48f24c4d IM |
6231 | |
6232 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6233 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6234 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6235 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6236 | register_cpu_notifier(&migration_notifier); | |
1da177e4 LT |
6237 | } |
6238 | #endif | |
6239 | ||
6240 | #ifdef CONFIG_SMP | |
476f3534 | 6241 | |
3e9830dc | 6242 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6243 | |
7c16ec58 MT |
6244 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
6245 | cpumask_t *groupmask) | |
1da177e4 | 6246 | { |
4dcf6aff | 6247 | struct sched_group *group = sd->groups; |
434d53b0 | 6248 | char str[256]; |
1da177e4 | 6249 | |
434d53b0 | 6250 | cpulist_scnprintf(str, sizeof(str), sd->span); |
7c16ec58 | 6251 | cpus_clear(*groupmask); |
4dcf6aff IM |
6252 | |
6253 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6254 | ||
6255 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
6256 | printk("does not load-balance\n"); | |
6257 | if (sd->parent) | |
6258 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" | |
6259 | " has parent"); | |
6260 | return -1; | |
41c7ce9a NP |
6261 | } |
6262 | ||
4dcf6aff IM |
6263 | printk(KERN_CONT "span %s\n", str); |
6264 | ||
6265 | if (!cpu_isset(cpu, sd->span)) { | |
6266 | printk(KERN_ERR "ERROR: domain->span does not contain " | |
6267 | "CPU%d\n", cpu); | |
6268 | } | |
6269 | if (!cpu_isset(cpu, group->cpumask)) { | |
6270 | printk(KERN_ERR "ERROR: domain->groups does not contain" | |
6271 | " CPU%d\n", cpu); | |
6272 | } | |
1da177e4 | 6273 | |
4dcf6aff | 6274 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6275 | do { |
4dcf6aff IM |
6276 | if (!group) { |
6277 | printk("\n"); | |
6278 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6279 | break; |
6280 | } | |
6281 | ||
4dcf6aff IM |
6282 | if (!group->__cpu_power) { |
6283 | printk(KERN_CONT "\n"); | |
6284 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6285 | "set\n"); | |
6286 | break; | |
6287 | } | |
1da177e4 | 6288 | |
4dcf6aff IM |
6289 | if (!cpus_weight(group->cpumask)) { |
6290 | printk(KERN_CONT "\n"); | |
6291 | printk(KERN_ERR "ERROR: empty group\n"); | |
6292 | break; | |
6293 | } | |
1da177e4 | 6294 | |
7c16ec58 | 6295 | if (cpus_intersects(*groupmask, group->cpumask)) { |
4dcf6aff IM |
6296 | printk(KERN_CONT "\n"); |
6297 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
6298 | break; | |
6299 | } | |
1da177e4 | 6300 | |
7c16ec58 | 6301 | cpus_or(*groupmask, *groupmask, group->cpumask); |
1da177e4 | 6302 | |
434d53b0 | 6303 | cpulist_scnprintf(str, sizeof(str), group->cpumask); |
4dcf6aff | 6304 | printk(KERN_CONT " %s", str); |
1da177e4 | 6305 | |
4dcf6aff IM |
6306 | group = group->next; |
6307 | } while (group != sd->groups); | |
6308 | printk(KERN_CONT "\n"); | |
1da177e4 | 6309 | |
7c16ec58 | 6310 | if (!cpus_equal(sd->span, *groupmask)) |
4dcf6aff | 6311 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6312 | |
7c16ec58 | 6313 | if (sd->parent && !cpus_subset(*groupmask, sd->parent->span)) |
4dcf6aff IM |
6314 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6315 | "of domain->span\n"); | |
6316 | return 0; | |
6317 | } | |
1da177e4 | 6318 | |
4dcf6aff IM |
6319 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6320 | { | |
7c16ec58 | 6321 | cpumask_t *groupmask; |
4dcf6aff | 6322 | int level = 0; |
1da177e4 | 6323 | |
4dcf6aff IM |
6324 | if (!sd) { |
6325 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6326 | return; | |
6327 | } | |
1da177e4 | 6328 | |
4dcf6aff IM |
6329 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6330 | ||
7c16ec58 MT |
6331 | groupmask = kmalloc(sizeof(cpumask_t), GFP_KERNEL); |
6332 | if (!groupmask) { | |
6333 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); | |
6334 | return; | |
6335 | } | |
6336 | ||
4dcf6aff | 6337 | for (;;) { |
7c16ec58 | 6338 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6339 | break; |
1da177e4 LT |
6340 | level++; |
6341 | sd = sd->parent; | |
33859f7f | 6342 | if (!sd) |
4dcf6aff IM |
6343 | break; |
6344 | } | |
7c16ec58 | 6345 | kfree(groupmask); |
1da177e4 LT |
6346 | } |
6347 | #else | |
48f24c4d | 6348 | # define sched_domain_debug(sd, cpu) do { } while (0) |
1da177e4 LT |
6349 | #endif |
6350 | ||
1a20ff27 | 6351 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 SS |
6352 | { |
6353 | if (cpus_weight(sd->span) == 1) | |
6354 | return 1; | |
6355 | ||
6356 | /* Following flags need at least 2 groups */ | |
6357 | if (sd->flags & (SD_LOAD_BALANCE | | |
6358 | SD_BALANCE_NEWIDLE | | |
6359 | SD_BALANCE_FORK | | |
89c4710e SS |
6360 | SD_BALANCE_EXEC | |
6361 | SD_SHARE_CPUPOWER | | |
6362 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6363 | if (sd->groups != sd->groups->next) |
6364 | return 0; | |
6365 | } | |
6366 | ||
6367 | /* Following flags don't use groups */ | |
6368 | if (sd->flags & (SD_WAKE_IDLE | | |
6369 | SD_WAKE_AFFINE | | |
6370 | SD_WAKE_BALANCE)) | |
6371 | return 0; | |
6372 | ||
6373 | return 1; | |
6374 | } | |
6375 | ||
48f24c4d IM |
6376 | static int |
6377 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6378 | { |
6379 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6380 | ||
6381 | if (sd_degenerate(parent)) | |
6382 | return 1; | |
6383 | ||
6384 | if (!cpus_equal(sd->span, parent->span)) | |
6385 | return 0; | |
6386 | ||
6387 | /* Does parent contain flags not in child? */ | |
6388 | /* WAKE_BALANCE is a subset of WAKE_AFFINE */ | |
6389 | if (cflags & SD_WAKE_AFFINE) | |
6390 | pflags &= ~SD_WAKE_BALANCE; | |
6391 | /* Flags needing groups don't count if only 1 group in parent */ | |
6392 | if (parent->groups == parent->groups->next) { | |
6393 | pflags &= ~(SD_LOAD_BALANCE | | |
6394 | SD_BALANCE_NEWIDLE | | |
6395 | SD_BALANCE_FORK | | |
89c4710e SS |
6396 | SD_BALANCE_EXEC | |
6397 | SD_SHARE_CPUPOWER | | |
6398 | SD_SHARE_PKG_RESOURCES); | |
245af2c7 SS |
6399 | } |
6400 | if (~cflags & pflags) | |
6401 | return 0; | |
6402 | ||
6403 | return 1; | |
6404 | } | |
6405 | ||
57d885fe GH |
6406 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6407 | { | |
6408 | unsigned long flags; | |
6409 | const struct sched_class *class; | |
6410 | ||
6411 | spin_lock_irqsave(&rq->lock, flags); | |
6412 | ||
6413 | if (rq->rd) { | |
6414 | struct root_domain *old_rd = rq->rd; | |
6415 | ||
0eab9146 | 6416 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6417 | if (class->leave_domain) |
6418 | class->leave_domain(rq); | |
0eab9146 | 6419 | } |
57d885fe | 6420 | |
dc938520 GH |
6421 | cpu_clear(rq->cpu, old_rd->span); |
6422 | cpu_clear(rq->cpu, old_rd->online); | |
6423 | ||
57d885fe GH |
6424 | if (atomic_dec_and_test(&old_rd->refcount)) |
6425 | kfree(old_rd); | |
6426 | } | |
6427 | ||
6428 | atomic_inc(&rd->refcount); | |
6429 | rq->rd = rd; | |
6430 | ||
dc938520 | 6431 | cpu_set(rq->cpu, rd->span); |
1f94ef59 GH |
6432 | if (cpu_isset(rq->cpu, cpu_online_map)) |
6433 | cpu_set(rq->cpu, rd->online); | |
dc938520 | 6434 | |
0eab9146 | 6435 | for (class = sched_class_highest; class; class = class->next) { |
57d885fe GH |
6436 | if (class->join_domain) |
6437 | class->join_domain(rq); | |
0eab9146 | 6438 | } |
57d885fe GH |
6439 | |
6440 | spin_unlock_irqrestore(&rq->lock, flags); | |
6441 | } | |
6442 | ||
dc938520 | 6443 | static void init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6444 | { |
6445 | memset(rd, 0, sizeof(*rd)); | |
6446 | ||
dc938520 GH |
6447 | cpus_clear(rd->span); |
6448 | cpus_clear(rd->online); | |
57d885fe GH |
6449 | } |
6450 | ||
6451 | static void init_defrootdomain(void) | |
6452 | { | |
dc938520 | 6453 | init_rootdomain(&def_root_domain); |
57d885fe GH |
6454 | atomic_set(&def_root_domain.refcount, 1); |
6455 | } | |
6456 | ||
dc938520 | 6457 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6458 | { |
6459 | struct root_domain *rd; | |
6460 | ||
6461 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6462 | if (!rd) | |
6463 | return NULL; | |
6464 | ||
dc938520 | 6465 | init_rootdomain(rd); |
57d885fe GH |
6466 | |
6467 | return rd; | |
6468 | } | |
6469 | ||
1da177e4 | 6470 | /* |
0eab9146 | 6471 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6472 | * hold the hotplug lock. |
6473 | */ | |
0eab9146 IM |
6474 | static void |
6475 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6476 | { |
70b97a7f | 6477 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6478 | struct sched_domain *tmp; |
6479 | ||
6480 | /* Remove the sched domains which do not contribute to scheduling. */ | |
6481 | for (tmp = sd; tmp; tmp = tmp->parent) { | |
6482 | struct sched_domain *parent = tmp->parent; | |
6483 | if (!parent) | |
6484 | break; | |
1a848870 | 6485 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6486 | tmp->parent = parent->parent; |
1a848870 SS |
6487 | if (parent->parent) |
6488 | parent->parent->child = tmp; | |
6489 | } | |
245af2c7 SS |
6490 | } |
6491 | ||
1a848870 | 6492 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6493 | sd = sd->parent; |
1a848870 SS |
6494 | if (sd) |
6495 | sd->child = NULL; | |
6496 | } | |
1da177e4 LT |
6497 | |
6498 | sched_domain_debug(sd, cpu); | |
6499 | ||
57d885fe | 6500 | rq_attach_root(rq, rd); |
674311d5 | 6501 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6502 | } |
6503 | ||
6504 | /* cpus with isolated domains */ | |
67af63a6 | 6505 | static cpumask_t cpu_isolated_map = CPU_MASK_NONE; |
1da177e4 LT |
6506 | |
6507 | /* Setup the mask of cpus configured for isolated domains */ | |
6508 | static int __init isolated_cpu_setup(char *str) | |
6509 | { | |
6510 | int ints[NR_CPUS], i; | |
6511 | ||
6512 | str = get_options(str, ARRAY_SIZE(ints), ints); | |
6513 | cpus_clear(cpu_isolated_map); | |
6514 | for (i = 1; i <= ints[0]; i++) | |
6515 | if (ints[i] < NR_CPUS) | |
6516 | cpu_set(ints[i], cpu_isolated_map); | |
6517 | return 1; | |
6518 | } | |
6519 | ||
8927f494 | 6520 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6521 | |
6522 | /* | |
6711cab4 SS |
6523 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6524 | * to a function which identifies what group(along with sched group) a CPU | |
6525 | * belongs to. The return value of group_fn must be a >= 0 and < NR_CPUS | |
6526 | * (due to the fact that we keep track of groups covered with a cpumask_t). | |
1da177e4 LT |
6527 | * |
6528 | * init_sched_build_groups will build a circular linked list of the groups | |
6529 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6530 | * and ->cpu_power to 0. | |
6531 | */ | |
a616058b | 6532 | static void |
7c16ec58 | 6533 | init_sched_build_groups(const cpumask_t *span, const cpumask_t *cpu_map, |
6711cab4 | 6534 | int (*group_fn)(int cpu, const cpumask_t *cpu_map, |
7c16ec58 MT |
6535 | struct sched_group **sg, |
6536 | cpumask_t *tmpmask), | |
6537 | cpumask_t *covered, cpumask_t *tmpmask) | |
1da177e4 LT |
6538 | { |
6539 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6540 | int i; |
6541 | ||
7c16ec58 MT |
6542 | cpus_clear(*covered); |
6543 | ||
6544 | for_each_cpu_mask(i, *span) { | |
6711cab4 | 6545 | struct sched_group *sg; |
7c16ec58 | 6546 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6547 | int j; |
6548 | ||
7c16ec58 | 6549 | if (cpu_isset(i, *covered)) |
1da177e4 LT |
6550 | continue; |
6551 | ||
7c16ec58 | 6552 | cpus_clear(sg->cpumask); |
5517d86b | 6553 | sg->__cpu_power = 0; |
1da177e4 | 6554 | |
7c16ec58 MT |
6555 | for_each_cpu_mask(j, *span) { |
6556 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) | |
1da177e4 LT |
6557 | continue; |
6558 | ||
7c16ec58 | 6559 | cpu_set(j, *covered); |
1da177e4 LT |
6560 | cpu_set(j, sg->cpumask); |
6561 | } | |
6562 | if (!first) | |
6563 | first = sg; | |
6564 | if (last) | |
6565 | last->next = sg; | |
6566 | last = sg; | |
6567 | } | |
6568 | last->next = first; | |
6569 | } | |
6570 | ||
9c1cfda2 | 6571 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6572 | |
9c1cfda2 | 6573 | #ifdef CONFIG_NUMA |
198e2f18 | 6574 | |
9c1cfda2 JH |
6575 | /** |
6576 | * find_next_best_node - find the next node to include in a sched_domain | |
6577 | * @node: node whose sched_domain we're building | |
6578 | * @used_nodes: nodes already in the sched_domain | |
6579 | * | |
41a2d6cf | 6580 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6581 | * finds the closest node not already in the @used_nodes map. |
6582 | * | |
6583 | * Should use nodemask_t. | |
6584 | */ | |
c5f59f08 | 6585 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6586 | { |
6587 | int i, n, val, min_val, best_node = 0; | |
6588 | ||
6589 | min_val = INT_MAX; | |
6590 | ||
6591 | for (i = 0; i < MAX_NUMNODES; i++) { | |
6592 | /* Start at @node */ | |
6593 | n = (node + i) % MAX_NUMNODES; | |
6594 | ||
6595 | if (!nr_cpus_node(n)) | |
6596 | continue; | |
6597 | ||
6598 | /* Skip already used nodes */ | |
c5f59f08 | 6599 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6600 | continue; |
6601 | ||
6602 | /* Simple min distance search */ | |
6603 | val = node_distance(node, n); | |
6604 | ||
6605 | if (val < min_val) { | |
6606 | min_val = val; | |
6607 | best_node = n; | |
6608 | } | |
6609 | } | |
6610 | ||
c5f59f08 | 6611 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6612 | return best_node; |
6613 | } | |
6614 | ||
6615 | /** | |
6616 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6617 | * @node: node whose cpumask we're constructing | |
73486722 | 6618 | * @span: resulting cpumask |
9c1cfda2 | 6619 | * |
41a2d6cf | 6620 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6621 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6622 | * out optimally. | |
6623 | */ | |
4bdbaad3 | 6624 | static void sched_domain_node_span(int node, cpumask_t *span) |
9c1cfda2 | 6625 | { |
c5f59f08 | 6626 | nodemask_t used_nodes; |
c5f59f08 | 6627 | node_to_cpumask_ptr(nodemask, node); |
48f24c4d | 6628 | int i; |
9c1cfda2 | 6629 | |
4bdbaad3 | 6630 | cpus_clear(*span); |
c5f59f08 | 6631 | nodes_clear(used_nodes); |
9c1cfda2 | 6632 | |
4bdbaad3 | 6633 | cpus_or(*span, *span, *nodemask); |
c5f59f08 | 6634 | node_set(node, used_nodes); |
9c1cfda2 JH |
6635 | |
6636 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6637 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6638 | |
c5f59f08 | 6639 | node_to_cpumask_ptr_next(nodemask, next_node); |
4bdbaad3 | 6640 | cpus_or(*span, *span, *nodemask); |
9c1cfda2 | 6641 | } |
9c1cfda2 JH |
6642 | } |
6643 | #endif | |
6644 | ||
5c45bf27 | 6645 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6646 | |
9c1cfda2 | 6647 | /* |
48f24c4d | 6648 | * SMT sched-domains: |
9c1cfda2 | 6649 | */ |
1da177e4 LT |
6650 | #ifdef CONFIG_SCHED_SMT |
6651 | static DEFINE_PER_CPU(struct sched_domain, cpu_domains); | |
6711cab4 | 6652 | static DEFINE_PER_CPU(struct sched_group, sched_group_cpus); |
48f24c4d | 6653 | |
41a2d6cf | 6654 | static int |
7c16ec58 MT |
6655 | cpu_to_cpu_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6656 | cpumask_t *unused) | |
1da177e4 | 6657 | { |
6711cab4 SS |
6658 | if (sg) |
6659 | *sg = &per_cpu(sched_group_cpus, cpu); | |
1da177e4 LT |
6660 | return cpu; |
6661 | } | |
6662 | #endif | |
6663 | ||
48f24c4d IM |
6664 | /* |
6665 | * multi-core sched-domains: | |
6666 | */ | |
1e9f28fa SS |
6667 | #ifdef CONFIG_SCHED_MC |
6668 | static DEFINE_PER_CPU(struct sched_domain, core_domains); | |
6711cab4 | 6669 | static DEFINE_PER_CPU(struct sched_group, sched_group_core); |
1e9f28fa SS |
6670 | #endif |
6671 | ||
6672 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6673 | static int |
7c16ec58 MT |
6674 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6675 | cpumask_t *mask) | |
1e9f28fa | 6676 | { |
6711cab4 | 6677 | int group; |
7c16ec58 MT |
6678 | |
6679 | *mask = per_cpu(cpu_sibling_map, cpu); | |
6680 | cpus_and(*mask, *mask, *cpu_map); | |
6681 | group = first_cpu(*mask); | |
6711cab4 SS |
6682 | if (sg) |
6683 | *sg = &per_cpu(sched_group_core, group); | |
6684 | return group; | |
1e9f28fa SS |
6685 | } |
6686 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6687 | static int |
7c16ec58 MT |
6688 | cpu_to_core_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6689 | cpumask_t *unused) | |
1e9f28fa | 6690 | { |
6711cab4 SS |
6691 | if (sg) |
6692 | *sg = &per_cpu(sched_group_core, cpu); | |
1e9f28fa SS |
6693 | return cpu; |
6694 | } | |
6695 | #endif | |
6696 | ||
1da177e4 | 6697 | static DEFINE_PER_CPU(struct sched_domain, phys_domains); |
6711cab4 | 6698 | static DEFINE_PER_CPU(struct sched_group, sched_group_phys); |
48f24c4d | 6699 | |
41a2d6cf | 6700 | static int |
7c16ec58 MT |
6701 | cpu_to_phys_group(int cpu, const cpumask_t *cpu_map, struct sched_group **sg, |
6702 | cpumask_t *mask) | |
1da177e4 | 6703 | { |
6711cab4 | 6704 | int group; |
48f24c4d | 6705 | #ifdef CONFIG_SCHED_MC |
7c16ec58 MT |
6706 | *mask = cpu_coregroup_map(cpu); |
6707 | cpus_and(*mask, *mask, *cpu_map); | |
6708 | group = first_cpu(*mask); | |
1e9f28fa | 6709 | #elif defined(CONFIG_SCHED_SMT) |
7c16ec58 MT |
6710 | *mask = per_cpu(cpu_sibling_map, cpu); |
6711 | cpus_and(*mask, *mask, *cpu_map); | |
6712 | group = first_cpu(*mask); | |
1da177e4 | 6713 | #else |
6711cab4 | 6714 | group = cpu; |
1da177e4 | 6715 | #endif |
6711cab4 SS |
6716 | if (sg) |
6717 | *sg = &per_cpu(sched_group_phys, group); | |
6718 | return group; | |
1da177e4 LT |
6719 | } |
6720 | ||
6721 | #ifdef CONFIG_NUMA | |
1da177e4 | 6722 | /* |
9c1cfda2 JH |
6723 | * The init_sched_build_groups can't handle what we want to do with node |
6724 | * groups, so roll our own. Now each node has its own list of groups which | |
6725 | * gets dynamically allocated. | |
1da177e4 | 6726 | */ |
9c1cfda2 | 6727 | static DEFINE_PER_CPU(struct sched_domain, node_domains); |
434d53b0 | 6728 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6729 | |
9c1cfda2 | 6730 | static DEFINE_PER_CPU(struct sched_domain, allnodes_domains); |
6711cab4 | 6731 | static DEFINE_PER_CPU(struct sched_group, sched_group_allnodes); |
9c1cfda2 | 6732 | |
6711cab4 | 6733 | static int cpu_to_allnodes_group(int cpu, const cpumask_t *cpu_map, |
7c16ec58 | 6734 | struct sched_group **sg, cpumask_t *nodemask) |
9c1cfda2 | 6735 | { |
6711cab4 SS |
6736 | int group; |
6737 | ||
7c16ec58 MT |
6738 | *nodemask = node_to_cpumask(cpu_to_node(cpu)); |
6739 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6740 | group = first_cpu(*nodemask); | |
6711cab4 SS |
6741 | |
6742 | if (sg) | |
6743 | *sg = &per_cpu(sched_group_allnodes, group); | |
6744 | return group; | |
1da177e4 | 6745 | } |
6711cab4 | 6746 | |
08069033 SS |
6747 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6748 | { | |
6749 | struct sched_group *sg = group_head; | |
6750 | int j; | |
6751 | ||
6752 | if (!sg) | |
6753 | return; | |
3a5c359a AK |
6754 | do { |
6755 | for_each_cpu_mask(j, sg->cpumask) { | |
6756 | struct sched_domain *sd; | |
08069033 | 6757 | |
3a5c359a AK |
6758 | sd = &per_cpu(phys_domains, j); |
6759 | if (j != first_cpu(sd->groups->cpumask)) { | |
6760 | /* | |
6761 | * Only add "power" once for each | |
6762 | * physical package. | |
6763 | */ | |
6764 | continue; | |
6765 | } | |
08069033 | 6766 | |
3a5c359a AK |
6767 | sg_inc_cpu_power(sg, sd->groups->__cpu_power); |
6768 | } | |
6769 | sg = sg->next; | |
6770 | } while (sg != group_head); | |
08069033 | 6771 | } |
1da177e4 LT |
6772 | #endif |
6773 | ||
a616058b | 6774 | #ifdef CONFIG_NUMA |
51888ca2 | 6775 | /* Free memory allocated for various sched_group structures */ |
7c16ec58 | 6776 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
51888ca2 | 6777 | { |
a616058b | 6778 | int cpu, i; |
51888ca2 SV |
6779 | |
6780 | for_each_cpu_mask(cpu, *cpu_map) { | |
51888ca2 SV |
6781 | struct sched_group **sched_group_nodes |
6782 | = sched_group_nodes_bycpu[cpu]; | |
6783 | ||
51888ca2 SV |
6784 | if (!sched_group_nodes) |
6785 | continue; | |
6786 | ||
6787 | for (i = 0; i < MAX_NUMNODES; i++) { | |
51888ca2 SV |
6788 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6789 | ||
7c16ec58 MT |
6790 | *nodemask = node_to_cpumask(i); |
6791 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
6792 | if (cpus_empty(*nodemask)) | |
51888ca2 SV |
6793 | continue; |
6794 | ||
6795 | if (sg == NULL) | |
6796 | continue; | |
6797 | sg = sg->next; | |
6798 | next_sg: | |
6799 | oldsg = sg; | |
6800 | sg = sg->next; | |
6801 | kfree(oldsg); | |
6802 | if (oldsg != sched_group_nodes[i]) | |
6803 | goto next_sg; | |
6804 | } | |
6805 | kfree(sched_group_nodes); | |
6806 | sched_group_nodes_bycpu[cpu] = NULL; | |
6807 | } | |
51888ca2 | 6808 | } |
a616058b | 6809 | #else |
7c16ec58 | 6810 | static void free_sched_groups(const cpumask_t *cpu_map, cpumask_t *nodemask) |
a616058b SS |
6811 | { |
6812 | } | |
6813 | #endif | |
51888ca2 | 6814 | |
89c4710e SS |
6815 | /* |
6816 | * Initialize sched groups cpu_power. | |
6817 | * | |
6818 | * cpu_power indicates the capacity of sched group, which is used while | |
6819 | * distributing the load between different sched groups in a sched domain. | |
6820 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6821 | * there are asymmetries in the topology. If there are asymmetries, group | |
6822 | * having more cpu_power will pickup more load compared to the group having | |
6823 | * less cpu_power. | |
6824 | * | |
6825 | * cpu_power will be a multiple of SCHED_LOAD_SCALE. This multiple represents | |
6826 | * the maximum number of tasks a group can handle in the presence of other idle | |
6827 | * or lightly loaded groups in the same sched domain. | |
6828 | */ | |
6829 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6830 | { | |
6831 | struct sched_domain *child; | |
6832 | struct sched_group *group; | |
6833 | ||
6834 | WARN_ON(!sd || !sd->groups); | |
6835 | ||
6836 | if (cpu != first_cpu(sd->groups->cpumask)) | |
6837 | return; | |
6838 | ||
6839 | child = sd->child; | |
6840 | ||
5517d86b ED |
6841 | sd->groups->__cpu_power = 0; |
6842 | ||
89c4710e SS |
6843 | /* |
6844 | * For perf policy, if the groups in child domain share resources | |
6845 | * (for example cores sharing some portions of the cache hierarchy | |
6846 | * or SMT), then set this domain groups cpu_power such that each group | |
6847 | * can handle only one task, when there are other idle groups in the | |
6848 | * same sched domain. | |
6849 | */ | |
6850 | if (!child || (!(sd->flags & SD_POWERSAVINGS_BALANCE) && | |
6851 | (child->flags & | |
6852 | (SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES)))) { | |
5517d86b | 6853 | sg_inc_cpu_power(sd->groups, SCHED_LOAD_SCALE); |
89c4710e SS |
6854 | return; |
6855 | } | |
6856 | ||
89c4710e SS |
6857 | /* |
6858 | * add cpu_power of each child group to this groups cpu_power | |
6859 | */ | |
6860 | group = child->groups; | |
6861 | do { | |
5517d86b | 6862 | sg_inc_cpu_power(sd->groups, group->__cpu_power); |
89c4710e SS |
6863 | group = group->next; |
6864 | } while (group != child->groups); | |
6865 | } | |
6866 | ||
7c16ec58 MT |
6867 | /* |
6868 | * Initializers for schedule domains | |
6869 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6870 | */ | |
6871 | ||
6872 | #define SD_INIT(sd, type) sd_init_##type(sd) | |
6873 | #define SD_INIT_FUNC(type) \ | |
6874 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6875 | { \ | |
6876 | memset(sd, 0, sizeof(*sd)); \ | |
6877 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6878 | sd->level = SD_LV_##type; \ |
7c16ec58 MT |
6879 | } |
6880 | ||
6881 | SD_INIT_FUNC(CPU) | |
6882 | #ifdef CONFIG_NUMA | |
6883 | SD_INIT_FUNC(ALLNODES) | |
6884 | SD_INIT_FUNC(NODE) | |
6885 | #endif | |
6886 | #ifdef CONFIG_SCHED_SMT | |
6887 | SD_INIT_FUNC(SIBLING) | |
6888 | #endif | |
6889 | #ifdef CONFIG_SCHED_MC | |
6890 | SD_INIT_FUNC(MC) | |
6891 | #endif | |
6892 | ||
6893 | /* | |
6894 | * To minimize stack usage kmalloc room for cpumasks and share the | |
6895 | * space as the usage in build_sched_domains() dictates. Used only | |
6896 | * if the amount of space is significant. | |
6897 | */ | |
6898 | struct allmasks { | |
6899 | cpumask_t tmpmask; /* make this one first */ | |
6900 | union { | |
6901 | cpumask_t nodemask; | |
6902 | cpumask_t this_sibling_map; | |
6903 | cpumask_t this_core_map; | |
6904 | }; | |
6905 | cpumask_t send_covered; | |
6906 | ||
6907 | #ifdef CONFIG_NUMA | |
6908 | cpumask_t domainspan; | |
6909 | cpumask_t covered; | |
6910 | cpumask_t notcovered; | |
6911 | #endif | |
6912 | }; | |
6913 | ||
6914 | #if NR_CPUS > 128 | |
6915 | #define SCHED_CPUMASK_ALLOC 1 | |
6916 | #define SCHED_CPUMASK_FREE(v) kfree(v) | |
6917 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks *v | |
6918 | #else | |
6919 | #define SCHED_CPUMASK_ALLOC 0 | |
6920 | #define SCHED_CPUMASK_FREE(v) | |
6921 | #define SCHED_CPUMASK_DECLARE(v) struct allmasks _v, *v = &_v | |
6922 | #endif | |
6923 | ||
6924 | #define SCHED_CPUMASK_VAR(v, a) cpumask_t *v = (cpumask_t *) \ | |
6925 | ((unsigned long)(a) + offsetof(struct allmasks, v)) | |
6926 | ||
1d3504fc HS |
6927 | static int default_relax_domain_level = -1; |
6928 | ||
6929 | static int __init setup_relax_domain_level(char *str) | |
6930 | { | |
30e0e178 LZ |
6931 | unsigned long val; |
6932 | ||
6933 | val = simple_strtoul(str, NULL, 0); | |
6934 | if (val < SD_LV_MAX) | |
6935 | default_relax_domain_level = val; | |
6936 | ||
1d3504fc HS |
6937 | return 1; |
6938 | } | |
6939 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6940 | ||
6941 | static void set_domain_attribute(struct sched_domain *sd, | |
6942 | struct sched_domain_attr *attr) | |
6943 | { | |
6944 | int request; | |
6945 | ||
6946 | if (!attr || attr->relax_domain_level < 0) { | |
6947 | if (default_relax_domain_level < 0) | |
6948 | return; | |
6949 | else | |
6950 | request = default_relax_domain_level; | |
6951 | } else | |
6952 | request = attr->relax_domain_level; | |
6953 | if (request < sd->level) { | |
6954 | /* turn off idle balance on this domain */ | |
6955 | sd->flags &= ~(SD_WAKE_IDLE|SD_BALANCE_NEWIDLE); | |
6956 | } else { | |
6957 | /* turn on idle balance on this domain */ | |
6958 | sd->flags |= (SD_WAKE_IDLE_FAR|SD_BALANCE_NEWIDLE); | |
6959 | } | |
6960 | } | |
6961 | ||
1da177e4 | 6962 | /* |
1a20ff27 DG |
6963 | * Build sched domains for a given set of cpus and attach the sched domains |
6964 | * to the individual cpus | |
1da177e4 | 6965 | */ |
1d3504fc HS |
6966 | static int __build_sched_domains(const cpumask_t *cpu_map, |
6967 | struct sched_domain_attr *attr) | |
1da177e4 LT |
6968 | { |
6969 | int i; | |
57d885fe | 6970 | struct root_domain *rd; |
7c16ec58 MT |
6971 | SCHED_CPUMASK_DECLARE(allmasks); |
6972 | cpumask_t *tmpmask; | |
d1b55138 JH |
6973 | #ifdef CONFIG_NUMA |
6974 | struct sched_group **sched_group_nodes = NULL; | |
6711cab4 | 6975 | int sd_allnodes = 0; |
d1b55138 JH |
6976 | |
6977 | /* | |
6978 | * Allocate the per-node list of sched groups | |
6979 | */ | |
5cf9f062 | 6980 | sched_group_nodes = kcalloc(MAX_NUMNODES, sizeof(struct sched_group *), |
41a2d6cf | 6981 | GFP_KERNEL); |
d1b55138 JH |
6982 | if (!sched_group_nodes) { |
6983 | printk(KERN_WARNING "Can not alloc sched group node list\n"); | |
51888ca2 | 6984 | return -ENOMEM; |
d1b55138 | 6985 | } |
d1b55138 | 6986 | #endif |
1da177e4 | 6987 | |
dc938520 | 6988 | rd = alloc_rootdomain(); |
57d885fe GH |
6989 | if (!rd) { |
6990 | printk(KERN_WARNING "Cannot alloc root domain\n"); | |
7c16ec58 MT |
6991 | #ifdef CONFIG_NUMA |
6992 | kfree(sched_group_nodes); | |
6993 | #endif | |
57d885fe GH |
6994 | return -ENOMEM; |
6995 | } | |
6996 | ||
7c16ec58 MT |
6997 | #if SCHED_CPUMASK_ALLOC |
6998 | /* get space for all scratch cpumask variables */ | |
6999 | allmasks = kmalloc(sizeof(*allmasks), GFP_KERNEL); | |
7000 | if (!allmasks) { | |
7001 | printk(KERN_WARNING "Cannot alloc cpumask array\n"); | |
7002 | kfree(rd); | |
7003 | #ifdef CONFIG_NUMA | |
7004 | kfree(sched_group_nodes); | |
7005 | #endif | |
7006 | return -ENOMEM; | |
7007 | } | |
7008 | #endif | |
7009 | tmpmask = (cpumask_t *)allmasks; | |
7010 | ||
7011 | ||
7012 | #ifdef CONFIG_NUMA | |
7013 | sched_group_nodes_bycpu[first_cpu(*cpu_map)] = sched_group_nodes; | |
7014 | #endif | |
7015 | ||
1da177e4 | 7016 | /* |
1a20ff27 | 7017 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7018 | */ |
1a20ff27 | 7019 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 | 7020 | struct sched_domain *sd = NULL, *p; |
7c16ec58 | 7021 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
1da177e4 | 7022 | |
7c16ec58 MT |
7023 | *nodemask = node_to_cpumask(cpu_to_node(i)); |
7024 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
1da177e4 LT |
7025 | |
7026 | #ifdef CONFIG_NUMA | |
dd41f596 | 7027 | if (cpus_weight(*cpu_map) > |
7c16ec58 | 7028 | SD_NODES_PER_DOMAIN*cpus_weight(*nodemask)) { |
9c1cfda2 | 7029 | sd = &per_cpu(allnodes_domains, i); |
7c16ec58 | 7030 | SD_INIT(sd, ALLNODES); |
1d3504fc | 7031 | set_domain_attribute(sd, attr); |
9c1cfda2 | 7032 | sd->span = *cpu_map; |
7c16ec58 | 7033 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, tmpmask); |
9c1cfda2 | 7034 | p = sd; |
6711cab4 | 7035 | sd_allnodes = 1; |
9c1cfda2 JH |
7036 | } else |
7037 | p = NULL; | |
7038 | ||
1da177e4 | 7039 | sd = &per_cpu(node_domains, i); |
7c16ec58 | 7040 | SD_INIT(sd, NODE); |
1d3504fc | 7041 | set_domain_attribute(sd, attr); |
4bdbaad3 | 7042 | sched_domain_node_span(cpu_to_node(i), &sd->span); |
9c1cfda2 | 7043 | sd->parent = p; |
1a848870 SS |
7044 | if (p) |
7045 | p->child = sd; | |
9c1cfda2 | 7046 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 LT |
7047 | #endif |
7048 | ||
7049 | p = sd; | |
7050 | sd = &per_cpu(phys_domains, i); | |
7c16ec58 | 7051 | SD_INIT(sd, CPU); |
1d3504fc | 7052 | set_domain_attribute(sd, attr); |
7c16ec58 | 7053 | sd->span = *nodemask; |
1da177e4 | 7054 | sd->parent = p; |
1a848870 SS |
7055 | if (p) |
7056 | p->child = sd; | |
7c16ec58 | 7057 | cpu_to_phys_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 | 7058 | |
1e9f28fa SS |
7059 | #ifdef CONFIG_SCHED_MC |
7060 | p = sd; | |
7061 | sd = &per_cpu(core_domains, i); | |
7c16ec58 | 7062 | SD_INIT(sd, MC); |
1d3504fc | 7063 | set_domain_attribute(sd, attr); |
1e9f28fa SS |
7064 | sd->span = cpu_coregroup_map(i); |
7065 | cpus_and(sd->span, sd->span, *cpu_map); | |
7066 | sd->parent = p; | |
1a848870 | 7067 | p->child = sd; |
7c16ec58 | 7068 | cpu_to_core_group(i, cpu_map, &sd->groups, tmpmask); |
1e9f28fa SS |
7069 | #endif |
7070 | ||
1da177e4 LT |
7071 | #ifdef CONFIG_SCHED_SMT |
7072 | p = sd; | |
7073 | sd = &per_cpu(cpu_domains, i); | |
7c16ec58 | 7074 | SD_INIT(sd, SIBLING); |
1d3504fc | 7075 | set_domain_attribute(sd, attr); |
d5a7430d | 7076 | sd->span = per_cpu(cpu_sibling_map, i); |
1a20ff27 | 7077 | cpus_and(sd->span, sd->span, *cpu_map); |
1da177e4 | 7078 | sd->parent = p; |
1a848870 | 7079 | p->child = sd; |
7c16ec58 | 7080 | cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); |
1da177e4 LT |
7081 | #endif |
7082 | } | |
7083 | ||
7084 | #ifdef CONFIG_SCHED_SMT | |
7085 | /* Set up CPU (sibling) groups */ | |
9c1cfda2 | 7086 | for_each_cpu_mask(i, *cpu_map) { |
7c16ec58 MT |
7087 | SCHED_CPUMASK_VAR(this_sibling_map, allmasks); |
7088 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7089 | ||
7090 | *this_sibling_map = per_cpu(cpu_sibling_map, i); | |
7091 | cpus_and(*this_sibling_map, *this_sibling_map, *cpu_map); | |
7092 | if (i != first_cpu(*this_sibling_map)) | |
1da177e4 LT |
7093 | continue; |
7094 | ||
dd41f596 | 7095 | init_sched_build_groups(this_sibling_map, cpu_map, |
7c16ec58 MT |
7096 | &cpu_to_cpu_group, |
7097 | send_covered, tmpmask); | |
1da177e4 LT |
7098 | } |
7099 | #endif | |
7100 | ||
1e9f28fa SS |
7101 | #ifdef CONFIG_SCHED_MC |
7102 | /* Set up multi-core groups */ | |
7103 | for_each_cpu_mask(i, *cpu_map) { | |
7c16ec58 MT |
7104 | SCHED_CPUMASK_VAR(this_core_map, allmasks); |
7105 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7106 | ||
7107 | *this_core_map = cpu_coregroup_map(i); | |
7108 | cpus_and(*this_core_map, *this_core_map, *cpu_map); | |
7109 | if (i != first_cpu(*this_core_map)) | |
1e9f28fa | 7110 | continue; |
7c16ec58 | 7111 | |
dd41f596 | 7112 | init_sched_build_groups(this_core_map, cpu_map, |
7c16ec58 MT |
7113 | &cpu_to_core_group, |
7114 | send_covered, tmpmask); | |
1e9f28fa SS |
7115 | } |
7116 | #endif | |
7117 | ||
1da177e4 LT |
7118 | /* Set up physical groups */ |
7119 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7c16ec58 MT |
7120 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7121 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
1da177e4 | 7122 | |
7c16ec58 MT |
7123 | *nodemask = node_to_cpumask(i); |
7124 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7125 | if (cpus_empty(*nodemask)) | |
1da177e4 LT |
7126 | continue; |
7127 | ||
7c16ec58 MT |
7128 | init_sched_build_groups(nodemask, cpu_map, |
7129 | &cpu_to_phys_group, | |
7130 | send_covered, tmpmask); | |
1da177e4 LT |
7131 | } |
7132 | ||
7133 | #ifdef CONFIG_NUMA | |
7134 | /* Set up node groups */ | |
7c16ec58 MT |
7135 | if (sd_allnodes) { |
7136 | SCHED_CPUMASK_VAR(send_covered, allmasks); | |
7137 | ||
7138 | init_sched_build_groups(cpu_map, cpu_map, | |
7139 | &cpu_to_allnodes_group, | |
7140 | send_covered, tmpmask); | |
7141 | } | |
9c1cfda2 JH |
7142 | |
7143 | for (i = 0; i < MAX_NUMNODES; i++) { | |
7144 | /* Set up node groups */ | |
7145 | struct sched_group *sg, *prev; | |
7c16ec58 MT |
7146 | SCHED_CPUMASK_VAR(nodemask, allmasks); |
7147 | SCHED_CPUMASK_VAR(domainspan, allmasks); | |
7148 | SCHED_CPUMASK_VAR(covered, allmasks); | |
9c1cfda2 JH |
7149 | int j; |
7150 | ||
7c16ec58 MT |
7151 | *nodemask = node_to_cpumask(i); |
7152 | cpus_clear(*covered); | |
7153 | ||
7154 | cpus_and(*nodemask, *nodemask, *cpu_map); | |
7155 | if (cpus_empty(*nodemask)) { | |
d1b55138 | 7156 | sched_group_nodes[i] = NULL; |
9c1cfda2 | 7157 | continue; |
d1b55138 | 7158 | } |
9c1cfda2 | 7159 | |
4bdbaad3 | 7160 | sched_domain_node_span(i, domainspan); |
7c16ec58 | 7161 | cpus_and(*domainspan, *domainspan, *cpu_map); |
9c1cfda2 | 7162 | |
15f0b676 | 7163 | sg = kmalloc_node(sizeof(struct sched_group), GFP_KERNEL, i); |
51888ca2 SV |
7164 | if (!sg) { |
7165 | printk(KERN_WARNING "Can not alloc domain group for " | |
7166 | "node %d\n", i); | |
7167 | goto error; | |
7168 | } | |
9c1cfda2 | 7169 | sched_group_nodes[i] = sg; |
7c16ec58 | 7170 | for_each_cpu_mask(j, *nodemask) { |
9c1cfda2 | 7171 | struct sched_domain *sd; |
9761eea8 | 7172 | |
9c1cfda2 JH |
7173 | sd = &per_cpu(node_domains, j); |
7174 | sd->groups = sg; | |
9c1cfda2 | 7175 | } |
5517d86b | 7176 | sg->__cpu_power = 0; |
7c16ec58 | 7177 | sg->cpumask = *nodemask; |
51888ca2 | 7178 | sg->next = sg; |
7c16ec58 | 7179 | cpus_or(*covered, *covered, *nodemask); |
9c1cfda2 JH |
7180 | prev = sg; |
7181 | ||
7182 | for (j = 0; j < MAX_NUMNODES; j++) { | |
7c16ec58 | 7183 | SCHED_CPUMASK_VAR(notcovered, allmasks); |
9c1cfda2 | 7184 | int n = (i + j) % MAX_NUMNODES; |
c5f59f08 | 7185 | node_to_cpumask_ptr(pnodemask, n); |
9c1cfda2 | 7186 | |
7c16ec58 MT |
7187 | cpus_complement(*notcovered, *covered); |
7188 | cpus_and(*tmpmask, *notcovered, *cpu_map); | |
7189 | cpus_and(*tmpmask, *tmpmask, *domainspan); | |
7190 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7191 | break; |
7192 | ||
7c16ec58 MT |
7193 | cpus_and(*tmpmask, *tmpmask, *pnodemask); |
7194 | if (cpus_empty(*tmpmask)) | |
9c1cfda2 JH |
7195 | continue; |
7196 | ||
15f0b676 SV |
7197 | sg = kmalloc_node(sizeof(struct sched_group), |
7198 | GFP_KERNEL, i); | |
9c1cfda2 JH |
7199 | if (!sg) { |
7200 | printk(KERN_WARNING | |
7201 | "Can not alloc domain group for node %d\n", j); | |
51888ca2 | 7202 | goto error; |
9c1cfda2 | 7203 | } |
5517d86b | 7204 | sg->__cpu_power = 0; |
7c16ec58 | 7205 | sg->cpumask = *tmpmask; |
51888ca2 | 7206 | sg->next = prev->next; |
7c16ec58 | 7207 | cpus_or(*covered, *covered, *tmpmask); |
9c1cfda2 JH |
7208 | prev->next = sg; |
7209 | prev = sg; | |
7210 | } | |
9c1cfda2 | 7211 | } |
1da177e4 LT |
7212 | #endif |
7213 | ||
7214 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7215 | #ifdef CONFIG_SCHED_SMT |
1a20ff27 | 7216 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7217 | struct sched_domain *sd = &per_cpu(cpu_domains, i); |
7218 | ||
89c4710e | 7219 | init_sched_groups_power(i, sd); |
5c45bf27 | 7220 | } |
1da177e4 | 7221 | #endif |
1e9f28fa | 7222 | #ifdef CONFIG_SCHED_MC |
5c45bf27 | 7223 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7224 | struct sched_domain *sd = &per_cpu(core_domains, i); |
7225 | ||
89c4710e | 7226 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7227 | } |
7228 | #endif | |
1e9f28fa | 7229 | |
5c45bf27 | 7230 | for_each_cpu_mask(i, *cpu_map) { |
dd41f596 IM |
7231 | struct sched_domain *sd = &per_cpu(phys_domains, i); |
7232 | ||
89c4710e | 7233 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7234 | } |
7235 | ||
9c1cfda2 | 7236 | #ifdef CONFIG_NUMA |
08069033 SS |
7237 | for (i = 0; i < MAX_NUMNODES; i++) |
7238 | init_numa_sched_groups_power(sched_group_nodes[i]); | |
9c1cfda2 | 7239 | |
6711cab4 SS |
7240 | if (sd_allnodes) { |
7241 | struct sched_group *sg; | |
f712c0c7 | 7242 | |
7c16ec58 MT |
7243 | cpu_to_allnodes_group(first_cpu(*cpu_map), cpu_map, &sg, |
7244 | tmpmask); | |
f712c0c7 SS |
7245 | init_numa_sched_groups_power(sg); |
7246 | } | |
9c1cfda2 JH |
7247 | #endif |
7248 | ||
1da177e4 | 7249 | /* Attach the domains */ |
1a20ff27 | 7250 | for_each_cpu_mask(i, *cpu_map) { |
1da177e4 LT |
7251 | struct sched_domain *sd; |
7252 | #ifdef CONFIG_SCHED_SMT | |
7253 | sd = &per_cpu(cpu_domains, i); | |
1e9f28fa SS |
7254 | #elif defined(CONFIG_SCHED_MC) |
7255 | sd = &per_cpu(core_domains, i); | |
1da177e4 LT |
7256 | #else |
7257 | sd = &per_cpu(phys_domains, i); | |
7258 | #endif | |
57d885fe | 7259 | cpu_attach_domain(sd, rd, i); |
1da177e4 | 7260 | } |
51888ca2 | 7261 | |
7c16ec58 | 7262 | SCHED_CPUMASK_FREE((void *)allmasks); |
51888ca2 SV |
7263 | return 0; |
7264 | ||
a616058b | 7265 | #ifdef CONFIG_NUMA |
51888ca2 | 7266 | error: |
7c16ec58 MT |
7267 | free_sched_groups(cpu_map, tmpmask); |
7268 | SCHED_CPUMASK_FREE((void *)allmasks); | |
51888ca2 | 7269 | return -ENOMEM; |
a616058b | 7270 | #endif |
1da177e4 | 7271 | } |
029190c5 | 7272 | |
1d3504fc HS |
7273 | static int build_sched_domains(const cpumask_t *cpu_map) |
7274 | { | |
7275 | return __build_sched_domains(cpu_map, NULL); | |
7276 | } | |
7277 | ||
029190c5 PJ |
7278 | static cpumask_t *doms_cur; /* current sched domains */ |
7279 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ | |
4285f594 IM |
7280 | static struct sched_domain_attr *dattr_cur; |
7281 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7282 | |
7283 | /* | |
7284 | * Special case: If a kmalloc of a doms_cur partition (array of | |
7285 | * cpumask_t) fails, then fallback to a single sched domain, | |
7286 | * as determined by the single cpumask_t fallback_doms. | |
7287 | */ | |
7288 | static cpumask_t fallback_doms; | |
7289 | ||
22e52b07 HC |
7290 | void __attribute__((weak)) arch_update_cpu_topology(void) |
7291 | { | |
7292 | } | |
7293 | ||
f18f982a MK |
7294 | /* |
7295 | * Free current domain masks. | |
7296 | * Called after all cpus are attached to NULL domain. | |
7297 | */ | |
7298 | static void free_sched_domains(void) | |
7299 | { | |
7300 | ndoms_cur = 0; | |
7301 | if (doms_cur != &fallback_doms) | |
7302 | kfree(doms_cur); | |
7303 | doms_cur = &fallback_doms; | |
7304 | } | |
7305 | ||
1a20ff27 | 7306 | /* |
41a2d6cf | 7307 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7308 | * For now this just excludes isolated cpus, but could be used to |
7309 | * exclude other special cases in the future. | |
1a20ff27 | 7310 | */ |
51888ca2 | 7311 | static int arch_init_sched_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7312 | { |
7378547f MM |
7313 | int err; |
7314 | ||
22e52b07 | 7315 | arch_update_cpu_topology(); |
029190c5 PJ |
7316 | ndoms_cur = 1; |
7317 | doms_cur = kmalloc(sizeof(cpumask_t), GFP_KERNEL); | |
7318 | if (!doms_cur) | |
7319 | doms_cur = &fallback_doms; | |
7320 | cpus_andnot(*doms_cur, *cpu_map, cpu_isolated_map); | |
1d3504fc | 7321 | dattr_cur = NULL; |
7378547f | 7322 | err = build_sched_domains(doms_cur); |
6382bc90 | 7323 | register_sched_domain_sysctl(); |
7378547f MM |
7324 | |
7325 | return err; | |
1a20ff27 DG |
7326 | } |
7327 | ||
7c16ec58 MT |
7328 | static void arch_destroy_sched_domains(const cpumask_t *cpu_map, |
7329 | cpumask_t *tmpmask) | |
1da177e4 | 7330 | { |
7c16ec58 | 7331 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7332 | } |
1da177e4 | 7333 | |
1a20ff27 DG |
7334 | /* |
7335 | * Detach sched domains from a group of cpus specified in cpu_map | |
7336 | * These cpus will now be attached to the NULL domain | |
7337 | */ | |
858119e1 | 7338 | static void detach_destroy_domains(const cpumask_t *cpu_map) |
1a20ff27 | 7339 | { |
7c16ec58 | 7340 | cpumask_t tmpmask; |
1a20ff27 DG |
7341 | int i; |
7342 | ||
6382bc90 MM |
7343 | unregister_sched_domain_sysctl(); |
7344 | ||
1a20ff27 | 7345 | for_each_cpu_mask(i, *cpu_map) |
57d885fe | 7346 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7347 | synchronize_sched(); |
7c16ec58 | 7348 | arch_destroy_sched_domains(cpu_map, &tmpmask); |
1a20ff27 DG |
7349 | } |
7350 | ||
1d3504fc HS |
7351 | /* handle null as "default" */ |
7352 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7353 | struct sched_domain_attr *new, int idx_new) | |
7354 | { | |
7355 | struct sched_domain_attr tmp; | |
7356 | ||
7357 | /* fast path */ | |
7358 | if (!new && !cur) | |
7359 | return 1; | |
7360 | ||
7361 | tmp = SD_ATTR_INIT; | |
7362 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7363 | new ? (new + idx_new) : &tmp, | |
7364 | sizeof(struct sched_domain_attr)); | |
7365 | } | |
7366 | ||
029190c5 PJ |
7367 | /* |
7368 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7369 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7370 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7371 | * It destroys each deleted domain and builds each new domain. | |
7372 | * | |
7373 | * 'doms_new' is an array of cpumask_t's of length 'ndoms_new'. | |
41a2d6cf IM |
7374 | * The masks don't intersect (don't overlap.) We should setup one |
7375 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7376 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7377 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7378 | * it as it is. | |
7379 | * | |
41a2d6cf IM |
7380 | * The passed in 'doms_new' should be kmalloc'd. This routine takes |
7381 | * ownership of it and will kfree it when done with it. If the caller | |
029190c5 PJ |
7382 | * failed the kmalloc call, then it can pass in doms_new == NULL, |
7383 | * and partition_sched_domains() will fallback to the single partition | |
7384 | * 'fallback_doms'. | |
7385 | * | |
7386 | * Call with hotplug lock held | |
7387 | */ | |
1d3504fc HS |
7388 | void partition_sched_domains(int ndoms_new, cpumask_t *doms_new, |
7389 | struct sched_domain_attr *dattr_new) | |
029190c5 PJ |
7390 | { |
7391 | int i, j; | |
7392 | ||
712555ee | 7393 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7394 | |
7378547f MM |
7395 | /* always unregister in case we don't destroy any domains */ |
7396 | unregister_sched_domain_sysctl(); | |
7397 | ||
029190c5 PJ |
7398 | if (doms_new == NULL) { |
7399 | ndoms_new = 1; | |
7400 | doms_new = &fallback_doms; | |
7401 | cpus_andnot(doms_new[0], cpu_online_map, cpu_isolated_map); | |
1d3504fc | 7402 | dattr_new = NULL; |
029190c5 PJ |
7403 | } |
7404 | ||
7405 | /* Destroy deleted domains */ | |
7406 | for (i = 0; i < ndoms_cur; i++) { | |
7407 | for (j = 0; j < ndoms_new; j++) { | |
1d3504fc HS |
7408 | if (cpus_equal(doms_cur[i], doms_new[j]) |
7409 | && dattrs_equal(dattr_cur, i, dattr_new, j)) | |
029190c5 PJ |
7410 | goto match1; |
7411 | } | |
7412 | /* no match - a current sched domain not in new doms_new[] */ | |
7413 | detach_destroy_domains(doms_cur + i); | |
7414 | match1: | |
7415 | ; | |
7416 | } | |
7417 | ||
7418 | /* Build new domains */ | |
7419 | for (i = 0; i < ndoms_new; i++) { | |
7420 | for (j = 0; j < ndoms_cur; j++) { | |
1d3504fc HS |
7421 | if (cpus_equal(doms_new[i], doms_cur[j]) |
7422 | && dattrs_equal(dattr_new, i, dattr_cur, j)) | |
029190c5 PJ |
7423 | goto match2; |
7424 | } | |
7425 | /* no match - add a new doms_new */ | |
1d3504fc HS |
7426 | __build_sched_domains(doms_new + i, |
7427 | dattr_new ? dattr_new + i : NULL); | |
029190c5 PJ |
7428 | match2: |
7429 | ; | |
7430 | } | |
7431 | ||
7432 | /* Remember the new sched domains */ | |
7433 | if (doms_cur != &fallback_doms) | |
7434 | kfree(doms_cur); | |
1d3504fc | 7435 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7436 | doms_cur = doms_new; |
1d3504fc | 7437 | dattr_cur = dattr_new; |
029190c5 | 7438 | ndoms_cur = ndoms_new; |
7378547f MM |
7439 | |
7440 | register_sched_domain_sysctl(); | |
a1835615 | 7441 | |
712555ee | 7442 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7443 | } |
7444 | ||
5c45bf27 | 7445 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
9aefd0ab | 7446 | int arch_reinit_sched_domains(void) |
5c45bf27 SS |
7447 | { |
7448 | int err; | |
7449 | ||
95402b38 | 7450 | get_online_cpus(); |
712555ee | 7451 | mutex_lock(&sched_domains_mutex); |
5c45bf27 | 7452 | detach_destroy_domains(&cpu_online_map); |
f18f982a | 7453 | free_sched_domains(); |
5c45bf27 | 7454 | err = arch_init_sched_domains(&cpu_online_map); |
712555ee | 7455 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7456 | put_online_cpus(); |
5c45bf27 SS |
7457 | |
7458 | return err; | |
7459 | } | |
7460 | ||
7461 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7462 | { | |
7463 | int ret; | |
7464 | ||
7465 | if (buf[0] != '0' && buf[0] != '1') | |
7466 | return -EINVAL; | |
7467 | ||
7468 | if (smt) | |
7469 | sched_smt_power_savings = (buf[0] == '1'); | |
7470 | else | |
7471 | sched_mc_power_savings = (buf[0] == '1'); | |
7472 | ||
7473 | ret = arch_reinit_sched_domains(); | |
7474 | ||
7475 | return ret ? ret : count; | |
7476 | } | |
7477 | ||
5c45bf27 SS |
7478 | #ifdef CONFIG_SCHED_MC |
7479 | static ssize_t sched_mc_power_savings_show(struct sys_device *dev, char *page) | |
7480 | { | |
7481 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7482 | } | |
48f24c4d IM |
7483 | static ssize_t sched_mc_power_savings_store(struct sys_device *dev, |
7484 | const char *buf, size_t count) | |
5c45bf27 SS |
7485 | { |
7486 | return sched_power_savings_store(buf, count, 0); | |
7487 | } | |
6707de00 AB |
7488 | static SYSDEV_ATTR(sched_mc_power_savings, 0644, sched_mc_power_savings_show, |
7489 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7490 | #endif |
7491 | ||
7492 | #ifdef CONFIG_SCHED_SMT | |
7493 | static ssize_t sched_smt_power_savings_show(struct sys_device *dev, char *page) | |
7494 | { | |
7495 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7496 | } | |
48f24c4d IM |
7497 | static ssize_t sched_smt_power_savings_store(struct sys_device *dev, |
7498 | const char *buf, size_t count) | |
5c45bf27 SS |
7499 | { |
7500 | return sched_power_savings_store(buf, count, 1); | |
7501 | } | |
6707de00 AB |
7502 | static SYSDEV_ATTR(sched_smt_power_savings, 0644, sched_smt_power_savings_show, |
7503 | sched_smt_power_savings_store); | |
7504 | #endif | |
7505 | ||
7506 | int sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) | |
7507 | { | |
7508 | int err = 0; | |
7509 | ||
7510 | #ifdef CONFIG_SCHED_SMT | |
7511 | if (smt_capable()) | |
7512 | err = sysfs_create_file(&cls->kset.kobj, | |
7513 | &attr_sched_smt_power_savings.attr); | |
7514 | #endif | |
7515 | #ifdef CONFIG_SCHED_MC | |
7516 | if (!err && mc_capable()) | |
7517 | err = sysfs_create_file(&cls->kset.kobj, | |
7518 | &attr_sched_mc_power_savings.attr); | |
7519 | #endif | |
7520 | return err; | |
7521 | } | |
5c45bf27 SS |
7522 | #endif |
7523 | ||
1da177e4 | 7524 | /* |
41a2d6cf | 7525 | * Force a reinitialization of the sched domains hierarchy. The domains |
1da177e4 | 7526 | * and groups cannot be updated in place without racing with the balancing |
41c7ce9a | 7527 | * code, so we temporarily attach all running cpus to the NULL domain |
1da177e4 LT |
7528 | * which will prevent rebalancing while the sched domains are recalculated. |
7529 | */ | |
7530 | static int update_sched_domains(struct notifier_block *nfb, | |
7531 | unsigned long action, void *hcpu) | |
7532 | { | |
1da177e4 LT |
7533 | switch (action) { |
7534 | case CPU_UP_PREPARE: | |
8bb78442 | 7535 | case CPU_UP_PREPARE_FROZEN: |
1da177e4 | 7536 | case CPU_DOWN_PREPARE: |
8bb78442 | 7537 | case CPU_DOWN_PREPARE_FROZEN: |
1a20ff27 | 7538 | detach_destroy_domains(&cpu_online_map); |
f18f982a | 7539 | free_sched_domains(); |
1da177e4 LT |
7540 | return NOTIFY_OK; |
7541 | ||
7542 | case CPU_UP_CANCELED: | |
8bb78442 | 7543 | case CPU_UP_CANCELED_FROZEN: |
1da177e4 | 7544 | case CPU_DOWN_FAILED: |
8bb78442 | 7545 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7546 | case CPU_ONLINE: |
8bb78442 | 7547 | case CPU_ONLINE_FROZEN: |
1da177e4 | 7548 | case CPU_DEAD: |
8bb78442 | 7549 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
7550 | /* |
7551 | * Fall through and re-initialise the domains. | |
7552 | */ | |
7553 | break; | |
7554 | default: | |
7555 | return NOTIFY_DONE; | |
7556 | } | |
7557 | ||
f18f982a MK |
7558 | #ifndef CONFIG_CPUSETS |
7559 | /* | |
7560 | * Create default domain partitioning if cpusets are disabled. | |
7561 | * Otherwise we let cpusets rebuild the domains based on the | |
7562 | * current setup. | |
7563 | */ | |
7564 | ||
1da177e4 | 7565 | /* The hotplug lock is already held by cpu_up/cpu_down */ |
1a20ff27 | 7566 | arch_init_sched_domains(&cpu_online_map); |
f18f982a | 7567 | #endif |
1da177e4 LT |
7568 | |
7569 | return NOTIFY_OK; | |
7570 | } | |
1da177e4 LT |
7571 | |
7572 | void __init sched_init_smp(void) | |
7573 | { | |
5c1e1767 NP |
7574 | cpumask_t non_isolated_cpus; |
7575 | ||
434d53b0 MT |
7576 | #if defined(CONFIG_NUMA) |
7577 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7578 | GFP_KERNEL); | |
7579 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7580 | #endif | |
95402b38 | 7581 | get_online_cpus(); |
712555ee | 7582 | mutex_lock(&sched_domains_mutex); |
1a20ff27 | 7583 | arch_init_sched_domains(&cpu_online_map); |
e5e5673f | 7584 | cpus_andnot(non_isolated_cpus, cpu_possible_map, cpu_isolated_map); |
5c1e1767 NP |
7585 | if (cpus_empty(non_isolated_cpus)) |
7586 | cpu_set(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7587 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7588 | put_online_cpus(); |
1da177e4 LT |
7589 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7590 | hotcpu_notifier(update_sched_domains, 0); | |
b328ca18 | 7591 | init_hrtick(); |
5c1e1767 NP |
7592 | |
7593 | /* Move init over to a non-isolated CPU */ | |
7c16ec58 | 7594 | if (set_cpus_allowed_ptr(current, &non_isolated_cpus) < 0) |
5c1e1767 | 7595 | BUG(); |
19978ca6 | 7596 | sched_init_granularity(); |
1da177e4 LT |
7597 | } |
7598 | #else | |
7599 | void __init sched_init_smp(void) | |
7600 | { | |
19978ca6 | 7601 | sched_init_granularity(); |
1da177e4 LT |
7602 | } |
7603 | #endif /* CONFIG_SMP */ | |
7604 | ||
7605 | int in_sched_functions(unsigned long addr) | |
7606 | { | |
1da177e4 LT |
7607 | return in_lock_functions(addr) || |
7608 | (addr >= (unsigned long)__sched_text_start | |
7609 | && addr < (unsigned long)__sched_text_end); | |
7610 | } | |
7611 | ||
a9957449 | 7612 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7613 | { |
7614 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7615 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7616 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7617 | cfs_rq->rq = rq; | |
7618 | #endif | |
67e9fb2a | 7619 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7620 | } |
7621 | ||
fa85ae24 PZ |
7622 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7623 | { | |
7624 | struct rt_prio_array *array; | |
7625 | int i; | |
7626 | ||
7627 | array = &rt_rq->active; | |
7628 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7629 | INIT_LIST_HEAD(array->queue + i); | |
7630 | __clear_bit(i, array->bitmap); | |
7631 | } | |
7632 | /* delimiter for bitsearch: */ | |
7633 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7634 | ||
052f1dc7 | 7635 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
48d5e258 PZ |
7636 | rt_rq->highest_prio = MAX_RT_PRIO; |
7637 | #endif | |
fa85ae24 PZ |
7638 | #ifdef CONFIG_SMP |
7639 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 PZ |
7640 | rt_rq->overloaded = 0; |
7641 | #endif | |
7642 | ||
7643 | rt_rq->rt_time = 0; | |
7644 | rt_rq->rt_throttled = 0; | |
ac086bc2 PZ |
7645 | rt_rq->rt_runtime = 0; |
7646 | spin_lock_init(&rt_rq->rt_runtime_lock); | |
6f505b16 | 7647 | |
052f1dc7 | 7648 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7649 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7650 | rt_rq->rq = rq; |
7651 | #endif | |
fa85ae24 PZ |
7652 | } |
7653 | ||
6f505b16 | 7654 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7655 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7656 | struct sched_entity *se, int cpu, int add, | |
7657 | struct sched_entity *parent) | |
6f505b16 | 7658 | { |
ec7dc8ac | 7659 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7660 | tg->cfs_rq[cpu] = cfs_rq; |
7661 | init_cfs_rq(cfs_rq, rq); | |
7662 | cfs_rq->tg = tg; | |
7663 | if (add) | |
7664 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7665 | ||
7666 | tg->se[cpu] = se; | |
354d60c2 DG |
7667 | /* se could be NULL for init_task_group */ |
7668 | if (!se) | |
7669 | return; | |
7670 | ||
ec7dc8ac DG |
7671 | if (!parent) |
7672 | se->cfs_rq = &rq->cfs; | |
7673 | else | |
7674 | se->cfs_rq = parent->my_q; | |
7675 | ||
6f505b16 PZ |
7676 | se->my_q = cfs_rq; |
7677 | se->load.weight = tg->shares; | |
e05510d0 | 7678 | se->load.inv_weight = 0; |
ec7dc8ac | 7679 | se->parent = parent; |
6f505b16 | 7680 | } |
052f1dc7 | 7681 | #endif |
6f505b16 | 7682 | |
052f1dc7 | 7683 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7684 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7685 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7686 | struct sched_rt_entity *parent) | |
6f505b16 | 7687 | { |
ec7dc8ac DG |
7688 | struct rq *rq = cpu_rq(cpu); |
7689 | ||
6f505b16 PZ |
7690 | tg->rt_rq[cpu] = rt_rq; |
7691 | init_rt_rq(rt_rq, rq); | |
7692 | rt_rq->tg = tg; | |
7693 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7694 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7695 | if (add) |
7696 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7697 | ||
7698 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7699 | if (!rt_se) |
7700 | return; | |
7701 | ||
ec7dc8ac DG |
7702 | if (!parent) |
7703 | rt_se->rt_rq = &rq->rt; | |
7704 | else | |
7705 | rt_se->rt_rq = parent->my_q; | |
7706 | ||
6f505b16 | 7707 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7708 | rt_se->parent = parent; |
6f505b16 PZ |
7709 | INIT_LIST_HEAD(&rt_se->run_list); |
7710 | } | |
7711 | #endif | |
7712 | ||
1da177e4 LT |
7713 | void __init sched_init(void) |
7714 | { | |
dd41f596 | 7715 | int i, j; |
434d53b0 MT |
7716 | unsigned long alloc_size = 0, ptr; |
7717 | ||
7718 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7719 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7720 | #endif | |
7721 | #ifdef CONFIG_RT_GROUP_SCHED | |
7722 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7723 | #endif |
7724 | #ifdef CONFIG_USER_SCHED | |
7725 | alloc_size *= 2; | |
434d53b0 MT |
7726 | #endif |
7727 | /* | |
7728 | * As sched_init() is called before page_alloc is setup, | |
7729 | * we use alloc_bootmem(). | |
7730 | */ | |
7731 | if (alloc_size) { | |
5a9d3225 | 7732 | ptr = (unsigned long)alloc_bootmem(alloc_size); |
434d53b0 MT |
7733 | |
7734 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7735 | init_task_group.se = (struct sched_entity **)ptr; | |
7736 | ptr += nr_cpu_ids * sizeof(void **); | |
7737 | ||
7738 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7739 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 PZ |
7740 | |
7741 | #ifdef CONFIG_USER_SCHED | |
7742 | root_task_group.se = (struct sched_entity **)ptr; | |
7743 | ptr += nr_cpu_ids * sizeof(void **); | |
7744 | ||
7745 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7746 | ptr += nr_cpu_ids * sizeof(void **); | |
7747 | #endif | |
434d53b0 MT |
7748 | #endif |
7749 | #ifdef CONFIG_RT_GROUP_SCHED | |
7750 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7751 | ptr += nr_cpu_ids * sizeof(void **); | |
7752 | ||
7753 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7754 | ptr += nr_cpu_ids * sizeof(void **); |
7755 | ||
7756 | #ifdef CONFIG_USER_SCHED | |
7757 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7758 | ptr += nr_cpu_ids * sizeof(void **); | |
7759 | ||
7760 | root_task_group.rt_rq = (struct rt_rq **)ptr; | |
7761 | ptr += nr_cpu_ids * sizeof(void **); | |
7762 | #endif | |
434d53b0 MT |
7763 | #endif |
7764 | } | |
dd41f596 | 7765 | |
57d885fe GH |
7766 | #ifdef CONFIG_SMP |
7767 | init_defrootdomain(); | |
7768 | #endif | |
7769 | ||
d0b27fa7 PZ |
7770 | init_rt_bandwidth(&def_rt_bandwidth, |
7771 | global_rt_period(), global_rt_runtime()); | |
7772 | ||
7773 | #ifdef CONFIG_RT_GROUP_SCHED | |
7774 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7775 | global_rt_period(), global_rt_runtime()); | |
eff766a6 PZ |
7776 | #ifdef CONFIG_USER_SCHED |
7777 | init_rt_bandwidth(&root_task_group.rt_bandwidth, | |
7778 | global_rt_period(), RUNTIME_INF); | |
7779 | #endif | |
d0b27fa7 PZ |
7780 | #endif |
7781 | ||
052f1dc7 | 7782 | #ifdef CONFIG_GROUP_SCHED |
6f505b16 | 7783 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7784 | INIT_LIST_HEAD(&init_task_group.children); |
7785 | ||
7786 | #ifdef CONFIG_USER_SCHED | |
7787 | INIT_LIST_HEAD(&root_task_group.children); | |
7788 | init_task_group.parent = &root_task_group; | |
7789 | list_add(&init_task_group.siblings, &root_task_group.children); | |
7790 | #endif | |
6f505b16 PZ |
7791 | #endif |
7792 | ||
0a945022 | 7793 | for_each_possible_cpu(i) { |
70b97a7f | 7794 | struct rq *rq; |
1da177e4 LT |
7795 | |
7796 | rq = cpu_rq(i); | |
7797 | spin_lock_init(&rq->lock); | |
fcb99371 | 7798 | lockdep_set_class(&rq->lock, &rq->rq_lock_key); |
7897986b | 7799 | rq->nr_running = 0; |
dd41f596 | 7800 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7801 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7802 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7803 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7804 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7805 | #ifdef CONFIG_CGROUP_SCHED |
7806 | /* | |
7807 | * How much cpu bandwidth does init_task_group get? | |
7808 | * | |
7809 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7810 | * gets 100% of the cpu resources in the system. This overall | |
7811 | * system cpu resource is divided among the tasks of | |
7812 | * init_task_group and its child task-groups in a fair manner, | |
7813 | * based on each entity's (task or task-group's) weight | |
7814 | * (se->load.weight). | |
7815 | * | |
7816 | * In other words, if init_task_group has 10 tasks of weight | |
7817 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7818 | * then A0's share of the cpu resource is: | |
7819 | * | |
7820 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% | |
7821 | * | |
7822 | * We achieve this by letting init_task_group's tasks sit | |
7823 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7824 | */ | |
ec7dc8ac | 7825 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
354d60c2 | 7826 | #elif defined CONFIG_USER_SCHED |
eff766a6 PZ |
7827 | root_task_group.shares = NICE_0_LOAD; |
7828 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, 0, NULL); | |
354d60c2 DG |
7829 | /* |
7830 | * In case of task-groups formed thr' the user id of tasks, | |
7831 | * init_task_group represents tasks belonging to root user. | |
7832 | * Hence it forms a sibling of all subsequent groups formed. | |
7833 | * In this case, init_task_group gets only a fraction of overall | |
7834 | * system cpu resource, based on the weight assigned to root | |
7835 | * user's cpu share (INIT_TASK_GROUP_LOAD). This is accomplished | |
7836 | * by letting tasks of init_task_group sit in a separate cfs_rq | |
7837 | * (init_cfs_rq) and having one entity represent this group of | |
7838 | * tasks in rq->cfs (i.e init_task_group->se[] != NULL). | |
7839 | */ | |
ec7dc8ac | 7840 | init_tg_cfs_entry(&init_task_group, |
6f505b16 | 7841 | &per_cpu(init_cfs_rq, i), |
eff766a6 PZ |
7842 | &per_cpu(init_sched_entity, i), i, 1, |
7843 | root_task_group.se[i]); | |
6f505b16 | 7844 | |
052f1dc7 | 7845 | #endif |
354d60c2 DG |
7846 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7847 | ||
7848 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7849 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7850 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7851 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7852 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7853 | #elif defined CONFIG_USER_SCHED |
eff766a6 | 7854 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, 0, NULL); |
ec7dc8ac | 7855 | init_tg_rt_entry(&init_task_group, |
6f505b16 | 7856 | &per_cpu(init_rt_rq, i), |
eff766a6 PZ |
7857 | &per_cpu(init_sched_rt_entity, i), i, 1, |
7858 | root_task_group.rt_se[i]); | |
354d60c2 | 7859 | #endif |
dd41f596 | 7860 | #endif |
1da177e4 | 7861 | |
dd41f596 IM |
7862 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7863 | rq->cpu_load[j] = 0; | |
1da177e4 | 7864 | #ifdef CONFIG_SMP |
41c7ce9a | 7865 | rq->sd = NULL; |
57d885fe | 7866 | rq->rd = NULL; |
1da177e4 | 7867 | rq->active_balance = 0; |
dd41f596 | 7868 | rq->next_balance = jiffies; |
1da177e4 | 7869 | rq->push_cpu = 0; |
0a2966b4 | 7870 | rq->cpu = i; |
1da177e4 LT |
7871 | rq->migration_thread = NULL; |
7872 | INIT_LIST_HEAD(&rq->migration_queue); | |
dc938520 | 7873 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7874 | #endif |
8f4d37ec | 7875 | init_rq_hrtick(rq); |
1da177e4 | 7876 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7877 | } |
7878 | ||
2dd73a4f | 7879 | set_load_weight(&init_task); |
b50f60ce | 7880 | |
e107be36 AK |
7881 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7882 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7883 | #endif | |
7884 | ||
c9819f45 CL |
7885 | #ifdef CONFIG_SMP |
7886 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains, NULL); | |
7887 | #endif | |
7888 | ||
b50f60ce HC |
7889 | #ifdef CONFIG_RT_MUTEXES |
7890 | plist_head_init(&init_task.pi_waiters, &init_task.pi_lock); | |
7891 | #endif | |
7892 | ||
1da177e4 LT |
7893 | /* |
7894 | * The boot idle thread does lazy MMU switching as well: | |
7895 | */ | |
7896 | atomic_inc(&init_mm.mm_count); | |
7897 | enter_lazy_tlb(&init_mm, current); | |
7898 | ||
7899 | /* | |
7900 | * Make us the idle thread. Technically, schedule() should not be | |
7901 | * called from this thread, however somewhere below it might be, | |
7902 | * but because we are the idle thread, we just pick up running again | |
7903 | * when this runqueue becomes "idle". | |
7904 | */ | |
7905 | init_idle(current, smp_processor_id()); | |
dd41f596 IM |
7906 | /* |
7907 | * During early bootup we pretend to be a normal task: | |
7908 | */ | |
7909 | current->sched_class = &fair_sched_class; | |
6892b75e IM |
7910 | |
7911 | scheduler_running = 1; | |
1da177e4 LT |
7912 | } |
7913 | ||
7914 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
7915 | void __might_sleep(char *file, int line) | |
7916 | { | |
48f24c4d | 7917 | #ifdef in_atomic |
1da177e4 LT |
7918 | static unsigned long prev_jiffy; /* ratelimiting */ |
7919 | ||
7920 | if ((in_atomic() || irqs_disabled()) && | |
7921 | system_state == SYSTEM_RUNNING && !oops_in_progress) { | |
7922 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7923 | return; | |
7924 | prev_jiffy = jiffies; | |
91368d73 | 7925 | printk(KERN_ERR "BUG: sleeping function called from invalid" |
1da177e4 LT |
7926 | " context at %s:%d\n", file, line); |
7927 | printk("in_atomic():%d, irqs_disabled():%d\n", | |
7928 | in_atomic(), irqs_disabled()); | |
a4c410f0 | 7929 | debug_show_held_locks(current); |
3117df04 IM |
7930 | if (irqs_disabled()) |
7931 | print_irqtrace_events(current); | |
1da177e4 LT |
7932 | dump_stack(); |
7933 | } | |
7934 | #endif | |
7935 | } | |
7936 | EXPORT_SYMBOL(__might_sleep); | |
7937 | #endif | |
7938 | ||
7939 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7940 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7941 | { | |
7942 | int on_rq; | |
3e51f33f | 7943 | |
3a5e4dc1 AK |
7944 | update_rq_clock(rq); |
7945 | on_rq = p->se.on_rq; | |
7946 | if (on_rq) | |
7947 | deactivate_task(rq, p, 0); | |
7948 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7949 | if (on_rq) { | |
7950 | activate_task(rq, p, 0); | |
7951 | resched_task(rq->curr); | |
7952 | } | |
7953 | } | |
7954 | ||
1da177e4 LT |
7955 | void normalize_rt_tasks(void) |
7956 | { | |
a0f98a1c | 7957 | struct task_struct *g, *p; |
1da177e4 | 7958 | unsigned long flags; |
70b97a7f | 7959 | struct rq *rq; |
1da177e4 | 7960 | |
4cf5d77a | 7961 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7962 | do_each_thread(g, p) { |
178be793 IM |
7963 | /* |
7964 | * Only normalize user tasks: | |
7965 | */ | |
7966 | if (!p->mm) | |
7967 | continue; | |
7968 | ||
6cfb0d5d | 7969 | p->se.exec_start = 0; |
6cfb0d5d | 7970 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7971 | p->se.wait_start = 0; |
dd41f596 | 7972 | p->se.sleep_start = 0; |
dd41f596 | 7973 | p->se.block_start = 0; |
6cfb0d5d | 7974 | #endif |
dd41f596 IM |
7975 | |
7976 | if (!rt_task(p)) { | |
7977 | /* | |
7978 | * Renice negative nice level userspace | |
7979 | * tasks back to 0: | |
7980 | */ | |
7981 | if (TASK_NICE(p) < 0 && p->mm) | |
7982 | set_user_nice(p, 0); | |
1da177e4 | 7983 | continue; |
dd41f596 | 7984 | } |
1da177e4 | 7985 | |
4cf5d77a | 7986 | spin_lock(&p->pi_lock); |
b29739f9 | 7987 | rq = __task_rq_lock(p); |
1da177e4 | 7988 | |
178be793 | 7989 | normalize_task(rq, p); |
3a5e4dc1 | 7990 | |
b29739f9 | 7991 | __task_rq_unlock(rq); |
4cf5d77a | 7992 | spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7993 | } while_each_thread(g, p); |
7994 | ||
4cf5d77a | 7995 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7996 | } |
7997 | ||
7998 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7999 | |
8000 | #ifdef CONFIG_IA64 | |
8001 | /* | |
8002 | * These functions are only useful for the IA64 MCA handling. | |
8003 | * | |
8004 | * They can only be called when the whole system has been | |
8005 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8006 | * activity can take place. Using them for anything else would | |
8007 | * be a serious bug, and as a result, they aren't even visible | |
8008 | * under any other configuration. | |
8009 | */ | |
8010 | ||
8011 | /** | |
8012 | * curr_task - return the current task for a given cpu. | |
8013 | * @cpu: the processor in question. | |
8014 | * | |
8015 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8016 | */ | |
36c8b586 | 8017 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8018 | { |
8019 | return cpu_curr(cpu); | |
8020 | } | |
8021 | ||
8022 | /** | |
8023 | * set_curr_task - set the current task for a given cpu. | |
8024 | * @cpu: the processor in question. | |
8025 | * @p: the task pointer to set. | |
8026 | * | |
8027 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8028 | * are serviced on a separate stack. It allows the architecture to switch the |
8029 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8030 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8031 | * and caller must save the original value of the current task (see | |
8032 | * curr_task() above) and restore that value before reenabling interrupts and | |
8033 | * re-starting the system. | |
8034 | * | |
8035 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8036 | */ | |
36c8b586 | 8037 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8038 | { |
8039 | cpu_curr(cpu) = p; | |
8040 | } | |
8041 | ||
8042 | #endif | |
29f59db3 | 8043 | |
bccbe08a PZ |
8044 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8045 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8046 | { |
8047 | int i; | |
8048 | ||
8049 | for_each_possible_cpu(i) { | |
8050 | if (tg->cfs_rq) | |
8051 | kfree(tg->cfs_rq[i]); | |
8052 | if (tg->se) | |
8053 | kfree(tg->se[i]); | |
6f505b16 PZ |
8054 | } |
8055 | ||
8056 | kfree(tg->cfs_rq); | |
8057 | kfree(tg->se); | |
6f505b16 PZ |
8058 | } |
8059 | ||
ec7dc8ac DG |
8060 | static |
8061 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8062 | { |
29f59db3 | 8063 | struct cfs_rq *cfs_rq; |
ec7dc8ac | 8064 | struct sched_entity *se, *parent_se; |
9b5b7751 | 8065 | struct rq *rq; |
29f59db3 SV |
8066 | int i; |
8067 | ||
434d53b0 | 8068 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8069 | if (!tg->cfs_rq) |
8070 | goto err; | |
434d53b0 | 8071 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8072 | if (!tg->se) |
8073 | goto err; | |
052f1dc7 PZ |
8074 | |
8075 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8076 | |
8077 | for_each_possible_cpu(i) { | |
9b5b7751 | 8078 | rq = cpu_rq(i); |
29f59db3 | 8079 | |
6f505b16 PZ |
8080 | cfs_rq = kmalloc_node(sizeof(struct cfs_rq), |
8081 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8082 | if (!cfs_rq) |
8083 | goto err; | |
8084 | ||
6f505b16 PZ |
8085 | se = kmalloc_node(sizeof(struct sched_entity), |
8086 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
29f59db3 SV |
8087 | if (!se) |
8088 | goto err; | |
8089 | ||
ec7dc8ac DG |
8090 | parent_se = parent ? parent->se[i] : NULL; |
8091 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent_se); | |
bccbe08a PZ |
8092 | } |
8093 | ||
8094 | return 1; | |
8095 | ||
8096 | err: | |
8097 | return 0; | |
8098 | } | |
8099 | ||
8100 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8101 | { | |
8102 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8103 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8104 | } | |
8105 | ||
8106 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8107 | { | |
8108 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8109 | } | |
8110 | #else | |
8111 | static inline void free_fair_sched_group(struct task_group *tg) | |
8112 | { | |
8113 | } | |
8114 | ||
ec7dc8ac DG |
8115 | static inline |
8116 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8117 | { |
8118 | return 1; | |
8119 | } | |
8120 | ||
8121 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8122 | { | |
8123 | } | |
8124 | ||
8125 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8126 | { | |
8127 | } | |
052f1dc7 PZ |
8128 | #endif |
8129 | ||
8130 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8131 | static void free_rt_sched_group(struct task_group *tg) |
8132 | { | |
8133 | int i; | |
8134 | ||
d0b27fa7 PZ |
8135 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8136 | ||
bccbe08a PZ |
8137 | for_each_possible_cpu(i) { |
8138 | if (tg->rt_rq) | |
8139 | kfree(tg->rt_rq[i]); | |
8140 | if (tg->rt_se) | |
8141 | kfree(tg->rt_se[i]); | |
8142 | } | |
8143 | ||
8144 | kfree(tg->rt_rq); | |
8145 | kfree(tg->rt_se); | |
8146 | } | |
8147 | ||
ec7dc8ac DG |
8148 | static |
8149 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8150 | { |
8151 | struct rt_rq *rt_rq; | |
ec7dc8ac | 8152 | struct sched_rt_entity *rt_se, *parent_se; |
bccbe08a PZ |
8153 | struct rq *rq; |
8154 | int i; | |
8155 | ||
434d53b0 | 8156 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8157 | if (!tg->rt_rq) |
8158 | goto err; | |
434d53b0 | 8159 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8160 | if (!tg->rt_se) |
8161 | goto err; | |
8162 | ||
d0b27fa7 PZ |
8163 | init_rt_bandwidth(&tg->rt_bandwidth, |
8164 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8165 | |
8166 | for_each_possible_cpu(i) { | |
8167 | rq = cpu_rq(i); | |
8168 | ||
6f505b16 PZ |
8169 | rt_rq = kmalloc_node(sizeof(struct rt_rq), |
8170 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8171 | if (!rt_rq) | |
8172 | goto err; | |
29f59db3 | 8173 | |
6f505b16 PZ |
8174 | rt_se = kmalloc_node(sizeof(struct sched_rt_entity), |
8175 | GFP_KERNEL|__GFP_ZERO, cpu_to_node(i)); | |
8176 | if (!rt_se) | |
8177 | goto err; | |
29f59db3 | 8178 | |
ec7dc8ac DG |
8179 | parent_se = parent ? parent->rt_se[i] : NULL; |
8180 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent_se); | |
29f59db3 SV |
8181 | } |
8182 | ||
bccbe08a PZ |
8183 | return 1; |
8184 | ||
8185 | err: | |
8186 | return 0; | |
8187 | } | |
8188 | ||
8189 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8190 | { | |
8191 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8192 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8193 | } | |
8194 | ||
8195 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8196 | { | |
8197 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8198 | } | |
8199 | #else | |
8200 | static inline void free_rt_sched_group(struct task_group *tg) | |
8201 | { | |
8202 | } | |
8203 | ||
ec7dc8ac DG |
8204 | static inline |
8205 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8206 | { |
8207 | return 1; | |
8208 | } | |
8209 | ||
8210 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8211 | { | |
8212 | } | |
8213 | ||
8214 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8215 | { | |
8216 | } | |
8217 | #endif | |
8218 | ||
d0b27fa7 | 8219 | #ifdef CONFIG_GROUP_SCHED |
bccbe08a PZ |
8220 | static void free_sched_group(struct task_group *tg) |
8221 | { | |
8222 | free_fair_sched_group(tg); | |
8223 | free_rt_sched_group(tg); | |
8224 | kfree(tg); | |
8225 | } | |
8226 | ||
8227 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8228 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8229 | { |
8230 | struct task_group *tg; | |
8231 | unsigned long flags; | |
8232 | int i; | |
8233 | ||
8234 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8235 | if (!tg) | |
8236 | return ERR_PTR(-ENOMEM); | |
8237 | ||
ec7dc8ac | 8238 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8239 | goto err; |
8240 | ||
ec7dc8ac | 8241 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8242 | goto err; |
8243 | ||
8ed36996 | 8244 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8245 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8246 | register_fair_sched_group(tg, i); |
8247 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8248 | } |
6f505b16 | 8249 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8250 | |
8251 | WARN_ON(!parent); /* root should already exist */ | |
8252 | ||
8253 | tg->parent = parent; | |
8254 | list_add_rcu(&tg->siblings, &parent->children); | |
8255 | INIT_LIST_HEAD(&tg->children); | |
8ed36996 | 8256 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8257 | |
9b5b7751 | 8258 | return tg; |
29f59db3 SV |
8259 | |
8260 | err: | |
6f505b16 | 8261 | free_sched_group(tg); |
29f59db3 SV |
8262 | return ERR_PTR(-ENOMEM); |
8263 | } | |
8264 | ||
9b5b7751 | 8265 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8266 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8267 | { |
29f59db3 | 8268 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8269 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8270 | } |
8271 | ||
9b5b7751 | 8272 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8273 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8274 | { |
8ed36996 | 8275 | unsigned long flags; |
9b5b7751 | 8276 | int i; |
29f59db3 | 8277 | |
8ed36996 | 8278 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8279 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8280 | unregister_fair_sched_group(tg, i); |
8281 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8282 | } |
6f505b16 | 8283 | list_del_rcu(&tg->list); |
f473aa5e | 8284 | list_del_rcu(&tg->siblings); |
8ed36996 | 8285 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8286 | |
9b5b7751 | 8287 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8288 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8289 | } |
8290 | ||
9b5b7751 | 8291 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8292 | * The caller of this function should have put the task in its new group |
8293 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8294 | * reflect its new group. | |
9b5b7751 SV |
8295 | */ |
8296 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8297 | { |
8298 | int on_rq, running; | |
8299 | unsigned long flags; | |
8300 | struct rq *rq; | |
8301 | ||
8302 | rq = task_rq_lock(tsk, &flags); | |
8303 | ||
29f59db3 SV |
8304 | update_rq_clock(rq); |
8305 | ||
051a1d1a | 8306 | running = task_current(rq, tsk); |
29f59db3 SV |
8307 | on_rq = tsk->se.on_rq; |
8308 | ||
0e1f3483 | 8309 | if (on_rq) |
29f59db3 | 8310 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8311 | if (unlikely(running)) |
8312 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8313 | |
6f505b16 | 8314 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8315 | |
810b3817 PZ |
8316 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8317 | if (tsk->sched_class->moved_group) | |
8318 | tsk->sched_class->moved_group(tsk); | |
8319 | #endif | |
8320 | ||
0e1f3483 HS |
8321 | if (unlikely(running)) |
8322 | tsk->sched_class->set_curr_task(rq); | |
8323 | if (on_rq) | |
7074badb | 8324 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8325 | |
29f59db3 SV |
8326 | task_rq_unlock(rq, &flags); |
8327 | } | |
d0b27fa7 | 8328 | #endif |
29f59db3 | 8329 | |
052f1dc7 | 8330 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6363ca57 | 8331 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8332 | { |
8333 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
6363ca57 | 8334 | struct rq *rq = cfs_rq->rq; |
29f59db3 SV |
8335 | int on_rq; |
8336 | ||
6363ca57 IM |
8337 | spin_lock_irq(&rq->lock); |
8338 | ||
29f59db3 | 8339 | on_rq = se->on_rq; |
62fb1851 | 8340 | if (on_rq) |
29f59db3 SV |
8341 | dequeue_entity(cfs_rq, se, 0); |
8342 | ||
8343 | se->load.weight = shares; | |
e05510d0 | 8344 | se->load.inv_weight = 0; |
29f59db3 | 8345 | |
62fb1851 | 8346 | if (on_rq) |
29f59db3 | 8347 | enqueue_entity(cfs_rq, se, 0); |
62fb1851 | 8348 | |
6363ca57 | 8349 | spin_unlock_irq(&rq->lock); |
29f59db3 SV |
8350 | } |
8351 | ||
8ed36996 PZ |
8352 | static DEFINE_MUTEX(shares_mutex); |
8353 | ||
4cf86d77 | 8354 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8355 | { |
8356 | int i; | |
8ed36996 | 8357 | unsigned long flags; |
c61935fd | 8358 | |
ec7dc8ac DG |
8359 | /* |
8360 | * We can't change the weight of the root cgroup. | |
8361 | */ | |
8362 | if (!tg->se[0]) | |
8363 | return -EINVAL; | |
8364 | ||
18d95a28 PZ |
8365 | if (shares < MIN_SHARES) |
8366 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8367 | else if (shares > MAX_SHARES) |
8368 | shares = MAX_SHARES; | |
62fb1851 | 8369 | |
8ed36996 | 8370 | mutex_lock(&shares_mutex); |
9b5b7751 | 8371 | if (tg->shares == shares) |
5cb350ba | 8372 | goto done; |
29f59db3 | 8373 | |
8ed36996 | 8374 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8375 | for_each_possible_cpu(i) |
8376 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8377 | list_del_rcu(&tg->siblings); |
8ed36996 | 8378 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8379 | |
8380 | /* wait for any ongoing reference to this group to finish */ | |
8381 | synchronize_sched(); | |
8382 | ||
8383 | /* | |
8384 | * Now we are free to modify the group's share on each cpu | |
8385 | * w/o tripping rebalance_share or load_balance_fair. | |
8386 | */ | |
9b5b7751 | 8387 | tg->shares = shares; |
6363ca57 | 8388 | for_each_possible_cpu(i) |
cb4ad1ff | 8389 | set_se_shares(tg->se[i], shares); |
29f59db3 | 8390 | |
6b2d7700 SV |
8391 | /* |
8392 | * Enable load balance activity on this group, by inserting it back on | |
8393 | * each cpu's rq->leaf_cfs_rq_list. | |
8394 | */ | |
8ed36996 | 8395 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8396 | for_each_possible_cpu(i) |
8397 | register_fair_sched_group(tg, i); | |
f473aa5e | 8398 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8399 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8400 | done: |
8ed36996 | 8401 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8402 | return 0; |
29f59db3 SV |
8403 | } |
8404 | ||
5cb350ba DG |
8405 | unsigned long sched_group_shares(struct task_group *tg) |
8406 | { | |
8407 | return tg->shares; | |
8408 | } | |
052f1dc7 | 8409 | #endif |
5cb350ba | 8410 | |
052f1dc7 | 8411 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8412 | /* |
9f0c1e56 | 8413 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8414 | */ |
9f0c1e56 PZ |
8415 | static DEFINE_MUTEX(rt_constraints_mutex); |
8416 | ||
8417 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8418 | { | |
8419 | if (runtime == RUNTIME_INF) | |
8420 | return 1ULL << 16; | |
8421 | ||
6f6d6a1a | 8422 | return div64_u64(runtime << 16, period); |
9f0c1e56 PZ |
8423 | } |
8424 | ||
b40b2e8e PZ |
8425 | #ifdef CONFIG_CGROUP_SCHED |
8426 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) | |
8427 | { | |
49307fd6 | 8428 | struct task_group *tgi, *parent = tg ? tg->parent : NULL; |
b40b2e8e PZ |
8429 | unsigned long total = 0; |
8430 | ||
8431 | if (!parent) { | |
8432 | if (global_rt_period() < period) | |
8433 | return 0; | |
8434 | ||
8435 | return to_ratio(period, runtime) < | |
8436 | to_ratio(global_rt_period(), global_rt_runtime()); | |
8437 | } | |
8438 | ||
8439 | if (ktime_to_ns(parent->rt_bandwidth.rt_period) < period) | |
8440 | return 0; | |
8441 | ||
8442 | rcu_read_lock(); | |
8443 | list_for_each_entry_rcu(tgi, &parent->children, siblings) { | |
8444 | if (tgi == tg) | |
8445 | continue; | |
8446 | ||
8447 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), | |
8448 | tgi->rt_bandwidth.rt_runtime); | |
8449 | } | |
8450 | rcu_read_unlock(); | |
8451 | ||
8452 | return total + to_ratio(period, runtime) < | |
8453 | to_ratio(ktime_to_ns(parent->rt_bandwidth.rt_period), | |
8454 | parent->rt_bandwidth.rt_runtime); | |
8455 | } | |
8456 | #elif defined CONFIG_USER_SCHED | |
9f0c1e56 | 8457 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
6f505b16 PZ |
8458 | { |
8459 | struct task_group *tgi; | |
8460 | unsigned long total = 0; | |
9f0c1e56 | 8461 | unsigned long global_ratio = |
d0b27fa7 | 8462 | to_ratio(global_rt_period(), global_rt_runtime()); |
6f505b16 PZ |
8463 | |
8464 | rcu_read_lock(); | |
9f0c1e56 PZ |
8465 | list_for_each_entry_rcu(tgi, &task_groups, list) { |
8466 | if (tgi == tg) | |
8467 | continue; | |
6f505b16 | 8468 | |
d0b27fa7 PZ |
8469 | total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period), |
8470 | tgi->rt_bandwidth.rt_runtime); | |
9f0c1e56 PZ |
8471 | } |
8472 | rcu_read_unlock(); | |
6f505b16 | 8473 | |
9f0c1e56 | 8474 | return total + to_ratio(period, runtime) < global_ratio; |
6f505b16 | 8475 | } |
b40b2e8e | 8476 | #endif |
6f505b16 | 8477 | |
521f1a24 DG |
8478 | /* Must be called with tasklist_lock held */ |
8479 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
8480 | { | |
8481 | struct task_struct *g, *p; | |
8482 | do_each_thread(g, p) { | |
8483 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8484 | return 1; | |
8485 | } while_each_thread(g, p); | |
8486 | return 0; | |
8487 | } | |
8488 | ||
d0b27fa7 PZ |
8489 | static int tg_set_bandwidth(struct task_group *tg, |
8490 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8491 | { |
ac086bc2 | 8492 | int i, err = 0; |
9f0c1e56 | 8493 | |
9f0c1e56 | 8494 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8495 | read_lock(&tasklist_lock); |
ac086bc2 | 8496 | if (rt_runtime == 0 && tg_has_rt_tasks(tg)) { |
521f1a24 DG |
8497 | err = -EBUSY; |
8498 | goto unlock; | |
8499 | } | |
9f0c1e56 PZ |
8500 | if (!__rt_schedulable(tg, rt_period, rt_runtime)) { |
8501 | err = -EINVAL; | |
8502 | goto unlock; | |
8503 | } | |
ac086bc2 PZ |
8504 | |
8505 | spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
d0b27fa7 PZ |
8506 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8507 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8508 | |
8509 | for_each_possible_cpu(i) { | |
8510 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8511 | ||
8512 | spin_lock(&rt_rq->rt_runtime_lock); | |
8513 | rt_rq->rt_runtime = rt_runtime; | |
8514 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8515 | } | |
8516 | spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); | |
9f0c1e56 | 8517 | unlock: |
521f1a24 | 8518 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8519 | mutex_unlock(&rt_constraints_mutex); |
8520 | ||
8521 | return err; | |
6f505b16 PZ |
8522 | } |
8523 | ||
d0b27fa7 PZ |
8524 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8525 | { | |
8526 | u64 rt_runtime, rt_period; | |
8527 | ||
8528 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8529 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8530 | if (rt_runtime_us < 0) | |
8531 | rt_runtime = RUNTIME_INF; | |
8532 | ||
8533 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8534 | } | |
8535 | ||
9f0c1e56 PZ |
8536 | long sched_group_rt_runtime(struct task_group *tg) |
8537 | { | |
8538 | u64 rt_runtime_us; | |
8539 | ||
d0b27fa7 | 8540 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8541 | return -1; |
8542 | ||
d0b27fa7 | 8543 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8544 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8545 | return rt_runtime_us; | |
8546 | } | |
d0b27fa7 PZ |
8547 | |
8548 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8549 | { | |
8550 | u64 rt_runtime, rt_period; | |
8551 | ||
8552 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8553 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8554 | ||
8555 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8556 | } | |
8557 | ||
8558 | long sched_group_rt_period(struct task_group *tg) | |
8559 | { | |
8560 | u64 rt_period_us; | |
8561 | ||
8562 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8563 | do_div(rt_period_us, NSEC_PER_USEC); | |
8564 | return rt_period_us; | |
8565 | } | |
8566 | ||
8567 | static int sched_rt_global_constraints(void) | |
8568 | { | |
8569 | int ret = 0; | |
8570 | ||
8571 | mutex_lock(&rt_constraints_mutex); | |
8572 | if (!__rt_schedulable(NULL, 1, 0)) | |
8573 | ret = -EINVAL; | |
8574 | mutex_unlock(&rt_constraints_mutex); | |
8575 | ||
8576 | return ret; | |
8577 | } | |
8578 | #else | |
8579 | static int sched_rt_global_constraints(void) | |
8580 | { | |
ac086bc2 PZ |
8581 | unsigned long flags; |
8582 | int i; | |
8583 | ||
8584 | spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8585 | for_each_possible_cpu(i) { | |
8586 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8587 | ||
8588 | spin_lock(&rt_rq->rt_runtime_lock); | |
8589 | rt_rq->rt_runtime = global_rt_runtime(); | |
8590 | spin_unlock(&rt_rq->rt_runtime_lock); | |
8591 | } | |
8592 | spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); | |
8593 | ||
d0b27fa7 PZ |
8594 | return 0; |
8595 | } | |
052f1dc7 | 8596 | #endif |
d0b27fa7 PZ |
8597 | |
8598 | int sched_rt_handler(struct ctl_table *table, int write, | |
8599 | struct file *filp, void __user *buffer, size_t *lenp, | |
8600 | loff_t *ppos) | |
8601 | { | |
8602 | int ret; | |
8603 | int old_period, old_runtime; | |
8604 | static DEFINE_MUTEX(mutex); | |
8605 | ||
8606 | mutex_lock(&mutex); | |
8607 | old_period = sysctl_sched_rt_period; | |
8608 | old_runtime = sysctl_sched_rt_runtime; | |
8609 | ||
8610 | ret = proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
8611 | ||
8612 | if (!ret && write) { | |
8613 | ret = sched_rt_global_constraints(); | |
8614 | if (ret) { | |
8615 | sysctl_sched_rt_period = old_period; | |
8616 | sysctl_sched_rt_runtime = old_runtime; | |
8617 | } else { | |
8618 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8619 | def_rt_bandwidth.rt_period = | |
8620 | ns_to_ktime(global_rt_period()); | |
8621 | } | |
8622 | } | |
8623 | mutex_unlock(&mutex); | |
8624 | ||
8625 | return ret; | |
8626 | } | |
68318b8e | 8627 | |
052f1dc7 | 8628 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8629 | |
8630 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8631 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8632 | { |
2b01dfe3 PM |
8633 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8634 | struct task_group, css); | |
68318b8e SV |
8635 | } |
8636 | ||
8637 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8638 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8639 | { |
ec7dc8ac | 8640 | struct task_group *tg, *parent; |
68318b8e | 8641 | |
2b01dfe3 | 8642 | if (!cgrp->parent) { |
68318b8e | 8643 | /* This is early initialization for the top cgroup */ |
2b01dfe3 | 8644 | init_task_group.css.cgroup = cgrp; |
68318b8e SV |
8645 | return &init_task_group.css; |
8646 | } | |
8647 | ||
ec7dc8ac DG |
8648 | parent = cgroup_tg(cgrp->parent); |
8649 | tg = sched_create_group(parent); | |
68318b8e SV |
8650 | if (IS_ERR(tg)) |
8651 | return ERR_PTR(-ENOMEM); | |
8652 | ||
8653 | /* Bind the cgroup to task_group object we just created */ | |
2b01dfe3 | 8654 | tg->css.cgroup = cgrp; |
68318b8e SV |
8655 | |
8656 | return &tg->css; | |
8657 | } | |
8658 | ||
41a2d6cf IM |
8659 | static void |
8660 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8661 | { |
2b01dfe3 | 8662 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8663 | |
8664 | sched_destroy_group(tg); | |
8665 | } | |
8666 | ||
41a2d6cf IM |
8667 | static int |
8668 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8669 | struct task_struct *tsk) | |
68318b8e | 8670 | { |
b68aa230 PZ |
8671 | #ifdef CONFIG_RT_GROUP_SCHED |
8672 | /* Don't accept realtime tasks when there is no way for them to run */ | |
d0b27fa7 | 8673 | if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0) |
b68aa230 PZ |
8674 | return -EINVAL; |
8675 | #else | |
68318b8e SV |
8676 | /* We don't support RT-tasks being in separate groups */ |
8677 | if (tsk->sched_class != &fair_sched_class) | |
8678 | return -EINVAL; | |
b68aa230 | 8679 | #endif |
68318b8e SV |
8680 | |
8681 | return 0; | |
8682 | } | |
8683 | ||
8684 | static void | |
2b01dfe3 | 8685 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
68318b8e SV |
8686 | struct cgroup *old_cont, struct task_struct *tsk) |
8687 | { | |
8688 | sched_move_task(tsk); | |
8689 | } | |
8690 | ||
052f1dc7 | 8691 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8692 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8693 | u64 shareval) |
68318b8e | 8694 | { |
2b01dfe3 | 8695 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8696 | } |
8697 | ||
f4c753b7 | 8698 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8699 | { |
2b01dfe3 | 8700 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8701 | |
8702 | return (u64) tg->shares; | |
8703 | } | |
052f1dc7 | 8704 | #endif |
68318b8e | 8705 | |
052f1dc7 | 8706 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8707 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8708 | s64 val) |
6f505b16 | 8709 | { |
06ecb27c | 8710 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8711 | } |
8712 | ||
06ecb27c | 8713 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8714 | { |
06ecb27c | 8715 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8716 | } |
d0b27fa7 PZ |
8717 | |
8718 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8719 | u64 rt_period_us) | |
8720 | { | |
8721 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8722 | } | |
8723 | ||
8724 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8725 | { | |
8726 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8727 | } | |
052f1dc7 | 8728 | #endif |
6f505b16 | 8729 | |
fe5c7cc2 | 8730 | static struct cftype cpu_files[] = { |
052f1dc7 | 8731 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8732 | { |
8733 | .name = "shares", | |
f4c753b7 PM |
8734 | .read_u64 = cpu_shares_read_u64, |
8735 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8736 | }, |
052f1dc7 PZ |
8737 | #endif |
8738 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8739 | { |
9f0c1e56 | 8740 | .name = "rt_runtime_us", |
06ecb27c PM |
8741 | .read_s64 = cpu_rt_runtime_read, |
8742 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8743 | }, |
d0b27fa7 PZ |
8744 | { |
8745 | .name = "rt_period_us", | |
f4c753b7 PM |
8746 | .read_u64 = cpu_rt_period_read_uint, |
8747 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8748 | }, |
052f1dc7 | 8749 | #endif |
68318b8e SV |
8750 | }; |
8751 | ||
8752 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8753 | { | |
fe5c7cc2 | 8754 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8755 | } |
8756 | ||
8757 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8758 | .name = "cpu", |
8759 | .create = cpu_cgroup_create, | |
8760 | .destroy = cpu_cgroup_destroy, | |
8761 | .can_attach = cpu_cgroup_can_attach, | |
8762 | .attach = cpu_cgroup_attach, | |
8763 | .populate = cpu_cgroup_populate, | |
8764 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8765 | .early_init = 1, |
8766 | }; | |
8767 | ||
052f1dc7 | 8768 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8769 | |
8770 | #ifdef CONFIG_CGROUP_CPUACCT | |
8771 | ||
8772 | /* | |
8773 | * CPU accounting code for task groups. | |
8774 | * | |
8775 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8776 | * (balbir@in.ibm.com). | |
8777 | */ | |
8778 | ||
8779 | /* track cpu usage of a group of tasks */ | |
8780 | struct cpuacct { | |
8781 | struct cgroup_subsys_state css; | |
8782 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8783 | u64 *cpuusage; | |
8784 | }; | |
8785 | ||
8786 | struct cgroup_subsys cpuacct_subsys; | |
8787 | ||
8788 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8789 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8790 | { |
32cd756a | 8791 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8792 | struct cpuacct, css); |
8793 | } | |
8794 | ||
8795 | /* return cpu accounting group to which this task belongs */ | |
8796 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8797 | { | |
8798 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8799 | struct cpuacct, css); | |
8800 | } | |
8801 | ||
8802 | /* create a new cpu accounting group */ | |
8803 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8804 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8805 | { |
8806 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
8807 | ||
8808 | if (!ca) | |
8809 | return ERR_PTR(-ENOMEM); | |
8810 | ||
8811 | ca->cpuusage = alloc_percpu(u64); | |
8812 | if (!ca->cpuusage) { | |
8813 | kfree(ca); | |
8814 | return ERR_PTR(-ENOMEM); | |
8815 | } | |
8816 | ||
8817 | return &ca->css; | |
8818 | } | |
8819 | ||
8820 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8821 | static void |
32cd756a | 8822 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8823 | { |
32cd756a | 8824 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8825 | |
8826 | free_percpu(ca->cpuusage); | |
8827 | kfree(ca); | |
8828 | } | |
8829 | ||
8830 | /* return total cpu usage (in nanoseconds) of a group */ | |
32cd756a | 8831 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8832 | { |
32cd756a | 8833 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8834 | u64 totalcpuusage = 0; |
8835 | int i; | |
8836 | ||
8837 | for_each_possible_cpu(i) { | |
8838 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8839 | ||
8840 | /* | |
8841 | * Take rq->lock to make 64-bit addition safe on 32-bit | |
8842 | * platforms. | |
8843 | */ | |
8844 | spin_lock_irq(&cpu_rq(i)->lock); | |
8845 | totalcpuusage += *cpuusage; | |
8846 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8847 | } | |
8848 | ||
8849 | return totalcpuusage; | |
8850 | } | |
8851 | ||
0297b803 DG |
8852 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8853 | u64 reset) | |
8854 | { | |
8855 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8856 | int err = 0; | |
8857 | int i; | |
8858 | ||
8859 | if (reset) { | |
8860 | err = -EINVAL; | |
8861 | goto out; | |
8862 | } | |
8863 | ||
8864 | for_each_possible_cpu(i) { | |
8865 | u64 *cpuusage = percpu_ptr(ca->cpuusage, i); | |
8866 | ||
8867 | spin_lock_irq(&cpu_rq(i)->lock); | |
8868 | *cpuusage = 0; | |
8869 | spin_unlock_irq(&cpu_rq(i)->lock); | |
8870 | } | |
8871 | out: | |
8872 | return err; | |
8873 | } | |
8874 | ||
d842de87 SV |
8875 | static struct cftype files[] = { |
8876 | { | |
8877 | .name = "usage", | |
f4c753b7 PM |
8878 | .read_u64 = cpuusage_read, |
8879 | .write_u64 = cpuusage_write, | |
d842de87 SV |
8880 | }, |
8881 | }; | |
8882 | ||
32cd756a | 8883 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8884 | { |
32cd756a | 8885 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8886 | } |
8887 | ||
8888 | /* | |
8889 | * charge this task's execution time to its accounting group. | |
8890 | * | |
8891 | * called with rq->lock held. | |
8892 | */ | |
8893 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8894 | { | |
8895 | struct cpuacct *ca; | |
8896 | ||
8897 | if (!cpuacct_subsys.active) | |
8898 | return; | |
8899 | ||
8900 | ca = task_ca(tsk); | |
8901 | if (ca) { | |
8902 | u64 *cpuusage = percpu_ptr(ca->cpuusage, task_cpu(tsk)); | |
8903 | ||
8904 | *cpuusage += cputime; | |
8905 | } | |
8906 | } | |
8907 | ||
8908 | struct cgroup_subsys cpuacct_subsys = { | |
8909 | .name = "cpuacct", | |
8910 | .create = cpuacct_create, | |
8911 | .destroy = cpuacct_destroy, | |
8912 | .populate = cpuacct_populate, | |
8913 | .subsys_id = cpuacct_subsys_id, | |
8914 | }; | |
8915 | #endif /* CONFIG_CGROUP_CPUACCT */ |