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> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
969c7921 | 60 | #include <linux/stop_machine.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
5a0e3ad6 | 74 | #include <linux/slab.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 | 79 | #include "sched_cpupri.h" |
21aa9af0 | 80 | #include "workqueue_sched.h" |
6e0534f2 | 81 | |
a8d154b0 | 82 | #define CREATE_TRACE_POINTS |
ad8d75ff | 83 | #include <trace/events/sched.h> |
a8d154b0 | 84 | |
1da177e4 LT |
85 | /* |
86 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
87 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
88 | * and back. | |
89 | */ | |
90 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
91 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
92 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
93 | ||
94 | /* | |
95 | * 'User priority' is the nice value converted to something we | |
96 | * can work with better when scaling various scheduler parameters, | |
97 | * it's a [ 0 ... 39 ] range. | |
98 | */ | |
99 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
100 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
101 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
102 | ||
103 | /* | |
d7876a08 | 104 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 105 | */ |
d6322faf | 106 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 107 | |
6aa645ea IM |
108 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
109 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
110 | ||
1da177e4 LT |
111 | /* |
112 | * These are the 'tuning knobs' of the scheduler: | |
113 | * | |
a4ec24b4 | 114 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
115 | * Timeslices get refilled after they expire. |
116 | */ | |
1da177e4 | 117 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 118 | |
d0b27fa7 PZ |
119 | /* |
120 | * single value that denotes runtime == period, ie unlimited time. | |
121 | */ | |
122 | #define RUNTIME_INF ((u64)~0ULL) | |
123 | ||
e05606d3 IM |
124 | static inline int rt_policy(int policy) |
125 | { | |
3f33a7ce | 126 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
127 | return 1; |
128 | return 0; | |
129 | } | |
130 | ||
131 | static inline int task_has_rt_policy(struct task_struct *p) | |
132 | { | |
133 | return rt_policy(p->policy); | |
134 | } | |
135 | ||
1da177e4 | 136 | /* |
6aa645ea | 137 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 138 | */ |
6aa645ea IM |
139 | struct rt_prio_array { |
140 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
141 | struct list_head queue[MAX_RT_PRIO]; | |
142 | }; | |
143 | ||
d0b27fa7 | 144 | struct rt_bandwidth { |
ea736ed5 | 145 | /* nests inside the rq lock: */ |
0986b11b | 146 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
147 | ktime_t rt_period; |
148 | u64 rt_runtime; | |
149 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
150 | }; |
151 | ||
152 | static struct rt_bandwidth def_rt_bandwidth; | |
153 | ||
154 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
155 | ||
156 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
157 | { | |
158 | struct rt_bandwidth *rt_b = | |
159 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
160 | ktime_t now; | |
161 | int overrun; | |
162 | int idle = 0; | |
163 | ||
164 | for (;;) { | |
165 | now = hrtimer_cb_get_time(timer); | |
166 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
167 | ||
168 | if (!overrun) | |
169 | break; | |
170 | ||
171 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
172 | } | |
173 | ||
174 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
175 | } | |
176 | ||
177 | static | |
178 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
179 | { | |
180 | rt_b->rt_period = ns_to_ktime(period); | |
181 | rt_b->rt_runtime = runtime; | |
182 | ||
0986b11b | 183 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 184 | |
d0b27fa7 PZ |
185 | hrtimer_init(&rt_b->rt_period_timer, |
186 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
187 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
188 | } |
189 | ||
c8bfff6d KH |
190 | static inline int rt_bandwidth_enabled(void) |
191 | { | |
192 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
193 | } |
194 | ||
195 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
196 | { | |
197 | ktime_t now; | |
198 | ||
cac64d00 | 199 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
200 | return; |
201 | ||
202 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
203 | return; | |
204 | ||
0986b11b | 205 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 206 | for (;;) { |
7f1e2ca9 PZ |
207 | unsigned long delta; |
208 | ktime_t soft, hard; | |
209 | ||
d0b27fa7 PZ |
210 | if (hrtimer_active(&rt_b->rt_period_timer)) |
211 | break; | |
212 | ||
213 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
214 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
215 | |
216 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
217 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
218 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
219 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 220 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 221 | } |
0986b11b | 222 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
223 | } |
224 | ||
225 | #ifdef CONFIG_RT_GROUP_SCHED | |
226 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
227 | { | |
228 | hrtimer_cancel(&rt_b->rt_period_timer); | |
229 | } | |
230 | #endif | |
231 | ||
712555ee HC |
232 | /* |
233 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
234 | * detach_destroy_domains and partition_sched_domains. | |
235 | */ | |
236 | static DEFINE_MUTEX(sched_domains_mutex); | |
237 | ||
7c941438 | 238 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 239 | |
68318b8e SV |
240 | #include <linux/cgroup.h> |
241 | ||
29f59db3 SV |
242 | struct cfs_rq; |
243 | ||
6f505b16 PZ |
244 | static LIST_HEAD(task_groups); |
245 | ||
29f59db3 | 246 | /* task group related information */ |
4cf86d77 | 247 | struct task_group { |
68318b8e | 248 | struct cgroup_subsys_state css; |
6c415b92 | 249 | |
052f1dc7 | 250 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
251 | /* schedulable entities of this group on each cpu */ |
252 | struct sched_entity **se; | |
253 | /* runqueue "owned" by this group on each cpu */ | |
254 | struct cfs_rq **cfs_rq; | |
255 | unsigned long shares; | |
2069dd75 PZ |
256 | |
257 | atomic_t load_weight; | |
052f1dc7 PZ |
258 | #endif |
259 | ||
260 | #ifdef CONFIG_RT_GROUP_SCHED | |
261 | struct sched_rt_entity **rt_se; | |
262 | struct rt_rq **rt_rq; | |
263 | ||
d0b27fa7 | 264 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 265 | #endif |
6b2d7700 | 266 | |
ae8393e5 | 267 | struct rcu_head rcu; |
6f505b16 | 268 | struct list_head list; |
f473aa5e PZ |
269 | |
270 | struct task_group *parent; | |
271 | struct list_head siblings; | |
272 | struct list_head children; | |
29f59db3 SV |
273 | }; |
274 | ||
eff766a6 | 275 | #define root_task_group init_task_group |
6f505b16 | 276 | |
3d4b47b4 | 277 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 278 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 279 | |
e9036b36 CG |
280 | #ifdef CONFIG_FAIR_GROUP_SCHED |
281 | ||
052f1dc7 | 282 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 283 | |
cb4ad1ff | 284 | /* |
2e084786 LJ |
285 | * A weight of 0 or 1 can cause arithmetics problems. |
286 | * A weight of a cfs_rq is the sum of weights of which entities | |
287 | * are queued on this cfs_rq, so a weight of a entity should not be | |
288 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
289 | * (The default weight is 1024 - so there's no practical |
290 | * limitation from this.) | |
291 | */ | |
18d95a28 | 292 | #define MIN_SHARES 2 |
2e084786 | 293 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 294 | |
052f1dc7 PZ |
295 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
296 | #endif | |
297 | ||
29f59db3 | 298 | /* Default task group. |
3a252015 | 299 | * Every task in system belong to this group at bootup. |
29f59db3 | 300 | */ |
434d53b0 | 301 | struct task_group init_task_group; |
29f59db3 | 302 | |
7c941438 | 303 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 304 | |
6aa645ea IM |
305 | /* CFS-related fields in a runqueue */ |
306 | struct cfs_rq { | |
307 | struct load_weight load; | |
308 | unsigned long nr_running; | |
309 | ||
6aa645ea | 310 | u64 exec_clock; |
e9acbff6 | 311 | u64 min_vruntime; |
6aa645ea IM |
312 | |
313 | struct rb_root tasks_timeline; | |
314 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
315 | |
316 | struct list_head tasks; | |
317 | struct list_head *balance_iterator; | |
318 | ||
319 | /* | |
320 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
321 | * It is set to NULL otherwise (i.e when none are currently running). |
322 | */ | |
4793241b | 323 | struct sched_entity *curr, *next, *last; |
ddc97297 | 324 | |
5ac5c4d6 | 325 | unsigned int nr_spread_over; |
ddc97297 | 326 | |
62160e3f | 327 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
328 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
329 | ||
41a2d6cf IM |
330 | /* |
331 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
332 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
333 | * (like users, containers etc.) | |
334 | * | |
335 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
336 | * list is used during load balance. | |
337 | */ | |
3d4b47b4 | 338 | int on_list; |
41a2d6cf IM |
339 | struct list_head leaf_cfs_rq_list; |
340 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
341 | |
342 | #ifdef CONFIG_SMP | |
c09595f6 | 343 | /* |
c8cba857 | 344 | * the part of load.weight contributed by tasks |
c09595f6 | 345 | */ |
c8cba857 | 346 | unsigned long task_weight; |
c09595f6 | 347 | |
c8cba857 PZ |
348 | /* |
349 | * h_load = weight * f(tg) | |
350 | * | |
351 | * Where f(tg) is the recursive weight fraction assigned to | |
352 | * this group. | |
353 | */ | |
354 | unsigned long h_load; | |
c09595f6 | 355 | |
3b3d190e PT |
356 | /* |
357 | * Maintaining per-cpu shares distribution for group scheduling | |
358 | * | |
359 | * load_stamp is the last time we updated the load average | |
360 | * load_last is the last time we updated the load average and saw load | |
361 | * load_unacc_exec_time is currently unaccounted execution time | |
362 | */ | |
2069dd75 PZ |
363 | u64 load_avg; |
364 | u64 load_period; | |
3b3d190e | 365 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 366 | |
2069dd75 | 367 | unsigned long load_contribution; |
c09595f6 | 368 | #endif |
6aa645ea IM |
369 | #endif |
370 | }; | |
1da177e4 | 371 | |
6aa645ea IM |
372 | /* Real-Time classes' related field in a runqueue: */ |
373 | struct rt_rq { | |
374 | struct rt_prio_array active; | |
63489e45 | 375 | unsigned long rt_nr_running; |
052f1dc7 | 376 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
377 | struct { |
378 | int curr; /* highest queued rt task prio */ | |
398a153b | 379 | #ifdef CONFIG_SMP |
e864c499 | 380 | int next; /* next highest */ |
398a153b | 381 | #endif |
e864c499 | 382 | } highest_prio; |
6f505b16 | 383 | #endif |
fa85ae24 | 384 | #ifdef CONFIG_SMP |
73fe6aae | 385 | unsigned long rt_nr_migratory; |
a1ba4d8b | 386 | unsigned long rt_nr_total; |
a22d7fc1 | 387 | int overloaded; |
917b627d | 388 | struct plist_head pushable_tasks; |
fa85ae24 | 389 | #endif |
6f505b16 | 390 | int rt_throttled; |
fa85ae24 | 391 | u64 rt_time; |
ac086bc2 | 392 | u64 rt_runtime; |
ea736ed5 | 393 | /* Nests inside the rq lock: */ |
0986b11b | 394 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 395 | |
052f1dc7 | 396 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
397 | unsigned long rt_nr_boosted; |
398 | ||
6f505b16 PZ |
399 | struct rq *rq; |
400 | struct list_head leaf_rt_rq_list; | |
401 | struct task_group *tg; | |
6f505b16 | 402 | #endif |
6aa645ea IM |
403 | }; |
404 | ||
57d885fe GH |
405 | #ifdef CONFIG_SMP |
406 | ||
407 | /* | |
408 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
409 | * variables. Each exclusive cpuset essentially defines an island domain by |
410 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
411 | * exclusive cpuset is created, we also create and attach a new root-domain |
412 | * object. | |
413 | * | |
57d885fe GH |
414 | */ |
415 | struct root_domain { | |
416 | atomic_t refcount; | |
c6c4927b RR |
417 | cpumask_var_t span; |
418 | cpumask_var_t online; | |
637f5085 | 419 | |
0eab9146 | 420 | /* |
637f5085 GH |
421 | * The "RT overload" flag: it gets set if a CPU has more than |
422 | * one runnable RT task. | |
423 | */ | |
c6c4927b | 424 | cpumask_var_t rto_mask; |
0eab9146 | 425 | atomic_t rto_count; |
6e0534f2 | 426 | struct cpupri cpupri; |
57d885fe GH |
427 | }; |
428 | ||
dc938520 GH |
429 | /* |
430 | * By default the system creates a single root-domain with all cpus as | |
431 | * members (mimicking the global state we have today). | |
432 | */ | |
57d885fe GH |
433 | static struct root_domain def_root_domain; |
434 | ||
ed2d372c | 435 | #endif /* CONFIG_SMP */ |
57d885fe | 436 | |
1da177e4 LT |
437 | /* |
438 | * This is the main, per-CPU runqueue data structure. | |
439 | * | |
440 | * Locking rule: those places that want to lock multiple runqueues | |
441 | * (such as the load balancing or the thread migration code), lock | |
442 | * acquire operations must be ordered by ascending &runqueue. | |
443 | */ | |
70b97a7f | 444 | struct rq { |
d8016491 | 445 | /* runqueue lock: */ |
05fa785c | 446 | raw_spinlock_t lock; |
1da177e4 LT |
447 | |
448 | /* | |
449 | * nr_running and cpu_load should be in the same cacheline because | |
450 | * remote CPUs use both these fields when doing load calculation. | |
451 | */ | |
452 | unsigned long nr_running; | |
6aa645ea IM |
453 | #define CPU_LOAD_IDX_MAX 5 |
454 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 455 | unsigned long last_load_update_tick; |
46cb4b7c | 456 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 457 | u64 nohz_stamp; |
83cd4fe2 | 458 | unsigned char nohz_balance_kick; |
46cb4b7c | 459 | #endif |
a64692a3 MG |
460 | unsigned int skip_clock_update; |
461 | ||
d8016491 IM |
462 | /* capture load from *all* tasks on this cpu: */ |
463 | struct load_weight load; | |
6aa645ea IM |
464 | unsigned long nr_load_updates; |
465 | u64 nr_switches; | |
466 | ||
467 | struct cfs_rq cfs; | |
6f505b16 | 468 | struct rt_rq rt; |
6f505b16 | 469 | |
6aa645ea | 470 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
471 | /* list of leaf cfs_rq on this cpu: */ |
472 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
473 | #endif |
474 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 475 | struct list_head leaf_rt_rq_list; |
1da177e4 | 476 | #endif |
1da177e4 LT |
477 | |
478 | /* | |
479 | * This is part of a global counter where only the total sum | |
480 | * over all CPUs matters. A task can increase this counter on | |
481 | * one CPU and if it got migrated afterwards it may decrease | |
482 | * it on another CPU. Always updated under the runqueue lock: | |
483 | */ | |
484 | unsigned long nr_uninterruptible; | |
485 | ||
34f971f6 | 486 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 487 | unsigned long next_balance; |
1da177e4 | 488 | struct mm_struct *prev_mm; |
6aa645ea | 489 | |
3e51f33f | 490 | u64 clock; |
305e6835 | 491 | u64 clock_task; |
6aa645ea | 492 | |
1da177e4 LT |
493 | atomic_t nr_iowait; |
494 | ||
495 | #ifdef CONFIG_SMP | |
0eab9146 | 496 | struct root_domain *rd; |
1da177e4 LT |
497 | struct sched_domain *sd; |
498 | ||
e51fd5e2 PZ |
499 | unsigned long cpu_power; |
500 | ||
a0a522ce | 501 | unsigned char idle_at_tick; |
1da177e4 | 502 | /* For active balancing */ |
3f029d3c | 503 | int post_schedule; |
1da177e4 LT |
504 | int active_balance; |
505 | int push_cpu; | |
969c7921 | 506 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
507 | /* cpu of this runqueue: */ |
508 | int cpu; | |
1f11eb6a | 509 | int online; |
1da177e4 | 510 | |
a8a51d5e | 511 | unsigned long avg_load_per_task; |
1da177e4 | 512 | |
e9e9250b PZ |
513 | u64 rt_avg; |
514 | u64 age_stamp; | |
1b9508f6 MG |
515 | u64 idle_stamp; |
516 | u64 avg_idle; | |
1da177e4 LT |
517 | #endif |
518 | ||
aa483808 VP |
519 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
520 | u64 prev_irq_time; | |
521 | #endif | |
522 | ||
dce48a84 TG |
523 | /* calc_load related fields */ |
524 | unsigned long calc_load_update; | |
525 | long calc_load_active; | |
526 | ||
8f4d37ec | 527 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
528 | #ifdef CONFIG_SMP |
529 | int hrtick_csd_pending; | |
530 | struct call_single_data hrtick_csd; | |
531 | #endif | |
8f4d37ec PZ |
532 | struct hrtimer hrtick_timer; |
533 | #endif | |
534 | ||
1da177e4 LT |
535 | #ifdef CONFIG_SCHEDSTATS |
536 | /* latency stats */ | |
537 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
538 | unsigned long long rq_cpu_time; |
539 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
540 | |
541 | /* sys_sched_yield() stats */ | |
480b9434 | 542 | unsigned int yld_count; |
1da177e4 LT |
543 | |
544 | /* schedule() stats */ | |
480b9434 KC |
545 | unsigned int sched_switch; |
546 | unsigned int sched_count; | |
547 | unsigned int sched_goidle; | |
1da177e4 LT |
548 | |
549 | /* try_to_wake_up() stats */ | |
480b9434 KC |
550 | unsigned int ttwu_count; |
551 | unsigned int ttwu_local; | |
b8efb561 IM |
552 | |
553 | /* BKL stats */ | |
480b9434 | 554 | unsigned int bkl_count; |
1da177e4 LT |
555 | #endif |
556 | }; | |
557 | ||
f34e3b61 | 558 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 559 | |
7d478721 PZ |
560 | static inline |
561 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 562 | { |
7d478721 | 563 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
a64692a3 MG |
564 | |
565 | /* | |
566 | * A queue event has occurred, and we're going to schedule. In | |
567 | * this case, we can save a useless back to back clock update. | |
568 | */ | |
569 | if (test_tsk_need_resched(p)) | |
570 | rq->skip_clock_update = 1; | |
dd41f596 IM |
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 | ||
497f0ab3 | 582 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
583 | rcu_dereference_check((p), \ |
584 | rcu_read_lock_sched_held() || \ | |
585 | lockdep_is_held(&sched_domains_mutex)) | |
586 | ||
674311d5 NP |
587 | /* |
588 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 589 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
590 | * |
591 | * The domain tree of any CPU may only be accessed from within | |
592 | * preempt-disabled sections. | |
593 | */ | |
48f24c4d | 594 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 595 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
596 | |
597 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
598 | #define this_rq() (&__get_cpu_var(runqueues)) | |
599 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
600 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 601 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 602 | |
dc61b1d6 PZ |
603 | #ifdef CONFIG_CGROUP_SCHED |
604 | ||
605 | /* | |
606 | * Return the group to which this tasks belongs. | |
607 | * | |
608 | * We use task_subsys_state_check() and extend the RCU verification | |
609 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
610 | * holds that lock for each task it moves into the cgroup. Therefore | |
611 | * by holding that lock, we pin the task to the current cgroup. | |
612 | */ | |
613 | static inline struct task_group *task_group(struct task_struct *p) | |
614 | { | |
615 | struct cgroup_subsys_state *css; | |
616 | ||
617 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
618 | lockdep_is_held(&task_rq(p)->lock)); | |
619 | return container_of(css, struct task_group, css); | |
620 | } | |
621 | ||
622 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
623 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
624 | { | |
625 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
626 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
627 | p->se.parent = task_group(p)->se[cpu]; | |
628 | #endif | |
629 | ||
630 | #ifdef CONFIG_RT_GROUP_SCHED | |
631 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
632 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
633 | #endif | |
634 | } | |
635 | ||
636 | #else /* CONFIG_CGROUP_SCHED */ | |
637 | ||
638 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
639 | static inline struct task_group *task_group(struct task_struct *p) | |
640 | { | |
641 | return NULL; | |
642 | } | |
643 | ||
644 | #endif /* CONFIG_CGROUP_SCHED */ | |
645 | ||
305e6835 | 646 | static u64 irq_time_cpu(int cpu); |
aa483808 | 647 | static void sched_irq_time_avg_update(struct rq *rq, u64 irq_time); |
305e6835 | 648 | |
aa9c4c0f | 649 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 650 | { |
305e6835 VP |
651 | if (!rq->skip_clock_update) { |
652 | int cpu = cpu_of(rq); | |
653 | u64 irq_time; | |
654 | ||
655 | rq->clock = sched_clock_cpu(cpu); | |
656 | irq_time = irq_time_cpu(cpu); | |
657 | if (rq->clock - irq_time > rq->clock_task) | |
658 | rq->clock_task = rq->clock - irq_time; | |
aa483808 VP |
659 | |
660 | sched_irq_time_avg_update(rq, irq_time); | |
305e6835 | 661 | } |
3e51f33f PZ |
662 | } |
663 | ||
bf5c91ba IM |
664 | /* |
665 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
666 | */ | |
667 | #ifdef CONFIG_SCHED_DEBUG | |
668 | # define const_debug __read_mostly | |
669 | #else | |
670 | # define const_debug static const | |
671 | #endif | |
672 | ||
017730c1 IM |
673 | /** |
674 | * runqueue_is_locked | |
e17b38bf | 675 | * @cpu: the processor in question. |
017730c1 IM |
676 | * |
677 | * Returns true if the current cpu runqueue is locked. | |
678 | * This interface allows printk to be called with the runqueue lock | |
679 | * held and know whether or not it is OK to wake up the klogd. | |
680 | */ | |
89f19f04 | 681 | int runqueue_is_locked(int cpu) |
017730c1 | 682 | { |
05fa785c | 683 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
684 | } |
685 | ||
bf5c91ba IM |
686 | /* |
687 | * Debugging: various feature bits | |
688 | */ | |
f00b45c1 PZ |
689 | |
690 | #define SCHED_FEAT(name, enabled) \ | |
691 | __SCHED_FEAT_##name , | |
692 | ||
bf5c91ba | 693 | enum { |
f00b45c1 | 694 | #include "sched_features.h" |
bf5c91ba IM |
695 | }; |
696 | ||
f00b45c1 PZ |
697 | #undef SCHED_FEAT |
698 | ||
699 | #define SCHED_FEAT(name, enabled) \ | |
700 | (1UL << __SCHED_FEAT_##name) * enabled | | |
701 | ||
bf5c91ba | 702 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
703 | #include "sched_features.h" |
704 | 0; | |
705 | ||
706 | #undef SCHED_FEAT | |
707 | ||
708 | #ifdef CONFIG_SCHED_DEBUG | |
709 | #define SCHED_FEAT(name, enabled) \ | |
710 | #name , | |
711 | ||
983ed7a6 | 712 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
713 | #include "sched_features.h" |
714 | NULL | |
715 | }; | |
716 | ||
717 | #undef SCHED_FEAT | |
718 | ||
34f3a814 | 719 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 720 | { |
f00b45c1 PZ |
721 | int i; |
722 | ||
723 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
724 | if (!(sysctl_sched_features & (1UL << i))) |
725 | seq_puts(m, "NO_"); | |
726 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 727 | } |
34f3a814 | 728 | seq_puts(m, "\n"); |
f00b45c1 | 729 | |
34f3a814 | 730 | return 0; |
f00b45c1 PZ |
731 | } |
732 | ||
733 | static ssize_t | |
734 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
735 | size_t cnt, loff_t *ppos) | |
736 | { | |
737 | char buf[64]; | |
7740191c | 738 | char *cmp; |
f00b45c1 PZ |
739 | int neg = 0; |
740 | int i; | |
741 | ||
742 | if (cnt > 63) | |
743 | cnt = 63; | |
744 | ||
745 | if (copy_from_user(&buf, ubuf, cnt)) | |
746 | return -EFAULT; | |
747 | ||
748 | buf[cnt] = 0; | |
7740191c | 749 | cmp = strstrip(buf); |
f00b45c1 | 750 | |
c24b7c52 | 751 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
752 | neg = 1; |
753 | cmp += 3; | |
754 | } | |
755 | ||
756 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 757 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
758 | if (neg) |
759 | sysctl_sched_features &= ~(1UL << i); | |
760 | else | |
761 | sysctl_sched_features |= (1UL << i); | |
762 | break; | |
763 | } | |
764 | } | |
765 | ||
766 | if (!sched_feat_names[i]) | |
767 | return -EINVAL; | |
768 | ||
42994724 | 769 | *ppos += cnt; |
f00b45c1 PZ |
770 | |
771 | return cnt; | |
772 | } | |
773 | ||
34f3a814 LZ |
774 | static int sched_feat_open(struct inode *inode, struct file *filp) |
775 | { | |
776 | return single_open(filp, sched_feat_show, NULL); | |
777 | } | |
778 | ||
828c0950 | 779 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
780 | .open = sched_feat_open, |
781 | .write = sched_feat_write, | |
782 | .read = seq_read, | |
783 | .llseek = seq_lseek, | |
784 | .release = single_release, | |
f00b45c1 PZ |
785 | }; |
786 | ||
787 | static __init int sched_init_debug(void) | |
788 | { | |
f00b45c1 PZ |
789 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
790 | &sched_feat_fops); | |
791 | ||
792 | return 0; | |
793 | } | |
794 | late_initcall(sched_init_debug); | |
795 | ||
796 | #endif | |
797 | ||
798 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 799 | |
b82d9fdd PZ |
800 | /* |
801 | * Number of tasks to iterate in a single balance run. | |
802 | * Limited because this is done with IRQs disabled. | |
803 | */ | |
804 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
805 | ||
e9e9250b PZ |
806 | /* |
807 | * period over which we average the RT time consumption, measured | |
808 | * in ms. | |
809 | * | |
810 | * default: 1s | |
811 | */ | |
812 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
813 | ||
fa85ae24 | 814 | /* |
9f0c1e56 | 815 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
816 | * default: 1s |
817 | */ | |
9f0c1e56 | 818 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 819 | |
6892b75e IM |
820 | static __read_mostly int scheduler_running; |
821 | ||
9f0c1e56 PZ |
822 | /* |
823 | * part of the period that we allow rt tasks to run in us. | |
824 | * default: 0.95s | |
825 | */ | |
826 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 827 | |
d0b27fa7 PZ |
828 | static inline u64 global_rt_period(void) |
829 | { | |
830 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
831 | } | |
832 | ||
833 | static inline u64 global_rt_runtime(void) | |
834 | { | |
e26873bb | 835 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
836 | return RUNTIME_INF; |
837 | ||
838 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
839 | } | |
fa85ae24 | 840 | |
1da177e4 | 841 | #ifndef prepare_arch_switch |
4866cde0 NP |
842 | # define prepare_arch_switch(next) do { } while (0) |
843 | #endif | |
844 | #ifndef finish_arch_switch | |
845 | # define finish_arch_switch(prev) do { } while (0) | |
846 | #endif | |
847 | ||
051a1d1a DA |
848 | static inline int task_current(struct rq *rq, struct task_struct *p) |
849 | { | |
850 | return rq->curr == p; | |
851 | } | |
852 | ||
4866cde0 | 853 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 854 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 855 | { |
051a1d1a | 856 | return task_current(rq, p); |
4866cde0 NP |
857 | } |
858 | ||
70b97a7f | 859 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
860 | { |
861 | } | |
862 | ||
70b97a7f | 863 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 864 | { |
da04c035 IM |
865 | #ifdef CONFIG_DEBUG_SPINLOCK |
866 | /* this is a valid case when another task releases the spinlock */ | |
867 | rq->lock.owner = current; | |
868 | #endif | |
8a25d5de IM |
869 | /* |
870 | * If we are tracking spinlock dependencies then we have to | |
871 | * fix up the runqueue lock - which gets 'carried over' from | |
872 | * prev into current: | |
873 | */ | |
874 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
875 | ||
05fa785c | 876 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
877 | } |
878 | ||
879 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 880 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
881 | { |
882 | #ifdef CONFIG_SMP | |
883 | return p->oncpu; | |
884 | #else | |
051a1d1a | 885 | return task_current(rq, p); |
4866cde0 NP |
886 | #endif |
887 | } | |
888 | ||
70b97a7f | 889 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
890 | { |
891 | #ifdef CONFIG_SMP | |
892 | /* | |
893 | * We can optimise this out completely for !SMP, because the | |
894 | * SMP rebalancing from interrupt is the only thing that cares | |
895 | * here. | |
896 | */ | |
897 | next->oncpu = 1; | |
898 | #endif | |
899 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 900 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 901 | #else |
05fa785c | 902 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
903 | #endif |
904 | } | |
905 | ||
70b97a7f | 906 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
907 | { |
908 | #ifdef CONFIG_SMP | |
909 | /* | |
910 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
911 | * We must ensure this doesn't happen until the switch is completely | |
912 | * finished. | |
913 | */ | |
914 | smp_wmb(); | |
915 | prev->oncpu = 0; | |
916 | #endif | |
917 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
918 | local_irq_enable(); | |
1da177e4 | 919 | #endif |
4866cde0 NP |
920 | } |
921 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 922 | |
0970d299 | 923 | /* |
65cc8e48 PZ |
924 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
925 | * against ttwu(). | |
0970d299 PZ |
926 | */ |
927 | static inline int task_is_waking(struct task_struct *p) | |
928 | { | |
0017d735 | 929 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
930 | } |
931 | ||
b29739f9 IM |
932 | /* |
933 | * __task_rq_lock - lock the runqueue a given task resides on. | |
934 | * Must be called interrupts disabled. | |
935 | */ | |
70b97a7f | 936 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
937 | __acquires(rq->lock) |
938 | { | |
0970d299 PZ |
939 | struct rq *rq; |
940 | ||
3a5c359a | 941 | for (;;) { |
0970d299 | 942 | rq = task_rq(p); |
05fa785c | 943 | raw_spin_lock(&rq->lock); |
65cc8e48 | 944 | if (likely(rq == task_rq(p))) |
3a5c359a | 945 | return rq; |
05fa785c | 946 | raw_spin_unlock(&rq->lock); |
b29739f9 | 947 | } |
b29739f9 IM |
948 | } |
949 | ||
1da177e4 LT |
950 | /* |
951 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 952 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
953 | * explicitly disabling preemption. |
954 | */ | |
70b97a7f | 955 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
956 | __acquires(rq->lock) |
957 | { | |
70b97a7f | 958 | struct rq *rq; |
1da177e4 | 959 | |
3a5c359a AK |
960 | for (;;) { |
961 | local_irq_save(*flags); | |
962 | rq = task_rq(p); | |
05fa785c | 963 | raw_spin_lock(&rq->lock); |
65cc8e48 | 964 | if (likely(rq == task_rq(p))) |
3a5c359a | 965 | return rq; |
05fa785c | 966 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 967 | } |
1da177e4 LT |
968 | } |
969 | ||
a9957449 | 970 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
971 | __releases(rq->lock) |
972 | { | |
05fa785c | 973 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
974 | } |
975 | ||
70b97a7f | 976 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
977 | __releases(rq->lock) |
978 | { | |
05fa785c | 979 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
980 | } |
981 | ||
1da177e4 | 982 | /* |
cc2a73b5 | 983 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 984 | */ |
a9957449 | 985 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
986 | __acquires(rq->lock) |
987 | { | |
70b97a7f | 988 | struct rq *rq; |
1da177e4 LT |
989 | |
990 | local_irq_disable(); | |
991 | rq = this_rq(); | |
05fa785c | 992 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
993 | |
994 | return rq; | |
995 | } | |
996 | ||
8f4d37ec PZ |
997 | #ifdef CONFIG_SCHED_HRTICK |
998 | /* | |
999 | * Use HR-timers to deliver accurate preemption points. | |
1000 | * | |
1001 | * Its all a bit involved since we cannot program an hrt while holding the | |
1002 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1003 | * reschedule event. | |
1004 | * | |
1005 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1006 | * rq->lock. | |
1007 | */ | |
8f4d37ec PZ |
1008 | |
1009 | /* | |
1010 | * Use hrtick when: | |
1011 | * - enabled by features | |
1012 | * - hrtimer is actually high res | |
1013 | */ | |
1014 | static inline int hrtick_enabled(struct rq *rq) | |
1015 | { | |
1016 | if (!sched_feat(HRTICK)) | |
1017 | return 0; | |
ba42059f | 1018 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1019 | return 0; |
8f4d37ec PZ |
1020 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1021 | } | |
1022 | ||
8f4d37ec PZ |
1023 | static void hrtick_clear(struct rq *rq) |
1024 | { | |
1025 | if (hrtimer_active(&rq->hrtick_timer)) | |
1026 | hrtimer_cancel(&rq->hrtick_timer); | |
1027 | } | |
1028 | ||
8f4d37ec PZ |
1029 | /* |
1030 | * High-resolution timer tick. | |
1031 | * Runs from hardirq context with interrupts disabled. | |
1032 | */ | |
1033 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1034 | { | |
1035 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1036 | ||
1037 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1038 | ||
05fa785c | 1039 | raw_spin_lock(&rq->lock); |
3e51f33f | 1040 | update_rq_clock(rq); |
8f4d37ec | 1041 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1042 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1043 | |
1044 | return HRTIMER_NORESTART; | |
1045 | } | |
1046 | ||
95e904c7 | 1047 | #ifdef CONFIG_SMP |
31656519 PZ |
1048 | /* |
1049 | * called from hardirq (IPI) context | |
1050 | */ | |
1051 | static void __hrtick_start(void *arg) | |
b328ca18 | 1052 | { |
31656519 | 1053 | struct rq *rq = arg; |
b328ca18 | 1054 | |
05fa785c | 1055 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1056 | hrtimer_restart(&rq->hrtick_timer); |
1057 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1058 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1059 | } |
1060 | ||
31656519 PZ |
1061 | /* |
1062 | * Called to set the hrtick timer state. | |
1063 | * | |
1064 | * called with rq->lock held and irqs disabled | |
1065 | */ | |
1066 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1067 | { |
31656519 PZ |
1068 | struct hrtimer *timer = &rq->hrtick_timer; |
1069 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1070 | |
cc584b21 | 1071 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1072 | |
1073 | if (rq == this_rq()) { | |
1074 | hrtimer_restart(timer); | |
1075 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1076 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1077 | rq->hrtick_csd_pending = 1; |
1078 | } | |
b328ca18 PZ |
1079 | } |
1080 | ||
1081 | static int | |
1082 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1083 | { | |
1084 | int cpu = (int)(long)hcpu; | |
1085 | ||
1086 | switch (action) { | |
1087 | case CPU_UP_CANCELED: | |
1088 | case CPU_UP_CANCELED_FROZEN: | |
1089 | case CPU_DOWN_PREPARE: | |
1090 | case CPU_DOWN_PREPARE_FROZEN: | |
1091 | case CPU_DEAD: | |
1092 | case CPU_DEAD_FROZEN: | |
31656519 | 1093 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1094 | return NOTIFY_OK; |
1095 | } | |
1096 | ||
1097 | return NOTIFY_DONE; | |
1098 | } | |
1099 | ||
fa748203 | 1100 | static __init void init_hrtick(void) |
b328ca18 PZ |
1101 | { |
1102 | hotcpu_notifier(hotplug_hrtick, 0); | |
1103 | } | |
31656519 PZ |
1104 | #else |
1105 | /* | |
1106 | * Called to set the hrtick timer state. | |
1107 | * | |
1108 | * called with rq->lock held and irqs disabled | |
1109 | */ | |
1110 | static void hrtick_start(struct rq *rq, u64 delay) | |
1111 | { | |
7f1e2ca9 | 1112 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1113 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1114 | } |
b328ca18 | 1115 | |
006c75f1 | 1116 | static inline void init_hrtick(void) |
8f4d37ec | 1117 | { |
8f4d37ec | 1118 | } |
31656519 | 1119 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1120 | |
31656519 | 1121 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1122 | { |
31656519 PZ |
1123 | #ifdef CONFIG_SMP |
1124 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1125 | |
31656519 PZ |
1126 | rq->hrtick_csd.flags = 0; |
1127 | rq->hrtick_csd.func = __hrtick_start; | |
1128 | rq->hrtick_csd.info = rq; | |
1129 | #endif | |
8f4d37ec | 1130 | |
31656519 PZ |
1131 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1132 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1133 | } |
006c75f1 | 1134 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1135 | static inline void hrtick_clear(struct rq *rq) |
1136 | { | |
1137 | } | |
1138 | ||
8f4d37ec PZ |
1139 | static inline void init_rq_hrtick(struct rq *rq) |
1140 | { | |
1141 | } | |
1142 | ||
b328ca18 PZ |
1143 | static inline void init_hrtick(void) |
1144 | { | |
1145 | } | |
006c75f1 | 1146 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1147 | |
c24d20db IM |
1148 | /* |
1149 | * resched_task - mark a task 'to be rescheduled now'. | |
1150 | * | |
1151 | * On UP this means the setting of the need_resched flag, on SMP it | |
1152 | * might also involve a cross-CPU call to trigger the scheduler on | |
1153 | * the target CPU. | |
1154 | */ | |
1155 | #ifdef CONFIG_SMP | |
1156 | ||
1157 | #ifndef tsk_is_polling | |
1158 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1159 | #endif | |
1160 | ||
31656519 | 1161 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1162 | { |
1163 | int cpu; | |
1164 | ||
05fa785c | 1165 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1166 | |
5ed0cec0 | 1167 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1168 | return; |
1169 | ||
5ed0cec0 | 1170 | set_tsk_need_resched(p); |
c24d20db IM |
1171 | |
1172 | cpu = task_cpu(p); | |
1173 | if (cpu == smp_processor_id()) | |
1174 | return; | |
1175 | ||
1176 | /* NEED_RESCHED must be visible before we test polling */ | |
1177 | smp_mb(); | |
1178 | if (!tsk_is_polling(p)) | |
1179 | smp_send_reschedule(cpu); | |
1180 | } | |
1181 | ||
1182 | static void resched_cpu(int cpu) | |
1183 | { | |
1184 | struct rq *rq = cpu_rq(cpu); | |
1185 | unsigned long flags; | |
1186 | ||
05fa785c | 1187 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1188 | return; |
1189 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1190 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1191 | } |
06d8308c TG |
1192 | |
1193 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1194 | /* |
1195 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1196 | * from an idle cpu. This is good for power-savings. | |
1197 | * | |
1198 | * We don't do similar optimization for completely idle system, as | |
1199 | * selecting an idle cpu will add more delays to the timers than intended | |
1200 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1201 | */ | |
1202 | int get_nohz_timer_target(void) | |
1203 | { | |
1204 | int cpu = smp_processor_id(); | |
1205 | int i; | |
1206 | struct sched_domain *sd; | |
1207 | ||
1208 | for_each_domain(cpu, sd) { | |
1209 | for_each_cpu(i, sched_domain_span(sd)) | |
1210 | if (!idle_cpu(i)) | |
1211 | return i; | |
1212 | } | |
1213 | return cpu; | |
1214 | } | |
06d8308c TG |
1215 | /* |
1216 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1217 | * idle CPU then this timer might expire before the next timer event | |
1218 | * which is scheduled to wake up that CPU. In case of a completely | |
1219 | * idle system the next event might even be infinite time into the | |
1220 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1221 | * leaves the inner idle loop so the newly added timer is taken into | |
1222 | * account when the CPU goes back to idle and evaluates the timer | |
1223 | * wheel for the next timer event. | |
1224 | */ | |
1225 | void wake_up_idle_cpu(int cpu) | |
1226 | { | |
1227 | struct rq *rq = cpu_rq(cpu); | |
1228 | ||
1229 | if (cpu == smp_processor_id()) | |
1230 | return; | |
1231 | ||
1232 | /* | |
1233 | * This is safe, as this function is called with the timer | |
1234 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1235 | * to idle and has not yet set rq->curr to idle then it will | |
1236 | * be serialized on the timer wheel base lock and take the new | |
1237 | * timer into account automatically. | |
1238 | */ | |
1239 | if (rq->curr != rq->idle) | |
1240 | return; | |
1241 | ||
1242 | /* | |
1243 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1244 | * lockless. The worst case is that the other CPU runs the | |
1245 | * idle task through an additional NOOP schedule() | |
1246 | */ | |
5ed0cec0 | 1247 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1248 | |
1249 | /* NEED_RESCHED must be visible before we test polling */ | |
1250 | smp_mb(); | |
1251 | if (!tsk_is_polling(rq->idle)) | |
1252 | smp_send_reschedule(cpu); | |
1253 | } | |
39c0cbe2 | 1254 | |
6d6bc0ad | 1255 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1256 | |
e9e9250b PZ |
1257 | static u64 sched_avg_period(void) |
1258 | { | |
1259 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1260 | } | |
1261 | ||
1262 | static void sched_avg_update(struct rq *rq) | |
1263 | { | |
1264 | s64 period = sched_avg_period(); | |
1265 | ||
1266 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1267 | /* |
1268 | * Inline assembly required to prevent the compiler | |
1269 | * optimising this loop into a divmod call. | |
1270 | * See __iter_div_u64_rem() for another example of this. | |
1271 | */ | |
1272 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1273 | rq->age_stamp += period; |
1274 | rq->rt_avg /= 2; | |
1275 | } | |
1276 | } | |
1277 | ||
1278 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1279 | { | |
1280 | rq->rt_avg += rt_delta; | |
1281 | sched_avg_update(rq); | |
1282 | } | |
1283 | ||
6d6bc0ad | 1284 | #else /* !CONFIG_SMP */ |
31656519 | 1285 | static void resched_task(struct task_struct *p) |
c24d20db | 1286 | { |
05fa785c | 1287 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1288 | set_tsk_need_resched(p); |
c24d20db | 1289 | } |
e9e9250b PZ |
1290 | |
1291 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1292 | { | |
1293 | } | |
da2b71ed SS |
1294 | |
1295 | static void sched_avg_update(struct rq *rq) | |
1296 | { | |
1297 | } | |
6d6bc0ad | 1298 | #endif /* CONFIG_SMP */ |
c24d20db | 1299 | |
45bf76df IM |
1300 | #if BITS_PER_LONG == 32 |
1301 | # define WMULT_CONST (~0UL) | |
1302 | #else | |
1303 | # define WMULT_CONST (1UL << 32) | |
1304 | #endif | |
1305 | ||
1306 | #define WMULT_SHIFT 32 | |
1307 | ||
194081eb IM |
1308 | /* |
1309 | * Shift right and round: | |
1310 | */ | |
cf2ab469 | 1311 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1312 | |
a7be37ac PZ |
1313 | /* |
1314 | * delta *= weight / lw | |
1315 | */ | |
cb1c4fc9 | 1316 | static unsigned long |
45bf76df IM |
1317 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1318 | struct load_weight *lw) | |
1319 | { | |
1320 | u64 tmp; | |
1321 | ||
7a232e03 LJ |
1322 | if (!lw->inv_weight) { |
1323 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1324 | lw->inv_weight = 1; | |
1325 | else | |
1326 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1327 | / (lw->weight+1); | |
1328 | } | |
45bf76df IM |
1329 | |
1330 | tmp = (u64)delta_exec * weight; | |
1331 | /* | |
1332 | * Check whether we'd overflow the 64-bit multiplication: | |
1333 | */ | |
194081eb | 1334 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1335 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1336 | WMULT_SHIFT/2); |
1337 | else | |
cf2ab469 | 1338 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1339 | |
ecf691da | 1340 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1341 | } |
1342 | ||
1091985b | 1343 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1344 | { |
1345 | lw->weight += inc; | |
e89996ae | 1346 | lw->inv_weight = 0; |
45bf76df IM |
1347 | } |
1348 | ||
1091985b | 1349 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1350 | { |
1351 | lw->weight -= dec; | |
e89996ae | 1352 | lw->inv_weight = 0; |
45bf76df IM |
1353 | } |
1354 | ||
2069dd75 PZ |
1355 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1356 | { | |
1357 | lw->weight = w; | |
1358 | lw->inv_weight = 0; | |
1359 | } | |
1360 | ||
2dd73a4f PW |
1361 | /* |
1362 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1363 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1364 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1365 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1366 | * scaled version of the new time slice allocation that they receive on time |
1367 | * slice expiry etc. | |
1368 | */ | |
1369 | ||
cce7ade8 PZ |
1370 | #define WEIGHT_IDLEPRIO 3 |
1371 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1372 | |
1373 | /* | |
1374 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1375 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1376 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1377 | * that remained on nice 0. | |
1378 | * | |
1379 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1380 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1381 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1382 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1383 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1384 | */ |
1385 | static const int prio_to_weight[40] = { | |
254753dc IM |
1386 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1387 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1388 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1389 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1390 | /* 0 */ 1024, 820, 655, 526, 423, | |
1391 | /* 5 */ 335, 272, 215, 172, 137, | |
1392 | /* 10 */ 110, 87, 70, 56, 45, | |
1393 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1394 | }; |
1395 | ||
5714d2de IM |
1396 | /* |
1397 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1398 | * | |
1399 | * In cases where the weight does not change often, we can use the | |
1400 | * precalculated inverse to speed up arithmetics by turning divisions | |
1401 | * into multiplications: | |
1402 | */ | |
dd41f596 | 1403 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1404 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1405 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1406 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1407 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1408 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1409 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1410 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1411 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1412 | }; |
2dd73a4f | 1413 | |
ef12fefa BR |
1414 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1415 | enum cpuacct_stat_index { | |
1416 | CPUACCT_STAT_USER, /* ... user mode */ | |
1417 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1418 | ||
1419 | CPUACCT_STAT_NSTATS, | |
1420 | }; | |
1421 | ||
d842de87 SV |
1422 | #ifdef CONFIG_CGROUP_CPUACCT |
1423 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1424 | static void cpuacct_update_stats(struct task_struct *tsk, |
1425 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1426 | #else |
1427 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1428 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1429 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1430 | #endif |
1431 | ||
18d95a28 PZ |
1432 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1433 | { | |
1434 | update_load_add(&rq->load, load); | |
1435 | } | |
1436 | ||
1437 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1438 | { | |
1439 | update_load_sub(&rq->load, load); | |
1440 | } | |
1441 | ||
7940ca36 | 1442 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1443 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1444 | |
1445 | /* | |
1446 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1447 | * leaving it for the final time. | |
1448 | */ | |
eb755805 | 1449 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1450 | { |
1451 | struct task_group *parent, *child; | |
eb755805 | 1452 | int ret; |
c09595f6 PZ |
1453 | |
1454 | rcu_read_lock(); | |
1455 | parent = &root_task_group; | |
1456 | down: | |
eb755805 PZ |
1457 | ret = (*down)(parent, data); |
1458 | if (ret) | |
1459 | goto out_unlock; | |
c09595f6 PZ |
1460 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1461 | parent = child; | |
1462 | goto down; | |
1463 | ||
1464 | up: | |
1465 | continue; | |
1466 | } | |
eb755805 PZ |
1467 | ret = (*up)(parent, data); |
1468 | if (ret) | |
1469 | goto out_unlock; | |
c09595f6 PZ |
1470 | |
1471 | child = parent; | |
1472 | parent = parent->parent; | |
1473 | if (parent) | |
1474 | goto up; | |
eb755805 | 1475 | out_unlock: |
c09595f6 | 1476 | rcu_read_unlock(); |
eb755805 PZ |
1477 | |
1478 | return ret; | |
c09595f6 PZ |
1479 | } |
1480 | ||
eb755805 PZ |
1481 | static int tg_nop(struct task_group *tg, void *data) |
1482 | { | |
1483 | return 0; | |
c09595f6 | 1484 | } |
eb755805 PZ |
1485 | #endif |
1486 | ||
1487 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1488 | /* Used instead of source_load when we know the type == 0 */ |
1489 | static unsigned long weighted_cpuload(const int cpu) | |
1490 | { | |
1491 | return cpu_rq(cpu)->load.weight; | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * Return a low guess at the load of a migration-source cpu weighted | |
1496 | * according to the scheduling class and "nice" value. | |
1497 | * | |
1498 | * We want to under-estimate the load of migration sources, to | |
1499 | * balance conservatively. | |
1500 | */ | |
1501 | static unsigned long source_load(int cpu, int type) | |
1502 | { | |
1503 | struct rq *rq = cpu_rq(cpu); | |
1504 | unsigned long total = weighted_cpuload(cpu); | |
1505 | ||
1506 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1507 | return total; | |
1508 | ||
1509 | return min(rq->cpu_load[type-1], total); | |
1510 | } | |
1511 | ||
1512 | /* | |
1513 | * Return a high guess at the load of a migration-target cpu weighted | |
1514 | * according to the scheduling class and "nice" value. | |
1515 | */ | |
1516 | static unsigned long target_load(int cpu, int type) | |
1517 | { | |
1518 | struct rq *rq = cpu_rq(cpu); | |
1519 | unsigned long total = weighted_cpuload(cpu); | |
1520 | ||
1521 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1522 | return total; | |
1523 | ||
1524 | return max(rq->cpu_load[type-1], total); | |
1525 | } | |
1526 | ||
ae154be1 PZ |
1527 | static unsigned long power_of(int cpu) |
1528 | { | |
e51fd5e2 | 1529 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1530 | } |
1531 | ||
eb755805 PZ |
1532 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1533 | ||
1534 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1535 | { | |
1536 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1537 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1538 | |
4cd42620 SR |
1539 | if (nr_running) |
1540 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1541 | else |
1542 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1543 | |
1544 | return rq->avg_load_per_task; | |
1545 | } | |
1546 | ||
1547 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1548 | |
c09595f6 | 1549 | /* |
c8cba857 PZ |
1550 | * Compute the cpu's hierarchical load factor for each task group. |
1551 | * This needs to be done in a top-down fashion because the load of a child | |
1552 | * group is a fraction of its parents load. | |
c09595f6 | 1553 | */ |
eb755805 | 1554 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1555 | { |
c8cba857 | 1556 | unsigned long load; |
eb755805 | 1557 | long cpu = (long)data; |
c09595f6 | 1558 | |
c8cba857 PZ |
1559 | if (!tg->parent) { |
1560 | load = cpu_rq(cpu)->load.weight; | |
1561 | } else { | |
1562 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1563 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1564 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1565 | } | |
c09595f6 | 1566 | |
c8cba857 | 1567 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1568 | |
eb755805 | 1569 | return 0; |
c09595f6 PZ |
1570 | } |
1571 | ||
eb755805 | 1572 | static void update_h_load(long cpu) |
c09595f6 | 1573 | { |
eb755805 | 1574 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1575 | } |
1576 | ||
18d95a28 PZ |
1577 | #endif |
1578 | ||
8f45e2b5 GH |
1579 | #ifdef CONFIG_PREEMPT |
1580 | ||
b78bb868 PZ |
1581 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1582 | ||
70574a99 | 1583 | /* |
8f45e2b5 GH |
1584 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1585 | * way at the expense of forcing extra atomic operations in all | |
1586 | * invocations. This assures that the double_lock is acquired using the | |
1587 | * same underlying policy as the spinlock_t on this architecture, which | |
1588 | * reduces latency compared to the unfair variant below. However, it | |
1589 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1590 | */ |
8f45e2b5 GH |
1591 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1592 | __releases(this_rq->lock) | |
1593 | __acquires(busiest->lock) | |
1594 | __acquires(this_rq->lock) | |
1595 | { | |
05fa785c | 1596 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1597 | double_rq_lock(this_rq, busiest); |
1598 | ||
1599 | return 1; | |
1600 | } | |
1601 | ||
1602 | #else | |
1603 | /* | |
1604 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1605 | * latency by eliminating extra atomic operations when the locks are | |
1606 | * already in proper order on entry. This favors lower cpu-ids and will | |
1607 | * grant the double lock to lower cpus over higher ids under contention, | |
1608 | * regardless of entry order into the function. | |
1609 | */ | |
1610 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1611 | __releases(this_rq->lock) |
1612 | __acquires(busiest->lock) | |
1613 | __acquires(this_rq->lock) | |
1614 | { | |
1615 | int ret = 0; | |
1616 | ||
05fa785c | 1617 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1618 | if (busiest < this_rq) { |
05fa785c TG |
1619 | raw_spin_unlock(&this_rq->lock); |
1620 | raw_spin_lock(&busiest->lock); | |
1621 | raw_spin_lock_nested(&this_rq->lock, | |
1622 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1623 | ret = 1; |
1624 | } else | |
05fa785c TG |
1625 | raw_spin_lock_nested(&busiest->lock, |
1626 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1627 | } |
1628 | return ret; | |
1629 | } | |
1630 | ||
8f45e2b5 GH |
1631 | #endif /* CONFIG_PREEMPT */ |
1632 | ||
1633 | /* | |
1634 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1635 | */ | |
1636 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1637 | { | |
1638 | if (unlikely(!irqs_disabled())) { | |
1639 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1640 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1641 | BUG_ON(1); |
1642 | } | |
1643 | ||
1644 | return _double_lock_balance(this_rq, busiest); | |
1645 | } | |
1646 | ||
70574a99 AD |
1647 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1648 | __releases(busiest->lock) | |
1649 | { | |
05fa785c | 1650 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1651 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1652 | } | |
1e3c88bd PZ |
1653 | |
1654 | /* | |
1655 | * double_rq_lock - safely lock two runqueues | |
1656 | * | |
1657 | * Note this does not disable interrupts like task_rq_lock, | |
1658 | * you need to do so manually before calling. | |
1659 | */ | |
1660 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1661 | __acquires(rq1->lock) | |
1662 | __acquires(rq2->lock) | |
1663 | { | |
1664 | BUG_ON(!irqs_disabled()); | |
1665 | if (rq1 == rq2) { | |
1666 | raw_spin_lock(&rq1->lock); | |
1667 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1668 | } else { | |
1669 | if (rq1 < rq2) { | |
1670 | raw_spin_lock(&rq1->lock); | |
1671 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1672 | } else { | |
1673 | raw_spin_lock(&rq2->lock); | |
1674 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1675 | } | |
1676 | } | |
1e3c88bd PZ |
1677 | } |
1678 | ||
1679 | /* | |
1680 | * double_rq_unlock - safely unlock two runqueues | |
1681 | * | |
1682 | * Note this does not restore interrupts like task_rq_unlock, | |
1683 | * you need to do so manually after calling. | |
1684 | */ | |
1685 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1686 | __releases(rq1->lock) | |
1687 | __releases(rq2->lock) | |
1688 | { | |
1689 | raw_spin_unlock(&rq1->lock); | |
1690 | if (rq1 != rq2) | |
1691 | raw_spin_unlock(&rq2->lock); | |
1692 | else | |
1693 | __release(rq2->lock); | |
1694 | } | |
1695 | ||
18d95a28 PZ |
1696 | #endif |
1697 | ||
74f5187a | 1698 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1699 | static void update_sysctl(void); |
acb4a848 | 1700 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1701 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1702 | |
cd29fe6f PZ |
1703 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1704 | { | |
1705 | set_task_rq(p, cpu); | |
1706 | #ifdef CONFIG_SMP | |
1707 | /* | |
1708 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1709 | * successfuly executed on another CPU. We must ensure that updates of | |
1710 | * per-task data have been completed by this moment. | |
1711 | */ | |
1712 | smp_wmb(); | |
1713 | task_thread_info(p)->cpu = cpu; | |
1714 | #endif | |
1715 | } | |
dce48a84 | 1716 | |
1e3c88bd | 1717 | static const struct sched_class rt_sched_class; |
dd41f596 | 1718 | |
34f971f6 | 1719 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1720 | #define for_each_class(class) \ |
1721 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1722 | |
1e3c88bd PZ |
1723 | #include "sched_stats.h" |
1724 | ||
c09595f6 | 1725 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1726 | { |
1727 | rq->nr_running++; | |
9c217245 IM |
1728 | } |
1729 | ||
c09595f6 | 1730 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1731 | { |
1732 | rq->nr_running--; | |
9c217245 IM |
1733 | } |
1734 | ||
45bf76df IM |
1735 | static void set_load_weight(struct task_struct *p) |
1736 | { | |
dd41f596 IM |
1737 | /* |
1738 | * SCHED_IDLE tasks get minimal weight: | |
1739 | */ | |
1740 | if (p->policy == SCHED_IDLE) { | |
1741 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1742 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1743 | return; | |
1744 | } | |
71f8bd46 | 1745 | |
dd41f596 IM |
1746 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1747 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1748 | } |
1749 | ||
371fd7e7 | 1750 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1751 | { |
a64692a3 | 1752 | update_rq_clock(rq); |
dd41f596 | 1753 | sched_info_queued(p); |
371fd7e7 | 1754 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1755 | p->se.on_rq = 1; |
71f8bd46 IM |
1756 | } |
1757 | ||
371fd7e7 | 1758 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1759 | { |
a64692a3 | 1760 | update_rq_clock(rq); |
46ac22ba | 1761 | sched_info_dequeued(p); |
371fd7e7 | 1762 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1763 | p->se.on_rq = 0; |
71f8bd46 IM |
1764 | } |
1765 | ||
1e3c88bd PZ |
1766 | /* |
1767 | * activate_task - move a task to the runqueue. | |
1768 | */ | |
371fd7e7 | 1769 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1770 | { |
1771 | if (task_contributes_to_load(p)) | |
1772 | rq->nr_uninterruptible--; | |
1773 | ||
371fd7e7 | 1774 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1775 | inc_nr_running(rq); |
1776 | } | |
1777 | ||
1778 | /* | |
1779 | * deactivate_task - remove a task from the runqueue. | |
1780 | */ | |
371fd7e7 | 1781 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1782 | { |
1783 | if (task_contributes_to_load(p)) | |
1784 | rq->nr_uninterruptible++; | |
1785 | ||
371fd7e7 | 1786 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1787 | dec_nr_running(rq); |
1788 | } | |
1789 | ||
b52bfee4 VP |
1790 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1791 | ||
305e6835 VP |
1792 | /* |
1793 | * There are no locks covering percpu hardirq/softirq time. | |
1794 | * They are only modified in account_system_vtime, on corresponding CPU | |
1795 | * with interrupts disabled. So, writes are safe. | |
1796 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1797 | * This may result in other CPU reading this CPU's irq time and can | |
1798 | * race with irq/account_system_vtime on this CPU. We would either get old | |
1799 | * or new value (or semi updated value on 32 bit) with a side effect of | |
1800 | * accounting a slice of irq time to wrong task when irq is in progress | |
1801 | * while we read rq->clock. That is a worthy compromise in place of having | |
1802 | * locks on each irq in account_system_time. | |
1803 | */ | |
b52bfee4 VP |
1804 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1805 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1806 | ||
1807 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1808 | static int sched_clock_irqtime; | |
1809 | ||
1810 | void enable_sched_clock_irqtime(void) | |
1811 | { | |
1812 | sched_clock_irqtime = 1; | |
1813 | } | |
1814 | ||
1815 | void disable_sched_clock_irqtime(void) | |
1816 | { | |
1817 | sched_clock_irqtime = 0; | |
1818 | } | |
1819 | ||
305e6835 VP |
1820 | static u64 irq_time_cpu(int cpu) |
1821 | { | |
1822 | if (!sched_clock_irqtime) | |
1823 | return 0; | |
1824 | ||
1825 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | |
1826 | } | |
1827 | ||
b52bfee4 VP |
1828 | void account_system_vtime(struct task_struct *curr) |
1829 | { | |
1830 | unsigned long flags; | |
1831 | int cpu; | |
1832 | u64 now, delta; | |
1833 | ||
1834 | if (!sched_clock_irqtime) | |
1835 | return; | |
1836 | ||
1837 | local_irq_save(flags); | |
1838 | ||
b52bfee4 | 1839 | cpu = smp_processor_id(); |
d267f87f | 1840 | now = sched_clock_cpu(cpu); |
b52bfee4 VP |
1841 | delta = now - per_cpu(irq_start_time, cpu); |
1842 | per_cpu(irq_start_time, cpu) = now; | |
1843 | /* | |
1844 | * We do not account for softirq time from ksoftirqd here. | |
1845 | * We want to continue accounting softirq time to ksoftirqd thread | |
1846 | * in that case, so as not to confuse scheduler with a special task | |
1847 | * that do not consume any time, but still wants to run. | |
1848 | */ | |
1849 | if (hardirq_count()) | |
1850 | per_cpu(cpu_hardirq_time, cpu) += delta; | |
1851 | else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD)) | |
1852 | per_cpu(cpu_softirq_time, cpu) += delta; | |
1853 | ||
1854 | local_irq_restore(flags); | |
1855 | } | |
b7dadc38 | 1856 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1857 | |
aa483808 VP |
1858 | static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) |
1859 | { | |
1860 | if (sched_clock_irqtime && sched_feat(NONIRQ_POWER)) { | |
1861 | u64 delta_irq = curr_irq_time - rq->prev_irq_time; | |
1862 | rq->prev_irq_time = curr_irq_time; | |
1863 | sched_rt_avg_update(rq, delta_irq); | |
1864 | } | |
1865 | } | |
1866 | ||
305e6835 VP |
1867 | #else |
1868 | ||
1869 | static u64 irq_time_cpu(int cpu) | |
1870 | { | |
1871 | return 0; | |
1872 | } | |
1873 | ||
aa483808 VP |
1874 | static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) { } |
1875 | ||
b52bfee4 VP |
1876 | #endif |
1877 | ||
1e3c88bd PZ |
1878 | #include "sched_idletask.c" |
1879 | #include "sched_fair.c" | |
1880 | #include "sched_rt.c" | |
34f971f6 | 1881 | #include "sched_stoptask.c" |
1e3c88bd PZ |
1882 | #ifdef CONFIG_SCHED_DEBUG |
1883 | # include "sched_debug.c" | |
1884 | #endif | |
1885 | ||
34f971f6 PZ |
1886 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
1887 | { | |
1888 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
1889 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
1890 | ||
1891 | if (stop) { | |
1892 | /* | |
1893 | * Make it appear like a SCHED_FIFO task, its something | |
1894 | * userspace knows about and won't get confused about. | |
1895 | * | |
1896 | * Also, it will make PI more or less work without too | |
1897 | * much confusion -- but then, stop work should not | |
1898 | * rely on PI working anyway. | |
1899 | */ | |
1900 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
1901 | ||
1902 | stop->sched_class = &stop_sched_class; | |
1903 | } | |
1904 | ||
1905 | cpu_rq(cpu)->stop = stop; | |
1906 | ||
1907 | if (old_stop) { | |
1908 | /* | |
1909 | * Reset it back to a normal scheduling class so that | |
1910 | * it can die in pieces. | |
1911 | */ | |
1912 | old_stop->sched_class = &rt_sched_class; | |
1913 | } | |
1914 | } | |
1915 | ||
14531189 | 1916 | /* |
dd41f596 | 1917 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1918 | */ |
14531189 IM |
1919 | static inline int __normal_prio(struct task_struct *p) |
1920 | { | |
dd41f596 | 1921 | return p->static_prio; |
14531189 IM |
1922 | } |
1923 | ||
b29739f9 IM |
1924 | /* |
1925 | * Calculate the expected normal priority: i.e. priority | |
1926 | * without taking RT-inheritance into account. Might be | |
1927 | * boosted by interactivity modifiers. Changes upon fork, | |
1928 | * setprio syscalls, and whenever the interactivity | |
1929 | * estimator recalculates. | |
1930 | */ | |
36c8b586 | 1931 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1932 | { |
1933 | int prio; | |
1934 | ||
e05606d3 | 1935 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1936 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1937 | else | |
1938 | prio = __normal_prio(p); | |
1939 | return prio; | |
1940 | } | |
1941 | ||
1942 | /* | |
1943 | * Calculate the current priority, i.e. the priority | |
1944 | * taken into account by the scheduler. This value might | |
1945 | * be boosted by RT tasks, or might be boosted by | |
1946 | * interactivity modifiers. Will be RT if the task got | |
1947 | * RT-boosted. If not then it returns p->normal_prio. | |
1948 | */ | |
36c8b586 | 1949 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1950 | { |
1951 | p->normal_prio = normal_prio(p); | |
1952 | /* | |
1953 | * If we are RT tasks or we were boosted to RT priority, | |
1954 | * keep the priority unchanged. Otherwise, update priority | |
1955 | * to the normal priority: | |
1956 | */ | |
1957 | if (!rt_prio(p->prio)) | |
1958 | return p->normal_prio; | |
1959 | return p->prio; | |
1960 | } | |
1961 | ||
1da177e4 LT |
1962 | /** |
1963 | * task_curr - is this task currently executing on a CPU? | |
1964 | * @p: the task in question. | |
1965 | */ | |
36c8b586 | 1966 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1967 | { |
1968 | return cpu_curr(task_cpu(p)) == p; | |
1969 | } | |
1970 | ||
cb469845 SR |
1971 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1972 | const struct sched_class *prev_class, | |
1973 | int oldprio, int running) | |
1974 | { | |
1975 | if (prev_class != p->sched_class) { | |
1976 | if (prev_class->switched_from) | |
1977 | prev_class->switched_from(rq, p, running); | |
1978 | p->sched_class->switched_to(rq, p, running); | |
1979 | } else | |
1980 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1981 | } | |
1982 | ||
1da177e4 | 1983 | #ifdef CONFIG_SMP |
cc367732 IM |
1984 | /* |
1985 | * Is this task likely cache-hot: | |
1986 | */ | |
e7693a36 | 1987 | static int |
cc367732 IM |
1988 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1989 | { | |
1990 | s64 delta; | |
1991 | ||
e6c8fba7 PZ |
1992 | if (p->sched_class != &fair_sched_class) |
1993 | return 0; | |
1994 | ||
ef8002f6 NR |
1995 | if (unlikely(p->policy == SCHED_IDLE)) |
1996 | return 0; | |
1997 | ||
f540a608 IM |
1998 | /* |
1999 | * Buddy candidates are cache hot: | |
2000 | */ | |
f685ceac | 2001 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2002 | (&p->se == cfs_rq_of(&p->se)->next || |
2003 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2004 | return 1; |
2005 | ||
6bc1665b IM |
2006 | if (sysctl_sched_migration_cost == -1) |
2007 | return 1; | |
2008 | if (sysctl_sched_migration_cost == 0) | |
2009 | return 0; | |
2010 | ||
cc367732 IM |
2011 | delta = now - p->se.exec_start; |
2012 | ||
2013 | return delta < (s64)sysctl_sched_migration_cost; | |
2014 | } | |
2015 | ||
dd41f596 | 2016 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2017 | { |
e2912009 PZ |
2018 | #ifdef CONFIG_SCHED_DEBUG |
2019 | /* | |
2020 | * We should never call set_task_cpu() on a blocked task, | |
2021 | * ttwu() will sort out the placement. | |
2022 | */ | |
077614ee PZ |
2023 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2024 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2025 | #endif |
2026 | ||
de1d7286 | 2027 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2028 | |
0c69774e PZ |
2029 | if (task_cpu(p) != new_cpu) { |
2030 | p->se.nr_migrations++; | |
2031 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2032 | } | |
dd41f596 IM |
2033 | |
2034 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2035 | } |
2036 | ||
969c7921 | 2037 | struct migration_arg { |
36c8b586 | 2038 | struct task_struct *task; |
1da177e4 | 2039 | int dest_cpu; |
70b97a7f | 2040 | }; |
1da177e4 | 2041 | |
969c7921 TH |
2042 | static int migration_cpu_stop(void *data); |
2043 | ||
1da177e4 LT |
2044 | /* |
2045 | * The task's runqueue lock must be held. | |
2046 | * Returns true if you have to wait for migration thread. | |
2047 | */ | |
969c7921 | 2048 | static bool migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2049 | { |
70b97a7f | 2050 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2051 | |
2052 | /* | |
2053 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2054 | * the next wake-up will properly place the task. |
1da177e4 | 2055 | */ |
969c7921 | 2056 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2057 | } |
2058 | ||
2059 | /* | |
2060 | * wait_task_inactive - wait for a thread to unschedule. | |
2061 | * | |
85ba2d86 RM |
2062 | * If @match_state is nonzero, it's the @p->state value just checked and |
2063 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2064 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2065 | * we return a positive number (its total switch count). If a second call | |
2066 | * a short while later returns the same number, the caller can be sure that | |
2067 | * @p has remained unscheduled the whole time. | |
2068 | * | |
1da177e4 LT |
2069 | * The caller must ensure that the task *will* unschedule sometime soon, |
2070 | * else this function might spin for a *long* time. This function can't | |
2071 | * be called with interrupts off, or it may introduce deadlock with | |
2072 | * smp_call_function() if an IPI is sent by the same process we are | |
2073 | * waiting to become inactive. | |
2074 | */ | |
85ba2d86 | 2075 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2076 | { |
2077 | unsigned long flags; | |
dd41f596 | 2078 | int running, on_rq; |
85ba2d86 | 2079 | unsigned long ncsw; |
70b97a7f | 2080 | struct rq *rq; |
1da177e4 | 2081 | |
3a5c359a AK |
2082 | for (;;) { |
2083 | /* | |
2084 | * We do the initial early heuristics without holding | |
2085 | * any task-queue locks at all. We'll only try to get | |
2086 | * the runqueue lock when things look like they will | |
2087 | * work out! | |
2088 | */ | |
2089 | rq = task_rq(p); | |
fa490cfd | 2090 | |
3a5c359a AK |
2091 | /* |
2092 | * If the task is actively running on another CPU | |
2093 | * still, just relax and busy-wait without holding | |
2094 | * any locks. | |
2095 | * | |
2096 | * NOTE! Since we don't hold any locks, it's not | |
2097 | * even sure that "rq" stays as the right runqueue! | |
2098 | * But we don't care, since "task_running()" will | |
2099 | * return false if the runqueue has changed and p | |
2100 | * is actually now running somewhere else! | |
2101 | */ | |
85ba2d86 RM |
2102 | while (task_running(rq, p)) { |
2103 | if (match_state && unlikely(p->state != match_state)) | |
2104 | return 0; | |
3a5c359a | 2105 | cpu_relax(); |
85ba2d86 | 2106 | } |
fa490cfd | 2107 | |
3a5c359a AK |
2108 | /* |
2109 | * Ok, time to look more closely! We need the rq | |
2110 | * lock now, to be *sure*. If we're wrong, we'll | |
2111 | * just go back and repeat. | |
2112 | */ | |
2113 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2114 | trace_sched_wait_task(p); |
3a5c359a AK |
2115 | running = task_running(rq, p); |
2116 | on_rq = p->se.on_rq; | |
85ba2d86 | 2117 | ncsw = 0; |
f31e11d8 | 2118 | if (!match_state || p->state == match_state) |
93dcf55f | 2119 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2120 | task_rq_unlock(rq, &flags); |
fa490cfd | 2121 | |
85ba2d86 RM |
2122 | /* |
2123 | * If it changed from the expected state, bail out now. | |
2124 | */ | |
2125 | if (unlikely(!ncsw)) | |
2126 | break; | |
2127 | ||
3a5c359a AK |
2128 | /* |
2129 | * Was it really running after all now that we | |
2130 | * checked with the proper locks actually held? | |
2131 | * | |
2132 | * Oops. Go back and try again.. | |
2133 | */ | |
2134 | if (unlikely(running)) { | |
2135 | cpu_relax(); | |
2136 | continue; | |
2137 | } | |
fa490cfd | 2138 | |
3a5c359a AK |
2139 | /* |
2140 | * It's not enough that it's not actively running, | |
2141 | * it must be off the runqueue _entirely_, and not | |
2142 | * preempted! | |
2143 | * | |
80dd99b3 | 2144 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2145 | * running right now), it's preempted, and we should |
2146 | * yield - it could be a while. | |
2147 | */ | |
2148 | if (unlikely(on_rq)) { | |
2149 | schedule_timeout_uninterruptible(1); | |
2150 | continue; | |
2151 | } | |
fa490cfd | 2152 | |
3a5c359a AK |
2153 | /* |
2154 | * Ahh, all good. It wasn't running, and it wasn't | |
2155 | * runnable, which means that it will never become | |
2156 | * running in the future either. We're all done! | |
2157 | */ | |
2158 | break; | |
2159 | } | |
85ba2d86 RM |
2160 | |
2161 | return ncsw; | |
1da177e4 LT |
2162 | } |
2163 | ||
2164 | /*** | |
2165 | * kick_process - kick a running thread to enter/exit the kernel | |
2166 | * @p: the to-be-kicked thread | |
2167 | * | |
2168 | * Cause a process which is running on another CPU to enter | |
2169 | * kernel-mode, without any delay. (to get signals handled.) | |
2170 | * | |
2171 | * NOTE: this function doesnt have to take the runqueue lock, | |
2172 | * because all it wants to ensure is that the remote task enters | |
2173 | * the kernel. If the IPI races and the task has been migrated | |
2174 | * to another CPU then no harm is done and the purpose has been | |
2175 | * achieved as well. | |
2176 | */ | |
36c8b586 | 2177 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2178 | { |
2179 | int cpu; | |
2180 | ||
2181 | preempt_disable(); | |
2182 | cpu = task_cpu(p); | |
2183 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2184 | smp_send_reschedule(cpu); | |
2185 | preempt_enable(); | |
2186 | } | |
b43e3521 | 2187 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2188 | #endif /* CONFIG_SMP */ |
1da177e4 | 2189 | |
0793a61d TG |
2190 | /** |
2191 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2192 | * @p: the task to evaluate | |
2193 | * @func: the function to be called | |
2194 | * @info: the function call argument | |
2195 | * | |
2196 | * Calls the function @func when the task is currently running. This might | |
2197 | * be on the current CPU, which just calls the function directly | |
2198 | */ | |
2199 | void task_oncpu_function_call(struct task_struct *p, | |
2200 | void (*func) (void *info), void *info) | |
2201 | { | |
2202 | int cpu; | |
2203 | ||
2204 | preempt_disable(); | |
2205 | cpu = task_cpu(p); | |
2206 | if (task_curr(p)) | |
2207 | smp_call_function_single(cpu, func, info, 1); | |
2208 | preempt_enable(); | |
2209 | } | |
2210 | ||
970b13ba | 2211 | #ifdef CONFIG_SMP |
30da688e ON |
2212 | /* |
2213 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2214 | */ | |
5da9a0fb PZ |
2215 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2216 | { | |
2217 | int dest_cpu; | |
2218 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2219 | ||
2220 | /* Look for allowed, online CPU in same node. */ | |
2221 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2222 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2223 | return dest_cpu; | |
2224 | ||
2225 | /* Any allowed, online CPU? */ | |
2226 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2227 | if (dest_cpu < nr_cpu_ids) | |
2228 | return dest_cpu; | |
2229 | ||
2230 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2231 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2232 | /* | |
2233 | * Don't tell them about moving exiting tasks or | |
2234 | * kernel threads (both mm NULL), since they never | |
2235 | * leave kernel. | |
2236 | */ | |
2237 | if (p->mm && printk_ratelimit()) { | |
2238 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2239 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2240 | } |
2241 | ||
2242 | return dest_cpu; | |
2243 | } | |
2244 | ||
e2912009 | 2245 | /* |
30da688e | 2246 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2247 | */ |
970b13ba | 2248 | static inline |
0017d735 | 2249 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2250 | { |
0017d735 | 2251 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2252 | |
2253 | /* | |
2254 | * In order not to call set_task_cpu() on a blocking task we need | |
2255 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2256 | * cpu. | |
2257 | * | |
2258 | * Since this is common to all placement strategies, this lives here. | |
2259 | * | |
2260 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2261 | * not worry about this generic constraint ] | |
2262 | */ | |
2263 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2264 | !cpu_online(cpu))) |
5da9a0fb | 2265 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2266 | |
2267 | return cpu; | |
970b13ba | 2268 | } |
09a40af5 MG |
2269 | |
2270 | static void update_avg(u64 *avg, u64 sample) | |
2271 | { | |
2272 | s64 diff = sample - *avg; | |
2273 | *avg += diff >> 3; | |
2274 | } | |
970b13ba PZ |
2275 | #endif |
2276 | ||
9ed3811a TH |
2277 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2278 | bool is_sync, bool is_migrate, bool is_local, | |
2279 | unsigned long en_flags) | |
2280 | { | |
2281 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2282 | if (is_sync) | |
2283 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2284 | if (is_migrate) | |
2285 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2286 | if (is_local) | |
2287 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2288 | else | |
2289 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2290 | ||
2291 | activate_task(rq, p, en_flags); | |
2292 | } | |
2293 | ||
2294 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2295 | int wake_flags, bool success) | |
2296 | { | |
2297 | trace_sched_wakeup(p, success); | |
2298 | check_preempt_curr(rq, p, wake_flags); | |
2299 | ||
2300 | p->state = TASK_RUNNING; | |
2301 | #ifdef CONFIG_SMP | |
2302 | if (p->sched_class->task_woken) | |
2303 | p->sched_class->task_woken(rq, p); | |
2304 | ||
2305 | if (unlikely(rq->idle_stamp)) { | |
2306 | u64 delta = rq->clock - rq->idle_stamp; | |
2307 | u64 max = 2*sysctl_sched_migration_cost; | |
2308 | ||
2309 | if (delta > max) | |
2310 | rq->avg_idle = max; | |
2311 | else | |
2312 | update_avg(&rq->avg_idle, delta); | |
2313 | rq->idle_stamp = 0; | |
2314 | } | |
2315 | #endif | |
21aa9af0 TH |
2316 | /* if a worker is waking up, notify workqueue */ |
2317 | if ((p->flags & PF_WQ_WORKER) && success) | |
2318 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2319 | } |
2320 | ||
2321 | /** | |
1da177e4 | 2322 | * try_to_wake_up - wake up a thread |
9ed3811a | 2323 | * @p: the thread to be awakened |
1da177e4 | 2324 | * @state: the mask of task states that can be woken |
9ed3811a | 2325 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2326 | * |
2327 | * Put it on the run-queue if it's not already there. The "current" | |
2328 | * thread is always on the run-queue (except when the actual | |
2329 | * re-schedule is in progress), and as such you're allowed to do | |
2330 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2331 | * runnable without the overhead of this. | |
2332 | * | |
9ed3811a TH |
2333 | * Returns %true if @p was woken up, %false if it was already running |
2334 | * or @state didn't match @p's state. | |
1da177e4 | 2335 | */ |
7d478721 PZ |
2336 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2337 | int wake_flags) | |
1da177e4 | 2338 | { |
cc367732 | 2339 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2340 | unsigned long flags; |
371fd7e7 | 2341 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2342 | struct rq *rq; |
1da177e4 | 2343 | |
e9c84311 | 2344 | this_cpu = get_cpu(); |
2398f2c6 | 2345 | |
04e2f174 | 2346 | smp_wmb(); |
ab3b3aa5 | 2347 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2348 | if (!(p->state & state)) |
1da177e4 LT |
2349 | goto out; |
2350 | ||
dd41f596 | 2351 | if (p->se.on_rq) |
1da177e4 LT |
2352 | goto out_running; |
2353 | ||
2354 | cpu = task_cpu(p); | |
cc367732 | 2355 | orig_cpu = cpu; |
1da177e4 LT |
2356 | |
2357 | #ifdef CONFIG_SMP | |
2358 | if (unlikely(task_running(rq, p))) | |
2359 | goto out_activate; | |
2360 | ||
e9c84311 PZ |
2361 | /* |
2362 | * In order to handle concurrent wakeups and release the rq->lock | |
2363 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2364 | * |
2365 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2366 | */ |
cc87f76a PZ |
2367 | if (task_contributes_to_load(p)) { |
2368 | if (likely(cpu_online(orig_cpu))) | |
2369 | rq->nr_uninterruptible--; | |
2370 | else | |
2371 | this_rq()->nr_uninterruptible--; | |
2372 | } | |
e9c84311 | 2373 | p->state = TASK_WAKING; |
efbbd05a | 2374 | |
371fd7e7 | 2375 | if (p->sched_class->task_waking) { |
efbbd05a | 2376 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2377 | en_flags |= ENQUEUE_WAKING; |
2378 | } | |
efbbd05a | 2379 | |
0017d735 PZ |
2380 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2381 | if (cpu != orig_cpu) | |
5d2f5a61 | 2382 | set_task_cpu(p, cpu); |
0017d735 | 2383 | __task_rq_unlock(rq); |
ab19cb23 | 2384 | |
0970d299 PZ |
2385 | rq = cpu_rq(cpu); |
2386 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2387 | |
0970d299 PZ |
2388 | /* |
2389 | * We migrated the task without holding either rq->lock, however | |
2390 | * since the task is not on the task list itself, nobody else | |
2391 | * will try and migrate the task, hence the rq should match the | |
2392 | * cpu we just moved it to. | |
2393 | */ | |
2394 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2395 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2396 | |
e7693a36 GH |
2397 | #ifdef CONFIG_SCHEDSTATS |
2398 | schedstat_inc(rq, ttwu_count); | |
2399 | if (cpu == this_cpu) | |
2400 | schedstat_inc(rq, ttwu_local); | |
2401 | else { | |
2402 | struct sched_domain *sd; | |
2403 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2404 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2405 | schedstat_inc(sd, ttwu_wake_remote); |
2406 | break; | |
2407 | } | |
2408 | } | |
2409 | } | |
6d6bc0ad | 2410 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2411 | |
1da177e4 LT |
2412 | out_activate: |
2413 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2414 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2415 | cpu == this_cpu, en_flags); | |
1da177e4 | 2416 | success = 1; |
1da177e4 | 2417 | out_running: |
9ed3811a | 2418 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2419 | out: |
2420 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2421 | put_cpu(); |
1da177e4 LT |
2422 | |
2423 | return success; | |
2424 | } | |
2425 | ||
21aa9af0 TH |
2426 | /** |
2427 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2428 | * @p: the thread to be awakened | |
2429 | * | |
2430 | * Put @p on the run-queue if it's not alredy there. The caller must | |
2431 | * ensure that this_rq() is locked, @p is bound to this_rq() and not | |
2432 | * the current task. this_rq() stays locked over invocation. | |
2433 | */ | |
2434 | static void try_to_wake_up_local(struct task_struct *p) | |
2435 | { | |
2436 | struct rq *rq = task_rq(p); | |
2437 | bool success = false; | |
2438 | ||
2439 | BUG_ON(rq != this_rq()); | |
2440 | BUG_ON(p == current); | |
2441 | lockdep_assert_held(&rq->lock); | |
2442 | ||
2443 | if (!(p->state & TASK_NORMAL)) | |
2444 | return; | |
2445 | ||
2446 | if (!p->se.on_rq) { | |
2447 | if (likely(!task_running(rq, p))) { | |
2448 | schedstat_inc(rq, ttwu_count); | |
2449 | schedstat_inc(rq, ttwu_local); | |
2450 | } | |
2451 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2452 | success = true; | |
2453 | } | |
2454 | ttwu_post_activation(p, rq, 0, success); | |
2455 | } | |
2456 | ||
50fa610a DH |
2457 | /** |
2458 | * wake_up_process - Wake up a specific process | |
2459 | * @p: The process to be woken up. | |
2460 | * | |
2461 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2462 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2463 | * running. | |
2464 | * | |
2465 | * It may be assumed that this function implies a write memory barrier before | |
2466 | * changing the task state if and only if any tasks are woken up. | |
2467 | */ | |
7ad5b3a5 | 2468 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2469 | { |
d9514f6c | 2470 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2471 | } |
1da177e4 LT |
2472 | EXPORT_SYMBOL(wake_up_process); |
2473 | ||
7ad5b3a5 | 2474 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2475 | { |
2476 | return try_to_wake_up(p, state, 0); | |
2477 | } | |
2478 | ||
1da177e4 LT |
2479 | /* |
2480 | * Perform scheduler related setup for a newly forked process p. | |
2481 | * p is forked by current. | |
dd41f596 IM |
2482 | * |
2483 | * __sched_fork() is basic setup used by init_idle() too: | |
2484 | */ | |
2485 | static void __sched_fork(struct task_struct *p) | |
2486 | { | |
dd41f596 IM |
2487 | p->se.exec_start = 0; |
2488 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2489 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2490 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2491 | |
2492 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2493 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2494 | #endif |
476d139c | 2495 | |
fa717060 | 2496 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2497 | p->se.on_rq = 0; |
4a55bd5e | 2498 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2499 | |
e107be36 AK |
2500 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2501 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2502 | #endif | |
dd41f596 IM |
2503 | } |
2504 | ||
2505 | /* | |
2506 | * fork()/clone()-time setup: | |
2507 | */ | |
2508 | void sched_fork(struct task_struct *p, int clone_flags) | |
2509 | { | |
2510 | int cpu = get_cpu(); | |
2511 | ||
2512 | __sched_fork(p); | |
06b83b5f | 2513 | /* |
0017d735 | 2514 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2515 | * nobody will actually run it, and a signal or other external |
2516 | * event cannot wake it up and insert it on the runqueue either. | |
2517 | */ | |
0017d735 | 2518 | p->state = TASK_RUNNING; |
dd41f596 | 2519 | |
b9dc29e7 MG |
2520 | /* |
2521 | * Revert to default priority/policy on fork if requested. | |
2522 | */ | |
2523 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2524 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2525 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2526 | p->normal_prio = p->static_prio; |
2527 | } | |
b9dc29e7 | 2528 | |
6c697bdf MG |
2529 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2530 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2531 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2532 | set_load_weight(p); |
2533 | } | |
2534 | ||
b9dc29e7 MG |
2535 | /* |
2536 | * We don't need the reset flag anymore after the fork. It has | |
2537 | * fulfilled its duty: | |
2538 | */ | |
2539 | p->sched_reset_on_fork = 0; | |
2540 | } | |
ca94c442 | 2541 | |
f83f9ac2 PW |
2542 | /* |
2543 | * Make sure we do not leak PI boosting priority to the child. | |
2544 | */ | |
2545 | p->prio = current->normal_prio; | |
2546 | ||
2ddbf952 HS |
2547 | if (!rt_prio(p->prio)) |
2548 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2549 | |
cd29fe6f PZ |
2550 | if (p->sched_class->task_fork) |
2551 | p->sched_class->task_fork(p); | |
2552 | ||
86951599 PZ |
2553 | /* |
2554 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2555 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2556 | * is ran before sched_fork(). | |
2557 | * | |
2558 | * Silence PROVE_RCU. | |
2559 | */ | |
2560 | rcu_read_lock(); | |
5f3edc1b | 2561 | set_task_cpu(p, cpu); |
86951599 | 2562 | rcu_read_unlock(); |
5f3edc1b | 2563 | |
52f17b6c | 2564 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2565 | if (likely(sched_info_on())) |
52f17b6c | 2566 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2567 | #endif |
d6077cb8 | 2568 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2569 | p->oncpu = 0; |
2570 | #endif | |
1da177e4 | 2571 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2572 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2573 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2574 | #endif |
917b627d GH |
2575 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2576 | ||
476d139c | 2577 | put_cpu(); |
1da177e4 LT |
2578 | } |
2579 | ||
2580 | /* | |
2581 | * wake_up_new_task - wake up a newly created task for the first time. | |
2582 | * | |
2583 | * This function will do some initial scheduler statistics housekeeping | |
2584 | * that must be done for every newly created context, then puts the task | |
2585 | * on the runqueue and wakes it. | |
2586 | */ | |
7ad5b3a5 | 2587 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2588 | { |
2589 | unsigned long flags; | |
dd41f596 | 2590 | struct rq *rq; |
c890692b | 2591 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2592 | |
2593 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2594 | rq = task_rq_lock(p, &flags); |
2595 | p->state = TASK_WAKING; | |
2596 | ||
fabf318e PZ |
2597 | /* |
2598 | * Fork balancing, do it here and not earlier because: | |
2599 | * - cpus_allowed can change in the fork path | |
2600 | * - any previously selected cpu might disappear through hotplug | |
2601 | * | |
0017d735 PZ |
2602 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2603 | * without people poking at ->cpus_allowed. | |
fabf318e | 2604 | */ |
0017d735 | 2605 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2606 | set_task_cpu(p, cpu); |
1da177e4 | 2607 | |
06b83b5f | 2608 | p->state = TASK_RUNNING; |
0017d735 PZ |
2609 | task_rq_unlock(rq, &flags); |
2610 | #endif | |
2611 | ||
2612 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2613 | activate_task(rq, p, 0); |
27a9da65 | 2614 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2615 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2616 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2617 | if (p->sched_class->task_woken) |
2618 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2619 | #endif |
dd41f596 | 2620 | task_rq_unlock(rq, &flags); |
fabf318e | 2621 | put_cpu(); |
1da177e4 LT |
2622 | } |
2623 | ||
e107be36 AK |
2624 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2625 | ||
2626 | /** | |
80dd99b3 | 2627 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2628 | * @notifier: notifier struct to register |
e107be36 AK |
2629 | */ |
2630 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2631 | { | |
2632 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2633 | } | |
2634 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2635 | ||
2636 | /** | |
2637 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2638 | * @notifier: notifier struct to unregister |
e107be36 AK |
2639 | * |
2640 | * This is safe to call from within a preemption notifier. | |
2641 | */ | |
2642 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2643 | { | |
2644 | hlist_del(¬ifier->link); | |
2645 | } | |
2646 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2647 | ||
2648 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2649 | { | |
2650 | struct preempt_notifier *notifier; | |
2651 | struct hlist_node *node; | |
2652 | ||
2653 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2654 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2655 | } | |
2656 | ||
2657 | static void | |
2658 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2659 | struct task_struct *next) | |
2660 | { | |
2661 | struct preempt_notifier *notifier; | |
2662 | struct hlist_node *node; | |
2663 | ||
2664 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2665 | notifier->ops->sched_out(notifier, next); | |
2666 | } | |
2667 | ||
6d6bc0ad | 2668 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2669 | |
2670 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2671 | { | |
2672 | } | |
2673 | ||
2674 | static void | |
2675 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2676 | struct task_struct *next) | |
2677 | { | |
2678 | } | |
2679 | ||
6d6bc0ad | 2680 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2681 | |
4866cde0 NP |
2682 | /** |
2683 | * prepare_task_switch - prepare to switch tasks | |
2684 | * @rq: the runqueue preparing to switch | |
421cee29 | 2685 | * @prev: the current task that is being switched out |
4866cde0 NP |
2686 | * @next: the task we are going to switch to. |
2687 | * | |
2688 | * This is called with the rq lock held and interrupts off. It must | |
2689 | * be paired with a subsequent finish_task_switch after the context | |
2690 | * switch. | |
2691 | * | |
2692 | * prepare_task_switch sets up locking and calls architecture specific | |
2693 | * hooks. | |
2694 | */ | |
e107be36 AK |
2695 | static inline void |
2696 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2697 | struct task_struct *next) | |
4866cde0 | 2698 | { |
e107be36 | 2699 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2700 | prepare_lock_switch(rq, next); |
2701 | prepare_arch_switch(next); | |
2702 | } | |
2703 | ||
1da177e4 LT |
2704 | /** |
2705 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2706 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2707 | * @prev: the thread we just switched away from. |
2708 | * | |
4866cde0 NP |
2709 | * finish_task_switch must be called after the context switch, paired |
2710 | * with a prepare_task_switch call before the context switch. | |
2711 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2712 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2713 | * |
2714 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2715 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2716 | * with the lock held can cause deadlocks; see schedule() for |
2717 | * details.) | |
2718 | */ | |
a9957449 | 2719 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2720 | __releases(rq->lock) |
2721 | { | |
1da177e4 | 2722 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2723 | long prev_state; |
1da177e4 LT |
2724 | |
2725 | rq->prev_mm = NULL; | |
2726 | ||
2727 | /* | |
2728 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2729 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2730 | * schedule one last time. The schedule call will never return, and |
2731 | * the scheduled task must drop that reference. | |
c394cc9f | 2732 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2733 | * still held, otherwise prev could be scheduled on another cpu, die |
2734 | * there before we look at prev->state, and then the reference would | |
2735 | * be dropped twice. | |
2736 | * Manfred Spraul <manfred@colorfullife.com> | |
2737 | */ | |
55a101f8 | 2738 | prev_state = prev->state; |
4866cde0 | 2739 | finish_arch_switch(prev); |
8381f65d JI |
2740 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2741 | local_irq_disable(); | |
2742 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2743 | perf_event_task_sched_in(current); |
8381f65d JI |
2744 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2745 | local_irq_enable(); | |
2746 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2747 | finish_lock_switch(rq, prev); |
e8fa1362 | 2748 | |
e107be36 | 2749 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2750 | if (mm) |
2751 | mmdrop(mm); | |
c394cc9f | 2752 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2753 | /* |
2754 | * Remove function-return probe instances associated with this | |
2755 | * task and put them back on the free list. | |
9761eea8 | 2756 | */ |
c6fd91f0 | 2757 | kprobe_flush_task(prev); |
1da177e4 | 2758 | put_task_struct(prev); |
c6fd91f0 | 2759 | } |
1da177e4 LT |
2760 | } |
2761 | ||
3f029d3c GH |
2762 | #ifdef CONFIG_SMP |
2763 | ||
2764 | /* assumes rq->lock is held */ | |
2765 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2766 | { | |
2767 | if (prev->sched_class->pre_schedule) | |
2768 | prev->sched_class->pre_schedule(rq, prev); | |
2769 | } | |
2770 | ||
2771 | /* rq->lock is NOT held, but preemption is disabled */ | |
2772 | static inline void post_schedule(struct rq *rq) | |
2773 | { | |
2774 | if (rq->post_schedule) { | |
2775 | unsigned long flags; | |
2776 | ||
05fa785c | 2777 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2778 | if (rq->curr->sched_class->post_schedule) |
2779 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2780 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2781 | |
2782 | rq->post_schedule = 0; | |
2783 | } | |
2784 | } | |
2785 | ||
2786 | #else | |
da19ab51 | 2787 | |
3f029d3c GH |
2788 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2789 | { | |
2790 | } | |
2791 | ||
2792 | static inline void post_schedule(struct rq *rq) | |
2793 | { | |
1da177e4 LT |
2794 | } |
2795 | ||
3f029d3c GH |
2796 | #endif |
2797 | ||
1da177e4 LT |
2798 | /** |
2799 | * schedule_tail - first thing a freshly forked thread must call. | |
2800 | * @prev: the thread we just switched away from. | |
2801 | */ | |
36c8b586 | 2802 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2803 | __releases(rq->lock) |
2804 | { | |
70b97a7f IM |
2805 | struct rq *rq = this_rq(); |
2806 | ||
4866cde0 | 2807 | finish_task_switch(rq, prev); |
da19ab51 | 2808 | |
3f029d3c GH |
2809 | /* |
2810 | * FIXME: do we need to worry about rq being invalidated by the | |
2811 | * task_switch? | |
2812 | */ | |
2813 | post_schedule(rq); | |
70b97a7f | 2814 | |
4866cde0 NP |
2815 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2816 | /* In this case, finish_task_switch does not reenable preemption */ | |
2817 | preempt_enable(); | |
2818 | #endif | |
1da177e4 | 2819 | if (current->set_child_tid) |
b488893a | 2820 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2821 | } |
2822 | ||
2823 | /* | |
2824 | * context_switch - switch to the new MM and the new | |
2825 | * thread's register state. | |
2826 | */ | |
dd41f596 | 2827 | static inline void |
70b97a7f | 2828 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2829 | struct task_struct *next) |
1da177e4 | 2830 | { |
dd41f596 | 2831 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2832 | |
e107be36 | 2833 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2834 | trace_sched_switch(prev, next); |
dd41f596 IM |
2835 | mm = next->mm; |
2836 | oldmm = prev->active_mm; | |
9226d125 ZA |
2837 | /* |
2838 | * For paravirt, this is coupled with an exit in switch_to to | |
2839 | * combine the page table reload and the switch backend into | |
2840 | * one hypercall. | |
2841 | */ | |
224101ed | 2842 | arch_start_context_switch(prev); |
9226d125 | 2843 | |
31915ab4 | 2844 | if (!mm) { |
1da177e4 LT |
2845 | next->active_mm = oldmm; |
2846 | atomic_inc(&oldmm->mm_count); | |
2847 | enter_lazy_tlb(oldmm, next); | |
2848 | } else | |
2849 | switch_mm(oldmm, mm, next); | |
2850 | ||
31915ab4 | 2851 | if (!prev->mm) { |
1da177e4 | 2852 | prev->active_mm = NULL; |
1da177e4 LT |
2853 | rq->prev_mm = oldmm; |
2854 | } | |
3a5f5e48 IM |
2855 | /* |
2856 | * Since the runqueue lock will be released by the next | |
2857 | * task (which is an invalid locking op but in the case | |
2858 | * of the scheduler it's an obvious special-case), so we | |
2859 | * do an early lockdep release here: | |
2860 | */ | |
2861 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2862 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2863 | #endif |
1da177e4 LT |
2864 | |
2865 | /* Here we just switch the register state and the stack. */ | |
2866 | switch_to(prev, next, prev); | |
2867 | ||
dd41f596 IM |
2868 | barrier(); |
2869 | /* | |
2870 | * this_rq must be evaluated again because prev may have moved | |
2871 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2872 | * frame will be invalid. | |
2873 | */ | |
2874 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2875 | } |
2876 | ||
2877 | /* | |
2878 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2879 | * | |
2880 | * externally visible scheduler statistics: current number of runnable | |
2881 | * threads, current number of uninterruptible-sleeping threads, total | |
2882 | * number of context switches performed since bootup. | |
2883 | */ | |
2884 | unsigned long nr_running(void) | |
2885 | { | |
2886 | unsigned long i, sum = 0; | |
2887 | ||
2888 | for_each_online_cpu(i) | |
2889 | sum += cpu_rq(i)->nr_running; | |
2890 | ||
2891 | return sum; | |
f711f609 | 2892 | } |
1da177e4 LT |
2893 | |
2894 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2895 | { |
1da177e4 | 2896 | unsigned long i, sum = 0; |
f711f609 | 2897 | |
0a945022 | 2898 | for_each_possible_cpu(i) |
1da177e4 | 2899 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2900 | |
2901 | /* | |
1da177e4 LT |
2902 | * Since we read the counters lockless, it might be slightly |
2903 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2904 | */ |
1da177e4 LT |
2905 | if (unlikely((long)sum < 0)) |
2906 | sum = 0; | |
f711f609 | 2907 | |
1da177e4 | 2908 | return sum; |
f711f609 | 2909 | } |
f711f609 | 2910 | |
1da177e4 | 2911 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2912 | { |
cc94abfc SR |
2913 | int i; |
2914 | unsigned long long sum = 0; | |
46cb4b7c | 2915 | |
0a945022 | 2916 | for_each_possible_cpu(i) |
1da177e4 | 2917 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2918 | |
1da177e4 LT |
2919 | return sum; |
2920 | } | |
483b4ee6 | 2921 | |
1da177e4 LT |
2922 | unsigned long nr_iowait(void) |
2923 | { | |
2924 | unsigned long i, sum = 0; | |
483b4ee6 | 2925 | |
0a945022 | 2926 | for_each_possible_cpu(i) |
1da177e4 | 2927 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2928 | |
1da177e4 LT |
2929 | return sum; |
2930 | } | |
483b4ee6 | 2931 | |
8c215bd3 | 2932 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2933 | { |
8c215bd3 | 2934 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2935 | return atomic_read(&this->nr_iowait); |
2936 | } | |
46cb4b7c | 2937 | |
69d25870 AV |
2938 | unsigned long this_cpu_load(void) |
2939 | { | |
2940 | struct rq *this = this_rq(); | |
2941 | return this->cpu_load[0]; | |
2942 | } | |
e790fb0b | 2943 | |
46cb4b7c | 2944 | |
dce48a84 TG |
2945 | /* Variables and functions for calc_load */ |
2946 | static atomic_long_t calc_load_tasks; | |
2947 | static unsigned long calc_load_update; | |
2948 | unsigned long avenrun[3]; | |
2949 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2950 | |
74f5187a PZ |
2951 | static long calc_load_fold_active(struct rq *this_rq) |
2952 | { | |
2953 | long nr_active, delta = 0; | |
2954 | ||
2955 | nr_active = this_rq->nr_running; | |
2956 | nr_active += (long) this_rq->nr_uninterruptible; | |
2957 | ||
2958 | if (nr_active != this_rq->calc_load_active) { | |
2959 | delta = nr_active - this_rq->calc_load_active; | |
2960 | this_rq->calc_load_active = nr_active; | |
2961 | } | |
2962 | ||
2963 | return delta; | |
2964 | } | |
2965 | ||
2966 | #ifdef CONFIG_NO_HZ | |
2967 | /* | |
2968 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2969 | * | |
2970 | * When making the ILB scale, we should try to pull this in as well. | |
2971 | */ | |
2972 | static atomic_long_t calc_load_tasks_idle; | |
2973 | ||
2974 | static void calc_load_account_idle(struct rq *this_rq) | |
2975 | { | |
2976 | long delta; | |
2977 | ||
2978 | delta = calc_load_fold_active(this_rq); | |
2979 | if (delta) | |
2980 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2981 | } | |
2982 | ||
2983 | static long calc_load_fold_idle(void) | |
2984 | { | |
2985 | long delta = 0; | |
2986 | ||
2987 | /* | |
2988 | * Its got a race, we don't care... | |
2989 | */ | |
2990 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2991 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2992 | ||
2993 | return delta; | |
2994 | } | |
2995 | #else | |
2996 | static void calc_load_account_idle(struct rq *this_rq) | |
2997 | { | |
2998 | } | |
2999 | ||
3000 | static inline long calc_load_fold_idle(void) | |
3001 | { | |
3002 | return 0; | |
3003 | } | |
3004 | #endif | |
3005 | ||
2d02494f TG |
3006 | /** |
3007 | * get_avenrun - get the load average array | |
3008 | * @loads: pointer to dest load array | |
3009 | * @offset: offset to add | |
3010 | * @shift: shift count to shift the result left | |
3011 | * | |
3012 | * These values are estimates at best, so no need for locking. | |
3013 | */ | |
3014 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3015 | { | |
3016 | loads[0] = (avenrun[0] + offset) << shift; | |
3017 | loads[1] = (avenrun[1] + offset) << shift; | |
3018 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3019 | } |
46cb4b7c | 3020 | |
dce48a84 TG |
3021 | static unsigned long |
3022 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3023 | { |
dce48a84 TG |
3024 | load *= exp; |
3025 | load += active * (FIXED_1 - exp); | |
3026 | return load >> FSHIFT; | |
3027 | } | |
46cb4b7c SS |
3028 | |
3029 | /* | |
dce48a84 TG |
3030 | * calc_load - update the avenrun load estimates 10 ticks after the |
3031 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3032 | */ |
dce48a84 | 3033 | void calc_global_load(void) |
7835b98b | 3034 | { |
dce48a84 TG |
3035 | unsigned long upd = calc_load_update + 10; |
3036 | long active; | |
1da177e4 | 3037 | |
dce48a84 TG |
3038 | if (time_before(jiffies, upd)) |
3039 | return; | |
1da177e4 | 3040 | |
dce48a84 TG |
3041 | active = atomic_long_read(&calc_load_tasks); |
3042 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3043 | |
dce48a84 TG |
3044 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3045 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3046 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3047 | |
dce48a84 TG |
3048 | calc_load_update += LOAD_FREQ; |
3049 | } | |
1da177e4 | 3050 | |
dce48a84 | 3051 | /* |
74f5187a PZ |
3052 | * Called from update_cpu_load() to periodically update this CPU's |
3053 | * active count. | |
dce48a84 TG |
3054 | */ |
3055 | static void calc_load_account_active(struct rq *this_rq) | |
3056 | { | |
74f5187a | 3057 | long delta; |
08c183f3 | 3058 | |
74f5187a PZ |
3059 | if (time_before(jiffies, this_rq->calc_load_update)) |
3060 | return; | |
783609c6 | 3061 | |
74f5187a PZ |
3062 | delta = calc_load_fold_active(this_rq); |
3063 | delta += calc_load_fold_idle(); | |
3064 | if (delta) | |
dce48a84 | 3065 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3066 | |
3067 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3068 | } |
3069 | ||
fdf3e95d VP |
3070 | /* |
3071 | * The exact cpuload at various idx values, calculated at every tick would be | |
3072 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3073 | * | |
3074 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3075 | * on nth tick when cpu may be busy, then we have: | |
3076 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3077 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3078 | * | |
3079 | * decay_load_missed() below does efficient calculation of | |
3080 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3081 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3082 | * | |
3083 | * The calculation is approximated on a 128 point scale. | |
3084 | * degrade_zero_ticks is the number of ticks after which load at any | |
3085 | * particular idx is approximated to be zero. | |
3086 | * degrade_factor is a precomputed table, a row for each load idx. | |
3087 | * Each column corresponds to degradation factor for a power of two ticks, | |
3088 | * based on 128 point scale. | |
3089 | * Example: | |
3090 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3091 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3092 | * | |
3093 | * With this power of 2 load factors, we can degrade the load n times | |
3094 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3095 | * n mult/shifts needed by the exact degradation. | |
3096 | */ | |
3097 | #define DEGRADE_SHIFT 7 | |
3098 | static const unsigned char | |
3099 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3100 | static const unsigned char | |
3101 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3102 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3103 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3104 | {96, 72, 40, 12, 1, 0, 0}, | |
3105 | {112, 98, 75, 43, 15, 1, 0}, | |
3106 | {120, 112, 98, 76, 45, 16, 2} }; | |
3107 | ||
3108 | /* | |
3109 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3110 | * would be when CPU is idle and so we just decay the old load without | |
3111 | * adding any new load. | |
3112 | */ | |
3113 | static unsigned long | |
3114 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3115 | { | |
3116 | int j = 0; | |
3117 | ||
3118 | if (!missed_updates) | |
3119 | return load; | |
3120 | ||
3121 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3122 | return 0; | |
3123 | ||
3124 | if (idx == 1) | |
3125 | return load >> missed_updates; | |
3126 | ||
3127 | while (missed_updates) { | |
3128 | if (missed_updates % 2) | |
3129 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3130 | ||
3131 | missed_updates >>= 1; | |
3132 | j++; | |
3133 | } | |
3134 | return load; | |
3135 | } | |
3136 | ||
46cb4b7c | 3137 | /* |
dd41f596 | 3138 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3139 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3140 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3141 | */ |
dd41f596 | 3142 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3143 | { |
495eca49 | 3144 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3145 | unsigned long curr_jiffies = jiffies; |
3146 | unsigned long pending_updates; | |
dd41f596 | 3147 | int i, scale; |
46cb4b7c | 3148 | |
dd41f596 | 3149 | this_rq->nr_load_updates++; |
46cb4b7c | 3150 | |
fdf3e95d VP |
3151 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3152 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3153 | return; | |
3154 | ||
3155 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3156 | this_rq->last_load_update_tick = curr_jiffies; | |
3157 | ||
dd41f596 | 3158 | /* Update our load: */ |
fdf3e95d VP |
3159 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3160 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3161 | unsigned long old_load, new_load; |
7d1e6a9b | 3162 | |
dd41f596 | 3163 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3164 | |
dd41f596 | 3165 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3166 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3167 | new_load = this_load; |
a25707f3 IM |
3168 | /* |
3169 | * Round up the averaging division if load is increasing. This | |
3170 | * prevents us from getting stuck on 9 if the load is 10, for | |
3171 | * example. | |
3172 | */ | |
3173 | if (new_load > old_load) | |
fdf3e95d VP |
3174 | new_load += scale - 1; |
3175 | ||
3176 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3177 | } |
da2b71ed SS |
3178 | |
3179 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3180 | } |
3181 | ||
3182 | static void update_cpu_load_active(struct rq *this_rq) | |
3183 | { | |
3184 | update_cpu_load(this_rq); | |
46cb4b7c | 3185 | |
74f5187a | 3186 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3187 | } |
3188 | ||
dd41f596 | 3189 | #ifdef CONFIG_SMP |
8a0be9ef | 3190 | |
46cb4b7c | 3191 | /* |
38022906 PZ |
3192 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3193 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3194 | */ |
38022906 | 3195 | void sched_exec(void) |
46cb4b7c | 3196 | { |
38022906 | 3197 | struct task_struct *p = current; |
1da177e4 | 3198 | unsigned long flags; |
70b97a7f | 3199 | struct rq *rq; |
0017d735 | 3200 | int dest_cpu; |
46cb4b7c | 3201 | |
1da177e4 | 3202 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3203 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3204 | if (dest_cpu == smp_processor_id()) | |
3205 | goto unlock; | |
38022906 | 3206 | |
46cb4b7c | 3207 | /* |
38022906 | 3208 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3209 | */ |
30da688e | 3210 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
969c7921 TH |
3211 | likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) { |
3212 | struct migration_arg arg = { p, dest_cpu }; | |
46cb4b7c | 3213 | |
1da177e4 | 3214 | task_rq_unlock(rq, &flags); |
969c7921 | 3215 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3216 | return; |
3217 | } | |
0017d735 | 3218 | unlock: |
1da177e4 | 3219 | task_rq_unlock(rq, &flags); |
1da177e4 | 3220 | } |
dd41f596 | 3221 | |
1da177e4 LT |
3222 | #endif |
3223 | ||
1da177e4 LT |
3224 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3225 | ||
3226 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3227 | ||
3228 | /* | |
c5f8d995 | 3229 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3230 | * @p in case that task is currently running. |
c5f8d995 HS |
3231 | * |
3232 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3233 | */ |
c5f8d995 HS |
3234 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3235 | { | |
3236 | u64 ns = 0; | |
3237 | ||
3238 | if (task_current(rq, p)) { | |
3239 | update_rq_clock(rq); | |
305e6835 | 3240 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3241 | if ((s64)ns < 0) |
3242 | ns = 0; | |
3243 | } | |
3244 | ||
3245 | return ns; | |
3246 | } | |
3247 | ||
bb34d92f | 3248 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3249 | { |
1da177e4 | 3250 | unsigned long flags; |
41b86e9c | 3251 | struct rq *rq; |
bb34d92f | 3252 | u64 ns = 0; |
48f24c4d | 3253 | |
41b86e9c | 3254 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3255 | ns = do_task_delta_exec(p, rq); |
3256 | task_rq_unlock(rq, &flags); | |
1508487e | 3257 | |
c5f8d995 HS |
3258 | return ns; |
3259 | } | |
f06febc9 | 3260 | |
c5f8d995 HS |
3261 | /* |
3262 | * Return accounted runtime for the task. | |
3263 | * In case the task is currently running, return the runtime plus current's | |
3264 | * pending runtime that have not been accounted yet. | |
3265 | */ | |
3266 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3267 | { | |
3268 | unsigned long flags; | |
3269 | struct rq *rq; | |
3270 | u64 ns = 0; | |
3271 | ||
3272 | rq = task_rq_lock(p, &flags); | |
3273 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3274 | task_rq_unlock(rq, &flags); | |
3275 | ||
3276 | return ns; | |
3277 | } | |
48f24c4d | 3278 | |
c5f8d995 HS |
3279 | /* |
3280 | * Return sum_exec_runtime for the thread group. | |
3281 | * In case the task is currently running, return the sum plus current's | |
3282 | * pending runtime that have not been accounted yet. | |
3283 | * | |
3284 | * Note that the thread group might have other running tasks as well, | |
3285 | * so the return value not includes other pending runtime that other | |
3286 | * running tasks might have. | |
3287 | */ | |
3288 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3289 | { | |
3290 | struct task_cputime totals; | |
3291 | unsigned long flags; | |
3292 | struct rq *rq; | |
3293 | u64 ns; | |
3294 | ||
3295 | rq = task_rq_lock(p, &flags); | |
3296 | thread_group_cputime(p, &totals); | |
3297 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3298 | task_rq_unlock(rq, &flags); |
48f24c4d | 3299 | |
1da177e4 LT |
3300 | return ns; |
3301 | } | |
3302 | ||
1da177e4 LT |
3303 | /* |
3304 | * Account user cpu time to a process. | |
3305 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3306 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3307 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3308 | */ |
457533a7 MS |
3309 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3310 | cputime_t cputime_scaled) | |
1da177e4 LT |
3311 | { |
3312 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3313 | cputime64_t tmp; | |
3314 | ||
457533a7 | 3315 | /* Add user time to process. */ |
1da177e4 | 3316 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3317 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3318 | account_group_user_time(p, cputime); |
1da177e4 LT |
3319 | |
3320 | /* Add user time to cpustat. */ | |
3321 | tmp = cputime_to_cputime64(cputime); | |
3322 | if (TASK_NICE(p) > 0) | |
3323 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3324 | else | |
3325 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3326 | |
3327 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3328 | /* Account for user time used */ |
3329 | acct_update_integrals(p); | |
1da177e4 LT |
3330 | } |
3331 | ||
94886b84 LV |
3332 | /* |
3333 | * Account guest cpu time to a process. | |
3334 | * @p: the process that the cpu time gets accounted to | |
3335 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3336 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3337 | */ |
457533a7 MS |
3338 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3339 | cputime_t cputime_scaled) | |
94886b84 LV |
3340 | { |
3341 | cputime64_t tmp; | |
3342 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3343 | ||
3344 | tmp = cputime_to_cputime64(cputime); | |
3345 | ||
457533a7 | 3346 | /* Add guest time to process. */ |
94886b84 | 3347 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3348 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3349 | account_group_user_time(p, cputime); |
94886b84 LV |
3350 | p->gtime = cputime_add(p->gtime, cputime); |
3351 | ||
457533a7 | 3352 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3353 | if (TASK_NICE(p) > 0) { |
3354 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3355 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3356 | } else { | |
3357 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3358 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3359 | } | |
94886b84 LV |
3360 | } |
3361 | ||
1da177e4 LT |
3362 | /* |
3363 | * Account system cpu time to a process. | |
3364 | * @p: the process that the cpu time gets accounted to | |
3365 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3366 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3367 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3368 | */ |
3369 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3370 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3371 | { |
3372 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3373 | cputime64_t tmp; |
3374 | ||
983ed7a6 | 3375 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3376 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3377 | return; |
3378 | } | |
94886b84 | 3379 | |
457533a7 | 3380 | /* Add system time to process. */ |
1da177e4 | 3381 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3382 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3383 | account_group_system_time(p, cputime); |
1da177e4 LT |
3384 | |
3385 | /* Add system time to cpustat. */ | |
3386 | tmp = cputime_to_cputime64(cputime); | |
3387 | if (hardirq_count() - hardirq_offset) | |
3388 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
75e1056f | 3389 | else if (in_serving_softirq()) |
1da177e4 | 3390 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); |
1da177e4 | 3391 | else |
79741dd3 MS |
3392 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3393 | ||
ef12fefa BR |
3394 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3395 | ||
1da177e4 LT |
3396 | /* Account for system time used */ |
3397 | acct_update_integrals(p); | |
1da177e4 LT |
3398 | } |
3399 | ||
c66f08be | 3400 | /* |
1da177e4 | 3401 | * Account for involuntary wait time. |
1da177e4 | 3402 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3403 | */ |
79741dd3 | 3404 | void account_steal_time(cputime_t cputime) |
c66f08be | 3405 | { |
79741dd3 MS |
3406 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3407 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3408 | ||
3409 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3410 | } |
3411 | ||
1da177e4 | 3412 | /* |
79741dd3 MS |
3413 | * Account for idle time. |
3414 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3415 | */ |
79741dd3 | 3416 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3417 | { |
3418 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3419 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3420 | struct rq *rq = this_rq(); |
1da177e4 | 3421 | |
79741dd3 MS |
3422 | if (atomic_read(&rq->nr_iowait) > 0) |
3423 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3424 | else | |
3425 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3426 | } |
3427 | ||
79741dd3 MS |
3428 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3429 | ||
3430 | /* | |
3431 | * Account a single tick of cpu time. | |
3432 | * @p: the process that the cpu time gets accounted to | |
3433 | * @user_tick: indicates if the tick is a user or a system tick | |
3434 | */ | |
3435 | void account_process_tick(struct task_struct *p, int user_tick) | |
3436 | { | |
a42548a1 | 3437 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3438 | struct rq *rq = this_rq(); |
3439 | ||
3440 | if (user_tick) | |
a42548a1 | 3441 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3442 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3443 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3444 | one_jiffy_scaled); |
3445 | else | |
a42548a1 | 3446 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3447 | } |
3448 | ||
3449 | /* | |
3450 | * Account multiple ticks of steal time. | |
3451 | * @p: the process from which the cpu time has been stolen | |
3452 | * @ticks: number of stolen ticks | |
3453 | */ | |
3454 | void account_steal_ticks(unsigned long ticks) | |
3455 | { | |
3456 | account_steal_time(jiffies_to_cputime(ticks)); | |
3457 | } | |
3458 | ||
3459 | /* | |
3460 | * Account multiple ticks of idle time. | |
3461 | * @ticks: number of stolen ticks | |
3462 | */ | |
3463 | void account_idle_ticks(unsigned long ticks) | |
3464 | { | |
3465 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3466 | } |
3467 | ||
79741dd3 MS |
3468 | #endif |
3469 | ||
49048622 BS |
3470 | /* |
3471 | * Use precise platform statistics if available: | |
3472 | */ | |
3473 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3474 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3475 | { |
d99ca3b9 HS |
3476 | *ut = p->utime; |
3477 | *st = p->stime; | |
49048622 BS |
3478 | } |
3479 | ||
0cf55e1e | 3480 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3481 | { |
0cf55e1e HS |
3482 | struct task_cputime cputime; |
3483 | ||
3484 | thread_group_cputime(p, &cputime); | |
3485 | ||
3486 | *ut = cputime.utime; | |
3487 | *st = cputime.stime; | |
49048622 BS |
3488 | } |
3489 | #else | |
761b1d26 HS |
3490 | |
3491 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3492 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3493 | #endif |
3494 | ||
d180c5bc | 3495 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3496 | { |
d99ca3b9 | 3497 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3498 | |
3499 | /* | |
3500 | * Use CFS's precise accounting: | |
3501 | */ | |
d180c5bc | 3502 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3503 | |
3504 | if (total) { | |
e75e863d | 3505 | u64 temp = rtime; |
d180c5bc | 3506 | |
e75e863d | 3507 | temp *= utime; |
49048622 | 3508 | do_div(temp, total); |
d180c5bc HS |
3509 | utime = (cputime_t)temp; |
3510 | } else | |
3511 | utime = rtime; | |
49048622 | 3512 | |
d180c5bc HS |
3513 | /* |
3514 | * Compare with previous values, to keep monotonicity: | |
3515 | */ | |
761b1d26 | 3516 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3517 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3518 | |
d99ca3b9 HS |
3519 | *ut = p->prev_utime; |
3520 | *st = p->prev_stime; | |
49048622 BS |
3521 | } |
3522 | ||
0cf55e1e HS |
3523 | /* |
3524 | * Must be called with siglock held. | |
3525 | */ | |
3526 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3527 | { |
0cf55e1e HS |
3528 | struct signal_struct *sig = p->signal; |
3529 | struct task_cputime cputime; | |
3530 | cputime_t rtime, utime, total; | |
49048622 | 3531 | |
0cf55e1e | 3532 | thread_group_cputime(p, &cputime); |
49048622 | 3533 | |
0cf55e1e HS |
3534 | total = cputime_add(cputime.utime, cputime.stime); |
3535 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3536 | |
0cf55e1e | 3537 | if (total) { |
e75e863d | 3538 | u64 temp = rtime; |
49048622 | 3539 | |
e75e863d | 3540 | temp *= cputime.utime; |
0cf55e1e HS |
3541 | do_div(temp, total); |
3542 | utime = (cputime_t)temp; | |
3543 | } else | |
3544 | utime = rtime; | |
3545 | ||
3546 | sig->prev_utime = max(sig->prev_utime, utime); | |
3547 | sig->prev_stime = max(sig->prev_stime, | |
3548 | cputime_sub(rtime, sig->prev_utime)); | |
3549 | ||
3550 | *ut = sig->prev_utime; | |
3551 | *st = sig->prev_stime; | |
49048622 | 3552 | } |
49048622 | 3553 | #endif |
49048622 | 3554 | |
7835b98b CL |
3555 | /* |
3556 | * This function gets called by the timer code, with HZ frequency. | |
3557 | * We call it with interrupts disabled. | |
3558 | * | |
3559 | * It also gets called by the fork code, when changing the parent's | |
3560 | * timeslices. | |
3561 | */ | |
3562 | void scheduler_tick(void) | |
3563 | { | |
7835b98b CL |
3564 | int cpu = smp_processor_id(); |
3565 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3566 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3567 | |
3568 | sched_clock_tick(); | |
dd41f596 | 3569 | |
05fa785c | 3570 | raw_spin_lock(&rq->lock); |
3e51f33f | 3571 | update_rq_clock(rq); |
fdf3e95d | 3572 | update_cpu_load_active(rq); |
fa85ae24 | 3573 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3574 | raw_spin_unlock(&rq->lock); |
7835b98b | 3575 | |
e9d2b064 | 3576 | perf_event_task_tick(); |
e220d2dc | 3577 | |
e418e1c2 | 3578 | #ifdef CONFIG_SMP |
dd41f596 IM |
3579 | rq->idle_at_tick = idle_cpu(cpu); |
3580 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3581 | #endif |
1da177e4 LT |
3582 | } |
3583 | ||
132380a0 | 3584 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3585 | { |
3586 | if (in_lock_functions(addr)) { | |
3587 | addr = CALLER_ADDR2; | |
3588 | if (in_lock_functions(addr)) | |
3589 | addr = CALLER_ADDR3; | |
3590 | } | |
3591 | return addr; | |
3592 | } | |
1da177e4 | 3593 | |
7e49fcce SR |
3594 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3595 | defined(CONFIG_PREEMPT_TRACER)) | |
3596 | ||
43627582 | 3597 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3598 | { |
6cd8a4bb | 3599 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3600 | /* |
3601 | * Underflow? | |
3602 | */ | |
9a11b49a IM |
3603 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3604 | return; | |
6cd8a4bb | 3605 | #endif |
1da177e4 | 3606 | preempt_count() += val; |
6cd8a4bb | 3607 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3608 | /* |
3609 | * Spinlock count overflowing soon? | |
3610 | */ | |
33859f7f MOS |
3611 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3612 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3613 | #endif |
3614 | if (preempt_count() == val) | |
3615 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3616 | } |
3617 | EXPORT_SYMBOL(add_preempt_count); | |
3618 | ||
43627582 | 3619 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3620 | { |
6cd8a4bb | 3621 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3622 | /* |
3623 | * Underflow? | |
3624 | */ | |
01e3eb82 | 3625 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3626 | return; |
1da177e4 LT |
3627 | /* |
3628 | * Is the spinlock portion underflowing? | |
3629 | */ | |
9a11b49a IM |
3630 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3631 | !(preempt_count() & PREEMPT_MASK))) | |
3632 | return; | |
6cd8a4bb | 3633 | #endif |
9a11b49a | 3634 | |
6cd8a4bb SR |
3635 | if (preempt_count() == val) |
3636 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3637 | preempt_count() -= val; |
3638 | } | |
3639 | EXPORT_SYMBOL(sub_preempt_count); | |
3640 | ||
3641 | #endif | |
3642 | ||
3643 | /* | |
dd41f596 | 3644 | * Print scheduling while atomic bug: |
1da177e4 | 3645 | */ |
dd41f596 | 3646 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3647 | { |
838225b4 SS |
3648 | struct pt_regs *regs = get_irq_regs(); |
3649 | ||
3df0fc5b PZ |
3650 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3651 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3652 | |
dd41f596 | 3653 | debug_show_held_locks(prev); |
e21f5b15 | 3654 | print_modules(); |
dd41f596 IM |
3655 | if (irqs_disabled()) |
3656 | print_irqtrace_events(prev); | |
838225b4 SS |
3657 | |
3658 | if (regs) | |
3659 | show_regs(regs); | |
3660 | else | |
3661 | dump_stack(); | |
dd41f596 | 3662 | } |
1da177e4 | 3663 | |
dd41f596 IM |
3664 | /* |
3665 | * Various schedule()-time debugging checks and statistics: | |
3666 | */ | |
3667 | static inline void schedule_debug(struct task_struct *prev) | |
3668 | { | |
1da177e4 | 3669 | /* |
41a2d6cf | 3670 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3671 | * schedule() atomically, we ignore that path for now. |
3672 | * Otherwise, whine if we are scheduling when we should not be. | |
3673 | */ | |
3f33a7ce | 3674 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3675 | __schedule_bug(prev); |
3676 | ||
1da177e4 LT |
3677 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3678 | ||
2d72376b | 3679 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3680 | #ifdef CONFIG_SCHEDSTATS |
3681 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3682 | schedstat_inc(this_rq(), bkl_count); |
3683 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3684 | } |
3685 | #endif | |
dd41f596 IM |
3686 | } |
3687 | ||
6cecd084 | 3688 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3689 | { |
a64692a3 MG |
3690 | if (prev->se.on_rq) |
3691 | update_rq_clock(rq); | |
3692 | rq->skip_clock_update = 0; | |
6cecd084 | 3693 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3694 | } |
3695 | ||
dd41f596 IM |
3696 | /* |
3697 | * Pick up the highest-prio task: | |
3698 | */ | |
3699 | static inline struct task_struct * | |
b67802ea | 3700 | pick_next_task(struct rq *rq) |
dd41f596 | 3701 | { |
5522d5d5 | 3702 | const struct sched_class *class; |
dd41f596 | 3703 | struct task_struct *p; |
1da177e4 LT |
3704 | |
3705 | /* | |
dd41f596 IM |
3706 | * Optimization: we know that if all tasks are in |
3707 | * the fair class we can call that function directly: | |
1da177e4 | 3708 | */ |
dd41f596 | 3709 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3710 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3711 | if (likely(p)) |
3712 | return p; | |
1da177e4 LT |
3713 | } |
3714 | ||
34f971f6 | 3715 | for_each_class(class) { |
fb8d4724 | 3716 | p = class->pick_next_task(rq); |
dd41f596 IM |
3717 | if (p) |
3718 | return p; | |
dd41f596 | 3719 | } |
34f971f6 PZ |
3720 | |
3721 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3722 | } |
1da177e4 | 3723 | |
dd41f596 IM |
3724 | /* |
3725 | * schedule() is the main scheduler function. | |
3726 | */ | |
ff743345 | 3727 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3728 | { |
3729 | struct task_struct *prev, *next; | |
67ca7bde | 3730 | unsigned long *switch_count; |
dd41f596 | 3731 | struct rq *rq; |
31656519 | 3732 | int cpu; |
dd41f596 | 3733 | |
ff743345 PZ |
3734 | need_resched: |
3735 | preempt_disable(); | |
dd41f596 IM |
3736 | cpu = smp_processor_id(); |
3737 | rq = cpu_rq(cpu); | |
25502a6c | 3738 | rcu_note_context_switch(cpu); |
dd41f596 | 3739 | prev = rq->curr; |
dd41f596 IM |
3740 | |
3741 | release_kernel_lock(prev); | |
3742 | need_resched_nonpreemptible: | |
3743 | ||
3744 | schedule_debug(prev); | |
1da177e4 | 3745 | |
31656519 | 3746 | if (sched_feat(HRTICK)) |
f333fdc9 | 3747 | hrtick_clear(rq); |
8f4d37ec | 3748 | |
05fa785c | 3749 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3750 | clear_tsk_need_resched(prev); |
1da177e4 | 3751 | |
246d86b5 | 3752 | switch_count = &prev->nivcsw; |
1da177e4 | 3753 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3754 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3755 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
3756 | } else { |
3757 | /* | |
3758 | * If a worker is going to sleep, notify and | |
3759 | * ask workqueue whether it wants to wake up a | |
3760 | * task to maintain concurrency. If so, wake | |
3761 | * up the task. | |
3762 | */ | |
3763 | if (prev->flags & PF_WQ_WORKER) { | |
3764 | struct task_struct *to_wakeup; | |
3765 | ||
3766 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3767 | if (to_wakeup) | |
3768 | try_to_wake_up_local(to_wakeup); | |
3769 | } | |
371fd7e7 | 3770 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 3771 | } |
dd41f596 | 3772 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3773 | } |
3774 | ||
3f029d3c | 3775 | pre_schedule(rq, prev); |
f65eda4f | 3776 | |
dd41f596 | 3777 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3778 | idle_balance(cpu, rq); |
1da177e4 | 3779 | |
df1c99d4 | 3780 | put_prev_task(rq, prev); |
b67802ea | 3781 | next = pick_next_task(rq); |
1da177e4 | 3782 | |
1da177e4 | 3783 | if (likely(prev != next)) { |
673a90a1 | 3784 | sched_info_switch(prev, next); |
49f47433 | 3785 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3786 | |
1da177e4 LT |
3787 | rq->nr_switches++; |
3788 | rq->curr = next; | |
3789 | ++*switch_count; | |
3790 | ||
dd41f596 | 3791 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3792 | /* |
246d86b5 ON |
3793 | * The context switch have flipped the stack from under us |
3794 | * and restored the local variables which were saved when | |
3795 | * this task called schedule() in the past. prev == current | |
3796 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3797 | */ |
3798 | cpu = smp_processor_id(); | |
3799 | rq = cpu_rq(cpu); | |
1da177e4 | 3800 | } else |
05fa785c | 3801 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3802 | |
3f029d3c | 3803 | post_schedule(rq); |
1da177e4 | 3804 | |
246d86b5 | 3805 | if (unlikely(reacquire_kernel_lock(prev))) |
1da177e4 | 3806 | goto need_resched_nonpreemptible; |
8f4d37ec | 3807 | |
1da177e4 | 3808 | preempt_enable_no_resched(); |
ff743345 | 3809 | if (need_resched()) |
1da177e4 LT |
3810 | goto need_resched; |
3811 | } | |
1da177e4 LT |
3812 | EXPORT_SYMBOL(schedule); |
3813 | ||
c08f7829 | 3814 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3815 | /* |
3816 | * Look out! "owner" is an entirely speculative pointer | |
3817 | * access and not reliable. | |
3818 | */ | |
3819 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3820 | { | |
3821 | unsigned int cpu; | |
3822 | struct rq *rq; | |
3823 | ||
3824 | if (!sched_feat(OWNER_SPIN)) | |
3825 | return 0; | |
3826 | ||
3827 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3828 | /* | |
3829 | * Need to access the cpu field knowing that | |
3830 | * DEBUG_PAGEALLOC could have unmapped it if | |
3831 | * the mutex owner just released it and exited. | |
3832 | */ | |
3833 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 3834 | return 0; |
0d66bf6d PZ |
3835 | #else |
3836 | cpu = owner->cpu; | |
3837 | #endif | |
3838 | ||
3839 | /* | |
3840 | * Even if the access succeeded (likely case), | |
3841 | * the cpu field may no longer be valid. | |
3842 | */ | |
3843 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 3844 | return 0; |
0d66bf6d PZ |
3845 | |
3846 | /* | |
3847 | * We need to validate that we can do a | |
3848 | * get_cpu() and that we have the percpu area. | |
3849 | */ | |
3850 | if (!cpu_online(cpu)) | |
4b402210 | 3851 | return 0; |
0d66bf6d PZ |
3852 | |
3853 | rq = cpu_rq(cpu); | |
3854 | ||
3855 | for (;;) { | |
3856 | /* | |
3857 | * Owner changed, break to re-assess state. | |
3858 | */ | |
9d0f4dcc TC |
3859 | if (lock->owner != owner) { |
3860 | /* | |
3861 | * If the lock has switched to a different owner, | |
3862 | * we likely have heavy contention. Return 0 to quit | |
3863 | * optimistic spinning and not contend further: | |
3864 | */ | |
3865 | if (lock->owner) | |
3866 | return 0; | |
0d66bf6d | 3867 | break; |
9d0f4dcc | 3868 | } |
0d66bf6d PZ |
3869 | |
3870 | /* | |
3871 | * Is that owner really running on that cpu? | |
3872 | */ | |
3873 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3874 | return 0; | |
3875 | ||
3876 | cpu_relax(); | |
3877 | } | |
4b402210 | 3878 | |
0d66bf6d PZ |
3879 | return 1; |
3880 | } | |
3881 | #endif | |
3882 | ||
1da177e4 LT |
3883 | #ifdef CONFIG_PREEMPT |
3884 | /* | |
2ed6e34f | 3885 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3886 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3887 | * occur there and call schedule directly. |
3888 | */ | |
d1f74e20 | 3889 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3890 | { |
3891 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3892 | |
1da177e4 LT |
3893 | /* |
3894 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3895 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3896 | */ |
beed33a8 | 3897 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3898 | return; |
3899 | ||
3a5c359a | 3900 | do { |
d1f74e20 | 3901 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 3902 | schedule(); |
d1f74e20 | 3903 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3904 | |
3a5c359a AK |
3905 | /* |
3906 | * Check again in case we missed a preemption opportunity | |
3907 | * between schedule and now. | |
3908 | */ | |
3909 | barrier(); | |
5ed0cec0 | 3910 | } while (need_resched()); |
1da177e4 | 3911 | } |
1da177e4 LT |
3912 | EXPORT_SYMBOL(preempt_schedule); |
3913 | ||
3914 | /* | |
2ed6e34f | 3915 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3916 | * off of irq context. |
3917 | * Note, that this is called and return with irqs disabled. This will | |
3918 | * protect us against recursive calling from irq. | |
3919 | */ | |
3920 | asmlinkage void __sched preempt_schedule_irq(void) | |
3921 | { | |
3922 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3923 | |
2ed6e34f | 3924 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3925 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3926 | ||
3a5c359a AK |
3927 | do { |
3928 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3929 | local_irq_enable(); |
3930 | schedule(); | |
3931 | local_irq_disable(); | |
3a5c359a | 3932 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3933 | |
3a5c359a AK |
3934 | /* |
3935 | * Check again in case we missed a preemption opportunity | |
3936 | * between schedule and now. | |
3937 | */ | |
3938 | barrier(); | |
5ed0cec0 | 3939 | } while (need_resched()); |
1da177e4 LT |
3940 | } |
3941 | ||
3942 | #endif /* CONFIG_PREEMPT */ | |
3943 | ||
63859d4f | 3944 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3945 | void *key) |
1da177e4 | 3946 | { |
63859d4f | 3947 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3948 | } |
1da177e4 LT |
3949 | EXPORT_SYMBOL(default_wake_function); |
3950 | ||
3951 | /* | |
41a2d6cf IM |
3952 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3953 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3954 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3955 | * | |
3956 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3957 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3958 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3959 | */ | |
78ddb08f | 3960 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3961 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3962 | { |
2e45874c | 3963 | wait_queue_t *curr, *next; |
1da177e4 | 3964 | |
2e45874c | 3965 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3966 | unsigned flags = curr->flags; |
3967 | ||
63859d4f | 3968 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3969 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3970 | break; |
3971 | } | |
3972 | } | |
3973 | ||
3974 | /** | |
3975 | * __wake_up - wake up threads blocked on a waitqueue. | |
3976 | * @q: the waitqueue | |
3977 | * @mode: which threads | |
3978 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3979 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3980 | * |
3981 | * It may be assumed that this function implies a write memory barrier before | |
3982 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3983 | */ |
7ad5b3a5 | 3984 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3985 | int nr_exclusive, void *key) |
1da177e4 LT |
3986 | { |
3987 | unsigned long flags; | |
3988 | ||
3989 | spin_lock_irqsave(&q->lock, flags); | |
3990 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3991 | spin_unlock_irqrestore(&q->lock, flags); | |
3992 | } | |
1da177e4 LT |
3993 | EXPORT_SYMBOL(__wake_up); |
3994 | ||
3995 | /* | |
3996 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3997 | */ | |
7ad5b3a5 | 3998 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3999 | { |
4000 | __wake_up_common(q, mode, 1, 0, NULL); | |
4001 | } | |
22c43c81 | 4002 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4003 | |
4ede816a DL |
4004 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4005 | { | |
4006 | __wake_up_common(q, mode, 1, 0, key); | |
4007 | } | |
4008 | ||
1da177e4 | 4009 | /** |
4ede816a | 4010 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4011 | * @q: the waitqueue |
4012 | * @mode: which threads | |
4013 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4014 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4015 | * |
4016 | * The sync wakeup differs that the waker knows that it will schedule | |
4017 | * away soon, so while the target thread will be woken up, it will not | |
4018 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4019 | * with each other. This can prevent needless bouncing between CPUs. | |
4020 | * | |
4021 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4022 | * |
4023 | * It may be assumed that this function implies a write memory barrier before | |
4024 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4025 | */ |
4ede816a DL |
4026 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4027 | int nr_exclusive, void *key) | |
1da177e4 LT |
4028 | { |
4029 | unsigned long flags; | |
7d478721 | 4030 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4031 | |
4032 | if (unlikely(!q)) | |
4033 | return; | |
4034 | ||
4035 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4036 | wake_flags = 0; |
1da177e4 LT |
4037 | |
4038 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4039 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4040 | spin_unlock_irqrestore(&q->lock, flags); |
4041 | } | |
4ede816a DL |
4042 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4043 | ||
4044 | /* | |
4045 | * __wake_up_sync - see __wake_up_sync_key() | |
4046 | */ | |
4047 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4048 | { | |
4049 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4050 | } | |
1da177e4 LT |
4051 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4052 | ||
65eb3dc6 KD |
4053 | /** |
4054 | * complete: - signals a single thread waiting on this completion | |
4055 | * @x: holds the state of this particular completion | |
4056 | * | |
4057 | * This will wake up a single thread waiting on this completion. Threads will be | |
4058 | * awakened in the same order in which they were queued. | |
4059 | * | |
4060 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4061 | * |
4062 | * It may be assumed that this function implies a write memory barrier before | |
4063 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4064 | */ |
b15136e9 | 4065 | void complete(struct completion *x) |
1da177e4 LT |
4066 | { |
4067 | unsigned long flags; | |
4068 | ||
4069 | spin_lock_irqsave(&x->wait.lock, flags); | |
4070 | x->done++; | |
d9514f6c | 4071 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4072 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4073 | } | |
4074 | EXPORT_SYMBOL(complete); | |
4075 | ||
65eb3dc6 KD |
4076 | /** |
4077 | * complete_all: - signals all threads waiting on this completion | |
4078 | * @x: holds the state of this particular completion | |
4079 | * | |
4080 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4081 | * |
4082 | * It may be assumed that this function implies a write memory barrier before | |
4083 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4084 | */ |
b15136e9 | 4085 | void complete_all(struct completion *x) |
1da177e4 LT |
4086 | { |
4087 | unsigned long flags; | |
4088 | ||
4089 | spin_lock_irqsave(&x->wait.lock, flags); | |
4090 | x->done += UINT_MAX/2; | |
d9514f6c | 4091 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4092 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4093 | } | |
4094 | EXPORT_SYMBOL(complete_all); | |
4095 | ||
8cbbe86d AK |
4096 | static inline long __sched |
4097 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4098 | { |
1da177e4 LT |
4099 | if (!x->done) { |
4100 | DECLARE_WAITQUEUE(wait, current); | |
4101 | ||
a93d2f17 | 4102 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4103 | do { |
94d3d824 | 4104 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4105 | timeout = -ERESTARTSYS; |
4106 | break; | |
8cbbe86d AK |
4107 | } |
4108 | __set_current_state(state); | |
1da177e4 LT |
4109 | spin_unlock_irq(&x->wait.lock); |
4110 | timeout = schedule_timeout(timeout); | |
4111 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4112 | } while (!x->done && timeout); |
1da177e4 | 4113 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4114 | if (!x->done) |
4115 | return timeout; | |
1da177e4 LT |
4116 | } |
4117 | x->done--; | |
ea71a546 | 4118 | return timeout ?: 1; |
1da177e4 | 4119 | } |
1da177e4 | 4120 | |
8cbbe86d AK |
4121 | static long __sched |
4122 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4123 | { |
1da177e4 LT |
4124 | might_sleep(); |
4125 | ||
4126 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4127 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4128 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4129 | return timeout; |
4130 | } | |
1da177e4 | 4131 | |
65eb3dc6 KD |
4132 | /** |
4133 | * wait_for_completion: - waits for completion of a task | |
4134 | * @x: holds the state of this particular completion | |
4135 | * | |
4136 | * This waits to be signaled for completion of a specific task. It is NOT | |
4137 | * interruptible and there is no timeout. | |
4138 | * | |
4139 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4140 | * and interrupt capability. Also see complete(). | |
4141 | */ | |
b15136e9 | 4142 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4143 | { |
4144 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4145 | } |
8cbbe86d | 4146 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4147 | |
65eb3dc6 KD |
4148 | /** |
4149 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4150 | * @x: holds the state of this particular completion | |
4151 | * @timeout: timeout value in jiffies | |
4152 | * | |
4153 | * This waits for either a completion of a specific task to be signaled or for a | |
4154 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4155 | * interruptible. | |
4156 | */ | |
b15136e9 | 4157 | unsigned long __sched |
8cbbe86d | 4158 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4159 | { |
8cbbe86d | 4160 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4161 | } |
8cbbe86d | 4162 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4163 | |
65eb3dc6 KD |
4164 | /** |
4165 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4166 | * @x: holds the state of this particular completion | |
4167 | * | |
4168 | * This waits for completion of a specific task to be signaled. It is | |
4169 | * interruptible. | |
4170 | */ | |
8cbbe86d | 4171 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4172 | { |
51e97990 AK |
4173 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4174 | if (t == -ERESTARTSYS) | |
4175 | return t; | |
4176 | return 0; | |
0fec171c | 4177 | } |
8cbbe86d | 4178 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4179 | |
65eb3dc6 KD |
4180 | /** |
4181 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4182 | * @x: holds the state of this particular completion | |
4183 | * @timeout: timeout value in jiffies | |
4184 | * | |
4185 | * This waits for either a completion of a specific task to be signaled or for a | |
4186 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4187 | */ | |
b15136e9 | 4188 | unsigned long __sched |
8cbbe86d AK |
4189 | wait_for_completion_interruptible_timeout(struct completion *x, |
4190 | unsigned long timeout) | |
0fec171c | 4191 | { |
8cbbe86d | 4192 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4193 | } |
8cbbe86d | 4194 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4195 | |
65eb3dc6 KD |
4196 | /** |
4197 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4198 | * @x: holds the state of this particular completion | |
4199 | * | |
4200 | * This waits to be signaled for completion of a specific task. It can be | |
4201 | * interrupted by a kill signal. | |
4202 | */ | |
009e577e MW |
4203 | int __sched wait_for_completion_killable(struct completion *x) |
4204 | { | |
4205 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4206 | if (t == -ERESTARTSYS) | |
4207 | return t; | |
4208 | return 0; | |
4209 | } | |
4210 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4211 | ||
0aa12fb4 SW |
4212 | /** |
4213 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4214 | * @x: holds the state of this particular completion | |
4215 | * @timeout: timeout value in jiffies | |
4216 | * | |
4217 | * This waits for either a completion of a specific task to be | |
4218 | * signaled or for a specified timeout to expire. It can be | |
4219 | * interrupted by a kill signal. The timeout is in jiffies. | |
4220 | */ | |
4221 | unsigned long __sched | |
4222 | wait_for_completion_killable_timeout(struct completion *x, | |
4223 | unsigned long timeout) | |
4224 | { | |
4225 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4226 | } | |
4227 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4228 | ||
be4de352 DC |
4229 | /** |
4230 | * try_wait_for_completion - try to decrement a completion without blocking | |
4231 | * @x: completion structure | |
4232 | * | |
4233 | * Returns: 0 if a decrement cannot be done without blocking | |
4234 | * 1 if a decrement succeeded. | |
4235 | * | |
4236 | * If a completion is being used as a counting completion, | |
4237 | * attempt to decrement the counter without blocking. This | |
4238 | * enables us to avoid waiting if the resource the completion | |
4239 | * is protecting is not available. | |
4240 | */ | |
4241 | bool try_wait_for_completion(struct completion *x) | |
4242 | { | |
7539a3b3 | 4243 | unsigned long flags; |
be4de352 DC |
4244 | int ret = 1; |
4245 | ||
7539a3b3 | 4246 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4247 | if (!x->done) |
4248 | ret = 0; | |
4249 | else | |
4250 | x->done--; | |
7539a3b3 | 4251 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4252 | return ret; |
4253 | } | |
4254 | EXPORT_SYMBOL(try_wait_for_completion); | |
4255 | ||
4256 | /** | |
4257 | * completion_done - Test to see if a completion has any waiters | |
4258 | * @x: completion structure | |
4259 | * | |
4260 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4261 | * 1 if there are no waiters. | |
4262 | * | |
4263 | */ | |
4264 | bool completion_done(struct completion *x) | |
4265 | { | |
7539a3b3 | 4266 | unsigned long flags; |
be4de352 DC |
4267 | int ret = 1; |
4268 | ||
7539a3b3 | 4269 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4270 | if (!x->done) |
4271 | ret = 0; | |
7539a3b3 | 4272 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4273 | return ret; |
4274 | } | |
4275 | EXPORT_SYMBOL(completion_done); | |
4276 | ||
8cbbe86d AK |
4277 | static long __sched |
4278 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4279 | { |
0fec171c IM |
4280 | unsigned long flags; |
4281 | wait_queue_t wait; | |
4282 | ||
4283 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4284 | |
8cbbe86d | 4285 | __set_current_state(state); |
1da177e4 | 4286 | |
8cbbe86d AK |
4287 | spin_lock_irqsave(&q->lock, flags); |
4288 | __add_wait_queue(q, &wait); | |
4289 | spin_unlock(&q->lock); | |
4290 | timeout = schedule_timeout(timeout); | |
4291 | spin_lock_irq(&q->lock); | |
4292 | __remove_wait_queue(q, &wait); | |
4293 | spin_unlock_irqrestore(&q->lock, flags); | |
4294 | ||
4295 | return timeout; | |
4296 | } | |
4297 | ||
4298 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4299 | { | |
4300 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4301 | } |
1da177e4 LT |
4302 | EXPORT_SYMBOL(interruptible_sleep_on); |
4303 | ||
0fec171c | 4304 | long __sched |
95cdf3b7 | 4305 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4306 | { |
8cbbe86d | 4307 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4308 | } |
1da177e4 LT |
4309 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4310 | ||
0fec171c | 4311 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4312 | { |
8cbbe86d | 4313 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4314 | } |
1da177e4 LT |
4315 | EXPORT_SYMBOL(sleep_on); |
4316 | ||
0fec171c | 4317 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4318 | { |
8cbbe86d | 4319 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4320 | } |
1da177e4 LT |
4321 | EXPORT_SYMBOL(sleep_on_timeout); |
4322 | ||
b29739f9 IM |
4323 | #ifdef CONFIG_RT_MUTEXES |
4324 | ||
4325 | /* | |
4326 | * rt_mutex_setprio - set the current priority of a task | |
4327 | * @p: task | |
4328 | * @prio: prio value (kernel-internal form) | |
4329 | * | |
4330 | * This function changes the 'effective' priority of a task. It does | |
4331 | * not touch ->normal_prio like __setscheduler(). | |
4332 | * | |
4333 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4334 | */ | |
36c8b586 | 4335 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4336 | { |
4337 | unsigned long flags; | |
83b699ed | 4338 | int oldprio, on_rq, running; |
70b97a7f | 4339 | struct rq *rq; |
83ab0aa0 | 4340 | const struct sched_class *prev_class; |
b29739f9 IM |
4341 | |
4342 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4343 | ||
4344 | rq = task_rq_lock(p, &flags); | |
4345 | ||
a8027073 | 4346 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4347 | oldprio = p->prio; |
83ab0aa0 | 4348 | prev_class = p->sched_class; |
dd41f596 | 4349 | on_rq = p->se.on_rq; |
051a1d1a | 4350 | running = task_current(rq, p); |
0e1f3483 | 4351 | if (on_rq) |
69be72c1 | 4352 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4353 | if (running) |
4354 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4355 | |
4356 | if (rt_prio(prio)) | |
4357 | p->sched_class = &rt_sched_class; | |
4358 | else | |
4359 | p->sched_class = &fair_sched_class; | |
4360 | ||
b29739f9 IM |
4361 | p->prio = prio; |
4362 | ||
0e1f3483 HS |
4363 | if (running) |
4364 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4365 | if (on_rq) { |
371fd7e7 | 4366 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4367 | |
4368 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4369 | } |
4370 | task_rq_unlock(rq, &flags); | |
4371 | } | |
4372 | ||
4373 | #endif | |
4374 | ||
36c8b586 | 4375 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4376 | { |
dd41f596 | 4377 | int old_prio, delta, on_rq; |
1da177e4 | 4378 | unsigned long flags; |
70b97a7f | 4379 | struct rq *rq; |
1da177e4 LT |
4380 | |
4381 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4382 | return; | |
4383 | /* | |
4384 | * We have to be careful, if called from sys_setpriority(), | |
4385 | * the task might be in the middle of scheduling on another CPU. | |
4386 | */ | |
4387 | rq = task_rq_lock(p, &flags); | |
4388 | /* | |
4389 | * The RT priorities are set via sched_setscheduler(), but we still | |
4390 | * allow the 'normal' nice value to be set - but as expected | |
4391 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4392 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4393 | */ |
e05606d3 | 4394 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4395 | p->static_prio = NICE_TO_PRIO(nice); |
4396 | goto out_unlock; | |
4397 | } | |
dd41f596 | 4398 | on_rq = p->se.on_rq; |
c09595f6 | 4399 | if (on_rq) |
69be72c1 | 4400 | dequeue_task(rq, p, 0); |
1da177e4 | 4401 | |
1da177e4 | 4402 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4403 | set_load_weight(p); |
b29739f9 IM |
4404 | old_prio = p->prio; |
4405 | p->prio = effective_prio(p); | |
4406 | delta = p->prio - old_prio; | |
1da177e4 | 4407 | |
dd41f596 | 4408 | if (on_rq) { |
371fd7e7 | 4409 | enqueue_task(rq, p, 0); |
1da177e4 | 4410 | /* |
d5f9f942 AM |
4411 | * If the task increased its priority or is running and |
4412 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4413 | */ |
d5f9f942 | 4414 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4415 | resched_task(rq->curr); |
4416 | } | |
4417 | out_unlock: | |
4418 | task_rq_unlock(rq, &flags); | |
4419 | } | |
1da177e4 LT |
4420 | EXPORT_SYMBOL(set_user_nice); |
4421 | ||
e43379f1 MM |
4422 | /* |
4423 | * can_nice - check if a task can reduce its nice value | |
4424 | * @p: task | |
4425 | * @nice: nice value | |
4426 | */ | |
36c8b586 | 4427 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4428 | { |
024f4747 MM |
4429 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4430 | int nice_rlim = 20 - nice; | |
48f24c4d | 4431 | |
78d7d407 | 4432 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4433 | capable(CAP_SYS_NICE)); |
4434 | } | |
4435 | ||
1da177e4 LT |
4436 | #ifdef __ARCH_WANT_SYS_NICE |
4437 | ||
4438 | /* | |
4439 | * sys_nice - change the priority of the current process. | |
4440 | * @increment: priority increment | |
4441 | * | |
4442 | * sys_setpriority is a more generic, but much slower function that | |
4443 | * does similar things. | |
4444 | */ | |
5add95d4 | 4445 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4446 | { |
48f24c4d | 4447 | long nice, retval; |
1da177e4 LT |
4448 | |
4449 | /* | |
4450 | * Setpriority might change our priority at the same moment. | |
4451 | * We don't have to worry. Conceptually one call occurs first | |
4452 | * and we have a single winner. | |
4453 | */ | |
e43379f1 MM |
4454 | if (increment < -40) |
4455 | increment = -40; | |
1da177e4 LT |
4456 | if (increment > 40) |
4457 | increment = 40; | |
4458 | ||
2b8f836f | 4459 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4460 | if (nice < -20) |
4461 | nice = -20; | |
4462 | if (nice > 19) | |
4463 | nice = 19; | |
4464 | ||
e43379f1 MM |
4465 | if (increment < 0 && !can_nice(current, nice)) |
4466 | return -EPERM; | |
4467 | ||
1da177e4 LT |
4468 | retval = security_task_setnice(current, nice); |
4469 | if (retval) | |
4470 | return retval; | |
4471 | ||
4472 | set_user_nice(current, nice); | |
4473 | return 0; | |
4474 | } | |
4475 | ||
4476 | #endif | |
4477 | ||
4478 | /** | |
4479 | * task_prio - return the priority value of a given task. | |
4480 | * @p: the task in question. | |
4481 | * | |
4482 | * This is the priority value as seen by users in /proc. | |
4483 | * RT tasks are offset by -200. Normal tasks are centered | |
4484 | * around 0, value goes from -16 to +15. | |
4485 | */ | |
36c8b586 | 4486 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4487 | { |
4488 | return p->prio - MAX_RT_PRIO; | |
4489 | } | |
4490 | ||
4491 | /** | |
4492 | * task_nice - return the nice value of a given task. | |
4493 | * @p: the task in question. | |
4494 | */ | |
36c8b586 | 4495 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4496 | { |
4497 | return TASK_NICE(p); | |
4498 | } | |
150d8bed | 4499 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4500 | |
4501 | /** | |
4502 | * idle_cpu - is a given cpu idle currently? | |
4503 | * @cpu: the processor in question. | |
4504 | */ | |
4505 | int idle_cpu(int cpu) | |
4506 | { | |
4507 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4508 | } | |
4509 | ||
1da177e4 LT |
4510 | /** |
4511 | * idle_task - return the idle task for a given cpu. | |
4512 | * @cpu: the processor in question. | |
4513 | */ | |
36c8b586 | 4514 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4515 | { |
4516 | return cpu_rq(cpu)->idle; | |
4517 | } | |
4518 | ||
4519 | /** | |
4520 | * find_process_by_pid - find a process with a matching PID value. | |
4521 | * @pid: the pid in question. | |
4522 | */ | |
a9957449 | 4523 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4524 | { |
228ebcbe | 4525 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4526 | } |
4527 | ||
4528 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4529 | static void |
4530 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4531 | { |
dd41f596 | 4532 | BUG_ON(p->se.on_rq); |
48f24c4d | 4533 | |
1da177e4 LT |
4534 | p->policy = policy; |
4535 | p->rt_priority = prio; | |
b29739f9 IM |
4536 | p->normal_prio = normal_prio(p); |
4537 | /* we are holding p->pi_lock already */ | |
4538 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4539 | if (rt_prio(p->prio)) |
4540 | p->sched_class = &rt_sched_class; | |
4541 | else | |
4542 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4543 | set_load_weight(p); |
1da177e4 LT |
4544 | } |
4545 | ||
c69e8d9c DH |
4546 | /* |
4547 | * check the target process has a UID that matches the current process's | |
4548 | */ | |
4549 | static bool check_same_owner(struct task_struct *p) | |
4550 | { | |
4551 | const struct cred *cred = current_cred(), *pcred; | |
4552 | bool match; | |
4553 | ||
4554 | rcu_read_lock(); | |
4555 | pcred = __task_cred(p); | |
4556 | match = (cred->euid == pcred->euid || | |
4557 | cred->euid == pcred->uid); | |
4558 | rcu_read_unlock(); | |
4559 | return match; | |
4560 | } | |
4561 | ||
961ccddd | 4562 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4563 | const struct sched_param *param, bool user) |
1da177e4 | 4564 | { |
83b699ed | 4565 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4566 | unsigned long flags; |
83ab0aa0 | 4567 | const struct sched_class *prev_class; |
70b97a7f | 4568 | struct rq *rq; |
ca94c442 | 4569 | int reset_on_fork; |
1da177e4 | 4570 | |
66e5393a SR |
4571 | /* may grab non-irq protected spin_locks */ |
4572 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4573 | recheck: |
4574 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4575 | if (policy < 0) { |
4576 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4577 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4578 | } else { |
4579 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4580 | policy &= ~SCHED_RESET_ON_FORK; | |
4581 | ||
4582 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4583 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4584 | policy != SCHED_IDLE) | |
4585 | return -EINVAL; | |
4586 | } | |
4587 | ||
1da177e4 LT |
4588 | /* |
4589 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4590 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4591 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4592 | */ |
4593 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4594 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4595 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4596 | return -EINVAL; |
e05606d3 | 4597 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4598 | return -EINVAL; |
4599 | ||
37e4ab3f OC |
4600 | /* |
4601 | * Allow unprivileged RT tasks to decrease priority: | |
4602 | */ | |
961ccddd | 4603 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4604 | if (rt_policy(policy)) { |
a44702e8 ON |
4605 | unsigned long rlim_rtprio = |
4606 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4607 | |
4608 | /* can't set/change the rt policy */ | |
4609 | if (policy != p->policy && !rlim_rtprio) | |
4610 | return -EPERM; | |
4611 | ||
4612 | /* can't increase priority */ | |
4613 | if (param->sched_priority > p->rt_priority && | |
4614 | param->sched_priority > rlim_rtprio) | |
4615 | return -EPERM; | |
4616 | } | |
dd41f596 IM |
4617 | /* |
4618 | * Like positive nice levels, dont allow tasks to | |
4619 | * move out of SCHED_IDLE either: | |
4620 | */ | |
4621 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4622 | return -EPERM; | |
5fe1d75f | 4623 | |
37e4ab3f | 4624 | /* can't change other user's priorities */ |
c69e8d9c | 4625 | if (!check_same_owner(p)) |
37e4ab3f | 4626 | return -EPERM; |
ca94c442 LP |
4627 | |
4628 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4629 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4630 | return -EPERM; | |
37e4ab3f | 4631 | } |
1da177e4 | 4632 | |
725aad24 | 4633 | if (user) { |
b0ae1981 | 4634 | retval = security_task_setscheduler(p); |
725aad24 JF |
4635 | if (retval) |
4636 | return retval; | |
4637 | } | |
4638 | ||
b29739f9 IM |
4639 | /* |
4640 | * make sure no PI-waiters arrive (or leave) while we are | |
4641 | * changing the priority of the task: | |
4642 | */ | |
1d615482 | 4643 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4644 | /* |
4645 | * To be able to change p->policy safely, the apropriate | |
4646 | * runqueue lock must be held. | |
4647 | */ | |
b29739f9 | 4648 | rq = __task_rq_lock(p); |
dc61b1d6 | 4649 | |
34f971f6 PZ |
4650 | /* |
4651 | * Changing the policy of the stop threads its a very bad idea | |
4652 | */ | |
4653 | if (p == rq->stop) { | |
4654 | __task_rq_unlock(rq); | |
4655 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4656 | return -EINVAL; | |
4657 | } | |
4658 | ||
dc61b1d6 PZ |
4659 | #ifdef CONFIG_RT_GROUP_SCHED |
4660 | if (user) { | |
4661 | /* | |
4662 | * Do not allow realtime tasks into groups that have no runtime | |
4663 | * assigned. | |
4664 | */ | |
4665 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
4666 | task_group(p)->rt_bandwidth.rt_runtime == 0) { | |
4667 | __task_rq_unlock(rq); | |
4668 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4669 | return -EPERM; | |
4670 | } | |
4671 | } | |
4672 | #endif | |
4673 | ||
1da177e4 LT |
4674 | /* recheck policy now with rq lock held */ |
4675 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4676 | policy = oldpolicy = -1; | |
b29739f9 | 4677 | __task_rq_unlock(rq); |
1d615482 | 4678 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4679 | goto recheck; |
4680 | } | |
dd41f596 | 4681 | on_rq = p->se.on_rq; |
051a1d1a | 4682 | running = task_current(rq, p); |
0e1f3483 | 4683 | if (on_rq) |
2e1cb74a | 4684 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4685 | if (running) |
4686 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4687 | |
ca94c442 LP |
4688 | p->sched_reset_on_fork = reset_on_fork; |
4689 | ||
1da177e4 | 4690 | oldprio = p->prio; |
83ab0aa0 | 4691 | prev_class = p->sched_class; |
dd41f596 | 4692 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4693 | |
0e1f3483 HS |
4694 | if (running) |
4695 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4696 | if (on_rq) { |
4697 | activate_task(rq, p, 0); | |
cb469845 SR |
4698 | |
4699 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4700 | } |
b29739f9 | 4701 | __task_rq_unlock(rq); |
1d615482 | 4702 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4703 | |
95e02ca9 TG |
4704 | rt_mutex_adjust_pi(p); |
4705 | ||
1da177e4 LT |
4706 | return 0; |
4707 | } | |
961ccddd RR |
4708 | |
4709 | /** | |
4710 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4711 | * @p: the task in question. | |
4712 | * @policy: new policy. | |
4713 | * @param: structure containing the new RT priority. | |
4714 | * | |
4715 | * NOTE that the task may be already dead. | |
4716 | */ | |
4717 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4718 | const struct sched_param *param) |
961ccddd RR |
4719 | { |
4720 | return __sched_setscheduler(p, policy, param, true); | |
4721 | } | |
1da177e4 LT |
4722 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4723 | ||
961ccddd RR |
4724 | /** |
4725 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4726 | * @p: the task in question. | |
4727 | * @policy: new policy. | |
4728 | * @param: structure containing the new RT priority. | |
4729 | * | |
4730 | * Just like sched_setscheduler, only don't bother checking if the | |
4731 | * current context has permission. For example, this is needed in | |
4732 | * stop_machine(): we create temporary high priority worker threads, | |
4733 | * but our caller might not have that capability. | |
4734 | */ | |
4735 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4736 | const struct sched_param *param) |
961ccddd RR |
4737 | { |
4738 | return __sched_setscheduler(p, policy, param, false); | |
4739 | } | |
4740 | ||
95cdf3b7 IM |
4741 | static int |
4742 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4743 | { |
1da177e4 LT |
4744 | struct sched_param lparam; |
4745 | struct task_struct *p; | |
36c8b586 | 4746 | int retval; |
1da177e4 LT |
4747 | |
4748 | if (!param || pid < 0) | |
4749 | return -EINVAL; | |
4750 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4751 | return -EFAULT; | |
5fe1d75f ON |
4752 | |
4753 | rcu_read_lock(); | |
4754 | retval = -ESRCH; | |
1da177e4 | 4755 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4756 | if (p != NULL) |
4757 | retval = sched_setscheduler(p, policy, &lparam); | |
4758 | rcu_read_unlock(); | |
36c8b586 | 4759 | |
1da177e4 LT |
4760 | return retval; |
4761 | } | |
4762 | ||
4763 | /** | |
4764 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4765 | * @pid: the pid in question. | |
4766 | * @policy: new policy. | |
4767 | * @param: structure containing the new RT priority. | |
4768 | */ | |
5add95d4 HC |
4769 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4770 | struct sched_param __user *, param) | |
1da177e4 | 4771 | { |
c21761f1 JB |
4772 | /* negative values for policy are not valid */ |
4773 | if (policy < 0) | |
4774 | return -EINVAL; | |
4775 | ||
1da177e4 LT |
4776 | return do_sched_setscheduler(pid, policy, param); |
4777 | } | |
4778 | ||
4779 | /** | |
4780 | * sys_sched_setparam - set/change the RT priority of a thread | |
4781 | * @pid: the pid in question. | |
4782 | * @param: structure containing the new RT priority. | |
4783 | */ | |
5add95d4 | 4784 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4785 | { |
4786 | return do_sched_setscheduler(pid, -1, param); | |
4787 | } | |
4788 | ||
4789 | /** | |
4790 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4791 | * @pid: the pid in question. | |
4792 | */ | |
5add95d4 | 4793 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4794 | { |
36c8b586 | 4795 | struct task_struct *p; |
3a5c359a | 4796 | int retval; |
1da177e4 LT |
4797 | |
4798 | if (pid < 0) | |
3a5c359a | 4799 | return -EINVAL; |
1da177e4 LT |
4800 | |
4801 | retval = -ESRCH; | |
5fe85be0 | 4802 | rcu_read_lock(); |
1da177e4 LT |
4803 | p = find_process_by_pid(pid); |
4804 | if (p) { | |
4805 | retval = security_task_getscheduler(p); | |
4806 | if (!retval) | |
ca94c442 LP |
4807 | retval = p->policy |
4808 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4809 | } |
5fe85be0 | 4810 | rcu_read_unlock(); |
1da177e4 LT |
4811 | return retval; |
4812 | } | |
4813 | ||
4814 | /** | |
ca94c442 | 4815 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4816 | * @pid: the pid in question. |
4817 | * @param: structure containing the RT priority. | |
4818 | */ | |
5add95d4 | 4819 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4820 | { |
4821 | struct sched_param lp; | |
36c8b586 | 4822 | struct task_struct *p; |
3a5c359a | 4823 | int retval; |
1da177e4 LT |
4824 | |
4825 | if (!param || pid < 0) | |
3a5c359a | 4826 | return -EINVAL; |
1da177e4 | 4827 | |
5fe85be0 | 4828 | rcu_read_lock(); |
1da177e4 LT |
4829 | p = find_process_by_pid(pid); |
4830 | retval = -ESRCH; | |
4831 | if (!p) | |
4832 | goto out_unlock; | |
4833 | ||
4834 | retval = security_task_getscheduler(p); | |
4835 | if (retval) | |
4836 | goto out_unlock; | |
4837 | ||
4838 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4839 | rcu_read_unlock(); |
1da177e4 LT |
4840 | |
4841 | /* | |
4842 | * This one might sleep, we cannot do it with a spinlock held ... | |
4843 | */ | |
4844 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4845 | ||
1da177e4 LT |
4846 | return retval; |
4847 | ||
4848 | out_unlock: | |
5fe85be0 | 4849 | rcu_read_unlock(); |
1da177e4 LT |
4850 | return retval; |
4851 | } | |
4852 | ||
96f874e2 | 4853 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4854 | { |
5a16f3d3 | 4855 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4856 | struct task_struct *p; |
4857 | int retval; | |
1da177e4 | 4858 | |
95402b38 | 4859 | get_online_cpus(); |
23f5d142 | 4860 | rcu_read_lock(); |
1da177e4 LT |
4861 | |
4862 | p = find_process_by_pid(pid); | |
4863 | if (!p) { | |
23f5d142 | 4864 | rcu_read_unlock(); |
95402b38 | 4865 | put_online_cpus(); |
1da177e4 LT |
4866 | return -ESRCH; |
4867 | } | |
4868 | ||
23f5d142 | 4869 | /* Prevent p going away */ |
1da177e4 | 4870 | get_task_struct(p); |
23f5d142 | 4871 | rcu_read_unlock(); |
1da177e4 | 4872 | |
5a16f3d3 RR |
4873 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4874 | retval = -ENOMEM; | |
4875 | goto out_put_task; | |
4876 | } | |
4877 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4878 | retval = -ENOMEM; | |
4879 | goto out_free_cpus_allowed; | |
4880 | } | |
1da177e4 | 4881 | retval = -EPERM; |
c69e8d9c | 4882 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4883 | goto out_unlock; |
4884 | ||
b0ae1981 | 4885 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4886 | if (retval) |
4887 | goto out_unlock; | |
4888 | ||
5a16f3d3 RR |
4889 | cpuset_cpus_allowed(p, cpus_allowed); |
4890 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4891 | again: |
5a16f3d3 | 4892 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4893 | |
8707d8b8 | 4894 | if (!retval) { |
5a16f3d3 RR |
4895 | cpuset_cpus_allowed(p, cpus_allowed); |
4896 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4897 | /* |
4898 | * We must have raced with a concurrent cpuset | |
4899 | * update. Just reset the cpus_allowed to the | |
4900 | * cpuset's cpus_allowed | |
4901 | */ | |
5a16f3d3 | 4902 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4903 | goto again; |
4904 | } | |
4905 | } | |
1da177e4 | 4906 | out_unlock: |
5a16f3d3 RR |
4907 | free_cpumask_var(new_mask); |
4908 | out_free_cpus_allowed: | |
4909 | free_cpumask_var(cpus_allowed); | |
4910 | out_put_task: | |
1da177e4 | 4911 | put_task_struct(p); |
95402b38 | 4912 | put_online_cpus(); |
1da177e4 LT |
4913 | return retval; |
4914 | } | |
4915 | ||
4916 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4917 | struct cpumask *new_mask) |
1da177e4 | 4918 | { |
96f874e2 RR |
4919 | if (len < cpumask_size()) |
4920 | cpumask_clear(new_mask); | |
4921 | else if (len > cpumask_size()) | |
4922 | len = cpumask_size(); | |
4923 | ||
1da177e4 LT |
4924 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4925 | } | |
4926 | ||
4927 | /** | |
4928 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4929 | * @pid: pid of the process | |
4930 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4931 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4932 | */ | |
5add95d4 HC |
4933 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4934 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4935 | { |
5a16f3d3 | 4936 | cpumask_var_t new_mask; |
1da177e4 LT |
4937 | int retval; |
4938 | ||
5a16f3d3 RR |
4939 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4940 | return -ENOMEM; | |
1da177e4 | 4941 | |
5a16f3d3 RR |
4942 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4943 | if (retval == 0) | |
4944 | retval = sched_setaffinity(pid, new_mask); | |
4945 | free_cpumask_var(new_mask); | |
4946 | return retval; | |
1da177e4 LT |
4947 | } |
4948 | ||
96f874e2 | 4949 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4950 | { |
36c8b586 | 4951 | struct task_struct *p; |
31605683 TG |
4952 | unsigned long flags; |
4953 | struct rq *rq; | |
1da177e4 | 4954 | int retval; |
1da177e4 | 4955 | |
95402b38 | 4956 | get_online_cpus(); |
23f5d142 | 4957 | rcu_read_lock(); |
1da177e4 LT |
4958 | |
4959 | retval = -ESRCH; | |
4960 | p = find_process_by_pid(pid); | |
4961 | if (!p) | |
4962 | goto out_unlock; | |
4963 | ||
e7834f8f DQ |
4964 | retval = security_task_getscheduler(p); |
4965 | if (retval) | |
4966 | goto out_unlock; | |
4967 | ||
31605683 | 4968 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4969 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4970 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4971 | |
4972 | out_unlock: | |
23f5d142 | 4973 | rcu_read_unlock(); |
95402b38 | 4974 | put_online_cpus(); |
1da177e4 | 4975 | |
9531b62f | 4976 | return retval; |
1da177e4 LT |
4977 | } |
4978 | ||
4979 | /** | |
4980 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4981 | * @pid: pid of the process | |
4982 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4983 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4984 | */ | |
5add95d4 HC |
4985 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4986 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4987 | { |
4988 | int ret; | |
f17c8607 | 4989 | cpumask_var_t mask; |
1da177e4 | 4990 | |
84fba5ec | 4991 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4992 | return -EINVAL; |
4993 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4994 | return -EINVAL; |
4995 | ||
f17c8607 RR |
4996 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4997 | return -ENOMEM; | |
1da177e4 | 4998 | |
f17c8607 RR |
4999 | ret = sched_getaffinity(pid, mask); |
5000 | if (ret == 0) { | |
8bc037fb | 5001 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5002 | |
5003 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5004 | ret = -EFAULT; |
5005 | else | |
cd3d8031 | 5006 | ret = retlen; |
f17c8607 RR |
5007 | } |
5008 | free_cpumask_var(mask); | |
1da177e4 | 5009 | |
f17c8607 | 5010 | return ret; |
1da177e4 LT |
5011 | } |
5012 | ||
5013 | /** | |
5014 | * sys_sched_yield - yield the current processor to other threads. | |
5015 | * | |
dd41f596 IM |
5016 | * This function yields the current CPU to other tasks. If there are no |
5017 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5018 | */ |
5add95d4 | 5019 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5020 | { |
70b97a7f | 5021 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5022 | |
2d72376b | 5023 | schedstat_inc(rq, yld_count); |
4530d7ab | 5024 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5025 | |
5026 | /* | |
5027 | * Since we are going to call schedule() anyway, there's | |
5028 | * no need to preempt or enable interrupts: | |
5029 | */ | |
5030 | __release(rq->lock); | |
8a25d5de | 5031 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5032 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5033 | preempt_enable_no_resched(); |
5034 | ||
5035 | schedule(); | |
5036 | ||
5037 | return 0; | |
5038 | } | |
5039 | ||
d86ee480 PZ |
5040 | static inline int should_resched(void) |
5041 | { | |
5042 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5043 | } | |
5044 | ||
e7b38404 | 5045 | static void __cond_resched(void) |
1da177e4 | 5046 | { |
e7aaaa69 FW |
5047 | add_preempt_count(PREEMPT_ACTIVE); |
5048 | schedule(); | |
5049 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5050 | } |
5051 | ||
02b67cc3 | 5052 | int __sched _cond_resched(void) |
1da177e4 | 5053 | { |
d86ee480 | 5054 | if (should_resched()) { |
1da177e4 LT |
5055 | __cond_resched(); |
5056 | return 1; | |
5057 | } | |
5058 | return 0; | |
5059 | } | |
02b67cc3 | 5060 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5061 | |
5062 | /* | |
613afbf8 | 5063 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5064 | * call schedule, and on return reacquire the lock. |
5065 | * | |
41a2d6cf | 5066 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5067 | * operations here to prevent schedule() from being called twice (once via |
5068 | * spin_unlock(), once by hand). | |
5069 | */ | |
613afbf8 | 5070 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5071 | { |
d86ee480 | 5072 | int resched = should_resched(); |
6df3cecb JK |
5073 | int ret = 0; |
5074 | ||
f607c668 PZ |
5075 | lockdep_assert_held(lock); |
5076 | ||
95c354fe | 5077 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5078 | spin_unlock(lock); |
d86ee480 | 5079 | if (resched) |
95c354fe NP |
5080 | __cond_resched(); |
5081 | else | |
5082 | cpu_relax(); | |
6df3cecb | 5083 | ret = 1; |
1da177e4 | 5084 | spin_lock(lock); |
1da177e4 | 5085 | } |
6df3cecb | 5086 | return ret; |
1da177e4 | 5087 | } |
613afbf8 | 5088 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5089 | |
613afbf8 | 5090 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5091 | { |
5092 | BUG_ON(!in_softirq()); | |
5093 | ||
d86ee480 | 5094 | if (should_resched()) { |
98d82567 | 5095 | local_bh_enable(); |
1da177e4 LT |
5096 | __cond_resched(); |
5097 | local_bh_disable(); | |
5098 | return 1; | |
5099 | } | |
5100 | return 0; | |
5101 | } | |
613afbf8 | 5102 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5103 | |
1da177e4 LT |
5104 | /** |
5105 | * yield - yield the current processor to other threads. | |
5106 | * | |
72fd4a35 | 5107 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5108 | * thread runnable and calls sys_sched_yield(). |
5109 | */ | |
5110 | void __sched yield(void) | |
5111 | { | |
5112 | set_current_state(TASK_RUNNING); | |
5113 | sys_sched_yield(); | |
5114 | } | |
1da177e4 LT |
5115 | EXPORT_SYMBOL(yield); |
5116 | ||
5117 | /* | |
41a2d6cf | 5118 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5119 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5120 | */ |
5121 | void __sched io_schedule(void) | |
5122 | { | |
54d35f29 | 5123 | struct rq *rq = raw_rq(); |
1da177e4 | 5124 | |
0ff92245 | 5125 | delayacct_blkio_start(); |
1da177e4 | 5126 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5127 | current->in_iowait = 1; |
1da177e4 | 5128 | schedule(); |
8f0dfc34 | 5129 | current->in_iowait = 0; |
1da177e4 | 5130 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5131 | delayacct_blkio_end(); |
1da177e4 | 5132 | } |
1da177e4 LT |
5133 | EXPORT_SYMBOL(io_schedule); |
5134 | ||
5135 | long __sched io_schedule_timeout(long timeout) | |
5136 | { | |
54d35f29 | 5137 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5138 | long ret; |
5139 | ||
0ff92245 | 5140 | delayacct_blkio_start(); |
1da177e4 | 5141 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5142 | current->in_iowait = 1; |
1da177e4 | 5143 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5144 | current->in_iowait = 0; |
1da177e4 | 5145 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5146 | delayacct_blkio_end(); |
1da177e4 LT |
5147 | return ret; |
5148 | } | |
5149 | ||
5150 | /** | |
5151 | * sys_sched_get_priority_max - return maximum RT priority. | |
5152 | * @policy: scheduling class. | |
5153 | * | |
5154 | * this syscall returns the maximum rt_priority that can be used | |
5155 | * by a given scheduling class. | |
5156 | */ | |
5add95d4 | 5157 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5158 | { |
5159 | int ret = -EINVAL; | |
5160 | ||
5161 | switch (policy) { | |
5162 | case SCHED_FIFO: | |
5163 | case SCHED_RR: | |
5164 | ret = MAX_USER_RT_PRIO-1; | |
5165 | break; | |
5166 | case SCHED_NORMAL: | |
b0a9499c | 5167 | case SCHED_BATCH: |
dd41f596 | 5168 | case SCHED_IDLE: |
1da177e4 LT |
5169 | ret = 0; |
5170 | break; | |
5171 | } | |
5172 | return ret; | |
5173 | } | |
5174 | ||
5175 | /** | |
5176 | * sys_sched_get_priority_min - return minimum RT priority. | |
5177 | * @policy: scheduling class. | |
5178 | * | |
5179 | * this syscall returns the minimum rt_priority that can be used | |
5180 | * by a given scheduling class. | |
5181 | */ | |
5add95d4 | 5182 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5183 | { |
5184 | int ret = -EINVAL; | |
5185 | ||
5186 | switch (policy) { | |
5187 | case SCHED_FIFO: | |
5188 | case SCHED_RR: | |
5189 | ret = 1; | |
5190 | break; | |
5191 | case SCHED_NORMAL: | |
b0a9499c | 5192 | case SCHED_BATCH: |
dd41f596 | 5193 | case SCHED_IDLE: |
1da177e4 LT |
5194 | ret = 0; |
5195 | } | |
5196 | return ret; | |
5197 | } | |
5198 | ||
5199 | /** | |
5200 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5201 | * @pid: pid of the process. | |
5202 | * @interval: userspace pointer to the timeslice value. | |
5203 | * | |
5204 | * this syscall writes the default timeslice value of a given process | |
5205 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5206 | */ | |
17da2bd9 | 5207 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5208 | struct timespec __user *, interval) |
1da177e4 | 5209 | { |
36c8b586 | 5210 | struct task_struct *p; |
a4ec24b4 | 5211 | unsigned int time_slice; |
dba091b9 TG |
5212 | unsigned long flags; |
5213 | struct rq *rq; | |
3a5c359a | 5214 | int retval; |
1da177e4 | 5215 | struct timespec t; |
1da177e4 LT |
5216 | |
5217 | if (pid < 0) | |
3a5c359a | 5218 | return -EINVAL; |
1da177e4 LT |
5219 | |
5220 | retval = -ESRCH; | |
1a551ae7 | 5221 | rcu_read_lock(); |
1da177e4 LT |
5222 | p = find_process_by_pid(pid); |
5223 | if (!p) | |
5224 | goto out_unlock; | |
5225 | ||
5226 | retval = security_task_getscheduler(p); | |
5227 | if (retval) | |
5228 | goto out_unlock; | |
5229 | ||
dba091b9 TG |
5230 | rq = task_rq_lock(p, &flags); |
5231 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5232 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5233 | |
1a551ae7 | 5234 | rcu_read_unlock(); |
a4ec24b4 | 5235 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5236 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5237 | return retval; |
3a5c359a | 5238 | |
1da177e4 | 5239 | out_unlock: |
1a551ae7 | 5240 | rcu_read_unlock(); |
1da177e4 LT |
5241 | return retval; |
5242 | } | |
5243 | ||
7c731e0a | 5244 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5245 | |
82a1fcb9 | 5246 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5247 | { |
1da177e4 | 5248 | unsigned long free = 0; |
36c8b586 | 5249 | unsigned state; |
1da177e4 | 5250 | |
1da177e4 | 5251 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5252 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5253 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5254 | #if BITS_PER_LONG == 32 |
1da177e4 | 5255 | if (state == TASK_RUNNING) |
3df0fc5b | 5256 | printk(KERN_CONT " running "); |
1da177e4 | 5257 | else |
3df0fc5b | 5258 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5259 | #else |
5260 | if (state == TASK_RUNNING) | |
3df0fc5b | 5261 | printk(KERN_CONT " running task "); |
1da177e4 | 5262 | else |
3df0fc5b | 5263 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5264 | #endif |
5265 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5266 | free = stack_not_used(p); |
1da177e4 | 5267 | #endif |
3df0fc5b | 5268 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5269 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5270 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5271 | |
5fb5e6de | 5272 | show_stack(p, NULL); |
1da177e4 LT |
5273 | } |
5274 | ||
e59e2ae2 | 5275 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5276 | { |
36c8b586 | 5277 | struct task_struct *g, *p; |
1da177e4 | 5278 | |
4bd77321 | 5279 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5280 | printk(KERN_INFO |
5281 | " task PC stack pid father\n"); | |
1da177e4 | 5282 | #else |
3df0fc5b PZ |
5283 | printk(KERN_INFO |
5284 | " task PC stack pid father\n"); | |
1da177e4 LT |
5285 | #endif |
5286 | read_lock(&tasklist_lock); | |
5287 | do_each_thread(g, p) { | |
5288 | /* | |
5289 | * reset the NMI-timeout, listing all files on a slow | |
5290 | * console might take alot of time: | |
5291 | */ | |
5292 | touch_nmi_watchdog(); | |
39bc89fd | 5293 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5294 | sched_show_task(p); |
1da177e4 LT |
5295 | } while_each_thread(g, p); |
5296 | ||
04c9167f JF |
5297 | touch_all_softlockup_watchdogs(); |
5298 | ||
dd41f596 IM |
5299 | #ifdef CONFIG_SCHED_DEBUG |
5300 | sysrq_sched_debug_show(); | |
5301 | #endif | |
1da177e4 | 5302 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5303 | /* |
5304 | * Only show locks if all tasks are dumped: | |
5305 | */ | |
93335a21 | 5306 | if (!state_filter) |
e59e2ae2 | 5307 | debug_show_all_locks(); |
1da177e4 LT |
5308 | } |
5309 | ||
1df21055 IM |
5310 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5311 | { | |
dd41f596 | 5312 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5313 | } |
5314 | ||
f340c0d1 IM |
5315 | /** |
5316 | * init_idle - set up an idle thread for a given CPU | |
5317 | * @idle: task in question | |
5318 | * @cpu: cpu the idle task belongs to | |
5319 | * | |
5320 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5321 | * flag, to make booting more robust. | |
5322 | */ | |
5c1e1767 | 5323 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5324 | { |
70b97a7f | 5325 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5326 | unsigned long flags; |
5327 | ||
05fa785c | 5328 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5329 | |
dd41f596 | 5330 | __sched_fork(idle); |
06b83b5f | 5331 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5332 | idle->se.exec_start = sched_clock(); |
5333 | ||
96f874e2 | 5334 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5335 | /* |
5336 | * We're having a chicken and egg problem, even though we are | |
5337 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5338 | * lockdep check in task_group() will fail. | |
5339 | * | |
5340 | * Similar case to sched_fork(). / Alternatively we could | |
5341 | * use task_rq_lock() here and obtain the other rq->lock. | |
5342 | * | |
5343 | * Silence PROVE_RCU | |
5344 | */ | |
5345 | rcu_read_lock(); | |
dd41f596 | 5346 | __set_task_cpu(idle, cpu); |
6506cf6c | 5347 | rcu_read_unlock(); |
1da177e4 | 5348 | |
1da177e4 | 5349 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5350 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5351 | idle->oncpu = 1; | |
5352 | #endif | |
05fa785c | 5353 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5354 | |
5355 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5356 | #if defined(CONFIG_PREEMPT) |
5357 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5358 | #else | |
a1261f54 | 5359 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5360 | #endif |
dd41f596 IM |
5361 | /* |
5362 | * The idle tasks have their own, simple scheduling class: | |
5363 | */ | |
5364 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5365 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5366 | } |
5367 | ||
5368 | /* | |
5369 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5370 | * indicates which cpus entered this state. This is used | |
5371 | * in the rcu update to wait only for active cpus. For system | |
5372 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5373 | * always be CPU_BITS_NONE. |
1da177e4 | 5374 | */ |
6a7b3dc3 | 5375 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5376 | |
19978ca6 IM |
5377 | /* |
5378 | * Increase the granularity value when there are more CPUs, | |
5379 | * because with more CPUs the 'effective latency' as visible | |
5380 | * to users decreases. But the relationship is not linear, | |
5381 | * so pick a second-best guess by going with the log2 of the | |
5382 | * number of CPUs. | |
5383 | * | |
5384 | * This idea comes from the SD scheduler of Con Kolivas: | |
5385 | */ | |
acb4a848 | 5386 | static int get_update_sysctl_factor(void) |
19978ca6 | 5387 | { |
4ca3ef71 | 5388 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5389 | unsigned int factor; |
5390 | ||
5391 | switch (sysctl_sched_tunable_scaling) { | |
5392 | case SCHED_TUNABLESCALING_NONE: | |
5393 | factor = 1; | |
5394 | break; | |
5395 | case SCHED_TUNABLESCALING_LINEAR: | |
5396 | factor = cpus; | |
5397 | break; | |
5398 | case SCHED_TUNABLESCALING_LOG: | |
5399 | default: | |
5400 | factor = 1 + ilog2(cpus); | |
5401 | break; | |
5402 | } | |
19978ca6 | 5403 | |
acb4a848 CE |
5404 | return factor; |
5405 | } | |
19978ca6 | 5406 | |
acb4a848 CE |
5407 | static void update_sysctl(void) |
5408 | { | |
5409 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5410 | |
0bcdcf28 CE |
5411 | #define SET_SYSCTL(name) \ |
5412 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5413 | SET_SYSCTL(sched_min_granularity); | |
5414 | SET_SYSCTL(sched_latency); | |
5415 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5416 | #undef SET_SYSCTL |
5417 | } | |
55cd5340 | 5418 | |
0bcdcf28 CE |
5419 | static inline void sched_init_granularity(void) |
5420 | { | |
5421 | update_sysctl(); | |
19978ca6 IM |
5422 | } |
5423 | ||
1da177e4 LT |
5424 | #ifdef CONFIG_SMP |
5425 | /* | |
5426 | * This is how migration works: | |
5427 | * | |
969c7921 TH |
5428 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5429 | * stop_one_cpu(). | |
5430 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5431 | * off the CPU) | |
5432 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5433 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5434 | * it and puts it into the right queue. |
969c7921 TH |
5435 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5436 | * is done. | |
1da177e4 LT |
5437 | */ |
5438 | ||
5439 | /* | |
5440 | * Change a given task's CPU affinity. Migrate the thread to a | |
5441 | * proper CPU and schedule it away if the CPU it's executing on | |
5442 | * is removed from the allowed bitmask. | |
5443 | * | |
5444 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5445 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5446 | * call is not atomic; no spinlocks may be held. |
5447 | */ | |
96f874e2 | 5448 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5449 | { |
5450 | unsigned long flags; | |
70b97a7f | 5451 | struct rq *rq; |
969c7921 | 5452 | unsigned int dest_cpu; |
48f24c4d | 5453 | int ret = 0; |
1da177e4 | 5454 | |
65cc8e48 PZ |
5455 | /* |
5456 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5457 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5458 | */ | |
5459 | again: | |
5460 | while (task_is_waking(p)) | |
5461 | cpu_relax(); | |
1da177e4 | 5462 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5463 | if (task_is_waking(p)) { |
5464 | task_rq_unlock(rq, &flags); | |
5465 | goto again; | |
5466 | } | |
e2912009 | 5467 | |
6ad4c188 | 5468 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5469 | ret = -EINVAL; |
5470 | goto out; | |
5471 | } | |
5472 | ||
9985b0ba | 5473 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5474 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5475 | ret = -EINVAL; |
5476 | goto out; | |
5477 | } | |
5478 | ||
73fe6aae | 5479 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5480 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5481 | else { |
96f874e2 RR |
5482 | cpumask_copy(&p->cpus_allowed, new_mask); |
5483 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5484 | } |
5485 | ||
1da177e4 | 5486 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5487 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5488 | goto out; |
5489 | ||
969c7921 TH |
5490 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
5491 | if (migrate_task(p, dest_cpu)) { | |
5492 | struct migration_arg arg = { p, dest_cpu }; | |
1da177e4 LT |
5493 | /* Need help from migration thread: drop lock and wait. */ |
5494 | task_rq_unlock(rq, &flags); | |
969c7921 | 5495 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5496 | tlb_migrate_finish(p->mm); |
5497 | return 0; | |
5498 | } | |
5499 | out: | |
5500 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5501 | |
1da177e4 LT |
5502 | return ret; |
5503 | } | |
cd8ba7cd | 5504 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5505 | |
5506 | /* | |
41a2d6cf | 5507 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5508 | * this because either it can't run here any more (set_cpus_allowed() |
5509 | * away from this CPU, or CPU going down), or because we're | |
5510 | * attempting to rebalance this task on exec (sched_exec). | |
5511 | * | |
5512 | * So we race with normal scheduler movements, but that's OK, as long | |
5513 | * as the task is no longer on this CPU. | |
efc30814 KK |
5514 | * |
5515 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5516 | */ |
efc30814 | 5517 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5518 | { |
70b97a7f | 5519 | struct rq *rq_dest, *rq_src; |
e2912009 | 5520 | int ret = 0; |
1da177e4 | 5521 | |
e761b772 | 5522 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5523 | return ret; |
1da177e4 LT |
5524 | |
5525 | rq_src = cpu_rq(src_cpu); | |
5526 | rq_dest = cpu_rq(dest_cpu); | |
5527 | ||
5528 | double_rq_lock(rq_src, rq_dest); | |
5529 | /* Already moved. */ | |
5530 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5531 | goto done; |
1da177e4 | 5532 | /* Affinity changed (again). */ |
96f874e2 | 5533 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5534 | goto fail; |
1da177e4 | 5535 | |
e2912009 PZ |
5536 | /* |
5537 | * If we're not on a rq, the next wake-up will ensure we're | |
5538 | * placed properly. | |
5539 | */ | |
5540 | if (p->se.on_rq) { | |
2e1cb74a | 5541 | deactivate_task(rq_src, p, 0); |
e2912009 | 5542 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5543 | activate_task(rq_dest, p, 0); |
15afe09b | 5544 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5545 | } |
b1e38734 | 5546 | done: |
efc30814 | 5547 | ret = 1; |
b1e38734 | 5548 | fail: |
1da177e4 | 5549 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5550 | return ret; |
1da177e4 LT |
5551 | } |
5552 | ||
5553 | /* | |
969c7921 TH |
5554 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5555 | * and performs thread migration by bumping thread off CPU then | |
5556 | * 'pushing' onto another runqueue. | |
1da177e4 | 5557 | */ |
969c7921 | 5558 | static int migration_cpu_stop(void *data) |
1da177e4 | 5559 | { |
969c7921 | 5560 | struct migration_arg *arg = data; |
f7b4cddc | 5561 | |
969c7921 TH |
5562 | /* |
5563 | * The original target cpu might have gone down and we might | |
5564 | * be on another cpu but it doesn't matter. | |
5565 | */ | |
f7b4cddc | 5566 | local_irq_disable(); |
969c7921 | 5567 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5568 | local_irq_enable(); |
1da177e4 | 5569 | return 0; |
f7b4cddc ON |
5570 | } |
5571 | ||
1da177e4 | 5572 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5573 | |
054b9108 | 5574 | /* |
48c5ccae PZ |
5575 | * Ensures that the idle task is using init_mm right before its cpu goes |
5576 | * offline. | |
054b9108 | 5577 | */ |
48c5ccae | 5578 | void idle_task_exit(void) |
1da177e4 | 5579 | { |
48c5ccae | 5580 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5581 | |
48c5ccae | 5582 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5583 | |
48c5ccae PZ |
5584 | if (mm != &init_mm) |
5585 | switch_mm(mm, &init_mm, current); | |
5586 | mmdrop(mm); | |
1da177e4 LT |
5587 | } |
5588 | ||
5589 | /* | |
5590 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5591 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5592 | * for performance reasons the counter is not stricly tracking tasks to | |
5593 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5594 | * to keep the global sum constant after CPU-down: | |
5595 | */ | |
70b97a7f | 5596 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5597 | { |
6ad4c188 | 5598 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 5599 | |
1da177e4 LT |
5600 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
5601 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
5602 | } |
5603 | ||
dd41f596 | 5604 | /* |
48c5ccae | 5605 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 5606 | */ |
48c5ccae | 5607 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 5608 | { |
48c5ccae PZ |
5609 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
5610 | rq->calc_load_active = 0; | |
1da177e4 LT |
5611 | } |
5612 | ||
48f24c4d | 5613 | /* |
48c5ccae PZ |
5614 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5615 | * try_to_wake_up()->select_task_rq(). | |
5616 | * | |
5617 | * Called with rq->lock held even though we'er in stop_machine() and | |
5618 | * there's no concurrency possible, we hold the required locks anyway | |
5619 | * because of lock validation efforts. | |
1da177e4 | 5620 | */ |
48c5ccae | 5621 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5622 | { |
70b97a7f | 5623 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5624 | struct task_struct *next, *stop = rq->stop; |
5625 | int dest_cpu; | |
1da177e4 LT |
5626 | |
5627 | /* | |
48c5ccae PZ |
5628 | * Fudge the rq selection such that the below task selection loop |
5629 | * doesn't get stuck on the currently eligible stop task. | |
5630 | * | |
5631 | * We're currently inside stop_machine() and the rq is either stuck | |
5632 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5633 | * either way we should never end up calling schedule() until we're | |
5634 | * done here. | |
1da177e4 | 5635 | */ |
48c5ccae | 5636 | rq->stop = NULL; |
48f24c4d | 5637 | |
dd41f596 | 5638 | for ( ; ; ) { |
48c5ccae PZ |
5639 | /* |
5640 | * There's this thread running, bail when that's the only | |
5641 | * remaining thread. | |
5642 | */ | |
5643 | if (rq->nr_running == 1) | |
dd41f596 | 5644 | break; |
48c5ccae | 5645 | |
b67802ea | 5646 | next = pick_next_task(rq); |
48c5ccae | 5647 | BUG_ON(!next); |
79c53799 | 5648 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5649 | |
48c5ccae PZ |
5650 | /* Find suitable destination for @next, with force if needed. */ |
5651 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5652 | raw_spin_unlock(&rq->lock); | |
5653 | ||
5654 | __migrate_task(next, dead_cpu, dest_cpu); | |
5655 | ||
5656 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5657 | } |
dce48a84 | 5658 | |
48c5ccae | 5659 | rq->stop = stop; |
dce48a84 | 5660 | } |
48c5ccae | 5661 | |
1da177e4 LT |
5662 | #endif /* CONFIG_HOTPLUG_CPU */ |
5663 | ||
e692ab53 NP |
5664 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5665 | ||
5666 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5667 | { |
5668 | .procname = "sched_domain", | |
c57baf1e | 5669 | .mode = 0555, |
e0361851 | 5670 | }, |
56992309 | 5671 | {} |
e692ab53 NP |
5672 | }; |
5673 | ||
5674 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5675 | { |
5676 | .procname = "kernel", | |
c57baf1e | 5677 | .mode = 0555, |
e0361851 AD |
5678 | .child = sd_ctl_dir, |
5679 | }, | |
56992309 | 5680 | {} |
e692ab53 NP |
5681 | }; |
5682 | ||
5683 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5684 | { | |
5685 | struct ctl_table *entry = | |
5cf9f062 | 5686 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5687 | |
e692ab53 NP |
5688 | return entry; |
5689 | } | |
5690 | ||
6382bc90 MM |
5691 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5692 | { | |
cd790076 | 5693 | struct ctl_table *entry; |
6382bc90 | 5694 | |
cd790076 MM |
5695 | /* |
5696 | * In the intermediate directories, both the child directory and | |
5697 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5698 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5699 | * static strings and all have proc handlers. |
5700 | */ | |
5701 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5702 | if (entry->child) |
5703 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5704 | if (entry->proc_handler == NULL) |
5705 | kfree(entry->procname); | |
5706 | } | |
6382bc90 MM |
5707 | |
5708 | kfree(*tablep); | |
5709 | *tablep = NULL; | |
5710 | } | |
5711 | ||
e692ab53 | 5712 | static void |
e0361851 | 5713 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5714 | const char *procname, void *data, int maxlen, |
5715 | mode_t mode, proc_handler *proc_handler) | |
5716 | { | |
e692ab53 NP |
5717 | entry->procname = procname; |
5718 | entry->data = data; | |
5719 | entry->maxlen = maxlen; | |
5720 | entry->mode = mode; | |
5721 | entry->proc_handler = proc_handler; | |
5722 | } | |
5723 | ||
5724 | static struct ctl_table * | |
5725 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5726 | { | |
a5d8c348 | 5727 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5728 | |
ad1cdc1d MM |
5729 | if (table == NULL) |
5730 | return NULL; | |
5731 | ||
e0361851 | 5732 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5733 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5734 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5735 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5736 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5737 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5738 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5739 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5740 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5741 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5742 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5743 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5744 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5745 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5746 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5747 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5748 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5749 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5750 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5751 | &sd->cache_nice_tries, |
5752 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5753 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5754 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5755 | set_table_entry(&table[11], "name", sd->name, |
5756 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5757 | /* &table[12] is terminator */ | |
e692ab53 NP |
5758 | |
5759 | return table; | |
5760 | } | |
5761 | ||
9a4e7159 | 5762 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5763 | { |
5764 | struct ctl_table *entry, *table; | |
5765 | struct sched_domain *sd; | |
5766 | int domain_num = 0, i; | |
5767 | char buf[32]; | |
5768 | ||
5769 | for_each_domain(cpu, sd) | |
5770 | domain_num++; | |
5771 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5772 | if (table == NULL) |
5773 | return NULL; | |
e692ab53 NP |
5774 | |
5775 | i = 0; | |
5776 | for_each_domain(cpu, sd) { | |
5777 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5778 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5779 | entry->mode = 0555; |
e692ab53 NP |
5780 | entry->child = sd_alloc_ctl_domain_table(sd); |
5781 | entry++; | |
5782 | i++; | |
5783 | } | |
5784 | return table; | |
5785 | } | |
5786 | ||
5787 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5788 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5789 | { |
6ad4c188 | 5790 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5791 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5792 | char buf[32]; | |
5793 | ||
7378547f MM |
5794 | WARN_ON(sd_ctl_dir[0].child); |
5795 | sd_ctl_dir[0].child = entry; | |
5796 | ||
ad1cdc1d MM |
5797 | if (entry == NULL) |
5798 | return; | |
5799 | ||
6ad4c188 | 5800 | for_each_possible_cpu(i) { |
e692ab53 | 5801 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5802 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5803 | entry->mode = 0555; |
e692ab53 | 5804 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5805 | entry++; |
e692ab53 | 5806 | } |
7378547f MM |
5807 | |
5808 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5809 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5810 | } | |
6382bc90 | 5811 | |
7378547f | 5812 | /* may be called multiple times per register */ |
6382bc90 MM |
5813 | static void unregister_sched_domain_sysctl(void) |
5814 | { | |
7378547f MM |
5815 | if (sd_sysctl_header) |
5816 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5817 | sd_sysctl_header = NULL; |
7378547f MM |
5818 | if (sd_ctl_dir[0].child) |
5819 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5820 | } |
e692ab53 | 5821 | #else |
6382bc90 MM |
5822 | static void register_sched_domain_sysctl(void) |
5823 | { | |
5824 | } | |
5825 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5826 | { |
5827 | } | |
5828 | #endif | |
5829 | ||
1f11eb6a GH |
5830 | static void set_rq_online(struct rq *rq) |
5831 | { | |
5832 | if (!rq->online) { | |
5833 | const struct sched_class *class; | |
5834 | ||
c6c4927b | 5835 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5836 | rq->online = 1; |
5837 | ||
5838 | for_each_class(class) { | |
5839 | if (class->rq_online) | |
5840 | class->rq_online(rq); | |
5841 | } | |
5842 | } | |
5843 | } | |
5844 | ||
5845 | static void set_rq_offline(struct rq *rq) | |
5846 | { | |
5847 | if (rq->online) { | |
5848 | const struct sched_class *class; | |
5849 | ||
5850 | for_each_class(class) { | |
5851 | if (class->rq_offline) | |
5852 | class->rq_offline(rq); | |
5853 | } | |
5854 | ||
c6c4927b | 5855 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5856 | rq->online = 0; |
5857 | } | |
5858 | } | |
5859 | ||
1da177e4 LT |
5860 | /* |
5861 | * migration_call - callback that gets triggered when a CPU is added. | |
5862 | * Here we can start up the necessary migration thread for the new CPU. | |
5863 | */ | |
48f24c4d IM |
5864 | static int __cpuinit |
5865 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5866 | { |
48f24c4d | 5867 | int cpu = (long)hcpu; |
1da177e4 | 5868 | unsigned long flags; |
969c7921 | 5869 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5870 | |
48c5ccae | 5871 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5872 | |
1da177e4 | 5873 | case CPU_UP_PREPARE: |
a468d389 | 5874 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5875 | break; |
48f24c4d | 5876 | |
1da177e4 | 5877 | case CPU_ONLINE: |
1f94ef59 | 5878 | /* Update our root-domain */ |
05fa785c | 5879 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5880 | if (rq->rd) { |
c6c4927b | 5881 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5882 | |
5883 | set_rq_online(rq); | |
1f94ef59 | 5884 | } |
05fa785c | 5885 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5886 | break; |
48f24c4d | 5887 | |
1da177e4 | 5888 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5889 | case CPU_DYING: |
57d885fe | 5890 | /* Update our root-domain */ |
05fa785c | 5891 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5892 | if (rq->rd) { |
c6c4927b | 5893 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5894 | set_rq_offline(rq); |
57d885fe | 5895 | } |
48c5ccae PZ |
5896 | migrate_tasks(cpu); |
5897 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5898 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
5899 | |
5900 | migrate_nr_uninterruptible(rq); | |
5901 | calc_global_load_remove(rq); | |
57d885fe | 5902 | break; |
1da177e4 LT |
5903 | #endif |
5904 | } | |
5905 | return NOTIFY_OK; | |
5906 | } | |
5907 | ||
f38b0820 PM |
5908 | /* |
5909 | * Register at high priority so that task migration (migrate_all_tasks) | |
5910 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5911 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5912 | */ |
26c2143b | 5913 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5914 | .notifier_call = migration_call, |
50a323b7 | 5915 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5916 | }; |
5917 | ||
3a101d05 TH |
5918 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5919 | unsigned long action, void *hcpu) | |
5920 | { | |
5921 | switch (action & ~CPU_TASKS_FROZEN) { | |
5922 | case CPU_ONLINE: | |
5923 | case CPU_DOWN_FAILED: | |
5924 | set_cpu_active((long)hcpu, true); | |
5925 | return NOTIFY_OK; | |
5926 | default: | |
5927 | return NOTIFY_DONE; | |
5928 | } | |
5929 | } | |
5930 | ||
5931 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5932 | unsigned long action, void *hcpu) | |
5933 | { | |
5934 | switch (action & ~CPU_TASKS_FROZEN) { | |
5935 | case CPU_DOWN_PREPARE: | |
5936 | set_cpu_active((long)hcpu, false); | |
5937 | return NOTIFY_OK; | |
5938 | default: | |
5939 | return NOTIFY_DONE; | |
5940 | } | |
5941 | } | |
5942 | ||
7babe8db | 5943 | static int __init migration_init(void) |
1da177e4 LT |
5944 | { |
5945 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5946 | int err; |
48f24c4d | 5947 | |
3a101d05 | 5948 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5949 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5950 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5951 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5952 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5953 | |
3a101d05 TH |
5954 | /* Register cpu active notifiers */ |
5955 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5956 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5957 | ||
a004cd42 | 5958 | return 0; |
1da177e4 | 5959 | } |
7babe8db | 5960 | early_initcall(migration_init); |
1da177e4 LT |
5961 | #endif |
5962 | ||
5963 | #ifdef CONFIG_SMP | |
476f3534 | 5964 | |
3e9830dc | 5965 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5966 | |
f6630114 MT |
5967 | static __read_mostly int sched_domain_debug_enabled; |
5968 | ||
5969 | static int __init sched_domain_debug_setup(char *str) | |
5970 | { | |
5971 | sched_domain_debug_enabled = 1; | |
5972 | ||
5973 | return 0; | |
5974 | } | |
5975 | early_param("sched_debug", sched_domain_debug_setup); | |
5976 | ||
7c16ec58 | 5977 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5978 | struct cpumask *groupmask) |
1da177e4 | 5979 | { |
4dcf6aff | 5980 | struct sched_group *group = sd->groups; |
434d53b0 | 5981 | char str[256]; |
1da177e4 | 5982 | |
968ea6d8 | 5983 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5984 | cpumask_clear(groupmask); |
4dcf6aff IM |
5985 | |
5986 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5987 | ||
5988 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5989 | printk("does not load-balance\n"); |
4dcf6aff | 5990 | if (sd->parent) |
3df0fc5b PZ |
5991 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5992 | " has parent"); | |
4dcf6aff | 5993 | return -1; |
41c7ce9a NP |
5994 | } |
5995 | ||
3df0fc5b | 5996 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5997 | |
758b2cdc | 5998 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5999 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6000 | "CPU%d\n", cpu); | |
4dcf6aff | 6001 | } |
758b2cdc | 6002 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6003 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6004 | " CPU%d\n", cpu); | |
4dcf6aff | 6005 | } |
1da177e4 | 6006 | |
4dcf6aff | 6007 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6008 | do { |
4dcf6aff | 6009 | if (!group) { |
3df0fc5b PZ |
6010 | printk("\n"); |
6011 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6012 | break; |
6013 | } | |
6014 | ||
18a3885f | 6015 | if (!group->cpu_power) { |
3df0fc5b PZ |
6016 | printk(KERN_CONT "\n"); |
6017 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6018 | "set\n"); | |
4dcf6aff IM |
6019 | break; |
6020 | } | |
1da177e4 | 6021 | |
758b2cdc | 6022 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6023 | printk(KERN_CONT "\n"); |
6024 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6025 | break; |
6026 | } | |
1da177e4 | 6027 | |
758b2cdc | 6028 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6029 | printk(KERN_CONT "\n"); |
6030 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6031 | break; |
6032 | } | |
1da177e4 | 6033 | |
758b2cdc | 6034 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6035 | |
968ea6d8 | 6036 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6037 | |
3df0fc5b | 6038 | printk(KERN_CONT " %s", str); |
18a3885f | 6039 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6040 | printk(KERN_CONT " (cpu_power = %d)", |
6041 | group->cpu_power); | |
381512cf | 6042 | } |
1da177e4 | 6043 | |
4dcf6aff IM |
6044 | group = group->next; |
6045 | } while (group != sd->groups); | |
3df0fc5b | 6046 | printk(KERN_CONT "\n"); |
1da177e4 | 6047 | |
758b2cdc | 6048 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6049 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6050 | |
758b2cdc RR |
6051 | if (sd->parent && |
6052 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6053 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6054 | "of domain->span\n"); | |
4dcf6aff IM |
6055 | return 0; |
6056 | } | |
1da177e4 | 6057 | |
4dcf6aff IM |
6058 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6059 | { | |
d5dd3db1 | 6060 | cpumask_var_t groupmask; |
4dcf6aff | 6061 | int level = 0; |
1da177e4 | 6062 | |
f6630114 MT |
6063 | if (!sched_domain_debug_enabled) |
6064 | return; | |
6065 | ||
4dcf6aff IM |
6066 | if (!sd) { |
6067 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6068 | return; | |
6069 | } | |
1da177e4 | 6070 | |
4dcf6aff IM |
6071 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6072 | ||
d5dd3db1 | 6073 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6074 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6075 | return; | |
6076 | } | |
6077 | ||
4dcf6aff | 6078 | for (;;) { |
7c16ec58 | 6079 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6080 | break; |
1da177e4 LT |
6081 | level++; |
6082 | sd = sd->parent; | |
33859f7f | 6083 | if (!sd) |
4dcf6aff IM |
6084 | break; |
6085 | } | |
d5dd3db1 | 6086 | free_cpumask_var(groupmask); |
1da177e4 | 6087 | } |
6d6bc0ad | 6088 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6089 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6090 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6091 | |
1a20ff27 | 6092 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6093 | { |
758b2cdc | 6094 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6095 | return 1; |
6096 | ||
6097 | /* Following flags need at least 2 groups */ | |
6098 | if (sd->flags & (SD_LOAD_BALANCE | | |
6099 | SD_BALANCE_NEWIDLE | | |
6100 | SD_BALANCE_FORK | | |
89c4710e SS |
6101 | SD_BALANCE_EXEC | |
6102 | SD_SHARE_CPUPOWER | | |
6103 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6104 | if (sd->groups != sd->groups->next) |
6105 | return 0; | |
6106 | } | |
6107 | ||
6108 | /* Following flags don't use groups */ | |
c88d5910 | 6109 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6110 | return 0; |
6111 | ||
6112 | return 1; | |
6113 | } | |
6114 | ||
48f24c4d IM |
6115 | static int |
6116 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6117 | { |
6118 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6119 | ||
6120 | if (sd_degenerate(parent)) | |
6121 | return 1; | |
6122 | ||
758b2cdc | 6123 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6124 | return 0; |
6125 | ||
245af2c7 SS |
6126 | /* Flags needing groups don't count if only 1 group in parent */ |
6127 | if (parent->groups == parent->groups->next) { | |
6128 | pflags &= ~(SD_LOAD_BALANCE | | |
6129 | SD_BALANCE_NEWIDLE | | |
6130 | SD_BALANCE_FORK | | |
89c4710e SS |
6131 | SD_BALANCE_EXEC | |
6132 | SD_SHARE_CPUPOWER | | |
6133 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6134 | if (nr_node_ids == 1) |
6135 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6136 | } |
6137 | if (~cflags & pflags) | |
6138 | return 0; | |
6139 | ||
6140 | return 1; | |
6141 | } | |
6142 | ||
c6c4927b RR |
6143 | static void free_rootdomain(struct root_domain *rd) |
6144 | { | |
047106ad PZ |
6145 | synchronize_sched(); |
6146 | ||
68e74568 RR |
6147 | cpupri_cleanup(&rd->cpupri); |
6148 | ||
c6c4927b RR |
6149 | free_cpumask_var(rd->rto_mask); |
6150 | free_cpumask_var(rd->online); | |
6151 | free_cpumask_var(rd->span); | |
6152 | kfree(rd); | |
6153 | } | |
6154 | ||
57d885fe GH |
6155 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6156 | { | |
a0490fa3 | 6157 | struct root_domain *old_rd = NULL; |
57d885fe | 6158 | unsigned long flags; |
57d885fe | 6159 | |
05fa785c | 6160 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6161 | |
6162 | if (rq->rd) { | |
a0490fa3 | 6163 | old_rd = rq->rd; |
57d885fe | 6164 | |
c6c4927b | 6165 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6166 | set_rq_offline(rq); |
57d885fe | 6167 | |
c6c4927b | 6168 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6169 | |
a0490fa3 IM |
6170 | /* |
6171 | * If we dont want to free the old_rt yet then | |
6172 | * set old_rd to NULL to skip the freeing later | |
6173 | * in this function: | |
6174 | */ | |
6175 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6176 | old_rd = NULL; | |
57d885fe GH |
6177 | } |
6178 | ||
6179 | atomic_inc(&rd->refcount); | |
6180 | rq->rd = rd; | |
6181 | ||
c6c4927b | 6182 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6183 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6184 | set_rq_online(rq); |
57d885fe | 6185 | |
05fa785c | 6186 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6187 | |
6188 | if (old_rd) | |
6189 | free_rootdomain(old_rd); | |
57d885fe GH |
6190 | } |
6191 | ||
68c38fc3 | 6192 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6193 | { |
6194 | memset(rd, 0, sizeof(*rd)); | |
6195 | ||
68c38fc3 | 6196 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6197 | goto out; |
68c38fc3 | 6198 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6199 | goto free_span; |
68c38fc3 | 6200 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6201 | goto free_online; |
6e0534f2 | 6202 | |
68c38fc3 | 6203 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6204 | goto free_rto_mask; |
c6c4927b | 6205 | return 0; |
6e0534f2 | 6206 | |
68e74568 RR |
6207 | free_rto_mask: |
6208 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6209 | free_online: |
6210 | free_cpumask_var(rd->online); | |
6211 | free_span: | |
6212 | free_cpumask_var(rd->span); | |
0c910d28 | 6213 | out: |
c6c4927b | 6214 | return -ENOMEM; |
57d885fe GH |
6215 | } |
6216 | ||
6217 | static void init_defrootdomain(void) | |
6218 | { | |
68c38fc3 | 6219 | init_rootdomain(&def_root_domain); |
c6c4927b | 6220 | |
57d885fe GH |
6221 | atomic_set(&def_root_domain.refcount, 1); |
6222 | } | |
6223 | ||
dc938520 | 6224 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6225 | { |
6226 | struct root_domain *rd; | |
6227 | ||
6228 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6229 | if (!rd) | |
6230 | return NULL; | |
6231 | ||
68c38fc3 | 6232 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6233 | kfree(rd); |
6234 | return NULL; | |
6235 | } | |
57d885fe GH |
6236 | |
6237 | return rd; | |
6238 | } | |
6239 | ||
1da177e4 | 6240 | /* |
0eab9146 | 6241 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6242 | * hold the hotplug lock. |
6243 | */ | |
0eab9146 IM |
6244 | static void |
6245 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6246 | { |
70b97a7f | 6247 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6248 | struct sched_domain *tmp; |
6249 | ||
669c55e9 PZ |
6250 | for (tmp = sd; tmp; tmp = tmp->parent) |
6251 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6252 | ||
245af2c7 | 6253 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6254 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6255 | struct sched_domain *parent = tmp->parent; |
6256 | if (!parent) | |
6257 | break; | |
f29c9b1c | 6258 | |
1a848870 | 6259 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6260 | tmp->parent = parent->parent; |
1a848870 SS |
6261 | if (parent->parent) |
6262 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6263 | } else |
6264 | tmp = tmp->parent; | |
245af2c7 SS |
6265 | } |
6266 | ||
1a848870 | 6267 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6268 | sd = sd->parent; |
1a848870 SS |
6269 | if (sd) |
6270 | sd->child = NULL; | |
6271 | } | |
1da177e4 LT |
6272 | |
6273 | sched_domain_debug(sd, cpu); | |
6274 | ||
57d885fe | 6275 | rq_attach_root(rq, rd); |
674311d5 | 6276 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6277 | } |
6278 | ||
6279 | /* cpus with isolated domains */ | |
dcc30a35 | 6280 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6281 | |
6282 | /* Setup the mask of cpus configured for isolated domains */ | |
6283 | static int __init isolated_cpu_setup(char *str) | |
6284 | { | |
bdddd296 | 6285 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6286 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6287 | return 1; |
6288 | } | |
6289 | ||
8927f494 | 6290 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6291 | |
6292 | /* | |
6711cab4 SS |
6293 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6294 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6295 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6296 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6297 | * |
6298 | * init_sched_build_groups will build a circular linked list of the groups | |
6299 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6300 | * and ->cpu_power to 0. | |
6301 | */ | |
a616058b | 6302 | static void |
96f874e2 RR |
6303 | init_sched_build_groups(const struct cpumask *span, |
6304 | const struct cpumask *cpu_map, | |
6305 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6306 | struct sched_group **sg, |
96f874e2 RR |
6307 | struct cpumask *tmpmask), |
6308 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6309 | { |
6310 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6311 | int i; |
6312 | ||
96f874e2 | 6313 | cpumask_clear(covered); |
7c16ec58 | 6314 | |
abcd083a | 6315 | for_each_cpu(i, span) { |
6711cab4 | 6316 | struct sched_group *sg; |
7c16ec58 | 6317 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6318 | int j; |
6319 | ||
758b2cdc | 6320 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6321 | continue; |
6322 | ||
758b2cdc | 6323 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6324 | sg->cpu_power = 0; |
1da177e4 | 6325 | |
abcd083a | 6326 | for_each_cpu(j, span) { |
7c16ec58 | 6327 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6328 | continue; |
6329 | ||
96f874e2 | 6330 | cpumask_set_cpu(j, covered); |
758b2cdc | 6331 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6332 | } |
6333 | if (!first) | |
6334 | first = sg; | |
6335 | if (last) | |
6336 | last->next = sg; | |
6337 | last = sg; | |
6338 | } | |
6339 | last->next = first; | |
6340 | } | |
6341 | ||
9c1cfda2 | 6342 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6343 | |
9c1cfda2 | 6344 | #ifdef CONFIG_NUMA |
198e2f18 | 6345 | |
9c1cfda2 JH |
6346 | /** |
6347 | * find_next_best_node - find the next node to include in a sched_domain | |
6348 | * @node: node whose sched_domain we're building | |
6349 | * @used_nodes: nodes already in the sched_domain | |
6350 | * | |
41a2d6cf | 6351 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6352 | * finds the closest node not already in the @used_nodes map. |
6353 | * | |
6354 | * Should use nodemask_t. | |
6355 | */ | |
c5f59f08 | 6356 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6357 | { |
6358 | int i, n, val, min_val, best_node = 0; | |
6359 | ||
6360 | min_val = INT_MAX; | |
6361 | ||
076ac2af | 6362 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6363 | /* Start at @node */ |
076ac2af | 6364 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6365 | |
6366 | if (!nr_cpus_node(n)) | |
6367 | continue; | |
6368 | ||
6369 | /* Skip already used nodes */ | |
c5f59f08 | 6370 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6371 | continue; |
6372 | ||
6373 | /* Simple min distance search */ | |
6374 | val = node_distance(node, n); | |
6375 | ||
6376 | if (val < min_val) { | |
6377 | min_val = val; | |
6378 | best_node = n; | |
6379 | } | |
6380 | } | |
6381 | ||
c5f59f08 | 6382 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6383 | return best_node; |
6384 | } | |
6385 | ||
6386 | /** | |
6387 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6388 | * @node: node whose cpumask we're constructing | |
73486722 | 6389 | * @span: resulting cpumask |
9c1cfda2 | 6390 | * |
41a2d6cf | 6391 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6392 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6393 | * out optimally. | |
6394 | */ | |
96f874e2 | 6395 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6396 | { |
c5f59f08 | 6397 | nodemask_t used_nodes; |
48f24c4d | 6398 | int i; |
9c1cfda2 | 6399 | |
6ca09dfc | 6400 | cpumask_clear(span); |
c5f59f08 | 6401 | nodes_clear(used_nodes); |
9c1cfda2 | 6402 | |
6ca09dfc | 6403 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6404 | node_set(node, used_nodes); |
9c1cfda2 JH |
6405 | |
6406 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6407 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6408 | |
6ca09dfc | 6409 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6410 | } |
9c1cfda2 | 6411 | } |
6d6bc0ad | 6412 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6413 | |
5c45bf27 | 6414 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6415 | |
6c99e9ad RR |
6416 | /* |
6417 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6418 | * |
6419 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6420 | * and struct sched_domain. ) | |
6c99e9ad RR |
6421 | */ |
6422 | struct static_sched_group { | |
6423 | struct sched_group sg; | |
6424 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6425 | }; | |
6426 | ||
6427 | struct static_sched_domain { | |
6428 | struct sched_domain sd; | |
6429 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6430 | }; | |
6431 | ||
49a02c51 AH |
6432 | struct s_data { |
6433 | #ifdef CONFIG_NUMA | |
6434 | int sd_allnodes; | |
6435 | cpumask_var_t domainspan; | |
6436 | cpumask_var_t covered; | |
6437 | cpumask_var_t notcovered; | |
6438 | #endif | |
6439 | cpumask_var_t nodemask; | |
6440 | cpumask_var_t this_sibling_map; | |
6441 | cpumask_var_t this_core_map; | |
01a08546 | 6442 | cpumask_var_t this_book_map; |
49a02c51 AH |
6443 | cpumask_var_t send_covered; |
6444 | cpumask_var_t tmpmask; | |
6445 | struct sched_group **sched_group_nodes; | |
6446 | struct root_domain *rd; | |
6447 | }; | |
6448 | ||
2109b99e AH |
6449 | enum s_alloc { |
6450 | sa_sched_groups = 0, | |
6451 | sa_rootdomain, | |
6452 | sa_tmpmask, | |
6453 | sa_send_covered, | |
01a08546 | 6454 | sa_this_book_map, |
2109b99e AH |
6455 | sa_this_core_map, |
6456 | sa_this_sibling_map, | |
6457 | sa_nodemask, | |
6458 | sa_sched_group_nodes, | |
6459 | #ifdef CONFIG_NUMA | |
6460 | sa_notcovered, | |
6461 | sa_covered, | |
6462 | sa_domainspan, | |
6463 | #endif | |
6464 | sa_none, | |
6465 | }; | |
6466 | ||
9c1cfda2 | 6467 | /* |
48f24c4d | 6468 | * SMT sched-domains: |
9c1cfda2 | 6469 | */ |
1da177e4 | 6470 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6471 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6472 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6473 | |
41a2d6cf | 6474 | static int |
96f874e2 RR |
6475 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6476 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6477 | { |
6711cab4 | 6478 | if (sg) |
1871e52c | 6479 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6480 | return cpu; |
6481 | } | |
6d6bc0ad | 6482 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6483 | |
48f24c4d IM |
6484 | /* |
6485 | * multi-core sched-domains: | |
6486 | */ | |
1e9f28fa | 6487 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6488 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6489 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6490 | |
41a2d6cf | 6491 | static int |
96f874e2 RR |
6492 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6493 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6494 | { |
6711cab4 | 6495 | int group; |
f269893c | 6496 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 6497 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6498 | group = cpumask_first(mask); |
f269893c HC |
6499 | #else |
6500 | group = cpu; | |
6501 | #endif | |
6711cab4 | 6502 | if (sg) |
6c99e9ad | 6503 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6504 | return group; |
1e9f28fa | 6505 | } |
f269893c | 6506 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 6507 | |
01a08546 HC |
6508 | /* |
6509 | * book sched-domains: | |
6510 | */ | |
6511 | #ifdef CONFIG_SCHED_BOOK | |
6512 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
6513 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
6514 | ||
41a2d6cf | 6515 | static int |
01a08546 HC |
6516 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
6517 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6518 | { |
01a08546 HC |
6519 | int group = cpu; |
6520 | #ifdef CONFIG_SCHED_MC | |
6521 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
6522 | group = cpumask_first(mask); | |
6523 | #elif defined(CONFIG_SCHED_SMT) | |
6524 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
6525 | group = cpumask_first(mask); | |
6526 | #endif | |
6711cab4 | 6527 | if (sg) |
01a08546 HC |
6528 | *sg = &per_cpu(sched_group_book, group).sg; |
6529 | return group; | |
1e9f28fa | 6530 | } |
01a08546 | 6531 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 6532 | |
6c99e9ad RR |
6533 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6534 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6535 | |
41a2d6cf | 6536 | static int |
96f874e2 RR |
6537 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6538 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6539 | { |
6711cab4 | 6540 | int group; |
01a08546 HC |
6541 | #ifdef CONFIG_SCHED_BOOK |
6542 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
6543 | group = cpumask_first(mask); | |
6544 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 6545 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6546 | group = cpumask_first(mask); |
1e9f28fa | 6547 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6548 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6549 | group = cpumask_first(mask); |
1da177e4 | 6550 | #else |
6711cab4 | 6551 | group = cpu; |
1da177e4 | 6552 | #endif |
6711cab4 | 6553 | if (sg) |
6c99e9ad | 6554 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6555 | return group; |
1da177e4 LT |
6556 | } |
6557 | ||
6558 | #ifdef CONFIG_NUMA | |
1da177e4 | 6559 | /* |
9c1cfda2 JH |
6560 | * The init_sched_build_groups can't handle what we want to do with node |
6561 | * groups, so roll our own. Now each node has its own list of groups which | |
6562 | * gets dynamically allocated. | |
1da177e4 | 6563 | */ |
62ea9ceb | 6564 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6565 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6566 | |
62ea9ceb | 6567 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6568 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6569 | |
96f874e2 RR |
6570 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6571 | struct sched_group **sg, | |
6572 | struct cpumask *nodemask) | |
9c1cfda2 | 6573 | { |
6711cab4 SS |
6574 | int group; |
6575 | ||
6ca09dfc | 6576 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6577 | group = cpumask_first(nodemask); |
6711cab4 SS |
6578 | |
6579 | if (sg) | |
6c99e9ad | 6580 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6581 | return group; |
1da177e4 | 6582 | } |
6711cab4 | 6583 | |
08069033 SS |
6584 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6585 | { | |
6586 | struct sched_group *sg = group_head; | |
6587 | int j; | |
6588 | ||
6589 | if (!sg) | |
6590 | return; | |
3a5c359a | 6591 | do { |
758b2cdc | 6592 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6593 | struct sched_domain *sd; |
08069033 | 6594 | |
6c99e9ad | 6595 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6596 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6597 | /* |
6598 | * Only add "power" once for each | |
6599 | * physical package. | |
6600 | */ | |
6601 | continue; | |
6602 | } | |
08069033 | 6603 | |
18a3885f | 6604 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6605 | } |
6606 | sg = sg->next; | |
6607 | } while (sg != group_head); | |
08069033 | 6608 | } |
0601a88d AH |
6609 | |
6610 | static int build_numa_sched_groups(struct s_data *d, | |
6611 | const struct cpumask *cpu_map, int num) | |
6612 | { | |
6613 | struct sched_domain *sd; | |
6614 | struct sched_group *sg, *prev; | |
6615 | int n, j; | |
6616 | ||
6617 | cpumask_clear(d->covered); | |
6618 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6619 | if (cpumask_empty(d->nodemask)) { | |
6620 | d->sched_group_nodes[num] = NULL; | |
6621 | goto out; | |
6622 | } | |
6623 | ||
6624 | sched_domain_node_span(num, d->domainspan); | |
6625 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6626 | ||
6627 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6628 | GFP_KERNEL, num); | |
6629 | if (!sg) { | |
3df0fc5b PZ |
6630 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6631 | num); | |
0601a88d AH |
6632 | return -ENOMEM; |
6633 | } | |
6634 | d->sched_group_nodes[num] = sg; | |
6635 | ||
6636 | for_each_cpu(j, d->nodemask) { | |
6637 | sd = &per_cpu(node_domains, j).sd; | |
6638 | sd->groups = sg; | |
6639 | } | |
6640 | ||
18a3885f | 6641 | sg->cpu_power = 0; |
0601a88d AH |
6642 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6643 | sg->next = sg; | |
6644 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6645 | ||
6646 | prev = sg; | |
6647 | for (j = 0; j < nr_node_ids; j++) { | |
6648 | n = (num + j) % nr_node_ids; | |
6649 | cpumask_complement(d->notcovered, d->covered); | |
6650 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6651 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6652 | if (cpumask_empty(d->tmpmask)) | |
6653 | break; | |
6654 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6655 | if (cpumask_empty(d->tmpmask)) | |
6656 | continue; | |
6657 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6658 | GFP_KERNEL, num); | |
6659 | if (!sg) { | |
3df0fc5b PZ |
6660 | printk(KERN_WARNING |
6661 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6662 | return -ENOMEM; |
6663 | } | |
18a3885f | 6664 | sg->cpu_power = 0; |
0601a88d AH |
6665 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6666 | sg->next = prev->next; | |
6667 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6668 | prev->next = sg; | |
6669 | prev = sg; | |
6670 | } | |
6671 | out: | |
6672 | return 0; | |
6673 | } | |
6d6bc0ad | 6674 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6675 | |
a616058b | 6676 | #ifdef CONFIG_NUMA |
51888ca2 | 6677 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6678 | static void free_sched_groups(const struct cpumask *cpu_map, |
6679 | struct cpumask *nodemask) | |
51888ca2 | 6680 | { |
a616058b | 6681 | int cpu, i; |
51888ca2 | 6682 | |
abcd083a | 6683 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6684 | struct sched_group **sched_group_nodes |
6685 | = sched_group_nodes_bycpu[cpu]; | |
6686 | ||
51888ca2 SV |
6687 | if (!sched_group_nodes) |
6688 | continue; | |
6689 | ||
076ac2af | 6690 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6691 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6692 | ||
6ca09dfc | 6693 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6694 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6695 | continue; |
6696 | ||
6697 | if (sg == NULL) | |
6698 | continue; | |
6699 | sg = sg->next; | |
6700 | next_sg: | |
6701 | oldsg = sg; | |
6702 | sg = sg->next; | |
6703 | kfree(oldsg); | |
6704 | if (oldsg != sched_group_nodes[i]) | |
6705 | goto next_sg; | |
6706 | } | |
6707 | kfree(sched_group_nodes); | |
6708 | sched_group_nodes_bycpu[cpu] = NULL; | |
6709 | } | |
51888ca2 | 6710 | } |
6d6bc0ad | 6711 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6712 | static void free_sched_groups(const struct cpumask *cpu_map, |
6713 | struct cpumask *nodemask) | |
a616058b SS |
6714 | { |
6715 | } | |
6d6bc0ad | 6716 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6717 | |
89c4710e SS |
6718 | /* |
6719 | * Initialize sched groups cpu_power. | |
6720 | * | |
6721 | * cpu_power indicates the capacity of sched group, which is used while | |
6722 | * distributing the load between different sched groups in a sched domain. | |
6723 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6724 | * there are asymmetries in the topology. If there are asymmetries, group | |
6725 | * having more cpu_power will pickup more load compared to the group having | |
6726 | * less cpu_power. | |
89c4710e SS |
6727 | */ |
6728 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6729 | { | |
6730 | struct sched_domain *child; | |
6731 | struct sched_group *group; | |
f93e65c1 PZ |
6732 | long power; |
6733 | int weight; | |
89c4710e SS |
6734 | |
6735 | WARN_ON(!sd || !sd->groups); | |
6736 | ||
13318a71 | 6737 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6738 | return; |
6739 | ||
6740 | child = sd->child; | |
6741 | ||
18a3885f | 6742 | sd->groups->cpu_power = 0; |
5517d86b | 6743 | |
f93e65c1 PZ |
6744 | if (!child) { |
6745 | power = SCHED_LOAD_SCALE; | |
6746 | weight = cpumask_weight(sched_domain_span(sd)); | |
6747 | /* | |
6748 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6749 | * Usually multiple threads get a better yield out of |
6750 | * that one core than a single thread would have, | |
6751 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6752 | */ |
a52bfd73 PZ |
6753 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6754 | power *= sd->smt_gain; | |
f93e65c1 | 6755 | power /= weight; |
a52bfd73 PZ |
6756 | power >>= SCHED_LOAD_SHIFT; |
6757 | } | |
18a3885f | 6758 | sd->groups->cpu_power += power; |
89c4710e SS |
6759 | return; |
6760 | } | |
6761 | ||
89c4710e | 6762 | /* |
f93e65c1 | 6763 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6764 | */ |
6765 | group = child->groups; | |
6766 | do { | |
18a3885f | 6767 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6768 | group = group->next; |
6769 | } while (group != child->groups); | |
6770 | } | |
6771 | ||
7c16ec58 MT |
6772 | /* |
6773 | * Initializers for schedule domains | |
6774 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6775 | */ | |
6776 | ||
a5d8c348 IM |
6777 | #ifdef CONFIG_SCHED_DEBUG |
6778 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6779 | #else | |
6780 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6781 | #endif | |
6782 | ||
7c16ec58 | 6783 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6784 | |
7c16ec58 MT |
6785 | #define SD_INIT_FUNC(type) \ |
6786 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6787 | { \ | |
6788 | memset(sd, 0, sizeof(*sd)); \ | |
6789 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6790 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6791 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6792 | } |
6793 | ||
6794 | SD_INIT_FUNC(CPU) | |
6795 | #ifdef CONFIG_NUMA | |
6796 | SD_INIT_FUNC(ALLNODES) | |
6797 | SD_INIT_FUNC(NODE) | |
6798 | #endif | |
6799 | #ifdef CONFIG_SCHED_SMT | |
6800 | SD_INIT_FUNC(SIBLING) | |
6801 | #endif | |
6802 | #ifdef CONFIG_SCHED_MC | |
6803 | SD_INIT_FUNC(MC) | |
6804 | #endif | |
01a08546 HC |
6805 | #ifdef CONFIG_SCHED_BOOK |
6806 | SD_INIT_FUNC(BOOK) | |
6807 | #endif | |
7c16ec58 | 6808 | |
1d3504fc HS |
6809 | static int default_relax_domain_level = -1; |
6810 | ||
6811 | static int __init setup_relax_domain_level(char *str) | |
6812 | { | |
30e0e178 LZ |
6813 | unsigned long val; |
6814 | ||
6815 | val = simple_strtoul(str, NULL, 0); | |
6816 | if (val < SD_LV_MAX) | |
6817 | default_relax_domain_level = val; | |
6818 | ||
1d3504fc HS |
6819 | return 1; |
6820 | } | |
6821 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6822 | ||
6823 | static void set_domain_attribute(struct sched_domain *sd, | |
6824 | struct sched_domain_attr *attr) | |
6825 | { | |
6826 | int request; | |
6827 | ||
6828 | if (!attr || attr->relax_domain_level < 0) { | |
6829 | if (default_relax_domain_level < 0) | |
6830 | return; | |
6831 | else | |
6832 | request = default_relax_domain_level; | |
6833 | } else | |
6834 | request = attr->relax_domain_level; | |
6835 | if (request < sd->level) { | |
6836 | /* turn off idle balance on this domain */ | |
c88d5910 | 6837 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6838 | } else { |
6839 | /* turn on idle balance on this domain */ | |
c88d5910 | 6840 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6841 | } |
6842 | } | |
6843 | ||
2109b99e AH |
6844 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6845 | const struct cpumask *cpu_map) | |
6846 | { | |
6847 | switch (what) { | |
6848 | case sa_sched_groups: | |
6849 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6850 | d->sched_group_nodes = NULL; | |
6851 | case sa_rootdomain: | |
6852 | free_rootdomain(d->rd); /* fall through */ | |
6853 | case sa_tmpmask: | |
6854 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6855 | case sa_send_covered: | |
6856 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
6857 | case sa_this_book_map: |
6858 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
6859 | case sa_this_core_map: |
6860 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6861 | case sa_this_sibling_map: | |
6862 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6863 | case sa_nodemask: | |
6864 | free_cpumask_var(d->nodemask); /* fall through */ | |
6865 | case sa_sched_group_nodes: | |
d1b55138 | 6866 | #ifdef CONFIG_NUMA |
2109b99e AH |
6867 | kfree(d->sched_group_nodes); /* fall through */ |
6868 | case sa_notcovered: | |
6869 | free_cpumask_var(d->notcovered); /* fall through */ | |
6870 | case sa_covered: | |
6871 | free_cpumask_var(d->covered); /* fall through */ | |
6872 | case sa_domainspan: | |
6873 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6874 | #endif |
2109b99e AH |
6875 | case sa_none: |
6876 | break; | |
6877 | } | |
6878 | } | |
3404c8d9 | 6879 | |
2109b99e AH |
6880 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6881 | const struct cpumask *cpu_map) | |
6882 | { | |
3404c8d9 | 6883 | #ifdef CONFIG_NUMA |
2109b99e AH |
6884 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6885 | return sa_none; | |
6886 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6887 | return sa_domainspan; | |
6888 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6889 | return sa_covered; | |
6890 | /* Allocate the per-node list of sched groups */ | |
6891 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6892 | sizeof(struct sched_group *), GFP_KERNEL); | |
6893 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6894 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6895 | return sa_notcovered; |
d1b55138 | 6896 | } |
2109b99e | 6897 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6898 | #endif |
2109b99e AH |
6899 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6900 | return sa_sched_group_nodes; | |
6901 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6902 | return sa_nodemask; | |
6903 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6904 | return sa_this_sibling_map; | |
01a08546 | 6905 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 6906 | return sa_this_core_map; |
01a08546 HC |
6907 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
6908 | return sa_this_book_map; | |
2109b99e AH |
6909 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
6910 | return sa_send_covered; | |
6911 | d->rd = alloc_rootdomain(); | |
6912 | if (!d->rd) { | |
3df0fc5b | 6913 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6914 | return sa_tmpmask; |
57d885fe | 6915 | } |
2109b99e AH |
6916 | return sa_rootdomain; |
6917 | } | |
57d885fe | 6918 | |
7f4588f3 AH |
6919 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6920 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6921 | { | |
6922 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6923 | #ifdef CONFIG_NUMA |
7f4588f3 | 6924 | struct sched_domain *parent; |
1da177e4 | 6925 | |
7f4588f3 AH |
6926 | d->sd_allnodes = 0; |
6927 | if (cpumask_weight(cpu_map) > | |
6928 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6929 | sd = &per_cpu(allnodes_domains, i).sd; | |
6930 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6931 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6932 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6933 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6934 | d->sd_allnodes = 1; | |
6935 | } | |
6936 | parent = sd; | |
6937 | ||
6938 | sd = &per_cpu(node_domains, i).sd; | |
6939 | SD_INIT(sd, NODE); | |
6940 | set_domain_attribute(sd, attr); | |
6941 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6942 | sd->parent = parent; | |
6943 | if (parent) | |
6944 | parent->child = sd; | |
6945 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6946 | #endif |
7f4588f3 AH |
6947 | return sd; |
6948 | } | |
1da177e4 | 6949 | |
87cce662 AH |
6950 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6951 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6952 | struct sched_domain *parent, int i) | |
6953 | { | |
6954 | struct sched_domain *sd; | |
6955 | sd = &per_cpu(phys_domains, i).sd; | |
6956 | SD_INIT(sd, CPU); | |
6957 | set_domain_attribute(sd, attr); | |
6958 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6959 | sd->parent = parent; | |
6960 | if (parent) | |
6961 | parent->child = sd; | |
6962 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6963 | return sd; | |
6964 | } | |
1da177e4 | 6965 | |
01a08546 HC |
6966 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
6967 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6968 | struct sched_domain *parent, int i) | |
6969 | { | |
6970 | struct sched_domain *sd = parent; | |
6971 | #ifdef CONFIG_SCHED_BOOK | |
6972 | sd = &per_cpu(book_domains, i).sd; | |
6973 | SD_INIT(sd, BOOK); | |
6974 | set_domain_attribute(sd, attr); | |
6975 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
6976 | sd->parent = parent; | |
6977 | parent->child = sd; | |
6978 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6979 | #endif | |
6980 | return sd; | |
6981 | } | |
6982 | ||
410c4081 AH |
6983 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
6984 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6985 | struct sched_domain *parent, int i) | |
6986 | { | |
6987 | struct sched_domain *sd = parent; | |
1e9f28fa | 6988 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
6989 | sd = &per_cpu(core_domains, i).sd; |
6990 | SD_INIT(sd, MC); | |
6991 | set_domain_attribute(sd, attr); | |
6992 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
6993 | sd->parent = parent; | |
6994 | parent->child = sd; | |
6995 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 6996 | #endif |
410c4081 AH |
6997 | return sd; |
6998 | } | |
1e9f28fa | 6999 | |
d8173535 AH |
7000 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7001 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7002 | struct sched_domain *parent, int i) | |
7003 | { | |
7004 | struct sched_domain *sd = parent; | |
1da177e4 | 7005 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7006 | sd = &per_cpu(cpu_domains, i).sd; |
7007 | SD_INIT(sd, SIBLING); | |
7008 | set_domain_attribute(sd, attr); | |
7009 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7010 | sd->parent = parent; | |
7011 | parent->child = sd; | |
7012 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7013 | #endif |
d8173535 AH |
7014 | return sd; |
7015 | } | |
1da177e4 | 7016 | |
0e8e85c9 AH |
7017 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7018 | const struct cpumask *cpu_map, int cpu) | |
7019 | { | |
7020 | switch (l) { | |
1da177e4 | 7021 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7022 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7023 | cpumask_and(d->this_sibling_map, cpu_map, | |
7024 | topology_thread_cpumask(cpu)); | |
7025 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7026 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7027 | &cpu_to_cpu_group, | |
7028 | d->send_covered, d->tmpmask); | |
7029 | break; | |
1da177e4 | 7030 | #endif |
1e9f28fa | 7031 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7032 | case SD_LV_MC: /* set up multi-core groups */ |
7033 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7034 | if (cpu == cpumask_first(d->this_core_map)) | |
7035 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7036 | &cpu_to_core_group, | |
7037 | d->send_covered, d->tmpmask); | |
7038 | break; | |
01a08546 HC |
7039 | #endif |
7040 | #ifdef CONFIG_SCHED_BOOK | |
7041 | case SD_LV_BOOK: /* set up book groups */ | |
7042 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7043 | if (cpu == cpumask_first(d->this_book_map)) | |
7044 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7045 | &cpu_to_book_group, | |
7046 | d->send_covered, d->tmpmask); | |
7047 | break; | |
1e9f28fa | 7048 | #endif |
86548096 AH |
7049 | case SD_LV_CPU: /* set up physical groups */ |
7050 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7051 | if (!cpumask_empty(d->nodemask)) | |
7052 | init_sched_build_groups(d->nodemask, cpu_map, | |
7053 | &cpu_to_phys_group, | |
7054 | d->send_covered, d->tmpmask); | |
7055 | break; | |
1da177e4 | 7056 | #ifdef CONFIG_NUMA |
de616e36 AH |
7057 | case SD_LV_ALLNODES: |
7058 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7059 | d->send_covered, d->tmpmask); | |
7060 | break; | |
7061 | #endif | |
0e8e85c9 AH |
7062 | default: |
7063 | break; | |
7c16ec58 | 7064 | } |
0e8e85c9 | 7065 | } |
9c1cfda2 | 7066 | |
2109b99e AH |
7067 | /* |
7068 | * Build sched domains for a given set of cpus and attach the sched domains | |
7069 | * to the individual cpus | |
7070 | */ | |
7071 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7072 | struct sched_domain_attr *attr) | |
7073 | { | |
7074 | enum s_alloc alloc_state = sa_none; | |
7075 | struct s_data d; | |
294b0c96 | 7076 | struct sched_domain *sd; |
2109b99e | 7077 | int i; |
7c16ec58 | 7078 | #ifdef CONFIG_NUMA |
2109b99e | 7079 | d.sd_allnodes = 0; |
7c16ec58 | 7080 | #endif |
9c1cfda2 | 7081 | |
2109b99e AH |
7082 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7083 | if (alloc_state != sa_rootdomain) | |
7084 | goto error; | |
7085 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7086 | |
1da177e4 | 7087 | /* |
1a20ff27 | 7088 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7089 | */ |
abcd083a | 7090 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7091 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7092 | cpu_map); | |
9761eea8 | 7093 | |
7f4588f3 | 7094 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7095 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7096 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7097 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7098 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7099 | } |
9c1cfda2 | 7100 | |
abcd083a | 7101 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7102 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7103 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7104 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7105 | } |
9c1cfda2 | 7106 | |
1da177e4 | 7107 | /* Set up physical groups */ |
86548096 AH |
7108 | for (i = 0; i < nr_node_ids; i++) |
7109 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7110 | |
1da177e4 LT |
7111 | #ifdef CONFIG_NUMA |
7112 | /* Set up node groups */ | |
de616e36 AH |
7113 | if (d.sd_allnodes) |
7114 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7115 | |
0601a88d AH |
7116 | for (i = 0; i < nr_node_ids; i++) |
7117 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7118 | goto error; |
1da177e4 LT |
7119 | #endif |
7120 | ||
7121 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7122 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7123 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7124 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7125 | init_sched_groups_power(i, sd); |
5c45bf27 | 7126 | } |
1da177e4 | 7127 | #endif |
1e9f28fa | 7128 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7129 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7130 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7131 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7132 | } |
7133 | #endif | |
01a08546 HC |
7134 | #ifdef CONFIG_SCHED_BOOK |
7135 | for_each_cpu(i, cpu_map) { | |
7136 | sd = &per_cpu(book_domains, i).sd; | |
7137 | init_sched_groups_power(i, sd); | |
7138 | } | |
7139 | #endif | |
1e9f28fa | 7140 | |
abcd083a | 7141 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7142 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7143 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7144 | } |
7145 | ||
9c1cfda2 | 7146 | #ifdef CONFIG_NUMA |
076ac2af | 7147 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7148 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7149 | |
49a02c51 | 7150 | if (d.sd_allnodes) { |
6711cab4 | 7151 | struct sched_group *sg; |
f712c0c7 | 7152 | |
96f874e2 | 7153 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7154 | d.tmpmask); |
f712c0c7 SS |
7155 | init_numa_sched_groups_power(sg); |
7156 | } | |
9c1cfda2 JH |
7157 | #endif |
7158 | ||
1da177e4 | 7159 | /* Attach the domains */ |
abcd083a | 7160 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7161 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7162 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7163 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7164 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7165 | #elif defined(CONFIG_SCHED_BOOK) |
7166 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7167 | #else |
6c99e9ad | 7168 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7169 | #endif |
49a02c51 | 7170 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7171 | } |
51888ca2 | 7172 | |
2109b99e AH |
7173 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7174 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7175 | return 0; | |
51888ca2 | 7176 | |
51888ca2 | 7177 | error: |
2109b99e AH |
7178 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7179 | return -ENOMEM; | |
1da177e4 | 7180 | } |
029190c5 | 7181 | |
96f874e2 | 7182 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7183 | { |
7184 | return __build_sched_domains(cpu_map, NULL); | |
7185 | } | |
7186 | ||
acc3f5d7 | 7187 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7188 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7189 | static struct sched_domain_attr *dattr_cur; |
7190 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7191 | |
7192 | /* | |
7193 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7194 | * cpumask) fails, then fallback to a single sched domain, |
7195 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7196 | */ |
4212823f | 7197 | static cpumask_var_t fallback_doms; |
029190c5 | 7198 | |
ee79d1bd HC |
7199 | /* |
7200 | * arch_update_cpu_topology lets virtualized architectures update the | |
7201 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7202 | * or 0 if it stayed the same. | |
7203 | */ | |
7204 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7205 | { |
ee79d1bd | 7206 | return 0; |
22e52b07 HC |
7207 | } |
7208 | ||
acc3f5d7 RR |
7209 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7210 | { | |
7211 | int i; | |
7212 | cpumask_var_t *doms; | |
7213 | ||
7214 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7215 | if (!doms) | |
7216 | return NULL; | |
7217 | for (i = 0; i < ndoms; i++) { | |
7218 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7219 | free_sched_domains(doms, i); | |
7220 | return NULL; | |
7221 | } | |
7222 | } | |
7223 | return doms; | |
7224 | } | |
7225 | ||
7226 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7227 | { | |
7228 | unsigned int i; | |
7229 | for (i = 0; i < ndoms; i++) | |
7230 | free_cpumask_var(doms[i]); | |
7231 | kfree(doms); | |
7232 | } | |
7233 | ||
1a20ff27 | 7234 | /* |
41a2d6cf | 7235 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7236 | * For now this just excludes isolated cpus, but could be used to |
7237 | * exclude other special cases in the future. | |
1a20ff27 | 7238 | */ |
96f874e2 | 7239 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7240 | { |
7378547f MM |
7241 | int err; |
7242 | ||
22e52b07 | 7243 | arch_update_cpu_topology(); |
029190c5 | 7244 | ndoms_cur = 1; |
acc3f5d7 | 7245 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7246 | if (!doms_cur) |
acc3f5d7 RR |
7247 | doms_cur = &fallback_doms; |
7248 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7249 | dattr_cur = NULL; |
acc3f5d7 | 7250 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7251 | register_sched_domain_sysctl(); |
7378547f MM |
7252 | |
7253 | return err; | |
1a20ff27 DG |
7254 | } |
7255 | ||
96f874e2 RR |
7256 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7257 | struct cpumask *tmpmask) | |
1da177e4 | 7258 | { |
7c16ec58 | 7259 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7260 | } |
1da177e4 | 7261 | |
1a20ff27 DG |
7262 | /* |
7263 | * Detach sched domains from a group of cpus specified in cpu_map | |
7264 | * These cpus will now be attached to the NULL domain | |
7265 | */ | |
96f874e2 | 7266 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7267 | { |
96f874e2 RR |
7268 | /* Save because hotplug lock held. */ |
7269 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7270 | int i; |
7271 | ||
abcd083a | 7272 | for_each_cpu(i, cpu_map) |
57d885fe | 7273 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7274 | synchronize_sched(); |
96f874e2 | 7275 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7276 | } |
7277 | ||
1d3504fc HS |
7278 | /* handle null as "default" */ |
7279 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7280 | struct sched_domain_attr *new, int idx_new) | |
7281 | { | |
7282 | struct sched_domain_attr tmp; | |
7283 | ||
7284 | /* fast path */ | |
7285 | if (!new && !cur) | |
7286 | return 1; | |
7287 | ||
7288 | tmp = SD_ATTR_INIT; | |
7289 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7290 | new ? (new + idx_new) : &tmp, | |
7291 | sizeof(struct sched_domain_attr)); | |
7292 | } | |
7293 | ||
029190c5 PJ |
7294 | /* |
7295 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7296 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7297 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7298 | * It destroys each deleted domain and builds each new domain. | |
7299 | * | |
acc3f5d7 | 7300 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7301 | * The masks don't intersect (don't overlap.) We should setup one |
7302 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7303 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7304 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7305 | * it as it is. | |
7306 | * | |
acc3f5d7 RR |
7307 | * The passed in 'doms_new' should be allocated using |
7308 | * alloc_sched_domains. This routine takes ownership of it and will | |
7309 | * free_sched_domains it when done with it. If the caller failed the | |
7310 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7311 | * and partition_sched_domains() will fallback to the single partition | |
7312 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7313 | * |
96f874e2 | 7314 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7315 | * ndoms_new == 0 is a special case for destroying existing domains, |
7316 | * and it will not create the default domain. | |
dfb512ec | 7317 | * |
029190c5 PJ |
7318 | * Call with hotplug lock held |
7319 | */ | |
acc3f5d7 | 7320 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7321 | struct sched_domain_attr *dattr_new) |
029190c5 | 7322 | { |
dfb512ec | 7323 | int i, j, n; |
d65bd5ec | 7324 | int new_topology; |
029190c5 | 7325 | |
712555ee | 7326 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7327 | |
7378547f MM |
7328 | /* always unregister in case we don't destroy any domains */ |
7329 | unregister_sched_domain_sysctl(); | |
7330 | ||
d65bd5ec HC |
7331 | /* Let architecture update cpu core mappings. */ |
7332 | new_topology = arch_update_cpu_topology(); | |
7333 | ||
dfb512ec | 7334 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7335 | |
7336 | /* Destroy deleted domains */ | |
7337 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7338 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7339 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7340 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7341 | goto match1; |
7342 | } | |
7343 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7344 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7345 | match1: |
7346 | ; | |
7347 | } | |
7348 | ||
e761b772 MK |
7349 | if (doms_new == NULL) { |
7350 | ndoms_cur = 0; | |
acc3f5d7 | 7351 | doms_new = &fallback_doms; |
6ad4c188 | 7352 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7353 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7354 | } |
7355 | ||
029190c5 PJ |
7356 | /* Build new domains */ |
7357 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7358 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7359 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7360 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7361 | goto match2; |
7362 | } | |
7363 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7364 | __build_sched_domains(doms_new[i], |
1d3504fc | 7365 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7366 | match2: |
7367 | ; | |
7368 | } | |
7369 | ||
7370 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7371 | if (doms_cur != &fallback_doms) |
7372 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7373 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7374 | doms_cur = doms_new; |
1d3504fc | 7375 | dattr_cur = dattr_new; |
029190c5 | 7376 | ndoms_cur = ndoms_new; |
7378547f MM |
7377 | |
7378 | register_sched_domain_sysctl(); | |
a1835615 | 7379 | |
712555ee | 7380 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7381 | } |
7382 | ||
5c45bf27 | 7383 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7384 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7385 | { |
95402b38 | 7386 | get_online_cpus(); |
dfb512ec MK |
7387 | |
7388 | /* Destroy domains first to force the rebuild */ | |
7389 | partition_sched_domains(0, NULL, NULL); | |
7390 | ||
e761b772 | 7391 | rebuild_sched_domains(); |
95402b38 | 7392 | put_online_cpus(); |
5c45bf27 SS |
7393 | } |
7394 | ||
7395 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7396 | { | |
afb8a9b7 | 7397 | unsigned int level = 0; |
5c45bf27 | 7398 | |
afb8a9b7 GS |
7399 | if (sscanf(buf, "%u", &level) != 1) |
7400 | return -EINVAL; | |
7401 | ||
7402 | /* | |
7403 | * level is always be positive so don't check for | |
7404 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7405 | * What happens on 0 or 1 byte write, | |
7406 | * need to check for count as well? | |
7407 | */ | |
7408 | ||
7409 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7410 | return -EINVAL; |
7411 | ||
7412 | if (smt) | |
afb8a9b7 | 7413 | sched_smt_power_savings = level; |
5c45bf27 | 7414 | else |
afb8a9b7 | 7415 | sched_mc_power_savings = level; |
5c45bf27 | 7416 | |
c70f22d2 | 7417 | arch_reinit_sched_domains(); |
5c45bf27 | 7418 | |
c70f22d2 | 7419 | return count; |
5c45bf27 SS |
7420 | } |
7421 | ||
5c45bf27 | 7422 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7423 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7424 | struct sysdev_class_attribute *attr, |
f718cd4a | 7425 | char *page) |
5c45bf27 SS |
7426 | { |
7427 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7428 | } | |
f718cd4a | 7429 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7430 | struct sysdev_class_attribute *attr, |
48f24c4d | 7431 | const char *buf, size_t count) |
5c45bf27 SS |
7432 | { |
7433 | return sched_power_savings_store(buf, count, 0); | |
7434 | } | |
f718cd4a AK |
7435 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7436 | sched_mc_power_savings_show, | |
7437 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7438 | #endif |
7439 | ||
7440 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7441 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7442 | struct sysdev_class_attribute *attr, |
f718cd4a | 7443 | char *page) |
5c45bf27 SS |
7444 | { |
7445 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7446 | } | |
f718cd4a | 7447 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7448 | struct sysdev_class_attribute *attr, |
48f24c4d | 7449 | const char *buf, size_t count) |
5c45bf27 SS |
7450 | { |
7451 | return sched_power_savings_store(buf, count, 1); | |
7452 | } | |
f718cd4a AK |
7453 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7454 | sched_smt_power_savings_show, | |
6707de00 AB |
7455 | sched_smt_power_savings_store); |
7456 | #endif | |
7457 | ||
39aac648 | 7458 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7459 | { |
7460 | int err = 0; | |
7461 | ||
7462 | #ifdef CONFIG_SCHED_SMT | |
7463 | if (smt_capable()) | |
7464 | err = sysfs_create_file(&cls->kset.kobj, | |
7465 | &attr_sched_smt_power_savings.attr); | |
7466 | #endif | |
7467 | #ifdef CONFIG_SCHED_MC | |
7468 | if (!err && mc_capable()) | |
7469 | err = sysfs_create_file(&cls->kset.kobj, | |
7470 | &attr_sched_mc_power_savings.attr); | |
7471 | #endif | |
7472 | return err; | |
7473 | } | |
6d6bc0ad | 7474 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7475 | |
1da177e4 | 7476 | /* |
3a101d05 TH |
7477 | * Update cpusets according to cpu_active mask. If cpusets are |
7478 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7479 | * around partition_sched_domains(). | |
1da177e4 | 7480 | */ |
0b2e918a TH |
7481 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7482 | void *hcpu) | |
e761b772 | 7483 | { |
3a101d05 | 7484 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7485 | case CPU_ONLINE: |
6ad4c188 | 7486 | case CPU_DOWN_FAILED: |
3a101d05 | 7487 | cpuset_update_active_cpus(); |
e761b772 | 7488 | return NOTIFY_OK; |
3a101d05 TH |
7489 | default: |
7490 | return NOTIFY_DONE; | |
7491 | } | |
7492 | } | |
e761b772 | 7493 | |
0b2e918a TH |
7494 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7495 | void *hcpu) | |
3a101d05 TH |
7496 | { |
7497 | switch (action & ~CPU_TASKS_FROZEN) { | |
7498 | case CPU_DOWN_PREPARE: | |
7499 | cpuset_update_active_cpus(); | |
7500 | return NOTIFY_OK; | |
e761b772 MK |
7501 | default: |
7502 | return NOTIFY_DONE; | |
7503 | } | |
7504 | } | |
e761b772 MK |
7505 | |
7506 | static int update_runtime(struct notifier_block *nfb, | |
7507 | unsigned long action, void *hcpu) | |
1da177e4 | 7508 | { |
7def2be1 PZ |
7509 | int cpu = (int)(long)hcpu; |
7510 | ||
1da177e4 | 7511 | switch (action) { |
1da177e4 | 7512 | case CPU_DOWN_PREPARE: |
8bb78442 | 7513 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7514 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7515 | return NOTIFY_OK; |
7516 | ||
1da177e4 | 7517 | case CPU_DOWN_FAILED: |
8bb78442 | 7518 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7519 | case CPU_ONLINE: |
8bb78442 | 7520 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7521 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7522 | return NOTIFY_OK; |
7523 | ||
1da177e4 LT |
7524 | default: |
7525 | return NOTIFY_DONE; | |
7526 | } | |
1da177e4 | 7527 | } |
1da177e4 LT |
7528 | |
7529 | void __init sched_init_smp(void) | |
7530 | { | |
dcc30a35 RR |
7531 | cpumask_var_t non_isolated_cpus; |
7532 | ||
7533 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7534 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7535 | |
434d53b0 MT |
7536 | #if defined(CONFIG_NUMA) |
7537 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7538 | GFP_KERNEL); | |
7539 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7540 | #endif | |
95402b38 | 7541 | get_online_cpus(); |
712555ee | 7542 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7543 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7544 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7545 | if (cpumask_empty(non_isolated_cpus)) | |
7546 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7547 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7548 | put_online_cpus(); |
e761b772 | 7549 | |
3a101d05 TH |
7550 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7551 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7552 | |
7553 | /* RT runtime code needs to handle some hotplug events */ | |
7554 | hotcpu_notifier(update_runtime, 0); | |
7555 | ||
b328ca18 | 7556 | init_hrtick(); |
5c1e1767 NP |
7557 | |
7558 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7559 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7560 | BUG(); |
19978ca6 | 7561 | sched_init_granularity(); |
dcc30a35 | 7562 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7563 | |
0e3900e6 | 7564 | init_sched_rt_class(); |
1da177e4 LT |
7565 | } |
7566 | #else | |
7567 | void __init sched_init_smp(void) | |
7568 | { | |
19978ca6 | 7569 | sched_init_granularity(); |
1da177e4 LT |
7570 | } |
7571 | #endif /* CONFIG_SMP */ | |
7572 | ||
cd1bb94b AB |
7573 | const_debug unsigned int sysctl_timer_migration = 1; |
7574 | ||
1da177e4 LT |
7575 | int in_sched_functions(unsigned long addr) |
7576 | { | |
1da177e4 LT |
7577 | return in_lock_functions(addr) || |
7578 | (addr >= (unsigned long)__sched_text_start | |
7579 | && addr < (unsigned long)__sched_text_end); | |
7580 | } | |
7581 | ||
a9957449 | 7582 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7583 | { |
7584 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7585 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7586 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7587 | cfs_rq->rq = rq; | |
7588 | #endif | |
67e9fb2a | 7589 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7590 | } |
7591 | ||
fa85ae24 PZ |
7592 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7593 | { | |
7594 | struct rt_prio_array *array; | |
7595 | int i; | |
7596 | ||
7597 | array = &rt_rq->active; | |
7598 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7599 | INIT_LIST_HEAD(array->queue + i); | |
7600 | __clear_bit(i, array->bitmap); | |
7601 | } | |
7602 | /* delimiter for bitsearch: */ | |
7603 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7604 | ||
052f1dc7 | 7605 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7606 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7607 | #ifdef CONFIG_SMP |
e864c499 | 7608 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7609 | #endif |
48d5e258 | 7610 | #endif |
fa85ae24 PZ |
7611 | #ifdef CONFIG_SMP |
7612 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7613 | rt_rq->overloaded = 0; |
05fa785c | 7614 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7615 | #endif |
7616 | ||
7617 | rt_rq->rt_time = 0; | |
7618 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7619 | rt_rq->rt_runtime = 0; |
0986b11b | 7620 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7621 | |
052f1dc7 | 7622 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7623 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7624 | rt_rq->rq = rq; |
7625 | #endif | |
fa85ae24 PZ |
7626 | } |
7627 | ||
6f505b16 | 7628 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 7629 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 7630 | struct sched_entity *se, int cpu, |
ec7dc8ac | 7631 | struct sched_entity *parent) |
6f505b16 | 7632 | { |
ec7dc8ac | 7633 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7634 | tg->cfs_rq[cpu] = cfs_rq; |
7635 | init_cfs_rq(cfs_rq, rq); | |
7636 | cfs_rq->tg = tg; | |
6f505b16 PZ |
7637 | |
7638 | tg->se[cpu] = se; | |
354d60c2 DG |
7639 | /* se could be NULL for init_task_group */ |
7640 | if (!se) | |
7641 | return; | |
7642 | ||
ec7dc8ac DG |
7643 | if (!parent) |
7644 | se->cfs_rq = &rq->cfs; | |
7645 | else | |
7646 | se->cfs_rq = parent->my_q; | |
7647 | ||
6f505b16 | 7648 | se->my_q = cfs_rq; |
2069dd75 | 7649 | update_load_set(&se->load, tg->shares); |
ec7dc8ac | 7650 | se->parent = parent; |
6f505b16 | 7651 | } |
052f1dc7 | 7652 | #endif |
6f505b16 | 7653 | |
052f1dc7 | 7654 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 7655 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 7656 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 7657 | struct sched_rt_entity *parent) |
6f505b16 | 7658 | { |
ec7dc8ac DG |
7659 | struct rq *rq = cpu_rq(cpu); |
7660 | ||
6f505b16 PZ |
7661 | tg->rt_rq[cpu] = rt_rq; |
7662 | init_rt_rq(rt_rq, rq); | |
7663 | rt_rq->tg = tg; | |
ac086bc2 | 7664 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7665 | |
7666 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7667 | if (!rt_se) |
7668 | return; | |
7669 | ||
ec7dc8ac DG |
7670 | if (!parent) |
7671 | rt_se->rt_rq = &rq->rt; | |
7672 | else | |
7673 | rt_se->rt_rq = parent->my_q; | |
7674 | ||
6f505b16 | 7675 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7676 | rt_se->parent = parent; |
6f505b16 PZ |
7677 | INIT_LIST_HEAD(&rt_se->run_list); |
7678 | } | |
7679 | #endif | |
7680 | ||
1da177e4 LT |
7681 | void __init sched_init(void) |
7682 | { | |
dd41f596 | 7683 | int i, j; |
434d53b0 MT |
7684 | unsigned long alloc_size = 0, ptr; |
7685 | ||
7686 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7687 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7688 | #endif | |
7689 | #ifdef CONFIG_RT_GROUP_SCHED | |
7690 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7691 | #endif |
df7c8e84 | 7692 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7693 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7694 | #endif |
434d53b0 | 7695 | if (alloc_size) { |
36b7b6d4 | 7696 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7697 | |
7698 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7699 | init_task_group.se = (struct sched_entity **)ptr; | |
7700 | ptr += nr_cpu_ids * sizeof(void **); | |
7701 | ||
7702 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7703 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7704 | |
6d6bc0ad | 7705 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7706 | #ifdef CONFIG_RT_GROUP_SCHED |
7707 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7708 | ptr += nr_cpu_ids * sizeof(void **); | |
7709 | ||
7710 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7711 | ptr += nr_cpu_ids * sizeof(void **); |
7712 | ||
6d6bc0ad | 7713 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7714 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7715 | for_each_possible_cpu(i) { | |
7716 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7717 | ptr += cpumask_size(); | |
7718 | } | |
7719 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7720 | } |
dd41f596 | 7721 | |
57d885fe GH |
7722 | #ifdef CONFIG_SMP |
7723 | init_defrootdomain(); | |
7724 | #endif | |
7725 | ||
d0b27fa7 PZ |
7726 | init_rt_bandwidth(&def_rt_bandwidth, |
7727 | global_rt_period(), global_rt_runtime()); | |
7728 | ||
7729 | #ifdef CONFIG_RT_GROUP_SCHED | |
7730 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7731 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7732 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7733 | |
7c941438 | 7734 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7735 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7736 | INIT_LIST_HEAD(&init_task_group.children); |
7737 | ||
7c941438 | 7738 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7739 | |
0a945022 | 7740 | for_each_possible_cpu(i) { |
70b97a7f | 7741 | struct rq *rq; |
1da177e4 LT |
7742 | |
7743 | rq = cpu_rq(i); | |
05fa785c | 7744 | raw_spin_lock_init(&rq->lock); |
7897986b | 7745 | rq->nr_running = 0; |
dce48a84 TG |
7746 | rq->calc_load_active = 0; |
7747 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7748 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7749 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7750 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7751 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7752 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7753 | #ifdef CONFIG_CGROUP_SCHED |
7754 | /* | |
7755 | * How much cpu bandwidth does init_task_group get? | |
7756 | * | |
7757 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7758 | * gets 100% of the cpu resources in the system. This overall | |
7759 | * system cpu resource is divided among the tasks of | |
7760 | * init_task_group and its child task-groups in a fair manner, | |
7761 | * based on each entity's (task or task-group's) weight | |
7762 | * (se->load.weight). | |
7763 | * | |
7764 | * In other words, if init_task_group has 10 tasks of weight | |
7765 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7766 | * then A0's share of the cpu resource is: | |
7767 | * | |
0d905bca | 7768 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7769 | * |
7770 | * We achieve this by letting init_task_group's tasks sit | |
7771 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7772 | */ | |
3d4b47b4 | 7773 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, NULL); |
052f1dc7 | 7774 | #endif |
354d60c2 DG |
7775 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7776 | ||
7777 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7778 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7779 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7780 | #ifdef CONFIG_CGROUP_SCHED |
3d4b47b4 | 7781 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, NULL); |
354d60c2 | 7782 | #endif |
dd41f596 | 7783 | #endif |
1da177e4 | 7784 | |
dd41f596 IM |
7785 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7786 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7787 | |
7788 | rq->last_load_update_tick = jiffies; | |
7789 | ||
1da177e4 | 7790 | #ifdef CONFIG_SMP |
41c7ce9a | 7791 | rq->sd = NULL; |
57d885fe | 7792 | rq->rd = NULL; |
e51fd5e2 | 7793 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 7794 | rq->post_schedule = 0; |
1da177e4 | 7795 | rq->active_balance = 0; |
dd41f596 | 7796 | rq->next_balance = jiffies; |
1da177e4 | 7797 | rq->push_cpu = 0; |
0a2966b4 | 7798 | rq->cpu = i; |
1f11eb6a | 7799 | rq->online = 0; |
eae0c9df MG |
7800 | rq->idle_stamp = 0; |
7801 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7802 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
7803 | #ifdef CONFIG_NO_HZ |
7804 | rq->nohz_balance_kick = 0; | |
7805 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
7806 | #endif | |
1da177e4 | 7807 | #endif |
8f4d37ec | 7808 | init_rq_hrtick(rq); |
1da177e4 | 7809 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7810 | } |
7811 | ||
2dd73a4f | 7812 | set_load_weight(&init_task); |
b50f60ce | 7813 | |
e107be36 AK |
7814 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7815 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7816 | #endif | |
7817 | ||
c9819f45 | 7818 | #ifdef CONFIG_SMP |
962cf36c | 7819 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7820 | #endif |
7821 | ||
b50f60ce | 7822 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7823 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7824 | #endif |
7825 | ||
1da177e4 LT |
7826 | /* |
7827 | * The boot idle thread does lazy MMU switching as well: | |
7828 | */ | |
7829 | atomic_inc(&init_mm.mm_count); | |
7830 | enter_lazy_tlb(&init_mm, current); | |
7831 | ||
7832 | /* | |
7833 | * Make us the idle thread. Technically, schedule() should not be | |
7834 | * called from this thread, however somewhere below it might be, | |
7835 | * but because we are the idle thread, we just pick up running again | |
7836 | * when this runqueue becomes "idle". | |
7837 | */ | |
7838 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7839 | |
7840 | calc_load_update = jiffies + LOAD_FREQ; | |
7841 | ||
dd41f596 IM |
7842 | /* |
7843 | * During early bootup we pretend to be a normal task: | |
7844 | */ | |
7845 | current->sched_class = &fair_sched_class; | |
6892b75e | 7846 | |
6a7b3dc3 | 7847 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7848 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7849 | #ifdef CONFIG_SMP |
7d1e6a9b | 7850 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
7851 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
7852 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
7853 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
7854 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
7855 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 7856 | #endif |
bdddd296 RR |
7857 | /* May be allocated at isolcpus cmdline parse time */ |
7858 | if (cpu_isolated_map == NULL) | |
7859 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7860 | #endif /* SMP */ |
6a7b3dc3 | 7861 | |
cdd6c482 | 7862 | perf_event_init(); |
0d905bca | 7863 | |
6892b75e | 7864 | scheduler_running = 1; |
1da177e4 LT |
7865 | } |
7866 | ||
7867 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7868 | static inline int preempt_count_equals(int preempt_offset) |
7869 | { | |
234da7bc | 7870 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7871 | |
7872 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7873 | } | |
7874 | ||
d894837f | 7875 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7876 | { |
48f24c4d | 7877 | #ifdef in_atomic |
1da177e4 LT |
7878 | static unsigned long prev_jiffy; /* ratelimiting */ |
7879 | ||
e4aafea2 FW |
7880 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7881 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7882 | return; |
7883 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7884 | return; | |
7885 | prev_jiffy = jiffies; | |
7886 | ||
3df0fc5b PZ |
7887 | printk(KERN_ERR |
7888 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7889 | file, line); | |
7890 | printk(KERN_ERR | |
7891 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7892 | in_atomic(), irqs_disabled(), | |
7893 | current->pid, current->comm); | |
aef745fc IM |
7894 | |
7895 | debug_show_held_locks(current); | |
7896 | if (irqs_disabled()) | |
7897 | print_irqtrace_events(current); | |
7898 | dump_stack(); | |
1da177e4 LT |
7899 | #endif |
7900 | } | |
7901 | EXPORT_SYMBOL(__might_sleep); | |
7902 | #endif | |
7903 | ||
7904 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7905 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7906 | { | |
7907 | int on_rq; | |
3e51f33f | 7908 | |
3a5e4dc1 AK |
7909 | on_rq = p->se.on_rq; |
7910 | if (on_rq) | |
7911 | deactivate_task(rq, p, 0); | |
7912 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7913 | if (on_rq) { | |
7914 | activate_task(rq, p, 0); | |
7915 | resched_task(rq->curr); | |
7916 | } | |
7917 | } | |
7918 | ||
1da177e4 LT |
7919 | void normalize_rt_tasks(void) |
7920 | { | |
a0f98a1c | 7921 | struct task_struct *g, *p; |
1da177e4 | 7922 | unsigned long flags; |
70b97a7f | 7923 | struct rq *rq; |
1da177e4 | 7924 | |
4cf5d77a | 7925 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7926 | do_each_thread(g, p) { |
178be793 IM |
7927 | /* |
7928 | * Only normalize user tasks: | |
7929 | */ | |
7930 | if (!p->mm) | |
7931 | continue; | |
7932 | ||
6cfb0d5d | 7933 | p->se.exec_start = 0; |
6cfb0d5d | 7934 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7935 | p->se.statistics.wait_start = 0; |
7936 | p->se.statistics.sleep_start = 0; | |
7937 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7938 | #endif |
dd41f596 IM |
7939 | |
7940 | if (!rt_task(p)) { | |
7941 | /* | |
7942 | * Renice negative nice level userspace | |
7943 | * tasks back to 0: | |
7944 | */ | |
7945 | if (TASK_NICE(p) < 0 && p->mm) | |
7946 | set_user_nice(p, 0); | |
1da177e4 | 7947 | continue; |
dd41f596 | 7948 | } |
1da177e4 | 7949 | |
1d615482 | 7950 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7951 | rq = __task_rq_lock(p); |
1da177e4 | 7952 | |
178be793 | 7953 | normalize_task(rq, p); |
3a5e4dc1 | 7954 | |
b29739f9 | 7955 | __task_rq_unlock(rq); |
1d615482 | 7956 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7957 | } while_each_thread(g, p); |
7958 | ||
4cf5d77a | 7959 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7960 | } |
7961 | ||
7962 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7963 | |
67fc4e0c | 7964 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7965 | /* |
67fc4e0c | 7966 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7967 | * |
7968 | * They can only be called when the whole system has been | |
7969 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7970 | * activity can take place. Using them for anything else would | |
7971 | * be a serious bug, and as a result, they aren't even visible | |
7972 | * under any other configuration. | |
7973 | */ | |
7974 | ||
7975 | /** | |
7976 | * curr_task - return the current task for a given cpu. | |
7977 | * @cpu: the processor in question. | |
7978 | * | |
7979 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7980 | */ | |
36c8b586 | 7981 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7982 | { |
7983 | return cpu_curr(cpu); | |
7984 | } | |
7985 | ||
67fc4e0c JW |
7986 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7987 | ||
7988 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7989 | /** |
7990 | * set_curr_task - set the current task for a given cpu. | |
7991 | * @cpu: the processor in question. | |
7992 | * @p: the task pointer to set. | |
7993 | * | |
7994 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7995 | * are serviced on a separate stack. It allows the architecture to switch the |
7996 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7997 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7998 | * and caller must save the original value of the current task (see | |
7999 | * curr_task() above) and restore that value before reenabling interrupts and | |
8000 | * re-starting the system. | |
8001 | * | |
8002 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8003 | */ | |
36c8b586 | 8004 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8005 | { |
8006 | cpu_curr(cpu) = p; | |
8007 | } | |
8008 | ||
8009 | #endif | |
29f59db3 | 8010 | |
bccbe08a PZ |
8011 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8012 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8013 | { |
8014 | int i; | |
8015 | ||
8016 | for_each_possible_cpu(i) { | |
8017 | if (tg->cfs_rq) | |
8018 | kfree(tg->cfs_rq[i]); | |
8019 | if (tg->se) | |
8020 | kfree(tg->se[i]); | |
6f505b16 PZ |
8021 | } |
8022 | ||
8023 | kfree(tg->cfs_rq); | |
8024 | kfree(tg->se); | |
6f505b16 PZ |
8025 | } |
8026 | ||
ec7dc8ac DG |
8027 | static |
8028 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8029 | { |
29f59db3 | 8030 | struct cfs_rq *cfs_rq; |
eab17229 | 8031 | struct sched_entity *se; |
9b5b7751 | 8032 | struct rq *rq; |
29f59db3 SV |
8033 | int i; |
8034 | ||
434d53b0 | 8035 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8036 | if (!tg->cfs_rq) |
8037 | goto err; | |
434d53b0 | 8038 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8039 | if (!tg->se) |
8040 | goto err; | |
052f1dc7 PZ |
8041 | |
8042 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8043 | |
8044 | for_each_possible_cpu(i) { | |
9b5b7751 | 8045 | rq = cpu_rq(i); |
29f59db3 | 8046 | |
eab17229 LZ |
8047 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8048 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8049 | if (!cfs_rq) |
8050 | goto err; | |
8051 | ||
eab17229 LZ |
8052 | se = kzalloc_node(sizeof(struct sched_entity), |
8053 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8054 | if (!se) |
dfc12eb2 | 8055 | goto err_free_rq; |
29f59db3 | 8056 | |
3d4b47b4 | 8057 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8058 | } |
8059 | ||
8060 | return 1; | |
8061 | ||
49246274 | 8062 | err_free_rq: |
dfc12eb2 | 8063 | kfree(cfs_rq); |
49246274 | 8064 | err: |
bccbe08a PZ |
8065 | return 0; |
8066 | } | |
8067 | ||
bccbe08a PZ |
8068 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8069 | { | |
3d4b47b4 PZ |
8070 | struct rq *rq = cpu_rq(cpu); |
8071 | unsigned long flags; | |
8072 | int i; | |
8073 | ||
8074 | /* | |
8075 | * Only empty task groups can be destroyed; so we can speculatively | |
8076 | * check on_list without danger of it being re-added. | |
8077 | */ | |
8078 | if (!tg->cfs_rq[cpu]->on_list) | |
8079 | return; | |
8080 | ||
8081 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8082 | list_del_leaf_cfs_rq(tg->cfs_rq[i]); | |
8083 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
bccbe08a | 8084 | } |
6d6bc0ad | 8085 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8086 | static inline void free_fair_sched_group(struct task_group *tg) |
8087 | { | |
8088 | } | |
8089 | ||
ec7dc8ac DG |
8090 | static inline |
8091 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8092 | { |
8093 | return 1; | |
8094 | } | |
8095 | ||
bccbe08a PZ |
8096 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8097 | { | |
8098 | } | |
6d6bc0ad | 8099 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8100 | |
8101 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8102 | static void free_rt_sched_group(struct task_group *tg) |
8103 | { | |
8104 | int i; | |
8105 | ||
d0b27fa7 PZ |
8106 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8107 | ||
bccbe08a PZ |
8108 | for_each_possible_cpu(i) { |
8109 | if (tg->rt_rq) | |
8110 | kfree(tg->rt_rq[i]); | |
8111 | if (tg->rt_se) | |
8112 | kfree(tg->rt_se[i]); | |
8113 | } | |
8114 | ||
8115 | kfree(tg->rt_rq); | |
8116 | kfree(tg->rt_se); | |
8117 | } | |
8118 | ||
ec7dc8ac DG |
8119 | static |
8120 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8121 | { |
8122 | struct rt_rq *rt_rq; | |
eab17229 | 8123 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8124 | struct rq *rq; |
8125 | int i; | |
8126 | ||
434d53b0 | 8127 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8128 | if (!tg->rt_rq) |
8129 | goto err; | |
434d53b0 | 8130 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8131 | if (!tg->rt_se) |
8132 | goto err; | |
8133 | ||
d0b27fa7 PZ |
8134 | init_rt_bandwidth(&tg->rt_bandwidth, |
8135 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8136 | |
8137 | for_each_possible_cpu(i) { | |
8138 | rq = cpu_rq(i); | |
8139 | ||
eab17229 LZ |
8140 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8141 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8142 | if (!rt_rq) |
8143 | goto err; | |
29f59db3 | 8144 | |
eab17229 LZ |
8145 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8146 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8147 | if (!rt_se) |
dfc12eb2 | 8148 | goto err_free_rq; |
29f59db3 | 8149 | |
3d4b47b4 | 8150 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8151 | } |
8152 | ||
bccbe08a PZ |
8153 | return 1; |
8154 | ||
49246274 | 8155 | err_free_rq: |
dfc12eb2 | 8156 | kfree(rt_rq); |
49246274 | 8157 | err: |
bccbe08a PZ |
8158 | return 0; |
8159 | } | |
6d6bc0ad | 8160 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8161 | static inline void free_rt_sched_group(struct task_group *tg) |
8162 | { | |
8163 | } | |
8164 | ||
ec7dc8ac DG |
8165 | static inline |
8166 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8167 | { |
8168 | return 1; | |
8169 | } | |
6d6bc0ad | 8170 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8171 | |
7c941438 | 8172 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8173 | static void free_sched_group(struct task_group *tg) |
8174 | { | |
8175 | free_fair_sched_group(tg); | |
8176 | free_rt_sched_group(tg); | |
8177 | kfree(tg); | |
8178 | } | |
8179 | ||
8180 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8181 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8182 | { |
8183 | struct task_group *tg; | |
8184 | unsigned long flags; | |
bccbe08a PZ |
8185 | |
8186 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8187 | if (!tg) | |
8188 | return ERR_PTR(-ENOMEM); | |
8189 | ||
ec7dc8ac | 8190 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8191 | goto err; |
8192 | ||
ec7dc8ac | 8193 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8194 | goto err; |
8195 | ||
8ed36996 | 8196 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8197 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8198 | |
8199 | WARN_ON(!parent); /* root should already exist */ | |
8200 | ||
8201 | tg->parent = parent; | |
f473aa5e | 8202 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8203 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8204 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8205 | |
9b5b7751 | 8206 | return tg; |
29f59db3 SV |
8207 | |
8208 | err: | |
6f505b16 | 8209 | free_sched_group(tg); |
29f59db3 SV |
8210 | return ERR_PTR(-ENOMEM); |
8211 | } | |
8212 | ||
9b5b7751 | 8213 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8214 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8215 | { |
29f59db3 | 8216 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8217 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8218 | } |
8219 | ||
9b5b7751 | 8220 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8221 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8222 | { |
8ed36996 | 8223 | unsigned long flags; |
9b5b7751 | 8224 | int i; |
29f59db3 | 8225 | |
3d4b47b4 PZ |
8226 | /* end participation in shares distribution */ |
8227 | for_each_possible_cpu(i) | |
bccbe08a | 8228 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8229 | |
8230 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8231 | list_del_rcu(&tg->list); |
f473aa5e | 8232 | list_del_rcu(&tg->siblings); |
8ed36996 | 8233 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8234 | |
9b5b7751 | 8235 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8236 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8237 | } |
8238 | ||
9b5b7751 | 8239 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8240 | * The caller of this function should have put the task in its new group |
8241 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8242 | * reflect its new group. | |
9b5b7751 SV |
8243 | */ |
8244 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8245 | { |
8246 | int on_rq, running; | |
8247 | unsigned long flags; | |
8248 | struct rq *rq; | |
8249 | ||
8250 | rq = task_rq_lock(tsk, &flags); | |
8251 | ||
051a1d1a | 8252 | running = task_current(rq, tsk); |
29f59db3 SV |
8253 | on_rq = tsk->se.on_rq; |
8254 | ||
0e1f3483 | 8255 | if (on_rq) |
29f59db3 | 8256 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8257 | if (unlikely(running)) |
8258 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8259 | |
810b3817 | 8260 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8261 | if (tsk->sched_class->task_move_group) |
8262 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8263 | else | |
810b3817 | 8264 | #endif |
b2b5ce02 | 8265 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8266 | |
0e1f3483 HS |
8267 | if (unlikely(running)) |
8268 | tsk->sched_class->set_curr_task(rq); | |
8269 | if (on_rq) | |
371fd7e7 | 8270 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8271 | |
29f59db3 SV |
8272 | task_rq_unlock(rq, &flags); |
8273 | } | |
7c941438 | 8274 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8275 | |
052f1dc7 | 8276 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8277 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8278 | { |
8279 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8280 | int on_rq; |
8281 | ||
29f59db3 | 8282 | on_rq = se->on_rq; |
62fb1851 | 8283 | if (on_rq) |
29f59db3 SV |
8284 | dequeue_entity(cfs_rq, se, 0); |
8285 | ||
2069dd75 | 8286 | update_load_set(&se->load, shares); |
29f59db3 | 8287 | |
62fb1851 | 8288 | if (on_rq) |
29f59db3 | 8289 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8290 | } |
62fb1851 | 8291 | |
c09595f6 PZ |
8292 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8293 | { | |
8294 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8295 | struct rq *rq = cfs_rq->rq; | |
8296 | unsigned long flags; | |
8297 | ||
05fa785c | 8298 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8299 | __set_se_shares(se, shares); |
05fa785c | 8300 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8301 | } |
8302 | ||
8ed36996 PZ |
8303 | static DEFINE_MUTEX(shares_mutex); |
8304 | ||
4cf86d77 | 8305 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8306 | { |
8307 | int i; | |
c61935fd | 8308 | |
ec7dc8ac DG |
8309 | /* |
8310 | * We can't change the weight of the root cgroup. | |
8311 | */ | |
8312 | if (!tg->se[0]) | |
8313 | return -EINVAL; | |
8314 | ||
18d95a28 PZ |
8315 | if (shares < MIN_SHARES) |
8316 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8317 | else if (shares > MAX_SHARES) |
8318 | shares = MAX_SHARES; | |
62fb1851 | 8319 | |
8ed36996 | 8320 | mutex_lock(&shares_mutex); |
9b5b7751 | 8321 | if (tg->shares == shares) |
5cb350ba | 8322 | goto done; |
29f59db3 | 8323 | |
9b5b7751 | 8324 | tg->shares = shares; |
c09595f6 PZ |
8325 | for_each_possible_cpu(i) { |
8326 | /* | |
8327 | * force a rebalance | |
8328 | */ | |
cb4ad1ff | 8329 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8330 | } |
29f59db3 | 8331 | |
5cb350ba | 8332 | done: |
8ed36996 | 8333 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8334 | return 0; |
29f59db3 SV |
8335 | } |
8336 | ||
5cb350ba DG |
8337 | unsigned long sched_group_shares(struct task_group *tg) |
8338 | { | |
8339 | return tg->shares; | |
8340 | } | |
052f1dc7 | 8341 | #endif |
5cb350ba | 8342 | |
052f1dc7 | 8343 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8344 | /* |
9f0c1e56 | 8345 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8346 | */ |
9f0c1e56 PZ |
8347 | static DEFINE_MUTEX(rt_constraints_mutex); |
8348 | ||
8349 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8350 | { | |
8351 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8352 | return 1ULL << 20; |
9f0c1e56 | 8353 | |
9a7e0b18 | 8354 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8355 | } |
8356 | ||
9a7e0b18 PZ |
8357 | /* Must be called with tasklist_lock held */ |
8358 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8359 | { |
9a7e0b18 | 8360 | struct task_struct *g, *p; |
b40b2e8e | 8361 | |
9a7e0b18 PZ |
8362 | do_each_thread(g, p) { |
8363 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8364 | return 1; | |
8365 | } while_each_thread(g, p); | |
b40b2e8e | 8366 | |
9a7e0b18 PZ |
8367 | return 0; |
8368 | } | |
b40b2e8e | 8369 | |
9a7e0b18 PZ |
8370 | struct rt_schedulable_data { |
8371 | struct task_group *tg; | |
8372 | u64 rt_period; | |
8373 | u64 rt_runtime; | |
8374 | }; | |
b40b2e8e | 8375 | |
9a7e0b18 PZ |
8376 | static int tg_schedulable(struct task_group *tg, void *data) |
8377 | { | |
8378 | struct rt_schedulable_data *d = data; | |
8379 | struct task_group *child; | |
8380 | unsigned long total, sum = 0; | |
8381 | u64 period, runtime; | |
b40b2e8e | 8382 | |
9a7e0b18 PZ |
8383 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8384 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8385 | |
9a7e0b18 PZ |
8386 | if (tg == d->tg) { |
8387 | period = d->rt_period; | |
8388 | runtime = d->rt_runtime; | |
b40b2e8e | 8389 | } |
b40b2e8e | 8390 | |
4653f803 PZ |
8391 | /* |
8392 | * Cannot have more runtime than the period. | |
8393 | */ | |
8394 | if (runtime > period && runtime != RUNTIME_INF) | |
8395 | return -EINVAL; | |
6f505b16 | 8396 | |
4653f803 PZ |
8397 | /* |
8398 | * Ensure we don't starve existing RT tasks. | |
8399 | */ | |
9a7e0b18 PZ |
8400 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8401 | return -EBUSY; | |
6f505b16 | 8402 | |
9a7e0b18 | 8403 | total = to_ratio(period, runtime); |
6f505b16 | 8404 | |
4653f803 PZ |
8405 | /* |
8406 | * Nobody can have more than the global setting allows. | |
8407 | */ | |
8408 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8409 | return -EINVAL; | |
6f505b16 | 8410 | |
4653f803 PZ |
8411 | /* |
8412 | * The sum of our children's runtime should not exceed our own. | |
8413 | */ | |
9a7e0b18 PZ |
8414 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8415 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8416 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8417 | |
9a7e0b18 PZ |
8418 | if (child == d->tg) { |
8419 | period = d->rt_period; | |
8420 | runtime = d->rt_runtime; | |
8421 | } | |
6f505b16 | 8422 | |
9a7e0b18 | 8423 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8424 | } |
6f505b16 | 8425 | |
9a7e0b18 PZ |
8426 | if (sum > total) |
8427 | return -EINVAL; | |
8428 | ||
8429 | return 0; | |
6f505b16 PZ |
8430 | } |
8431 | ||
9a7e0b18 | 8432 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8433 | { |
9a7e0b18 PZ |
8434 | struct rt_schedulable_data data = { |
8435 | .tg = tg, | |
8436 | .rt_period = period, | |
8437 | .rt_runtime = runtime, | |
8438 | }; | |
8439 | ||
8440 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8441 | } |
8442 | ||
d0b27fa7 PZ |
8443 | static int tg_set_bandwidth(struct task_group *tg, |
8444 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8445 | { |
ac086bc2 | 8446 | int i, err = 0; |
9f0c1e56 | 8447 | |
9f0c1e56 | 8448 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8449 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8450 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8451 | if (err) | |
9f0c1e56 | 8452 | goto unlock; |
ac086bc2 | 8453 | |
0986b11b | 8454 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8455 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8456 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8457 | |
8458 | for_each_possible_cpu(i) { | |
8459 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8460 | ||
0986b11b | 8461 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8462 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8463 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8464 | } |
0986b11b | 8465 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8466 | unlock: |
521f1a24 | 8467 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8468 | mutex_unlock(&rt_constraints_mutex); |
8469 | ||
8470 | return err; | |
6f505b16 PZ |
8471 | } |
8472 | ||
d0b27fa7 PZ |
8473 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8474 | { | |
8475 | u64 rt_runtime, rt_period; | |
8476 | ||
8477 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8478 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8479 | if (rt_runtime_us < 0) | |
8480 | rt_runtime = RUNTIME_INF; | |
8481 | ||
8482 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8483 | } | |
8484 | ||
9f0c1e56 PZ |
8485 | long sched_group_rt_runtime(struct task_group *tg) |
8486 | { | |
8487 | u64 rt_runtime_us; | |
8488 | ||
d0b27fa7 | 8489 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8490 | return -1; |
8491 | ||
d0b27fa7 | 8492 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8493 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8494 | return rt_runtime_us; | |
8495 | } | |
d0b27fa7 PZ |
8496 | |
8497 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8498 | { | |
8499 | u64 rt_runtime, rt_period; | |
8500 | ||
8501 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8502 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8503 | ||
619b0488 R |
8504 | if (rt_period == 0) |
8505 | return -EINVAL; | |
8506 | ||
d0b27fa7 PZ |
8507 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8508 | } | |
8509 | ||
8510 | long sched_group_rt_period(struct task_group *tg) | |
8511 | { | |
8512 | u64 rt_period_us; | |
8513 | ||
8514 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8515 | do_div(rt_period_us, NSEC_PER_USEC); | |
8516 | return rt_period_us; | |
8517 | } | |
8518 | ||
8519 | static int sched_rt_global_constraints(void) | |
8520 | { | |
4653f803 | 8521 | u64 runtime, period; |
d0b27fa7 PZ |
8522 | int ret = 0; |
8523 | ||
ec5d4989 HS |
8524 | if (sysctl_sched_rt_period <= 0) |
8525 | return -EINVAL; | |
8526 | ||
4653f803 PZ |
8527 | runtime = global_rt_runtime(); |
8528 | period = global_rt_period(); | |
8529 | ||
8530 | /* | |
8531 | * Sanity check on the sysctl variables. | |
8532 | */ | |
8533 | if (runtime > period && runtime != RUNTIME_INF) | |
8534 | return -EINVAL; | |
10b612f4 | 8535 | |
d0b27fa7 | 8536 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8537 | read_lock(&tasklist_lock); |
4653f803 | 8538 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8539 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8540 | mutex_unlock(&rt_constraints_mutex); |
8541 | ||
8542 | return ret; | |
8543 | } | |
54e99124 DG |
8544 | |
8545 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8546 | { | |
8547 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8548 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8549 | return 0; | |
8550 | ||
8551 | return 1; | |
8552 | } | |
8553 | ||
6d6bc0ad | 8554 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8555 | static int sched_rt_global_constraints(void) |
8556 | { | |
ac086bc2 PZ |
8557 | unsigned long flags; |
8558 | int i; | |
8559 | ||
ec5d4989 HS |
8560 | if (sysctl_sched_rt_period <= 0) |
8561 | return -EINVAL; | |
8562 | ||
60aa605d PZ |
8563 | /* |
8564 | * There's always some RT tasks in the root group | |
8565 | * -- migration, kstopmachine etc.. | |
8566 | */ | |
8567 | if (sysctl_sched_rt_runtime == 0) | |
8568 | return -EBUSY; | |
8569 | ||
0986b11b | 8570 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8571 | for_each_possible_cpu(i) { |
8572 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8573 | ||
0986b11b | 8574 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8575 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8576 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8577 | } |
0986b11b | 8578 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8579 | |
d0b27fa7 PZ |
8580 | return 0; |
8581 | } | |
6d6bc0ad | 8582 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8583 | |
8584 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8585 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8586 | loff_t *ppos) |
8587 | { | |
8588 | int ret; | |
8589 | int old_period, old_runtime; | |
8590 | static DEFINE_MUTEX(mutex); | |
8591 | ||
8592 | mutex_lock(&mutex); | |
8593 | old_period = sysctl_sched_rt_period; | |
8594 | old_runtime = sysctl_sched_rt_runtime; | |
8595 | ||
8d65af78 | 8596 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8597 | |
8598 | if (!ret && write) { | |
8599 | ret = sched_rt_global_constraints(); | |
8600 | if (ret) { | |
8601 | sysctl_sched_rt_period = old_period; | |
8602 | sysctl_sched_rt_runtime = old_runtime; | |
8603 | } else { | |
8604 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8605 | def_rt_bandwidth.rt_period = | |
8606 | ns_to_ktime(global_rt_period()); | |
8607 | } | |
8608 | } | |
8609 | mutex_unlock(&mutex); | |
8610 | ||
8611 | return ret; | |
8612 | } | |
68318b8e | 8613 | |
052f1dc7 | 8614 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8615 | |
8616 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8617 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8618 | { |
2b01dfe3 PM |
8619 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8620 | struct task_group, css); | |
68318b8e SV |
8621 | } |
8622 | ||
8623 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8624 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8625 | { |
ec7dc8ac | 8626 | struct task_group *tg, *parent; |
68318b8e | 8627 | |
2b01dfe3 | 8628 | if (!cgrp->parent) { |
68318b8e | 8629 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8630 | return &init_task_group.css; |
8631 | } | |
8632 | ||
ec7dc8ac DG |
8633 | parent = cgroup_tg(cgrp->parent); |
8634 | tg = sched_create_group(parent); | |
68318b8e SV |
8635 | if (IS_ERR(tg)) |
8636 | return ERR_PTR(-ENOMEM); | |
8637 | ||
68318b8e SV |
8638 | return &tg->css; |
8639 | } | |
8640 | ||
41a2d6cf IM |
8641 | static void |
8642 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8643 | { |
2b01dfe3 | 8644 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8645 | |
8646 | sched_destroy_group(tg); | |
8647 | } | |
8648 | ||
41a2d6cf | 8649 | static int |
be367d09 | 8650 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8651 | { |
b68aa230 | 8652 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8653 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8654 | return -EINVAL; |
8655 | #else | |
68318b8e SV |
8656 | /* We don't support RT-tasks being in separate groups */ |
8657 | if (tsk->sched_class != &fair_sched_class) | |
8658 | return -EINVAL; | |
b68aa230 | 8659 | #endif |
be367d09 BB |
8660 | return 0; |
8661 | } | |
68318b8e | 8662 | |
be367d09 BB |
8663 | static int |
8664 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8665 | struct task_struct *tsk, bool threadgroup) | |
8666 | { | |
8667 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8668 | if (retval) | |
8669 | return retval; | |
8670 | if (threadgroup) { | |
8671 | struct task_struct *c; | |
8672 | rcu_read_lock(); | |
8673 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8674 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8675 | if (retval) { | |
8676 | rcu_read_unlock(); | |
8677 | return retval; | |
8678 | } | |
8679 | } | |
8680 | rcu_read_unlock(); | |
8681 | } | |
68318b8e SV |
8682 | return 0; |
8683 | } | |
8684 | ||
8685 | static void | |
2b01dfe3 | 8686 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8687 | struct cgroup *old_cont, struct task_struct *tsk, |
8688 | bool threadgroup) | |
68318b8e SV |
8689 | { |
8690 | sched_move_task(tsk); | |
be367d09 BB |
8691 | if (threadgroup) { |
8692 | struct task_struct *c; | |
8693 | rcu_read_lock(); | |
8694 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8695 | sched_move_task(c); | |
8696 | } | |
8697 | rcu_read_unlock(); | |
8698 | } | |
68318b8e SV |
8699 | } |
8700 | ||
052f1dc7 | 8701 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8702 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8703 | u64 shareval) |
68318b8e | 8704 | { |
2b01dfe3 | 8705 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8706 | } |
8707 | ||
f4c753b7 | 8708 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8709 | { |
2b01dfe3 | 8710 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8711 | |
8712 | return (u64) tg->shares; | |
8713 | } | |
6d6bc0ad | 8714 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8715 | |
052f1dc7 | 8716 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8717 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8718 | s64 val) |
6f505b16 | 8719 | { |
06ecb27c | 8720 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8721 | } |
8722 | ||
06ecb27c | 8723 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8724 | { |
06ecb27c | 8725 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8726 | } |
d0b27fa7 PZ |
8727 | |
8728 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8729 | u64 rt_period_us) | |
8730 | { | |
8731 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8732 | } | |
8733 | ||
8734 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8735 | { | |
8736 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8737 | } | |
6d6bc0ad | 8738 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8739 | |
fe5c7cc2 | 8740 | static struct cftype cpu_files[] = { |
052f1dc7 | 8741 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8742 | { |
8743 | .name = "shares", | |
f4c753b7 PM |
8744 | .read_u64 = cpu_shares_read_u64, |
8745 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8746 | }, |
052f1dc7 PZ |
8747 | #endif |
8748 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8749 | { |
9f0c1e56 | 8750 | .name = "rt_runtime_us", |
06ecb27c PM |
8751 | .read_s64 = cpu_rt_runtime_read, |
8752 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8753 | }, |
d0b27fa7 PZ |
8754 | { |
8755 | .name = "rt_period_us", | |
f4c753b7 PM |
8756 | .read_u64 = cpu_rt_period_read_uint, |
8757 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8758 | }, |
052f1dc7 | 8759 | #endif |
68318b8e SV |
8760 | }; |
8761 | ||
8762 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8763 | { | |
fe5c7cc2 | 8764 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8765 | } |
8766 | ||
8767 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8768 | .name = "cpu", |
8769 | .create = cpu_cgroup_create, | |
8770 | .destroy = cpu_cgroup_destroy, | |
8771 | .can_attach = cpu_cgroup_can_attach, | |
8772 | .attach = cpu_cgroup_attach, | |
8773 | .populate = cpu_cgroup_populate, | |
8774 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8775 | .early_init = 1, |
8776 | }; | |
8777 | ||
052f1dc7 | 8778 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8779 | |
8780 | #ifdef CONFIG_CGROUP_CPUACCT | |
8781 | ||
8782 | /* | |
8783 | * CPU accounting code for task groups. | |
8784 | * | |
8785 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8786 | * (balbir@in.ibm.com). | |
8787 | */ | |
8788 | ||
934352f2 | 8789 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8790 | struct cpuacct { |
8791 | struct cgroup_subsys_state css; | |
8792 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8793 | u64 __percpu *cpuusage; |
ef12fefa | 8794 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8795 | struct cpuacct *parent; |
d842de87 SV |
8796 | }; |
8797 | ||
8798 | struct cgroup_subsys cpuacct_subsys; | |
8799 | ||
8800 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8801 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8802 | { |
32cd756a | 8803 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8804 | struct cpuacct, css); |
8805 | } | |
8806 | ||
8807 | /* return cpu accounting group to which this task belongs */ | |
8808 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8809 | { | |
8810 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8811 | struct cpuacct, css); | |
8812 | } | |
8813 | ||
8814 | /* create a new cpu accounting group */ | |
8815 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8816 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8817 | { |
8818 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8819 | int i; |
d842de87 SV |
8820 | |
8821 | if (!ca) | |
ef12fefa | 8822 | goto out; |
d842de87 SV |
8823 | |
8824 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8825 | if (!ca->cpuusage) |
8826 | goto out_free_ca; | |
8827 | ||
8828 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8829 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8830 | goto out_free_counters; | |
d842de87 | 8831 | |
934352f2 BR |
8832 | if (cgrp->parent) |
8833 | ca->parent = cgroup_ca(cgrp->parent); | |
8834 | ||
d842de87 | 8835 | return &ca->css; |
ef12fefa BR |
8836 | |
8837 | out_free_counters: | |
8838 | while (--i >= 0) | |
8839 | percpu_counter_destroy(&ca->cpustat[i]); | |
8840 | free_percpu(ca->cpuusage); | |
8841 | out_free_ca: | |
8842 | kfree(ca); | |
8843 | out: | |
8844 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8845 | } |
8846 | ||
8847 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8848 | static void |
32cd756a | 8849 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8850 | { |
32cd756a | 8851 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8852 | int i; |
d842de87 | 8853 | |
ef12fefa BR |
8854 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8855 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8856 | free_percpu(ca->cpuusage); |
8857 | kfree(ca); | |
8858 | } | |
8859 | ||
720f5498 KC |
8860 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8861 | { | |
b36128c8 | 8862 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8863 | u64 data; |
8864 | ||
8865 | #ifndef CONFIG_64BIT | |
8866 | /* | |
8867 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8868 | */ | |
05fa785c | 8869 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8870 | data = *cpuusage; |
05fa785c | 8871 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8872 | #else |
8873 | data = *cpuusage; | |
8874 | #endif | |
8875 | ||
8876 | return data; | |
8877 | } | |
8878 | ||
8879 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8880 | { | |
b36128c8 | 8881 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8882 | |
8883 | #ifndef CONFIG_64BIT | |
8884 | /* | |
8885 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8886 | */ | |
05fa785c | 8887 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8888 | *cpuusage = val; |
05fa785c | 8889 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8890 | #else |
8891 | *cpuusage = val; | |
8892 | #endif | |
8893 | } | |
8894 | ||
d842de87 | 8895 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8896 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8897 | { |
32cd756a | 8898 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8899 | u64 totalcpuusage = 0; |
8900 | int i; | |
8901 | ||
720f5498 KC |
8902 | for_each_present_cpu(i) |
8903 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8904 | |
8905 | return totalcpuusage; | |
8906 | } | |
8907 | ||
0297b803 DG |
8908 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8909 | u64 reset) | |
8910 | { | |
8911 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8912 | int err = 0; | |
8913 | int i; | |
8914 | ||
8915 | if (reset) { | |
8916 | err = -EINVAL; | |
8917 | goto out; | |
8918 | } | |
8919 | ||
720f5498 KC |
8920 | for_each_present_cpu(i) |
8921 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8922 | |
0297b803 DG |
8923 | out: |
8924 | return err; | |
8925 | } | |
8926 | ||
e9515c3c KC |
8927 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8928 | struct seq_file *m) | |
8929 | { | |
8930 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8931 | u64 percpu; | |
8932 | int i; | |
8933 | ||
8934 | for_each_present_cpu(i) { | |
8935 | percpu = cpuacct_cpuusage_read(ca, i); | |
8936 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8937 | } | |
8938 | seq_printf(m, "\n"); | |
8939 | return 0; | |
8940 | } | |
8941 | ||
ef12fefa BR |
8942 | static const char *cpuacct_stat_desc[] = { |
8943 | [CPUACCT_STAT_USER] = "user", | |
8944 | [CPUACCT_STAT_SYSTEM] = "system", | |
8945 | }; | |
8946 | ||
8947 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8948 | struct cgroup_map_cb *cb) | |
8949 | { | |
8950 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8951 | int i; | |
8952 | ||
8953 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8954 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8955 | val = cputime64_to_clock_t(val); | |
8956 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8957 | } | |
8958 | return 0; | |
8959 | } | |
8960 | ||
d842de87 SV |
8961 | static struct cftype files[] = { |
8962 | { | |
8963 | .name = "usage", | |
f4c753b7 PM |
8964 | .read_u64 = cpuusage_read, |
8965 | .write_u64 = cpuusage_write, | |
d842de87 | 8966 | }, |
e9515c3c KC |
8967 | { |
8968 | .name = "usage_percpu", | |
8969 | .read_seq_string = cpuacct_percpu_seq_read, | |
8970 | }, | |
ef12fefa BR |
8971 | { |
8972 | .name = "stat", | |
8973 | .read_map = cpuacct_stats_show, | |
8974 | }, | |
d842de87 SV |
8975 | }; |
8976 | ||
32cd756a | 8977 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8978 | { |
32cd756a | 8979 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8980 | } |
8981 | ||
8982 | /* | |
8983 | * charge this task's execution time to its accounting group. | |
8984 | * | |
8985 | * called with rq->lock held. | |
8986 | */ | |
8987 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8988 | { | |
8989 | struct cpuacct *ca; | |
934352f2 | 8990 | int cpu; |
d842de87 | 8991 | |
c40c6f85 | 8992 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8993 | return; |
8994 | ||
934352f2 | 8995 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8996 | |
8997 | rcu_read_lock(); | |
8998 | ||
d842de87 | 8999 | ca = task_ca(tsk); |
d842de87 | 9000 | |
934352f2 | 9001 | for (; ca; ca = ca->parent) { |
b36128c8 | 9002 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9003 | *cpuusage += cputime; |
9004 | } | |
a18b83b7 BR |
9005 | |
9006 | rcu_read_unlock(); | |
d842de87 SV |
9007 | } |
9008 | ||
fa535a77 AB |
9009 | /* |
9010 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9011 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9012 | * percpu_counter_add with values large enough to always overflow the | |
9013 | * per cpu batch limit causing bad SMP scalability. | |
9014 | * | |
9015 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9016 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9017 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9018 | */ | |
9019 | #ifdef CONFIG_SMP | |
9020 | #define CPUACCT_BATCH \ | |
9021 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9022 | #else | |
9023 | #define CPUACCT_BATCH 0 | |
9024 | #endif | |
9025 | ||
ef12fefa BR |
9026 | /* |
9027 | * Charge the system/user time to the task's accounting group. | |
9028 | */ | |
9029 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9030 | enum cpuacct_stat_index idx, cputime_t val) | |
9031 | { | |
9032 | struct cpuacct *ca; | |
fa535a77 | 9033 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9034 | |
9035 | if (unlikely(!cpuacct_subsys.active)) | |
9036 | return; | |
9037 | ||
9038 | rcu_read_lock(); | |
9039 | ca = task_ca(tsk); | |
9040 | ||
9041 | do { | |
fa535a77 | 9042 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9043 | ca = ca->parent; |
9044 | } while (ca); | |
9045 | rcu_read_unlock(); | |
9046 | } | |
9047 | ||
d842de87 SV |
9048 | struct cgroup_subsys cpuacct_subsys = { |
9049 | .name = "cpuacct", | |
9050 | .create = cpuacct_create, | |
9051 | .destroy = cpuacct_destroy, | |
9052 | .populate = cpuacct_populate, | |
9053 | .subsys_id = cpuacct_subsys_id, | |
9054 | }; | |
9055 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
9056 | |
9057 | #ifndef CONFIG_SMP | |
9058 | ||
03b042bf PM |
9059 | void synchronize_sched_expedited(void) |
9060 | { | |
fc390cde | 9061 | barrier(); |
03b042bf PM |
9062 | } |
9063 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9064 | ||
9065 | #else /* #ifndef CONFIG_SMP */ | |
9066 | ||
cc631fb7 | 9067 | static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0); |
03b042bf | 9068 | |
cc631fb7 | 9069 | static int synchronize_sched_expedited_cpu_stop(void *data) |
03b042bf | 9070 | { |
969c7921 TH |
9071 | /* |
9072 | * There must be a full memory barrier on each affected CPU | |
9073 | * between the time that try_stop_cpus() is called and the | |
9074 | * time that it returns. | |
9075 | * | |
9076 | * In the current initial implementation of cpu_stop, the | |
9077 | * above condition is already met when the control reaches | |
9078 | * this point and the following smp_mb() is not strictly | |
9079 | * necessary. Do smp_mb() anyway for documentation and | |
9080 | * robustness against future implementation changes. | |
9081 | */ | |
cc631fb7 | 9082 | smp_mb(); /* See above comment block. */ |
969c7921 | 9083 | return 0; |
03b042bf | 9084 | } |
03b042bf PM |
9085 | |
9086 | /* | |
9087 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9088 | * approach to force grace period to end quickly. This consumes | |
9089 | * significant time on all CPUs, and is thus not recommended for | |
9090 | * any sort of common-case code. | |
9091 | * | |
9092 | * Note that it is illegal to call this function while holding any | |
9093 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9094 | * observe this restriction will result in deadlock. | |
9095 | */ | |
9096 | void synchronize_sched_expedited(void) | |
9097 | { | |
969c7921 | 9098 | int snap, trycount = 0; |
03b042bf PM |
9099 | |
9100 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
969c7921 | 9101 | snap = atomic_read(&synchronize_sched_expedited_count) + 1; |
03b042bf | 9102 | get_online_cpus(); |
969c7921 TH |
9103 | while (try_stop_cpus(cpu_online_mask, |
9104 | synchronize_sched_expedited_cpu_stop, | |
94458d5e | 9105 | NULL) == -EAGAIN) { |
03b042bf PM |
9106 | put_online_cpus(); |
9107 | if (trycount++ < 10) | |
9108 | udelay(trycount * num_online_cpus()); | |
9109 | else { | |
9110 | synchronize_sched(); | |
9111 | return; | |
9112 | } | |
969c7921 | 9113 | if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) { |
03b042bf PM |
9114 | smp_mb(); /* ensure test happens before caller kfree */ |
9115 | return; | |
9116 | } | |
9117 | get_online_cpus(); | |
9118 | } | |
969c7921 | 9119 | atomic_inc(&synchronize_sched_expedited_count); |
cc631fb7 | 9120 | smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */ |
03b042bf | 9121 | put_online_cpus(); |
03b042bf PM |
9122 | } |
9123 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9124 | ||
9125 | #endif /* #else #ifndef CONFIG_SMP */ |