]>
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 | 34 | #include <linux/highmem.h> |
1da177e4 LT |
35 | #include <asm/mmu_context.h> |
36 | #include <linux/interrupt.h> | |
c59ede7b | 37 | #include <linux/capability.h> |
1da177e4 LT |
38 | #include <linux/completion.h> |
39 | #include <linux/kernel_stat.h> | |
9a11b49a | 40 | #include <linux/debug_locks.h> |
cdd6c482 | 41 | #include <linux/perf_event.h> |
1da177e4 LT |
42 | #include <linux/security.h> |
43 | #include <linux/notifier.h> | |
44 | #include <linux/profile.h> | |
7dfb7103 | 45 | #include <linux/freezer.h> |
198e2f18 | 46 | #include <linux/vmalloc.h> |
1da177e4 LT |
47 | #include <linux/blkdev.h> |
48 | #include <linux/delay.h> | |
b488893a | 49 | #include <linux/pid_namespace.h> |
1da177e4 LT |
50 | #include <linux/smp.h> |
51 | #include <linux/threads.h> | |
52 | #include <linux/timer.h> | |
53 | #include <linux/rcupdate.h> | |
54 | #include <linux/cpu.h> | |
55 | #include <linux/cpuset.h> | |
56 | #include <linux/percpu.h> | |
b5aadf7f | 57 | #include <linux/proc_fs.h> |
1da177e4 | 58 | #include <linux/seq_file.h> |
969c7921 | 59 | #include <linux/stop_machine.h> |
e692ab53 | 60 | #include <linux/sysctl.h> |
1da177e4 LT |
61 | #include <linux/syscalls.h> |
62 | #include <linux/times.h> | |
8f0ab514 | 63 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 64 | #include <linux/kprobes.h> |
0ff92245 | 65 | #include <linux/delayacct.h> |
dff06c15 | 66 | #include <linux/unistd.h> |
f5ff8422 | 67 | #include <linux/pagemap.h> |
8f4d37ec | 68 | #include <linux/hrtimer.h> |
30914a58 | 69 | #include <linux/tick.h> |
f00b45c1 PZ |
70 | #include <linux/debugfs.h> |
71 | #include <linux/ctype.h> | |
6cd8a4bb | 72 | #include <linux/ftrace.h> |
5a0e3ad6 | 73 | #include <linux/slab.h> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
335d7afb | 77 | #include <asm/mutex.h> |
1da177e4 | 78 | |
6e0534f2 | 79 | #include "sched_cpupri.h" |
21aa9af0 | 80 | #include "workqueue_sched.h" |
5091faa4 | 81 | #include "sched_autogroup.h" |
6e0534f2 | 82 | |
a8d154b0 | 83 | #define CREATE_TRACE_POINTS |
ad8d75ff | 84 | #include <trace/events/sched.h> |
a8d154b0 | 85 | |
1da177e4 LT |
86 | /* |
87 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
88 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
89 | * and back. | |
90 | */ | |
91 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
92 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
93 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
94 | ||
95 | /* | |
96 | * 'User priority' is the nice value converted to something we | |
97 | * can work with better when scaling various scheduler parameters, | |
98 | * it's a [ 0 ... 39 ] range. | |
99 | */ | |
100 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
101 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
102 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
103 | ||
104 | /* | |
d7876a08 | 105 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 106 | */ |
d6322faf | 107 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 108 | |
6aa645ea IM |
109 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
110 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
111 | ||
1da177e4 LT |
112 | /* |
113 | * These are the 'tuning knobs' of the scheduler: | |
114 | * | |
a4ec24b4 | 115 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
116 | * Timeslices get refilled after they expire. |
117 | */ | |
1da177e4 | 118 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 119 | |
d0b27fa7 PZ |
120 | /* |
121 | * single value that denotes runtime == period, ie unlimited time. | |
122 | */ | |
123 | #define RUNTIME_INF ((u64)~0ULL) | |
124 | ||
e05606d3 IM |
125 | static inline int rt_policy(int policy) |
126 | { | |
3f33a7ce | 127 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
128 | return 1; |
129 | return 0; | |
130 | } | |
131 | ||
132 | static inline int task_has_rt_policy(struct task_struct *p) | |
133 | { | |
134 | return rt_policy(p->policy); | |
135 | } | |
136 | ||
1da177e4 | 137 | /* |
6aa645ea | 138 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 139 | */ |
6aa645ea IM |
140 | struct rt_prio_array { |
141 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
142 | struct list_head queue[MAX_RT_PRIO]; | |
143 | }; | |
144 | ||
d0b27fa7 | 145 | struct rt_bandwidth { |
ea736ed5 | 146 | /* nests inside the rq lock: */ |
0986b11b | 147 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
148 | ktime_t rt_period; |
149 | u64 rt_runtime; | |
150 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
151 | }; |
152 | ||
153 | static struct rt_bandwidth def_rt_bandwidth; | |
154 | ||
155 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
156 | ||
157 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
158 | { | |
159 | struct rt_bandwidth *rt_b = | |
160 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
161 | ktime_t now; | |
162 | int overrun; | |
163 | int idle = 0; | |
164 | ||
165 | for (;;) { | |
166 | now = hrtimer_cb_get_time(timer); | |
167 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
168 | ||
169 | if (!overrun) | |
170 | break; | |
171 | ||
172 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
173 | } | |
174 | ||
175 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
176 | } | |
177 | ||
178 | static | |
179 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
180 | { | |
181 | rt_b->rt_period = ns_to_ktime(period); | |
182 | rt_b->rt_runtime = runtime; | |
183 | ||
0986b11b | 184 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 185 | |
d0b27fa7 PZ |
186 | hrtimer_init(&rt_b->rt_period_timer, |
187 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
188 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
189 | } |
190 | ||
c8bfff6d KH |
191 | static inline int rt_bandwidth_enabled(void) |
192 | { | |
193 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
194 | } |
195 | ||
196 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
197 | { | |
198 | ktime_t now; | |
199 | ||
cac64d00 | 200 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
201 | return; |
202 | ||
203 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
204 | return; | |
205 | ||
0986b11b | 206 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 207 | for (;;) { |
7f1e2ca9 PZ |
208 | unsigned long delta; |
209 | ktime_t soft, hard; | |
210 | ||
d0b27fa7 PZ |
211 | if (hrtimer_active(&rt_b->rt_period_timer)) |
212 | break; | |
213 | ||
214 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
215 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
216 | |
217 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
218 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
219 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
220 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 221 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 222 | } |
0986b11b | 223 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
224 | } |
225 | ||
226 | #ifdef CONFIG_RT_GROUP_SCHED | |
227 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
228 | { | |
229 | hrtimer_cancel(&rt_b->rt_period_timer); | |
230 | } | |
231 | #endif | |
232 | ||
712555ee | 233 | /* |
c4a8849a | 234 | * sched_domains_mutex serializes calls to init_sched_domains, |
712555ee HC |
235 | * detach_destroy_domains and partition_sched_domains. |
236 | */ | |
237 | static DEFINE_MUTEX(sched_domains_mutex); | |
238 | ||
7c941438 | 239 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 240 | |
68318b8e SV |
241 | #include <linux/cgroup.h> |
242 | ||
29f59db3 SV |
243 | struct cfs_rq; |
244 | ||
6f505b16 PZ |
245 | static LIST_HEAD(task_groups); |
246 | ||
29f59db3 | 247 | /* task group related information */ |
4cf86d77 | 248 | struct task_group { |
68318b8e | 249 | struct cgroup_subsys_state css; |
6c415b92 | 250 | |
052f1dc7 | 251 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
252 | /* schedulable entities of this group on each cpu */ |
253 | struct sched_entity **se; | |
254 | /* runqueue "owned" by this group on each cpu */ | |
255 | struct cfs_rq **cfs_rq; | |
256 | unsigned long shares; | |
2069dd75 PZ |
257 | |
258 | atomic_t load_weight; | |
052f1dc7 PZ |
259 | #endif |
260 | ||
261 | #ifdef CONFIG_RT_GROUP_SCHED | |
262 | struct sched_rt_entity **rt_se; | |
263 | struct rt_rq **rt_rq; | |
264 | ||
d0b27fa7 | 265 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 266 | #endif |
6b2d7700 | 267 | |
ae8393e5 | 268 | struct rcu_head rcu; |
6f505b16 | 269 | struct list_head list; |
f473aa5e PZ |
270 | |
271 | struct task_group *parent; | |
272 | struct list_head siblings; | |
273 | struct list_head children; | |
5091faa4 MG |
274 | |
275 | #ifdef CONFIG_SCHED_AUTOGROUP | |
276 | struct autogroup *autogroup; | |
277 | #endif | |
29f59db3 SV |
278 | }; |
279 | ||
3d4b47b4 | 280 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 281 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 282 | |
e9036b36 CG |
283 | #ifdef CONFIG_FAIR_GROUP_SCHED |
284 | ||
07e06b01 | 285 | # define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 286 | |
cb4ad1ff | 287 | /* |
2e084786 LJ |
288 | * A weight of 0 or 1 can cause arithmetics problems. |
289 | * A weight of a cfs_rq is the sum of weights of which entities | |
290 | * are queued on this cfs_rq, so a weight of a entity should not be | |
291 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
292 | * (The default weight is 1024 - so there's no practical |
293 | * limitation from this.) | |
294 | */ | |
18d95a28 | 295 | #define MIN_SHARES 2 |
c8b28116 | 296 | #define MAX_SHARES (1UL << (18 + SCHED_LOAD_RESOLUTION)) |
18d95a28 | 297 | |
07e06b01 | 298 | static int root_task_group_load = ROOT_TASK_GROUP_LOAD; |
052f1dc7 PZ |
299 | #endif |
300 | ||
29f59db3 | 301 | /* Default task group. |
3a252015 | 302 | * Every task in system belong to this group at bootup. |
29f59db3 | 303 | */ |
07e06b01 | 304 | struct task_group root_task_group; |
29f59db3 | 305 | |
7c941438 | 306 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 307 | |
6aa645ea IM |
308 | /* CFS-related fields in a runqueue */ |
309 | struct cfs_rq { | |
310 | struct load_weight load; | |
311 | unsigned long nr_running; | |
312 | ||
6aa645ea | 313 | u64 exec_clock; |
e9acbff6 | 314 | u64 min_vruntime; |
3fe1698b PZ |
315 | #ifndef CONFIG_64BIT |
316 | u64 min_vruntime_copy; | |
317 | #endif | |
6aa645ea IM |
318 | |
319 | struct rb_root tasks_timeline; | |
320 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
321 | |
322 | struct list_head tasks; | |
323 | struct list_head *balance_iterator; | |
324 | ||
325 | /* | |
326 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
327 | * It is set to NULL otherwise (i.e when none are currently running). |
328 | */ | |
ac53db59 | 329 | struct sched_entity *curr, *next, *last, *skip; |
ddc97297 | 330 | |
4934a4d3 | 331 | #ifdef CONFIG_SCHED_DEBUG |
5ac5c4d6 | 332 | unsigned int nr_spread_over; |
4934a4d3 | 333 | #endif |
ddc97297 | 334 | |
62160e3f | 335 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
336 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
337 | ||
41a2d6cf IM |
338 | /* |
339 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
340 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
341 | * (like users, containers etc.) | |
342 | * | |
343 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
344 | * list is used during load balance. | |
345 | */ | |
3d4b47b4 | 346 | int on_list; |
41a2d6cf IM |
347 | struct list_head leaf_cfs_rq_list; |
348 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
349 | |
350 | #ifdef CONFIG_SMP | |
c09595f6 | 351 | /* |
c8cba857 | 352 | * the part of load.weight contributed by tasks |
c09595f6 | 353 | */ |
c8cba857 | 354 | unsigned long task_weight; |
c09595f6 | 355 | |
c8cba857 PZ |
356 | /* |
357 | * h_load = weight * f(tg) | |
358 | * | |
359 | * Where f(tg) is the recursive weight fraction assigned to | |
360 | * this group. | |
361 | */ | |
362 | unsigned long h_load; | |
c09595f6 | 363 | |
c8cba857 | 364 | /* |
3b3d190e PT |
365 | * Maintaining per-cpu shares distribution for group scheduling |
366 | * | |
367 | * load_stamp is the last time we updated the load average | |
368 | * load_last is the last time we updated the load average and saw load | |
369 | * load_unacc_exec_time is currently unaccounted execution time | |
c8cba857 | 370 | */ |
2069dd75 PZ |
371 | u64 load_avg; |
372 | u64 load_period; | |
3b3d190e | 373 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 374 | |
2069dd75 | 375 | unsigned long load_contribution; |
c09595f6 | 376 | #endif |
6aa645ea IM |
377 | #endif |
378 | }; | |
1da177e4 | 379 | |
6aa645ea IM |
380 | /* Real-Time classes' related field in a runqueue: */ |
381 | struct rt_rq { | |
382 | struct rt_prio_array active; | |
63489e45 | 383 | unsigned long rt_nr_running; |
052f1dc7 | 384 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
385 | struct { |
386 | int curr; /* highest queued rt task prio */ | |
398a153b | 387 | #ifdef CONFIG_SMP |
e864c499 | 388 | int next; /* next highest */ |
398a153b | 389 | #endif |
e864c499 | 390 | } highest_prio; |
6f505b16 | 391 | #endif |
fa85ae24 | 392 | #ifdef CONFIG_SMP |
73fe6aae | 393 | unsigned long rt_nr_migratory; |
a1ba4d8b | 394 | unsigned long rt_nr_total; |
a22d7fc1 | 395 | int overloaded; |
917b627d | 396 | struct plist_head pushable_tasks; |
fa85ae24 | 397 | #endif |
6f505b16 | 398 | int rt_throttled; |
fa85ae24 | 399 | u64 rt_time; |
ac086bc2 | 400 | u64 rt_runtime; |
ea736ed5 | 401 | /* Nests inside the rq lock: */ |
0986b11b | 402 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 403 | |
052f1dc7 | 404 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
405 | unsigned long rt_nr_boosted; |
406 | ||
6f505b16 PZ |
407 | struct rq *rq; |
408 | struct list_head leaf_rt_rq_list; | |
409 | struct task_group *tg; | |
6f505b16 | 410 | #endif |
6aa645ea IM |
411 | }; |
412 | ||
57d885fe GH |
413 | #ifdef CONFIG_SMP |
414 | ||
415 | /* | |
416 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
417 | * variables. Each exclusive cpuset essentially defines an island domain by |
418 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
419 | * exclusive cpuset is created, we also create and attach a new root-domain |
420 | * object. | |
421 | * | |
57d885fe GH |
422 | */ |
423 | struct root_domain { | |
424 | atomic_t refcount; | |
dce840a0 | 425 | struct rcu_head rcu; |
c6c4927b RR |
426 | cpumask_var_t span; |
427 | cpumask_var_t online; | |
637f5085 | 428 | |
0eab9146 | 429 | /* |
637f5085 GH |
430 | * The "RT overload" flag: it gets set if a CPU has more than |
431 | * one runnable RT task. | |
432 | */ | |
c6c4927b | 433 | cpumask_var_t rto_mask; |
0eab9146 | 434 | atomic_t rto_count; |
6e0534f2 | 435 | struct cpupri cpupri; |
57d885fe GH |
436 | }; |
437 | ||
dc938520 GH |
438 | /* |
439 | * By default the system creates a single root-domain with all cpus as | |
440 | * members (mimicking the global state we have today). | |
441 | */ | |
57d885fe GH |
442 | static struct root_domain def_root_domain; |
443 | ||
ed2d372c | 444 | #endif /* CONFIG_SMP */ |
57d885fe | 445 | |
1da177e4 LT |
446 | /* |
447 | * This is the main, per-CPU runqueue data structure. | |
448 | * | |
449 | * Locking rule: those places that want to lock multiple runqueues | |
450 | * (such as the load balancing or the thread migration code), lock | |
451 | * acquire operations must be ordered by ascending &runqueue. | |
452 | */ | |
70b97a7f | 453 | struct rq { |
d8016491 | 454 | /* runqueue lock: */ |
05fa785c | 455 | raw_spinlock_t lock; |
1da177e4 LT |
456 | |
457 | /* | |
458 | * nr_running and cpu_load should be in the same cacheline because | |
459 | * remote CPUs use both these fields when doing load calculation. | |
460 | */ | |
461 | unsigned long nr_running; | |
6aa645ea IM |
462 | #define CPU_LOAD_IDX_MAX 5 |
463 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 464 | unsigned long last_load_update_tick; |
46cb4b7c | 465 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 466 | u64 nohz_stamp; |
83cd4fe2 | 467 | unsigned char nohz_balance_kick; |
46cb4b7c | 468 | #endif |
61eadef6 | 469 | int skip_clock_update; |
a64692a3 | 470 | |
d8016491 IM |
471 | /* capture load from *all* tasks on this cpu: */ |
472 | struct load_weight load; | |
6aa645ea IM |
473 | unsigned long nr_load_updates; |
474 | u64 nr_switches; | |
475 | ||
476 | struct cfs_rq cfs; | |
6f505b16 | 477 | struct rt_rq rt; |
6f505b16 | 478 | |
6aa645ea | 479 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
480 | /* list of leaf cfs_rq on this cpu: */ |
481 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
482 | #endif |
483 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 484 | struct list_head leaf_rt_rq_list; |
1da177e4 | 485 | #endif |
1da177e4 LT |
486 | |
487 | /* | |
488 | * This is part of a global counter where only the total sum | |
489 | * over all CPUs matters. A task can increase this counter on | |
490 | * one CPU and if it got migrated afterwards it may decrease | |
491 | * it on another CPU. Always updated under the runqueue lock: | |
492 | */ | |
493 | unsigned long nr_uninterruptible; | |
494 | ||
34f971f6 | 495 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 496 | unsigned long next_balance; |
1da177e4 | 497 | struct mm_struct *prev_mm; |
6aa645ea | 498 | |
3e51f33f | 499 | u64 clock; |
305e6835 | 500 | u64 clock_task; |
6aa645ea | 501 | |
1da177e4 LT |
502 | atomic_t nr_iowait; |
503 | ||
504 | #ifdef CONFIG_SMP | |
0eab9146 | 505 | struct root_domain *rd; |
1da177e4 LT |
506 | struct sched_domain *sd; |
507 | ||
e51fd5e2 PZ |
508 | unsigned long cpu_power; |
509 | ||
a0a522ce | 510 | unsigned char idle_at_tick; |
1da177e4 | 511 | /* For active balancing */ |
3f029d3c | 512 | int post_schedule; |
1da177e4 LT |
513 | int active_balance; |
514 | int push_cpu; | |
969c7921 | 515 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
516 | /* cpu of this runqueue: */ |
517 | int cpu; | |
1f11eb6a | 518 | int online; |
1da177e4 | 519 | |
a8a51d5e | 520 | unsigned long avg_load_per_task; |
1da177e4 | 521 | |
e9e9250b PZ |
522 | u64 rt_avg; |
523 | u64 age_stamp; | |
1b9508f6 MG |
524 | u64 idle_stamp; |
525 | u64 avg_idle; | |
1da177e4 LT |
526 | #endif |
527 | ||
aa483808 VP |
528 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
529 | u64 prev_irq_time; | |
530 | #endif | |
531 | ||
dce48a84 TG |
532 | /* calc_load related fields */ |
533 | unsigned long calc_load_update; | |
534 | long calc_load_active; | |
535 | ||
8f4d37ec | 536 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
537 | #ifdef CONFIG_SMP |
538 | int hrtick_csd_pending; | |
539 | struct call_single_data hrtick_csd; | |
540 | #endif | |
8f4d37ec PZ |
541 | struct hrtimer hrtick_timer; |
542 | #endif | |
543 | ||
1da177e4 LT |
544 | #ifdef CONFIG_SCHEDSTATS |
545 | /* latency stats */ | |
546 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
547 | unsigned long long rq_cpu_time; |
548 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
549 | |
550 | /* sys_sched_yield() stats */ | |
480b9434 | 551 | unsigned int yld_count; |
1da177e4 LT |
552 | |
553 | /* schedule() stats */ | |
480b9434 KC |
554 | unsigned int sched_switch; |
555 | unsigned int sched_count; | |
556 | unsigned int sched_goidle; | |
1da177e4 LT |
557 | |
558 | /* try_to_wake_up() stats */ | |
480b9434 KC |
559 | unsigned int ttwu_count; |
560 | unsigned int ttwu_local; | |
1da177e4 | 561 | #endif |
317f3941 PZ |
562 | |
563 | #ifdef CONFIG_SMP | |
564 | struct task_struct *wake_list; | |
565 | #endif | |
1da177e4 LT |
566 | }; |
567 | ||
f34e3b61 | 568 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 569 | |
a64692a3 | 570 | |
1e5a7405 | 571 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 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 | 583 | rcu_dereference_check((p), \ |
dce840a0 | 584 | rcu_read_lock_held() || \ |
d11c563d PM |
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 | |
0122ec5b | 609 | * with lockdep_is_held(&p->pi_lock) because cpu_cgroup_attach() |
dc61b1d6 PZ |
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 | { | |
5091faa4 | 615 | struct task_group *tg; |
dc61b1d6 PZ |
616 | struct cgroup_subsys_state *css; |
617 | ||
618 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
0122ec5b | 619 | lockdep_is_held(&p->pi_lock)); |
5091faa4 MG |
620 | tg = container_of(css, struct task_group, css); |
621 | ||
622 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
623 | } |
624 | ||
625 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
626 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
627 | { | |
628 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
629 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
630 | p->se.parent = task_group(p)->se[cpu]; | |
631 | #endif | |
632 | ||
633 | #ifdef CONFIG_RT_GROUP_SCHED | |
634 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
635 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
636 | #endif | |
637 | } | |
638 | ||
639 | #else /* CONFIG_CGROUP_SCHED */ | |
640 | ||
641 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
642 | static inline struct task_group *task_group(struct task_struct *p) | |
643 | { | |
644 | return NULL; | |
645 | } | |
646 | ||
647 | #endif /* CONFIG_CGROUP_SCHED */ | |
648 | ||
fe44d621 | 649 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 650 | |
fe44d621 | 651 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 652 | { |
fe44d621 | 653 | s64 delta; |
305e6835 | 654 | |
61eadef6 | 655 | if (rq->skip_clock_update > 0) |
f26f9aff | 656 | return; |
aa483808 | 657 | |
fe44d621 PZ |
658 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
659 | rq->clock += delta; | |
660 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
661 | } |
662 | ||
bf5c91ba IM |
663 | /* |
664 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
665 | */ | |
666 | #ifdef CONFIG_SCHED_DEBUG | |
667 | # define const_debug __read_mostly | |
668 | #else | |
669 | # define const_debug static const | |
670 | #endif | |
671 | ||
017730c1 | 672 | /** |
1fd06bb1 | 673 | * runqueue_is_locked - Returns true if the current cpu runqueue is locked |
e17b38bf | 674 | * @cpu: the processor in question. |
017730c1 | 675 | * |
017730c1 IM |
676 | * This interface allows printk to be called with the runqueue lock |
677 | * held and know whether or not it is OK to wake up the klogd. | |
678 | */ | |
89f19f04 | 679 | int runqueue_is_locked(int cpu) |
017730c1 | 680 | { |
05fa785c | 681 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
682 | } |
683 | ||
bf5c91ba IM |
684 | /* |
685 | * Debugging: various feature bits | |
686 | */ | |
f00b45c1 PZ |
687 | |
688 | #define SCHED_FEAT(name, enabled) \ | |
689 | __SCHED_FEAT_##name , | |
690 | ||
bf5c91ba | 691 | enum { |
f00b45c1 | 692 | #include "sched_features.h" |
bf5c91ba IM |
693 | }; |
694 | ||
f00b45c1 PZ |
695 | #undef SCHED_FEAT |
696 | ||
697 | #define SCHED_FEAT(name, enabled) \ | |
698 | (1UL << __SCHED_FEAT_##name) * enabled | | |
699 | ||
bf5c91ba | 700 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
701 | #include "sched_features.h" |
702 | 0; | |
703 | ||
704 | #undef SCHED_FEAT | |
705 | ||
706 | #ifdef CONFIG_SCHED_DEBUG | |
707 | #define SCHED_FEAT(name, enabled) \ | |
708 | #name , | |
709 | ||
983ed7a6 | 710 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
711 | #include "sched_features.h" |
712 | NULL | |
713 | }; | |
714 | ||
715 | #undef SCHED_FEAT | |
716 | ||
34f3a814 | 717 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 718 | { |
f00b45c1 PZ |
719 | int i; |
720 | ||
721 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
722 | if (!(sysctl_sched_features & (1UL << i))) |
723 | seq_puts(m, "NO_"); | |
724 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 725 | } |
34f3a814 | 726 | seq_puts(m, "\n"); |
f00b45c1 | 727 | |
34f3a814 | 728 | return 0; |
f00b45c1 PZ |
729 | } |
730 | ||
731 | static ssize_t | |
732 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
733 | size_t cnt, loff_t *ppos) | |
734 | { | |
735 | char buf[64]; | |
7740191c | 736 | char *cmp; |
f00b45c1 PZ |
737 | int neg = 0; |
738 | int i; | |
739 | ||
740 | if (cnt > 63) | |
741 | cnt = 63; | |
742 | ||
743 | if (copy_from_user(&buf, ubuf, cnt)) | |
744 | return -EFAULT; | |
745 | ||
746 | buf[cnt] = 0; | |
7740191c | 747 | cmp = strstrip(buf); |
f00b45c1 | 748 | |
524429c3 | 749 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
750 | neg = 1; |
751 | cmp += 3; | |
752 | } | |
753 | ||
754 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 755 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
756 | if (neg) |
757 | sysctl_sched_features &= ~(1UL << i); | |
758 | else | |
759 | sysctl_sched_features |= (1UL << i); | |
760 | break; | |
761 | } | |
762 | } | |
763 | ||
764 | if (!sched_feat_names[i]) | |
765 | return -EINVAL; | |
766 | ||
42994724 | 767 | *ppos += cnt; |
f00b45c1 PZ |
768 | |
769 | return cnt; | |
770 | } | |
771 | ||
34f3a814 LZ |
772 | static int sched_feat_open(struct inode *inode, struct file *filp) |
773 | { | |
774 | return single_open(filp, sched_feat_show, NULL); | |
775 | } | |
776 | ||
828c0950 | 777 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
778 | .open = sched_feat_open, |
779 | .write = sched_feat_write, | |
780 | .read = seq_read, | |
781 | .llseek = seq_lseek, | |
782 | .release = single_release, | |
f00b45c1 PZ |
783 | }; |
784 | ||
785 | static __init int sched_init_debug(void) | |
786 | { | |
f00b45c1 PZ |
787 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
788 | &sched_feat_fops); | |
789 | ||
790 | return 0; | |
791 | } | |
792 | late_initcall(sched_init_debug); | |
793 | ||
794 | #endif | |
795 | ||
796 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 797 | |
b82d9fdd PZ |
798 | /* |
799 | * Number of tasks to iterate in a single balance run. | |
800 | * Limited because this is done with IRQs disabled. | |
801 | */ | |
802 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
803 | ||
e9e9250b PZ |
804 | /* |
805 | * period over which we average the RT time consumption, measured | |
806 | * in ms. | |
807 | * | |
808 | * default: 1s | |
809 | */ | |
810 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
811 | ||
fa85ae24 | 812 | /* |
9f0c1e56 | 813 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
814 | * default: 1s |
815 | */ | |
9f0c1e56 | 816 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 817 | |
6892b75e IM |
818 | static __read_mostly int scheduler_running; |
819 | ||
9f0c1e56 PZ |
820 | /* |
821 | * part of the period that we allow rt tasks to run in us. | |
822 | * default: 0.95s | |
823 | */ | |
824 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 825 | |
d0b27fa7 PZ |
826 | static inline u64 global_rt_period(void) |
827 | { | |
828 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
829 | } | |
830 | ||
831 | static inline u64 global_rt_runtime(void) | |
832 | { | |
e26873bb | 833 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
834 | return RUNTIME_INF; |
835 | ||
836 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
837 | } | |
fa85ae24 | 838 | |
1da177e4 | 839 | #ifndef prepare_arch_switch |
4866cde0 NP |
840 | # define prepare_arch_switch(next) do { } while (0) |
841 | #endif | |
842 | #ifndef finish_arch_switch | |
843 | # define finish_arch_switch(prev) do { } while (0) | |
844 | #endif | |
845 | ||
051a1d1a DA |
846 | static inline int task_current(struct rq *rq, struct task_struct *p) |
847 | { | |
848 | return rq->curr == p; | |
849 | } | |
850 | ||
70b97a7f | 851 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 852 | { |
3ca7a440 PZ |
853 | #ifdef CONFIG_SMP |
854 | return p->on_cpu; | |
855 | #else | |
051a1d1a | 856 | return task_current(rq, p); |
3ca7a440 | 857 | #endif |
4866cde0 NP |
858 | } |
859 | ||
3ca7a440 | 860 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 861 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 | 862 | { |
3ca7a440 PZ |
863 | #ifdef CONFIG_SMP |
864 | /* | |
865 | * We can optimise this out completely for !SMP, because the | |
866 | * SMP rebalancing from interrupt is the only thing that cares | |
867 | * here. | |
868 | */ | |
869 | next->on_cpu = 1; | |
870 | #endif | |
4866cde0 NP |
871 | } |
872 | ||
70b97a7f | 873 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 874 | { |
3ca7a440 PZ |
875 | #ifdef CONFIG_SMP |
876 | /* | |
877 | * After ->on_cpu is cleared, the task can be moved to a different CPU. | |
878 | * We must ensure this doesn't happen until the switch is completely | |
879 | * finished. | |
880 | */ | |
881 | smp_wmb(); | |
882 | prev->on_cpu = 0; | |
883 | #endif | |
da04c035 IM |
884 | #ifdef CONFIG_DEBUG_SPINLOCK |
885 | /* this is a valid case when another task releases the spinlock */ | |
886 | rq->lock.owner = current; | |
887 | #endif | |
8a25d5de IM |
888 | /* |
889 | * If we are tracking spinlock dependencies then we have to | |
890 | * fix up the runqueue lock - which gets 'carried over' from | |
891 | * prev into current: | |
892 | */ | |
893 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
894 | ||
05fa785c | 895 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
896 | } |
897 | ||
898 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 899 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
900 | { |
901 | #ifdef CONFIG_SMP | |
902 | /* | |
903 | * We can optimise this out completely for !SMP, because the | |
904 | * SMP rebalancing from interrupt is the only thing that cares | |
905 | * here. | |
906 | */ | |
3ca7a440 | 907 | next->on_cpu = 1; |
4866cde0 NP |
908 | #endif |
909 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 910 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 911 | #else |
05fa785c | 912 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
913 | #endif |
914 | } | |
915 | ||
70b97a7f | 916 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
917 | { |
918 | #ifdef CONFIG_SMP | |
919 | /* | |
3ca7a440 | 920 | * After ->on_cpu is cleared, the task can be moved to a different CPU. |
4866cde0 NP |
921 | * We must ensure this doesn't happen until the switch is completely |
922 | * finished. | |
923 | */ | |
924 | smp_wmb(); | |
3ca7a440 | 925 | prev->on_cpu = 0; |
4866cde0 NP |
926 | #endif |
927 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
928 | local_irq_enable(); | |
1da177e4 | 929 | #endif |
4866cde0 NP |
930 | } |
931 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 932 | |
0970d299 | 933 | /* |
0122ec5b | 934 | * __task_rq_lock - lock the rq @p resides on. |
b29739f9 | 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 | ||
0122ec5b PZ |
941 | lockdep_assert_held(&p->pi_lock); |
942 | ||
3a5c359a | 943 | for (;;) { |
0970d299 | 944 | rq = task_rq(p); |
05fa785c | 945 | raw_spin_lock(&rq->lock); |
65cc8e48 | 946 | if (likely(rq == task_rq(p))) |
3a5c359a | 947 | return rq; |
05fa785c | 948 | raw_spin_unlock(&rq->lock); |
b29739f9 | 949 | } |
b29739f9 IM |
950 | } |
951 | ||
1da177e4 | 952 | /* |
0122ec5b | 953 | * task_rq_lock - lock p->pi_lock and lock the rq @p resides on. |
1da177e4 | 954 | */ |
70b97a7f | 955 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
0122ec5b | 956 | __acquires(p->pi_lock) |
1da177e4 LT |
957 | __acquires(rq->lock) |
958 | { | |
70b97a7f | 959 | struct rq *rq; |
1da177e4 | 960 | |
3a5c359a | 961 | for (;;) { |
0122ec5b | 962 | raw_spin_lock_irqsave(&p->pi_lock, *flags); |
3a5c359a | 963 | rq = task_rq(p); |
05fa785c | 964 | raw_spin_lock(&rq->lock); |
65cc8e48 | 965 | if (likely(rq == task_rq(p))) |
3a5c359a | 966 | return rq; |
0122ec5b PZ |
967 | raw_spin_unlock(&rq->lock); |
968 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 | 969 | } |
1da177e4 LT |
970 | } |
971 | ||
a9957449 | 972 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
973 | __releases(rq->lock) |
974 | { | |
05fa785c | 975 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
976 | } |
977 | ||
0122ec5b PZ |
978 | static inline void |
979 | task_rq_unlock(struct rq *rq, struct task_struct *p, unsigned long *flags) | |
1da177e4 | 980 | __releases(rq->lock) |
0122ec5b | 981 | __releases(p->pi_lock) |
1da177e4 | 982 | { |
0122ec5b PZ |
983 | raw_spin_unlock(&rq->lock); |
984 | raw_spin_unlock_irqrestore(&p->pi_lock, *flags); | |
1da177e4 LT |
985 | } |
986 | ||
1da177e4 | 987 | /* |
cc2a73b5 | 988 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 989 | */ |
a9957449 | 990 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
991 | __acquires(rq->lock) |
992 | { | |
70b97a7f | 993 | struct rq *rq; |
1da177e4 LT |
994 | |
995 | local_irq_disable(); | |
996 | rq = this_rq(); | |
05fa785c | 997 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
998 | |
999 | return rq; | |
1000 | } | |
1001 | ||
8f4d37ec PZ |
1002 | #ifdef CONFIG_SCHED_HRTICK |
1003 | /* | |
1004 | * Use HR-timers to deliver accurate preemption points. | |
1005 | * | |
1006 | * Its all a bit involved since we cannot program an hrt while holding the | |
1007 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1008 | * reschedule event. | |
1009 | * | |
1010 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1011 | * rq->lock. | |
1012 | */ | |
8f4d37ec PZ |
1013 | |
1014 | /* | |
1015 | * Use hrtick when: | |
1016 | * - enabled by features | |
1017 | * - hrtimer is actually high res | |
1018 | */ | |
1019 | static inline int hrtick_enabled(struct rq *rq) | |
1020 | { | |
1021 | if (!sched_feat(HRTICK)) | |
1022 | return 0; | |
ba42059f | 1023 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1024 | return 0; |
8f4d37ec PZ |
1025 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1026 | } | |
1027 | ||
8f4d37ec PZ |
1028 | static void hrtick_clear(struct rq *rq) |
1029 | { | |
1030 | if (hrtimer_active(&rq->hrtick_timer)) | |
1031 | hrtimer_cancel(&rq->hrtick_timer); | |
1032 | } | |
1033 | ||
8f4d37ec PZ |
1034 | /* |
1035 | * High-resolution timer tick. | |
1036 | * Runs from hardirq context with interrupts disabled. | |
1037 | */ | |
1038 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1039 | { | |
1040 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1041 | ||
1042 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1043 | ||
05fa785c | 1044 | raw_spin_lock(&rq->lock); |
3e51f33f | 1045 | update_rq_clock(rq); |
8f4d37ec | 1046 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1047 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1048 | |
1049 | return HRTIMER_NORESTART; | |
1050 | } | |
1051 | ||
95e904c7 | 1052 | #ifdef CONFIG_SMP |
31656519 PZ |
1053 | /* |
1054 | * called from hardirq (IPI) context | |
1055 | */ | |
1056 | static void __hrtick_start(void *arg) | |
b328ca18 | 1057 | { |
31656519 | 1058 | struct rq *rq = arg; |
b328ca18 | 1059 | |
05fa785c | 1060 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1061 | hrtimer_restart(&rq->hrtick_timer); |
1062 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1063 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1064 | } |
1065 | ||
31656519 PZ |
1066 | /* |
1067 | * Called to set the hrtick timer state. | |
1068 | * | |
1069 | * called with rq->lock held and irqs disabled | |
1070 | */ | |
1071 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1072 | { |
31656519 PZ |
1073 | struct hrtimer *timer = &rq->hrtick_timer; |
1074 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1075 | |
cc584b21 | 1076 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1077 | |
1078 | if (rq == this_rq()) { | |
1079 | hrtimer_restart(timer); | |
1080 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1081 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1082 | rq->hrtick_csd_pending = 1; |
1083 | } | |
b328ca18 PZ |
1084 | } |
1085 | ||
1086 | static int | |
1087 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1088 | { | |
1089 | int cpu = (int)(long)hcpu; | |
1090 | ||
1091 | switch (action) { | |
1092 | case CPU_UP_CANCELED: | |
1093 | case CPU_UP_CANCELED_FROZEN: | |
1094 | case CPU_DOWN_PREPARE: | |
1095 | case CPU_DOWN_PREPARE_FROZEN: | |
1096 | case CPU_DEAD: | |
1097 | case CPU_DEAD_FROZEN: | |
31656519 | 1098 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1099 | return NOTIFY_OK; |
1100 | } | |
1101 | ||
1102 | return NOTIFY_DONE; | |
1103 | } | |
1104 | ||
fa748203 | 1105 | static __init void init_hrtick(void) |
b328ca18 PZ |
1106 | { |
1107 | hotcpu_notifier(hotplug_hrtick, 0); | |
1108 | } | |
31656519 PZ |
1109 | #else |
1110 | /* | |
1111 | * Called to set the hrtick timer state. | |
1112 | * | |
1113 | * called with rq->lock held and irqs disabled | |
1114 | */ | |
1115 | static void hrtick_start(struct rq *rq, u64 delay) | |
1116 | { | |
7f1e2ca9 | 1117 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1118 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1119 | } |
b328ca18 | 1120 | |
006c75f1 | 1121 | static inline void init_hrtick(void) |
8f4d37ec | 1122 | { |
8f4d37ec | 1123 | } |
31656519 | 1124 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1125 | |
31656519 | 1126 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1127 | { |
31656519 PZ |
1128 | #ifdef CONFIG_SMP |
1129 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1130 | |
31656519 PZ |
1131 | rq->hrtick_csd.flags = 0; |
1132 | rq->hrtick_csd.func = __hrtick_start; | |
1133 | rq->hrtick_csd.info = rq; | |
1134 | #endif | |
8f4d37ec | 1135 | |
31656519 PZ |
1136 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1137 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1138 | } |
006c75f1 | 1139 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1140 | static inline void hrtick_clear(struct rq *rq) |
1141 | { | |
1142 | } | |
1143 | ||
8f4d37ec PZ |
1144 | static inline void init_rq_hrtick(struct rq *rq) |
1145 | { | |
1146 | } | |
1147 | ||
b328ca18 PZ |
1148 | static inline void init_hrtick(void) |
1149 | { | |
1150 | } | |
006c75f1 | 1151 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1152 | |
c24d20db IM |
1153 | /* |
1154 | * resched_task - mark a task 'to be rescheduled now'. | |
1155 | * | |
1156 | * On UP this means the setting of the need_resched flag, on SMP it | |
1157 | * might also involve a cross-CPU call to trigger the scheduler on | |
1158 | * the target CPU. | |
1159 | */ | |
1160 | #ifdef CONFIG_SMP | |
1161 | ||
1162 | #ifndef tsk_is_polling | |
1163 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1164 | #endif | |
1165 | ||
31656519 | 1166 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1167 | { |
1168 | int cpu; | |
1169 | ||
05fa785c | 1170 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1171 | |
5ed0cec0 | 1172 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1173 | return; |
1174 | ||
5ed0cec0 | 1175 | set_tsk_need_resched(p); |
c24d20db IM |
1176 | |
1177 | cpu = task_cpu(p); | |
1178 | if (cpu == smp_processor_id()) | |
1179 | return; | |
1180 | ||
1181 | /* NEED_RESCHED must be visible before we test polling */ | |
1182 | smp_mb(); | |
1183 | if (!tsk_is_polling(p)) | |
1184 | smp_send_reschedule(cpu); | |
1185 | } | |
1186 | ||
1187 | static void resched_cpu(int cpu) | |
1188 | { | |
1189 | struct rq *rq = cpu_rq(cpu); | |
1190 | unsigned long flags; | |
1191 | ||
05fa785c | 1192 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1193 | return; |
1194 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1195 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1196 | } |
06d8308c TG |
1197 | |
1198 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1199 | /* |
1200 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1201 | * from an idle cpu. This is good for power-savings. | |
1202 | * | |
1203 | * We don't do similar optimization for completely idle system, as | |
1204 | * selecting an idle cpu will add more delays to the timers than intended | |
1205 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1206 | */ | |
1207 | int get_nohz_timer_target(void) | |
1208 | { | |
1209 | int cpu = smp_processor_id(); | |
1210 | int i; | |
1211 | struct sched_domain *sd; | |
1212 | ||
057f3fad | 1213 | rcu_read_lock(); |
83cd4fe2 | 1214 | for_each_domain(cpu, sd) { |
057f3fad PZ |
1215 | for_each_cpu(i, sched_domain_span(sd)) { |
1216 | if (!idle_cpu(i)) { | |
1217 | cpu = i; | |
1218 | goto unlock; | |
1219 | } | |
1220 | } | |
83cd4fe2 | 1221 | } |
057f3fad PZ |
1222 | unlock: |
1223 | rcu_read_unlock(); | |
83cd4fe2 VP |
1224 | return cpu; |
1225 | } | |
06d8308c TG |
1226 | /* |
1227 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1228 | * idle CPU then this timer might expire before the next timer event | |
1229 | * which is scheduled to wake up that CPU. In case of a completely | |
1230 | * idle system the next event might even be infinite time into the | |
1231 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1232 | * leaves the inner idle loop so the newly added timer is taken into | |
1233 | * account when the CPU goes back to idle and evaluates the timer | |
1234 | * wheel for the next timer event. | |
1235 | */ | |
1236 | void wake_up_idle_cpu(int cpu) | |
1237 | { | |
1238 | struct rq *rq = cpu_rq(cpu); | |
1239 | ||
1240 | if (cpu == smp_processor_id()) | |
1241 | return; | |
1242 | ||
1243 | /* | |
1244 | * This is safe, as this function is called with the timer | |
1245 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1246 | * to idle and has not yet set rq->curr to idle then it will | |
1247 | * be serialized on the timer wheel base lock and take the new | |
1248 | * timer into account automatically. | |
1249 | */ | |
1250 | if (rq->curr != rq->idle) | |
1251 | return; | |
1252 | ||
1253 | /* | |
1254 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1255 | * lockless. The worst case is that the other CPU runs the | |
1256 | * idle task through an additional NOOP schedule() | |
1257 | */ | |
5ed0cec0 | 1258 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1259 | |
1260 | /* NEED_RESCHED must be visible before we test polling */ | |
1261 | smp_mb(); | |
1262 | if (!tsk_is_polling(rq->idle)) | |
1263 | smp_send_reschedule(cpu); | |
1264 | } | |
39c0cbe2 | 1265 | |
6d6bc0ad | 1266 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1267 | |
e9e9250b PZ |
1268 | static u64 sched_avg_period(void) |
1269 | { | |
1270 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1271 | } | |
1272 | ||
1273 | static void sched_avg_update(struct rq *rq) | |
1274 | { | |
1275 | s64 period = sched_avg_period(); | |
1276 | ||
1277 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1278 | /* |
1279 | * Inline assembly required to prevent the compiler | |
1280 | * optimising this loop into a divmod call. | |
1281 | * See __iter_div_u64_rem() for another example of this. | |
1282 | */ | |
1283 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1284 | rq->age_stamp += period; |
1285 | rq->rt_avg /= 2; | |
1286 | } | |
1287 | } | |
1288 | ||
1289 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1290 | { | |
1291 | rq->rt_avg += rt_delta; | |
1292 | sched_avg_update(rq); | |
1293 | } | |
1294 | ||
6d6bc0ad | 1295 | #else /* !CONFIG_SMP */ |
31656519 | 1296 | static void resched_task(struct task_struct *p) |
c24d20db | 1297 | { |
05fa785c | 1298 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1299 | set_tsk_need_resched(p); |
c24d20db | 1300 | } |
e9e9250b PZ |
1301 | |
1302 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1303 | { | |
1304 | } | |
da2b71ed SS |
1305 | |
1306 | static void sched_avg_update(struct rq *rq) | |
1307 | { | |
1308 | } | |
6d6bc0ad | 1309 | #endif /* CONFIG_SMP */ |
c24d20db | 1310 | |
45bf76df IM |
1311 | #if BITS_PER_LONG == 32 |
1312 | # define WMULT_CONST (~0UL) | |
1313 | #else | |
1314 | # define WMULT_CONST (1UL << 32) | |
1315 | #endif | |
1316 | ||
1317 | #define WMULT_SHIFT 32 | |
1318 | ||
194081eb IM |
1319 | /* |
1320 | * Shift right and round: | |
1321 | */ | |
cf2ab469 | 1322 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1323 | |
a7be37ac PZ |
1324 | /* |
1325 | * delta *= weight / lw | |
1326 | */ | |
cb1c4fc9 | 1327 | static unsigned long |
45bf76df IM |
1328 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1329 | struct load_weight *lw) | |
1330 | { | |
1331 | u64 tmp; | |
1332 | ||
c8b28116 NR |
1333 | /* |
1334 | * weight can be less than 2^SCHED_LOAD_RESOLUTION for task group sched | |
1335 | * entities since MIN_SHARES = 2. Treat weight as 1 if less than | |
1336 | * 2^SCHED_LOAD_RESOLUTION. | |
1337 | */ | |
1338 | if (likely(weight > (1UL << SCHED_LOAD_RESOLUTION))) | |
1339 | tmp = (u64)delta_exec * scale_load_down(weight); | |
1340 | else | |
1341 | tmp = (u64)delta_exec; | |
db670dac | 1342 | |
7a232e03 | 1343 | if (!lw->inv_weight) { |
c8b28116 NR |
1344 | unsigned long w = scale_load_down(lw->weight); |
1345 | ||
1346 | if (BITS_PER_LONG > 32 && unlikely(w >= WMULT_CONST)) | |
7a232e03 | 1347 | lw->inv_weight = 1; |
c8b28116 NR |
1348 | else if (unlikely(!w)) |
1349 | lw->inv_weight = WMULT_CONST; | |
7a232e03 | 1350 | else |
c8b28116 | 1351 | lw->inv_weight = WMULT_CONST / w; |
7a232e03 | 1352 | } |
45bf76df | 1353 | |
45bf76df IM |
1354 | /* |
1355 | * Check whether we'd overflow the 64-bit multiplication: | |
1356 | */ | |
194081eb | 1357 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1358 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1359 | WMULT_SHIFT/2); |
1360 | else | |
cf2ab469 | 1361 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1362 | |
ecf691da | 1363 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1364 | } |
1365 | ||
1091985b | 1366 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1367 | { |
1368 | lw->weight += inc; | |
e89996ae | 1369 | lw->inv_weight = 0; |
45bf76df IM |
1370 | } |
1371 | ||
1091985b | 1372 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1373 | { |
1374 | lw->weight -= dec; | |
e89996ae | 1375 | lw->inv_weight = 0; |
45bf76df IM |
1376 | } |
1377 | ||
2069dd75 PZ |
1378 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1379 | { | |
1380 | lw->weight = w; | |
1381 | lw->inv_weight = 0; | |
1382 | } | |
1383 | ||
2dd73a4f PW |
1384 | /* |
1385 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1386 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1387 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1388 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1389 | * scaled version of the new time slice allocation that they receive on time |
1390 | * slice expiry etc. | |
1391 | */ | |
1392 | ||
cce7ade8 PZ |
1393 | #define WEIGHT_IDLEPRIO 3 |
1394 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1395 | |
1396 | /* | |
1397 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1398 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1399 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1400 | * that remained on nice 0. | |
1401 | * | |
1402 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1403 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1404 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1405 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1406 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1407 | */ |
1408 | static const int prio_to_weight[40] = { | |
254753dc IM |
1409 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1410 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1411 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1412 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1413 | /* 0 */ 1024, 820, 655, 526, 423, | |
1414 | /* 5 */ 335, 272, 215, 172, 137, | |
1415 | /* 10 */ 110, 87, 70, 56, 45, | |
1416 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1417 | }; |
1418 | ||
5714d2de IM |
1419 | /* |
1420 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1421 | * | |
1422 | * In cases where the weight does not change often, we can use the | |
1423 | * precalculated inverse to speed up arithmetics by turning divisions | |
1424 | * into multiplications: | |
1425 | */ | |
dd41f596 | 1426 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1427 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1428 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1429 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1430 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1431 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1432 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1433 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1434 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1435 | }; |
2dd73a4f | 1436 | |
ef12fefa BR |
1437 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1438 | enum cpuacct_stat_index { | |
1439 | CPUACCT_STAT_USER, /* ... user mode */ | |
1440 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1441 | ||
1442 | CPUACCT_STAT_NSTATS, | |
1443 | }; | |
1444 | ||
d842de87 SV |
1445 | #ifdef CONFIG_CGROUP_CPUACCT |
1446 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1447 | static void cpuacct_update_stats(struct task_struct *tsk, |
1448 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1449 | #else |
1450 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1451 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1452 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1453 | #endif |
1454 | ||
18d95a28 PZ |
1455 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1456 | { | |
1457 | update_load_add(&rq->load, load); | |
1458 | } | |
1459 | ||
1460 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1461 | { | |
1462 | update_load_sub(&rq->load, load); | |
1463 | } | |
1464 | ||
7940ca36 | 1465 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1466 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1467 | |
1468 | /* | |
1469 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1470 | * leaving it for the final time. | |
1471 | */ | |
eb755805 | 1472 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1473 | { |
1474 | struct task_group *parent, *child; | |
eb755805 | 1475 | int ret; |
c09595f6 PZ |
1476 | |
1477 | rcu_read_lock(); | |
1478 | parent = &root_task_group; | |
1479 | down: | |
eb755805 PZ |
1480 | ret = (*down)(parent, data); |
1481 | if (ret) | |
1482 | goto out_unlock; | |
c09595f6 PZ |
1483 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1484 | parent = child; | |
1485 | goto down; | |
1486 | ||
1487 | up: | |
1488 | continue; | |
1489 | } | |
eb755805 PZ |
1490 | ret = (*up)(parent, data); |
1491 | if (ret) | |
1492 | goto out_unlock; | |
c09595f6 PZ |
1493 | |
1494 | child = parent; | |
1495 | parent = parent->parent; | |
1496 | if (parent) | |
1497 | goto up; | |
eb755805 | 1498 | out_unlock: |
c09595f6 | 1499 | rcu_read_unlock(); |
eb755805 PZ |
1500 | |
1501 | return ret; | |
c09595f6 PZ |
1502 | } |
1503 | ||
eb755805 PZ |
1504 | static int tg_nop(struct task_group *tg, void *data) |
1505 | { | |
1506 | return 0; | |
c09595f6 | 1507 | } |
eb755805 PZ |
1508 | #endif |
1509 | ||
1510 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1511 | /* Used instead of source_load when we know the type == 0 */ |
1512 | static unsigned long weighted_cpuload(const int cpu) | |
1513 | { | |
1514 | return cpu_rq(cpu)->load.weight; | |
1515 | } | |
1516 | ||
1517 | /* | |
1518 | * Return a low guess at the load of a migration-source cpu weighted | |
1519 | * according to the scheduling class and "nice" value. | |
1520 | * | |
1521 | * We want to under-estimate the load of migration sources, to | |
1522 | * balance conservatively. | |
1523 | */ | |
1524 | static unsigned long source_load(int cpu, int type) | |
1525 | { | |
1526 | struct rq *rq = cpu_rq(cpu); | |
1527 | unsigned long total = weighted_cpuload(cpu); | |
1528 | ||
1529 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1530 | return total; | |
1531 | ||
1532 | return min(rq->cpu_load[type-1], total); | |
1533 | } | |
1534 | ||
1535 | /* | |
1536 | * Return a high guess at the load of a migration-target cpu weighted | |
1537 | * according to the scheduling class and "nice" value. | |
1538 | */ | |
1539 | static unsigned long target_load(int cpu, int type) | |
1540 | { | |
1541 | struct rq *rq = cpu_rq(cpu); | |
1542 | unsigned long total = weighted_cpuload(cpu); | |
1543 | ||
1544 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1545 | return total; | |
1546 | ||
1547 | return max(rq->cpu_load[type-1], total); | |
1548 | } | |
1549 | ||
ae154be1 PZ |
1550 | static unsigned long power_of(int cpu) |
1551 | { | |
e51fd5e2 | 1552 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1553 | } |
1554 | ||
eb755805 PZ |
1555 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1556 | ||
1557 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1558 | { | |
1559 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1560 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1561 | |
4cd42620 SR |
1562 | if (nr_running) |
1563 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1564 | else |
1565 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1566 | |
1567 | return rq->avg_load_per_task; | |
1568 | } | |
1569 | ||
1570 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1571 | |
c09595f6 | 1572 | /* |
c8cba857 PZ |
1573 | * Compute the cpu's hierarchical load factor for each task group. |
1574 | * This needs to be done in a top-down fashion because the load of a child | |
1575 | * group is a fraction of its parents load. | |
c09595f6 | 1576 | */ |
eb755805 | 1577 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1578 | { |
c8cba857 | 1579 | unsigned long load; |
eb755805 | 1580 | long cpu = (long)data; |
c09595f6 | 1581 | |
c8cba857 PZ |
1582 | if (!tg->parent) { |
1583 | load = cpu_rq(cpu)->load.weight; | |
1584 | } else { | |
1585 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1586 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1587 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1588 | } | |
c09595f6 | 1589 | |
c8cba857 | 1590 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1591 | |
eb755805 | 1592 | return 0; |
c09595f6 PZ |
1593 | } |
1594 | ||
eb755805 | 1595 | static void update_h_load(long cpu) |
c09595f6 | 1596 | { |
eb755805 | 1597 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1598 | } |
1599 | ||
18d95a28 PZ |
1600 | #endif |
1601 | ||
8f45e2b5 GH |
1602 | #ifdef CONFIG_PREEMPT |
1603 | ||
b78bb868 PZ |
1604 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1605 | ||
70574a99 | 1606 | /* |
8f45e2b5 GH |
1607 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1608 | * way at the expense of forcing extra atomic operations in all | |
1609 | * invocations. This assures that the double_lock is acquired using the | |
1610 | * same underlying policy as the spinlock_t on this architecture, which | |
1611 | * reduces latency compared to the unfair variant below. However, it | |
1612 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1613 | */ |
8f45e2b5 GH |
1614 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1615 | __releases(this_rq->lock) | |
1616 | __acquires(busiest->lock) | |
1617 | __acquires(this_rq->lock) | |
1618 | { | |
05fa785c | 1619 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1620 | double_rq_lock(this_rq, busiest); |
1621 | ||
1622 | return 1; | |
1623 | } | |
1624 | ||
1625 | #else | |
1626 | /* | |
1627 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1628 | * latency by eliminating extra atomic operations when the locks are | |
1629 | * already in proper order on entry. This favors lower cpu-ids and will | |
1630 | * grant the double lock to lower cpus over higher ids under contention, | |
1631 | * regardless of entry order into the function. | |
1632 | */ | |
1633 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1634 | __releases(this_rq->lock) |
1635 | __acquires(busiest->lock) | |
1636 | __acquires(this_rq->lock) | |
1637 | { | |
1638 | int ret = 0; | |
1639 | ||
05fa785c | 1640 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1641 | if (busiest < this_rq) { |
05fa785c TG |
1642 | raw_spin_unlock(&this_rq->lock); |
1643 | raw_spin_lock(&busiest->lock); | |
1644 | raw_spin_lock_nested(&this_rq->lock, | |
1645 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1646 | ret = 1; |
1647 | } else | |
05fa785c TG |
1648 | raw_spin_lock_nested(&busiest->lock, |
1649 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1650 | } |
1651 | return ret; | |
1652 | } | |
1653 | ||
8f45e2b5 GH |
1654 | #endif /* CONFIG_PREEMPT */ |
1655 | ||
1656 | /* | |
1657 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1658 | */ | |
1659 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1660 | { | |
1661 | if (unlikely(!irqs_disabled())) { | |
1662 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1663 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1664 | BUG_ON(1); |
1665 | } | |
1666 | ||
1667 | return _double_lock_balance(this_rq, busiest); | |
1668 | } | |
1669 | ||
70574a99 AD |
1670 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1671 | __releases(busiest->lock) | |
1672 | { | |
05fa785c | 1673 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1674 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1675 | } | |
1e3c88bd PZ |
1676 | |
1677 | /* | |
1678 | * double_rq_lock - safely lock two runqueues | |
1679 | * | |
1680 | * Note this does not disable interrupts like task_rq_lock, | |
1681 | * you need to do so manually before calling. | |
1682 | */ | |
1683 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1684 | __acquires(rq1->lock) | |
1685 | __acquires(rq2->lock) | |
1686 | { | |
1687 | BUG_ON(!irqs_disabled()); | |
1688 | if (rq1 == rq2) { | |
1689 | raw_spin_lock(&rq1->lock); | |
1690 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1691 | } else { | |
1692 | if (rq1 < rq2) { | |
1693 | raw_spin_lock(&rq1->lock); | |
1694 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1695 | } else { | |
1696 | raw_spin_lock(&rq2->lock); | |
1697 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1698 | } | |
1699 | } | |
1e3c88bd PZ |
1700 | } |
1701 | ||
1702 | /* | |
1703 | * double_rq_unlock - safely unlock two runqueues | |
1704 | * | |
1705 | * Note this does not restore interrupts like task_rq_unlock, | |
1706 | * you need to do so manually after calling. | |
1707 | */ | |
1708 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1709 | __releases(rq1->lock) | |
1710 | __releases(rq2->lock) | |
1711 | { | |
1712 | raw_spin_unlock(&rq1->lock); | |
1713 | if (rq1 != rq2) | |
1714 | raw_spin_unlock(&rq2->lock); | |
1715 | else | |
1716 | __release(rq2->lock); | |
1717 | } | |
1718 | ||
d95f4122 MG |
1719 | #else /* CONFIG_SMP */ |
1720 | ||
1721 | /* | |
1722 | * double_rq_lock - safely lock two runqueues | |
1723 | * | |
1724 | * Note this does not disable interrupts like task_rq_lock, | |
1725 | * you need to do so manually before calling. | |
1726 | */ | |
1727 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1728 | __acquires(rq1->lock) | |
1729 | __acquires(rq2->lock) | |
1730 | { | |
1731 | BUG_ON(!irqs_disabled()); | |
1732 | BUG_ON(rq1 != rq2); | |
1733 | raw_spin_lock(&rq1->lock); | |
1734 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1735 | } | |
1736 | ||
1737 | /* | |
1738 | * double_rq_unlock - safely unlock two runqueues | |
1739 | * | |
1740 | * Note this does not restore interrupts like task_rq_unlock, | |
1741 | * you need to do so manually after calling. | |
1742 | */ | |
1743 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1744 | __releases(rq1->lock) | |
1745 | __releases(rq2->lock) | |
1746 | { | |
1747 | BUG_ON(rq1 != rq2); | |
1748 | raw_spin_unlock(&rq1->lock); | |
1749 | __release(rq2->lock); | |
1750 | } | |
1751 | ||
18d95a28 PZ |
1752 | #endif |
1753 | ||
74f5187a | 1754 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1755 | static void update_sysctl(void); |
acb4a848 | 1756 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1757 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1758 | |
cd29fe6f PZ |
1759 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1760 | { | |
1761 | set_task_rq(p, cpu); | |
1762 | #ifdef CONFIG_SMP | |
1763 | /* | |
1764 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1765 | * successfuly executed on another CPU. We must ensure that updates of | |
1766 | * per-task data have been completed by this moment. | |
1767 | */ | |
1768 | smp_wmb(); | |
1769 | task_thread_info(p)->cpu = cpu; | |
1770 | #endif | |
1771 | } | |
dce48a84 | 1772 | |
1e3c88bd | 1773 | static const struct sched_class rt_sched_class; |
dd41f596 | 1774 | |
34f971f6 | 1775 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1776 | #define for_each_class(class) \ |
1777 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1778 | |
1e3c88bd PZ |
1779 | #include "sched_stats.h" |
1780 | ||
c09595f6 | 1781 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1782 | { |
1783 | rq->nr_running++; | |
9c217245 IM |
1784 | } |
1785 | ||
c09595f6 | 1786 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1787 | { |
1788 | rq->nr_running--; | |
9c217245 IM |
1789 | } |
1790 | ||
45bf76df IM |
1791 | static void set_load_weight(struct task_struct *p) |
1792 | { | |
f05998d4 NR |
1793 | int prio = p->static_prio - MAX_RT_PRIO; |
1794 | struct load_weight *load = &p->se.load; | |
1795 | ||
dd41f596 IM |
1796 | /* |
1797 | * SCHED_IDLE tasks get minimal weight: | |
1798 | */ | |
1799 | if (p->policy == SCHED_IDLE) { | |
c8b28116 | 1800 | load->weight = scale_load(WEIGHT_IDLEPRIO); |
f05998d4 | 1801 | load->inv_weight = WMULT_IDLEPRIO; |
dd41f596 IM |
1802 | return; |
1803 | } | |
71f8bd46 | 1804 | |
c8b28116 | 1805 | load->weight = scale_load(prio_to_weight[prio]); |
f05998d4 | 1806 | load->inv_weight = prio_to_wmult[prio]; |
71f8bd46 IM |
1807 | } |
1808 | ||
371fd7e7 | 1809 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1810 | { |
a64692a3 | 1811 | update_rq_clock(rq); |
dd41f596 | 1812 | sched_info_queued(p); |
371fd7e7 | 1813 | p->sched_class->enqueue_task(rq, p, flags); |
71f8bd46 IM |
1814 | } |
1815 | ||
371fd7e7 | 1816 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1817 | { |
a64692a3 | 1818 | update_rq_clock(rq); |
46ac22ba | 1819 | sched_info_dequeued(p); |
371fd7e7 | 1820 | p->sched_class->dequeue_task(rq, p, flags); |
71f8bd46 IM |
1821 | } |
1822 | ||
1e3c88bd PZ |
1823 | /* |
1824 | * activate_task - move a task to the runqueue. | |
1825 | */ | |
371fd7e7 | 1826 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1827 | { |
1828 | if (task_contributes_to_load(p)) | |
1829 | rq->nr_uninterruptible--; | |
1830 | ||
371fd7e7 | 1831 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1832 | inc_nr_running(rq); |
1833 | } | |
1834 | ||
1835 | /* | |
1836 | * deactivate_task - remove a task from the runqueue. | |
1837 | */ | |
371fd7e7 | 1838 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1839 | { |
1840 | if (task_contributes_to_load(p)) | |
1841 | rq->nr_uninterruptible++; | |
1842 | ||
371fd7e7 | 1843 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1844 | dec_nr_running(rq); |
1845 | } | |
1846 | ||
b52bfee4 VP |
1847 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1848 | ||
305e6835 VP |
1849 | /* |
1850 | * There are no locks covering percpu hardirq/softirq time. | |
1851 | * They are only modified in account_system_vtime, on corresponding CPU | |
1852 | * with interrupts disabled. So, writes are safe. | |
1853 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1854 | * This may result in other CPU reading this CPU's irq time and can | |
1855 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1856 | * or new value with a side effect of accounting a slice of irq time to wrong |
1857 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1858 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1859 | */ |
b52bfee4 VP |
1860 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1861 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1862 | ||
1863 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1864 | static int sched_clock_irqtime; | |
1865 | ||
1866 | void enable_sched_clock_irqtime(void) | |
1867 | { | |
1868 | sched_clock_irqtime = 1; | |
1869 | } | |
1870 | ||
1871 | void disable_sched_clock_irqtime(void) | |
1872 | { | |
1873 | sched_clock_irqtime = 0; | |
1874 | } | |
1875 | ||
8e92c201 PZ |
1876 | #ifndef CONFIG_64BIT |
1877 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1878 | ||
1879 | static inline void irq_time_write_begin(void) | |
1880 | { | |
1881 | __this_cpu_inc(irq_time_seq.sequence); | |
1882 | smp_wmb(); | |
1883 | } | |
1884 | ||
1885 | static inline void irq_time_write_end(void) | |
1886 | { | |
1887 | smp_wmb(); | |
1888 | __this_cpu_inc(irq_time_seq.sequence); | |
1889 | } | |
1890 | ||
1891 | static inline u64 irq_time_read(int cpu) | |
1892 | { | |
1893 | u64 irq_time; | |
1894 | unsigned seq; | |
1895 | ||
1896 | do { | |
1897 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1898 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1899 | per_cpu(cpu_hardirq_time, cpu); | |
1900 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1901 | ||
1902 | return irq_time; | |
1903 | } | |
1904 | #else /* CONFIG_64BIT */ | |
1905 | static inline void irq_time_write_begin(void) | |
1906 | { | |
1907 | } | |
1908 | ||
1909 | static inline void irq_time_write_end(void) | |
1910 | { | |
1911 | } | |
1912 | ||
1913 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1914 | { |
305e6835 VP |
1915 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1916 | } | |
8e92c201 | 1917 | #endif /* CONFIG_64BIT */ |
305e6835 | 1918 | |
fe44d621 PZ |
1919 | /* |
1920 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1921 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1922 | */ | |
b52bfee4 VP |
1923 | void account_system_vtime(struct task_struct *curr) |
1924 | { | |
1925 | unsigned long flags; | |
fe44d621 | 1926 | s64 delta; |
b52bfee4 | 1927 | int cpu; |
b52bfee4 VP |
1928 | |
1929 | if (!sched_clock_irqtime) | |
1930 | return; | |
1931 | ||
1932 | local_irq_save(flags); | |
1933 | ||
b52bfee4 | 1934 | cpu = smp_processor_id(); |
fe44d621 PZ |
1935 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1936 | __this_cpu_add(irq_start_time, delta); | |
1937 | ||
8e92c201 | 1938 | irq_time_write_begin(); |
b52bfee4 VP |
1939 | /* |
1940 | * We do not account for softirq time from ksoftirqd here. | |
1941 | * We want to continue accounting softirq time to ksoftirqd thread | |
1942 | * in that case, so as not to confuse scheduler with a special task | |
1943 | * that do not consume any time, but still wants to run. | |
1944 | */ | |
1945 | if (hardirq_count()) | |
fe44d621 | 1946 | __this_cpu_add(cpu_hardirq_time, delta); |
4dd53d89 | 1947 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) |
fe44d621 | 1948 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1949 | |
8e92c201 | 1950 | irq_time_write_end(); |
b52bfee4 VP |
1951 | local_irq_restore(flags); |
1952 | } | |
b7dadc38 | 1953 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1954 | |
fe44d621 | 1955 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1956 | { |
fe44d621 PZ |
1957 | s64 irq_delta; |
1958 | ||
8e92c201 | 1959 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
1960 | |
1961 | /* | |
1962 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
1963 | * this case when a previous update_rq_clock() happened inside a | |
1964 | * {soft,}irq region. | |
1965 | * | |
1966 | * When this happens, we stop ->clock_task and only update the | |
1967 | * prev_irq_time stamp to account for the part that fit, so that a next | |
1968 | * update will consume the rest. This ensures ->clock_task is | |
1969 | * monotonic. | |
1970 | * | |
1971 | * It does however cause some slight miss-attribution of {soft,}irq | |
1972 | * time, a more accurate solution would be to update the irq_time using | |
1973 | * the current rq->clock timestamp, except that would require using | |
1974 | * atomic ops. | |
1975 | */ | |
1976 | if (irq_delta > delta) | |
1977 | irq_delta = delta; | |
1978 | ||
1979 | rq->prev_irq_time += irq_delta; | |
1980 | delta -= irq_delta; | |
1981 | rq->clock_task += delta; | |
1982 | ||
1983 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
1984 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
1985 | } |
1986 | ||
abb74cef VP |
1987 | static int irqtime_account_hi_update(void) |
1988 | { | |
1989 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1990 | unsigned long flags; | |
1991 | u64 latest_ns; | |
1992 | int ret = 0; | |
1993 | ||
1994 | local_irq_save(flags); | |
1995 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
1996 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->irq)) | |
1997 | ret = 1; | |
1998 | local_irq_restore(flags); | |
1999 | return ret; | |
2000 | } | |
2001 | ||
2002 | static int irqtime_account_si_update(void) | |
2003 | { | |
2004 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
2005 | unsigned long flags; | |
2006 | u64 latest_ns; | |
2007 | int ret = 0; | |
2008 | ||
2009 | local_irq_save(flags); | |
2010 | latest_ns = this_cpu_read(cpu_softirq_time); | |
2011 | if (cputime64_gt(nsecs_to_cputime64(latest_ns), cpustat->softirq)) | |
2012 | ret = 1; | |
2013 | local_irq_restore(flags); | |
2014 | return ret; | |
2015 | } | |
2016 | ||
fe44d621 | 2017 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 2018 | |
abb74cef VP |
2019 | #define sched_clock_irqtime (0) |
2020 | ||
fe44d621 | 2021 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 2022 | { |
fe44d621 | 2023 | rq->clock_task += delta; |
305e6835 VP |
2024 | } |
2025 | ||
fe44d621 | 2026 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 2027 | |
1e3c88bd PZ |
2028 | #include "sched_idletask.c" |
2029 | #include "sched_fair.c" | |
2030 | #include "sched_rt.c" | |
5091faa4 | 2031 | #include "sched_autogroup.c" |
34f971f6 | 2032 | #include "sched_stoptask.c" |
1e3c88bd PZ |
2033 | #ifdef CONFIG_SCHED_DEBUG |
2034 | # include "sched_debug.c" | |
2035 | #endif | |
2036 | ||
34f971f6 PZ |
2037 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2038 | { | |
2039 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2040 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2041 | ||
2042 | if (stop) { | |
2043 | /* | |
2044 | * Make it appear like a SCHED_FIFO task, its something | |
2045 | * userspace knows about and won't get confused about. | |
2046 | * | |
2047 | * Also, it will make PI more or less work without too | |
2048 | * much confusion -- but then, stop work should not | |
2049 | * rely on PI working anyway. | |
2050 | */ | |
2051 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2052 | ||
2053 | stop->sched_class = &stop_sched_class; | |
2054 | } | |
2055 | ||
2056 | cpu_rq(cpu)->stop = stop; | |
2057 | ||
2058 | if (old_stop) { | |
2059 | /* | |
2060 | * Reset it back to a normal scheduling class so that | |
2061 | * it can die in pieces. | |
2062 | */ | |
2063 | old_stop->sched_class = &rt_sched_class; | |
2064 | } | |
2065 | } | |
2066 | ||
14531189 | 2067 | /* |
dd41f596 | 2068 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2069 | */ |
14531189 IM |
2070 | static inline int __normal_prio(struct task_struct *p) |
2071 | { | |
dd41f596 | 2072 | return p->static_prio; |
14531189 IM |
2073 | } |
2074 | ||
b29739f9 IM |
2075 | /* |
2076 | * Calculate the expected normal priority: i.e. priority | |
2077 | * without taking RT-inheritance into account. Might be | |
2078 | * boosted by interactivity modifiers. Changes upon fork, | |
2079 | * setprio syscalls, and whenever the interactivity | |
2080 | * estimator recalculates. | |
2081 | */ | |
36c8b586 | 2082 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2083 | { |
2084 | int prio; | |
2085 | ||
e05606d3 | 2086 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2087 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2088 | else | |
2089 | prio = __normal_prio(p); | |
2090 | return prio; | |
2091 | } | |
2092 | ||
2093 | /* | |
2094 | * Calculate the current priority, i.e. the priority | |
2095 | * taken into account by the scheduler. This value might | |
2096 | * be boosted by RT tasks, or might be boosted by | |
2097 | * interactivity modifiers. Will be RT if the task got | |
2098 | * RT-boosted. If not then it returns p->normal_prio. | |
2099 | */ | |
36c8b586 | 2100 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2101 | { |
2102 | p->normal_prio = normal_prio(p); | |
2103 | /* | |
2104 | * If we are RT tasks or we were boosted to RT priority, | |
2105 | * keep the priority unchanged. Otherwise, update priority | |
2106 | * to the normal priority: | |
2107 | */ | |
2108 | if (!rt_prio(p->prio)) | |
2109 | return p->normal_prio; | |
2110 | return p->prio; | |
2111 | } | |
2112 | ||
1da177e4 LT |
2113 | /** |
2114 | * task_curr - is this task currently executing on a CPU? | |
2115 | * @p: the task in question. | |
2116 | */ | |
36c8b586 | 2117 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2118 | { |
2119 | return cpu_curr(task_cpu(p)) == p; | |
2120 | } | |
2121 | ||
cb469845 SR |
2122 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2123 | const struct sched_class *prev_class, | |
da7a735e | 2124 | int oldprio) |
cb469845 SR |
2125 | { |
2126 | if (prev_class != p->sched_class) { | |
2127 | if (prev_class->switched_from) | |
da7a735e PZ |
2128 | prev_class->switched_from(rq, p); |
2129 | p->sched_class->switched_to(rq, p); | |
2130 | } else if (oldprio != p->prio) | |
2131 | p->sched_class->prio_changed(rq, p, oldprio); | |
cb469845 SR |
2132 | } |
2133 | ||
1e5a7405 PZ |
2134 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2135 | { | |
2136 | const struct sched_class *class; | |
2137 | ||
2138 | if (p->sched_class == rq->curr->sched_class) { | |
2139 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2140 | } else { | |
2141 | for_each_class(class) { | |
2142 | if (class == rq->curr->sched_class) | |
2143 | break; | |
2144 | if (class == p->sched_class) { | |
2145 | resched_task(rq->curr); | |
2146 | break; | |
2147 | } | |
2148 | } | |
2149 | } | |
2150 | ||
2151 | /* | |
2152 | * A queue event has occurred, and we're going to schedule. In | |
2153 | * this case, we can save a useless back to back clock update. | |
2154 | */ | |
fd2f4419 | 2155 | if (rq->curr->on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2156 | rq->skip_clock_update = 1; |
2157 | } | |
2158 | ||
1da177e4 | 2159 | #ifdef CONFIG_SMP |
cc367732 IM |
2160 | /* |
2161 | * Is this task likely cache-hot: | |
2162 | */ | |
e7693a36 | 2163 | static int |
cc367732 IM |
2164 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2165 | { | |
2166 | s64 delta; | |
2167 | ||
e6c8fba7 PZ |
2168 | if (p->sched_class != &fair_sched_class) |
2169 | return 0; | |
2170 | ||
ef8002f6 NR |
2171 | if (unlikely(p->policy == SCHED_IDLE)) |
2172 | return 0; | |
2173 | ||
f540a608 IM |
2174 | /* |
2175 | * Buddy candidates are cache hot: | |
2176 | */ | |
f685ceac | 2177 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2178 | (&p->se == cfs_rq_of(&p->se)->next || |
2179 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2180 | return 1; |
2181 | ||
6bc1665b IM |
2182 | if (sysctl_sched_migration_cost == -1) |
2183 | return 1; | |
2184 | if (sysctl_sched_migration_cost == 0) | |
2185 | return 0; | |
2186 | ||
cc367732 IM |
2187 | delta = now - p->se.exec_start; |
2188 | ||
2189 | return delta < (s64)sysctl_sched_migration_cost; | |
2190 | } | |
2191 | ||
dd41f596 | 2192 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2193 | { |
e2912009 PZ |
2194 | #ifdef CONFIG_SCHED_DEBUG |
2195 | /* | |
2196 | * We should never call set_task_cpu() on a blocked task, | |
2197 | * ttwu() will sort out the placement. | |
2198 | */ | |
077614ee PZ |
2199 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2200 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
0122ec5b PZ |
2201 | |
2202 | #ifdef CONFIG_LOCKDEP | |
2203 | WARN_ON_ONCE(debug_locks && !(lockdep_is_held(&p->pi_lock) || | |
2204 | lockdep_is_held(&task_rq(p)->lock))); | |
2205 | #endif | |
e2912009 PZ |
2206 | #endif |
2207 | ||
de1d7286 | 2208 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2209 | |
0c69774e PZ |
2210 | if (task_cpu(p) != new_cpu) { |
2211 | p->se.nr_migrations++; | |
2212 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2213 | } | |
dd41f596 IM |
2214 | |
2215 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2216 | } |
2217 | ||
969c7921 | 2218 | struct migration_arg { |
36c8b586 | 2219 | struct task_struct *task; |
1da177e4 | 2220 | int dest_cpu; |
70b97a7f | 2221 | }; |
1da177e4 | 2222 | |
969c7921 TH |
2223 | static int migration_cpu_stop(void *data); |
2224 | ||
1da177e4 LT |
2225 | /* |
2226 | * wait_task_inactive - wait for a thread to unschedule. | |
2227 | * | |
85ba2d86 RM |
2228 | * If @match_state is nonzero, it's the @p->state value just checked and |
2229 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2230 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2231 | * we return a positive number (its total switch count). If a second call | |
2232 | * a short while later returns the same number, the caller can be sure that | |
2233 | * @p has remained unscheduled the whole time. | |
2234 | * | |
1da177e4 LT |
2235 | * The caller must ensure that the task *will* unschedule sometime soon, |
2236 | * else this function might spin for a *long* time. This function can't | |
2237 | * be called with interrupts off, or it may introduce deadlock with | |
2238 | * smp_call_function() if an IPI is sent by the same process we are | |
2239 | * waiting to become inactive. | |
2240 | */ | |
85ba2d86 | 2241 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2242 | { |
2243 | unsigned long flags; | |
dd41f596 | 2244 | int running, on_rq; |
85ba2d86 | 2245 | unsigned long ncsw; |
70b97a7f | 2246 | struct rq *rq; |
1da177e4 | 2247 | |
3a5c359a AK |
2248 | for (;;) { |
2249 | /* | |
2250 | * We do the initial early heuristics without holding | |
2251 | * any task-queue locks at all. We'll only try to get | |
2252 | * the runqueue lock when things look like they will | |
2253 | * work out! | |
2254 | */ | |
2255 | rq = task_rq(p); | |
fa490cfd | 2256 | |
3a5c359a AK |
2257 | /* |
2258 | * If the task is actively running on another CPU | |
2259 | * still, just relax and busy-wait without holding | |
2260 | * any locks. | |
2261 | * | |
2262 | * NOTE! Since we don't hold any locks, it's not | |
2263 | * even sure that "rq" stays as the right runqueue! | |
2264 | * But we don't care, since "task_running()" will | |
2265 | * return false if the runqueue has changed and p | |
2266 | * is actually now running somewhere else! | |
2267 | */ | |
85ba2d86 RM |
2268 | while (task_running(rq, p)) { |
2269 | if (match_state && unlikely(p->state != match_state)) | |
2270 | return 0; | |
3a5c359a | 2271 | cpu_relax(); |
85ba2d86 | 2272 | } |
fa490cfd | 2273 | |
3a5c359a AK |
2274 | /* |
2275 | * Ok, time to look more closely! We need the rq | |
2276 | * lock now, to be *sure*. If we're wrong, we'll | |
2277 | * just go back and repeat. | |
2278 | */ | |
2279 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2280 | trace_sched_wait_task(p); |
3a5c359a | 2281 | running = task_running(rq, p); |
fd2f4419 | 2282 | on_rq = p->on_rq; |
85ba2d86 | 2283 | ncsw = 0; |
f31e11d8 | 2284 | if (!match_state || p->state == match_state) |
93dcf55f | 2285 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
0122ec5b | 2286 | task_rq_unlock(rq, p, &flags); |
fa490cfd | 2287 | |
85ba2d86 RM |
2288 | /* |
2289 | * If it changed from the expected state, bail out now. | |
2290 | */ | |
2291 | if (unlikely(!ncsw)) | |
2292 | break; | |
2293 | ||
3a5c359a AK |
2294 | /* |
2295 | * Was it really running after all now that we | |
2296 | * checked with the proper locks actually held? | |
2297 | * | |
2298 | * Oops. Go back and try again.. | |
2299 | */ | |
2300 | if (unlikely(running)) { | |
2301 | cpu_relax(); | |
2302 | continue; | |
2303 | } | |
fa490cfd | 2304 | |
3a5c359a AK |
2305 | /* |
2306 | * It's not enough that it's not actively running, | |
2307 | * it must be off the runqueue _entirely_, and not | |
2308 | * preempted! | |
2309 | * | |
80dd99b3 | 2310 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2311 | * running right now), it's preempted, and we should |
2312 | * yield - it could be a while. | |
2313 | */ | |
2314 | if (unlikely(on_rq)) { | |
8eb90c30 TG |
2315 | ktime_t to = ktime_set(0, NSEC_PER_SEC/HZ); |
2316 | ||
2317 | set_current_state(TASK_UNINTERRUPTIBLE); | |
2318 | schedule_hrtimeout(&to, HRTIMER_MODE_REL); | |
3a5c359a AK |
2319 | continue; |
2320 | } | |
fa490cfd | 2321 | |
3a5c359a AK |
2322 | /* |
2323 | * Ahh, all good. It wasn't running, and it wasn't | |
2324 | * runnable, which means that it will never become | |
2325 | * running in the future either. We're all done! | |
2326 | */ | |
2327 | break; | |
2328 | } | |
85ba2d86 RM |
2329 | |
2330 | return ncsw; | |
1da177e4 LT |
2331 | } |
2332 | ||
2333 | /*** | |
2334 | * kick_process - kick a running thread to enter/exit the kernel | |
2335 | * @p: the to-be-kicked thread | |
2336 | * | |
2337 | * Cause a process which is running on another CPU to enter | |
2338 | * kernel-mode, without any delay. (to get signals handled.) | |
2339 | * | |
25985edc | 2340 | * NOTE: this function doesn't have to take the runqueue lock, |
1da177e4 LT |
2341 | * because all it wants to ensure is that the remote task enters |
2342 | * the kernel. If the IPI races and the task has been migrated | |
2343 | * to another CPU then no harm is done and the purpose has been | |
2344 | * achieved as well. | |
2345 | */ | |
36c8b586 | 2346 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2347 | { |
2348 | int cpu; | |
2349 | ||
2350 | preempt_disable(); | |
2351 | cpu = task_cpu(p); | |
2352 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2353 | smp_send_reschedule(cpu); | |
2354 | preempt_enable(); | |
2355 | } | |
b43e3521 | 2356 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2357 | #endif /* CONFIG_SMP */ |
1da177e4 | 2358 | |
970b13ba | 2359 | #ifdef CONFIG_SMP |
30da688e | 2360 | /* |
013fdb80 | 2361 | * ->cpus_allowed is protected by both rq->lock and p->pi_lock |
30da688e | 2362 | */ |
5da9a0fb PZ |
2363 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2364 | { | |
2365 | int dest_cpu; | |
2366 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2367 | ||
2368 | /* Look for allowed, online CPU in same node. */ | |
2369 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2370 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2371 | return dest_cpu; | |
2372 | ||
2373 | /* Any allowed, online CPU? */ | |
2374 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2375 | if (dest_cpu < nr_cpu_ids) | |
2376 | return dest_cpu; | |
2377 | ||
2378 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2379 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2380 | /* | |
2381 | * Don't tell them about moving exiting tasks or | |
2382 | * kernel threads (both mm NULL), since they never | |
2383 | * leave kernel. | |
2384 | */ | |
2385 | if (p->mm && printk_ratelimit()) { | |
2386 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2387 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2388 | } |
2389 | ||
2390 | return dest_cpu; | |
2391 | } | |
2392 | ||
e2912009 | 2393 | /* |
013fdb80 | 2394 | * The caller (fork, wakeup) owns p->pi_lock, ->cpus_allowed is stable. |
e2912009 | 2395 | */ |
970b13ba | 2396 | static inline |
7608dec2 | 2397 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2398 | { |
7608dec2 | 2399 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
e2912009 PZ |
2400 | |
2401 | /* | |
2402 | * In order not to call set_task_cpu() on a blocking task we need | |
2403 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2404 | * cpu. | |
2405 | * | |
2406 | * Since this is common to all placement strategies, this lives here. | |
2407 | * | |
2408 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2409 | * not worry about this generic constraint ] | |
2410 | */ | |
2411 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2412 | !cpu_online(cpu))) |
5da9a0fb | 2413 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2414 | |
2415 | return cpu; | |
970b13ba | 2416 | } |
09a40af5 MG |
2417 | |
2418 | static void update_avg(u64 *avg, u64 sample) | |
2419 | { | |
2420 | s64 diff = sample - *avg; | |
2421 | *avg += diff >> 3; | |
2422 | } | |
970b13ba PZ |
2423 | #endif |
2424 | ||
d7c01d27 | 2425 | static void |
b84cb5df | 2426 | ttwu_stat(struct task_struct *p, int cpu, int wake_flags) |
9ed3811a | 2427 | { |
d7c01d27 | 2428 | #ifdef CONFIG_SCHEDSTATS |
b84cb5df PZ |
2429 | struct rq *rq = this_rq(); |
2430 | ||
d7c01d27 PZ |
2431 | #ifdef CONFIG_SMP |
2432 | int this_cpu = smp_processor_id(); | |
2433 | ||
2434 | if (cpu == this_cpu) { | |
2435 | schedstat_inc(rq, ttwu_local); | |
2436 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2437 | } else { | |
2438 | struct sched_domain *sd; | |
2439 | ||
2440 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
057f3fad | 2441 | rcu_read_lock(); |
d7c01d27 PZ |
2442 | for_each_domain(this_cpu, sd) { |
2443 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { | |
2444 | schedstat_inc(sd, ttwu_wake_remote); | |
2445 | break; | |
2446 | } | |
2447 | } | |
057f3fad | 2448 | rcu_read_unlock(); |
d7c01d27 PZ |
2449 | } |
2450 | #endif /* CONFIG_SMP */ | |
2451 | ||
2452 | schedstat_inc(rq, ttwu_count); | |
9ed3811a | 2453 | schedstat_inc(p, se.statistics.nr_wakeups); |
d7c01d27 PZ |
2454 | |
2455 | if (wake_flags & WF_SYNC) | |
9ed3811a | 2456 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
d7c01d27 PZ |
2457 | |
2458 | if (cpu != task_cpu(p)) | |
9ed3811a | 2459 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
9ed3811a | 2460 | |
d7c01d27 PZ |
2461 | #endif /* CONFIG_SCHEDSTATS */ |
2462 | } | |
2463 | ||
2464 | static void ttwu_activate(struct rq *rq, struct task_struct *p, int en_flags) | |
2465 | { | |
9ed3811a | 2466 | activate_task(rq, p, en_flags); |
fd2f4419 | 2467 | p->on_rq = 1; |
c2f7115e PZ |
2468 | |
2469 | /* if a worker is waking up, notify workqueue */ | |
2470 | if (p->flags & PF_WQ_WORKER) | |
2471 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2472 | } |
2473 | ||
23f41eeb PZ |
2474 | /* |
2475 | * Mark the task runnable and perform wakeup-preemption. | |
2476 | */ | |
89363381 | 2477 | static void |
23f41eeb | 2478 | ttwu_do_wakeup(struct rq *rq, struct task_struct *p, int wake_flags) |
9ed3811a | 2479 | { |
89363381 | 2480 | trace_sched_wakeup(p, true); |
9ed3811a TH |
2481 | check_preempt_curr(rq, p, wake_flags); |
2482 | ||
2483 | p->state = TASK_RUNNING; | |
2484 | #ifdef CONFIG_SMP | |
2485 | if (p->sched_class->task_woken) | |
2486 | p->sched_class->task_woken(rq, p); | |
2487 | ||
2488 | if (unlikely(rq->idle_stamp)) { | |
2489 | u64 delta = rq->clock - rq->idle_stamp; | |
2490 | u64 max = 2*sysctl_sched_migration_cost; | |
2491 | ||
2492 | if (delta > max) | |
2493 | rq->avg_idle = max; | |
2494 | else | |
2495 | update_avg(&rq->avg_idle, delta); | |
2496 | rq->idle_stamp = 0; | |
2497 | } | |
2498 | #endif | |
2499 | } | |
2500 | ||
c05fbafb PZ |
2501 | static void |
2502 | ttwu_do_activate(struct rq *rq, struct task_struct *p, int wake_flags) | |
2503 | { | |
2504 | #ifdef CONFIG_SMP | |
2505 | if (p->sched_contributes_to_load) | |
2506 | rq->nr_uninterruptible--; | |
2507 | #endif | |
2508 | ||
2509 | ttwu_activate(rq, p, ENQUEUE_WAKEUP | ENQUEUE_WAKING); | |
2510 | ttwu_do_wakeup(rq, p, wake_flags); | |
2511 | } | |
2512 | ||
2513 | /* | |
2514 | * Called in case the task @p isn't fully descheduled from its runqueue, | |
2515 | * in this case we must do a remote wakeup. Its a 'light' wakeup though, | |
2516 | * since all we need to do is flip p->state to TASK_RUNNING, since | |
2517 | * the task is still ->on_rq. | |
2518 | */ | |
2519 | static int ttwu_remote(struct task_struct *p, int wake_flags) | |
2520 | { | |
2521 | struct rq *rq; | |
2522 | int ret = 0; | |
2523 | ||
2524 | rq = __task_rq_lock(p); | |
2525 | if (p->on_rq) { | |
2526 | ttwu_do_wakeup(rq, p, wake_flags); | |
2527 | ret = 1; | |
2528 | } | |
2529 | __task_rq_unlock(rq); | |
2530 | ||
2531 | return ret; | |
2532 | } | |
2533 | ||
317f3941 PZ |
2534 | #ifdef CONFIG_SMP |
2535 | static void sched_ttwu_pending(void) | |
2536 | { | |
2537 | struct rq *rq = this_rq(); | |
2538 | struct task_struct *list = xchg(&rq->wake_list, NULL); | |
2539 | ||
2540 | if (!list) | |
2541 | return; | |
2542 | ||
2543 | raw_spin_lock(&rq->lock); | |
2544 | ||
2545 | while (list) { | |
2546 | struct task_struct *p = list; | |
2547 | list = list->wake_entry; | |
2548 | ttwu_do_activate(rq, p, 0); | |
2549 | } | |
2550 | ||
2551 | raw_spin_unlock(&rq->lock); | |
2552 | } | |
2553 | ||
2554 | void scheduler_ipi(void) | |
2555 | { | |
2556 | sched_ttwu_pending(); | |
2557 | } | |
2558 | ||
2559 | static void ttwu_queue_remote(struct task_struct *p, int cpu) | |
2560 | { | |
2561 | struct rq *rq = cpu_rq(cpu); | |
2562 | struct task_struct *next = rq->wake_list; | |
2563 | ||
2564 | for (;;) { | |
2565 | struct task_struct *old = next; | |
2566 | ||
2567 | p->wake_entry = next; | |
2568 | next = cmpxchg(&rq->wake_list, old, p); | |
2569 | if (next == old) | |
2570 | break; | |
2571 | } | |
2572 | ||
2573 | if (!next) | |
2574 | smp_send_reschedule(cpu); | |
2575 | } | |
d6aa8f85 PZ |
2576 | |
2577 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
2578 | static int ttwu_activate_remote(struct task_struct *p, int wake_flags) | |
2579 | { | |
2580 | struct rq *rq; | |
2581 | int ret = 0; | |
2582 | ||
2583 | rq = __task_rq_lock(p); | |
2584 | if (p->on_cpu) { | |
2585 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); | |
2586 | ttwu_do_wakeup(rq, p, wake_flags); | |
2587 | ret = 1; | |
2588 | } | |
2589 | __task_rq_unlock(rq); | |
2590 | ||
2591 | return ret; | |
2592 | ||
2593 | } | |
2594 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
2595 | #endif /* CONFIG_SMP */ | |
317f3941 | 2596 | |
c05fbafb PZ |
2597 | static void ttwu_queue(struct task_struct *p, int cpu) |
2598 | { | |
2599 | struct rq *rq = cpu_rq(cpu); | |
2600 | ||
17d9f311 | 2601 | #if defined(CONFIG_SMP) |
317f3941 | 2602 | if (sched_feat(TTWU_QUEUE) && cpu != smp_processor_id()) { |
f01114cb | 2603 | sched_clock_cpu(cpu); /* sync clocks x-cpu */ |
317f3941 PZ |
2604 | ttwu_queue_remote(p, cpu); |
2605 | return; | |
2606 | } | |
2607 | #endif | |
2608 | ||
c05fbafb PZ |
2609 | raw_spin_lock(&rq->lock); |
2610 | ttwu_do_activate(rq, p, 0); | |
2611 | raw_spin_unlock(&rq->lock); | |
9ed3811a TH |
2612 | } |
2613 | ||
2614 | /** | |
1da177e4 | 2615 | * try_to_wake_up - wake up a thread |
9ed3811a | 2616 | * @p: the thread to be awakened |
1da177e4 | 2617 | * @state: the mask of task states that can be woken |
9ed3811a | 2618 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2619 | * |
2620 | * Put it on the run-queue if it's not already there. The "current" | |
2621 | * thread is always on the run-queue (except when the actual | |
2622 | * re-schedule is in progress), and as such you're allowed to do | |
2623 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2624 | * runnable without the overhead of this. | |
2625 | * | |
9ed3811a TH |
2626 | * Returns %true if @p was woken up, %false if it was already running |
2627 | * or @state didn't match @p's state. | |
1da177e4 | 2628 | */ |
e4a52bcb PZ |
2629 | static int |
2630 | try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags) | |
1da177e4 | 2631 | { |
1da177e4 | 2632 | unsigned long flags; |
c05fbafb | 2633 | int cpu, success = 0; |
2398f2c6 | 2634 | |
04e2f174 | 2635 | smp_wmb(); |
013fdb80 | 2636 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
e9c84311 | 2637 | if (!(p->state & state)) |
1da177e4 LT |
2638 | goto out; |
2639 | ||
c05fbafb | 2640 | success = 1; /* we're going to change ->state */ |
1da177e4 | 2641 | cpu = task_cpu(p); |
1da177e4 | 2642 | |
c05fbafb PZ |
2643 | if (p->on_rq && ttwu_remote(p, wake_flags)) |
2644 | goto stat; | |
1da177e4 | 2645 | |
1da177e4 | 2646 | #ifdef CONFIG_SMP |
e9c84311 | 2647 | /* |
c05fbafb PZ |
2648 | * If the owning (remote) cpu is still in the middle of schedule() with |
2649 | * this task as prev, wait until its done referencing the task. | |
e9c84311 | 2650 | */ |
e4a52bcb PZ |
2651 | while (p->on_cpu) { |
2652 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
2653 | /* | |
d6aa8f85 PZ |
2654 | * In case the architecture enables interrupts in |
2655 | * context_switch(), we cannot busy wait, since that | |
2656 | * would lead to deadlocks when an interrupt hits and | |
2657 | * tries to wake up @prev. So bail and do a complete | |
2658 | * remote wakeup. | |
e4a52bcb | 2659 | */ |
d6aa8f85 | 2660 | if (ttwu_activate_remote(p, wake_flags)) |
c05fbafb | 2661 | goto stat; |
d6aa8f85 | 2662 | #else |
e4a52bcb | 2663 | cpu_relax(); |
d6aa8f85 | 2664 | #endif |
371fd7e7 | 2665 | } |
0970d299 | 2666 | /* |
e4a52bcb | 2667 | * Pairs with the smp_wmb() in finish_lock_switch(). |
0970d299 | 2668 | */ |
e4a52bcb | 2669 | smp_rmb(); |
1da177e4 | 2670 | |
a8e4f2ea | 2671 | p->sched_contributes_to_load = !!task_contributes_to_load(p); |
e9c84311 | 2672 | p->state = TASK_WAKING; |
e7693a36 | 2673 | |
e4a52bcb | 2674 | if (p->sched_class->task_waking) |
74f8e4b2 | 2675 | p->sched_class->task_waking(p); |
efbbd05a | 2676 | |
7608dec2 | 2677 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
c05fbafb | 2678 | if (task_cpu(p) != cpu) |
e4a52bcb | 2679 | set_task_cpu(p, cpu); |
1da177e4 | 2680 | #endif /* CONFIG_SMP */ |
1da177e4 | 2681 | |
c05fbafb PZ |
2682 | ttwu_queue(p, cpu); |
2683 | stat: | |
b84cb5df | 2684 | ttwu_stat(p, cpu, wake_flags); |
1da177e4 | 2685 | out: |
013fdb80 | 2686 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
2687 | |
2688 | return success; | |
2689 | } | |
2690 | ||
21aa9af0 TH |
2691 | /** |
2692 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2693 | * @p: the thread to be awakened | |
2694 | * | |
2acca55e | 2695 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 | 2696 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2acca55e | 2697 | * the current task. |
21aa9af0 TH |
2698 | */ |
2699 | static void try_to_wake_up_local(struct task_struct *p) | |
2700 | { | |
2701 | struct rq *rq = task_rq(p); | |
21aa9af0 TH |
2702 | |
2703 | BUG_ON(rq != this_rq()); | |
2704 | BUG_ON(p == current); | |
2705 | lockdep_assert_held(&rq->lock); | |
2706 | ||
2acca55e PZ |
2707 | if (!raw_spin_trylock(&p->pi_lock)) { |
2708 | raw_spin_unlock(&rq->lock); | |
2709 | raw_spin_lock(&p->pi_lock); | |
2710 | raw_spin_lock(&rq->lock); | |
2711 | } | |
2712 | ||
21aa9af0 | 2713 | if (!(p->state & TASK_NORMAL)) |
2acca55e | 2714 | goto out; |
21aa9af0 | 2715 | |
fd2f4419 | 2716 | if (!p->on_rq) |
d7c01d27 PZ |
2717 | ttwu_activate(rq, p, ENQUEUE_WAKEUP); |
2718 | ||
23f41eeb | 2719 | ttwu_do_wakeup(rq, p, 0); |
b84cb5df | 2720 | ttwu_stat(p, smp_processor_id(), 0); |
2acca55e PZ |
2721 | out: |
2722 | raw_spin_unlock(&p->pi_lock); | |
21aa9af0 TH |
2723 | } |
2724 | ||
50fa610a DH |
2725 | /** |
2726 | * wake_up_process - Wake up a specific process | |
2727 | * @p: The process to be woken up. | |
2728 | * | |
2729 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2730 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2731 | * running. | |
2732 | * | |
2733 | * It may be assumed that this function implies a write memory barrier before | |
2734 | * changing the task state if and only if any tasks are woken up. | |
2735 | */ | |
7ad5b3a5 | 2736 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2737 | { |
d9514f6c | 2738 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2739 | } |
1da177e4 LT |
2740 | EXPORT_SYMBOL(wake_up_process); |
2741 | ||
7ad5b3a5 | 2742 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2743 | { |
2744 | return try_to_wake_up(p, state, 0); | |
2745 | } | |
2746 | ||
1da177e4 LT |
2747 | /* |
2748 | * Perform scheduler related setup for a newly forked process p. | |
2749 | * p is forked by current. | |
dd41f596 IM |
2750 | * |
2751 | * __sched_fork() is basic setup used by init_idle() too: | |
2752 | */ | |
2753 | static void __sched_fork(struct task_struct *p) | |
2754 | { | |
fd2f4419 PZ |
2755 | p->on_rq = 0; |
2756 | ||
2757 | p->se.on_rq = 0; | |
dd41f596 IM |
2758 | p->se.exec_start = 0; |
2759 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2760 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2761 | p->se.nr_migrations = 0; |
da7a735e | 2762 | p->se.vruntime = 0; |
fd2f4419 | 2763 | INIT_LIST_HEAD(&p->se.group_node); |
6cfb0d5d IM |
2764 | |
2765 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2766 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2767 | #endif |
476d139c | 2768 | |
fa717060 | 2769 | INIT_LIST_HEAD(&p->rt.run_list); |
476d139c | 2770 | |
e107be36 AK |
2771 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2772 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2773 | #endif | |
dd41f596 IM |
2774 | } |
2775 | ||
2776 | /* | |
2777 | * fork()/clone()-time setup: | |
2778 | */ | |
3e51e3ed | 2779 | void sched_fork(struct task_struct *p) |
dd41f596 | 2780 | { |
0122ec5b | 2781 | unsigned long flags; |
dd41f596 IM |
2782 | int cpu = get_cpu(); |
2783 | ||
2784 | __sched_fork(p); | |
06b83b5f | 2785 | /* |
0017d735 | 2786 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2787 | * nobody will actually run it, and a signal or other external |
2788 | * event cannot wake it up and insert it on the runqueue either. | |
2789 | */ | |
0017d735 | 2790 | p->state = TASK_RUNNING; |
dd41f596 | 2791 | |
b9dc29e7 MG |
2792 | /* |
2793 | * Revert to default priority/policy on fork if requested. | |
2794 | */ | |
2795 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2796 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2797 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2798 | p->normal_prio = p->static_prio; |
2799 | } | |
b9dc29e7 | 2800 | |
6c697bdf MG |
2801 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2802 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2803 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2804 | set_load_weight(p); |
2805 | } | |
2806 | ||
b9dc29e7 MG |
2807 | /* |
2808 | * We don't need the reset flag anymore after the fork. It has | |
2809 | * fulfilled its duty: | |
2810 | */ | |
2811 | p->sched_reset_on_fork = 0; | |
2812 | } | |
ca94c442 | 2813 | |
f83f9ac2 PW |
2814 | /* |
2815 | * Make sure we do not leak PI boosting priority to the child. | |
2816 | */ | |
2817 | p->prio = current->normal_prio; | |
2818 | ||
2ddbf952 HS |
2819 | if (!rt_prio(p->prio)) |
2820 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2821 | |
cd29fe6f PZ |
2822 | if (p->sched_class->task_fork) |
2823 | p->sched_class->task_fork(p); | |
2824 | ||
86951599 PZ |
2825 | /* |
2826 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2827 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2828 | * is ran before sched_fork(). | |
2829 | * | |
2830 | * Silence PROVE_RCU. | |
2831 | */ | |
0122ec5b | 2832 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
5f3edc1b | 2833 | set_task_cpu(p, cpu); |
0122ec5b | 2834 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
5f3edc1b | 2835 | |
52f17b6c | 2836 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2837 | if (likely(sched_info_on())) |
52f17b6c | 2838 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2839 | #endif |
3ca7a440 PZ |
2840 | #if defined(CONFIG_SMP) |
2841 | p->on_cpu = 0; | |
4866cde0 | 2842 | #endif |
1da177e4 | 2843 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2844 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2845 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2846 | #endif |
806c09a7 | 2847 | #ifdef CONFIG_SMP |
917b627d | 2848 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2849 | #endif |
917b627d | 2850 | |
476d139c | 2851 | put_cpu(); |
1da177e4 LT |
2852 | } |
2853 | ||
2854 | /* | |
2855 | * wake_up_new_task - wake up a newly created task for the first time. | |
2856 | * | |
2857 | * This function will do some initial scheduler statistics housekeeping | |
2858 | * that must be done for every newly created context, then puts the task | |
2859 | * on the runqueue and wakes it. | |
2860 | */ | |
3e51e3ed | 2861 | void wake_up_new_task(struct task_struct *p) |
1da177e4 LT |
2862 | { |
2863 | unsigned long flags; | |
dd41f596 | 2864 | struct rq *rq; |
fabf318e | 2865 | |
ab2515c4 | 2866 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
fabf318e PZ |
2867 | #ifdef CONFIG_SMP |
2868 | /* | |
2869 | * Fork balancing, do it here and not earlier because: | |
2870 | * - cpus_allowed can change in the fork path | |
2871 | * - any previously selected cpu might disappear through hotplug | |
fabf318e | 2872 | */ |
ab2515c4 | 2873 | set_task_cpu(p, select_task_rq(p, SD_BALANCE_FORK, 0)); |
0017d735 PZ |
2874 | #endif |
2875 | ||
ab2515c4 | 2876 | rq = __task_rq_lock(p); |
cd29fe6f | 2877 | activate_task(rq, p, 0); |
fd2f4419 | 2878 | p->on_rq = 1; |
89363381 | 2879 | trace_sched_wakeup_new(p, true); |
a7558e01 | 2880 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2881 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2882 | if (p->sched_class->task_woken) |
2883 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2884 | #endif |
0122ec5b | 2885 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
2886 | } |
2887 | ||
e107be36 AK |
2888 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2889 | ||
2890 | /** | |
80dd99b3 | 2891 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2892 | * @notifier: notifier struct to register |
e107be36 AK |
2893 | */ |
2894 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2895 | { | |
2896 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2897 | } | |
2898 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2899 | ||
2900 | /** | |
2901 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2902 | * @notifier: notifier struct to unregister |
e107be36 AK |
2903 | * |
2904 | * This is safe to call from within a preemption notifier. | |
2905 | */ | |
2906 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2907 | { | |
2908 | hlist_del(¬ifier->link); | |
2909 | } | |
2910 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2911 | ||
2912 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2913 | { | |
2914 | struct preempt_notifier *notifier; | |
2915 | struct hlist_node *node; | |
2916 | ||
2917 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2918 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2919 | } | |
2920 | ||
2921 | static void | |
2922 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2923 | struct task_struct *next) | |
2924 | { | |
2925 | struct preempt_notifier *notifier; | |
2926 | struct hlist_node *node; | |
2927 | ||
2928 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2929 | notifier->ops->sched_out(notifier, next); | |
2930 | } | |
2931 | ||
6d6bc0ad | 2932 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2933 | |
2934 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2935 | { | |
2936 | } | |
2937 | ||
2938 | static void | |
2939 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2940 | struct task_struct *next) | |
2941 | { | |
2942 | } | |
2943 | ||
6d6bc0ad | 2944 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2945 | |
4866cde0 NP |
2946 | /** |
2947 | * prepare_task_switch - prepare to switch tasks | |
2948 | * @rq: the runqueue preparing to switch | |
421cee29 | 2949 | * @prev: the current task that is being switched out |
4866cde0 NP |
2950 | * @next: the task we are going to switch to. |
2951 | * | |
2952 | * This is called with the rq lock held and interrupts off. It must | |
2953 | * be paired with a subsequent finish_task_switch after the context | |
2954 | * switch. | |
2955 | * | |
2956 | * prepare_task_switch sets up locking and calls architecture specific | |
2957 | * hooks. | |
2958 | */ | |
e107be36 AK |
2959 | static inline void |
2960 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2961 | struct task_struct *next) | |
4866cde0 | 2962 | { |
fe4b04fa PZ |
2963 | sched_info_switch(prev, next); |
2964 | perf_event_task_sched_out(prev, next); | |
e107be36 | 2965 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2966 | prepare_lock_switch(rq, next); |
2967 | prepare_arch_switch(next); | |
fe4b04fa | 2968 | trace_sched_switch(prev, next); |
4866cde0 NP |
2969 | } |
2970 | ||
1da177e4 LT |
2971 | /** |
2972 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2973 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2974 | * @prev: the thread we just switched away from. |
2975 | * | |
4866cde0 NP |
2976 | * finish_task_switch must be called after the context switch, paired |
2977 | * with a prepare_task_switch call before the context switch. | |
2978 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2979 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2980 | * |
2981 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2982 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2983 | * with the lock held can cause deadlocks; see schedule() for |
2984 | * details.) | |
2985 | */ | |
a9957449 | 2986 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2987 | __releases(rq->lock) |
2988 | { | |
1da177e4 | 2989 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2990 | long prev_state; |
1da177e4 LT |
2991 | |
2992 | rq->prev_mm = NULL; | |
2993 | ||
2994 | /* | |
2995 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2996 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2997 | * schedule one last time. The schedule call will never return, and |
2998 | * the scheduled task must drop that reference. | |
c394cc9f | 2999 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
3000 | * still held, otherwise prev could be scheduled on another cpu, die |
3001 | * there before we look at prev->state, and then the reference would | |
3002 | * be dropped twice. | |
3003 | * Manfred Spraul <manfred@colorfullife.com> | |
3004 | */ | |
55a101f8 | 3005 | prev_state = prev->state; |
4866cde0 | 3006 | finish_arch_switch(prev); |
8381f65d JI |
3007 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
3008 | local_irq_disable(); | |
3009 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 3010 | perf_event_task_sched_in(current); |
8381f65d JI |
3011 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
3012 | local_irq_enable(); | |
3013 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 3014 | finish_lock_switch(rq, prev); |
e8fa1362 | 3015 | |
e107be36 | 3016 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
3017 | if (mm) |
3018 | mmdrop(mm); | |
c394cc9f | 3019 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 3020 | /* |
3021 | * Remove function-return probe instances associated with this | |
3022 | * task and put them back on the free list. | |
9761eea8 | 3023 | */ |
c6fd91f0 | 3024 | kprobe_flush_task(prev); |
1da177e4 | 3025 | put_task_struct(prev); |
c6fd91f0 | 3026 | } |
1da177e4 LT |
3027 | } |
3028 | ||
3f029d3c GH |
3029 | #ifdef CONFIG_SMP |
3030 | ||
3031 | /* assumes rq->lock is held */ | |
3032 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
3033 | { | |
3034 | if (prev->sched_class->pre_schedule) | |
3035 | prev->sched_class->pre_schedule(rq, prev); | |
3036 | } | |
3037 | ||
3038 | /* rq->lock is NOT held, but preemption is disabled */ | |
3039 | static inline void post_schedule(struct rq *rq) | |
3040 | { | |
3041 | if (rq->post_schedule) { | |
3042 | unsigned long flags; | |
3043 | ||
05fa785c | 3044 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
3045 | if (rq->curr->sched_class->post_schedule) |
3046 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 3047 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
3048 | |
3049 | rq->post_schedule = 0; | |
3050 | } | |
3051 | } | |
3052 | ||
3053 | #else | |
da19ab51 | 3054 | |
3f029d3c GH |
3055 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
3056 | { | |
3057 | } | |
3058 | ||
3059 | static inline void post_schedule(struct rq *rq) | |
3060 | { | |
1da177e4 LT |
3061 | } |
3062 | ||
3f029d3c GH |
3063 | #endif |
3064 | ||
1da177e4 LT |
3065 | /** |
3066 | * schedule_tail - first thing a freshly forked thread must call. | |
3067 | * @prev: the thread we just switched away from. | |
3068 | */ | |
36c8b586 | 3069 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
3070 | __releases(rq->lock) |
3071 | { | |
70b97a7f IM |
3072 | struct rq *rq = this_rq(); |
3073 | ||
4866cde0 | 3074 | finish_task_switch(rq, prev); |
da19ab51 | 3075 | |
3f029d3c GH |
3076 | /* |
3077 | * FIXME: do we need to worry about rq being invalidated by the | |
3078 | * task_switch? | |
3079 | */ | |
3080 | post_schedule(rq); | |
70b97a7f | 3081 | |
4866cde0 NP |
3082 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
3083 | /* In this case, finish_task_switch does not reenable preemption */ | |
3084 | preempt_enable(); | |
3085 | #endif | |
1da177e4 | 3086 | if (current->set_child_tid) |
b488893a | 3087 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
3088 | } |
3089 | ||
3090 | /* | |
3091 | * context_switch - switch to the new MM and the new | |
3092 | * thread's register state. | |
3093 | */ | |
dd41f596 | 3094 | static inline void |
70b97a7f | 3095 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 3096 | struct task_struct *next) |
1da177e4 | 3097 | { |
dd41f596 | 3098 | struct mm_struct *mm, *oldmm; |
1da177e4 | 3099 | |
e107be36 | 3100 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 3101 | |
dd41f596 IM |
3102 | mm = next->mm; |
3103 | oldmm = prev->active_mm; | |
9226d125 ZA |
3104 | /* |
3105 | * For paravirt, this is coupled with an exit in switch_to to | |
3106 | * combine the page table reload and the switch backend into | |
3107 | * one hypercall. | |
3108 | */ | |
224101ed | 3109 | arch_start_context_switch(prev); |
9226d125 | 3110 | |
31915ab4 | 3111 | if (!mm) { |
1da177e4 LT |
3112 | next->active_mm = oldmm; |
3113 | atomic_inc(&oldmm->mm_count); | |
3114 | enter_lazy_tlb(oldmm, next); | |
3115 | } else | |
3116 | switch_mm(oldmm, mm, next); | |
3117 | ||
31915ab4 | 3118 | if (!prev->mm) { |
1da177e4 | 3119 | prev->active_mm = NULL; |
1da177e4 LT |
3120 | rq->prev_mm = oldmm; |
3121 | } | |
3a5f5e48 IM |
3122 | /* |
3123 | * Since the runqueue lock will be released by the next | |
3124 | * task (which is an invalid locking op but in the case | |
3125 | * of the scheduler it's an obvious special-case), so we | |
3126 | * do an early lockdep release here: | |
3127 | */ | |
3128 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3129 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3130 | #endif |
1da177e4 LT |
3131 | |
3132 | /* Here we just switch the register state and the stack. */ | |
3133 | switch_to(prev, next, prev); | |
3134 | ||
dd41f596 IM |
3135 | barrier(); |
3136 | /* | |
3137 | * this_rq must be evaluated again because prev may have moved | |
3138 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3139 | * frame will be invalid. | |
3140 | */ | |
3141 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3142 | } |
3143 | ||
3144 | /* | |
3145 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3146 | * | |
3147 | * externally visible scheduler statistics: current number of runnable | |
3148 | * threads, current number of uninterruptible-sleeping threads, total | |
3149 | * number of context switches performed since bootup. | |
3150 | */ | |
3151 | unsigned long nr_running(void) | |
3152 | { | |
3153 | unsigned long i, sum = 0; | |
3154 | ||
3155 | for_each_online_cpu(i) | |
3156 | sum += cpu_rq(i)->nr_running; | |
3157 | ||
3158 | return sum; | |
f711f609 | 3159 | } |
1da177e4 LT |
3160 | |
3161 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3162 | { |
1da177e4 | 3163 | unsigned long i, sum = 0; |
f711f609 | 3164 | |
0a945022 | 3165 | for_each_possible_cpu(i) |
1da177e4 | 3166 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3167 | |
3168 | /* | |
1da177e4 LT |
3169 | * Since we read the counters lockless, it might be slightly |
3170 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3171 | */ |
1da177e4 LT |
3172 | if (unlikely((long)sum < 0)) |
3173 | sum = 0; | |
f711f609 | 3174 | |
1da177e4 | 3175 | return sum; |
f711f609 | 3176 | } |
f711f609 | 3177 | |
1da177e4 | 3178 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3179 | { |
cc94abfc SR |
3180 | int i; |
3181 | unsigned long long sum = 0; | |
46cb4b7c | 3182 | |
0a945022 | 3183 | for_each_possible_cpu(i) |
1da177e4 | 3184 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3185 | |
1da177e4 LT |
3186 | return sum; |
3187 | } | |
483b4ee6 | 3188 | |
1da177e4 LT |
3189 | unsigned long nr_iowait(void) |
3190 | { | |
3191 | unsigned long i, sum = 0; | |
483b4ee6 | 3192 | |
0a945022 | 3193 | for_each_possible_cpu(i) |
1da177e4 | 3194 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3195 | |
1da177e4 LT |
3196 | return sum; |
3197 | } | |
483b4ee6 | 3198 | |
8c215bd3 | 3199 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3200 | { |
8c215bd3 | 3201 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3202 | return atomic_read(&this->nr_iowait); |
3203 | } | |
46cb4b7c | 3204 | |
69d25870 AV |
3205 | unsigned long this_cpu_load(void) |
3206 | { | |
3207 | struct rq *this = this_rq(); | |
3208 | return this->cpu_load[0]; | |
3209 | } | |
e790fb0b | 3210 | |
46cb4b7c | 3211 | |
dce48a84 TG |
3212 | /* Variables and functions for calc_load */ |
3213 | static atomic_long_t calc_load_tasks; | |
3214 | static unsigned long calc_load_update; | |
3215 | unsigned long avenrun[3]; | |
3216 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3217 | |
74f5187a PZ |
3218 | static long calc_load_fold_active(struct rq *this_rq) |
3219 | { | |
3220 | long nr_active, delta = 0; | |
3221 | ||
3222 | nr_active = this_rq->nr_running; | |
3223 | nr_active += (long) this_rq->nr_uninterruptible; | |
3224 | ||
3225 | if (nr_active != this_rq->calc_load_active) { | |
3226 | delta = nr_active - this_rq->calc_load_active; | |
3227 | this_rq->calc_load_active = nr_active; | |
3228 | } | |
3229 | ||
3230 | return delta; | |
3231 | } | |
3232 | ||
0f004f5a PZ |
3233 | static unsigned long |
3234 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3235 | { | |
3236 | load *= exp; | |
3237 | load += active * (FIXED_1 - exp); | |
3238 | load += 1UL << (FSHIFT - 1); | |
3239 | return load >> FSHIFT; | |
3240 | } | |
3241 | ||
74f5187a PZ |
3242 | #ifdef CONFIG_NO_HZ |
3243 | /* | |
3244 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3245 | * | |
3246 | * When making the ILB scale, we should try to pull this in as well. | |
3247 | */ | |
3248 | static atomic_long_t calc_load_tasks_idle; | |
3249 | ||
3250 | static void calc_load_account_idle(struct rq *this_rq) | |
3251 | { | |
3252 | long delta; | |
3253 | ||
3254 | delta = calc_load_fold_active(this_rq); | |
3255 | if (delta) | |
3256 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3257 | } | |
3258 | ||
3259 | static long calc_load_fold_idle(void) | |
3260 | { | |
3261 | long delta = 0; | |
3262 | ||
3263 | /* | |
3264 | * Its got a race, we don't care... | |
3265 | */ | |
3266 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3267 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3268 | ||
3269 | return delta; | |
3270 | } | |
0f004f5a PZ |
3271 | |
3272 | /** | |
3273 | * fixed_power_int - compute: x^n, in O(log n) time | |
3274 | * | |
3275 | * @x: base of the power | |
3276 | * @frac_bits: fractional bits of @x | |
3277 | * @n: power to raise @x to. | |
3278 | * | |
3279 | * By exploiting the relation between the definition of the natural power | |
3280 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3281 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3282 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3283 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3284 | * of course trivially computable in O(log_2 n), the length of our binary | |
3285 | * vector. | |
3286 | */ | |
3287 | static unsigned long | |
3288 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3289 | { | |
3290 | unsigned long result = 1UL << frac_bits; | |
3291 | ||
3292 | if (n) for (;;) { | |
3293 | if (n & 1) { | |
3294 | result *= x; | |
3295 | result += 1UL << (frac_bits - 1); | |
3296 | result >>= frac_bits; | |
3297 | } | |
3298 | n >>= 1; | |
3299 | if (!n) | |
3300 | break; | |
3301 | x *= x; | |
3302 | x += 1UL << (frac_bits - 1); | |
3303 | x >>= frac_bits; | |
3304 | } | |
3305 | ||
3306 | return result; | |
3307 | } | |
3308 | ||
3309 | /* | |
3310 | * a1 = a0 * e + a * (1 - e) | |
3311 | * | |
3312 | * a2 = a1 * e + a * (1 - e) | |
3313 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3314 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3315 | * | |
3316 | * a3 = a2 * e + a * (1 - e) | |
3317 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3318 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3319 | * | |
3320 | * ... | |
3321 | * | |
3322 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3323 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3324 | * = a0 * e^n + a * (1 - e^n) | |
3325 | * | |
3326 | * [1] application of the geometric series: | |
3327 | * | |
3328 | * n 1 - x^(n+1) | |
3329 | * S_n := \Sum x^i = ------------- | |
3330 | * i=0 1 - x | |
3331 | */ | |
3332 | static unsigned long | |
3333 | calc_load_n(unsigned long load, unsigned long exp, | |
3334 | unsigned long active, unsigned int n) | |
3335 | { | |
3336 | ||
3337 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3338 | } | |
3339 | ||
3340 | /* | |
3341 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3342 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3343 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3344 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3345 | * | |
3346 | * Once we've updated the global active value, we need to apply the exponential | |
3347 | * weights adjusted to the number of cycles missed. | |
3348 | */ | |
3349 | static void calc_global_nohz(unsigned long ticks) | |
3350 | { | |
3351 | long delta, active, n; | |
3352 | ||
3353 | if (time_before(jiffies, calc_load_update)) | |
3354 | return; | |
3355 | ||
3356 | /* | |
3357 | * If we crossed a calc_load_update boundary, make sure to fold | |
3358 | * any pending idle changes, the respective CPUs might have | |
3359 | * missed the tick driven calc_load_account_active() update | |
3360 | * due to NO_HZ. | |
3361 | */ | |
3362 | delta = calc_load_fold_idle(); | |
3363 | if (delta) | |
3364 | atomic_long_add(delta, &calc_load_tasks); | |
3365 | ||
3366 | /* | |
3367 | * If we were idle for multiple load cycles, apply them. | |
3368 | */ | |
3369 | if (ticks >= LOAD_FREQ) { | |
3370 | n = ticks / LOAD_FREQ; | |
3371 | ||
3372 | active = atomic_long_read(&calc_load_tasks); | |
3373 | active = active > 0 ? active * FIXED_1 : 0; | |
3374 | ||
3375 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3376 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3377 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3378 | ||
3379 | calc_load_update += n * LOAD_FREQ; | |
3380 | } | |
3381 | ||
3382 | /* | |
3383 | * Its possible the remainder of the above division also crosses | |
3384 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3385 | * which comes after this will take care of that. | |
3386 | * | |
3387 | * Consider us being 11 ticks before a cycle completion, and us | |
3388 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3389 | * age us 4 cycles, and the test in calc_global_load() will | |
3390 | * pick up the final one. | |
3391 | */ | |
3392 | } | |
74f5187a PZ |
3393 | #else |
3394 | static void calc_load_account_idle(struct rq *this_rq) | |
3395 | { | |
3396 | } | |
3397 | ||
3398 | static inline long calc_load_fold_idle(void) | |
3399 | { | |
3400 | return 0; | |
3401 | } | |
0f004f5a PZ |
3402 | |
3403 | static void calc_global_nohz(unsigned long ticks) | |
3404 | { | |
3405 | } | |
74f5187a PZ |
3406 | #endif |
3407 | ||
2d02494f TG |
3408 | /** |
3409 | * get_avenrun - get the load average array | |
3410 | * @loads: pointer to dest load array | |
3411 | * @offset: offset to add | |
3412 | * @shift: shift count to shift the result left | |
3413 | * | |
3414 | * These values are estimates at best, so no need for locking. | |
3415 | */ | |
3416 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3417 | { | |
3418 | loads[0] = (avenrun[0] + offset) << shift; | |
3419 | loads[1] = (avenrun[1] + offset) << shift; | |
3420 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3421 | } |
46cb4b7c | 3422 | |
46cb4b7c | 3423 | /* |
dce48a84 TG |
3424 | * calc_load - update the avenrun load estimates 10 ticks after the |
3425 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3426 | */ |
0f004f5a | 3427 | void calc_global_load(unsigned long ticks) |
7835b98b | 3428 | { |
dce48a84 | 3429 | long active; |
1da177e4 | 3430 | |
0f004f5a PZ |
3431 | calc_global_nohz(ticks); |
3432 | ||
3433 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3434 | return; |
1da177e4 | 3435 | |
dce48a84 TG |
3436 | active = atomic_long_read(&calc_load_tasks); |
3437 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3438 | |
dce48a84 TG |
3439 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3440 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3441 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3442 | |
dce48a84 TG |
3443 | calc_load_update += LOAD_FREQ; |
3444 | } | |
1da177e4 | 3445 | |
dce48a84 | 3446 | /* |
74f5187a PZ |
3447 | * Called from update_cpu_load() to periodically update this CPU's |
3448 | * active count. | |
dce48a84 TG |
3449 | */ |
3450 | static void calc_load_account_active(struct rq *this_rq) | |
3451 | { | |
74f5187a | 3452 | long delta; |
08c183f3 | 3453 | |
74f5187a PZ |
3454 | if (time_before(jiffies, this_rq->calc_load_update)) |
3455 | return; | |
783609c6 | 3456 | |
74f5187a PZ |
3457 | delta = calc_load_fold_active(this_rq); |
3458 | delta += calc_load_fold_idle(); | |
3459 | if (delta) | |
dce48a84 | 3460 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3461 | |
3462 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3463 | } |
3464 | ||
fdf3e95d VP |
3465 | /* |
3466 | * The exact cpuload at various idx values, calculated at every tick would be | |
3467 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3468 | * | |
3469 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3470 | * on nth tick when cpu may be busy, then we have: | |
3471 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3472 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3473 | * | |
3474 | * decay_load_missed() below does efficient calculation of | |
3475 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3476 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3477 | * | |
3478 | * The calculation is approximated on a 128 point scale. | |
3479 | * degrade_zero_ticks is the number of ticks after which load at any | |
3480 | * particular idx is approximated to be zero. | |
3481 | * degrade_factor is a precomputed table, a row for each load idx. | |
3482 | * Each column corresponds to degradation factor for a power of two ticks, | |
3483 | * based on 128 point scale. | |
3484 | * Example: | |
3485 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3486 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3487 | * | |
3488 | * With this power of 2 load factors, we can degrade the load n times | |
3489 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3490 | * n mult/shifts needed by the exact degradation. | |
3491 | */ | |
3492 | #define DEGRADE_SHIFT 7 | |
3493 | static const unsigned char | |
3494 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3495 | static const unsigned char | |
3496 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3497 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3498 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3499 | {96, 72, 40, 12, 1, 0, 0}, | |
3500 | {112, 98, 75, 43, 15, 1, 0}, | |
3501 | {120, 112, 98, 76, 45, 16, 2} }; | |
3502 | ||
3503 | /* | |
3504 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3505 | * would be when CPU is idle and so we just decay the old load without | |
3506 | * adding any new load. | |
3507 | */ | |
3508 | static unsigned long | |
3509 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3510 | { | |
3511 | int j = 0; | |
3512 | ||
3513 | if (!missed_updates) | |
3514 | return load; | |
3515 | ||
3516 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3517 | return 0; | |
3518 | ||
3519 | if (idx == 1) | |
3520 | return load >> missed_updates; | |
3521 | ||
3522 | while (missed_updates) { | |
3523 | if (missed_updates % 2) | |
3524 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3525 | ||
3526 | missed_updates >>= 1; | |
3527 | j++; | |
3528 | } | |
3529 | return load; | |
3530 | } | |
3531 | ||
46cb4b7c | 3532 | /* |
dd41f596 | 3533 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3534 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3535 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3536 | */ |
dd41f596 | 3537 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3538 | { |
495eca49 | 3539 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3540 | unsigned long curr_jiffies = jiffies; |
3541 | unsigned long pending_updates; | |
dd41f596 | 3542 | int i, scale; |
46cb4b7c | 3543 | |
dd41f596 | 3544 | this_rq->nr_load_updates++; |
46cb4b7c | 3545 | |
fdf3e95d VP |
3546 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3547 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3548 | return; | |
3549 | ||
3550 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3551 | this_rq->last_load_update_tick = curr_jiffies; | |
3552 | ||
dd41f596 | 3553 | /* Update our load: */ |
fdf3e95d VP |
3554 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3555 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3556 | unsigned long old_load, new_load; |
7d1e6a9b | 3557 | |
dd41f596 | 3558 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3559 | |
dd41f596 | 3560 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3561 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3562 | new_load = this_load; |
a25707f3 IM |
3563 | /* |
3564 | * Round up the averaging division if load is increasing. This | |
3565 | * prevents us from getting stuck on 9 if the load is 10, for | |
3566 | * example. | |
3567 | */ | |
3568 | if (new_load > old_load) | |
fdf3e95d VP |
3569 | new_load += scale - 1; |
3570 | ||
3571 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3572 | } |
da2b71ed SS |
3573 | |
3574 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3575 | } |
3576 | ||
3577 | static void update_cpu_load_active(struct rq *this_rq) | |
3578 | { | |
3579 | update_cpu_load(this_rq); | |
46cb4b7c | 3580 | |
74f5187a | 3581 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3582 | } |
3583 | ||
dd41f596 | 3584 | #ifdef CONFIG_SMP |
8a0be9ef | 3585 | |
46cb4b7c | 3586 | /* |
38022906 PZ |
3587 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3588 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3589 | */ |
38022906 | 3590 | void sched_exec(void) |
46cb4b7c | 3591 | { |
38022906 | 3592 | struct task_struct *p = current; |
1da177e4 | 3593 | unsigned long flags; |
0017d735 | 3594 | int dest_cpu; |
46cb4b7c | 3595 | |
8f42ced9 | 3596 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
7608dec2 | 3597 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
0017d735 PZ |
3598 | if (dest_cpu == smp_processor_id()) |
3599 | goto unlock; | |
38022906 | 3600 | |
8f42ced9 | 3601 | if (likely(cpu_active(dest_cpu))) { |
969c7921 | 3602 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3603 | |
8f42ced9 PZ |
3604 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
3605 | stop_one_cpu(task_cpu(p), migration_cpu_stop, &arg); | |
1da177e4 LT |
3606 | return; |
3607 | } | |
0017d735 | 3608 | unlock: |
8f42ced9 | 3609 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 | 3610 | } |
dd41f596 | 3611 | |
1da177e4 LT |
3612 | #endif |
3613 | ||
1da177e4 LT |
3614 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3615 | ||
3616 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3617 | ||
3618 | /* | |
c5f8d995 | 3619 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3620 | * @p in case that task is currently running. |
c5f8d995 HS |
3621 | * |
3622 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3623 | */ |
c5f8d995 HS |
3624 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3625 | { | |
3626 | u64 ns = 0; | |
3627 | ||
3628 | if (task_current(rq, p)) { | |
3629 | update_rq_clock(rq); | |
305e6835 | 3630 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3631 | if ((s64)ns < 0) |
3632 | ns = 0; | |
3633 | } | |
3634 | ||
3635 | return ns; | |
3636 | } | |
3637 | ||
bb34d92f | 3638 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3639 | { |
1da177e4 | 3640 | unsigned long flags; |
41b86e9c | 3641 | struct rq *rq; |
bb34d92f | 3642 | u64 ns = 0; |
48f24c4d | 3643 | |
41b86e9c | 3644 | rq = task_rq_lock(p, &flags); |
c5f8d995 | 3645 | ns = do_task_delta_exec(p, rq); |
0122ec5b | 3646 | task_rq_unlock(rq, p, &flags); |
1508487e | 3647 | |
c5f8d995 HS |
3648 | return ns; |
3649 | } | |
f06febc9 | 3650 | |
c5f8d995 HS |
3651 | /* |
3652 | * Return accounted runtime for the task. | |
3653 | * In case the task is currently running, return the runtime plus current's | |
3654 | * pending runtime that have not been accounted yet. | |
3655 | */ | |
3656 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3657 | { | |
3658 | unsigned long flags; | |
3659 | struct rq *rq; | |
3660 | u64 ns = 0; | |
3661 | ||
3662 | rq = task_rq_lock(p, &flags); | |
3663 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 3664 | task_rq_unlock(rq, p, &flags); |
c5f8d995 HS |
3665 | |
3666 | return ns; | |
3667 | } | |
48f24c4d | 3668 | |
c5f8d995 HS |
3669 | /* |
3670 | * Return sum_exec_runtime for the thread group. | |
3671 | * In case the task is currently running, return the sum plus current's | |
3672 | * pending runtime that have not been accounted yet. | |
3673 | * | |
3674 | * Note that the thread group might have other running tasks as well, | |
3675 | * so the return value not includes other pending runtime that other | |
3676 | * running tasks might have. | |
3677 | */ | |
3678 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3679 | { | |
3680 | struct task_cputime totals; | |
3681 | unsigned long flags; | |
3682 | struct rq *rq; | |
3683 | u64 ns; | |
3684 | ||
3685 | rq = task_rq_lock(p, &flags); | |
3686 | thread_group_cputime(p, &totals); | |
3687 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
0122ec5b | 3688 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 3689 | |
1da177e4 LT |
3690 | return ns; |
3691 | } | |
3692 | ||
1da177e4 LT |
3693 | /* |
3694 | * Account user cpu time to a process. | |
3695 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3696 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3697 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3698 | */ |
457533a7 MS |
3699 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3700 | cputime_t cputime_scaled) | |
1da177e4 LT |
3701 | { |
3702 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3703 | cputime64_t tmp; | |
3704 | ||
457533a7 | 3705 | /* Add user time to process. */ |
1da177e4 | 3706 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3707 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3708 | account_group_user_time(p, cputime); |
1da177e4 LT |
3709 | |
3710 | /* Add user time to cpustat. */ | |
3711 | tmp = cputime_to_cputime64(cputime); | |
3712 | if (TASK_NICE(p) > 0) | |
3713 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3714 | else | |
3715 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3716 | |
3717 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3718 | /* Account for user time used */ |
3719 | acct_update_integrals(p); | |
1da177e4 LT |
3720 | } |
3721 | ||
94886b84 LV |
3722 | /* |
3723 | * Account guest cpu time to a process. | |
3724 | * @p: the process that the cpu time gets accounted to | |
3725 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3726 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3727 | */ |
457533a7 MS |
3728 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3729 | cputime_t cputime_scaled) | |
94886b84 LV |
3730 | { |
3731 | cputime64_t tmp; | |
3732 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3733 | ||
3734 | tmp = cputime_to_cputime64(cputime); | |
3735 | ||
457533a7 | 3736 | /* Add guest time to process. */ |
94886b84 | 3737 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3738 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3739 | account_group_user_time(p, cputime); |
94886b84 LV |
3740 | p->gtime = cputime_add(p->gtime, cputime); |
3741 | ||
457533a7 | 3742 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3743 | if (TASK_NICE(p) > 0) { |
3744 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3745 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3746 | } else { | |
3747 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3748 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3749 | } | |
94886b84 LV |
3750 | } |
3751 | ||
70a89a66 VP |
3752 | /* |
3753 | * Account system cpu time to a process and desired cpustat field | |
3754 | * @p: the process that the cpu time gets accounted to | |
3755 | * @cputime: the cpu time spent in kernel space since the last update | |
3756 | * @cputime_scaled: cputime scaled by cpu frequency | |
3757 | * @target_cputime64: pointer to cpustat field that has to be updated | |
3758 | */ | |
3759 | static inline | |
3760 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
3761 | cputime_t cputime_scaled, cputime64_t *target_cputime64) | |
3762 | { | |
3763 | cputime64_t tmp = cputime_to_cputime64(cputime); | |
3764 | ||
3765 | /* Add system time to process. */ | |
3766 | p->stime = cputime_add(p->stime, cputime); | |
3767 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); | |
3768 | account_group_system_time(p, cputime); | |
3769 | ||
3770 | /* Add system time to cpustat. */ | |
3771 | *target_cputime64 = cputime64_add(*target_cputime64, tmp); | |
3772 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); | |
3773 | ||
3774 | /* Account for system time used */ | |
3775 | acct_update_integrals(p); | |
3776 | } | |
3777 | ||
1da177e4 LT |
3778 | /* |
3779 | * Account system cpu time to a process. | |
3780 | * @p: the process that the cpu time gets accounted to | |
3781 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3782 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3783 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3784 | */ |
3785 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3786 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3787 | { |
3788 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
70a89a66 | 3789 | cputime64_t *target_cputime64; |
1da177e4 | 3790 | |
983ed7a6 | 3791 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3792 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3793 | return; |
3794 | } | |
94886b84 | 3795 | |
1da177e4 | 3796 | if (hardirq_count() - hardirq_offset) |
70a89a66 | 3797 | target_cputime64 = &cpustat->irq; |
75e1056f | 3798 | else if (in_serving_softirq()) |
70a89a66 | 3799 | target_cputime64 = &cpustat->softirq; |
1da177e4 | 3800 | else |
70a89a66 | 3801 | target_cputime64 = &cpustat->system; |
ef12fefa | 3802 | |
70a89a66 | 3803 | __account_system_time(p, cputime, cputime_scaled, target_cputime64); |
1da177e4 LT |
3804 | } |
3805 | ||
c66f08be | 3806 | /* |
1da177e4 | 3807 | * Account for involuntary wait time. |
544b4a1f | 3808 | * @cputime: the cpu time spent in involuntary wait |
c66f08be | 3809 | */ |
79741dd3 | 3810 | void account_steal_time(cputime_t cputime) |
c66f08be | 3811 | { |
79741dd3 MS |
3812 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3813 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3814 | ||
3815 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3816 | } |
3817 | ||
1da177e4 | 3818 | /* |
79741dd3 MS |
3819 | * Account for idle time. |
3820 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3821 | */ |
79741dd3 | 3822 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3823 | { |
3824 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3825 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3826 | struct rq *rq = this_rq(); |
1da177e4 | 3827 | |
79741dd3 MS |
3828 | if (atomic_read(&rq->nr_iowait) > 0) |
3829 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3830 | else | |
3831 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3832 | } |
3833 | ||
79741dd3 MS |
3834 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3835 | ||
abb74cef VP |
3836 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
3837 | /* | |
3838 | * Account a tick to a process and cpustat | |
3839 | * @p: the process that the cpu time gets accounted to | |
3840 | * @user_tick: is the tick from userspace | |
3841 | * @rq: the pointer to rq | |
3842 | * | |
3843 | * Tick demultiplexing follows the order | |
3844 | * - pending hardirq update | |
3845 | * - pending softirq update | |
3846 | * - user_time | |
3847 | * - idle_time | |
3848 | * - system time | |
3849 | * - check for guest_time | |
3850 | * - else account as system_time | |
3851 | * | |
3852 | * Check for hardirq is done both for system and user time as there is | |
3853 | * no timer going off while we are on hardirq and hence we may never get an | |
3854 | * opportunity to update it solely in system time. | |
3855 | * p->stime and friends are only updated on system time and not on irq | |
3856 | * softirq as those do not count in task exec_runtime any more. | |
3857 | */ | |
3858 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3859 | struct rq *rq) | |
3860 | { | |
3861 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
3862 | cputime64_t tmp = cputime_to_cputime64(cputime_one_jiffy); | |
3863 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3864 | ||
3865 | if (irqtime_account_hi_update()) { | |
3866 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3867 | } else if (irqtime_account_si_update()) { | |
3868 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
414bee9b VP |
3869 | } else if (this_cpu_ksoftirqd() == p) { |
3870 | /* | |
3871 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
3872 | * So, we have to handle it separately here. | |
3873 | * Also, p->stime needs to be updated for ksoftirqd. | |
3874 | */ | |
3875 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3876 | &cpustat->softirq); | |
abb74cef VP |
3877 | } else if (user_tick) { |
3878 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3879 | } else if (p == rq->idle) { | |
3880 | account_idle_time(cputime_one_jiffy); | |
3881 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
3882 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
3883 | } else { | |
3884 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
3885 | &cpustat->system); | |
3886 | } | |
3887 | } | |
3888 | ||
3889 | static void irqtime_account_idle_ticks(int ticks) | |
3890 | { | |
3891 | int i; | |
3892 | struct rq *rq = this_rq(); | |
3893 | ||
3894 | for (i = 0; i < ticks; i++) | |
3895 | irqtime_account_process_tick(current, 0, rq); | |
3896 | } | |
544b4a1f | 3897 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
abb74cef VP |
3898 | static void irqtime_account_idle_ticks(int ticks) {} |
3899 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
3900 | struct rq *rq) {} | |
544b4a1f | 3901 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
79741dd3 MS |
3902 | |
3903 | /* | |
3904 | * Account a single tick of cpu time. | |
3905 | * @p: the process that the cpu time gets accounted to | |
3906 | * @user_tick: indicates if the tick is a user or a system tick | |
3907 | */ | |
3908 | void account_process_tick(struct task_struct *p, int user_tick) | |
3909 | { | |
a42548a1 | 3910 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3911 | struct rq *rq = this_rq(); |
3912 | ||
abb74cef VP |
3913 | if (sched_clock_irqtime) { |
3914 | irqtime_account_process_tick(p, user_tick, rq); | |
3915 | return; | |
3916 | } | |
3917 | ||
79741dd3 | 3918 | if (user_tick) |
a42548a1 | 3919 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3920 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3921 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3922 | one_jiffy_scaled); |
3923 | else | |
a42548a1 | 3924 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3925 | } |
3926 | ||
3927 | /* | |
3928 | * Account multiple ticks of steal time. | |
3929 | * @p: the process from which the cpu time has been stolen | |
3930 | * @ticks: number of stolen ticks | |
3931 | */ | |
3932 | void account_steal_ticks(unsigned long ticks) | |
3933 | { | |
3934 | account_steal_time(jiffies_to_cputime(ticks)); | |
3935 | } | |
3936 | ||
3937 | /* | |
3938 | * Account multiple ticks of idle time. | |
3939 | * @ticks: number of stolen ticks | |
3940 | */ | |
3941 | void account_idle_ticks(unsigned long ticks) | |
3942 | { | |
abb74cef VP |
3943 | |
3944 | if (sched_clock_irqtime) { | |
3945 | irqtime_account_idle_ticks(ticks); | |
3946 | return; | |
3947 | } | |
3948 | ||
79741dd3 | 3949 | account_idle_time(jiffies_to_cputime(ticks)); |
1da177e4 LT |
3950 | } |
3951 | ||
79741dd3 MS |
3952 | #endif |
3953 | ||
49048622 BS |
3954 | /* |
3955 | * Use precise platform statistics if available: | |
3956 | */ | |
3957 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3958 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3959 | { |
d99ca3b9 HS |
3960 | *ut = p->utime; |
3961 | *st = p->stime; | |
49048622 BS |
3962 | } |
3963 | ||
0cf55e1e | 3964 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3965 | { |
0cf55e1e HS |
3966 | struct task_cputime cputime; |
3967 | ||
3968 | thread_group_cputime(p, &cputime); | |
3969 | ||
3970 | *ut = cputime.utime; | |
3971 | *st = cputime.stime; | |
49048622 BS |
3972 | } |
3973 | #else | |
761b1d26 HS |
3974 | |
3975 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3976 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3977 | #endif |
3978 | ||
d180c5bc | 3979 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3980 | { |
d99ca3b9 | 3981 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3982 | |
3983 | /* | |
3984 | * Use CFS's precise accounting: | |
3985 | */ | |
d180c5bc | 3986 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3987 | |
3988 | if (total) { | |
e75e863d | 3989 | u64 temp = rtime; |
d180c5bc | 3990 | |
e75e863d | 3991 | temp *= utime; |
49048622 | 3992 | do_div(temp, total); |
d180c5bc HS |
3993 | utime = (cputime_t)temp; |
3994 | } else | |
3995 | utime = rtime; | |
49048622 | 3996 | |
d180c5bc HS |
3997 | /* |
3998 | * Compare with previous values, to keep monotonicity: | |
3999 | */ | |
761b1d26 | 4000 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 4001 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 4002 | |
d99ca3b9 HS |
4003 | *ut = p->prev_utime; |
4004 | *st = p->prev_stime; | |
49048622 BS |
4005 | } |
4006 | ||
0cf55e1e HS |
4007 | /* |
4008 | * Must be called with siglock held. | |
4009 | */ | |
4010 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 4011 | { |
0cf55e1e HS |
4012 | struct signal_struct *sig = p->signal; |
4013 | struct task_cputime cputime; | |
4014 | cputime_t rtime, utime, total; | |
49048622 | 4015 | |
0cf55e1e | 4016 | thread_group_cputime(p, &cputime); |
49048622 | 4017 | |
0cf55e1e HS |
4018 | total = cputime_add(cputime.utime, cputime.stime); |
4019 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 4020 | |
0cf55e1e | 4021 | if (total) { |
e75e863d | 4022 | u64 temp = rtime; |
49048622 | 4023 | |
e75e863d | 4024 | temp *= cputime.utime; |
0cf55e1e HS |
4025 | do_div(temp, total); |
4026 | utime = (cputime_t)temp; | |
4027 | } else | |
4028 | utime = rtime; | |
4029 | ||
4030 | sig->prev_utime = max(sig->prev_utime, utime); | |
4031 | sig->prev_stime = max(sig->prev_stime, | |
4032 | cputime_sub(rtime, sig->prev_utime)); | |
4033 | ||
4034 | *ut = sig->prev_utime; | |
4035 | *st = sig->prev_stime; | |
49048622 | 4036 | } |
49048622 | 4037 | #endif |
49048622 | 4038 | |
7835b98b CL |
4039 | /* |
4040 | * This function gets called by the timer code, with HZ frequency. | |
4041 | * We call it with interrupts disabled. | |
7835b98b CL |
4042 | */ |
4043 | void scheduler_tick(void) | |
4044 | { | |
7835b98b CL |
4045 | int cpu = smp_processor_id(); |
4046 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 4047 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
4048 | |
4049 | sched_clock_tick(); | |
dd41f596 | 4050 | |
05fa785c | 4051 | raw_spin_lock(&rq->lock); |
3e51f33f | 4052 | update_rq_clock(rq); |
fdf3e95d | 4053 | update_cpu_load_active(rq); |
fa85ae24 | 4054 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 4055 | raw_spin_unlock(&rq->lock); |
7835b98b | 4056 | |
e9d2b064 | 4057 | perf_event_task_tick(); |
e220d2dc | 4058 | |
e418e1c2 | 4059 | #ifdef CONFIG_SMP |
dd41f596 IM |
4060 | rq->idle_at_tick = idle_cpu(cpu); |
4061 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 4062 | #endif |
1da177e4 LT |
4063 | } |
4064 | ||
132380a0 | 4065 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
4066 | { |
4067 | if (in_lock_functions(addr)) { | |
4068 | addr = CALLER_ADDR2; | |
4069 | if (in_lock_functions(addr)) | |
4070 | addr = CALLER_ADDR3; | |
4071 | } | |
4072 | return addr; | |
4073 | } | |
1da177e4 | 4074 | |
7e49fcce SR |
4075 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
4076 | defined(CONFIG_PREEMPT_TRACER)) | |
4077 | ||
43627582 | 4078 | void __kprobes add_preempt_count(int val) |
1da177e4 | 4079 | { |
6cd8a4bb | 4080 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4081 | /* |
4082 | * Underflow? | |
4083 | */ | |
9a11b49a IM |
4084 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
4085 | return; | |
6cd8a4bb | 4086 | #endif |
1da177e4 | 4087 | preempt_count() += val; |
6cd8a4bb | 4088 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4089 | /* |
4090 | * Spinlock count overflowing soon? | |
4091 | */ | |
33859f7f MOS |
4092 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
4093 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
4094 | #endif |
4095 | if (preempt_count() == val) | |
4096 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4097 | } |
4098 | EXPORT_SYMBOL(add_preempt_count); | |
4099 | ||
43627582 | 4100 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 4101 | { |
6cd8a4bb | 4102 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
4103 | /* |
4104 | * Underflow? | |
4105 | */ | |
01e3eb82 | 4106 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 4107 | return; |
1da177e4 LT |
4108 | /* |
4109 | * Is the spinlock portion underflowing? | |
4110 | */ | |
9a11b49a IM |
4111 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
4112 | !(preempt_count() & PREEMPT_MASK))) | |
4113 | return; | |
6cd8a4bb | 4114 | #endif |
9a11b49a | 4115 | |
6cd8a4bb SR |
4116 | if (preempt_count() == val) |
4117 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
4118 | preempt_count() -= val; |
4119 | } | |
4120 | EXPORT_SYMBOL(sub_preempt_count); | |
4121 | ||
4122 | #endif | |
4123 | ||
4124 | /* | |
dd41f596 | 4125 | * Print scheduling while atomic bug: |
1da177e4 | 4126 | */ |
dd41f596 | 4127 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 4128 | { |
838225b4 SS |
4129 | struct pt_regs *regs = get_irq_regs(); |
4130 | ||
3df0fc5b PZ |
4131 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
4132 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 4133 | |
dd41f596 | 4134 | debug_show_held_locks(prev); |
e21f5b15 | 4135 | print_modules(); |
dd41f596 IM |
4136 | if (irqs_disabled()) |
4137 | print_irqtrace_events(prev); | |
838225b4 SS |
4138 | |
4139 | if (regs) | |
4140 | show_regs(regs); | |
4141 | else | |
4142 | dump_stack(); | |
dd41f596 | 4143 | } |
1da177e4 | 4144 | |
dd41f596 IM |
4145 | /* |
4146 | * Various schedule()-time debugging checks and statistics: | |
4147 | */ | |
4148 | static inline void schedule_debug(struct task_struct *prev) | |
4149 | { | |
1da177e4 | 4150 | /* |
41a2d6cf | 4151 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4152 | * schedule() atomically, we ignore that path for now. |
4153 | * Otherwise, whine if we are scheduling when we should not be. | |
4154 | */ | |
3f33a7ce | 4155 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4156 | __schedule_bug(prev); |
4157 | ||
1da177e4 LT |
4158 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4159 | ||
2d72376b | 4160 | schedstat_inc(this_rq(), sched_count); |
dd41f596 IM |
4161 | } |
4162 | ||
6cecd084 | 4163 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4164 | { |
61eadef6 | 4165 | if (prev->on_rq || rq->skip_clock_update < 0) |
a64692a3 | 4166 | update_rq_clock(rq); |
6cecd084 | 4167 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4168 | } |
4169 | ||
dd41f596 IM |
4170 | /* |
4171 | * Pick up the highest-prio task: | |
4172 | */ | |
4173 | static inline struct task_struct * | |
b67802ea | 4174 | pick_next_task(struct rq *rq) |
dd41f596 | 4175 | { |
5522d5d5 | 4176 | const struct sched_class *class; |
dd41f596 | 4177 | struct task_struct *p; |
1da177e4 LT |
4178 | |
4179 | /* | |
dd41f596 IM |
4180 | * Optimization: we know that if all tasks are in |
4181 | * the fair class we can call that function directly: | |
1da177e4 | 4182 | */ |
dd41f596 | 4183 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4184 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4185 | if (likely(p)) |
4186 | return p; | |
1da177e4 LT |
4187 | } |
4188 | ||
34f971f6 | 4189 | for_each_class(class) { |
fb8d4724 | 4190 | p = class->pick_next_task(rq); |
dd41f596 IM |
4191 | if (p) |
4192 | return p; | |
dd41f596 | 4193 | } |
34f971f6 PZ |
4194 | |
4195 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4196 | } |
1da177e4 | 4197 | |
dd41f596 IM |
4198 | /* |
4199 | * schedule() is the main scheduler function. | |
4200 | */ | |
ff743345 | 4201 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4202 | { |
4203 | struct task_struct *prev, *next; | |
67ca7bde | 4204 | unsigned long *switch_count; |
dd41f596 | 4205 | struct rq *rq; |
31656519 | 4206 | int cpu; |
dd41f596 | 4207 | |
ff743345 PZ |
4208 | need_resched: |
4209 | preempt_disable(); | |
dd41f596 IM |
4210 | cpu = smp_processor_id(); |
4211 | rq = cpu_rq(cpu); | |
25502a6c | 4212 | rcu_note_context_switch(cpu); |
dd41f596 | 4213 | prev = rq->curr; |
dd41f596 | 4214 | |
dd41f596 | 4215 | schedule_debug(prev); |
1da177e4 | 4216 | |
31656519 | 4217 | if (sched_feat(HRTICK)) |
f333fdc9 | 4218 | hrtick_clear(rq); |
8f4d37ec | 4219 | |
05fa785c | 4220 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4221 | |
246d86b5 | 4222 | switch_count = &prev->nivcsw; |
1da177e4 | 4223 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4224 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4225 | prev->state = TASK_RUNNING; |
21aa9af0 | 4226 | } else { |
2acca55e PZ |
4227 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
4228 | prev->on_rq = 0; | |
4229 | ||
21aa9af0 | 4230 | /* |
2acca55e PZ |
4231 | * If a worker went to sleep, notify and ask workqueue |
4232 | * whether it wants to wake up a task to maintain | |
4233 | * concurrency. | |
21aa9af0 TH |
4234 | */ |
4235 | if (prev->flags & PF_WQ_WORKER) { | |
4236 | struct task_struct *to_wakeup; | |
4237 | ||
4238 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4239 | if (to_wakeup) | |
4240 | try_to_wake_up_local(to_wakeup); | |
4241 | } | |
fd2f4419 | 4242 | |
6631e635 | 4243 | /* |
2acca55e PZ |
4244 | * If we are going to sleep and we have plugged IO |
4245 | * queued, make sure to submit it to avoid deadlocks. | |
6631e635 LT |
4246 | */ |
4247 | if (blk_needs_flush_plug(prev)) { | |
4248 | raw_spin_unlock(&rq->lock); | |
a237c1c5 | 4249 | blk_schedule_flush_plug(prev); |
6631e635 LT |
4250 | raw_spin_lock(&rq->lock); |
4251 | } | |
21aa9af0 | 4252 | } |
dd41f596 | 4253 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4254 | } |
4255 | ||
3f029d3c | 4256 | pre_schedule(rq, prev); |
f65eda4f | 4257 | |
dd41f596 | 4258 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4259 | idle_balance(cpu, rq); |
1da177e4 | 4260 | |
df1c99d4 | 4261 | put_prev_task(rq, prev); |
b67802ea | 4262 | next = pick_next_task(rq); |
f26f9aff MG |
4263 | clear_tsk_need_resched(prev); |
4264 | rq->skip_clock_update = 0; | |
1da177e4 | 4265 | |
1da177e4 | 4266 | if (likely(prev != next)) { |
1da177e4 LT |
4267 | rq->nr_switches++; |
4268 | rq->curr = next; | |
4269 | ++*switch_count; | |
4270 | ||
dd41f596 | 4271 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4272 | /* |
246d86b5 ON |
4273 | * The context switch have flipped the stack from under us |
4274 | * and restored the local variables which were saved when | |
4275 | * this task called schedule() in the past. prev == current | |
4276 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4277 | */ |
4278 | cpu = smp_processor_id(); | |
4279 | rq = cpu_rq(cpu); | |
1da177e4 | 4280 | } else |
05fa785c | 4281 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4282 | |
3f029d3c | 4283 | post_schedule(rq); |
1da177e4 | 4284 | |
1da177e4 | 4285 | preempt_enable_no_resched(); |
ff743345 | 4286 | if (need_resched()) |
1da177e4 LT |
4287 | goto need_resched; |
4288 | } | |
1da177e4 LT |
4289 | EXPORT_SYMBOL(schedule); |
4290 | ||
c08f7829 | 4291 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d | 4292 | |
c6eb3dda PZ |
4293 | static inline bool owner_running(struct mutex *lock, struct task_struct *owner) |
4294 | { | |
4295 | bool ret = false; | |
0d66bf6d | 4296 | |
c6eb3dda PZ |
4297 | rcu_read_lock(); |
4298 | if (lock->owner != owner) | |
4299 | goto fail; | |
0d66bf6d PZ |
4300 | |
4301 | /* | |
c6eb3dda PZ |
4302 | * Ensure we emit the owner->on_cpu, dereference _after_ checking |
4303 | * lock->owner still matches owner, if that fails, owner might | |
4304 | * point to free()d memory, if it still matches, the rcu_read_lock() | |
4305 | * ensures the memory stays valid. | |
0d66bf6d | 4306 | */ |
c6eb3dda | 4307 | barrier(); |
0d66bf6d | 4308 | |
c6eb3dda PZ |
4309 | ret = owner->on_cpu; |
4310 | fail: | |
4311 | rcu_read_unlock(); | |
0d66bf6d | 4312 | |
c6eb3dda PZ |
4313 | return ret; |
4314 | } | |
0d66bf6d | 4315 | |
c6eb3dda PZ |
4316 | /* |
4317 | * Look out! "owner" is an entirely speculative pointer | |
4318 | * access and not reliable. | |
4319 | */ | |
4320 | int mutex_spin_on_owner(struct mutex *lock, struct task_struct *owner) | |
4321 | { | |
4322 | if (!sched_feat(OWNER_SPIN)) | |
4323 | return 0; | |
0d66bf6d | 4324 | |
c6eb3dda PZ |
4325 | while (owner_running(lock, owner)) { |
4326 | if (need_resched()) | |
0d66bf6d PZ |
4327 | return 0; |
4328 | ||
335d7afb | 4329 | arch_mutex_cpu_relax(); |
0d66bf6d | 4330 | } |
4b402210 | 4331 | |
c6eb3dda PZ |
4332 | /* |
4333 | * If the owner changed to another task there is likely | |
4334 | * heavy contention, stop spinning. | |
4335 | */ | |
4336 | if (lock->owner) | |
4337 | return 0; | |
4338 | ||
0d66bf6d PZ |
4339 | return 1; |
4340 | } | |
4341 | #endif | |
4342 | ||
1da177e4 LT |
4343 | #ifdef CONFIG_PREEMPT |
4344 | /* | |
2ed6e34f | 4345 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4346 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4347 | * occur there and call schedule directly. |
4348 | */ | |
d1f74e20 | 4349 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4350 | { |
4351 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4352 | |
1da177e4 LT |
4353 | /* |
4354 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4355 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4356 | */ |
beed33a8 | 4357 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4358 | return; |
4359 | ||
3a5c359a | 4360 | do { |
d1f74e20 | 4361 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4362 | schedule(); |
d1f74e20 | 4363 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4364 | |
3a5c359a AK |
4365 | /* |
4366 | * Check again in case we missed a preemption opportunity | |
4367 | * between schedule and now. | |
4368 | */ | |
4369 | barrier(); | |
5ed0cec0 | 4370 | } while (need_resched()); |
1da177e4 | 4371 | } |
1da177e4 LT |
4372 | EXPORT_SYMBOL(preempt_schedule); |
4373 | ||
4374 | /* | |
2ed6e34f | 4375 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4376 | * off of irq context. |
4377 | * Note, that this is called and return with irqs disabled. This will | |
4378 | * protect us against recursive calling from irq. | |
4379 | */ | |
4380 | asmlinkage void __sched preempt_schedule_irq(void) | |
4381 | { | |
4382 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4383 | |
2ed6e34f | 4384 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4385 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4386 | ||
3a5c359a AK |
4387 | do { |
4388 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4389 | local_irq_enable(); |
4390 | schedule(); | |
4391 | local_irq_disable(); | |
3a5c359a | 4392 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4393 | |
3a5c359a AK |
4394 | /* |
4395 | * Check again in case we missed a preemption opportunity | |
4396 | * between schedule and now. | |
4397 | */ | |
4398 | barrier(); | |
5ed0cec0 | 4399 | } while (need_resched()); |
1da177e4 LT |
4400 | } |
4401 | ||
4402 | #endif /* CONFIG_PREEMPT */ | |
4403 | ||
63859d4f | 4404 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4405 | void *key) |
1da177e4 | 4406 | { |
63859d4f | 4407 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4408 | } |
1da177e4 LT |
4409 | EXPORT_SYMBOL(default_wake_function); |
4410 | ||
4411 | /* | |
41a2d6cf IM |
4412 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4413 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4414 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4415 | * | |
4416 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4417 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4418 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4419 | */ | |
78ddb08f | 4420 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4421 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4422 | { |
2e45874c | 4423 | wait_queue_t *curr, *next; |
1da177e4 | 4424 | |
2e45874c | 4425 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4426 | unsigned flags = curr->flags; |
4427 | ||
63859d4f | 4428 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4429 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4430 | break; |
4431 | } | |
4432 | } | |
4433 | ||
4434 | /** | |
4435 | * __wake_up - wake up threads blocked on a waitqueue. | |
4436 | * @q: the waitqueue | |
4437 | * @mode: which threads | |
4438 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4439 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4440 | * |
4441 | * It may be assumed that this function implies a write memory barrier before | |
4442 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4443 | */ |
7ad5b3a5 | 4444 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4445 | int nr_exclusive, void *key) |
1da177e4 LT |
4446 | { |
4447 | unsigned long flags; | |
4448 | ||
4449 | spin_lock_irqsave(&q->lock, flags); | |
4450 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4451 | spin_unlock_irqrestore(&q->lock, flags); | |
4452 | } | |
1da177e4 LT |
4453 | EXPORT_SYMBOL(__wake_up); |
4454 | ||
4455 | /* | |
4456 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4457 | */ | |
7ad5b3a5 | 4458 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4459 | { |
4460 | __wake_up_common(q, mode, 1, 0, NULL); | |
4461 | } | |
22c43c81 | 4462 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4463 | |
4ede816a DL |
4464 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4465 | { | |
4466 | __wake_up_common(q, mode, 1, 0, key); | |
4467 | } | |
bf294b41 | 4468 | EXPORT_SYMBOL_GPL(__wake_up_locked_key); |
4ede816a | 4469 | |
1da177e4 | 4470 | /** |
4ede816a | 4471 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4472 | * @q: the waitqueue |
4473 | * @mode: which threads | |
4474 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4475 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4476 | * |
4477 | * The sync wakeup differs that the waker knows that it will schedule | |
4478 | * away soon, so while the target thread will be woken up, it will not | |
4479 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4480 | * with each other. This can prevent needless bouncing between CPUs. | |
4481 | * | |
4482 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4483 | * |
4484 | * It may be assumed that this function implies a write memory barrier before | |
4485 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4486 | */ |
4ede816a DL |
4487 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4488 | int nr_exclusive, void *key) | |
1da177e4 LT |
4489 | { |
4490 | unsigned long flags; | |
7d478721 | 4491 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4492 | |
4493 | if (unlikely(!q)) | |
4494 | return; | |
4495 | ||
4496 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4497 | wake_flags = 0; |
1da177e4 LT |
4498 | |
4499 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4500 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4501 | spin_unlock_irqrestore(&q->lock, flags); |
4502 | } | |
4ede816a DL |
4503 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4504 | ||
4505 | /* | |
4506 | * __wake_up_sync - see __wake_up_sync_key() | |
4507 | */ | |
4508 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4509 | { | |
4510 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4511 | } | |
1da177e4 LT |
4512 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4513 | ||
65eb3dc6 KD |
4514 | /** |
4515 | * complete: - signals a single thread waiting on this completion | |
4516 | * @x: holds the state of this particular completion | |
4517 | * | |
4518 | * This will wake up a single thread waiting on this completion. Threads will be | |
4519 | * awakened in the same order in which they were queued. | |
4520 | * | |
4521 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4522 | * |
4523 | * It may be assumed that this function implies a write memory barrier before | |
4524 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4525 | */ |
b15136e9 | 4526 | void complete(struct completion *x) |
1da177e4 LT |
4527 | { |
4528 | unsigned long flags; | |
4529 | ||
4530 | spin_lock_irqsave(&x->wait.lock, flags); | |
4531 | x->done++; | |
d9514f6c | 4532 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4533 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4534 | } | |
4535 | EXPORT_SYMBOL(complete); | |
4536 | ||
65eb3dc6 KD |
4537 | /** |
4538 | * complete_all: - signals all threads waiting on this completion | |
4539 | * @x: holds the state of this particular completion | |
4540 | * | |
4541 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4542 | * |
4543 | * It may be assumed that this function implies a write memory barrier before | |
4544 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4545 | */ |
b15136e9 | 4546 | void complete_all(struct completion *x) |
1da177e4 LT |
4547 | { |
4548 | unsigned long flags; | |
4549 | ||
4550 | spin_lock_irqsave(&x->wait.lock, flags); | |
4551 | x->done += UINT_MAX/2; | |
d9514f6c | 4552 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4553 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4554 | } | |
4555 | EXPORT_SYMBOL(complete_all); | |
4556 | ||
8cbbe86d AK |
4557 | static inline long __sched |
4558 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4559 | { |
1da177e4 LT |
4560 | if (!x->done) { |
4561 | DECLARE_WAITQUEUE(wait, current); | |
4562 | ||
a93d2f17 | 4563 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4564 | do { |
94d3d824 | 4565 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4566 | timeout = -ERESTARTSYS; |
4567 | break; | |
8cbbe86d AK |
4568 | } |
4569 | __set_current_state(state); | |
1da177e4 LT |
4570 | spin_unlock_irq(&x->wait.lock); |
4571 | timeout = schedule_timeout(timeout); | |
4572 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4573 | } while (!x->done && timeout); |
1da177e4 | 4574 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4575 | if (!x->done) |
4576 | return timeout; | |
1da177e4 LT |
4577 | } |
4578 | x->done--; | |
ea71a546 | 4579 | return timeout ?: 1; |
1da177e4 | 4580 | } |
1da177e4 | 4581 | |
8cbbe86d AK |
4582 | static long __sched |
4583 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4584 | { |
1da177e4 LT |
4585 | might_sleep(); |
4586 | ||
4587 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4588 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4589 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4590 | return timeout; |
4591 | } | |
1da177e4 | 4592 | |
65eb3dc6 KD |
4593 | /** |
4594 | * wait_for_completion: - waits for completion of a task | |
4595 | * @x: holds the state of this particular completion | |
4596 | * | |
4597 | * This waits to be signaled for completion of a specific task. It is NOT | |
4598 | * interruptible and there is no timeout. | |
4599 | * | |
4600 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4601 | * and interrupt capability. Also see complete(). | |
4602 | */ | |
b15136e9 | 4603 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4604 | { |
4605 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4606 | } |
8cbbe86d | 4607 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4608 | |
65eb3dc6 KD |
4609 | /** |
4610 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4611 | * @x: holds the state of this particular completion | |
4612 | * @timeout: timeout value in jiffies | |
4613 | * | |
4614 | * This waits for either a completion of a specific task to be signaled or for a | |
4615 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4616 | * interruptible. | |
4617 | */ | |
b15136e9 | 4618 | unsigned long __sched |
8cbbe86d | 4619 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4620 | { |
8cbbe86d | 4621 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4622 | } |
8cbbe86d | 4623 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4624 | |
65eb3dc6 KD |
4625 | /** |
4626 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4627 | * @x: holds the state of this particular completion | |
4628 | * | |
4629 | * This waits for completion of a specific task to be signaled. It is | |
4630 | * interruptible. | |
4631 | */ | |
8cbbe86d | 4632 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4633 | { |
51e97990 AK |
4634 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4635 | if (t == -ERESTARTSYS) | |
4636 | return t; | |
4637 | return 0; | |
0fec171c | 4638 | } |
8cbbe86d | 4639 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4640 | |
65eb3dc6 KD |
4641 | /** |
4642 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4643 | * @x: holds the state of this particular completion | |
4644 | * @timeout: timeout value in jiffies | |
4645 | * | |
4646 | * This waits for either a completion of a specific task to be signaled or for a | |
4647 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4648 | */ | |
6bf41237 | 4649 | long __sched |
8cbbe86d AK |
4650 | wait_for_completion_interruptible_timeout(struct completion *x, |
4651 | unsigned long timeout) | |
0fec171c | 4652 | { |
8cbbe86d | 4653 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4654 | } |
8cbbe86d | 4655 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4656 | |
65eb3dc6 KD |
4657 | /** |
4658 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4659 | * @x: holds the state of this particular completion | |
4660 | * | |
4661 | * This waits to be signaled for completion of a specific task. It can be | |
4662 | * interrupted by a kill signal. | |
4663 | */ | |
009e577e MW |
4664 | int __sched wait_for_completion_killable(struct completion *x) |
4665 | { | |
4666 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4667 | if (t == -ERESTARTSYS) | |
4668 | return t; | |
4669 | return 0; | |
4670 | } | |
4671 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4672 | ||
0aa12fb4 SW |
4673 | /** |
4674 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4675 | * @x: holds the state of this particular completion | |
4676 | * @timeout: timeout value in jiffies | |
4677 | * | |
4678 | * This waits for either a completion of a specific task to be | |
4679 | * signaled or for a specified timeout to expire. It can be | |
4680 | * interrupted by a kill signal. The timeout is in jiffies. | |
4681 | */ | |
6bf41237 | 4682 | long __sched |
0aa12fb4 SW |
4683 | wait_for_completion_killable_timeout(struct completion *x, |
4684 | unsigned long timeout) | |
4685 | { | |
4686 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4687 | } | |
4688 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4689 | ||
be4de352 DC |
4690 | /** |
4691 | * try_wait_for_completion - try to decrement a completion without blocking | |
4692 | * @x: completion structure | |
4693 | * | |
4694 | * Returns: 0 if a decrement cannot be done without blocking | |
4695 | * 1 if a decrement succeeded. | |
4696 | * | |
4697 | * If a completion is being used as a counting completion, | |
4698 | * attempt to decrement the counter without blocking. This | |
4699 | * enables us to avoid waiting if the resource the completion | |
4700 | * is protecting is not available. | |
4701 | */ | |
4702 | bool try_wait_for_completion(struct completion *x) | |
4703 | { | |
7539a3b3 | 4704 | unsigned long flags; |
be4de352 DC |
4705 | int ret = 1; |
4706 | ||
7539a3b3 | 4707 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4708 | if (!x->done) |
4709 | ret = 0; | |
4710 | else | |
4711 | x->done--; | |
7539a3b3 | 4712 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4713 | return ret; |
4714 | } | |
4715 | EXPORT_SYMBOL(try_wait_for_completion); | |
4716 | ||
4717 | /** | |
4718 | * completion_done - Test to see if a completion has any waiters | |
4719 | * @x: completion structure | |
4720 | * | |
4721 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4722 | * 1 if there are no waiters. | |
4723 | * | |
4724 | */ | |
4725 | bool completion_done(struct completion *x) | |
4726 | { | |
7539a3b3 | 4727 | unsigned long flags; |
be4de352 DC |
4728 | int ret = 1; |
4729 | ||
7539a3b3 | 4730 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4731 | if (!x->done) |
4732 | ret = 0; | |
7539a3b3 | 4733 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4734 | return ret; |
4735 | } | |
4736 | EXPORT_SYMBOL(completion_done); | |
4737 | ||
8cbbe86d AK |
4738 | static long __sched |
4739 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4740 | { |
0fec171c IM |
4741 | unsigned long flags; |
4742 | wait_queue_t wait; | |
4743 | ||
4744 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4745 | |
8cbbe86d | 4746 | __set_current_state(state); |
1da177e4 | 4747 | |
8cbbe86d AK |
4748 | spin_lock_irqsave(&q->lock, flags); |
4749 | __add_wait_queue(q, &wait); | |
4750 | spin_unlock(&q->lock); | |
4751 | timeout = schedule_timeout(timeout); | |
4752 | spin_lock_irq(&q->lock); | |
4753 | __remove_wait_queue(q, &wait); | |
4754 | spin_unlock_irqrestore(&q->lock, flags); | |
4755 | ||
4756 | return timeout; | |
4757 | } | |
4758 | ||
4759 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4760 | { | |
4761 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4762 | } |
1da177e4 LT |
4763 | EXPORT_SYMBOL(interruptible_sleep_on); |
4764 | ||
0fec171c | 4765 | long __sched |
95cdf3b7 | 4766 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4767 | { |
8cbbe86d | 4768 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4769 | } |
1da177e4 LT |
4770 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4771 | ||
0fec171c | 4772 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4773 | { |
8cbbe86d | 4774 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4775 | } |
1da177e4 LT |
4776 | EXPORT_SYMBOL(sleep_on); |
4777 | ||
0fec171c | 4778 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4779 | { |
8cbbe86d | 4780 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4781 | } |
1da177e4 LT |
4782 | EXPORT_SYMBOL(sleep_on_timeout); |
4783 | ||
b29739f9 IM |
4784 | #ifdef CONFIG_RT_MUTEXES |
4785 | ||
4786 | /* | |
4787 | * rt_mutex_setprio - set the current priority of a task | |
4788 | * @p: task | |
4789 | * @prio: prio value (kernel-internal form) | |
4790 | * | |
4791 | * This function changes the 'effective' priority of a task. It does | |
4792 | * not touch ->normal_prio like __setscheduler(). | |
4793 | * | |
4794 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4795 | */ | |
36c8b586 | 4796 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 | 4797 | { |
83b699ed | 4798 | int oldprio, on_rq, running; |
70b97a7f | 4799 | struct rq *rq; |
83ab0aa0 | 4800 | const struct sched_class *prev_class; |
b29739f9 IM |
4801 | |
4802 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4803 | ||
0122ec5b | 4804 | rq = __task_rq_lock(p); |
b29739f9 | 4805 | |
a8027073 | 4806 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4807 | oldprio = p->prio; |
83ab0aa0 | 4808 | prev_class = p->sched_class; |
fd2f4419 | 4809 | on_rq = p->on_rq; |
051a1d1a | 4810 | running = task_current(rq, p); |
0e1f3483 | 4811 | if (on_rq) |
69be72c1 | 4812 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4813 | if (running) |
4814 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4815 | |
4816 | if (rt_prio(prio)) | |
4817 | p->sched_class = &rt_sched_class; | |
4818 | else | |
4819 | p->sched_class = &fair_sched_class; | |
4820 | ||
b29739f9 IM |
4821 | p->prio = prio; |
4822 | ||
0e1f3483 HS |
4823 | if (running) |
4824 | p->sched_class->set_curr_task(rq); | |
da7a735e | 4825 | if (on_rq) |
371fd7e7 | 4826 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 | 4827 | |
da7a735e | 4828 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 4829 | __task_rq_unlock(rq); |
b29739f9 IM |
4830 | } |
4831 | ||
4832 | #endif | |
4833 | ||
36c8b586 | 4834 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4835 | { |
dd41f596 | 4836 | int old_prio, delta, on_rq; |
1da177e4 | 4837 | unsigned long flags; |
70b97a7f | 4838 | struct rq *rq; |
1da177e4 LT |
4839 | |
4840 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4841 | return; | |
4842 | /* | |
4843 | * We have to be careful, if called from sys_setpriority(), | |
4844 | * the task might be in the middle of scheduling on another CPU. | |
4845 | */ | |
4846 | rq = task_rq_lock(p, &flags); | |
4847 | /* | |
4848 | * The RT priorities are set via sched_setscheduler(), but we still | |
4849 | * allow the 'normal' nice value to be set - but as expected | |
4850 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4851 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4852 | */ |
e05606d3 | 4853 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4854 | p->static_prio = NICE_TO_PRIO(nice); |
4855 | goto out_unlock; | |
4856 | } | |
fd2f4419 | 4857 | on_rq = p->on_rq; |
c09595f6 | 4858 | if (on_rq) |
69be72c1 | 4859 | dequeue_task(rq, p, 0); |
1da177e4 | 4860 | |
1da177e4 | 4861 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4862 | set_load_weight(p); |
b29739f9 IM |
4863 | old_prio = p->prio; |
4864 | p->prio = effective_prio(p); | |
4865 | delta = p->prio - old_prio; | |
1da177e4 | 4866 | |
dd41f596 | 4867 | if (on_rq) { |
371fd7e7 | 4868 | enqueue_task(rq, p, 0); |
1da177e4 | 4869 | /* |
d5f9f942 AM |
4870 | * If the task increased its priority or is running and |
4871 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4872 | */ |
d5f9f942 | 4873 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4874 | resched_task(rq->curr); |
4875 | } | |
4876 | out_unlock: | |
0122ec5b | 4877 | task_rq_unlock(rq, p, &flags); |
1da177e4 | 4878 | } |
1da177e4 LT |
4879 | EXPORT_SYMBOL(set_user_nice); |
4880 | ||
e43379f1 MM |
4881 | /* |
4882 | * can_nice - check if a task can reduce its nice value | |
4883 | * @p: task | |
4884 | * @nice: nice value | |
4885 | */ | |
36c8b586 | 4886 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4887 | { |
024f4747 MM |
4888 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4889 | int nice_rlim = 20 - nice; | |
48f24c4d | 4890 | |
78d7d407 | 4891 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4892 | capable(CAP_SYS_NICE)); |
4893 | } | |
4894 | ||
1da177e4 LT |
4895 | #ifdef __ARCH_WANT_SYS_NICE |
4896 | ||
4897 | /* | |
4898 | * sys_nice - change the priority of the current process. | |
4899 | * @increment: priority increment | |
4900 | * | |
4901 | * sys_setpriority is a more generic, but much slower function that | |
4902 | * does similar things. | |
4903 | */ | |
5add95d4 | 4904 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4905 | { |
48f24c4d | 4906 | long nice, retval; |
1da177e4 LT |
4907 | |
4908 | /* | |
4909 | * Setpriority might change our priority at the same moment. | |
4910 | * We don't have to worry. Conceptually one call occurs first | |
4911 | * and we have a single winner. | |
4912 | */ | |
e43379f1 MM |
4913 | if (increment < -40) |
4914 | increment = -40; | |
1da177e4 LT |
4915 | if (increment > 40) |
4916 | increment = 40; | |
4917 | ||
2b8f836f | 4918 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4919 | if (nice < -20) |
4920 | nice = -20; | |
4921 | if (nice > 19) | |
4922 | nice = 19; | |
4923 | ||
e43379f1 MM |
4924 | if (increment < 0 && !can_nice(current, nice)) |
4925 | return -EPERM; | |
4926 | ||
1da177e4 LT |
4927 | retval = security_task_setnice(current, nice); |
4928 | if (retval) | |
4929 | return retval; | |
4930 | ||
4931 | set_user_nice(current, nice); | |
4932 | return 0; | |
4933 | } | |
4934 | ||
4935 | #endif | |
4936 | ||
4937 | /** | |
4938 | * task_prio - return the priority value of a given task. | |
4939 | * @p: the task in question. | |
4940 | * | |
4941 | * This is the priority value as seen by users in /proc. | |
4942 | * RT tasks are offset by -200. Normal tasks are centered | |
4943 | * around 0, value goes from -16 to +15. | |
4944 | */ | |
36c8b586 | 4945 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4946 | { |
4947 | return p->prio - MAX_RT_PRIO; | |
4948 | } | |
4949 | ||
4950 | /** | |
4951 | * task_nice - return the nice value of a given task. | |
4952 | * @p: the task in question. | |
4953 | */ | |
36c8b586 | 4954 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4955 | { |
4956 | return TASK_NICE(p); | |
4957 | } | |
150d8bed | 4958 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4959 | |
4960 | /** | |
4961 | * idle_cpu - is a given cpu idle currently? | |
4962 | * @cpu: the processor in question. | |
4963 | */ | |
4964 | int idle_cpu(int cpu) | |
4965 | { | |
4966 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4967 | } | |
4968 | ||
1da177e4 LT |
4969 | /** |
4970 | * idle_task - return the idle task for a given cpu. | |
4971 | * @cpu: the processor in question. | |
4972 | */ | |
36c8b586 | 4973 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4974 | { |
4975 | return cpu_rq(cpu)->idle; | |
4976 | } | |
4977 | ||
4978 | /** | |
4979 | * find_process_by_pid - find a process with a matching PID value. | |
4980 | * @pid: the pid in question. | |
4981 | */ | |
a9957449 | 4982 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4983 | { |
228ebcbe | 4984 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4985 | } |
4986 | ||
4987 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4988 | static void |
4989 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4990 | { |
1da177e4 LT |
4991 | p->policy = policy; |
4992 | p->rt_priority = prio; | |
b29739f9 IM |
4993 | p->normal_prio = normal_prio(p); |
4994 | /* we are holding p->pi_lock already */ | |
4995 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4996 | if (rt_prio(p->prio)) |
4997 | p->sched_class = &rt_sched_class; | |
4998 | else | |
4999 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 5000 | set_load_weight(p); |
1da177e4 LT |
5001 | } |
5002 | ||
c69e8d9c DH |
5003 | /* |
5004 | * check the target process has a UID that matches the current process's | |
5005 | */ | |
5006 | static bool check_same_owner(struct task_struct *p) | |
5007 | { | |
5008 | const struct cred *cred = current_cred(), *pcred; | |
5009 | bool match; | |
5010 | ||
5011 | rcu_read_lock(); | |
5012 | pcred = __task_cred(p); | |
b0e77598 SH |
5013 | if (cred->user->user_ns == pcred->user->user_ns) |
5014 | match = (cred->euid == pcred->euid || | |
5015 | cred->euid == pcred->uid); | |
5016 | else | |
5017 | match = false; | |
c69e8d9c DH |
5018 | rcu_read_unlock(); |
5019 | return match; | |
5020 | } | |
5021 | ||
961ccddd | 5022 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 5023 | const struct sched_param *param, bool user) |
1da177e4 | 5024 | { |
83b699ed | 5025 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 5026 | unsigned long flags; |
83ab0aa0 | 5027 | const struct sched_class *prev_class; |
70b97a7f | 5028 | struct rq *rq; |
ca94c442 | 5029 | int reset_on_fork; |
1da177e4 | 5030 | |
66e5393a SR |
5031 | /* may grab non-irq protected spin_locks */ |
5032 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
5033 | recheck: |
5034 | /* double check policy once rq lock held */ | |
ca94c442 LP |
5035 | if (policy < 0) { |
5036 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 5037 | policy = oldpolicy = p->policy; |
ca94c442 LP |
5038 | } else { |
5039 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
5040 | policy &= ~SCHED_RESET_ON_FORK; | |
5041 | ||
5042 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
5043 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
5044 | policy != SCHED_IDLE) | |
5045 | return -EINVAL; | |
5046 | } | |
5047 | ||
1da177e4 LT |
5048 | /* |
5049 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
5050 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
5051 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
5052 | */ |
5053 | if (param->sched_priority < 0 || | |
95cdf3b7 | 5054 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 5055 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 5056 | return -EINVAL; |
e05606d3 | 5057 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
5058 | return -EINVAL; |
5059 | ||
37e4ab3f OC |
5060 | /* |
5061 | * Allow unprivileged RT tasks to decrease priority: | |
5062 | */ | |
961ccddd | 5063 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 5064 | if (rt_policy(policy)) { |
a44702e8 ON |
5065 | unsigned long rlim_rtprio = |
5066 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
5067 | |
5068 | /* can't set/change the rt policy */ | |
5069 | if (policy != p->policy && !rlim_rtprio) | |
5070 | return -EPERM; | |
5071 | ||
5072 | /* can't increase priority */ | |
5073 | if (param->sched_priority > p->rt_priority && | |
5074 | param->sched_priority > rlim_rtprio) | |
5075 | return -EPERM; | |
5076 | } | |
c02aa73b | 5077 | |
dd41f596 | 5078 | /* |
c02aa73b DH |
5079 | * Treat SCHED_IDLE as nice 20. Only allow a switch to |
5080 | * SCHED_NORMAL if the RLIMIT_NICE would normally permit it. | |
dd41f596 | 5081 | */ |
c02aa73b DH |
5082 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) { |
5083 | if (!can_nice(p, TASK_NICE(p))) | |
5084 | return -EPERM; | |
5085 | } | |
5fe1d75f | 5086 | |
37e4ab3f | 5087 | /* can't change other user's priorities */ |
c69e8d9c | 5088 | if (!check_same_owner(p)) |
37e4ab3f | 5089 | return -EPERM; |
ca94c442 LP |
5090 | |
5091 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
5092 | if (p->sched_reset_on_fork && !reset_on_fork) | |
5093 | return -EPERM; | |
37e4ab3f | 5094 | } |
1da177e4 | 5095 | |
725aad24 | 5096 | if (user) { |
b0ae1981 | 5097 | retval = security_task_setscheduler(p); |
725aad24 JF |
5098 | if (retval) |
5099 | return retval; | |
5100 | } | |
5101 | ||
b29739f9 IM |
5102 | /* |
5103 | * make sure no PI-waiters arrive (or leave) while we are | |
5104 | * changing the priority of the task: | |
0122ec5b | 5105 | * |
25985edc | 5106 | * To be able to change p->policy safely, the appropriate |
1da177e4 LT |
5107 | * runqueue lock must be held. |
5108 | */ | |
0122ec5b | 5109 | rq = task_rq_lock(p, &flags); |
dc61b1d6 | 5110 | |
34f971f6 PZ |
5111 | /* |
5112 | * Changing the policy of the stop threads its a very bad idea | |
5113 | */ | |
5114 | if (p == rq->stop) { | |
0122ec5b | 5115 | task_rq_unlock(rq, p, &flags); |
34f971f6 PZ |
5116 | return -EINVAL; |
5117 | } | |
5118 | ||
a51e9198 DF |
5119 | /* |
5120 | * If not changing anything there's no need to proceed further: | |
5121 | */ | |
5122 | if (unlikely(policy == p->policy && (!rt_policy(policy) || | |
5123 | param->sched_priority == p->rt_priority))) { | |
5124 | ||
5125 | __task_rq_unlock(rq); | |
5126 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5127 | return 0; | |
5128 | } | |
5129 | ||
dc61b1d6 PZ |
5130 | #ifdef CONFIG_RT_GROUP_SCHED |
5131 | if (user) { | |
5132 | /* | |
5133 | * Do not allow realtime tasks into groups that have no runtime | |
5134 | * assigned. | |
5135 | */ | |
5136 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
5137 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
5138 | !task_group_is_autogroup(task_group(p))) { | |
0122ec5b | 5139 | task_rq_unlock(rq, p, &flags); |
dc61b1d6 PZ |
5140 | return -EPERM; |
5141 | } | |
5142 | } | |
5143 | #endif | |
5144 | ||
1da177e4 LT |
5145 | /* recheck policy now with rq lock held */ |
5146 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5147 | policy = oldpolicy = -1; | |
0122ec5b | 5148 | task_rq_unlock(rq, p, &flags); |
1da177e4 LT |
5149 | goto recheck; |
5150 | } | |
fd2f4419 | 5151 | on_rq = p->on_rq; |
051a1d1a | 5152 | running = task_current(rq, p); |
0e1f3483 | 5153 | if (on_rq) |
2e1cb74a | 5154 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5155 | if (running) |
5156 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5157 | |
ca94c442 LP |
5158 | p->sched_reset_on_fork = reset_on_fork; |
5159 | ||
1da177e4 | 5160 | oldprio = p->prio; |
83ab0aa0 | 5161 | prev_class = p->sched_class; |
dd41f596 | 5162 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5163 | |
0e1f3483 HS |
5164 | if (running) |
5165 | p->sched_class->set_curr_task(rq); | |
da7a735e | 5166 | if (on_rq) |
dd41f596 | 5167 | activate_task(rq, p, 0); |
cb469845 | 5168 | |
da7a735e | 5169 | check_class_changed(rq, p, prev_class, oldprio); |
0122ec5b | 5170 | task_rq_unlock(rq, p, &flags); |
b29739f9 | 5171 | |
95e02ca9 TG |
5172 | rt_mutex_adjust_pi(p); |
5173 | ||
1da177e4 LT |
5174 | return 0; |
5175 | } | |
961ccddd RR |
5176 | |
5177 | /** | |
5178 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5179 | * @p: the task in question. | |
5180 | * @policy: new policy. | |
5181 | * @param: structure containing the new RT priority. | |
5182 | * | |
5183 | * NOTE that the task may be already dead. | |
5184 | */ | |
5185 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 5186 | const struct sched_param *param) |
961ccddd RR |
5187 | { |
5188 | return __sched_setscheduler(p, policy, param, true); | |
5189 | } | |
1da177e4 LT |
5190 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5191 | ||
961ccddd RR |
5192 | /** |
5193 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5194 | * @p: the task in question. | |
5195 | * @policy: new policy. | |
5196 | * @param: structure containing the new RT priority. | |
5197 | * | |
5198 | * Just like sched_setscheduler, only don't bother checking if the | |
5199 | * current context has permission. For example, this is needed in | |
5200 | * stop_machine(): we create temporary high priority worker threads, | |
5201 | * but our caller might not have that capability. | |
5202 | */ | |
5203 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 5204 | const struct sched_param *param) |
961ccddd RR |
5205 | { |
5206 | return __sched_setscheduler(p, policy, param, false); | |
5207 | } | |
5208 | ||
95cdf3b7 IM |
5209 | static int |
5210 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5211 | { |
1da177e4 LT |
5212 | struct sched_param lparam; |
5213 | struct task_struct *p; | |
36c8b586 | 5214 | int retval; |
1da177e4 LT |
5215 | |
5216 | if (!param || pid < 0) | |
5217 | return -EINVAL; | |
5218 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5219 | return -EFAULT; | |
5fe1d75f ON |
5220 | |
5221 | rcu_read_lock(); | |
5222 | retval = -ESRCH; | |
1da177e4 | 5223 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5224 | if (p != NULL) |
5225 | retval = sched_setscheduler(p, policy, &lparam); | |
5226 | rcu_read_unlock(); | |
36c8b586 | 5227 | |
1da177e4 LT |
5228 | return retval; |
5229 | } | |
5230 | ||
5231 | /** | |
5232 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5233 | * @pid: the pid in question. | |
5234 | * @policy: new policy. | |
5235 | * @param: structure containing the new RT priority. | |
5236 | */ | |
5add95d4 HC |
5237 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5238 | struct sched_param __user *, param) | |
1da177e4 | 5239 | { |
c21761f1 JB |
5240 | /* negative values for policy are not valid */ |
5241 | if (policy < 0) | |
5242 | return -EINVAL; | |
5243 | ||
1da177e4 LT |
5244 | return do_sched_setscheduler(pid, policy, param); |
5245 | } | |
5246 | ||
5247 | /** | |
5248 | * sys_sched_setparam - set/change the RT priority of a thread | |
5249 | * @pid: the pid in question. | |
5250 | * @param: structure containing the new RT priority. | |
5251 | */ | |
5add95d4 | 5252 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5253 | { |
5254 | return do_sched_setscheduler(pid, -1, param); | |
5255 | } | |
5256 | ||
5257 | /** | |
5258 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5259 | * @pid: the pid in question. | |
5260 | */ | |
5add95d4 | 5261 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5262 | { |
36c8b586 | 5263 | struct task_struct *p; |
3a5c359a | 5264 | int retval; |
1da177e4 LT |
5265 | |
5266 | if (pid < 0) | |
3a5c359a | 5267 | return -EINVAL; |
1da177e4 LT |
5268 | |
5269 | retval = -ESRCH; | |
5fe85be0 | 5270 | rcu_read_lock(); |
1da177e4 LT |
5271 | p = find_process_by_pid(pid); |
5272 | if (p) { | |
5273 | retval = security_task_getscheduler(p); | |
5274 | if (!retval) | |
ca94c442 LP |
5275 | retval = p->policy |
5276 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5277 | } |
5fe85be0 | 5278 | rcu_read_unlock(); |
1da177e4 LT |
5279 | return retval; |
5280 | } | |
5281 | ||
5282 | /** | |
ca94c442 | 5283 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5284 | * @pid: the pid in question. |
5285 | * @param: structure containing the RT priority. | |
5286 | */ | |
5add95d4 | 5287 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5288 | { |
5289 | struct sched_param lp; | |
36c8b586 | 5290 | struct task_struct *p; |
3a5c359a | 5291 | int retval; |
1da177e4 LT |
5292 | |
5293 | if (!param || pid < 0) | |
3a5c359a | 5294 | return -EINVAL; |
1da177e4 | 5295 | |
5fe85be0 | 5296 | rcu_read_lock(); |
1da177e4 LT |
5297 | p = find_process_by_pid(pid); |
5298 | retval = -ESRCH; | |
5299 | if (!p) | |
5300 | goto out_unlock; | |
5301 | ||
5302 | retval = security_task_getscheduler(p); | |
5303 | if (retval) | |
5304 | goto out_unlock; | |
5305 | ||
5306 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5307 | rcu_read_unlock(); |
1da177e4 LT |
5308 | |
5309 | /* | |
5310 | * This one might sleep, we cannot do it with a spinlock held ... | |
5311 | */ | |
5312 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5313 | ||
1da177e4 LT |
5314 | return retval; |
5315 | ||
5316 | out_unlock: | |
5fe85be0 | 5317 | rcu_read_unlock(); |
1da177e4 LT |
5318 | return retval; |
5319 | } | |
5320 | ||
96f874e2 | 5321 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5322 | { |
5a16f3d3 | 5323 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5324 | struct task_struct *p; |
5325 | int retval; | |
1da177e4 | 5326 | |
95402b38 | 5327 | get_online_cpus(); |
23f5d142 | 5328 | rcu_read_lock(); |
1da177e4 LT |
5329 | |
5330 | p = find_process_by_pid(pid); | |
5331 | if (!p) { | |
23f5d142 | 5332 | rcu_read_unlock(); |
95402b38 | 5333 | put_online_cpus(); |
1da177e4 LT |
5334 | return -ESRCH; |
5335 | } | |
5336 | ||
23f5d142 | 5337 | /* Prevent p going away */ |
1da177e4 | 5338 | get_task_struct(p); |
23f5d142 | 5339 | rcu_read_unlock(); |
1da177e4 | 5340 | |
5a16f3d3 RR |
5341 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5342 | retval = -ENOMEM; | |
5343 | goto out_put_task; | |
5344 | } | |
5345 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5346 | retval = -ENOMEM; | |
5347 | goto out_free_cpus_allowed; | |
5348 | } | |
1da177e4 | 5349 | retval = -EPERM; |
b0e77598 | 5350 | if (!check_same_owner(p) && !task_ns_capable(p, CAP_SYS_NICE)) |
1da177e4 LT |
5351 | goto out_unlock; |
5352 | ||
b0ae1981 | 5353 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5354 | if (retval) |
5355 | goto out_unlock; | |
5356 | ||
5a16f3d3 RR |
5357 | cpuset_cpus_allowed(p, cpus_allowed); |
5358 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5359 | again: |
5a16f3d3 | 5360 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5361 | |
8707d8b8 | 5362 | if (!retval) { |
5a16f3d3 RR |
5363 | cpuset_cpus_allowed(p, cpus_allowed); |
5364 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5365 | /* |
5366 | * We must have raced with a concurrent cpuset | |
5367 | * update. Just reset the cpus_allowed to the | |
5368 | * cpuset's cpus_allowed | |
5369 | */ | |
5a16f3d3 | 5370 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5371 | goto again; |
5372 | } | |
5373 | } | |
1da177e4 | 5374 | out_unlock: |
5a16f3d3 RR |
5375 | free_cpumask_var(new_mask); |
5376 | out_free_cpus_allowed: | |
5377 | free_cpumask_var(cpus_allowed); | |
5378 | out_put_task: | |
1da177e4 | 5379 | put_task_struct(p); |
95402b38 | 5380 | put_online_cpus(); |
1da177e4 LT |
5381 | return retval; |
5382 | } | |
5383 | ||
5384 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5385 | struct cpumask *new_mask) |
1da177e4 | 5386 | { |
96f874e2 RR |
5387 | if (len < cpumask_size()) |
5388 | cpumask_clear(new_mask); | |
5389 | else if (len > cpumask_size()) | |
5390 | len = cpumask_size(); | |
5391 | ||
1da177e4 LT |
5392 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5393 | } | |
5394 | ||
5395 | /** | |
5396 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5397 | * @pid: pid of the process | |
5398 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5399 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5400 | */ | |
5add95d4 HC |
5401 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5402 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5403 | { |
5a16f3d3 | 5404 | cpumask_var_t new_mask; |
1da177e4 LT |
5405 | int retval; |
5406 | ||
5a16f3d3 RR |
5407 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5408 | return -ENOMEM; | |
1da177e4 | 5409 | |
5a16f3d3 RR |
5410 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5411 | if (retval == 0) | |
5412 | retval = sched_setaffinity(pid, new_mask); | |
5413 | free_cpumask_var(new_mask); | |
5414 | return retval; | |
1da177e4 LT |
5415 | } |
5416 | ||
96f874e2 | 5417 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5418 | { |
36c8b586 | 5419 | struct task_struct *p; |
31605683 | 5420 | unsigned long flags; |
1da177e4 | 5421 | int retval; |
1da177e4 | 5422 | |
95402b38 | 5423 | get_online_cpus(); |
23f5d142 | 5424 | rcu_read_lock(); |
1da177e4 LT |
5425 | |
5426 | retval = -ESRCH; | |
5427 | p = find_process_by_pid(pid); | |
5428 | if (!p) | |
5429 | goto out_unlock; | |
5430 | ||
e7834f8f DQ |
5431 | retval = security_task_getscheduler(p); |
5432 | if (retval) | |
5433 | goto out_unlock; | |
5434 | ||
013fdb80 | 5435 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
96f874e2 | 5436 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
013fdb80 | 5437 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5438 | |
5439 | out_unlock: | |
23f5d142 | 5440 | rcu_read_unlock(); |
95402b38 | 5441 | put_online_cpus(); |
1da177e4 | 5442 | |
9531b62f | 5443 | return retval; |
1da177e4 LT |
5444 | } |
5445 | ||
5446 | /** | |
5447 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5448 | * @pid: pid of the process | |
5449 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5450 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5451 | */ | |
5add95d4 HC |
5452 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5453 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5454 | { |
5455 | int ret; | |
f17c8607 | 5456 | cpumask_var_t mask; |
1da177e4 | 5457 | |
84fba5ec | 5458 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5459 | return -EINVAL; |
5460 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5461 | return -EINVAL; |
5462 | ||
f17c8607 RR |
5463 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5464 | return -ENOMEM; | |
1da177e4 | 5465 | |
f17c8607 RR |
5466 | ret = sched_getaffinity(pid, mask); |
5467 | if (ret == 0) { | |
8bc037fb | 5468 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5469 | |
5470 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5471 | ret = -EFAULT; |
5472 | else | |
cd3d8031 | 5473 | ret = retlen; |
f17c8607 RR |
5474 | } |
5475 | free_cpumask_var(mask); | |
1da177e4 | 5476 | |
f17c8607 | 5477 | return ret; |
1da177e4 LT |
5478 | } |
5479 | ||
5480 | /** | |
5481 | * sys_sched_yield - yield the current processor to other threads. | |
5482 | * | |
dd41f596 IM |
5483 | * This function yields the current CPU to other tasks. If there are no |
5484 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5485 | */ |
5add95d4 | 5486 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5487 | { |
70b97a7f | 5488 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5489 | |
2d72376b | 5490 | schedstat_inc(rq, yld_count); |
4530d7ab | 5491 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5492 | |
5493 | /* | |
5494 | * Since we are going to call schedule() anyway, there's | |
5495 | * no need to preempt or enable interrupts: | |
5496 | */ | |
5497 | __release(rq->lock); | |
8a25d5de | 5498 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5499 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5500 | preempt_enable_no_resched(); |
5501 | ||
5502 | schedule(); | |
5503 | ||
5504 | return 0; | |
5505 | } | |
5506 | ||
d86ee480 PZ |
5507 | static inline int should_resched(void) |
5508 | { | |
5509 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5510 | } | |
5511 | ||
e7b38404 | 5512 | static void __cond_resched(void) |
1da177e4 | 5513 | { |
e7aaaa69 FW |
5514 | add_preempt_count(PREEMPT_ACTIVE); |
5515 | schedule(); | |
5516 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5517 | } |
5518 | ||
02b67cc3 | 5519 | int __sched _cond_resched(void) |
1da177e4 | 5520 | { |
d86ee480 | 5521 | if (should_resched()) { |
1da177e4 LT |
5522 | __cond_resched(); |
5523 | return 1; | |
5524 | } | |
5525 | return 0; | |
5526 | } | |
02b67cc3 | 5527 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5528 | |
5529 | /* | |
613afbf8 | 5530 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5531 | * call schedule, and on return reacquire the lock. |
5532 | * | |
41a2d6cf | 5533 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5534 | * operations here to prevent schedule() from being called twice (once via |
5535 | * spin_unlock(), once by hand). | |
5536 | */ | |
613afbf8 | 5537 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5538 | { |
d86ee480 | 5539 | int resched = should_resched(); |
6df3cecb JK |
5540 | int ret = 0; |
5541 | ||
f607c668 PZ |
5542 | lockdep_assert_held(lock); |
5543 | ||
95c354fe | 5544 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5545 | spin_unlock(lock); |
d86ee480 | 5546 | if (resched) |
95c354fe NP |
5547 | __cond_resched(); |
5548 | else | |
5549 | cpu_relax(); | |
6df3cecb | 5550 | ret = 1; |
1da177e4 | 5551 | spin_lock(lock); |
1da177e4 | 5552 | } |
6df3cecb | 5553 | return ret; |
1da177e4 | 5554 | } |
613afbf8 | 5555 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5556 | |
613afbf8 | 5557 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5558 | { |
5559 | BUG_ON(!in_softirq()); | |
5560 | ||
d86ee480 | 5561 | if (should_resched()) { |
98d82567 | 5562 | local_bh_enable(); |
1da177e4 LT |
5563 | __cond_resched(); |
5564 | local_bh_disable(); | |
5565 | return 1; | |
5566 | } | |
5567 | return 0; | |
5568 | } | |
613afbf8 | 5569 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5570 | |
1da177e4 LT |
5571 | /** |
5572 | * yield - yield the current processor to other threads. | |
5573 | * | |
72fd4a35 | 5574 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5575 | * thread runnable and calls sys_sched_yield(). |
5576 | */ | |
5577 | void __sched yield(void) | |
5578 | { | |
5579 | set_current_state(TASK_RUNNING); | |
5580 | sys_sched_yield(); | |
5581 | } | |
1da177e4 LT |
5582 | EXPORT_SYMBOL(yield); |
5583 | ||
d95f4122 MG |
5584 | /** |
5585 | * yield_to - yield the current processor to another thread in | |
5586 | * your thread group, or accelerate that thread toward the | |
5587 | * processor it's on. | |
16addf95 RD |
5588 | * @p: target task |
5589 | * @preempt: whether task preemption is allowed or not | |
d95f4122 MG |
5590 | * |
5591 | * It's the caller's job to ensure that the target task struct | |
5592 | * can't go away on us before we can do any checks. | |
5593 | * | |
5594 | * Returns true if we indeed boosted the target task. | |
5595 | */ | |
5596 | bool __sched yield_to(struct task_struct *p, bool preempt) | |
5597 | { | |
5598 | struct task_struct *curr = current; | |
5599 | struct rq *rq, *p_rq; | |
5600 | unsigned long flags; | |
5601 | bool yielded = 0; | |
5602 | ||
5603 | local_irq_save(flags); | |
5604 | rq = this_rq(); | |
5605 | ||
5606 | again: | |
5607 | p_rq = task_rq(p); | |
5608 | double_rq_lock(rq, p_rq); | |
5609 | while (task_rq(p) != p_rq) { | |
5610 | double_rq_unlock(rq, p_rq); | |
5611 | goto again; | |
5612 | } | |
5613 | ||
5614 | if (!curr->sched_class->yield_to_task) | |
5615 | goto out; | |
5616 | ||
5617 | if (curr->sched_class != p->sched_class) | |
5618 | goto out; | |
5619 | ||
5620 | if (task_running(p_rq, p) || p->state) | |
5621 | goto out; | |
5622 | ||
5623 | yielded = curr->sched_class->yield_to_task(rq, p, preempt); | |
6d1cafd8 | 5624 | if (yielded) { |
d95f4122 | 5625 | schedstat_inc(rq, yld_count); |
6d1cafd8 VP |
5626 | /* |
5627 | * Make p's CPU reschedule; pick_next_entity takes care of | |
5628 | * fairness. | |
5629 | */ | |
5630 | if (preempt && rq != p_rq) | |
5631 | resched_task(p_rq->curr); | |
5632 | } | |
d95f4122 MG |
5633 | |
5634 | out: | |
5635 | double_rq_unlock(rq, p_rq); | |
5636 | local_irq_restore(flags); | |
5637 | ||
5638 | if (yielded) | |
5639 | schedule(); | |
5640 | ||
5641 | return yielded; | |
5642 | } | |
5643 | EXPORT_SYMBOL_GPL(yield_to); | |
5644 | ||
1da177e4 | 5645 | /* |
41a2d6cf | 5646 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5647 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5648 | */ |
5649 | void __sched io_schedule(void) | |
5650 | { | |
54d35f29 | 5651 | struct rq *rq = raw_rq(); |
1da177e4 | 5652 | |
0ff92245 | 5653 | delayacct_blkio_start(); |
1da177e4 | 5654 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5655 | blk_flush_plug(current); |
8f0dfc34 | 5656 | current->in_iowait = 1; |
1da177e4 | 5657 | schedule(); |
8f0dfc34 | 5658 | current->in_iowait = 0; |
1da177e4 | 5659 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5660 | delayacct_blkio_end(); |
1da177e4 | 5661 | } |
1da177e4 LT |
5662 | EXPORT_SYMBOL(io_schedule); |
5663 | ||
5664 | long __sched io_schedule_timeout(long timeout) | |
5665 | { | |
54d35f29 | 5666 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5667 | long ret; |
5668 | ||
0ff92245 | 5669 | delayacct_blkio_start(); |
1da177e4 | 5670 | atomic_inc(&rq->nr_iowait); |
73c10101 | 5671 | blk_flush_plug(current); |
8f0dfc34 | 5672 | current->in_iowait = 1; |
1da177e4 | 5673 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5674 | current->in_iowait = 0; |
1da177e4 | 5675 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5676 | delayacct_blkio_end(); |
1da177e4 LT |
5677 | return ret; |
5678 | } | |
5679 | ||
5680 | /** | |
5681 | * sys_sched_get_priority_max - return maximum RT priority. | |
5682 | * @policy: scheduling class. | |
5683 | * | |
5684 | * this syscall returns the maximum rt_priority that can be used | |
5685 | * by a given scheduling class. | |
5686 | */ | |
5add95d4 | 5687 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5688 | { |
5689 | int ret = -EINVAL; | |
5690 | ||
5691 | switch (policy) { | |
5692 | case SCHED_FIFO: | |
5693 | case SCHED_RR: | |
5694 | ret = MAX_USER_RT_PRIO-1; | |
5695 | break; | |
5696 | case SCHED_NORMAL: | |
b0a9499c | 5697 | case SCHED_BATCH: |
dd41f596 | 5698 | case SCHED_IDLE: |
1da177e4 LT |
5699 | ret = 0; |
5700 | break; | |
5701 | } | |
5702 | return ret; | |
5703 | } | |
5704 | ||
5705 | /** | |
5706 | * sys_sched_get_priority_min - return minimum RT priority. | |
5707 | * @policy: scheduling class. | |
5708 | * | |
5709 | * this syscall returns the minimum rt_priority that can be used | |
5710 | * by a given scheduling class. | |
5711 | */ | |
5add95d4 | 5712 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5713 | { |
5714 | int ret = -EINVAL; | |
5715 | ||
5716 | switch (policy) { | |
5717 | case SCHED_FIFO: | |
5718 | case SCHED_RR: | |
5719 | ret = 1; | |
5720 | break; | |
5721 | case SCHED_NORMAL: | |
b0a9499c | 5722 | case SCHED_BATCH: |
dd41f596 | 5723 | case SCHED_IDLE: |
1da177e4 LT |
5724 | ret = 0; |
5725 | } | |
5726 | return ret; | |
5727 | } | |
5728 | ||
5729 | /** | |
5730 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5731 | * @pid: pid of the process. | |
5732 | * @interval: userspace pointer to the timeslice value. | |
5733 | * | |
5734 | * this syscall writes the default timeslice value of a given process | |
5735 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5736 | */ | |
17da2bd9 | 5737 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5738 | struct timespec __user *, interval) |
1da177e4 | 5739 | { |
36c8b586 | 5740 | struct task_struct *p; |
a4ec24b4 | 5741 | unsigned int time_slice; |
dba091b9 TG |
5742 | unsigned long flags; |
5743 | struct rq *rq; | |
3a5c359a | 5744 | int retval; |
1da177e4 | 5745 | struct timespec t; |
1da177e4 LT |
5746 | |
5747 | if (pid < 0) | |
3a5c359a | 5748 | return -EINVAL; |
1da177e4 LT |
5749 | |
5750 | retval = -ESRCH; | |
1a551ae7 | 5751 | rcu_read_lock(); |
1da177e4 LT |
5752 | p = find_process_by_pid(pid); |
5753 | if (!p) | |
5754 | goto out_unlock; | |
5755 | ||
5756 | retval = security_task_getscheduler(p); | |
5757 | if (retval) | |
5758 | goto out_unlock; | |
5759 | ||
dba091b9 TG |
5760 | rq = task_rq_lock(p, &flags); |
5761 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
0122ec5b | 5762 | task_rq_unlock(rq, p, &flags); |
a4ec24b4 | 5763 | |
1a551ae7 | 5764 | rcu_read_unlock(); |
a4ec24b4 | 5765 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5766 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5767 | return retval; |
3a5c359a | 5768 | |
1da177e4 | 5769 | out_unlock: |
1a551ae7 | 5770 | rcu_read_unlock(); |
1da177e4 LT |
5771 | return retval; |
5772 | } | |
5773 | ||
7c731e0a | 5774 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5775 | |
82a1fcb9 | 5776 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5777 | { |
1da177e4 | 5778 | unsigned long free = 0; |
36c8b586 | 5779 | unsigned state; |
1da177e4 | 5780 | |
1da177e4 | 5781 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5782 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5783 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5784 | #if BITS_PER_LONG == 32 |
1da177e4 | 5785 | if (state == TASK_RUNNING) |
3df0fc5b | 5786 | printk(KERN_CONT " running "); |
1da177e4 | 5787 | else |
3df0fc5b | 5788 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5789 | #else |
5790 | if (state == TASK_RUNNING) | |
3df0fc5b | 5791 | printk(KERN_CONT " running task "); |
1da177e4 | 5792 | else |
3df0fc5b | 5793 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5794 | #endif |
5795 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5796 | free = stack_not_used(p); |
1da177e4 | 5797 | #endif |
3df0fc5b | 5798 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5799 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5800 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5801 | |
5fb5e6de | 5802 | show_stack(p, NULL); |
1da177e4 LT |
5803 | } |
5804 | ||
e59e2ae2 | 5805 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5806 | { |
36c8b586 | 5807 | struct task_struct *g, *p; |
1da177e4 | 5808 | |
4bd77321 | 5809 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5810 | printk(KERN_INFO |
5811 | " task PC stack pid father\n"); | |
1da177e4 | 5812 | #else |
3df0fc5b PZ |
5813 | printk(KERN_INFO |
5814 | " task PC stack pid father\n"); | |
1da177e4 LT |
5815 | #endif |
5816 | read_lock(&tasklist_lock); | |
5817 | do_each_thread(g, p) { | |
5818 | /* | |
5819 | * reset the NMI-timeout, listing all files on a slow | |
25985edc | 5820 | * console might take a lot of time: |
1da177e4 LT |
5821 | */ |
5822 | touch_nmi_watchdog(); | |
39bc89fd | 5823 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5824 | sched_show_task(p); |
1da177e4 LT |
5825 | } while_each_thread(g, p); |
5826 | ||
04c9167f JF |
5827 | touch_all_softlockup_watchdogs(); |
5828 | ||
dd41f596 IM |
5829 | #ifdef CONFIG_SCHED_DEBUG |
5830 | sysrq_sched_debug_show(); | |
5831 | #endif | |
1da177e4 | 5832 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5833 | /* |
5834 | * Only show locks if all tasks are dumped: | |
5835 | */ | |
93335a21 | 5836 | if (!state_filter) |
e59e2ae2 | 5837 | debug_show_all_locks(); |
1da177e4 LT |
5838 | } |
5839 | ||
1df21055 IM |
5840 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5841 | { | |
dd41f596 | 5842 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5843 | } |
5844 | ||
f340c0d1 IM |
5845 | /** |
5846 | * init_idle - set up an idle thread for a given CPU | |
5847 | * @idle: task in question | |
5848 | * @cpu: cpu the idle task belongs to | |
5849 | * | |
5850 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5851 | * flag, to make booting more robust. | |
5852 | */ | |
5c1e1767 | 5853 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5854 | { |
70b97a7f | 5855 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5856 | unsigned long flags; |
5857 | ||
05fa785c | 5858 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5859 | |
dd41f596 | 5860 | __sched_fork(idle); |
06b83b5f | 5861 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5862 | idle->se.exec_start = sched_clock(); |
5863 | ||
1e1b6c51 | 5864 | do_set_cpus_allowed(idle, cpumask_of(cpu)); |
6506cf6c PZ |
5865 | /* |
5866 | * We're having a chicken and egg problem, even though we are | |
5867 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5868 | * lockdep check in task_group() will fail. | |
5869 | * | |
5870 | * Similar case to sched_fork(). / Alternatively we could | |
5871 | * use task_rq_lock() here and obtain the other rq->lock. | |
5872 | * | |
5873 | * Silence PROVE_RCU | |
5874 | */ | |
5875 | rcu_read_lock(); | |
dd41f596 | 5876 | __set_task_cpu(idle, cpu); |
6506cf6c | 5877 | rcu_read_unlock(); |
1da177e4 | 5878 | |
1da177e4 | 5879 | rq->curr = rq->idle = idle; |
3ca7a440 PZ |
5880 | #if defined(CONFIG_SMP) |
5881 | idle->on_cpu = 1; | |
4866cde0 | 5882 | #endif |
05fa785c | 5883 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5884 | |
5885 | /* Set the preempt count _outside_ the spinlocks! */ | |
a1261f54 | 5886 | task_thread_info(idle)->preempt_count = 0; |
625f2a37 | 5887 | |
dd41f596 IM |
5888 | /* |
5889 | * The idle tasks have their own, simple scheduling class: | |
5890 | */ | |
5891 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5892 | ftrace_graph_init_idle_task(idle, cpu); |
1da177e4 LT |
5893 | } |
5894 | ||
5895 | /* | |
5896 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5897 | * indicates which cpus entered this state. This is used | |
5898 | * in the rcu update to wait only for active cpus. For system | |
5899 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5900 | * always be CPU_BITS_NONE. |
1da177e4 | 5901 | */ |
6a7b3dc3 | 5902 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5903 | |
19978ca6 IM |
5904 | /* |
5905 | * Increase the granularity value when there are more CPUs, | |
5906 | * because with more CPUs the 'effective latency' as visible | |
5907 | * to users decreases. But the relationship is not linear, | |
5908 | * so pick a second-best guess by going with the log2 of the | |
5909 | * number of CPUs. | |
5910 | * | |
5911 | * This idea comes from the SD scheduler of Con Kolivas: | |
5912 | */ | |
acb4a848 | 5913 | static int get_update_sysctl_factor(void) |
19978ca6 | 5914 | { |
4ca3ef71 | 5915 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5916 | unsigned int factor; |
5917 | ||
5918 | switch (sysctl_sched_tunable_scaling) { | |
5919 | case SCHED_TUNABLESCALING_NONE: | |
5920 | factor = 1; | |
5921 | break; | |
5922 | case SCHED_TUNABLESCALING_LINEAR: | |
5923 | factor = cpus; | |
5924 | break; | |
5925 | case SCHED_TUNABLESCALING_LOG: | |
5926 | default: | |
5927 | factor = 1 + ilog2(cpus); | |
5928 | break; | |
5929 | } | |
19978ca6 | 5930 | |
acb4a848 CE |
5931 | return factor; |
5932 | } | |
19978ca6 | 5933 | |
acb4a848 CE |
5934 | static void update_sysctl(void) |
5935 | { | |
5936 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5937 | |
0bcdcf28 CE |
5938 | #define SET_SYSCTL(name) \ |
5939 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5940 | SET_SYSCTL(sched_min_granularity); | |
5941 | SET_SYSCTL(sched_latency); | |
5942 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5943 | #undef SET_SYSCTL |
5944 | } | |
55cd5340 | 5945 | |
0bcdcf28 CE |
5946 | static inline void sched_init_granularity(void) |
5947 | { | |
5948 | update_sysctl(); | |
19978ca6 IM |
5949 | } |
5950 | ||
1da177e4 | 5951 | #ifdef CONFIG_SMP |
1e1b6c51 KM |
5952 | void do_set_cpus_allowed(struct task_struct *p, const struct cpumask *new_mask) |
5953 | { | |
5954 | if (p->sched_class && p->sched_class->set_cpus_allowed) | |
5955 | p->sched_class->set_cpus_allowed(p, new_mask); | |
5956 | else { | |
5957 | cpumask_copy(&p->cpus_allowed, new_mask); | |
5958 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
5959 | } | |
5960 | } | |
5961 | ||
1da177e4 LT |
5962 | /* |
5963 | * This is how migration works: | |
5964 | * | |
969c7921 TH |
5965 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5966 | * stop_one_cpu(). | |
5967 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5968 | * off the CPU) | |
5969 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5970 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5971 | * it and puts it into the right queue. |
969c7921 TH |
5972 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5973 | * is done. | |
1da177e4 LT |
5974 | */ |
5975 | ||
5976 | /* | |
5977 | * Change a given task's CPU affinity. Migrate the thread to a | |
5978 | * proper CPU and schedule it away if the CPU it's executing on | |
5979 | * is removed from the allowed bitmask. | |
5980 | * | |
5981 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5982 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5983 | * call is not atomic; no spinlocks may be held. |
5984 | */ | |
96f874e2 | 5985 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5986 | { |
5987 | unsigned long flags; | |
70b97a7f | 5988 | struct rq *rq; |
969c7921 | 5989 | unsigned int dest_cpu; |
48f24c4d | 5990 | int ret = 0; |
1da177e4 LT |
5991 | |
5992 | rq = task_rq_lock(p, &flags); | |
e2912009 | 5993 | |
db44fc01 YZ |
5994 | if (cpumask_equal(&p->cpus_allowed, new_mask)) |
5995 | goto out; | |
5996 | ||
6ad4c188 | 5997 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5998 | ret = -EINVAL; |
5999 | goto out; | |
6000 | } | |
6001 | ||
db44fc01 | 6002 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current)) { |
9985b0ba DR |
6003 | ret = -EINVAL; |
6004 | goto out; | |
6005 | } | |
6006 | ||
1e1b6c51 | 6007 | do_set_cpus_allowed(p, new_mask); |
73fe6aae | 6008 | |
1da177e4 | 6009 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 6010 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
6011 | goto out; |
6012 | ||
969c7921 | 6013 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
bd8e7dde | 6014 | if (p->on_rq) { |
969c7921 | 6015 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 | 6016 | /* Need help from migration thread: drop lock and wait. */ |
0122ec5b | 6017 | task_rq_unlock(rq, p, &flags); |
969c7921 | 6018 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
6019 | tlb_migrate_finish(p->mm); |
6020 | return 0; | |
6021 | } | |
6022 | out: | |
0122ec5b | 6023 | task_rq_unlock(rq, p, &flags); |
48f24c4d | 6024 | |
1da177e4 LT |
6025 | return ret; |
6026 | } | |
cd8ba7cd | 6027 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
6028 | |
6029 | /* | |
41a2d6cf | 6030 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
6031 | * this because either it can't run here any more (set_cpus_allowed() |
6032 | * away from this CPU, or CPU going down), or because we're | |
6033 | * attempting to rebalance this task on exec (sched_exec). | |
6034 | * | |
6035 | * So we race with normal scheduler movements, but that's OK, as long | |
6036 | * as the task is no longer on this CPU. | |
efc30814 KK |
6037 | * |
6038 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 6039 | */ |
efc30814 | 6040 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 6041 | { |
70b97a7f | 6042 | struct rq *rq_dest, *rq_src; |
e2912009 | 6043 | int ret = 0; |
1da177e4 | 6044 | |
e761b772 | 6045 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 6046 | return ret; |
1da177e4 LT |
6047 | |
6048 | rq_src = cpu_rq(src_cpu); | |
6049 | rq_dest = cpu_rq(dest_cpu); | |
6050 | ||
0122ec5b | 6051 | raw_spin_lock(&p->pi_lock); |
1da177e4 LT |
6052 | double_rq_lock(rq_src, rq_dest); |
6053 | /* Already moved. */ | |
6054 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 6055 | goto done; |
1da177e4 | 6056 | /* Affinity changed (again). */ |
96f874e2 | 6057 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 6058 | goto fail; |
1da177e4 | 6059 | |
e2912009 PZ |
6060 | /* |
6061 | * If we're not on a rq, the next wake-up will ensure we're | |
6062 | * placed properly. | |
6063 | */ | |
fd2f4419 | 6064 | if (p->on_rq) { |
2e1cb74a | 6065 | deactivate_task(rq_src, p, 0); |
e2912009 | 6066 | set_task_cpu(p, dest_cpu); |
dd41f596 | 6067 | activate_task(rq_dest, p, 0); |
15afe09b | 6068 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 6069 | } |
b1e38734 | 6070 | done: |
efc30814 | 6071 | ret = 1; |
b1e38734 | 6072 | fail: |
1da177e4 | 6073 | double_rq_unlock(rq_src, rq_dest); |
0122ec5b | 6074 | raw_spin_unlock(&p->pi_lock); |
efc30814 | 6075 | return ret; |
1da177e4 LT |
6076 | } |
6077 | ||
6078 | /* | |
969c7921 TH |
6079 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
6080 | * and performs thread migration by bumping thread off CPU then | |
6081 | * 'pushing' onto another runqueue. | |
1da177e4 | 6082 | */ |
969c7921 | 6083 | static int migration_cpu_stop(void *data) |
1da177e4 | 6084 | { |
969c7921 | 6085 | struct migration_arg *arg = data; |
f7b4cddc | 6086 | |
969c7921 TH |
6087 | /* |
6088 | * The original target cpu might have gone down and we might | |
6089 | * be on another cpu but it doesn't matter. | |
6090 | */ | |
f7b4cddc | 6091 | local_irq_disable(); |
969c7921 | 6092 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 6093 | local_irq_enable(); |
1da177e4 | 6094 | return 0; |
f7b4cddc ON |
6095 | } |
6096 | ||
1da177e4 | 6097 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 6098 | |
054b9108 | 6099 | /* |
48c5ccae PZ |
6100 | * Ensures that the idle task is using init_mm right before its cpu goes |
6101 | * offline. | |
054b9108 | 6102 | */ |
48c5ccae | 6103 | void idle_task_exit(void) |
1da177e4 | 6104 | { |
48c5ccae | 6105 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 6106 | |
48c5ccae | 6107 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 6108 | |
48c5ccae PZ |
6109 | if (mm != &init_mm) |
6110 | switch_mm(mm, &init_mm, current); | |
6111 | mmdrop(mm); | |
1da177e4 LT |
6112 | } |
6113 | ||
6114 | /* | |
6115 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
6116 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
6117 | * for performance reasons the counter is not stricly tracking tasks to | |
6118 | * their home CPUs. So we just add the counter to another CPU's counter, | |
6119 | * to keep the global sum constant after CPU-down: | |
6120 | */ | |
70b97a7f | 6121 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 6122 | { |
6ad4c188 | 6123 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 6124 | |
1da177e4 LT |
6125 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
6126 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
6127 | } |
6128 | ||
dd41f596 | 6129 | /* |
48c5ccae | 6130 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 6131 | */ |
48c5ccae | 6132 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 6133 | { |
48c5ccae PZ |
6134 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
6135 | rq->calc_load_active = 0; | |
1da177e4 LT |
6136 | } |
6137 | ||
48f24c4d | 6138 | /* |
48c5ccae PZ |
6139 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
6140 | * try_to_wake_up()->select_task_rq(). | |
6141 | * | |
6142 | * Called with rq->lock held even though we'er in stop_machine() and | |
6143 | * there's no concurrency possible, we hold the required locks anyway | |
6144 | * because of lock validation efforts. | |
1da177e4 | 6145 | */ |
48c5ccae | 6146 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 6147 | { |
70b97a7f | 6148 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
6149 | struct task_struct *next, *stop = rq->stop; |
6150 | int dest_cpu; | |
1da177e4 LT |
6151 | |
6152 | /* | |
48c5ccae PZ |
6153 | * Fudge the rq selection such that the below task selection loop |
6154 | * doesn't get stuck on the currently eligible stop task. | |
6155 | * | |
6156 | * We're currently inside stop_machine() and the rq is either stuck | |
6157 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
6158 | * either way we should never end up calling schedule() until we're | |
6159 | * done here. | |
1da177e4 | 6160 | */ |
48c5ccae | 6161 | rq->stop = NULL; |
48f24c4d | 6162 | |
dd41f596 | 6163 | for ( ; ; ) { |
48c5ccae PZ |
6164 | /* |
6165 | * There's this thread running, bail when that's the only | |
6166 | * remaining thread. | |
6167 | */ | |
6168 | if (rq->nr_running == 1) | |
dd41f596 | 6169 | break; |
48c5ccae | 6170 | |
b67802ea | 6171 | next = pick_next_task(rq); |
48c5ccae | 6172 | BUG_ON(!next); |
79c53799 | 6173 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 6174 | |
48c5ccae PZ |
6175 | /* Find suitable destination for @next, with force if needed. */ |
6176 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
6177 | raw_spin_unlock(&rq->lock); | |
6178 | ||
6179 | __migrate_task(next, dead_cpu, dest_cpu); | |
6180 | ||
6181 | raw_spin_lock(&rq->lock); | |
1da177e4 | 6182 | } |
dce48a84 | 6183 | |
48c5ccae | 6184 | rq->stop = stop; |
dce48a84 | 6185 | } |
48c5ccae | 6186 | |
1da177e4 LT |
6187 | #endif /* CONFIG_HOTPLUG_CPU */ |
6188 | ||
e692ab53 NP |
6189 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6190 | ||
6191 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6192 | { |
6193 | .procname = "sched_domain", | |
c57baf1e | 6194 | .mode = 0555, |
e0361851 | 6195 | }, |
56992309 | 6196 | {} |
e692ab53 NP |
6197 | }; |
6198 | ||
6199 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6200 | { |
6201 | .procname = "kernel", | |
c57baf1e | 6202 | .mode = 0555, |
e0361851 AD |
6203 | .child = sd_ctl_dir, |
6204 | }, | |
56992309 | 6205 | {} |
e692ab53 NP |
6206 | }; |
6207 | ||
6208 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6209 | { | |
6210 | struct ctl_table *entry = | |
5cf9f062 | 6211 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6212 | |
e692ab53 NP |
6213 | return entry; |
6214 | } | |
6215 | ||
6382bc90 MM |
6216 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6217 | { | |
cd790076 | 6218 | struct ctl_table *entry; |
6382bc90 | 6219 | |
cd790076 MM |
6220 | /* |
6221 | * In the intermediate directories, both the child directory and | |
6222 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6223 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6224 | * static strings and all have proc handlers. |
6225 | */ | |
6226 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6227 | if (entry->child) |
6228 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6229 | if (entry->proc_handler == NULL) |
6230 | kfree(entry->procname); | |
6231 | } | |
6382bc90 MM |
6232 | |
6233 | kfree(*tablep); | |
6234 | *tablep = NULL; | |
6235 | } | |
6236 | ||
e692ab53 | 6237 | static void |
e0361851 | 6238 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6239 | const char *procname, void *data, int maxlen, |
6240 | mode_t mode, proc_handler *proc_handler) | |
6241 | { | |
e692ab53 NP |
6242 | entry->procname = procname; |
6243 | entry->data = data; | |
6244 | entry->maxlen = maxlen; | |
6245 | entry->mode = mode; | |
6246 | entry->proc_handler = proc_handler; | |
6247 | } | |
6248 | ||
6249 | static struct ctl_table * | |
6250 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6251 | { | |
a5d8c348 | 6252 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6253 | |
ad1cdc1d MM |
6254 | if (table == NULL) |
6255 | return NULL; | |
6256 | ||
e0361851 | 6257 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6258 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6259 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6260 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6261 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6262 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6263 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6264 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6265 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6266 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6267 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6268 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6269 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6270 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6271 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6272 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6273 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6274 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6275 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6276 | &sd->cache_nice_tries, |
6277 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6278 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6279 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6280 | set_table_entry(&table[11], "name", sd->name, |
6281 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6282 | /* &table[12] is terminator */ | |
e692ab53 NP |
6283 | |
6284 | return table; | |
6285 | } | |
6286 | ||
9a4e7159 | 6287 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6288 | { |
6289 | struct ctl_table *entry, *table; | |
6290 | struct sched_domain *sd; | |
6291 | int domain_num = 0, i; | |
6292 | char buf[32]; | |
6293 | ||
6294 | for_each_domain(cpu, sd) | |
6295 | domain_num++; | |
6296 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6297 | if (table == NULL) |
6298 | return NULL; | |
e692ab53 NP |
6299 | |
6300 | i = 0; | |
6301 | for_each_domain(cpu, sd) { | |
6302 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6303 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6304 | entry->mode = 0555; |
e692ab53 NP |
6305 | entry->child = sd_alloc_ctl_domain_table(sd); |
6306 | entry++; | |
6307 | i++; | |
6308 | } | |
6309 | return table; | |
6310 | } | |
6311 | ||
6312 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6313 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6314 | { |
6ad4c188 | 6315 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6316 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6317 | char buf[32]; | |
6318 | ||
7378547f MM |
6319 | WARN_ON(sd_ctl_dir[0].child); |
6320 | sd_ctl_dir[0].child = entry; | |
6321 | ||
ad1cdc1d MM |
6322 | if (entry == NULL) |
6323 | return; | |
6324 | ||
6ad4c188 | 6325 | for_each_possible_cpu(i) { |
e692ab53 | 6326 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6327 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6328 | entry->mode = 0555; |
e692ab53 | 6329 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6330 | entry++; |
e692ab53 | 6331 | } |
7378547f MM |
6332 | |
6333 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6334 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6335 | } | |
6382bc90 | 6336 | |
7378547f | 6337 | /* may be called multiple times per register */ |
6382bc90 MM |
6338 | static void unregister_sched_domain_sysctl(void) |
6339 | { | |
7378547f MM |
6340 | if (sd_sysctl_header) |
6341 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6342 | sd_sysctl_header = NULL; |
7378547f MM |
6343 | if (sd_ctl_dir[0].child) |
6344 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6345 | } |
e692ab53 | 6346 | #else |
6382bc90 MM |
6347 | static void register_sched_domain_sysctl(void) |
6348 | { | |
6349 | } | |
6350 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6351 | { |
6352 | } | |
6353 | #endif | |
6354 | ||
1f11eb6a GH |
6355 | static void set_rq_online(struct rq *rq) |
6356 | { | |
6357 | if (!rq->online) { | |
6358 | const struct sched_class *class; | |
6359 | ||
c6c4927b | 6360 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6361 | rq->online = 1; |
6362 | ||
6363 | for_each_class(class) { | |
6364 | if (class->rq_online) | |
6365 | class->rq_online(rq); | |
6366 | } | |
6367 | } | |
6368 | } | |
6369 | ||
6370 | static void set_rq_offline(struct rq *rq) | |
6371 | { | |
6372 | if (rq->online) { | |
6373 | const struct sched_class *class; | |
6374 | ||
6375 | for_each_class(class) { | |
6376 | if (class->rq_offline) | |
6377 | class->rq_offline(rq); | |
6378 | } | |
6379 | ||
c6c4927b | 6380 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6381 | rq->online = 0; |
6382 | } | |
6383 | } | |
6384 | ||
1da177e4 LT |
6385 | /* |
6386 | * migration_call - callback that gets triggered when a CPU is added. | |
6387 | * Here we can start up the necessary migration thread for the new CPU. | |
6388 | */ | |
48f24c4d IM |
6389 | static int __cpuinit |
6390 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6391 | { |
48f24c4d | 6392 | int cpu = (long)hcpu; |
1da177e4 | 6393 | unsigned long flags; |
969c7921 | 6394 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6395 | |
48c5ccae | 6396 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6397 | |
1da177e4 | 6398 | case CPU_UP_PREPARE: |
a468d389 | 6399 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6400 | break; |
48f24c4d | 6401 | |
1da177e4 | 6402 | case CPU_ONLINE: |
1f94ef59 | 6403 | /* Update our root-domain */ |
05fa785c | 6404 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6405 | if (rq->rd) { |
c6c4927b | 6406 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6407 | |
6408 | set_rq_online(rq); | |
1f94ef59 | 6409 | } |
05fa785c | 6410 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6411 | break; |
48f24c4d | 6412 | |
1da177e4 | 6413 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6414 | case CPU_DYING: |
317f3941 | 6415 | sched_ttwu_pending(); |
57d885fe | 6416 | /* Update our root-domain */ |
05fa785c | 6417 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6418 | if (rq->rd) { |
c6c4927b | 6419 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6420 | set_rq_offline(rq); |
57d885fe | 6421 | } |
48c5ccae PZ |
6422 | migrate_tasks(cpu); |
6423 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6424 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6425 | |
6426 | migrate_nr_uninterruptible(rq); | |
6427 | calc_global_load_remove(rq); | |
57d885fe | 6428 | break; |
1da177e4 LT |
6429 | #endif |
6430 | } | |
49c022e6 PZ |
6431 | |
6432 | update_max_interval(); | |
6433 | ||
1da177e4 LT |
6434 | return NOTIFY_OK; |
6435 | } | |
6436 | ||
f38b0820 PM |
6437 | /* |
6438 | * Register at high priority so that task migration (migrate_all_tasks) | |
6439 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6440 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6441 | */ |
26c2143b | 6442 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6443 | .notifier_call = migration_call, |
50a323b7 | 6444 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6445 | }; |
6446 | ||
3a101d05 TH |
6447 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6448 | unsigned long action, void *hcpu) | |
6449 | { | |
6450 | switch (action & ~CPU_TASKS_FROZEN) { | |
6451 | case CPU_ONLINE: | |
6452 | case CPU_DOWN_FAILED: | |
6453 | set_cpu_active((long)hcpu, true); | |
6454 | return NOTIFY_OK; | |
6455 | default: | |
6456 | return NOTIFY_DONE; | |
6457 | } | |
6458 | } | |
6459 | ||
6460 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6461 | unsigned long action, void *hcpu) | |
6462 | { | |
6463 | switch (action & ~CPU_TASKS_FROZEN) { | |
6464 | case CPU_DOWN_PREPARE: | |
6465 | set_cpu_active((long)hcpu, false); | |
6466 | return NOTIFY_OK; | |
6467 | default: | |
6468 | return NOTIFY_DONE; | |
6469 | } | |
6470 | } | |
6471 | ||
7babe8db | 6472 | static int __init migration_init(void) |
1da177e4 LT |
6473 | { |
6474 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6475 | int err; |
48f24c4d | 6476 | |
3a101d05 | 6477 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6478 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6479 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6480 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6481 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6482 | |
3a101d05 TH |
6483 | /* Register cpu active notifiers */ |
6484 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6485 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6486 | ||
a004cd42 | 6487 | return 0; |
1da177e4 | 6488 | } |
7babe8db | 6489 | early_initcall(migration_init); |
1da177e4 LT |
6490 | #endif |
6491 | ||
6492 | #ifdef CONFIG_SMP | |
476f3534 | 6493 | |
4cb98839 PZ |
6494 | static cpumask_var_t sched_domains_tmpmask; /* sched_domains_mutex */ |
6495 | ||
3e9830dc | 6496 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6497 | |
f6630114 MT |
6498 | static __read_mostly int sched_domain_debug_enabled; |
6499 | ||
6500 | static int __init sched_domain_debug_setup(char *str) | |
6501 | { | |
6502 | sched_domain_debug_enabled = 1; | |
6503 | ||
6504 | return 0; | |
6505 | } | |
6506 | early_param("sched_debug", sched_domain_debug_setup); | |
6507 | ||
7c16ec58 | 6508 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6509 | struct cpumask *groupmask) |
1da177e4 | 6510 | { |
4dcf6aff | 6511 | struct sched_group *group = sd->groups; |
434d53b0 | 6512 | char str[256]; |
1da177e4 | 6513 | |
968ea6d8 | 6514 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6515 | cpumask_clear(groupmask); |
4dcf6aff IM |
6516 | |
6517 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6518 | ||
6519 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6520 | printk("does not load-balance\n"); |
4dcf6aff | 6521 | if (sd->parent) |
3df0fc5b PZ |
6522 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6523 | " has parent"); | |
4dcf6aff | 6524 | return -1; |
41c7ce9a NP |
6525 | } |
6526 | ||
3df0fc5b | 6527 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6528 | |
758b2cdc | 6529 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6530 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6531 | "CPU%d\n", cpu); | |
4dcf6aff | 6532 | } |
758b2cdc | 6533 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6534 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6535 | " CPU%d\n", cpu); | |
4dcf6aff | 6536 | } |
1da177e4 | 6537 | |
4dcf6aff | 6538 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6539 | do { |
4dcf6aff | 6540 | if (!group) { |
3df0fc5b PZ |
6541 | printk("\n"); |
6542 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6543 | break; |
6544 | } | |
6545 | ||
18a3885f | 6546 | if (!group->cpu_power) { |
3df0fc5b PZ |
6547 | printk(KERN_CONT "\n"); |
6548 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6549 | "set\n"); | |
4dcf6aff IM |
6550 | break; |
6551 | } | |
1da177e4 | 6552 | |
758b2cdc | 6553 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6554 | printk(KERN_CONT "\n"); |
6555 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6556 | break; |
6557 | } | |
1da177e4 | 6558 | |
758b2cdc | 6559 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6560 | printk(KERN_CONT "\n"); |
6561 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6562 | break; |
6563 | } | |
1da177e4 | 6564 | |
758b2cdc | 6565 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6566 | |
968ea6d8 | 6567 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6568 | |
3df0fc5b | 6569 | printk(KERN_CONT " %s", str); |
1399fa78 | 6570 | if (group->cpu_power != SCHED_POWER_SCALE) { |
3df0fc5b PZ |
6571 | printk(KERN_CONT " (cpu_power = %d)", |
6572 | group->cpu_power); | |
381512cf | 6573 | } |
1da177e4 | 6574 | |
4dcf6aff IM |
6575 | group = group->next; |
6576 | } while (group != sd->groups); | |
3df0fc5b | 6577 | printk(KERN_CONT "\n"); |
1da177e4 | 6578 | |
758b2cdc | 6579 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6580 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6581 | |
758b2cdc RR |
6582 | if (sd->parent && |
6583 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6584 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6585 | "of domain->span\n"); | |
4dcf6aff IM |
6586 | return 0; |
6587 | } | |
1da177e4 | 6588 | |
4dcf6aff IM |
6589 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6590 | { | |
6591 | int level = 0; | |
1da177e4 | 6592 | |
f6630114 MT |
6593 | if (!sched_domain_debug_enabled) |
6594 | return; | |
6595 | ||
4dcf6aff IM |
6596 | if (!sd) { |
6597 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6598 | return; | |
6599 | } | |
1da177e4 | 6600 | |
4dcf6aff IM |
6601 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6602 | ||
6603 | for (;;) { | |
4cb98839 | 6604 | if (sched_domain_debug_one(sd, cpu, level, sched_domains_tmpmask)) |
4dcf6aff | 6605 | break; |
1da177e4 LT |
6606 | level++; |
6607 | sd = sd->parent; | |
33859f7f | 6608 | if (!sd) |
4dcf6aff IM |
6609 | break; |
6610 | } | |
1da177e4 | 6611 | } |
6d6bc0ad | 6612 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6613 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6614 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6615 | |
1a20ff27 | 6616 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6617 | { |
758b2cdc | 6618 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6619 | return 1; |
6620 | ||
6621 | /* Following flags need at least 2 groups */ | |
6622 | if (sd->flags & (SD_LOAD_BALANCE | | |
6623 | SD_BALANCE_NEWIDLE | | |
6624 | SD_BALANCE_FORK | | |
89c4710e SS |
6625 | SD_BALANCE_EXEC | |
6626 | SD_SHARE_CPUPOWER | | |
6627 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6628 | if (sd->groups != sd->groups->next) |
6629 | return 0; | |
6630 | } | |
6631 | ||
6632 | /* Following flags don't use groups */ | |
c88d5910 | 6633 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6634 | return 0; |
6635 | ||
6636 | return 1; | |
6637 | } | |
6638 | ||
48f24c4d IM |
6639 | static int |
6640 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6641 | { |
6642 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6643 | ||
6644 | if (sd_degenerate(parent)) | |
6645 | return 1; | |
6646 | ||
758b2cdc | 6647 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6648 | return 0; |
6649 | ||
245af2c7 SS |
6650 | /* Flags needing groups don't count if only 1 group in parent */ |
6651 | if (parent->groups == parent->groups->next) { | |
6652 | pflags &= ~(SD_LOAD_BALANCE | | |
6653 | SD_BALANCE_NEWIDLE | | |
6654 | SD_BALANCE_FORK | | |
89c4710e SS |
6655 | SD_BALANCE_EXEC | |
6656 | SD_SHARE_CPUPOWER | | |
6657 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6658 | if (nr_node_ids == 1) |
6659 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6660 | } |
6661 | if (~cflags & pflags) | |
6662 | return 0; | |
6663 | ||
6664 | return 1; | |
6665 | } | |
6666 | ||
dce840a0 | 6667 | static void free_rootdomain(struct rcu_head *rcu) |
c6c4927b | 6668 | { |
dce840a0 | 6669 | struct root_domain *rd = container_of(rcu, struct root_domain, rcu); |
047106ad | 6670 | |
68e74568 | 6671 | cpupri_cleanup(&rd->cpupri); |
c6c4927b RR |
6672 | free_cpumask_var(rd->rto_mask); |
6673 | free_cpumask_var(rd->online); | |
6674 | free_cpumask_var(rd->span); | |
6675 | kfree(rd); | |
6676 | } | |
6677 | ||
57d885fe GH |
6678 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6679 | { | |
a0490fa3 | 6680 | struct root_domain *old_rd = NULL; |
57d885fe | 6681 | unsigned long flags; |
57d885fe | 6682 | |
05fa785c | 6683 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6684 | |
6685 | if (rq->rd) { | |
a0490fa3 | 6686 | old_rd = rq->rd; |
57d885fe | 6687 | |
c6c4927b | 6688 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6689 | set_rq_offline(rq); |
57d885fe | 6690 | |
c6c4927b | 6691 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6692 | |
a0490fa3 IM |
6693 | /* |
6694 | * If we dont want to free the old_rt yet then | |
6695 | * set old_rd to NULL to skip the freeing later | |
6696 | * in this function: | |
6697 | */ | |
6698 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6699 | old_rd = NULL; | |
57d885fe GH |
6700 | } |
6701 | ||
6702 | atomic_inc(&rd->refcount); | |
6703 | rq->rd = rd; | |
6704 | ||
c6c4927b | 6705 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6706 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6707 | set_rq_online(rq); |
57d885fe | 6708 | |
05fa785c | 6709 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6710 | |
6711 | if (old_rd) | |
dce840a0 | 6712 | call_rcu_sched(&old_rd->rcu, free_rootdomain); |
57d885fe GH |
6713 | } |
6714 | ||
68c38fc3 | 6715 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6716 | { |
6717 | memset(rd, 0, sizeof(*rd)); | |
6718 | ||
68c38fc3 | 6719 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6720 | goto out; |
68c38fc3 | 6721 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6722 | goto free_span; |
68c38fc3 | 6723 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6724 | goto free_online; |
6e0534f2 | 6725 | |
68c38fc3 | 6726 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6727 | goto free_rto_mask; |
c6c4927b | 6728 | return 0; |
6e0534f2 | 6729 | |
68e74568 RR |
6730 | free_rto_mask: |
6731 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6732 | free_online: |
6733 | free_cpumask_var(rd->online); | |
6734 | free_span: | |
6735 | free_cpumask_var(rd->span); | |
0c910d28 | 6736 | out: |
c6c4927b | 6737 | return -ENOMEM; |
57d885fe GH |
6738 | } |
6739 | ||
6740 | static void init_defrootdomain(void) | |
6741 | { | |
68c38fc3 | 6742 | init_rootdomain(&def_root_domain); |
c6c4927b | 6743 | |
57d885fe GH |
6744 | atomic_set(&def_root_domain.refcount, 1); |
6745 | } | |
6746 | ||
dc938520 | 6747 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6748 | { |
6749 | struct root_domain *rd; | |
6750 | ||
6751 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6752 | if (!rd) | |
6753 | return NULL; | |
6754 | ||
68c38fc3 | 6755 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6756 | kfree(rd); |
6757 | return NULL; | |
6758 | } | |
57d885fe GH |
6759 | |
6760 | return rd; | |
6761 | } | |
6762 | ||
dce840a0 PZ |
6763 | static void free_sched_domain(struct rcu_head *rcu) |
6764 | { | |
6765 | struct sched_domain *sd = container_of(rcu, struct sched_domain, rcu); | |
6766 | if (atomic_dec_and_test(&sd->groups->ref)) | |
6767 | kfree(sd->groups); | |
6768 | kfree(sd); | |
6769 | } | |
6770 | ||
6771 | static void destroy_sched_domain(struct sched_domain *sd, int cpu) | |
6772 | { | |
6773 | call_rcu(&sd->rcu, free_sched_domain); | |
6774 | } | |
6775 | ||
6776 | static void destroy_sched_domains(struct sched_domain *sd, int cpu) | |
6777 | { | |
6778 | for (; sd; sd = sd->parent) | |
6779 | destroy_sched_domain(sd, cpu); | |
6780 | } | |
6781 | ||
1da177e4 | 6782 | /* |
0eab9146 | 6783 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6784 | * hold the hotplug lock. |
6785 | */ | |
0eab9146 IM |
6786 | static void |
6787 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6788 | { |
70b97a7f | 6789 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6790 | struct sched_domain *tmp; |
6791 | ||
6792 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6793 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6794 | struct sched_domain *parent = tmp->parent; |
6795 | if (!parent) | |
6796 | break; | |
f29c9b1c | 6797 | |
1a848870 | 6798 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6799 | tmp->parent = parent->parent; |
1a848870 SS |
6800 | if (parent->parent) |
6801 | parent->parent->child = tmp; | |
dce840a0 | 6802 | destroy_sched_domain(parent, cpu); |
f29c9b1c LZ |
6803 | } else |
6804 | tmp = tmp->parent; | |
245af2c7 SS |
6805 | } |
6806 | ||
1a848870 | 6807 | if (sd && sd_degenerate(sd)) { |
dce840a0 | 6808 | tmp = sd; |
245af2c7 | 6809 | sd = sd->parent; |
dce840a0 | 6810 | destroy_sched_domain(tmp, cpu); |
1a848870 SS |
6811 | if (sd) |
6812 | sd->child = NULL; | |
6813 | } | |
1da177e4 | 6814 | |
4cb98839 | 6815 | sched_domain_debug(sd, cpu); |
1da177e4 | 6816 | |
57d885fe | 6817 | rq_attach_root(rq, rd); |
dce840a0 | 6818 | tmp = rq->sd; |
674311d5 | 6819 | rcu_assign_pointer(rq->sd, sd); |
dce840a0 | 6820 | destroy_sched_domains(tmp, cpu); |
1da177e4 LT |
6821 | } |
6822 | ||
6823 | /* cpus with isolated domains */ | |
dcc30a35 | 6824 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6825 | |
6826 | /* Setup the mask of cpus configured for isolated domains */ | |
6827 | static int __init isolated_cpu_setup(char *str) | |
6828 | { | |
bdddd296 | 6829 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6830 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6831 | return 1; |
6832 | } | |
6833 | ||
8927f494 | 6834 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 | 6835 | |
9c1cfda2 | 6836 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6837 | |
9c1cfda2 | 6838 | #ifdef CONFIG_NUMA |
198e2f18 | 6839 | |
9c1cfda2 JH |
6840 | /** |
6841 | * find_next_best_node - find the next node to include in a sched_domain | |
6842 | * @node: node whose sched_domain we're building | |
6843 | * @used_nodes: nodes already in the sched_domain | |
6844 | * | |
41a2d6cf | 6845 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6846 | * finds the closest node not already in the @used_nodes map. |
6847 | * | |
6848 | * Should use nodemask_t. | |
6849 | */ | |
c5f59f08 | 6850 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 | 6851 | { |
7142d17e | 6852 | int i, n, val, min_val, best_node = -1; |
9c1cfda2 JH |
6853 | |
6854 | min_val = INT_MAX; | |
6855 | ||
076ac2af | 6856 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6857 | /* Start at @node */ |
076ac2af | 6858 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6859 | |
6860 | if (!nr_cpus_node(n)) | |
6861 | continue; | |
6862 | ||
6863 | /* Skip already used nodes */ | |
c5f59f08 | 6864 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6865 | continue; |
6866 | ||
6867 | /* Simple min distance search */ | |
6868 | val = node_distance(node, n); | |
6869 | ||
6870 | if (val < min_val) { | |
6871 | min_val = val; | |
6872 | best_node = n; | |
6873 | } | |
6874 | } | |
6875 | ||
7142d17e HD |
6876 | if (best_node != -1) |
6877 | node_set(best_node, *used_nodes); | |
9c1cfda2 JH |
6878 | return best_node; |
6879 | } | |
6880 | ||
6881 | /** | |
6882 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6883 | * @node: node whose cpumask we're constructing | |
73486722 | 6884 | * @span: resulting cpumask |
9c1cfda2 | 6885 | * |
41a2d6cf | 6886 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6887 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6888 | * out optimally. | |
6889 | */ | |
96f874e2 | 6890 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6891 | { |
c5f59f08 | 6892 | nodemask_t used_nodes; |
48f24c4d | 6893 | int i; |
9c1cfda2 | 6894 | |
6ca09dfc | 6895 | cpumask_clear(span); |
c5f59f08 | 6896 | nodes_clear(used_nodes); |
9c1cfda2 | 6897 | |
6ca09dfc | 6898 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6899 | node_set(node, used_nodes); |
9c1cfda2 JH |
6900 | |
6901 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6902 | int next_node = find_next_best_node(node, &used_nodes); |
7142d17e HD |
6903 | if (next_node < 0) |
6904 | break; | |
6ca09dfc | 6905 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6906 | } |
9c1cfda2 | 6907 | } |
d3081f52 PZ |
6908 | |
6909 | static const struct cpumask *cpu_node_mask(int cpu) | |
6910 | { | |
6911 | lockdep_assert_held(&sched_domains_mutex); | |
6912 | ||
6913 | sched_domain_node_span(cpu_to_node(cpu), sched_domains_tmpmask); | |
6914 | ||
6915 | return sched_domains_tmpmask; | |
6916 | } | |
2c402dc3 PZ |
6917 | |
6918 | static const struct cpumask *cpu_allnodes_mask(int cpu) | |
6919 | { | |
6920 | return cpu_possible_mask; | |
6921 | } | |
6d6bc0ad | 6922 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6923 | |
d3081f52 PZ |
6924 | static const struct cpumask *cpu_cpu_mask(int cpu) |
6925 | { | |
6926 | return cpumask_of_node(cpu_to_node(cpu)); | |
6927 | } | |
6928 | ||
5c45bf27 | 6929 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6930 | |
dce840a0 PZ |
6931 | struct sd_data { |
6932 | struct sched_domain **__percpu sd; | |
6933 | struct sched_group **__percpu sg; | |
6934 | }; | |
6935 | ||
49a02c51 | 6936 | struct s_data { |
21d42ccf | 6937 | struct sched_domain ** __percpu sd; |
49a02c51 AH |
6938 | struct root_domain *rd; |
6939 | }; | |
6940 | ||
2109b99e | 6941 | enum s_alloc { |
2109b99e | 6942 | sa_rootdomain, |
21d42ccf | 6943 | sa_sd, |
dce840a0 | 6944 | sa_sd_storage, |
2109b99e AH |
6945 | sa_none, |
6946 | }; | |
6947 | ||
54ab4ff4 PZ |
6948 | struct sched_domain_topology_level; |
6949 | ||
6950 | typedef struct sched_domain *(*sched_domain_init_f)(struct sched_domain_topology_level *tl, int cpu); | |
eb7a74e6 PZ |
6951 | typedef const struct cpumask *(*sched_domain_mask_f)(int cpu); |
6952 | ||
6953 | struct sched_domain_topology_level { | |
2c402dc3 PZ |
6954 | sched_domain_init_f init; |
6955 | sched_domain_mask_f mask; | |
54ab4ff4 | 6956 | struct sd_data data; |
eb7a74e6 PZ |
6957 | }; |
6958 | ||
9c1cfda2 | 6959 | /* |
dce840a0 | 6960 | * Assumes the sched_domain tree is fully constructed |
9c1cfda2 | 6961 | */ |
dce840a0 | 6962 | static int get_group(int cpu, struct sd_data *sdd, struct sched_group **sg) |
1da177e4 | 6963 | { |
dce840a0 PZ |
6964 | struct sched_domain *sd = *per_cpu_ptr(sdd->sd, cpu); |
6965 | struct sched_domain *child = sd->child; | |
1da177e4 | 6966 | |
dce840a0 PZ |
6967 | if (child) |
6968 | cpu = cpumask_first(sched_domain_span(child)); | |
1e9f28fa | 6969 | |
6711cab4 | 6970 | if (sg) |
dce840a0 PZ |
6971 | *sg = *per_cpu_ptr(sdd->sg, cpu); |
6972 | ||
6973 | return cpu; | |
1e9f28fa | 6974 | } |
1e9f28fa | 6975 | |
01a08546 | 6976 | /* |
dce840a0 PZ |
6977 | * build_sched_groups takes the cpumask we wish to span, and a pointer |
6978 | * to a function which identifies what group(along with sched group) a CPU | |
6979 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids | |
6980 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
6981 | * | |
6982 | * build_sched_groups will build a circular linked list of the groups | |
6983 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6984 | * and ->cpu_power to 0. | |
01a08546 | 6985 | */ |
dce840a0 | 6986 | static void |
f96225fd | 6987 | build_sched_groups(struct sched_domain *sd) |
1da177e4 | 6988 | { |
dce840a0 PZ |
6989 | struct sched_group *first = NULL, *last = NULL; |
6990 | struct sd_data *sdd = sd->private; | |
6991 | const struct cpumask *span = sched_domain_span(sd); | |
f96225fd | 6992 | struct cpumask *covered; |
dce840a0 | 6993 | int i; |
9c1cfda2 | 6994 | |
f96225fd PZ |
6995 | lockdep_assert_held(&sched_domains_mutex); |
6996 | covered = sched_domains_tmpmask; | |
6997 | ||
dce840a0 | 6998 | cpumask_clear(covered); |
6711cab4 | 6999 | |
dce840a0 PZ |
7000 | for_each_cpu(i, span) { |
7001 | struct sched_group *sg; | |
7002 | int group = get_group(i, sdd, &sg); | |
7003 | int j; | |
6711cab4 | 7004 | |
dce840a0 PZ |
7005 | if (cpumask_test_cpu(i, covered)) |
7006 | continue; | |
6711cab4 | 7007 | |
dce840a0 PZ |
7008 | cpumask_clear(sched_group_cpus(sg)); |
7009 | sg->cpu_power = 0; | |
0601a88d | 7010 | |
dce840a0 PZ |
7011 | for_each_cpu(j, span) { |
7012 | if (get_group(j, sdd, NULL) != group) | |
7013 | continue; | |
0601a88d | 7014 | |
dce840a0 PZ |
7015 | cpumask_set_cpu(j, covered); |
7016 | cpumask_set_cpu(j, sched_group_cpus(sg)); | |
7017 | } | |
0601a88d | 7018 | |
dce840a0 PZ |
7019 | if (!first) |
7020 | first = sg; | |
7021 | if (last) | |
7022 | last->next = sg; | |
7023 | last = sg; | |
7024 | } | |
7025 | last->next = first; | |
0601a88d | 7026 | } |
51888ca2 | 7027 | |
89c4710e SS |
7028 | /* |
7029 | * Initialize sched groups cpu_power. | |
7030 | * | |
7031 | * cpu_power indicates the capacity of sched group, which is used while | |
7032 | * distributing the load between different sched groups in a sched domain. | |
7033 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7034 | * there are asymmetries in the topology. If there are asymmetries, group | |
7035 | * having more cpu_power will pickup more load compared to the group having | |
7036 | * less cpu_power. | |
89c4710e SS |
7037 | */ |
7038 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7039 | { | |
89c4710e SS |
7040 | WARN_ON(!sd || !sd->groups); |
7041 | ||
13318a71 | 7042 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
7043 | return; |
7044 | ||
aae6d3dd SS |
7045 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
7046 | ||
d274cb30 | 7047 | update_group_power(sd, cpu); |
89c4710e SS |
7048 | } |
7049 | ||
7c16ec58 MT |
7050 | /* |
7051 | * Initializers for schedule domains | |
7052 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7053 | */ | |
7054 | ||
a5d8c348 IM |
7055 | #ifdef CONFIG_SCHED_DEBUG |
7056 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7057 | #else | |
7058 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7059 | #endif | |
7060 | ||
54ab4ff4 PZ |
7061 | #define SD_INIT_FUNC(type) \ |
7062 | static noinline struct sched_domain * \ | |
7063 | sd_init_##type(struct sched_domain_topology_level *tl, int cpu) \ | |
7064 | { \ | |
7065 | struct sched_domain *sd = *per_cpu_ptr(tl->data.sd, cpu); \ | |
7066 | *sd = SD_##type##_INIT; \ | |
54ab4ff4 PZ |
7067 | SD_INIT_NAME(sd, type); \ |
7068 | sd->private = &tl->data; \ | |
7069 | return sd; \ | |
7c16ec58 MT |
7070 | } |
7071 | ||
7072 | SD_INIT_FUNC(CPU) | |
7073 | #ifdef CONFIG_NUMA | |
7074 | SD_INIT_FUNC(ALLNODES) | |
7075 | SD_INIT_FUNC(NODE) | |
7076 | #endif | |
7077 | #ifdef CONFIG_SCHED_SMT | |
7078 | SD_INIT_FUNC(SIBLING) | |
7079 | #endif | |
7080 | #ifdef CONFIG_SCHED_MC | |
7081 | SD_INIT_FUNC(MC) | |
7082 | #endif | |
01a08546 HC |
7083 | #ifdef CONFIG_SCHED_BOOK |
7084 | SD_INIT_FUNC(BOOK) | |
7085 | #endif | |
7c16ec58 | 7086 | |
1d3504fc | 7087 | static int default_relax_domain_level = -1; |
60495e77 | 7088 | int sched_domain_level_max; |
1d3504fc HS |
7089 | |
7090 | static int __init setup_relax_domain_level(char *str) | |
7091 | { | |
30e0e178 LZ |
7092 | unsigned long val; |
7093 | ||
7094 | val = simple_strtoul(str, NULL, 0); | |
60495e77 | 7095 | if (val < sched_domain_level_max) |
30e0e178 LZ |
7096 | default_relax_domain_level = val; |
7097 | ||
1d3504fc HS |
7098 | return 1; |
7099 | } | |
7100 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7101 | ||
7102 | static void set_domain_attribute(struct sched_domain *sd, | |
7103 | struct sched_domain_attr *attr) | |
7104 | { | |
7105 | int request; | |
7106 | ||
7107 | if (!attr || attr->relax_domain_level < 0) { | |
7108 | if (default_relax_domain_level < 0) | |
7109 | return; | |
7110 | else | |
7111 | request = default_relax_domain_level; | |
7112 | } else | |
7113 | request = attr->relax_domain_level; | |
7114 | if (request < sd->level) { | |
7115 | /* turn off idle balance on this domain */ | |
c88d5910 | 7116 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7117 | } else { |
7118 | /* turn on idle balance on this domain */ | |
c88d5910 | 7119 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7120 | } |
7121 | } | |
7122 | ||
54ab4ff4 PZ |
7123 | static void __sdt_free(const struct cpumask *cpu_map); |
7124 | static int __sdt_alloc(const struct cpumask *cpu_map); | |
7125 | ||
2109b99e AH |
7126 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7127 | const struct cpumask *cpu_map) | |
7128 | { | |
7129 | switch (what) { | |
2109b99e | 7130 | case sa_rootdomain: |
822ff793 PZ |
7131 | if (!atomic_read(&d->rd->refcount)) |
7132 | free_rootdomain(&d->rd->rcu); /* fall through */ | |
21d42ccf PZ |
7133 | case sa_sd: |
7134 | free_percpu(d->sd); /* fall through */ | |
dce840a0 | 7135 | case sa_sd_storage: |
54ab4ff4 | 7136 | __sdt_free(cpu_map); /* fall through */ |
2109b99e AH |
7137 | case sa_none: |
7138 | break; | |
7139 | } | |
7140 | } | |
3404c8d9 | 7141 | |
2109b99e AH |
7142 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7143 | const struct cpumask *cpu_map) | |
7144 | { | |
dce840a0 PZ |
7145 | memset(d, 0, sizeof(*d)); |
7146 | ||
54ab4ff4 PZ |
7147 | if (__sdt_alloc(cpu_map)) |
7148 | return sa_sd_storage; | |
dce840a0 PZ |
7149 | d->sd = alloc_percpu(struct sched_domain *); |
7150 | if (!d->sd) | |
7151 | return sa_sd_storage; | |
2109b99e | 7152 | d->rd = alloc_rootdomain(); |
dce840a0 | 7153 | if (!d->rd) |
21d42ccf | 7154 | return sa_sd; |
2109b99e AH |
7155 | return sa_rootdomain; |
7156 | } | |
57d885fe | 7157 | |
dce840a0 PZ |
7158 | /* |
7159 | * NULL the sd_data elements we've used to build the sched_domain and | |
7160 | * sched_group structure so that the subsequent __free_domain_allocs() | |
7161 | * will not free the data we're using. | |
7162 | */ | |
7163 | static void claim_allocations(int cpu, struct sched_domain *sd) | |
7164 | { | |
7165 | struct sd_data *sdd = sd->private; | |
7166 | struct sched_group *sg = sd->groups; | |
7167 | ||
7168 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, cpu) != sd); | |
7169 | *per_cpu_ptr(sdd->sd, cpu) = NULL; | |
7170 | ||
7171 | if (cpu == cpumask_first(sched_group_cpus(sg))) { | |
7172 | WARN_ON_ONCE(*per_cpu_ptr(sdd->sg, cpu) != sg); | |
7173 | *per_cpu_ptr(sdd->sg, cpu) = NULL; | |
7174 | } | |
7175 | } | |
7176 | ||
2c402dc3 PZ |
7177 | #ifdef CONFIG_SCHED_SMT |
7178 | static const struct cpumask *cpu_smt_mask(int cpu) | |
7f4588f3 | 7179 | { |
2c402dc3 | 7180 | return topology_thread_cpumask(cpu); |
3bd65a80 | 7181 | } |
2c402dc3 | 7182 | #endif |
7f4588f3 | 7183 | |
d069b916 PZ |
7184 | /* |
7185 | * Topology list, bottom-up. | |
7186 | */ | |
2c402dc3 | 7187 | static struct sched_domain_topology_level default_topology[] = { |
d069b916 PZ |
7188 | #ifdef CONFIG_SCHED_SMT |
7189 | { sd_init_SIBLING, cpu_smt_mask, }, | |
01a08546 | 7190 | #endif |
1e9f28fa | 7191 | #ifdef CONFIG_SCHED_MC |
2c402dc3 | 7192 | { sd_init_MC, cpu_coregroup_mask, }, |
1e9f28fa | 7193 | #endif |
d069b916 PZ |
7194 | #ifdef CONFIG_SCHED_BOOK |
7195 | { sd_init_BOOK, cpu_book_mask, }, | |
7196 | #endif | |
7197 | { sd_init_CPU, cpu_cpu_mask, }, | |
7198 | #ifdef CONFIG_NUMA | |
7199 | { sd_init_NODE, cpu_node_mask, }, | |
7200 | { sd_init_ALLNODES, cpu_allnodes_mask, }, | |
1da177e4 | 7201 | #endif |
eb7a74e6 PZ |
7202 | { NULL, }, |
7203 | }; | |
7204 | ||
7205 | static struct sched_domain_topology_level *sched_domain_topology = default_topology; | |
7206 | ||
54ab4ff4 PZ |
7207 | static int __sdt_alloc(const struct cpumask *cpu_map) |
7208 | { | |
7209 | struct sched_domain_topology_level *tl; | |
7210 | int j; | |
7211 | ||
7212 | for (tl = sched_domain_topology; tl->init; tl++) { | |
7213 | struct sd_data *sdd = &tl->data; | |
7214 | ||
7215 | sdd->sd = alloc_percpu(struct sched_domain *); | |
7216 | if (!sdd->sd) | |
7217 | return -ENOMEM; | |
7218 | ||
7219 | sdd->sg = alloc_percpu(struct sched_group *); | |
7220 | if (!sdd->sg) | |
7221 | return -ENOMEM; | |
7222 | ||
7223 | for_each_cpu(j, cpu_map) { | |
7224 | struct sched_domain *sd; | |
7225 | struct sched_group *sg; | |
7226 | ||
7227 | sd = kzalloc_node(sizeof(struct sched_domain) + cpumask_size(), | |
7228 | GFP_KERNEL, cpu_to_node(j)); | |
7229 | if (!sd) | |
7230 | return -ENOMEM; | |
7231 | ||
7232 | *per_cpu_ptr(sdd->sd, j) = sd; | |
7233 | ||
7234 | sg = kzalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7235 | GFP_KERNEL, cpu_to_node(j)); | |
7236 | if (!sg) | |
7237 | return -ENOMEM; | |
7238 | ||
7239 | *per_cpu_ptr(sdd->sg, j) = sg; | |
7240 | } | |
7241 | } | |
7242 | ||
7243 | return 0; | |
7244 | } | |
7245 | ||
7246 | static void __sdt_free(const struct cpumask *cpu_map) | |
7247 | { | |
7248 | struct sched_domain_topology_level *tl; | |
7249 | int j; | |
7250 | ||
7251 | for (tl = sched_domain_topology; tl->init; tl++) { | |
7252 | struct sd_data *sdd = &tl->data; | |
7253 | ||
7254 | for_each_cpu(j, cpu_map) { | |
7255 | kfree(*per_cpu_ptr(sdd->sd, j)); | |
7256 | kfree(*per_cpu_ptr(sdd->sg, j)); | |
7257 | } | |
7258 | free_percpu(sdd->sd); | |
7259 | free_percpu(sdd->sg); | |
7260 | } | |
7261 | } | |
7262 | ||
2c402dc3 PZ |
7263 | struct sched_domain *build_sched_domain(struct sched_domain_topology_level *tl, |
7264 | struct s_data *d, const struct cpumask *cpu_map, | |
d069b916 | 7265 | struct sched_domain_attr *attr, struct sched_domain *child, |
2c402dc3 PZ |
7266 | int cpu) |
7267 | { | |
54ab4ff4 | 7268 | struct sched_domain *sd = tl->init(tl, cpu); |
2c402dc3 | 7269 | if (!sd) |
d069b916 | 7270 | return child; |
2c402dc3 PZ |
7271 | |
7272 | set_domain_attribute(sd, attr); | |
7273 | cpumask_and(sched_domain_span(sd), cpu_map, tl->mask(cpu)); | |
60495e77 PZ |
7274 | if (child) { |
7275 | sd->level = child->level + 1; | |
7276 | sched_domain_level_max = max(sched_domain_level_max, sd->level); | |
d069b916 | 7277 | child->parent = sd; |
60495e77 | 7278 | } |
d069b916 | 7279 | sd->child = child; |
2c402dc3 PZ |
7280 | |
7281 | return sd; | |
7282 | } | |
7283 | ||
2109b99e AH |
7284 | /* |
7285 | * Build sched domains for a given set of cpus and attach the sched domains | |
7286 | * to the individual cpus | |
7287 | */ | |
dce840a0 PZ |
7288 | static int build_sched_domains(const struct cpumask *cpu_map, |
7289 | struct sched_domain_attr *attr) | |
2109b99e AH |
7290 | { |
7291 | enum s_alloc alloc_state = sa_none; | |
dce840a0 | 7292 | struct sched_domain *sd; |
2109b99e | 7293 | struct s_data d; |
822ff793 | 7294 | int i, ret = -ENOMEM; |
9c1cfda2 | 7295 | |
2109b99e AH |
7296 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7297 | if (alloc_state != sa_rootdomain) | |
7298 | goto error; | |
9c1cfda2 | 7299 | |
dce840a0 | 7300 | /* Set up domains for cpus specified by the cpu_map. */ |
abcd083a | 7301 | for_each_cpu(i, cpu_map) { |
eb7a74e6 PZ |
7302 | struct sched_domain_topology_level *tl; |
7303 | ||
3bd65a80 | 7304 | sd = NULL; |
2c402dc3 PZ |
7305 | for (tl = sched_domain_topology; tl->init; tl++) |
7306 | sd = build_sched_domain(tl, &d, cpu_map, attr, sd, i); | |
d274cb30 | 7307 | |
d069b916 PZ |
7308 | while (sd->child) |
7309 | sd = sd->child; | |
7310 | ||
21d42ccf | 7311 | *per_cpu_ptr(d.sd, i) = sd; |
dce840a0 PZ |
7312 | } |
7313 | ||
7314 | /* Build the groups for the domains */ | |
7315 | for_each_cpu(i, cpu_map) { | |
7316 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { | |
7317 | sd->span_weight = cpumask_weight(sched_domain_span(sd)); | |
7318 | get_group(i, sd->private, &sd->groups); | |
7319 | atomic_inc(&sd->groups->ref); | |
21d42ccf | 7320 | |
dce840a0 PZ |
7321 | if (i != cpumask_first(sched_domain_span(sd))) |
7322 | continue; | |
7323 | ||
f96225fd | 7324 | build_sched_groups(sd); |
1cf51902 | 7325 | } |
a06dadbe | 7326 | } |
9c1cfda2 | 7327 | |
1da177e4 | 7328 | /* Calculate CPU power for physical packages and nodes */ |
a9c9a9b6 PZ |
7329 | for (i = nr_cpumask_bits-1; i >= 0; i--) { |
7330 | if (!cpumask_test_cpu(i, cpu_map)) | |
7331 | continue; | |
9c1cfda2 | 7332 | |
dce840a0 PZ |
7333 | for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) { |
7334 | claim_allocations(i, sd); | |
cd4ea6ae | 7335 | init_sched_groups_power(i, sd); |
dce840a0 | 7336 | } |
f712c0c7 | 7337 | } |
9c1cfda2 | 7338 | |
1da177e4 | 7339 | /* Attach the domains */ |
dce840a0 | 7340 | rcu_read_lock(); |
abcd083a | 7341 | for_each_cpu(i, cpu_map) { |
21d42ccf | 7342 | sd = *per_cpu_ptr(d.sd, i); |
49a02c51 | 7343 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7344 | } |
dce840a0 | 7345 | rcu_read_unlock(); |
51888ca2 | 7346 | |
822ff793 | 7347 | ret = 0; |
51888ca2 | 7348 | error: |
2109b99e | 7349 | __free_domain_allocs(&d, alloc_state, cpu_map); |
822ff793 | 7350 | return ret; |
1da177e4 | 7351 | } |
029190c5 | 7352 | |
acc3f5d7 | 7353 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7354 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7355 | static struct sched_domain_attr *dattr_cur; |
7356 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7357 | |
7358 | /* | |
7359 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7360 | * cpumask) fails, then fallback to a single sched domain, |
7361 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7362 | */ |
4212823f | 7363 | static cpumask_var_t fallback_doms; |
029190c5 | 7364 | |
ee79d1bd HC |
7365 | /* |
7366 | * arch_update_cpu_topology lets virtualized architectures update the | |
7367 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7368 | * or 0 if it stayed the same. | |
7369 | */ | |
7370 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7371 | { |
ee79d1bd | 7372 | return 0; |
22e52b07 HC |
7373 | } |
7374 | ||
acc3f5d7 RR |
7375 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7376 | { | |
7377 | int i; | |
7378 | cpumask_var_t *doms; | |
7379 | ||
7380 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7381 | if (!doms) | |
7382 | return NULL; | |
7383 | for (i = 0; i < ndoms; i++) { | |
7384 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7385 | free_sched_domains(doms, i); | |
7386 | return NULL; | |
7387 | } | |
7388 | } | |
7389 | return doms; | |
7390 | } | |
7391 | ||
7392 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7393 | { | |
7394 | unsigned int i; | |
7395 | for (i = 0; i < ndoms; i++) | |
7396 | free_cpumask_var(doms[i]); | |
7397 | kfree(doms); | |
7398 | } | |
7399 | ||
1a20ff27 | 7400 | /* |
41a2d6cf | 7401 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7402 | * For now this just excludes isolated cpus, but could be used to |
7403 | * exclude other special cases in the future. | |
1a20ff27 | 7404 | */ |
c4a8849a | 7405 | static int init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7406 | { |
7378547f MM |
7407 | int err; |
7408 | ||
22e52b07 | 7409 | arch_update_cpu_topology(); |
029190c5 | 7410 | ndoms_cur = 1; |
acc3f5d7 | 7411 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7412 | if (!doms_cur) |
acc3f5d7 RR |
7413 | doms_cur = &fallback_doms; |
7414 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7415 | dattr_cur = NULL; |
dce840a0 | 7416 | err = build_sched_domains(doms_cur[0], NULL); |
6382bc90 | 7417 | register_sched_domain_sysctl(); |
7378547f MM |
7418 | |
7419 | return err; | |
1a20ff27 DG |
7420 | } |
7421 | ||
1a20ff27 DG |
7422 | /* |
7423 | * Detach sched domains from a group of cpus specified in cpu_map | |
7424 | * These cpus will now be attached to the NULL domain | |
7425 | */ | |
96f874e2 | 7426 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 DG |
7427 | { |
7428 | int i; | |
7429 | ||
dce840a0 | 7430 | rcu_read_lock(); |
abcd083a | 7431 | for_each_cpu(i, cpu_map) |
57d885fe | 7432 | cpu_attach_domain(NULL, &def_root_domain, i); |
dce840a0 | 7433 | rcu_read_unlock(); |
1a20ff27 DG |
7434 | } |
7435 | ||
1d3504fc HS |
7436 | /* handle null as "default" */ |
7437 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7438 | struct sched_domain_attr *new, int idx_new) | |
7439 | { | |
7440 | struct sched_domain_attr tmp; | |
7441 | ||
7442 | /* fast path */ | |
7443 | if (!new && !cur) | |
7444 | return 1; | |
7445 | ||
7446 | tmp = SD_ATTR_INIT; | |
7447 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7448 | new ? (new + idx_new) : &tmp, | |
7449 | sizeof(struct sched_domain_attr)); | |
7450 | } | |
7451 | ||
029190c5 PJ |
7452 | /* |
7453 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7454 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7455 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7456 | * It destroys each deleted domain and builds each new domain. | |
7457 | * | |
acc3f5d7 | 7458 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7459 | * The masks don't intersect (don't overlap.) We should setup one |
7460 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7461 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7462 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7463 | * it as it is. | |
7464 | * | |
acc3f5d7 RR |
7465 | * The passed in 'doms_new' should be allocated using |
7466 | * alloc_sched_domains. This routine takes ownership of it and will | |
7467 | * free_sched_domains it when done with it. If the caller failed the | |
7468 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7469 | * and partition_sched_domains() will fallback to the single partition | |
7470 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7471 | * |
96f874e2 | 7472 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7473 | * ndoms_new == 0 is a special case for destroying existing domains, |
7474 | * and it will not create the default domain. | |
dfb512ec | 7475 | * |
029190c5 PJ |
7476 | * Call with hotplug lock held |
7477 | */ | |
acc3f5d7 | 7478 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7479 | struct sched_domain_attr *dattr_new) |
029190c5 | 7480 | { |
dfb512ec | 7481 | int i, j, n; |
d65bd5ec | 7482 | int new_topology; |
029190c5 | 7483 | |
712555ee | 7484 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7485 | |
7378547f MM |
7486 | /* always unregister in case we don't destroy any domains */ |
7487 | unregister_sched_domain_sysctl(); | |
7488 | ||
d65bd5ec HC |
7489 | /* Let architecture update cpu core mappings. */ |
7490 | new_topology = arch_update_cpu_topology(); | |
7491 | ||
dfb512ec | 7492 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7493 | |
7494 | /* Destroy deleted domains */ | |
7495 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7496 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7497 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7498 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7499 | goto match1; |
7500 | } | |
7501 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7502 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7503 | match1: |
7504 | ; | |
7505 | } | |
7506 | ||
e761b772 MK |
7507 | if (doms_new == NULL) { |
7508 | ndoms_cur = 0; | |
acc3f5d7 | 7509 | doms_new = &fallback_doms; |
6ad4c188 | 7510 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7511 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7512 | } |
7513 | ||
029190c5 PJ |
7514 | /* Build new domains */ |
7515 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7516 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7517 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7518 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7519 | goto match2; |
7520 | } | |
7521 | /* no match - add a new doms_new */ | |
dce840a0 | 7522 | build_sched_domains(doms_new[i], dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7523 | match2: |
7524 | ; | |
7525 | } | |
7526 | ||
7527 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7528 | if (doms_cur != &fallback_doms) |
7529 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7530 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7531 | doms_cur = doms_new; |
1d3504fc | 7532 | dattr_cur = dattr_new; |
029190c5 | 7533 | ndoms_cur = ndoms_new; |
7378547f MM |
7534 | |
7535 | register_sched_domain_sysctl(); | |
a1835615 | 7536 | |
712555ee | 7537 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7538 | } |
7539 | ||
5c45bf27 | 7540 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c4a8849a | 7541 | static void reinit_sched_domains(void) |
5c45bf27 | 7542 | { |
95402b38 | 7543 | get_online_cpus(); |
dfb512ec MK |
7544 | |
7545 | /* Destroy domains first to force the rebuild */ | |
7546 | partition_sched_domains(0, NULL, NULL); | |
7547 | ||
e761b772 | 7548 | rebuild_sched_domains(); |
95402b38 | 7549 | put_online_cpus(); |
5c45bf27 SS |
7550 | } |
7551 | ||
7552 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7553 | { | |
afb8a9b7 | 7554 | unsigned int level = 0; |
5c45bf27 | 7555 | |
afb8a9b7 GS |
7556 | if (sscanf(buf, "%u", &level) != 1) |
7557 | return -EINVAL; | |
7558 | ||
7559 | /* | |
7560 | * level is always be positive so don't check for | |
7561 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7562 | * What happens on 0 or 1 byte write, | |
7563 | * need to check for count as well? | |
7564 | */ | |
7565 | ||
7566 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7567 | return -EINVAL; |
7568 | ||
7569 | if (smt) | |
afb8a9b7 | 7570 | sched_smt_power_savings = level; |
5c45bf27 | 7571 | else |
afb8a9b7 | 7572 | sched_mc_power_savings = level; |
5c45bf27 | 7573 | |
c4a8849a | 7574 | reinit_sched_domains(); |
5c45bf27 | 7575 | |
c70f22d2 | 7576 | return count; |
5c45bf27 SS |
7577 | } |
7578 | ||
5c45bf27 | 7579 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7580 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7581 | struct sysdev_class_attribute *attr, |
f718cd4a | 7582 | char *page) |
5c45bf27 SS |
7583 | { |
7584 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7585 | } | |
f718cd4a | 7586 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7587 | struct sysdev_class_attribute *attr, |
48f24c4d | 7588 | const char *buf, size_t count) |
5c45bf27 SS |
7589 | { |
7590 | return sched_power_savings_store(buf, count, 0); | |
7591 | } | |
f718cd4a AK |
7592 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7593 | sched_mc_power_savings_show, | |
7594 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7595 | #endif |
7596 | ||
7597 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7598 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7599 | struct sysdev_class_attribute *attr, |
f718cd4a | 7600 | char *page) |
5c45bf27 SS |
7601 | { |
7602 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7603 | } | |
f718cd4a | 7604 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7605 | struct sysdev_class_attribute *attr, |
48f24c4d | 7606 | const char *buf, size_t count) |
5c45bf27 SS |
7607 | { |
7608 | return sched_power_savings_store(buf, count, 1); | |
7609 | } | |
f718cd4a AK |
7610 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7611 | sched_smt_power_savings_show, | |
6707de00 AB |
7612 | sched_smt_power_savings_store); |
7613 | #endif | |
7614 | ||
39aac648 | 7615 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7616 | { |
7617 | int err = 0; | |
7618 | ||
7619 | #ifdef CONFIG_SCHED_SMT | |
7620 | if (smt_capable()) | |
7621 | err = sysfs_create_file(&cls->kset.kobj, | |
7622 | &attr_sched_smt_power_savings.attr); | |
7623 | #endif | |
7624 | #ifdef CONFIG_SCHED_MC | |
7625 | if (!err && mc_capable()) | |
7626 | err = sysfs_create_file(&cls->kset.kobj, | |
7627 | &attr_sched_mc_power_savings.attr); | |
7628 | #endif | |
7629 | return err; | |
7630 | } | |
6d6bc0ad | 7631 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7632 | |
1da177e4 | 7633 | /* |
3a101d05 TH |
7634 | * Update cpusets according to cpu_active mask. If cpusets are |
7635 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7636 | * around partition_sched_domains(). | |
1da177e4 | 7637 | */ |
0b2e918a TH |
7638 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7639 | void *hcpu) | |
e761b772 | 7640 | { |
3a101d05 | 7641 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7642 | case CPU_ONLINE: |
6ad4c188 | 7643 | case CPU_DOWN_FAILED: |
3a101d05 | 7644 | cpuset_update_active_cpus(); |
e761b772 | 7645 | return NOTIFY_OK; |
3a101d05 TH |
7646 | default: |
7647 | return NOTIFY_DONE; | |
7648 | } | |
7649 | } | |
e761b772 | 7650 | |
0b2e918a TH |
7651 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7652 | void *hcpu) | |
3a101d05 TH |
7653 | { |
7654 | switch (action & ~CPU_TASKS_FROZEN) { | |
7655 | case CPU_DOWN_PREPARE: | |
7656 | cpuset_update_active_cpus(); | |
7657 | return NOTIFY_OK; | |
e761b772 MK |
7658 | default: |
7659 | return NOTIFY_DONE; | |
7660 | } | |
7661 | } | |
e761b772 MK |
7662 | |
7663 | static int update_runtime(struct notifier_block *nfb, | |
7664 | unsigned long action, void *hcpu) | |
1da177e4 | 7665 | { |
7def2be1 PZ |
7666 | int cpu = (int)(long)hcpu; |
7667 | ||
1da177e4 | 7668 | switch (action) { |
1da177e4 | 7669 | case CPU_DOWN_PREPARE: |
8bb78442 | 7670 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7671 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7672 | return NOTIFY_OK; |
7673 | ||
1da177e4 | 7674 | case CPU_DOWN_FAILED: |
8bb78442 | 7675 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7676 | case CPU_ONLINE: |
8bb78442 | 7677 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7678 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7679 | return NOTIFY_OK; |
7680 | ||
1da177e4 LT |
7681 | default: |
7682 | return NOTIFY_DONE; | |
7683 | } | |
1da177e4 | 7684 | } |
1da177e4 LT |
7685 | |
7686 | void __init sched_init_smp(void) | |
7687 | { | |
dcc30a35 RR |
7688 | cpumask_var_t non_isolated_cpus; |
7689 | ||
7690 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7691 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7692 | |
95402b38 | 7693 | get_online_cpus(); |
712555ee | 7694 | mutex_lock(&sched_domains_mutex); |
c4a8849a | 7695 | init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7696 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7697 | if (cpumask_empty(non_isolated_cpus)) | |
7698 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7699 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7700 | put_online_cpus(); |
e761b772 | 7701 | |
3a101d05 TH |
7702 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7703 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7704 | |
7705 | /* RT runtime code needs to handle some hotplug events */ | |
7706 | hotcpu_notifier(update_runtime, 0); | |
7707 | ||
b328ca18 | 7708 | init_hrtick(); |
5c1e1767 NP |
7709 | |
7710 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7711 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7712 | BUG(); |
19978ca6 | 7713 | sched_init_granularity(); |
dcc30a35 | 7714 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7715 | |
0e3900e6 | 7716 | init_sched_rt_class(); |
1da177e4 LT |
7717 | } |
7718 | #else | |
7719 | void __init sched_init_smp(void) | |
7720 | { | |
19978ca6 | 7721 | sched_init_granularity(); |
1da177e4 LT |
7722 | } |
7723 | #endif /* CONFIG_SMP */ | |
7724 | ||
cd1bb94b AB |
7725 | const_debug unsigned int sysctl_timer_migration = 1; |
7726 | ||
1da177e4 LT |
7727 | int in_sched_functions(unsigned long addr) |
7728 | { | |
1da177e4 LT |
7729 | return in_lock_functions(addr) || |
7730 | (addr >= (unsigned long)__sched_text_start | |
7731 | && addr < (unsigned long)__sched_text_end); | |
7732 | } | |
7733 | ||
a9957449 | 7734 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7735 | { |
7736 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7737 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7738 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7739 | cfs_rq->rq = rq; | |
f07333bf | 7740 | /* allow initial update_cfs_load() to truncate */ |
6ea72f12 | 7741 | #ifdef CONFIG_SMP |
f07333bf | 7742 | cfs_rq->load_stamp = 1; |
6ea72f12 | 7743 | #endif |
dd41f596 | 7744 | #endif |
67e9fb2a | 7745 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7746 | } |
7747 | ||
fa85ae24 PZ |
7748 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7749 | { | |
7750 | struct rt_prio_array *array; | |
7751 | int i; | |
7752 | ||
7753 | array = &rt_rq->active; | |
7754 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7755 | INIT_LIST_HEAD(array->queue + i); | |
7756 | __clear_bit(i, array->bitmap); | |
7757 | } | |
7758 | /* delimiter for bitsearch: */ | |
7759 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7760 | ||
052f1dc7 | 7761 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7762 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7763 | #ifdef CONFIG_SMP |
e864c499 | 7764 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7765 | #endif |
48d5e258 | 7766 | #endif |
fa85ae24 PZ |
7767 | #ifdef CONFIG_SMP |
7768 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7769 | rt_rq->overloaded = 0; |
05fa785c | 7770 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7771 | #endif |
7772 | ||
7773 | rt_rq->rt_time = 0; | |
7774 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7775 | rt_rq->rt_runtime = 0; |
0986b11b | 7776 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7777 | |
052f1dc7 | 7778 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7779 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7780 | rt_rq->rq = rq; |
7781 | #endif | |
fa85ae24 PZ |
7782 | } |
7783 | ||
6f505b16 | 7784 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 7785 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 7786 | struct sched_entity *se, int cpu, |
ec7dc8ac | 7787 | struct sched_entity *parent) |
6f505b16 | 7788 | { |
ec7dc8ac | 7789 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7790 | tg->cfs_rq[cpu] = cfs_rq; |
7791 | init_cfs_rq(cfs_rq, rq); | |
7792 | cfs_rq->tg = tg; | |
6f505b16 PZ |
7793 | |
7794 | tg->se[cpu] = se; | |
07e06b01 | 7795 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
7796 | if (!se) |
7797 | return; | |
7798 | ||
ec7dc8ac DG |
7799 | if (!parent) |
7800 | se->cfs_rq = &rq->cfs; | |
7801 | else | |
7802 | se->cfs_rq = parent->my_q; | |
7803 | ||
6f505b16 | 7804 | se->my_q = cfs_rq; |
9437178f | 7805 | update_load_set(&se->load, 0); |
ec7dc8ac | 7806 | se->parent = parent; |
6f505b16 | 7807 | } |
052f1dc7 | 7808 | #endif |
6f505b16 | 7809 | |
052f1dc7 | 7810 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 7811 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 7812 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 7813 | struct sched_rt_entity *parent) |
6f505b16 | 7814 | { |
ec7dc8ac DG |
7815 | struct rq *rq = cpu_rq(cpu); |
7816 | ||
6f505b16 PZ |
7817 | tg->rt_rq[cpu] = rt_rq; |
7818 | init_rt_rq(rt_rq, rq); | |
7819 | rt_rq->tg = tg; | |
ac086bc2 | 7820 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7821 | |
7822 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7823 | if (!rt_se) |
7824 | return; | |
7825 | ||
ec7dc8ac DG |
7826 | if (!parent) |
7827 | rt_se->rt_rq = &rq->rt; | |
7828 | else | |
7829 | rt_se->rt_rq = parent->my_q; | |
7830 | ||
6f505b16 | 7831 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7832 | rt_se->parent = parent; |
6f505b16 PZ |
7833 | INIT_LIST_HEAD(&rt_se->run_list); |
7834 | } | |
7835 | #endif | |
7836 | ||
1da177e4 LT |
7837 | void __init sched_init(void) |
7838 | { | |
dd41f596 | 7839 | int i, j; |
434d53b0 MT |
7840 | unsigned long alloc_size = 0, ptr; |
7841 | ||
7842 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7843 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7844 | #endif | |
7845 | #ifdef CONFIG_RT_GROUP_SCHED | |
7846 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7847 | #endif |
df7c8e84 | 7848 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7849 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7850 | #endif |
434d53b0 | 7851 | if (alloc_size) { |
36b7b6d4 | 7852 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7853 | |
7854 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7855 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7856 | ptr += nr_cpu_ids * sizeof(void **); |
7857 | ||
07e06b01 | 7858 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7859 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7860 | |
6d6bc0ad | 7861 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7862 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7863 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7864 | ptr += nr_cpu_ids * sizeof(void **); |
7865 | ||
07e06b01 | 7866 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7867 | ptr += nr_cpu_ids * sizeof(void **); |
7868 | ||
6d6bc0ad | 7869 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7870 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7871 | for_each_possible_cpu(i) { | |
7872 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7873 | ptr += cpumask_size(); | |
7874 | } | |
7875 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7876 | } |
dd41f596 | 7877 | |
57d885fe GH |
7878 | #ifdef CONFIG_SMP |
7879 | init_defrootdomain(); | |
7880 | #endif | |
7881 | ||
d0b27fa7 PZ |
7882 | init_rt_bandwidth(&def_rt_bandwidth, |
7883 | global_rt_period(), global_rt_runtime()); | |
7884 | ||
7885 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 7886 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7887 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7888 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7889 | |
7c941438 | 7890 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7891 | list_add(&root_task_group.list, &task_groups); |
7892 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 7893 | autogroup_init(&init_task); |
7c941438 | 7894 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7895 | |
0a945022 | 7896 | for_each_possible_cpu(i) { |
70b97a7f | 7897 | struct rq *rq; |
1da177e4 LT |
7898 | |
7899 | rq = cpu_rq(i); | |
05fa785c | 7900 | raw_spin_lock_init(&rq->lock); |
7897986b | 7901 | rq->nr_running = 0; |
dce48a84 TG |
7902 | rq->calc_load_active = 0; |
7903 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7904 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7905 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7906 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 7907 | root_task_group.shares = root_task_group_load; |
6f505b16 | 7908 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7909 | /* |
07e06b01 | 7910 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7911 | * |
7912 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7913 | * gets 100% of the cpu resources in the system. This overall | |
7914 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7915 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7916 | * based on each entity's (task or task-group's) weight |
7917 | * (se->load.weight). | |
7918 | * | |
07e06b01 | 7919 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7920 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7921 | * then A0's share of the cpu resource is: | |
7922 | * | |
0d905bca | 7923 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7924 | * |
07e06b01 YZ |
7925 | * We achieve this by letting root_task_group's tasks sit |
7926 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7927 | */ |
07e06b01 | 7928 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7929 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7930 | ||
7931 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7932 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7933 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 7934 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7935 | #endif |
1da177e4 | 7936 | |
dd41f596 IM |
7937 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7938 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7939 | |
7940 | rq->last_load_update_tick = jiffies; | |
7941 | ||
1da177e4 | 7942 | #ifdef CONFIG_SMP |
41c7ce9a | 7943 | rq->sd = NULL; |
57d885fe | 7944 | rq->rd = NULL; |
1399fa78 | 7945 | rq->cpu_power = SCHED_POWER_SCALE; |
3f029d3c | 7946 | rq->post_schedule = 0; |
1da177e4 | 7947 | rq->active_balance = 0; |
dd41f596 | 7948 | rq->next_balance = jiffies; |
1da177e4 | 7949 | rq->push_cpu = 0; |
0a2966b4 | 7950 | rq->cpu = i; |
1f11eb6a | 7951 | rq->online = 0; |
eae0c9df MG |
7952 | rq->idle_stamp = 0; |
7953 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7954 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
7955 | #ifdef CONFIG_NO_HZ |
7956 | rq->nohz_balance_kick = 0; | |
7957 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
7958 | #endif | |
1da177e4 | 7959 | #endif |
8f4d37ec | 7960 | init_rq_hrtick(rq); |
1da177e4 | 7961 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7962 | } |
7963 | ||
2dd73a4f | 7964 | set_load_weight(&init_task); |
b50f60ce | 7965 | |
e107be36 AK |
7966 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7967 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7968 | #endif | |
7969 | ||
c9819f45 | 7970 | #ifdef CONFIG_SMP |
962cf36c | 7971 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7972 | #endif |
7973 | ||
b50f60ce | 7974 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7975 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7976 | #endif |
7977 | ||
1da177e4 LT |
7978 | /* |
7979 | * The boot idle thread does lazy MMU switching as well: | |
7980 | */ | |
7981 | atomic_inc(&init_mm.mm_count); | |
7982 | enter_lazy_tlb(&init_mm, current); | |
7983 | ||
7984 | /* | |
7985 | * Make us the idle thread. Technically, schedule() should not be | |
7986 | * called from this thread, however somewhere below it might be, | |
7987 | * but because we are the idle thread, we just pick up running again | |
7988 | * when this runqueue becomes "idle". | |
7989 | */ | |
7990 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7991 | |
7992 | calc_load_update = jiffies + LOAD_FREQ; | |
7993 | ||
dd41f596 IM |
7994 | /* |
7995 | * During early bootup we pretend to be a normal task: | |
7996 | */ | |
7997 | current->sched_class = &fair_sched_class; | |
6892b75e | 7998 | |
6a7b3dc3 | 7999 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8000 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8001 | #ifdef CONFIG_SMP |
4cb98839 | 8002 | zalloc_cpumask_var(&sched_domains_tmpmask, GFP_NOWAIT); |
7d1e6a9b | 8003 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8004 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8005 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8006 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8007 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8008 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8009 | #endif |
bdddd296 RR |
8010 | /* May be allocated at isolcpus cmdline parse time */ |
8011 | if (cpu_isolated_map == NULL) | |
8012 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8013 | #endif /* SMP */ |
6a7b3dc3 | 8014 | |
6892b75e | 8015 | scheduler_running = 1; |
1da177e4 LT |
8016 | } |
8017 | ||
8018 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
8019 | static inline int preempt_count_equals(int preempt_offset) |
8020 | { | |
234da7bc | 8021 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 | 8022 | |
4ba8216c | 8023 | return (nested == preempt_offset); |
e4aafea2 FW |
8024 | } |
8025 | ||
d894837f | 8026 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8027 | { |
48f24c4d | 8028 | #ifdef in_atomic |
1da177e4 LT |
8029 | static unsigned long prev_jiffy; /* ratelimiting */ |
8030 | ||
e4aafea2 FW |
8031 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8032 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8033 | return; |
8034 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8035 | return; | |
8036 | prev_jiffy = jiffies; | |
8037 | ||
3df0fc5b PZ |
8038 | printk(KERN_ERR |
8039 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8040 | file, line); | |
8041 | printk(KERN_ERR | |
8042 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8043 | in_atomic(), irqs_disabled(), | |
8044 | current->pid, current->comm); | |
aef745fc IM |
8045 | |
8046 | debug_show_held_locks(current); | |
8047 | if (irqs_disabled()) | |
8048 | print_irqtrace_events(current); | |
8049 | dump_stack(); | |
1da177e4 LT |
8050 | #endif |
8051 | } | |
8052 | EXPORT_SYMBOL(__might_sleep); | |
8053 | #endif | |
8054 | ||
8055 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8056 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8057 | { | |
da7a735e PZ |
8058 | const struct sched_class *prev_class = p->sched_class; |
8059 | int old_prio = p->prio; | |
3a5e4dc1 | 8060 | int on_rq; |
3e51f33f | 8061 | |
fd2f4419 | 8062 | on_rq = p->on_rq; |
3a5e4dc1 AK |
8063 | if (on_rq) |
8064 | deactivate_task(rq, p, 0); | |
8065 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8066 | if (on_rq) { | |
8067 | activate_task(rq, p, 0); | |
8068 | resched_task(rq->curr); | |
8069 | } | |
da7a735e PZ |
8070 | |
8071 | check_class_changed(rq, p, prev_class, old_prio); | |
3a5e4dc1 AK |
8072 | } |
8073 | ||
1da177e4 LT |
8074 | void normalize_rt_tasks(void) |
8075 | { | |
a0f98a1c | 8076 | struct task_struct *g, *p; |
1da177e4 | 8077 | unsigned long flags; |
70b97a7f | 8078 | struct rq *rq; |
1da177e4 | 8079 | |
4cf5d77a | 8080 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8081 | do_each_thread(g, p) { |
178be793 IM |
8082 | /* |
8083 | * Only normalize user tasks: | |
8084 | */ | |
8085 | if (!p->mm) | |
8086 | continue; | |
8087 | ||
6cfb0d5d | 8088 | p->se.exec_start = 0; |
6cfb0d5d | 8089 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8090 | p->se.statistics.wait_start = 0; |
8091 | p->se.statistics.sleep_start = 0; | |
8092 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8093 | #endif |
dd41f596 IM |
8094 | |
8095 | if (!rt_task(p)) { | |
8096 | /* | |
8097 | * Renice negative nice level userspace | |
8098 | * tasks back to 0: | |
8099 | */ | |
8100 | if (TASK_NICE(p) < 0 && p->mm) | |
8101 | set_user_nice(p, 0); | |
1da177e4 | 8102 | continue; |
dd41f596 | 8103 | } |
1da177e4 | 8104 | |
1d615482 | 8105 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8106 | rq = __task_rq_lock(p); |
1da177e4 | 8107 | |
178be793 | 8108 | normalize_task(rq, p); |
3a5e4dc1 | 8109 | |
b29739f9 | 8110 | __task_rq_unlock(rq); |
1d615482 | 8111 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8112 | } while_each_thread(g, p); |
8113 | ||
4cf5d77a | 8114 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8115 | } |
8116 | ||
8117 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8118 | |
67fc4e0c | 8119 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8120 | /* |
67fc4e0c | 8121 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8122 | * |
8123 | * They can only be called when the whole system has been | |
8124 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8125 | * activity can take place. Using them for anything else would | |
8126 | * be a serious bug, and as a result, they aren't even visible | |
8127 | * under any other configuration. | |
8128 | */ | |
8129 | ||
8130 | /** | |
8131 | * curr_task - return the current task for a given cpu. | |
8132 | * @cpu: the processor in question. | |
8133 | * | |
8134 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8135 | */ | |
36c8b586 | 8136 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8137 | { |
8138 | return cpu_curr(cpu); | |
8139 | } | |
8140 | ||
67fc4e0c JW |
8141 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8142 | ||
8143 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8144 | /** |
8145 | * set_curr_task - set the current task for a given cpu. | |
8146 | * @cpu: the processor in question. | |
8147 | * @p: the task pointer to set. | |
8148 | * | |
8149 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8150 | * are serviced on a separate stack. It allows the architecture to switch the |
8151 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8152 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8153 | * and caller must save the original value of the current task (see | |
8154 | * curr_task() above) and restore that value before reenabling interrupts and | |
8155 | * re-starting the system. | |
8156 | * | |
8157 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8158 | */ | |
36c8b586 | 8159 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8160 | { |
8161 | cpu_curr(cpu) = p; | |
8162 | } | |
8163 | ||
8164 | #endif | |
29f59db3 | 8165 | |
bccbe08a PZ |
8166 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8167 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8168 | { |
8169 | int i; | |
8170 | ||
8171 | for_each_possible_cpu(i) { | |
8172 | if (tg->cfs_rq) | |
8173 | kfree(tg->cfs_rq[i]); | |
8174 | if (tg->se) | |
8175 | kfree(tg->se[i]); | |
6f505b16 PZ |
8176 | } |
8177 | ||
8178 | kfree(tg->cfs_rq); | |
8179 | kfree(tg->se); | |
6f505b16 PZ |
8180 | } |
8181 | ||
ec7dc8ac DG |
8182 | static |
8183 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8184 | { |
29f59db3 | 8185 | struct cfs_rq *cfs_rq; |
eab17229 | 8186 | struct sched_entity *se; |
29f59db3 SV |
8187 | int i; |
8188 | ||
434d53b0 | 8189 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8190 | if (!tg->cfs_rq) |
8191 | goto err; | |
434d53b0 | 8192 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8193 | if (!tg->se) |
8194 | goto err; | |
052f1dc7 PZ |
8195 | |
8196 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8197 | |
8198 | for_each_possible_cpu(i) { | |
eab17229 LZ |
8199 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8200 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8201 | if (!cfs_rq) |
8202 | goto err; | |
8203 | ||
eab17229 LZ |
8204 | se = kzalloc_node(sizeof(struct sched_entity), |
8205 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8206 | if (!se) |
dfc12eb2 | 8207 | goto err_free_rq; |
29f59db3 | 8208 | |
3d4b47b4 | 8209 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8210 | } |
8211 | ||
8212 | return 1; | |
8213 | ||
49246274 | 8214 | err_free_rq: |
dfc12eb2 | 8215 | kfree(cfs_rq); |
49246274 | 8216 | err: |
bccbe08a PZ |
8217 | return 0; |
8218 | } | |
8219 | ||
bccbe08a PZ |
8220 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8221 | { | |
3d4b47b4 PZ |
8222 | struct rq *rq = cpu_rq(cpu); |
8223 | unsigned long flags; | |
3d4b47b4 PZ |
8224 | |
8225 | /* | |
8226 | * Only empty task groups can be destroyed; so we can speculatively | |
8227 | * check on_list without danger of it being re-added. | |
8228 | */ | |
8229 | if (!tg->cfs_rq[cpu]->on_list) | |
8230 | return; | |
8231 | ||
8232 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8233 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8234 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8235 | } |
6d6bc0ad | 8236 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8237 | static inline void free_fair_sched_group(struct task_group *tg) |
8238 | { | |
8239 | } | |
8240 | ||
ec7dc8ac DG |
8241 | static inline |
8242 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8243 | { |
8244 | return 1; | |
8245 | } | |
8246 | ||
bccbe08a PZ |
8247 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8248 | { | |
8249 | } | |
6d6bc0ad | 8250 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8251 | |
8252 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8253 | static void free_rt_sched_group(struct task_group *tg) |
8254 | { | |
8255 | int i; | |
8256 | ||
d0b27fa7 PZ |
8257 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8258 | ||
bccbe08a PZ |
8259 | for_each_possible_cpu(i) { |
8260 | if (tg->rt_rq) | |
8261 | kfree(tg->rt_rq[i]); | |
8262 | if (tg->rt_se) | |
8263 | kfree(tg->rt_se[i]); | |
8264 | } | |
8265 | ||
8266 | kfree(tg->rt_rq); | |
8267 | kfree(tg->rt_se); | |
8268 | } | |
8269 | ||
ec7dc8ac DG |
8270 | static |
8271 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8272 | { |
8273 | struct rt_rq *rt_rq; | |
eab17229 | 8274 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8275 | int i; |
8276 | ||
434d53b0 | 8277 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8278 | if (!tg->rt_rq) |
8279 | goto err; | |
434d53b0 | 8280 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8281 | if (!tg->rt_se) |
8282 | goto err; | |
8283 | ||
d0b27fa7 PZ |
8284 | init_rt_bandwidth(&tg->rt_bandwidth, |
8285 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8286 | |
8287 | for_each_possible_cpu(i) { | |
eab17229 LZ |
8288 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8289 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8290 | if (!rt_rq) |
8291 | goto err; | |
29f59db3 | 8292 | |
eab17229 LZ |
8293 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8294 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8295 | if (!rt_se) |
dfc12eb2 | 8296 | goto err_free_rq; |
29f59db3 | 8297 | |
3d4b47b4 | 8298 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8299 | } |
8300 | ||
bccbe08a PZ |
8301 | return 1; |
8302 | ||
49246274 | 8303 | err_free_rq: |
dfc12eb2 | 8304 | kfree(rt_rq); |
49246274 | 8305 | err: |
bccbe08a PZ |
8306 | return 0; |
8307 | } | |
6d6bc0ad | 8308 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8309 | static inline void free_rt_sched_group(struct task_group *tg) |
8310 | { | |
8311 | } | |
8312 | ||
ec7dc8ac DG |
8313 | static inline |
8314 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8315 | { |
8316 | return 1; | |
8317 | } | |
6d6bc0ad | 8318 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8319 | |
7c941438 | 8320 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8321 | static void free_sched_group(struct task_group *tg) |
8322 | { | |
8323 | free_fair_sched_group(tg); | |
8324 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8325 | autogroup_free(tg); |
bccbe08a PZ |
8326 | kfree(tg); |
8327 | } | |
8328 | ||
8329 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8330 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8331 | { |
8332 | struct task_group *tg; | |
8333 | unsigned long flags; | |
bccbe08a PZ |
8334 | |
8335 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8336 | if (!tg) | |
8337 | return ERR_PTR(-ENOMEM); | |
8338 | ||
ec7dc8ac | 8339 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8340 | goto err; |
8341 | ||
ec7dc8ac | 8342 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8343 | goto err; |
8344 | ||
8ed36996 | 8345 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8346 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8347 | |
8348 | WARN_ON(!parent); /* root should already exist */ | |
8349 | ||
8350 | tg->parent = parent; | |
f473aa5e | 8351 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8352 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8353 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8354 | |
9b5b7751 | 8355 | return tg; |
29f59db3 SV |
8356 | |
8357 | err: | |
6f505b16 | 8358 | free_sched_group(tg); |
29f59db3 SV |
8359 | return ERR_PTR(-ENOMEM); |
8360 | } | |
8361 | ||
9b5b7751 | 8362 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8363 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8364 | { |
29f59db3 | 8365 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8366 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8367 | } |
8368 | ||
9b5b7751 | 8369 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8370 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8371 | { |
8ed36996 | 8372 | unsigned long flags; |
9b5b7751 | 8373 | int i; |
29f59db3 | 8374 | |
3d4b47b4 PZ |
8375 | /* end participation in shares distribution */ |
8376 | for_each_possible_cpu(i) | |
bccbe08a | 8377 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8378 | |
8379 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8380 | list_del_rcu(&tg->list); |
f473aa5e | 8381 | list_del_rcu(&tg->siblings); |
8ed36996 | 8382 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8383 | |
9b5b7751 | 8384 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8385 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8386 | } |
8387 | ||
9b5b7751 | 8388 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8389 | * The caller of this function should have put the task in its new group |
8390 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8391 | * reflect its new group. | |
9b5b7751 SV |
8392 | */ |
8393 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8394 | { |
8395 | int on_rq, running; | |
8396 | unsigned long flags; | |
8397 | struct rq *rq; | |
8398 | ||
8399 | rq = task_rq_lock(tsk, &flags); | |
8400 | ||
051a1d1a | 8401 | running = task_current(rq, tsk); |
fd2f4419 | 8402 | on_rq = tsk->on_rq; |
29f59db3 | 8403 | |
0e1f3483 | 8404 | if (on_rq) |
29f59db3 | 8405 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8406 | if (unlikely(running)) |
8407 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8408 | |
810b3817 | 8409 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8410 | if (tsk->sched_class->task_move_group) |
8411 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8412 | else | |
810b3817 | 8413 | #endif |
b2b5ce02 | 8414 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8415 | |
0e1f3483 HS |
8416 | if (unlikely(running)) |
8417 | tsk->sched_class->set_curr_task(rq); | |
8418 | if (on_rq) | |
371fd7e7 | 8419 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8420 | |
0122ec5b | 8421 | task_rq_unlock(rq, tsk, &flags); |
29f59db3 | 8422 | } |
7c941438 | 8423 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8424 | |
052f1dc7 | 8425 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8426 | static DEFINE_MUTEX(shares_mutex); |
8427 | ||
4cf86d77 | 8428 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8429 | { |
8430 | int i; | |
8ed36996 | 8431 | unsigned long flags; |
c61935fd | 8432 | |
ec7dc8ac DG |
8433 | /* |
8434 | * We can't change the weight of the root cgroup. | |
8435 | */ | |
8436 | if (!tg->se[0]) | |
8437 | return -EINVAL; | |
8438 | ||
18d95a28 PZ |
8439 | if (shares < MIN_SHARES) |
8440 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8441 | else if (shares > MAX_SHARES) |
8442 | shares = MAX_SHARES; | |
62fb1851 | 8443 | |
8ed36996 | 8444 | mutex_lock(&shares_mutex); |
9b5b7751 | 8445 | if (tg->shares == shares) |
5cb350ba | 8446 | goto done; |
29f59db3 | 8447 | |
9b5b7751 | 8448 | tg->shares = shares; |
c09595f6 | 8449 | for_each_possible_cpu(i) { |
9437178f PT |
8450 | struct rq *rq = cpu_rq(i); |
8451 | struct sched_entity *se; | |
8452 | ||
8453 | se = tg->se[i]; | |
8454 | /* Propagate contribution to hierarchy */ | |
8455 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8456 | for_each_sched_entity(se) | |
6d5ab293 | 8457 | update_cfs_shares(group_cfs_rq(se)); |
9437178f | 8458 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c09595f6 | 8459 | } |
29f59db3 | 8460 | |
5cb350ba | 8461 | done: |
8ed36996 | 8462 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8463 | return 0; |
29f59db3 SV |
8464 | } |
8465 | ||
5cb350ba DG |
8466 | unsigned long sched_group_shares(struct task_group *tg) |
8467 | { | |
8468 | return tg->shares; | |
8469 | } | |
052f1dc7 | 8470 | #endif |
5cb350ba | 8471 | |
052f1dc7 | 8472 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8473 | /* |
9f0c1e56 | 8474 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8475 | */ |
9f0c1e56 PZ |
8476 | static DEFINE_MUTEX(rt_constraints_mutex); |
8477 | ||
8478 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8479 | { | |
8480 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8481 | return 1ULL << 20; |
9f0c1e56 | 8482 | |
9a7e0b18 | 8483 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8484 | } |
8485 | ||
9a7e0b18 PZ |
8486 | /* Must be called with tasklist_lock held */ |
8487 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8488 | { |
9a7e0b18 | 8489 | struct task_struct *g, *p; |
b40b2e8e | 8490 | |
9a7e0b18 PZ |
8491 | do_each_thread(g, p) { |
8492 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8493 | return 1; | |
8494 | } while_each_thread(g, p); | |
b40b2e8e | 8495 | |
9a7e0b18 PZ |
8496 | return 0; |
8497 | } | |
b40b2e8e | 8498 | |
9a7e0b18 PZ |
8499 | struct rt_schedulable_data { |
8500 | struct task_group *tg; | |
8501 | u64 rt_period; | |
8502 | u64 rt_runtime; | |
8503 | }; | |
b40b2e8e | 8504 | |
9a7e0b18 PZ |
8505 | static int tg_schedulable(struct task_group *tg, void *data) |
8506 | { | |
8507 | struct rt_schedulable_data *d = data; | |
8508 | struct task_group *child; | |
8509 | unsigned long total, sum = 0; | |
8510 | u64 period, runtime; | |
b40b2e8e | 8511 | |
9a7e0b18 PZ |
8512 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8513 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8514 | |
9a7e0b18 PZ |
8515 | if (tg == d->tg) { |
8516 | period = d->rt_period; | |
8517 | runtime = d->rt_runtime; | |
b40b2e8e | 8518 | } |
b40b2e8e | 8519 | |
4653f803 PZ |
8520 | /* |
8521 | * Cannot have more runtime than the period. | |
8522 | */ | |
8523 | if (runtime > period && runtime != RUNTIME_INF) | |
8524 | return -EINVAL; | |
6f505b16 | 8525 | |
4653f803 PZ |
8526 | /* |
8527 | * Ensure we don't starve existing RT tasks. | |
8528 | */ | |
9a7e0b18 PZ |
8529 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8530 | return -EBUSY; | |
6f505b16 | 8531 | |
9a7e0b18 | 8532 | total = to_ratio(period, runtime); |
6f505b16 | 8533 | |
4653f803 PZ |
8534 | /* |
8535 | * Nobody can have more than the global setting allows. | |
8536 | */ | |
8537 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8538 | return -EINVAL; | |
6f505b16 | 8539 | |
4653f803 PZ |
8540 | /* |
8541 | * The sum of our children's runtime should not exceed our own. | |
8542 | */ | |
9a7e0b18 PZ |
8543 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8544 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8545 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8546 | |
9a7e0b18 PZ |
8547 | if (child == d->tg) { |
8548 | period = d->rt_period; | |
8549 | runtime = d->rt_runtime; | |
8550 | } | |
6f505b16 | 8551 | |
9a7e0b18 | 8552 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8553 | } |
6f505b16 | 8554 | |
9a7e0b18 PZ |
8555 | if (sum > total) |
8556 | return -EINVAL; | |
8557 | ||
8558 | return 0; | |
6f505b16 PZ |
8559 | } |
8560 | ||
9a7e0b18 | 8561 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8562 | { |
9a7e0b18 PZ |
8563 | struct rt_schedulable_data data = { |
8564 | .tg = tg, | |
8565 | .rt_period = period, | |
8566 | .rt_runtime = runtime, | |
8567 | }; | |
8568 | ||
8569 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8570 | } |
8571 | ||
d0b27fa7 PZ |
8572 | static int tg_set_bandwidth(struct task_group *tg, |
8573 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8574 | { |
ac086bc2 | 8575 | int i, err = 0; |
9f0c1e56 | 8576 | |
9f0c1e56 | 8577 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8578 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8579 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8580 | if (err) | |
9f0c1e56 | 8581 | goto unlock; |
ac086bc2 | 8582 | |
0986b11b | 8583 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8584 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8585 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8586 | |
8587 | for_each_possible_cpu(i) { | |
8588 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8589 | ||
0986b11b | 8590 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8591 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8592 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8593 | } |
0986b11b | 8594 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8595 | unlock: |
521f1a24 | 8596 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8597 | mutex_unlock(&rt_constraints_mutex); |
8598 | ||
8599 | return err; | |
6f505b16 PZ |
8600 | } |
8601 | ||
d0b27fa7 PZ |
8602 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8603 | { | |
8604 | u64 rt_runtime, rt_period; | |
8605 | ||
8606 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8607 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8608 | if (rt_runtime_us < 0) | |
8609 | rt_runtime = RUNTIME_INF; | |
8610 | ||
8611 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8612 | } | |
8613 | ||
9f0c1e56 PZ |
8614 | long sched_group_rt_runtime(struct task_group *tg) |
8615 | { | |
8616 | u64 rt_runtime_us; | |
8617 | ||
d0b27fa7 | 8618 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8619 | return -1; |
8620 | ||
d0b27fa7 | 8621 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8622 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8623 | return rt_runtime_us; | |
8624 | } | |
d0b27fa7 PZ |
8625 | |
8626 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8627 | { | |
8628 | u64 rt_runtime, rt_period; | |
8629 | ||
8630 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8631 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8632 | ||
619b0488 R |
8633 | if (rt_period == 0) |
8634 | return -EINVAL; | |
8635 | ||
d0b27fa7 PZ |
8636 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8637 | } | |
8638 | ||
8639 | long sched_group_rt_period(struct task_group *tg) | |
8640 | { | |
8641 | u64 rt_period_us; | |
8642 | ||
8643 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8644 | do_div(rt_period_us, NSEC_PER_USEC); | |
8645 | return rt_period_us; | |
8646 | } | |
8647 | ||
8648 | static int sched_rt_global_constraints(void) | |
8649 | { | |
4653f803 | 8650 | u64 runtime, period; |
d0b27fa7 PZ |
8651 | int ret = 0; |
8652 | ||
ec5d4989 HS |
8653 | if (sysctl_sched_rt_period <= 0) |
8654 | return -EINVAL; | |
8655 | ||
4653f803 PZ |
8656 | runtime = global_rt_runtime(); |
8657 | period = global_rt_period(); | |
8658 | ||
8659 | /* | |
8660 | * Sanity check on the sysctl variables. | |
8661 | */ | |
8662 | if (runtime > period && runtime != RUNTIME_INF) | |
8663 | return -EINVAL; | |
10b612f4 | 8664 | |
d0b27fa7 | 8665 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8666 | read_lock(&tasklist_lock); |
4653f803 | 8667 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8668 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8669 | mutex_unlock(&rt_constraints_mutex); |
8670 | ||
8671 | return ret; | |
8672 | } | |
54e99124 DG |
8673 | |
8674 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8675 | { | |
8676 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8677 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8678 | return 0; | |
8679 | ||
8680 | return 1; | |
8681 | } | |
8682 | ||
6d6bc0ad | 8683 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8684 | static int sched_rt_global_constraints(void) |
8685 | { | |
ac086bc2 PZ |
8686 | unsigned long flags; |
8687 | int i; | |
8688 | ||
ec5d4989 HS |
8689 | if (sysctl_sched_rt_period <= 0) |
8690 | return -EINVAL; | |
8691 | ||
60aa605d PZ |
8692 | /* |
8693 | * There's always some RT tasks in the root group | |
8694 | * -- migration, kstopmachine etc.. | |
8695 | */ | |
8696 | if (sysctl_sched_rt_runtime == 0) | |
8697 | return -EBUSY; | |
8698 | ||
0986b11b | 8699 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8700 | for_each_possible_cpu(i) { |
8701 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8702 | ||
0986b11b | 8703 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8704 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8705 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8706 | } |
0986b11b | 8707 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8708 | |
d0b27fa7 PZ |
8709 | return 0; |
8710 | } | |
6d6bc0ad | 8711 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8712 | |
8713 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8714 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8715 | loff_t *ppos) |
8716 | { | |
8717 | int ret; | |
8718 | int old_period, old_runtime; | |
8719 | static DEFINE_MUTEX(mutex); | |
8720 | ||
8721 | mutex_lock(&mutex); | |
8722 | old_period = sysctl_sched_rt_period; | |
8723 | old_runtime = sysctl_sched_rt_runtime; | |
8724 | ||
8d65af78 | 8725 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8726 | |
8727 | if (!ret && write) { | |
8728 | ret = sched_rt_global_constraints(); | |
8729 | if (ret) { | |
8730 | sysctl_sched_rt_period = old_period; | |
8731 | sysctl_sched_rt_runtime = old_runtime; | |
8732 | } else { | |
8733 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8734 | def_rt_bandwidth.rt_period = | |
8735 | ns_to_ktime(global_rt_period()); | |
8736 | } | |
8737 | } | |
8738 | mutex_unlock(&mutex); | |
8739 | ||
8740 | return ret; | |
8741 | } | |
68318b8e | 8742 | |
052f1dc7 | 8743 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8744 | |
8745 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8746 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8747 | { |
2b01dfe3 PM |
8748 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8749 | struct task_group, css); | |
68318b8e SV |
8750 | } |
8751 | ||
8752 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8753 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8754 | { |
ec7dc8ac | 8755 | struct task_group *tg, *parent; |
68318b8e | 8756 | |
2b01dfe3 | 8757 | if (!cgrp->parent) { |
68318b8e | 8758 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8759 | return &root_task_group.css; |
68318b8e SV |
8760 | } |
8761 | ||
ec7dc8ac DG |
8762 | parent = cgroup_tg(cgrp->parent); |
8763 | tg = sched_create_group(parent); | |
68318b8e SV |
8764 | if (IS_ERR(tg)) |
8765 | return ERR_PTR(-ENOMEM); | |
8766 | ||
68318b8e SV |
8767 | return &tg->css; |
8768 | } | |
8769 | ||
41a2d6cf IM |
8770 | static void |
8771 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8772 | { |
2b01dfe3 | 8773 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8774 | |
8775 | sched_destroy_group(tg); | |
8776 | } | |
8777 | ||
41a2d6cf | 8778 | static int |
be367d09 | 8779 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8780 | { |
b68aa230 | 8781 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8782 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8783 | return -EINVAL; |
8784 | #else | |
68318b8e SV |
8785 | /* We don't support RT-tasks being in separate groups */ |
8786 | if (tsk->sched_class != &fair_sched_class) | |
8787 | return -EINVAL; | |
b68aa230 | 8788 | #endif |
be367d09 BB |
8789 | return 0; |
8790 | } | |
68318b8e | 8791 | |
68318b8e | 8792 | static void |
f780bdb7 | 8793 | cpu_cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e SV |
8794 | { |
8795 | sched_move_task(tsk); | |
8796 | } | |
8797 | ||
068c5cc5 | 8798 | static void |
d41d5a01 PZ |
8799 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct cgroup *cgrp, |
8800 | struct cgroup *old_cgrp, struct task_struct *task) | |
068c5cc5 PZ |
8801 | { |
8802 | /* | |
8803 | * cgroup_exit() is called in the copy_process() failure path. | |
8804 | * Ignore this case since the task hasn't ran yet, this avoids | |
8805 | * trying to poke a half freed task state from generic code. | |
8806 | */ | |
8807 | if (!(task->flags & PF_EXITING)) | |
8808 | return; | |
8809 | ||
8810 | sched_move_task(task); | |
8811 | } | |
8812 | ||
052f1dc7 | 8813 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8814 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8815 | u64 shareval) |
68318b8e | 8816 | { |
c8b28116 | 8817 | return sched_group_set_shares(cgroup_tg(cgrp), scale_load(shareval)); |
68318b8e SV |
8818 | } |
8819 | ||
f4c753b7 | 8820 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8821 | { |
2b01dfe3 | 8822 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e | 8823 | |
c8b28116 | 8824 | return (u64) scale_load_down(tg->shares); |
68318b8e | 8825 | } |
6d6bc0ad | 8826 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8827 | |
052f1dc7 | 8828 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8829 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8830 | s64 val) |
6f505b16 | 8831 | { |
06ecb27c | 8832 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8833 | } |
8834 | ||
06ecb27c | 8835 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8836 | { |
06ecb27c | 8837 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8838 | } |
d0b27fa7 PZ |
8839 | |
8840 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8841 | u64 rt_period_us) | |
8842 | { | |
8843 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8844 | } | |
8845 | ||
8846 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8847 | { | |
8848 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8849 | } | |
6d6bc0ad | 8850 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8851 | |
fe5c7cc2 | 8852 | static struct cftype cpu_files[] = { |
052f1dc7 | 8853 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8854 | { |
8855 | .name = "shares", | |
f4c753b7 PM |
8856 | .read_u64 = cpu_shares_read_u64, |
8857 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8858 | }, |
052f1dc7 PZ |
8859 | #endif |
8860 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8861 | { |
9f0c1e56 | 8862 | .name = "rt_runtime_us", |
06ecb27c PM |
8863 | .read_s64 = cpu_rt_runtime_read, |
8864 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8865 | }, |
d0b27fa7 PZ |
8866 | { |
8867 | .name = "rt_period_us", | |
f4c753b7 PM |
8868 | .read_u64 = cpu_rt_period_read_uint, |
8869 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8870 | }, |
052f1dc7 | 8871 | #endif |
68318b8e SV |
8872 | }; |
8873 | ||
8874 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8875 | { | |
fe5c7cc2 | 8876 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8877 | } |
8878 | ||
8879 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8880 | .name = "cpu", |
8881 | .create = cpu_cgroup_create, | |
8882 | .destroy = cpu_cgroup_destroy, | |
f780bdb7 BB |
8883 | .can_attach_task = cpu_cgroup_can_attach_task, |
8884 | .attach_task = cpu_cgroup_attach_task, | |
068c5cc5 | 8885 | .exit = cpu_cgroup_exit, |
38605cae IM |
8886 | .populate = cpu_cgroup_populate, |
8887 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8888 | .early_init = 1, |
8889 | }; | |
8890 | ||
052f1dc7 | 8891 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8892 | |
8893 | #ifdef CONFIG_CGROUP_CPUACCT | |
8894 | ||
8895 | /* | |
8896 | * CPU accounting code for task groups. | |
8897 | * | |
8898 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8899 | * (balbir@in.ibm.com). | |
8900 | */ | |
8901 | ||
934352f2 | 8902 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8903 | struct cpuacct { |
8904 | struct cgroup_subsys_state css; | |
8905 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8906 | u64 __percpu *cpuusage; |
ef12fefa | 8907 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8908 | struct cpuacct *parent; |
d842de87 SV |
8909 | }; |
8910 | ||
8911 | struct cgroup_subsys cpuacct_subsys; | |
8912 | ||
8913 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8914 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8915 | { |
32cd756a | 8916 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8917 | struct cpuacct, css); |
8918 | } | |
8919 | ||
8920 | /* return cpu accounting group to which this task belongs */ | |
8921 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8922 | { | |
8923 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8924 | struct cpuacct, css); | |
8925 | } | |
8926 | ||
8927 | /* create a new cpu accounting group */ | |
8928 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8929 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8930 | { |
8931 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8932 | int i; |
d842de87 SV |
8933 | |
8934 | if (!ca) | |
ef12fefa | 8935 | goto out; |
d842de87 SV |
8936 | |
8937 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8938 | if (!ca->cpuusage) |
8939 | goto out_free_ca; | |
8940 | ||
8941 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8942 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8943 | goto out_free_counters; | |
d842de87 | 8944 | |
934352f2 BR |
8945 | if (cgrp->parent) |
8946 | ca->parent = cgroup_ca(cgrp->parent); | |
8947 | ||
d842de87 | 8948 | return &ca->css; |
ef12fefa BR |
8949 | |
8950 | out_free_counters: | |
8951 | while (--i >= 0) | |
8952 | percpu_counter_destroy(&ca->cpustat[i]); | |
8953 | free_percpu(ca->cpuusage); | |
8954 | out_free_ca: | |
8955 | kfree(ca); | |
8956 | out: | |
8957 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8958 | } |
8959 | ||
8960 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8961 | static void |
32cd756a | 8962 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8963 | { |
32cd756a | 8964 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8965 | int i; |
d842de87 | 8966 | |
ef12fefa BR |
8967 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8968 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8969 | free_percpu(ca->cpuusage); |
8970 | kfree(ca); | |
8971 | } | |
8972 | ||
720f5498 KC |
8973 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8974 | { | |
b36128c8 | 8975 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8976 | u64 data; |
8977 | ||
8978 | #ifndef CONFIG_64BIT | |
8979 | /* | |
8980 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8981 | */ | |
05fa785c | 8982 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8983 | data = *cpuusage; |
05fa785c | 8984 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8985 | #else |
8986 | data = *cpuusage; | |
8987 | #endif | |
8988 | ||
8989 | return data; | |
8990 | } | |
8991 | ||
8992 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8993 | { | |
b36128c8 | 8994 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8995 | |
8996 | #ifndef CONFIG_64BIT | |
8997 | /* | |
8998 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8999 | */ | |
05fa785c | 9000 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9001 | *cpuusage = val; |
05fa785c | 9002 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9003 | #else |
9004 | *cpuusage = val; | |
9005 | #endif | |
9006 | } | |
9007 | ||
d842de87 | 9008 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9009 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9010 | { |
32cd756a | 9011 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9012 | u64 totalcpuusage = 0; |
9013 | int i; | |
9014 | ||
720f5498 KC |
9015 | for_each_present_cpu(i) |
9016 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9017 | |
9018 | return totalcpuusage; | |
9019 | } | |
9020 | ||
0297b803 DG |
9021 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9022 | u64 reset) | |
9023 | { | |
9024 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9025 | int err = 0; | |
9026 | int i; | |
9027 | ||
9028 | if (reset) { | |
9029 | err = -EINVAL; | |
9030 | goto out; | |
9031 | } | |
9032 | ||
720f5498 KC |
9033 | for_each_present_cpu(i) |
9034 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9035 | |
0297b803 DG |
9036 | out: |
9037 | return err; | |
9038 | } | |
9039 | ||
e9515c3c KC |
9040 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9041 | struct seq_file *m) | |
9042 | { | |
9043 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9044 | u64 percpu; | |
9045 | int i; | |
9046 | ||
9047 | for_each_present_cpu(i) { | |
9048 | percpu = cpuacct_cpuusage_read(ca, i); | |
9049 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9050 | } | |
9051 | seq_printf(m, "\n"); | |
9052 | return 0; | |
9053 | } | |
9054 | ||
ef12fefa BR |
9055 | static const char *cpuacct_stat_desc[] = { |
9056 | [CPUACCT_STAT_USER] = "user", | |
9057 | [CPUACCT_STAT_SYSTEM] = "system", | |
9058 | }; | |
9059 | ||
9060 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9061 | struct cgroup_map_cb *cb) | |
9062 | { | |
9063 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9064 | int i; | |
9065 | ||
9066 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9067 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9068 | val = cputime64_to_clock_t(val); | |
9069 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9070 | } | |
9071 | return 0; | |
9072 | } | |
9073 | ||
d842de87 SV |
9074 | static struct cftype files[] = { |
9075 | { | |
9076 | .name = "usage", | |
f4c753b7 PM |
9077 | .read_u64 = cpuusage_read, |
9078 | .write_u64 = cpuusage_write, | |
d842de87 | 9079 | }, |
e9515c3c KC |
9080 | { |
9081 | .name = "usage_percpu", | |
9082 | .read_seq_string = cpuacct_percpu_seq_read, | |
9083 | }, | |
ef12fefa BR |
9084 | { |
9085 | .name = "stat", | |
9086 | .read_map = cpuacct_stats_show, | |
9087 | }, | |
d842de87 SV |
9088 | }; |
9089 | ||
32cd756a | 9090 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9091 | { |
32cd756a | 9092 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9093 | } |
9094 | ||
9095 | /* | |
9096 | * charge this task's execution time to its accounting group. | |
9097 | * | |
9098 | * called with rq->lock held. | |
9099 | */ | |
9100 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9101 | { | |
9102 | struct cpuacct *ca; | |
934352f2 | 9103 | int cpu; |
d842de87 | 9104 | |
c40c6f85 | 9105 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9106 | return; |
9107 | ||
934352f2 | 9108 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9109 | |
9110 | rcu_read_lock(); | |
9111 | ||
d842de87 | 9112 | ca = task_ca(tsk); |
d842de87 | 9113 | |
934352f2 | 9114 | for (; ca; ca = ca->parent) { |
b36128c8 | 9115 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9116 | *cpuusage += cputime; |
9117 | } | |
a18b83b7 BR |
9118 | |
9119 | rcu_read_unlock(); | |
d842de87 SV |
9120 | } |
9121 | ||
fa535a77 AB |
9122 | /* |
9123 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9124 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9125 | * percpu_counter_add with values large enough to always overflow the | |
9126 | * per cpu batch limit causing bad SMP scalability. | |
9127 | * | |
9128 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9129 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9130 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9131 | */ | |
9132 | #ifdef CONFIG_SMP | |
9133 | #define CPUACCT_BATCH \ | |
9134 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9135 | #else | |
9136 | #define CPUACCT_BATCH 0 | |
9137 | #endif | |
9138 | ||
ef12fefa BR |
9139 | /* |
9140 | * Charge the system/user time to the task's accounting group. | |
9141 | */ | |
9142 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9143 | enum cpuacct_stat_index idx, cputime_t val) | |
9144 | { | |
9145 | struct cpuacct *ca; | |
fa535a77 | 9146 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9147 | |
9148 | if (unlikely(!cpuacct_subsys.active)) | |
9149 | return; | |
9150 | ||
9151 | rcu_read_lock(); | |
9152 | ca = task_ca(tsk); | |
9153 | ||
9154 | do { | |
fa535a77 | 9155 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9156 | ca = ca->parent; |
9157 | } while (ca); | |
9158 | rcu_read_unlock(); | |
9159 | } | |
9160 | ||
d842de87 SV |
9161 | struct cgroup_subsys cpuacct_subsys = { |
9162 | .name = "cpuacct", | |
9163 | .create = cpuacct_create, | |
9164 | .destroy = cpuacct_destroy, | |
9165 | .populate = cpuacct_populate, | |
9166 | .subsys_id = cpuacct_subsys_id, | |
9167 | }; | |
9168 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf | 9169 |