Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
58 | #include <linux/kthread.h> | |
b5aadf7f | 59 | #include <linux/proc_fs.h> |
1da177e4 | 60 | #include <linux/seq_file.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
5a0e3ad6 | 74 | #include <linux/slab.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 GH |
79 | #include "sched_cpupri.h" |
80 | ||
a8d154b0 | 81 | #define CREATE_TRACE_POINTS |
ad8d75ff | 82 | #include <trace/events/sched.h> |
a8d154b0 | 83 | |
1da177e4 LT |
84 | /* |
85 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
86 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
87 | * and back. | |
88 | */ | |
89 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
90 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
91 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
92 | ||
93 | /* | |
94 | * 'User priority' is the nice value converted to something we | |
95 | * can work with better when scaling various scheduler parameters, | |
96 | * it's a [ 0 ... 39 ] range. | |
97 | */ | |
98 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
99 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
100 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
101 | ||
102 | /* | |
d7876a08 | 103 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 104 | */ |
d6322faf | 105 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 106 | |
6aa645ea IM |
107 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
108 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
109 | ||
1da177e4 LT |
110 | /* |
111 | * These are the 'tuning knobs' of the scheduler: | |
112 | * | |
a4ec24b4 | 113 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
114 | * Timeslices get refilled after they expire. |
115 | */ | |
1da177e4 | 116 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 117 | |
d0b27fa7 PZ |
118 | /* |
119 | * single value that denotes runtime == period, ie unlimited time. | |
120 | */ | |
121 | #define RUNTIME_INF ((u64)~0ULL) | |
122 | ||
e05606d3 IM |
123 | static inline int rt_policy(int policy) |
124 | { | |
3f33a7ce | 125 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
126 | return 1; |
127 | return 0; | |
128 | } | |
129 | ||
130 | static inline int task_has_rt_policy(struct task_struct *p) | |
131 | { | |
132 | return rt_policy(p->policy); | |
133 | } | |
134 | ||
1da177e4 | 135 | /* |
6aa645ea | 136 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 137 | */ |
6aa645ea IM |
138 | struct rt_prio_array { |
139 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
140 | struct list_head queue[MAX_RT_PRIO]; | |
141 | }; | |
142 | ||
d0b27fa7 | 143 | struct rt_bandwidth { |
ea736ed5 | 144 | /* nests inside the rq lock: */ |
0986b11b | 145 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
146 | ktime_t rt_period; |
147 | u64 rt_runtime; | |
148 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
149 | }; |
150 | ||
151 | static struct rt_bandwidth def_rt_bandwidth; | |
152 | ||
153 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
154 | ||
155 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
156 | { | |
157 | struct rt_bandwidth *rt_b = | |
158 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
159 | ktime_t now; | |
160 | int overrun; | |
161 | int idle = 0; | |
162 | ||
163 | for (;;) { | |
164 | now = hrtimer_cb_get_time(timer); | |
165 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
166 | ||
167 | if (!overrun) | |
168 | break; | |
169 | ||
170 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
171 | } | |
172 | ||
173 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
174 | } | |
175 | ||
176 | static | |
177 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
178 | { | |
179 | rt_b->rt_period = ns_to_ktime(period); | |
180 | rt_b->rt_runtime = runtime; | |
181 | ||
0986b11b | 182 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 183 | |
d0b27fa7 PZ |
184 | hrtimer_init(&rt_b->rt_period_timer, |
185 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
186 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
187 | } |
188 | ||
c8bfff6d KH |
189 | static inline int rt_bandwidth_enabled(void) |
190 | { | |
191 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
192 | } |
193 | ||
194 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
195 | { | |
196 | ktime_t now; | |
197 | ||
cac64d00 | 198 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
199 | return; |
200 | ||
201 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
202 | return; | |
203 | ||
0986b11b | 204 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 205 | for (;;) { |
7f1e2ca9 PZ |
206 | unsigned long delta; |
207 | ktime_t soft, hard; | |
208 | ||
d0b27fa7 PZ |
209 | if (hrtimer_active(&rt_b->rt_period_timer)) |
210 | break; | |
211 | ||
212 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
213 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
214 | |
215 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
216 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
217 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
218 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 219 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 220 | } |
0986b11b | 221 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
222 | } |
223 | ||
224 | #ifdef CONFIG_RT_GROUP_SCHED | |
225 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
226 | { | |
227 | hrtimer_cancel(&rt_b->rt_period_timer); | |
228 | } | |
229 | #endif | |
230 | ||
712555ee HC |
231 | /* |
232 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
233 | * detach_destroy_domains and partition_sched_domains. | |
234 | */ | |
235 | static DEFINE_MUTEX(sched_domains_mutex); | |
236 | ||
7c941438 | 237 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 238 | |
68318b8e SV |
239 | #include <linux/cgroup.h> |
240 | ||
29f59db3 SV |
241 | struct cfs_rq; |
242 | ||
6f505b16 PZ |
243 | static LIST_HEAD(task_groups); |
244 | ||
29f59db3 | 245 | /* task group related information */ |
4cf86d77 | 246 | struct task_group { |
68318b8e | 247 | struct cgroup_subsys_state css; |
6c415b92 | 248 | |
052f1dc7 | 249 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
250 | /* schedulable entities of this group on each cpu */ |
251 | struct sched_entity **se; | |
252 | /* runqueue "owned" by this group on each cpu */ | |
253 | struct cfs_rq **cfs_rq; | |
254 | unsigned long shares; | |
052f1dc7 PZ |
255 | #endif |
256 | ||
257 | #ifdef CONFIG_RT_GROUP_SCHED | |
258 | struct sched_rt_entity **rt_se; | |
259 | struct rt_rq **rt_rq; | |
260 | ||
d0b27fa7 | 261 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 262 | #endif |
6b2d7700 | 263 | |
ae8393e5 | 264 | struct rcu_head rcu; |
6f505b16 | 265 | struct list_head list; |
f473aa5e PZ |
266 | |
267 | struct task_group *parent; | |
268 | struct list_head siblings; | |
269 | struct list_head children; | |
29f59db3 SV |
270 | }; |
271 | ||
eff766a6 | 272 | #define root_task_group init_task_group |
6f505b16 | 273 | |
8ed36996 | 274 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
275 | * a task group's cpu shares. |
276 | */ | |
8ed36996 | 277 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 278 | |
e9036b36 CG |
279 | #ifdef CONFIG_FAIR_GROUP_SCHED |
280 | ||
57310a98 PZ |
281 | #ifdef CONFIG_SMP |
282 | static int root_task_group_empty(void) | |
283 | { | |
284 | return list_empty(&root_task_group.children); | |
285 | } | |
286 | #endif | |
287 | ||
052f1dc7 | 288 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 289 | |
cb4ad1ff | 290 | /* |
2e084786 LJ |
291 | * A weight of 0 or 1 can cause arithmetics problems. |
292 | * A weight of a cfs_rq is the sum of weights of which entities | |
293 | * are queued on this cfs_rq, so a weight of a entity should not be | |
294 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
295 | * (The default weight is 1024 - so there's no practical |
296 | * limitation from this.) | |
297 | */ | |
18d95a28 | 298 | #define MIN_SHARES 2 |
2e084786 | 299 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 300 | |
052f1dc7 PZ |
301 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
302 | #endif | |
303 | ||
29f59db3 | 304 | /* Default task group. |
3a252015 | 305 | * Every task in system belong to this group at bootup. |
29f59db3 | 306 | */ |
434d53b0 | 307 | struct task_group init_task_group; |
29f59db3 SV |
308 | |
309 | /* return group to which a task belongs */ | |
4cf86d77 | 310 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 311 | { |
4cf86d77 | 312 | struct task_group *tg; |
9b5b7751 | 313 | |
7c941438 | 314 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
315 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
316 | struct task_group, css); | |
24e377a8 | 317 | #else |
41a2d6cf | 318 | tg = &init_task_group; |
24e377a8 | 319 | #endif |
9b5b7751 | 320 | return tg; |
29f59db3 SV |
321 | } |
322 | ||
323 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 324 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 325 | { |
052f1dc7 | 326 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
327 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
328 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 329 | #endif |
6f505b16 | 330 | |
052f1dc7 | 331 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
332 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
333 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 334 | #endif |
29f59db3 SV |
335 | } |
336 | ||
337 | #else | |
338 | ||
6f505b16 | 339 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
340 | static inline struct task_group *task_group(struct task_struct *p) |
341 | { | |
342 | return NULL; | |
343 | } | |
29f59db3 | 344 | |
7c941438 | 345 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 346 | |
6aa645ea IM |
347 | /* CFS-related fields in a runqueue */ |
348 | struct cfs_rq { | |
349 | struct load_weight load; | |
350 | unsigned long nr_running; | |
351 | ||
6aa645ea | 352 | u64 exec_clock; |
e9acbff6 | 353 | u64 min_vruntime; |
6aa645ea IM |
354 | |
355 | struct rb_root tasks_timeline; | |
356 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
357 | |
358 | struct list_head tasks; | |
359 | struct list_head *balance_iterator; | |
360 | ||
361 | /* | |
362 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
363 | * It is set to NULL otherwise (i.e when none are currently running). |
364 | */ | |
4793241b | 365 | struct sched_entity *curr, *next, *last; |
ddc97297 | 366 | |
5ac5c4d6 | 367 | unsigned int nr_spread_over; |
ddc97297 | 368 | |
62160e3f | 369 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
370 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
371 | ||
41a2d6cf IM |
372 | /* |
373 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
374 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
375 | * (like users, containers etc.) | |
376 | * | |
377 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
378 | * list is used during load balance. | |
379 | */ | |
41a2d6cf IM |
380 | struct list_head leaf_cfs_rq_list; |
381 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
382 | |
383 | #ifdef CONFIG_SMP | |
c09595f6 | 384 | /* |
c8cba857 | 385 | * the part of load.weight contributed by tasks |
c09595f6 | 386 | */ |
c8cba857 | 387 | unsigned long task_weight; |
c09595f6 | 388 | |
c8cba857 PZ |
389 | /* |
390 | * h_load = weight * f(tg) | |
391 | * | |
392 | * Where f(tg) is the recursive weight fraction assigned to | |
393 | * this group. | |
394 | */ | |
395 | unsigned long h_load; | |
c09595f6 | 396 | |
c8cba857 PZ |
397 | /* |
398 | * this cpu's part of tg->shares | |
399 | */ | |
400 | unsigned long shares; | |
f1d239f7 PZ |
401 | |
402 | /* | |
403 | * load.weight at the time we set shares | |
404 | */ | |
405 | unsigned long rq_weight; | |
c09595f6 | 406 | #endif |
6aa645ea IM |
407 | #endif |
408 | }; | |
1da177e4 | 409 | |
6aa645ea IM |
410 | /* Real-Time classes' related field in a runqueue: */ |
411 | struct rt_rq { | |
412 | struct rt_prio_array active; | |
63489e45 | 413 | unsigned long rt_nr_running; |
052f1dc7 | 414 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
415 | struct { |
416 | int curr; /* highest queued rt task prio */ | |
398a153b | 417 | #ifdef CONFIG_SMP |
e864c499 | 418 | int next; /* next highest */ |
398a153b | 419 | #endif |
e864c499 | 420 | } highest_prio; |
6f505b16 | 421 | #endif |
fa85ae24 | 422 | #ifdef CONFIG_SMP |
73fe6aae | 423 | unsigned long rt_nr_migratory; |
a1ba4d8b | 424 | unsigned long rt_nr_total; |
a22d7fc1 | 425 | int overloaded; |
917b627d | 426 | struct plist_head pushable_tasks; |
fa85ae24 | 427 | #endif |
6f505b16 | 428 | int rt_throttled; |
fa85ae24 | 429 | u64 rt_time; |
ac086bc2 | 430 | u64 rt_runtime; |
ea736ed5 | 431 | /* Nests inside the rq lock: */ |
0986b11b | 432 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 433 | |
052f1dc7 | 434 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
435 | unsigned long rt_nr_boosted; |
436 | ||
6f505b16 PZ |
437 | struct rq *rq; |
438 | struct list_head leaf_rt_rq_list; | |
439 | struct task_group *tg; | |
6f505b16 | 440 | #endif |
6aa645ea IM |
441 | }; |
442 | ||
57d885fe GH |
443 | #ifdef CONFIG_SMP |
444 | ||
445 | /* | |
446 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
447 | * variables. Each exclusive cpuset essentially defines an island domain by |
448 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
449 | * exclusive cpuset is created, we also create and attach a new root-domain |
450 | * object. | |
451 | * | |
57d885fe GH |
452 | */ |
453 | struct root_domain { | |
454 | atomic_t refcount; | |
c6c4927b RR |
455 | cpumask_var_t span; |
456 | cpumask_var_t online; | |
637f5085 | 457 | |
0eab9146 | 458 | /* |
637f5085 GH |
459 | * The "RT overload" flag: it gets set if a CPU has more than |
460 | * one runnable RT task. | |
461 | */ | |
c6c4927b | 462 | cpumask_var_t rto_mask; |
0eab9146 | 463 | atomic_t rto_count; |
6e0534f2 GH |
464 | #ifdef CONFIG_SMP |
465 | struct cpupri cpupri; | |
466 | #endif | |
57d885fe GH |
467 | }; |
468 | ||
dc938520 GH |
469 | /* |
470 | * By default the system creates a single root-domain with all cpus as | |
471 | * members (mimicking the global state we have today). | |
472 | */ | |
57d885fe GH |
473 | static struct root_domain def_root_domain; |
474 | ||
475 | #endif | |
476 | ||
1da177e4 LT |
477 | /* |
478 | * This is the main, per-CPU runqueue data structure. | |
479 | * | |
480 | * Locking rule: those places that want to lock multiple runqueues | |
481 | * (such as the load balancing or the thread migration code), lock | |
482 | * acquire operations must be ordered by ascending &runqueue. | |
483 | */ | |
70b97a7f | 484 | struct rq { |
d8016491 | 485 | /* runqueue lock: */ |
05fa785c | 486 | raw_spinlock_t lock; |
1da177e4 LT |
487 | |
488 | /* | |
489 | * nr_running and cpu_load should be in the same cacheline because | |
490 | * remote CPUs use both these fields when doing load calculation. | |
491 | */ | |
492 | unsigned long nr_running; | |
6aa645ea IM |
493 | #define CPU_LOAD_IDX_MAX 5 |
494 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 495 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 496 | u64 nohz_stamp; |
46cb4b7c SS |
497 | unsigned char in_nohz_recently; |
498 | #endif | |
a64692a3 MG |
499 | unsigned int skip_clock_update; |
500 | ||
d8016491 IM |
501 | /* capture load from *all* tasks on this cpu: */ |
502 | struct load_weight load; | |
6aa645ea IM |
503 | unsigned long nr_load_updates; |
504 | u64 nr_switches; | |
505 | ||
506 | struct cfs_rq cfs; | |
6f505b16 | 507 | struct rt_rq rt; |
6f505b16 | 508 | |
6aa645ea | 509 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
510 | /* list of leaf cfs_rq on this cpu: */ |
511 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
512 | #endif |
513 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 514 | struct list_head leaf_rt_rq_list; |
1da177e4 | 515 | #endif |
1da177e4 LT |
516 | |
517 | /* | |
518 | * This is part of a global counter where only the total sum | |
519 | * over all CPUs matters. A task can increase this counter on | |
520 | * one CPU and if it got migrated afterwards it may decrease | |
521 | * it on another CPU. Always updated under the runqueue lock: | |
522 | */ | |
523 | unsigned long nr_uninterruptible; | |
524 | ||
36c8b586 | 525 | struct task_struct *curr, *idle; |
c9819f45 | 526 | unsigned long next_balance; |
1da177e4 | 527 | struct mm_struct *prev_mm; |
6aa645ea | 528 | |
3e51f33f | 529 | u64 clock; |
6aa645ea | 530 | |
1da177e4 LT |
531 | atomic_t nr_iowait; |
532 | ||
533 | #ifdef CONFIG_SMP | |
0eab9146 | 534 | struct root_domain *rd; |
1da177e4 LT |
535 | struct sched_domain *sd; |
536 | ||
a0a522ce | 537 | unsigned char idle_at_tick; |
1da177e4 | 538 | /* For active balancing */ |
3f029d3c | 539 | int post_schedule; |
1da177e4 LT |
540 | int active_balance; |
541 | int push_cpu; | |
d8016491 IM |
542 | /* cpu of this runqueue: */ |
543 | int cpu; | |
1f11eb6a | 544 | int online; |
1da177e4 | 545 | |
a8a51d5e | 546 | unsigned long avg_load_per_task; |
1da177e4 | 547 | |
36c8b586 | 548 | struct task_struct *migration_thread; |
1da177e4 | 549 | struct list_head migration_queue; |
e9e9250b PZ |
550 | |
551 | u64 rt_avg; | |
552 | u64 age_stamp; | |
1b9508f6 MG |
553 | u64 idle_stamp; |
554 | u64 avg_idle; | |
1da177e4 LT |
555 | #endif |
556 | ||
dce48a84 TG |
557 | /* calc_load related fields */ |
558 | unsigned long calc_load_update; | |
559 | long calc_load_active; | |
560 | ||
8f4d37ec | 561 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
562 | #ifdef CONFIG_SMP |
563 | int hrtick_csd_pending; | |
564 | struct call_single_data hrtick_csd; | |
565 | #endif | |
8f4d37ec PZ |
566 | struct hrtimer hrtick_timer; |
567 | #endif | |
568 | ||
1da177e4 LT |
569 | #ifdef CONFIG_SCHEDSTATS |
570 | /* latency stats */ | |
571 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
572 | unsigned long long rq_cpu_time; |
573 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
574 | |
575 | /* sys_sched_yield() stats */ | |
480b9434 | 576 | unsigned int yld_count; |
1da177e4 LT |
577 | |
578 | /* schedule() stats */ | |
480b9434 KC |
579 | unsigned int sched_switch; |
580 | unsigned int sched_count; | |
581 | unsigned int sched_goidle; | |
1da177e4 LT |
582 | |
583 | /* try_to_wake_up() stats */ | |
480b9434 KC |
584 | unsigned int ttwu_count; |
585 | unsigned int ttwu_local; | |
b8efb561 IM |
586 | |
587 | /* BKL stats */ | |
480b9434 | 588 | unsigned int bkl_count; |
1da177e4 LT |
589 | #endif |
590 | }; | |
591 | ||
f34e3b61 | 592 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 593 | |
7d478721 PZ |
594 | static inline |
595 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 596 | { |
7d478721 | 597 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
a64692a3 MG |
598 | |
599 | /* | |
600 | * A queue event has occurred, and we're going to schedule. In | |
601 | * this case, we can save a useless back to back clock update. | |
602 | */ | |
603 | if (test_tsk_need_resched(p)) | |
604 | rq->skip_clock_update = 1; | |
dd41f596 IM |
605 | } |
606 | ||
0a2966b4 CL |
607 | static inline int cpu_of(struct rq *rq) |
608 | { | |
609 | #ifdef CONFIG_SMP | |
610 | return rq->cpu; | |
611 | #else | |
612 | return 0; | |
613 | #endif | |
614 | } | |
615 | ||
497f0ab3 | 616 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
617 | rcu_dereference_check((p), \ |
618 | rcu_read_lock_sched_held() || \ | |
619 | lockdep_is_held(&sched_domains_mutex)) | |
620 | ||
674311d5 NP |
621 | /* |
622 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 623 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
624 | * |
625 | * The domain tree of any CPU may only be accessed from within | |
626 | * preempt-disabled sections. | |
627 | */ | |
48f24c4d | 628 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 629 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
630 | |
631 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
632 | #define this_rq() (&__get_cpu_var(runqueues)) | |
633 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
634 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 635 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 636 | |
aa9c4c0f | 637 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 638 | { |
a64692a3 MG |
639 | if (!rq->skip_clock_update) |
640 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
3e51f33f PZ |
641 | } |
642 | ||
bf5c91ba IM |
643 | /* |
644 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
645 | */ | |
646 | #ifdef CONFIG_SCHED_DEBUG | |
647 | # define const_debug __read_mostly | |
648 | #else | |
649 | # define const_debug static const | |
650 | #endif | |
651 | ||
017730c1 IM |
652 | /** |
653 | * runqueue_is_locked | |
e17b38bf | 654 | * @cpu: the processor in question. |
017730c1 IM |
655 | * |
656 | * Returns true if the current cpu runqueue is locked. | |
657 | * This interface allows printk to be called with the runqueue lock | |
658 | * held and know whether or not it is OK to wake up the klogd. | |
659 | */ | |
89f19f04 | 660 | int runqueue_is_locked(int cpu) |
017730c1 | 661 | { |
05fa785c | 662 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
663 | } |
664 | ||
bf5c91ba IM |
665 | /* |
666 | * Debugging: various feature bits | |
667 | */ | |
f00b45c1 PZ |
668 | |
669 | #define SCHED_FEAT(name, enabled) \ | |
670 | __SCHED_FEAT_##name , | |
671 | ||
bf5c91ba | 672 | enum { |
f00b45c1 | 673 | #include "sched_features.h" |
bf5c91ba IM |
674 | }; |
675 | ||
f00b45c1 PZ |
676 | #undef SCHED_FEAT |
677 | ||
678 | #define SCHED_FEAT(name, enabled) \ | |
679 | (1UL << __SCHED_FEAT_##name) * enabled | | |
680 | ||
bf5c91ba | 681 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
682 | #include "sched_features.h" |
683 | 0; | |
684 | ||
685 | #undef SCHED_FEAT | |
686 | ||
687 | #ifdef CONFIG_SCHED_DEBUG | |
688 | #define SCHED_FEAT(name, enabled) \ | |
689 | #name , | |
690 | ||
983ed7a6 | 691 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
692 | #include "sched_features.h" |
693 | NULL | |
694 | }; | |
695 | ||
696 | #undef SCHED_FEAT | |
697 | ||
34f3a814 | 698 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 699 | { |
f00b45c1 PZ |
700 | int i; |
701 | ||
702 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
703 | if (!(sysctl_sched_features & (1UL << i))) |
704 | seq_puts(m, "NO_"); | |
705 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 706 | } |
34f3a814 | 707 | seq_puts(m, "\n"); |
f00b45c1 | 708 | |
34f3a814 | 709 | return 0; |
f00b45c1 PZ |
710 | } |
711 | ||
712 | static ssize_t | |
713 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
714 | size_t cnt, loff_t *ppos) | |
715 | { | |
716 | char buf[64]; | |
717 | char *cmp = buf; | |
718 | int neg = 0; | |
719 | int i; | |
720 | ||
721 | if (cnt > 63) | |
722 | cnt = 63; | |
723 | ||
724 | if (copy_from_user(&buf, ubuf, cnt)) | |
725 | return -EFAULT; | |
726 | ||
727 | buf[cnt] = 0; | |
728 | ||
c24b7c52 | 729 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
730 | neg = 1; |
731 | cmp += 3; | |
732 | } | |
733 | ||
734 | for (i = 0; sched_feat_names[i]; i++) { | |
735 | int len = strlen(sched_feat_names[i]); | |
736 | ||
737 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
738 | if (neg) | |
739 | sysctl_sched_features &= ~(1UL << i); | |
740 | else | |
741 | sysctl_sched_features |= (1UL << i); | |
742 | break; | |
743 | } | |
744 | } | |
745 | ||
746 | if (!sched_feat_names[i]) | |
747 | return -EINVAL; | |
748 | ||
42994724 | 749 | *ppos += cnt; |
f00b45c1 PZ |
750 | |
751 | return cnt; | |
752 | } | |
753 | ||
34f3a814 LZ |
754 | static int sched_feat_open(struct inode *inode, struct file *filp) |
755 | { | |
756 | return single_open(filp, sched_feat_show, NULL); | |
757 | } | |
758 | ||
828c0950 | 759 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
760 | .open = sched_feat_open, |
761 | .write = sched_feat_write, | |
762 | .read = seq_read, | |
763 | .llseek = seq_lseek, | |
764 | .release = single_release, | |
f00b45c1 PZ |
765 | }; |
766 | ||
767 | static __init int sched_init_debug(void) | |
768 | { | |
f00b45c1 PZ |
769 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
770 | &sched_feat_fops); | |
771 | ||
772 | return 0; | |
773 | } | |
774 | late_initcall(sched_init_debug); | |
775 | ||
776 | #endif | |
777 | ||
778 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 779 | |
b82d9fdd PZ |
780 | /* |
781 | * Number of tasks to iterate in a single balance run. | |
782 | * Limited because this is done with IRQs disabled. | |
783 | */ | |
784 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
785 | ||
2398f2c6 PZ |
786 | /* |
787 | * ratelimit for updating the group shares. | |
55cd5340 | 788 | * default: 0.25ms |
2398f2c6 | 789 | */ |
55cd5340 | 790 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 791 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 792 | |
ffda12a1 PZ |
793 | /* |
794 | * Inject some fuzzyness into changing the per-cpu group shares | |
795 | * this avoids remote rq-locks at the expense of fairness. | |
796 | * default: 4 | |
797 | */ | |
798 | unsigned int sysctl_sched_shares_thresh = 4; | |
799 | ||
e9e9250b PZ |
800 | /* |
801 | * period over which we average the RT time consumption, measured | |
802 | * in ms. | |
803 | * | |
804 | * default: 1s | |
805 | */ | |
806 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
807 | ||
fa85ae24 | 808 | /* |
9f0c1e56 | 809 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
810 | * default: 1s |
811 | */ | |
9f0c1e56 | 812 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 813 | |
6892b75e IM |
814 | static __read_mostly int scheduler_running; |
815 | ||
9f0c1e56 PZ |
816 | /* |
817 | * part of the period that we allow rt tasks to run in us. | |
818 | * default: 0.95s | |
819 | */ | |
820 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 821 | |
d0b27fa7 PZ |
822 | static inline u64 global_rt_period(void) |
823 | { | |
824 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
825 | } | |
826 | ||
827 | static inline u64 global_rt_runtime(void) | |
828 | { | |
e26873bb | 829 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
830 | return RUNTIME_INF; |
831 | ||
832 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
833 | } | |
fa85ae24 | 834 | |
1da177e4 | 835 | #ifndef prepare_arch_switch |
4866cde0 NP |
836 | # define prepare_arch_switch(next) do { } while (0) |
837 | #endif | |
838 | #ifndef finish_arch_switch | |
839 | # define finish_arch_switch(prev) do { } while (0) | |
840 | #endif | |
841 | ||
051a1d1a DA |
842 | static inline int task_current(struct rq *rq, struct task_struct *p) |
843 | { | |
844 | return rq->curr == p; | |
845 | } | |
846 | ||
4866cde0 | 847 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 848 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 849 | { |
051a1d1a | 850 | return task_current(rq, p); |
4866cde0 NP |
851 | } |
852 | ||
70b97a7f | 853 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
854 | { |
855 | } | |
856 | ||
70b97a7f | 857 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 858 | { |
da04c035 IM |
859 | #ifdef CONFIG_DEBUG_SPINLOCK |
860 | /* this is a valid case when another task releases the spinlock */ | |
861 | rq->lock.owner = current; | |
862 | #endif | |
8a25d5de IM |
863 | /* |
864 | * If we are tracking spinlock dependencies then we have to | |
865 | * fix up the runqueue lock - which gets 'carried over' from | |
866 | * prev into current: | |
867 | */ | |
868 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
869 | ||
05fa785c | 870 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
871 | } |
872 | ||
873 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 874 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
875 | { |
876 | #ifdef CONFIG_SMP | |
877 | return p->oncpu; | |
878 | #else | |
051a1d1a | 879 | return task_current(rq, p); |
4866cde0 NP |
880 | #endif |
881 | } | |
882 | ||
70b97a7f | 883 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
884 | { |
885 | #ifdef CONFIG_SMP | |
886 | /* | |
887 | * We can optimise this out completely for !SMP, because the | |
888 | * SMP rebalancing from interrupt is the only thing that cares | |
889 | * here. | |
890 | */ | |
891 | next->oncpu = 1; | |
892 | #endif | |
893 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 894 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 895 | #else |
05fa785c | 896 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
897 | #endif |
898 | } | |
899 | ||
70b97a7f | 900 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
901 | { |
902 | #ifdef CONFIG_SMP | |
903 | /* | |
904 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
905 | * We must ensure this doesn't happen until the switch is completely | |
906 | * finished. | |
907 | */ | |
908 | smp_wmb(); | |
909 | prev->oncpu = 0; | |
910 | #endif | |
911 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
912 | local_irq_enable(); | |
1da177e4 | 913 | #endif |
4866cde0 NP |
914 | } |
915 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 916 | |
0970d299 | 917 | /* |
65cc8e48 PZ |
918 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
919 | * against ttwu(). | |
0970d299 PZ |
920 | */ |
921 | static inline int task_is_waking(struct task_struct *p) | |
922 | { | |
0017d735 | 923 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
924 | } |
925 | ||
b29739f9 IM |
926 | /* |
927 | * __task_rq_lock - lock the runqueue a given task resides on. | |
928 | * Must be called interrupts disabled. | |
929 | */ | |
70b97a7f | 930 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
931 | __acquires(rq->lock) |
932 | { | |
0970d299 PZ |
933 | struct rq *rq; |
934 | ||
3a5c359a | 935 | for (;;) { |
0970d299 | 936 | rq = task_rq(p); |
05fa785c | 937 | raw_spin_lock(&rq->lock); |
65cc8e48 | 938 | if (likely(rq == task_rq(p))) |
3a5c359a | 939 | return rq; |
05fa785c | 940 | raw_spin_unlock(&rq->lock); |
b29739f9 | 941 | } |
b29739f9 IM |
942 | } |
943 | ||
1da177e4 LT |
944 | /* |
945 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 946 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
947 | * explicitly disabling preemption. |
948 | */ | |
70b97a7f | 949 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
950 | __acquires(rq->lock) |
951 | { | |
70b97a7f | 952 | struct rq *rq; |
1da177e4 | 953 | |
3a5c359a AK |
954 | for (;;) { |
955 | local_irq_save(*flags); | |
956 | rq = task_rq(p); | |
05fa785c | 957 | raw_spin_lock(&rq->lock); |
65cc8e48 | 958 | if (likely(rq == task_rq(p))) |
3a5c359a | 959 | return rq; |
05fa785c | 960 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 961 | } |
1da177e4 LT |
962 | } |
963 | ||
ad474cac ON |
964 | void task_rq_unlock_wait(struct task_struct *p) |
965 | { | |
966 | struct rq *rq = task_rq(p); | |
967 | ||
968 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
05fa785c | 969 | raw_spin_unlock_wait(&rq->lock); |
ad474cac ON |
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 | ||
70b97a7f | 978 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
979 | __releases(rq->lock) |
980 | { | |
05fa785c | 981 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
982 | } |
983 | ||
1da177e4 | 984 | /* |
cc2a73b5 | 985 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 986 | */ |
a9957449 | 987 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
988 | __acquires(rq->lock) |
989 | { | |
70b97a7f | 990 | struct rq *rq; |
1da177e4 LT |
991 | |
992 | local_irq_disable(); | |
993 | rq = this_rq(); | |
05fa785c | 994 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
995 | |
996 | return rq; | |
997 | } | |
998 | ||
8f4d37ec PZ |
999 | #ifdef CONFIG_SCHED_HRTICK |
1000 | /* | |
1001 | * Use HR-timers to deliver accurate preemption points. | |
1002 | * | |
1003 | * Its all a bit involved since we cannot program an hrt while holding the | |
1004 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1005 | * reschedule event. | |
1006 | * | |
1007 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1008 | * rq->lock. | |
1009 | */ | |
8f4d37ec PZ |
1010 | |
1011 | /* | |
1012 | * Use hrtick when: | |
1013 | * - enabled by features | |
1014 | * - hrtimer is actually high res | |
1015 | */ | |
1016 | static inline int hrtick_enabled(struct rq *rq) | |
1017 | { | |
1018 | if (!sched_feat(HRTICK)) | |
1019 | return 0; | |
ba42059f | 1020 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1021 | return 0; |
8f4d37ec PZ |
1022 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1023 | } | |
1024 | ||
8f4d37ec PZ |
1025 | static void hrtick_clear(struct rq *rq) |
1026 | { | |
1027 | if (hrtimer_active(&rq->hrtick_timer)) | |
1028 | hrtimer_cancel(&rq->hrtick_timer); | |
1029 | } | |
1030 | ||
8f4d37ec PZ |
1031 | /* |
1032 | * High-resolution timer tick. | |
1033 | * Runs from hardirq context with interrupts disabled. | |
1034 | */ | |
1035 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1036 | { | |
1037 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1038 | ||
1039 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1040 | ||
05fa785c | 1041 | raw_spin_lock(&rq->lock); |
3e51f33f | 1042 | update_rq_clock(rq); |
8f4d37ec | 1043 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1044 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1045 | |
1046 | return HRTIMER_NORESTART; | |
1047 | } | |
1048 | ||
95e904c7 | 1049 | #ifdef CONFIG_SMP |
31656519 PZ |
1050 | /* |
1051 | * called from hardirq (IPI) context | |
1052 | */ | |
1053 | static void __hrtick_start(void *arg) | |
b328ca18 | 1054 | { |
31656519 | 1055 | struct rq *rq = arg; |
b328ca18 | 1056 | |
05fa785c | 1057 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1058 | hrtimer_restart(&rq->hrtick_timer); |
1059 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1060 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1061 | } |
1062 | ||
31656519 PZ |
1063 | /* |
1064 | * Called to set the hrtick timer state. | |
1065 | * | |
1066 | * called with rq->lock held and irqs disabled | |
1067 | */ | |
1068 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1069 | { |
31656519 PZ |
1070 | struct hrtimer *timer = &rq->hrtick_timer; |
1071 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1072 | |
cc584b21 | 1073 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1074 | |
1075 | if (rq == this_rq()) { | |
1076 | hrtimer_restart(timer); | |
1077 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1078 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1079 | rq->hrtick_csd_pending = 1; |
1080 | } | |
b328ca18 PZ |
1081 | } |
1082 | ||
1083 | static int | |
1084 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1085 | { | |
1086 | int cpu = (int)(long)hcpu; | |
1087 | ||
1088 | switch (action) { | |
1089 | case CPU_UP_CANCELED: | |
1090 | case CPU_UP_CANCELED_FROZEN: | |
1091 | case CPU_DOWN_PREPARE: | |
1092 | case CPU_DOWN_PREPARE_FROZEN: | |
1093 | case CPU_DEAD: | |
1094 | case CPU_DEAD_FROZEN: | |
31656519 | 1095 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1096 | return NOTIFY_OK; |
1097 | } | |
1098 | ||
1099 | return NOTIFY_DONE; | |
1100 | } | |
1101 | ||
fa748203 | 1102 | static __init void init_hrtick(void) |
b328ca18 PZ |
1103 | { |
1104 | hotcpu_notifier(hotplug_hrtick, 0); | |
1105 | } | |
31656519 PZ |
1106 | #else |
1107 | /* | |
1108 | * Called to set the hrtick timer state. | |
1109 | * | |
1110 | * called with rq->lock held and irqs disabled | |
1111 | */ | |
1112 | static void hrtick_start(struct rq *rq, u64 delay) | |
1113 | { | |
7f1e2ca9 | 1114 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1115 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1116 | } |
b328ca18 | 1117 | |
006c75f1 | 1118 | static inline void init_hrtick(void) |
8f4d37ec | 1119 | { |
8f4d37ec | 1120 | } |
31656519 | 1121 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1122 | |
31656519 | 1123 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1124 | { |
31656519 PZ |
1125 | #ifdef CONFIG_SMP |
1126 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1127 | |
31656519 PZ |
1128 | rq->hrtick_csd.flags = 0; |
1129 | rq->hrtick_csd.func = __hrtick_start; | |
1130 | rq->hrtick_csd.info = rq; | |
1131 | #endif | |
8f4d37ec | 1132 | |
31656519 PZ |
1133 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1134 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1135 | } |
006c75f1 | 1136 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1137 | static inline void hrtick_clear(struct rq *rq) |
1138 | { | |
1139 | } | |
1140 | ||
8f4d37ec PZ |
1141 | static inline void init_rq_hrtick(struct rq *rq) |
1142 | { | |
1143 | } | |
1144 | ||
b328ca18 PZ |
1145 | static inline void init_hrtick(void) |
1146 | { | |
1147 | } | |
006c75f1 | 1148 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1149 | |
c24d20db IM |
1150 | /* |
1151 | * resched_task - mark a task 'to be rescheduled now'. | |
1152 | * | |
1153 | * On UP this means the setting of the need_resched flag, on SMP it | |
1154 | * might also involve a cross-CPU call to trigger the scheduler on | |
1155 | * the target CPU. | |
1156 | */ | |
1157 | #ifdef CONFIG_SMP | |
1158 | ||
1159 | #ifndef tsk_is_polling | |
1160 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1161 | #endif | |
1162 | ||
31656519 | 1163 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1164 | { |
1165 | int cpu; | |
1166 | ||
05fa785c | 1167 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1168 | |
5ed0cec0 | 1169 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1170 | return; |
1171 | ||
5ed0cec0 | 1172 | set_tsk_need_resched(p); |
c24d20db IM |
1173 | |
1174 | cpu = task_cpu(p); | |
1175 | if (cpu == smp_processor_id()) | |
1176 | return; | |
1177 | ||
1178 | /* NEED_RESCHED must be visible before we test polling */ | |
1179 | smp_mb(); | |
1180 | if (!tsk_is_polling(p)) | |
1181 | smp_send_reschedule(cpu); | |
1182 | } | |
1183 | ||
1184 | static void resched_cpu(int cpu) | |
1185 | { | |
1186 | struct rq *rq = cpu_rq(cpu); | |
1187 | unsigned long flags; | |
1188 | ||
05fa785c | 1189 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1190 | return; |
1191 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1192 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1193 | } |
06d8308c TG |
1194 | |
1195 | #ifdef CONFIG_NO_HZ | |
1196 | /* | |
1197 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1198 | * idle CPU then this timer might expire before the next timer event | |
1199 | * which is scheduled to wake up that CPU. In case of a completely | |
1200 | * idle system the next event might even be infinite time into the | |
1201 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1202 | * leaves the inner idle loop so the newly added timer is taken into | |
1203 | * account when the CPU goes back to idle and evaluates the timer | |
1204 | * wheel for the next timer event. | |
1205 | */ | |
1206 | void wake_up_idle_cpu(int cpu) | |
1207 | { | |
1208 | struct rq *rq = cpu_rq(cpu); | |
1209 | ||
1210 | if (cpu == smp_processor_id()) | |
1211 | return; | |
1212 | ||
1213 | /* | |
1214 | * This is safe, as this function is called with the timer | |
1215 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1216 | * to idle and has not yet set rq->curr to idle then it will | |
1217 | * be serialized on the timer wheel base lock and take the new | |
1218 | * timer into account automatically. | |
1219 | */ | |
1220 | if (rq->curr != rq->idle) | |
1221 | return; | |
1222 | ||
1223 | /* | |
1224 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1225 | * lockless. The worst case is that the other CPU runs the | |
1226 | * idle task through an additional NOOP schedule() | |
1227 | */ | |
5ed0cec0 | 1228 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1229 | |
1230 | /* NEED_RESCHED must be visible before we test polling */ | |
1231 | smp_mb(); | |
1232 | if (!tsk_is_polling(rq->idle)) | |
1233 | smp_send_reschedule(cpu); | |
1234 | } | |
39c0cbe2 MG |
1235 | |
1236 | int nohz_ratelimit(int cpu) | |
1237 | { | |
1238 | struct rq *rq = cpu_rq(cpu); | |
1239 | u64 diff = rq->clock - rq->nohz_stamp; | |
1240 | ||
1241 | rq->nohz_stamp = rq->clock; | |
1242 | ||
1243 | return diff < (NSEC_PER_SEC / HZ) >> 1; | |
1244 | } | |
1245 | ||
6d6bc0ad | 1246 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1247 | |
e9e9250b PZ |
1248 | static u64 sched_avg_period(void) |
1249 | { | |
1250 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1251 | } | |
1252 | ||
1253 | static void sched_avg_update(struct rq *rq) | |
1254 | { | |
1255 | s64 period = sched_avg_period(); | |
1256 | ||
1257 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1258 | rq->age_stamp += period; | |
1259 | rq->rt_avg /= 2; | |
1260 | } | |
1261 | } | |
1262 | ||
1263 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1264 | { | |
1265 | rq->rt_avg += rt_delta; | |
1266 | sched_avg_update(rq); | |
1267 | } | |
1268 | ||
6d6bc0ad | 1269 | #else /* !CONFIG_SMP */ |
31656519 | 1270 | static void resched_task(struct task_struct *p) |
c24d20db | 1271 | { |
05fa785c | 1272 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1273 | set_tsk_need_resched(p); |
c24d20db | 1274 | } |
e9e9250b PZ |
1275 | |
1276 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1277 | { | |
1278 | } | |
6d6bc0ad | 1279 | #endif /* CONFIG_SMP */ |
c24d20db | 1280 | |
45bf76df IM |
1281 | #if BITS_PER_LONG == 32 |
1282 | # define WMULT_CONST (~0UL) | |
1283 | #else | |
1284 | # define WMULT_CONST (1UL << 32) | |
1285 | #endif | |
1286 | ||
1287 | #define WMULT_SHIFT 32 | |
1288 | ||
194081eb IM |
1289 | /* |
1290 | * Shift right and round: | |
1291 | */ | |
cf2ab469 | 1292 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1293 | |
a7be37ac PZ |
1294 | /* |
1295 | * delta *= weight / lw | |
1296 | */ | |
cb1c4fc9 | 1297 | static unsigned long |
45bf76df IM |
1298 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1299 | struct load_weight *lw) | |
1300 | { | |
1301 | u64 tmp; | |
1302 | ||
7a232e03 LJ |
1303 | if (!lw->inv_weight) { |
1304 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1305 | lw->inv_weight = 1; | |
1306 | else | |
1307 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1308 | / (lw->weight+1); | |
1309 | } | |
45bf76df IM |
1310 | |
1311 | tmp = (u64)delta_exec * weight; | |
1312 | /* | |
1313 | * Check whether we'd overflow the 64-bit multiplication: | |
1314 | */ | |
194081eb | 1315 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1316 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1317 | WMULT_SHIFT/2); |
1318 | else | |
cf2ab469 | 1319 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1320 | |
ecf691da | 1321 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1322 | } |
1323 | ||
1091985b | 1324 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1325 | { |
1326 | lw->weight += inc; | |
e89996ae | 1327 | lw->inv_weight = 0; |
45bf76df IM |
1328 | } |
1329 | ||
1091985b | 1330 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1331 | { |
1332 | lw->weight -= dec; | |
e89996ae | 1333 | lw->inv_weight = 0; |
45bf76df IM |
1334 | } |
1335 | ||
2dd73a4f PW |
1336 | /* |
1337 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1338 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1339 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1340 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1341 | * scaled version of the new time slice allocation that they receive on time |
1342 | * slice expiry etc. | |
1343 | */ | |
1344 | ||
cce7ade8 PZ |
1345 | #define WEIGHT_IDLEPRIO 3 |
1346 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1347 | |
1348 | /* | |
1349 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1350 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1351 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1352 | * that remained on nice 0. | |
1353 | * | |
1354 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1355 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1356 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1357 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1358 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1359 | */ |
1360 | static const int prio_to_weight[40] = { | |
254753dc IM |
1361 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1362 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1363 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1364 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1365 | /* 0 */ 1024, 820, 655, 526, 423, | |
1366 | /* 5 */ 335, 272, 215, 172, 137, | |
1367 | /* 10 */ 110, 87, 70, 56, 45, | |
1368 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1369 | }; |
1370 | ||
5714d2de IM |
1371 | /* |
1372 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1373 | * | |
1374 | * In cases where the weight does not change often, we can use the | |
1375 | * precalculated inverse to speed up arithmetics by turning divisions | |
1376 | * into multiplications: | |
1377 | */ | |
dd41f596 | 1378 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1379 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1380 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1381 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1382 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1383 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1384 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1385 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1386 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1387 | }; |
2dd73a4f | 1388 | |
ef12fefa BR |
1389 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1390 | enum cpuacct_stat_index { | |
1391 | CPUACCT_STAT_USER, /* ... user mode */ | |
1392 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1393 | ||
1394 | CPUACCT_STAT_NSTATS, | |
1395 | }; | |
1396 | ||
d842de87 SV |
1397 | #ifdef CONFIG_CGROUP_CPUACCT |
1398 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1399 | static void cpuacct_update_stats(struct task_struct *tsk, |
1400 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1401 | #else |
1402 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1403 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1404 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1405 | #endif |
1406 | ||
18d95a28 PZ |
1407 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1408 | { | |
1409 | update_load_add(&rq->load, load); | |
1410 | } | |
1411 | ||
1412 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1413 | { | |
1414 | update_load_sub(&rq->load, load); | |
1415 | } | |
1416 | ||
7940ca36 | 1417 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1418 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1419 | |
1420 | /* | |
1421 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1422 | * leaving it for the final time. | |
1423 | */ | |
eb755805 | 1424 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1425 | { |
1426 | struct task_group *parent, *child; | |
eb755805 | 1427 | int ret; |
c09595f6 PZ |
1428 | |
1429 | rcu_read_lock(); | |
1430 | parent = &root_task_group; | |
1431 | down: | |
eb755805 PZ |
1432 | ret = (*down)(parent, data); |
1433 | if (ret) | |
1434 | goto out_unlock; | |
c09595f6 PZ |
1435 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1436 | parent = child; | |
1437 | goto down; | |
1438 | ||
1439 | up: | |
1440 | continue; | |
1441 | } | |
eb755805 PZ |
1442 | ret = (*up)(parent, data); |
1443 | if (ret) | |
1444 | goto out_unlock; | |
c09595f6 PZ |
1445 | |
1446 | child = parent; | |
1447 | parent = parent->parent; | |
1448 | if (parent) | |
1449 | goto up; | |
eb755805 | 1450 | out_unlock: |
c09595f6 | 1451 | rcu_read_unlock(); |
eb755805 PZ |
1452 | |
1453 | return ret; | |
c09595f6 PZ |
1454 | } |
1455 | ||
eb755805 PZ |
1456 | static int tg_nop(struct task_group *tg, void *data) |
1457 | { | |
1458 | return 0; | |
c09595f6 | 1459 | } |
eb755805 PZ |
1460 | #endif |
1461 | ||
1462 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1463 | /* Used instead of source_load when we know the type == 0 */ |
1464 | static unsigned long weighted_cpuload(const int cpu) | |
1465 | { | |
1466 | return cpu_rq(cpu)->load.weight; | |
1467 | } | |
1468 | ||
1469 | /* | |
1470 | * Return a low guess at the load of a migration-source cpu weighted | |
1471 | * according to the scheduling class and "nice" value. | |
1472 | * | |
1473 | * We want to under-estimate the load of migration sources, to | |
1474 | * balance conservatively. | |
1475 | */ | |
1476 | static unsigned long source_load(int cpu, int type) | |
1477 | { | |
1478 | struct rq *rq = cpu_rq(cpu); | |
1479 | unsigned long total = weighted_cpuload(cpu); | |
1480 | ||
1481 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1482 | return total; | |
1483 | ||
1484 | return min(rq->cpu_load[type-1], total); | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Return a high guess at the load of a migration-target cpu weighted | |
1489 | * according to the scheduling class and "nice" value. | |
1490 | */ | |
1491 | static unsigned long target_load(int cpu, int type) | |
1492 | { | |
1493 | struct rq *rq = cpu_rq(cpu); | |
1494 | unsigned long total = weighted_cpuload(cpu); | |
1495 | ||
1496 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1497 | return total; | |
1498 | ||
1499 | return max(rq->cpu_load[type-1], total); | |
1500 | } | |
1501 | ||
ae154be1 PZ |
1502 | static struct sched_group *group_of(int cpu) |
1503 | { | |
d11c563d | 1504 | struct sched_domain *sd = rcu_dereference_sched(cpu_rq(cpu)->sd); |
ae154be1 PZ |
1505 | |
1506 | if (!sd) | |
1507 | return NULL; | |
1508 | ||
1509 | return sd->groups; | |
1510 | } | |
1511 | ||
1512 | static unsigned long power_of(int cpu) | |
1513 | { | |
1514 | struct sched_group *group = group_of(cpu); | |
1515 | ||
1516 | if (!group) | |
1517 | return SCHED_LOAD_SCALE; | |
1518 | ||
1519 | return group->cpu_power; | |
1520 | } | |
1521 | ||
eb755805 PZ |
1522 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1523 | ||
1524 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1525 | { | |
1526 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1527 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1528 | |
4cd42620 SR |
1529 | if (nr_running) |
1530 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1531 | else |
1532 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1533 | |
1534 | return rq->avg_load_per_task; | |
1535 | } | |
1536 | ||
1537 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1538 | |
43cf38eb | 1539 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1540 | |
c09595f6 PZ |
1541 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1542 | ||
1543 | /* | |
1544 | * Calculate and set the cpu's group shares. | |
1545 | */ | |
34d76c41 PZ |
1546 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1547 | unsigned long sd_shares, | |
1548 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1549 | unsigned long *usd_rq_weight) |
18d95a28 | 1550 | { |
34d76c41 | 1551 | unsigned long shares, rq_weight; |
a5004278 | 1552 | int boost = 0; |
c09595f6 | 1553 | |
4a6cc4bd | 1554 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1555 | if (!rq_weight) { |
1556 | boost = 1; | |
1557 | rq_weight = NICE_0_LOAD; | |
1558 | } | |
c8cba857 | 1559 | |
c09595f6 | 1560 | /* |
a8af7246 PZ |
1561 | * \Sum_j shares_j * rq_weight_i |
1562 | * shares_i = ----------------------------- | |
1563 | * \Sum_j rq_weight_j | |
c09595f6 | 1564 | */ |
ec4e0e2f | 1565 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1566 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1567 | |
ffda12a1 PZ |
1568 | if (abs(shares - tg->se[cpu]->load.weight) > |
1569 | sysctl_sched_shares_thresh) { | |
1570 | struct rq *rq = cpu_rq(cpu); | |
1571 | unsigned long flags; | |
c09595f6 | 1572 | |
05fa785c | 1573 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1574 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1575 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1576 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1577 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1578 | } |
18d95a28 | 1579 | } |
c09595f6 PZ |
1580 | |
1581 | /* | |
c8cba857 PZ |
1582 | * Re-compute the task group their per cpu shares over the given domain. |
1583 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1584 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1585 | */ |
eb755805 | 1586 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1587 | { |
cd8ad40d | 1588 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1589 | unsigned long *usd_rq_weight; |
eb755805 | 1590 | struct sched_domain *sd = data; |
34d76c41 | 1591 | unsigned long flags; |
c8cba857 | 1592 | int i; |
c09595f6 | 1593 | |
34d76c41 PZ |
1594 | if (!tg->se[0]) |
1595 | return 0; | |
1596 | ||
1597 | local_irq_save(flags); | |
4a6cc4bd | 1598 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1599 | |
758b2cdc | 1600 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1601 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1602 | usd_rq_weight[i] = weight; |
34d76c41 | 1603 | |
cd8ad40d | 1604 | rq_weight += weight; |
ec4e0e2f KC |
1605 | /* |
1606 | * If there are currently no tasks on the cpu pretend there | |
1607 | * is one of average load so that when a new task gets to | |
1608 | * run here it will not get delayed by group starvation. | |
1609 | */ | |
ec4e0e2f KC |
1610 | if (!weight) |
1611 | weight = NICE_0_LOAD; | |
1612 | ||
cd8ad40d | 1613 | sum_weight += weight; |
c8cba857 | 1614 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1615 | } |
c09595f6 | 1616 | |
cd8ad40d PZ |
1617 | if (!rq_weight) |
1618 | rq_weight = sum_weight; | |
1619 | ||
c8cba857 PZ |
1620 | if ((!shares && rq_weight) || shares > tg->shares) |
1621 | shares = tg->shares; | |
1622 | ||
1623 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1624 | shares = tg->shares; | |
c09595f6 | 1625 | |
758b2cdc | 1626 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1627 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1628 | |
1629 | local_irq_restore(flags); | |
eb755805 PZ |
1630 | |
1631 | return 0; | |
c09595f6 PZ |
1632 | } |
1633 | ||
1634 | /* | |
c8cba857 PZ |
1635 | * Compute the cpu's hierarchical load factor for each task group. |
1636 | * This needs to be done in a top-down fashion because the load of a child | |
1637 | * group is a fraction of its parents load. | |
c09595f6 | 1638 | */ |
eb755805 | 1639 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1640 | { |
c8cba857 | 1641 | unsigned long load; |
eb755805 | 1642 | long cpu = (long)data; |
c09595f6 | 1643 | |
c8cba857 PZ |
1644 | if (!tg->parent) { |
1645 | load = cpu_rq(cpu)->load.weight; | |
1646 | } else { | |
1647 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1648 | load *= tg->cfs_rq[cpu]->shares; | |
1649 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1650 | } | |
c09595f6 | 1651 | |
c8cba857 | 1652 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1653 | |
eb755805 | 1654 | return 0; |
c09595f6 PZ |
1655 | } |
1656 | ||
c8cba857 | 1657 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1658 | { |
e7097159 PZ |
1659 | s64 elapsed; |
1660 | u64 now; | |
1661 | ||
1662 | if (root_task_group_empty()) | |
1663 | return; | |
1664 | ||
1665 | now = cpu_clock(raw_smp_processor_id()); | |
1666 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1667 | |
1668 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1669 | sd->last_update = now; | |
eb755805 | 1670 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1671 | } |
4d8d595d PZ |
1672 | } |
1673 | ||
eb755805 | 1674 | static void update_h_load(long cpu) |
c09595f6 | 1675 | { |
e7097159 PZ |
1676 | if (root_task_group_empty()) |
1677 | return; | |
1678 | ||
eb755805 | 1679 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1680 | } |
1681 | ||
c09595f6 PZ |
1682 | #else |
1683 | ||
c8cba857 | 1684 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1685 | { |
1686 | } | |
1687 | ||
18d95a28 PZ |
1688 | #endif |
1689 | ||
8f45e2b5 GH |
1690 | #ifdef CONFIG_PREEMPT |
1691 | ||
b78bb868 PZ |
1692 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1693 | ||
70574a99 | 1694 | /* |
8f45e2b5 GH |
1695 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1696 | * way at the expense of forcing extra atomic operations in all | |
1697 | * invocations. This assures that the double_lock is acquired using the | |
1698 | * same underlying policy as the spinlock_t on this architecture, which | |
1699 | * reduces latency compared to the unfair variant below. However, it | |
1700 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1701 | */ |
8f45e2b5 GH |
1702 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1703 | __releases(this_rq->lock) | |
1704 | __acquires(busiest->lock) | |
1705 | __acquires(this_rq->lock) | |
1706 | { | |
05fa785c | 1707 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1708 | double_rq_lock(this_rq, busiest); |
1709 | ||
1710 | return 1; | |
1711 | } | |
1712 | ||
1713 | #else | |
1714 | /* | |
1715 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1716 | * latency by eliminating extra atomic operations when the locks are | |
1717 | * already in proper order on entry. This favors lower cpu-ids and will | |
1718 | * grant the double lock to lower cpus over higher ids under contention, | |
1719 | * regardless of entry order into the function. | |
1720 | */ | |
1721 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1722 | __releases(this_rq->lock) |
1723 | __acquires(busiest->lock) | |
1724 | __acquires(this_rq->lock) | |
1725 | { | |
1726 | int ret = 0; | |
1727 | ||
05fa785c | 1728 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1729 | if (busiest < this_rq) { |
05fa785c TG |
1730 | raw_spin_unlock(&this_rq->lock); |
1731 | raw_spin_lock(&busiest->lock); | |
1732 | raw_spin_lock_nested(&this_rq->lock, | |
1733 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1734 | ret = 1; |
1735 | } else | |
05fa785c TG |
1736 | raw_spin_lock_nested(&busiest->lock, |
1737 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1738 | } |
1739 | return ret; | |
1740 | } | |
1741 | ||
8f45e2b5 GH |
1742 | #endif /* CONFIG_PREEMPT */ |
1743 | ||
1744 | /* | |
1745 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1746 | */ | |
1747 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1748 | { | |
1749 | if (unlikely(!irqs_disabled())) { | |
1750 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1751 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1752 | BUG_ON(1); |
1753 | } | |
1754 | ||
1755 | return _double_lock_balance(this_rq, busiest); | |
1756 | } | |
1757 | ||
70574a99 AD |
1758 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1759 | __releases(busiest->lock) | |
1760 | { | |
05fa785c | 1761 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1762 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1763 | } | |
1e3c88bd PZ |
1764 | |
1765 | /* | |
1766 | * double_rq_lock - safely lock two runqueues | |
1767 | * | |
1768 | * Note this does not disable interrupts like task_rq_lock, | |
1769 | * you need to do so manually before calling. | |
1770 | */ | |
1771 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1772 | __acquires(rq1->lock) | |
1773 | __acquires(rq2->lock) | |
1774 | { | |
1775 | BUG_ON(!irqs_disabled()); | |
1776 | if (rq1 == rq2) { | |
1777 | raw_spin_lock(&rq1->lock); | |
1778 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1779 | } else { | |
1780 | if (rq1 < rq2) { | |
1781 | raw_spin_lock(&rq1->lock); | |
1782 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1783 | } else { | |
1784 | raw_spin_lock(&rq2->lock); | |
1785 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1786 | } | |
1787 | } | |
1e3c88bd PZ |
1788 | } |
1789 | ||
1790 | /* | |
1791 | * double_rq_unlock - safely unlock two runqueues | |
1792 | * | |
1793 | * Note this does not restore interrupts like task_rq_unlock, | |
1794 | * you need to do so manually after calling. | |
1795 | */ | |
1796 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1797 | __releases(rq1->lock) | |
1798 | __releases(rq2->lock) | |
1799 | { | |
1800 | raw_spin_unlock(&rq1->lock); | |
1801 | if (rq1 != rq2) | |
1802 | raw_spin_unlock(&rq2->lock); | |
1803 | else | |
1804 | __release(rq2->lock); | |
1805 | } | |
1806 | ||
18d95a28 PZ |
1807 | #endif |
1808 | ||
30432094 | 1809 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1810 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1811 | { | |
30432094 | 1812 | #ifdef CONFIG_SMP |
34e83e85 IM |
1813 | cfs_rq->shares = shares; |
1814 | #endif | |
1815 | } | |
30432094 | 1816 | #endif |
e7693a36 | 1817 | |
dce48a84 | 1818 | static void calc_load_account_active(struct rq *this_rq); |
0bcdcf28 | 1819 | static void update_sysctl(void); |
acb4a848 | 1820 | static int get_update_sysctl_factor(void); |
dce48a84 | 1821 | |
cd29fe6f PZ |
1822 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1823 | { | |
1824 | set_task_rq(p, cpu); | |
1825 | #ifdef CONFIG_SMP | |
1826 | /* | |
1827 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1828 | * successfuly executed on another CPU. We must ensure that updates of | |
1829 | * per-task data have been completed by this moment. | |
1830 | */ | |
1831 | smp_wmb(); | |
1832 | task_thread_info(p)->cpu = cpu; | |
1833 | #endif | |
1834 | } | |
dce48a84 | 1835 | |
1e3c88bd | 1836 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1837 | |
1838 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1839 | #define for_each_class(class) \ |
1840 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1841 | |
1e3c88bd PZ |
1842 | #include "sched_stats.h" |
1843 | ||
c09595f6 | 1844 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1845 | { |
1846 | rq->nr_running++; | |
9c217245 IM |
1847 | } |
1848 | ||
c09595f6 | 1849 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1850 | { |
1851 | rq->nr_running--; | |
9c217245 IM |
1852 | } |
1853 | ||
45bf76df IM |
1854 | static void set_load_weight(struct task_struct *p) |
1855 | { | |
1856 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1857 | p->se.load.weight = prio_to_weight[0] * 2; |
1858 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1859 | return; | |
1860 | } | |
45bf76df | 1861 | |
dd41f596 IM |
1862 | /* |
1863 | * SCHED_IDLE tasks get minimal weight: | |
1864 | */ | |
1865 | if (p->policy == SCHED_IDLE) { | |
1866 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1867 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1868 | return; | |
1869 | } | |
71f8bd46 | 1870 | |
dd41f596 IM |
1871 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1872 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1873 | } |
1874 | ||
2087a1ad GH |
1875 | static void update_avg(u64 *avg, u64 sample) |
1876 | { | |
1877 | s64 diff = sample - *avg; | |
1878 | *avg += diff >> 3; | |
1879 | } | |
1880 | ||
371fd7e7 | 1881 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1882 | { |
a64692a3 | 1883 | update_rq_clock(rq); |
dd41f596 | 1884 | sched_info_queued(p); |
371fd7e7 | 1885 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1886 | p->se.on_rq = 1; |
71f8bd46 IM |
1887 | } |
1888 | ||
371fd7e7 | 1889 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1890 | { |
a64692a3 | 1891 | update_rq_clock(rq); |
46ac22ba | 1892 | sched_info_dequeued(p); |
371fd7e7 | 1893 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1894 | p->se.on_rq = 0; |
71f8bd46 IM |
1895 | } |
1896 | ||
1e3c88bd PZ |
1897 | /* |
1898 | * activate_task - move a task to the runqueue. | |
1899 | */ | |
371fd7e7 | 1900 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1901 | { |
1902 | if (task_contributes_to_load(p)) | |
1903 | rq->nr_uninterruptible--; | |
1904 | ||
371fd7e7 | 1905 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1906 | inc_nr_running(rq); |
1907 | } | |
1908 | ||
1909 | /* | |
1910 | * deactivate_task - remove a task from the runqueue. | |
1911 | */ | |
371fd7e7 | 1912 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1913 | { |
1914 | if (task_contributes_to_load(p)) | |
1915 | rq->nr_uninterruptible++; | |
1916 | ||
371fd7e7 | 1917 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1918 | dec_nr_running(rq); |
1919 | } | |
1920 | ||
1921 | #include "sched_idletask.c" | |
1922 | #include "sched_fair.c" | |
1923 | #include "sched_rt.c" | |
1924 | #ifdef CONFIG_SCHED_DEBUG | |
1925 | # include "sched_debug.c" | |
1926 | #endif | |
1927 | ||
14531189 | 1928 | /* |
dd41f596 | 1929 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1930 | */ |
14531189 IM |
1931 | static inline int __normal_prio(struct task_struct *p) |
1932 | { | |
dd41f596 | 1933 | return p->static_prio; |
14531189 IM |
1934 | } |
1935 | ||
b29739f9 IM |
1936 | /* |
1937 | * Calculate the expected normal priority: i.e. priority | |
1938 | * without taking RT-inheritance into account. Might be | |
1939 | * boosted by interactivity modifiers. Changes upon fork, | |
1940 | * setprio syscalls, and whenever the interactivity | |
1941 | * estimator recalculates. | |
1942 | */ | |
36c8b586 | 1943 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1944 | { |
1945 | int prio; | |
1946 | ||
e05606d3 | 1947 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1948 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1949 | else | |
1950 | prio = __normal_prio(p); | |
1951 | return prio; | |
1952 | } | |
1953 | ||
1954 | /* | |
1955 | * Calculate the current priority, i.e. the priority | |
1956 | * taken into account by the scheduler. This value might | |
1957 | * be boosted by RT tasks, or might be boosted by | |
1958 | * interactivity modifiers. Will be RT if the task got | |
1959 | * RT-boosted. If not then it returns p->normal_prio. | |
1960 | */ | |
36c8b586 | 1961 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1962 | { |
1963 | p->normal_prio = normal_prio(p); | |
1964 | /* | |
1965 | * If we are RT tasks or we were boosted to RT priority, | |
1966 | * keep the priority unchanged. Otherwise, update priority | |
1967 | * to the normal priority: | |
1968 | */ | |
1969 | if (!rt_prio(p->prio)) | |
1970 | return p->normal_prio; | |
1971 | return p->prio; | |
1972 | } | |
1973 | ||
1da177e4 LT |
1974 | /** |
1975 | * task_curr - is this task currently executing on a CPU? | |
1976 | * @p: the task in question. | |
1977 | */ | |
36c8b586 | 1978 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1979 | { |
1980 | return cpu_curr(task_cpu(p)) == p; | |
1981 | } | |
1982 | ||
cb469845 SR |
1983 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1984 | const struct sched_class *prev_class, | |
1985 | int oldprio, int running) | |
1986 | { | |
1987 | if (prev_class != p->sched_class) { | |
1988 | if (prev_class->switched_from) | |
1989 | prev_class->switched_from(rq, p, running); | |
1990 | p->sched_class->switched_to(rq, p, running); | |
1991 | } else | |
1992 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1993 | } | |
1994 | ||
1da177e4 | 1995 | #ifdef CONFIG_SMP |
cc367732 IM |
1996 | /* |
1997 | * Is this task likely cache-hot: | |
1998 | */ | |
e7693a36 | 1999 | static int |
cc367732 IM |
2000 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2001 | { | |
2002 | s64 delta; | |
2003 | ||
e6c8fba7 PZ |
2004 | if (p->sched_class != &fair_sched_class) |
2005 | return 0; | |
2006 | ||
f540a608 IM |
2007 | /* |
2008 | * Buddy candidates are cache hot: | |
2009 | */ | |
f685ceac | 2010 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2011 | (&p->se == cfs_rq_of(&p->se)->next || |
2012 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2013 | return 1; |
2014 | ||
6bc1665b IM |
2015 | if (sysctl_sched_migration_cost == -1) |
2016 | return 1; | |
2017 | if (sysctl_sched_migration_cost == 0) | |
2018 | return 0; | |
2019 | ||
cc367732 IM |
2020 | delta = now - p->se.exec_start; |
2021 | ||
2022 | return delta < (s64)sysctl_sched_migration_cost; | |
2023 | } | |
2024 | ||
dd41f596 | 2025 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2026 | { |
e2912009 PZ |
2027 | #ifdef CONFIG_SCHED_DEBUG |
2028 | /* | |
2029 | * We should never call set_task_cpu() on a blocked task, | |
2030 | * ttwu() will sort out the placement. | |
2031 | */ | |
077614ee PZ |
2032 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2033 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2034 | #endif |
2035 | ||
de1d7286 | 2036 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2037 | |
0c69774e PZ |
2038 | if (task_cpu(p) != new_cpu) { |
2039 | p->se.nr_migrations++; | |
2040 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2041 | } | |
dd41f596 IM |
2042 | |
2043 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2044 | } |
2045 | ||
70b97a7f | 2046 | struct migration_req { |
1da177e4 | 2047 | struct list_head list; |
1da177e4 | 2048 | |
36c8b586 | 2049 | struct task_struct *task; |
1da177e4 LT |
2050 | int dest_cpu; |
2051 | ||
1da177e4 | 2052 | struct completion done; |
70b97a7f | 2053 | }; |
1da177e4 LT |
2054 | |
2055 | /* | |
2056 | * The task's runqueue lock must be held. | |
2057 | * Returns true if you have to wait for migration thread. | |
2058 | */ | |
36c8b586 | 2059 | static int |
70b97a7f | 2060 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2061 | { |
70b97a7f | 2062 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2063 | |
2064 | /* | |
2065 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2066 | * the next wake-up will properly place the task. |
1da177e4 | 2067 | */ |
e2912009 | 2068 | if (!p->se.on_rq && !task_running(rq, p)) |
1da177e4 | 2069 | return 0; |
1da177e4 LT |
2070 | |
2071 | init_completion(&req->done); | |
1da177e4 LT |
2072 | req->task = p; |
2073 | req->dest_cpu = dest_cpu; | |
2074 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2075 | |
1da177e4 LT |
2076 | return 1; |
2077 | } | |
2078 | ||
a26b89f0 MM |
2079 | /* |
2080 | * wait_task_context_switch - wait for a thread to complete at least one | |
2081 | * context switch. | |
2082 | * | |
2083 | * @p must not be current. | |
2084 | */ | |
2085 | void wait_task_context_switch(struct task_struct *p) | |
2086 | { | |
2087 | unsigned long nvcsw, nivcsw, flags; | |
2088 | int running; | |
2089 | struct rq *rq; | |
2090 | ||
2091 | nvcsw = p->nvcsw; | |
2092 | nivcsw = p->nivcsw; | |
2093 | for (;;) { | |
2094 | /* | |
2095 | * The runqueue is assigned before the actual context | |
2096 | * switch. We need to take the runqueue lock. | |
2097 | * | |
2098 | * We could check initially without the lock but it is | |
2099 | * very likely that we need to take the lock in every | |
2100 | * iteration. | |
2101 | */ | |
2102 | rq = task_rq_lock(p, &flags); | |
2103 | running = task_running(rq, p); | |
2104 | task_rq_unlock(rq, &flags); | |
2105 | ||
2106 | if (likely(!running)) | |
2107 | break; | |
2108 | /* | |
2109 | * The switch count is incremented before the actual | |
2110 | * context switch. We thus wait for two switches to be | |
2111 | * sure at least one completed. | |
2112 | */ | |
2113 | if ((p->nvcsw - nvcsw) > 1) | |
2114 | break; | |
2115 | if ((p->nivcsw - nivcsw) > 1) | |
2116 | break; | |
2117 | ||
2118 | cpu_relax(); | |
2119 | } | |
2120 | } | |
2121 | ||
1da177e4 LT |
2122 | /* |
2123 | * wait_task_inactive - wait for a thread to unschedule. | |
2124 | * | |
85ba2d86 RM |
2125 | * If @match_state is nonzero, it's the @p->state value just checked and |
2126 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2127 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2128 | * we return a positive number (its total switch count). If a second call | |
2129 | * a short while later returns the same number, the caller can be sure that | |
2130 | * @p has remained unscheduled the whole time. | |
2131 | * | |
1da177e4 LT |
2132 | * The caller must ensure that the task *will* unschedule sometime soon, |
2133 | * else this function might spin for a *long* time. This function can't | |
2134 | * be called with interrupts off, or it may introduce deadlock with | |
2135 | * smp_call_function() if an IPI is sent by the same process we are | |
2136 | * waiting to become inactive. | |
2137 | */ | |
85ba2d86 | 2138 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2139 | { |
2140 | unsigned long flags; | |
dd41f596 | 2141 | int running, on_rq; |
85ba2d86 | 2142 | unsigned long ncsw; |
70b97a7f | 2143 | struct rq *rq; |
1da177e4 | 2144 | |
3a5c359a AK |
2145 | for (;;) { |
2146 | /* | |
2147 | * We do the initial early heuristics without holding | |
2148 | * any task-queue locks at all. We'll only try to get | |
2149 | * the runqueue lock when things look like they will | |
2150 | * work out! | |
2151 | */ | |
2152 | rq = task_rq(p); | |
fa490cfd | 2153 | |
3a5c359a AK |
2154 | /* |
2155 | * If the task is actively running on another CPU | |
2156 | * still, just relax and busy-wait without holding | |
2157 | * any locks. | |
2158 | * | |
2159 | * NOTE! Since we don't hold any locks, it's not | |
2160 | * even sure that "rq" stays as the right runqueue! | |
2161 | * But we don't care, since "task_running()" will | |
2162 | * return false if the runqueue has changed and p | |
2163 | * is actually now running somewhere else! | |
2164 | */ | |
85ba2d86 RM |
2165 | while (task_running(rq, p)) { |
2166 | if (match_state && unlikely(p->state != match_state)) | |
2167 | return 0; | |
3a5c359a | 2168 | cpu_relax(); |
85ba2d86 | 2169 | } |
fa490cfd | 2170 | |
3a5c359a AK |
2171 | /* |
2172 | * Ok, time to look more closely! We need the rq | |
2173 | * lock now, to be *sure*. If we're wrong, we'll | |
2174 | * just go back and repeat. | |
2175 | */ | |
2176 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2177 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2178 | running = task_running(rq, p); |
2179 | on_rq = p->se.on_rq; | |
85ba2d86 | 2180 | ncsw = 0; |
f31e11d8 | 2181 | if (!match_state || p->state == match_state) |
93dcf55f | 2182 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2183 | task_rq_unlock(rq, &flags); |
fa490cfd | 2184 | |
85ba2d86 RM |
2185 | /* |
2186 | * If it changed from the expected state, bail out now. | |
2187 | */ | |
2188 | if (unlikely(!ncsw)) | |
2189 | break; | |
2190 | ||
3a5c359a AK |
2191 | /* |
2192 | * Was it really running after all now that we | |
2193 | * checked with the proper locks actually held? | |
2194 | * | |
2195 | * Oops. Go back and try again.. | |
2196 | */ | |
2197 | if (unlikely(running)) { | |
2198 | cpu_relax(); | |
2199 | continue; | |
2200 | } | |
fa490cfd | 2201 | |
3a5c359a AK |
2202 | /* |
2203 | * It's not enough that it's not actively running, | |
2204 | * it must be off the runqueue _entirely_, and not | |
2205 | * preempted! | |
2206 | * | |
80dd99b3 | 2207 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2208 | * running right now), it's preempted, and we should |
2209 | * yield - it could be a while. | |
2210 | */ | |
2211 | if (unlikely(on_rq)) { | |
2212 | schedule_timeout_uninterruptible(1); | |
2213 | continue; | |
2214 | } | |
fa490cfd | 2215 | |
3a5c359a AK |
2216 | /* |
2217 | * Ahh, all good. It wasn't running, and it wasn't | |
2218 | * runnable, which means that it will never become | |
2219 | * running in the future either. We're all done! | |
2220 | */ | |
2221 | break; | |
2222 | } | |
85ba2d86 RM |
2223 | |
2224 | return ncsw; | |
1da177e4 LT |
2225 | } |
2226 | ||
2227 | /*** | |
2228 | * kick_process - kick a running thread to enter/exit the kernel | |
2229 | * @p: the to-be-kicked thread | |
2230 | * | |
2231 | * Cause a process which is running on another CPU to enter | |
2232 | * kernel-mode, without any delay. (to get signals handled.) | |
2233 | * | |
2234 | * NOTE: this function doesnt have to take the runqueue lock, | |
2235 | * because all it wants to ensure is that the remote task enters | |
2236 | * the kernel. If the IPI races and the task has been migrated | |
2237 | * to another CPU then no harm is done and the purpose has been | |
2238 | * achieved as well. | |
2239 | */ | |
36c8b586 | 2240 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2241 | { |
2242 | int cpu; | |
2243 | ||
2244 | preempt_disable(); | |
2245 | cpu = task_cpu(p); | |
2246 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2247 | smp_send_reschedule(cpu); | |
2248 | preempt_enable(); | |
2249 | } | |
b43e3521 | 2250 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2251 | #endif /* CONFIG_SMP */ |
1da177e4 | 2252 | |
0793a61d TG |
2253 | /** |
2254 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2255 | * @p: the task to evaluate | |
2256 | * @func: the function to be called | |
2257 | * @info: the function call argument | |
2258 | * | |
2259 | * Calls the function @func when the task is currently running. This might | |
2260 | * be on the current CPU, which just calls the function directly | |
2261 | */ | |
2262 | void task_oncpu_function_call(struct task_struct *p, | |
2263 | void (*func) (void *info), void *info) | |
2264 | { | |
2265 | int cpu; | |
2266 | ||
2267 | preempt_disable(); | |
2268 | cpu = task_cpu(p); | |
2269 | if (task_curr(p)) | |
2270 | smp_call_function_single(cpu, func, info, 1); | |
2271 | preempt_enable(); | |
2272 | } | |
2273 | ||
970b13ba | 2274 | #ifdef CONFIG_SMP |
30da688e ON |
2275 | /* |
2276 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2277 | */ | |
5da9a0fb PZ |
2278 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2279 | { | |
2280 | int dest_cpu; | |
2281 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2282 | ||
2283 | /* Look for allowed, online CPU in same node. */ | |
2284 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2285 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2286 | return dest_cpu; | |
2287 | ||
2288 | /* Any allowed, online CPU? */ | |
2289 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2290 | if (dest_cpu < nr_cpu_ids) | |
2291 | return dest_cpu; | |
2292 | ||
2293 | /* No more Mr. Nice Guy. */ | |
897f0b3c | 2294 | if (unlikely(dest_cpu >= nr_cpu_ids)) { |
9084bb82 | 2295 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
5da9a0fb PZ |
2296 | /* |
2297 | * Don't tell them about moving exiting tasks or | |
2298 | * kernel threads (both mm NULL), since they never | |
2299 | * leave kernel. | |
2300 | */ | |
2301 | if (p->mm && printk_ratelimit()) { | |
2302 | printk(KERN_INFO "process %d (%s) no " | |
2303 | "longer affine to cpu%d\n", | |
2304 | task_pid_nr(p), p->comm, cpu); | |
2305 | } | |
2306 | } | |
2307 | ||
2308 | return dest_cpu; | |
2309 | } | |
2310 | ||
e2912009 | 2311 | /* |
30da688e | 2312 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2313 | */ |
970b13ba | 2314 | static inline |
0017d735 | 2315 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2316 | { |
0017d735 | 2317 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2318 | |
2319 | /* | |
2320 | * In order not to call set_task_cpu() on a blocking task we need | |
2321 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2322 | * cpu. | |
2323 | * | |
2324 | * Since this is common to all placement strategies, this lives here. | |
2325 | * | |
2326 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2327 | * not worry about this generic constraint ] | |
2328 | */ | |
2329 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2330 | !cpu_online(cpu))) |
5da9a0fb | 2331 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2332 | |
2333 | return cpu; | |
970b13ba PZ |
2334 | } |
2335 | #endif | |
2336 | ||
1da177e4 LT |
2337 | /*** |
2338 | * try_to_wake_up - wake up a thread | |
2339 | * @p: the to-be-woken-up thread | |
2340 | * @state: the mask of task states that can be woken | |
2341 | * @sync: do a synchronous wakeup? | |
2342 | * | |
2343 | * Put it on the run-queue if it's not already there. The "current" | |
2344 | * thread is always on the run-queue (except when the actual | |
2345 | * re-schedule is in progress), and as such you're allowed to do | |
2346 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2347 | * runnable without the overhead of this. | |
2348 | * | |
2349 | * returns failure only if the task is already active. | |
2350 | */ | |
7d478721 PZ |
2351 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2352 | int wake_flags) | |
1da177e4 | 2353 | { |
cc367732 | 2354 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2355 | unsigned long flags; |
371fd7e7 | 2356 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2357 | struct rq *rq; |
1da177e4 | 2358 | |
e9c84311 | 2359 | this_cpu = get_cpu(); |
2398f2c6 | 2360 | |
04e2f174 | 2361 | smp_wmb(); |
ab3b3aa5 | 2362 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2363 | if (!(p->state & state)) |
1da177e4 LT |
2364 | goto out; |
2365 | ||
dd41f596 | 2366 | if (p->se.on_rq) |
1da177e4 LT |
2367 | goto out_running; |
2368 | ||
2369 | cpu = task_cpu(p); | |
cc367732 | 2370 | orig_cpu = cpu; |
1da177e4 LT |
2371 | |
2372 | #ifdef CONFIG_SMP | |
2373 | if (unlikely(task_running(rq, p))) | |
2374 | goto out_activate; | |
2375 | ||
e9c84311 PZ |
2376 | /* |
2377 | * In order to handle concurrent wakeups and release the rq->lock | |
2378 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2379 | * |
2380 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2381 | */ |
cc87f76a PZ |
2382 | if (task_contributes_to_load(p)) { |
2383 | if (likely(cpu_online(orig_cpu))) | |
2384 | rq->nr_uninterruptible--; | |
2385 | else | |
2386 | this_rq()->nr_uninterruptible--; | |
2387 | } | |
e9c84311 | 2388 | p->state = TASK_WAKING; |
efbbd05a | 2389 | |
371fd7e7 | 2390 | if (p->sched_class->task_waking) { |
efbbd05a | 2391 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2392 | en_flags |= ENQUEUE_WAKING; |
2393 | } | |
efbbd05a | 2394 | |
0017d735 PZ |
2395 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2396 | if (cpu != orig_cpu) | |
5d2f5a61 | 2397 | set_task_cpu(p, cpu); |
0017d735 | 2398 | __task_rq_unlock(rq); |
ab19cb23 | 2399 | |
0970d299 PZ |
2400 | rq = cpu_rq(cpu); |
2401 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2402 | |
0970d299 PZ |
2403 | /* |
2404 | * We migrated the task without holding either rq->lock, however | |
2405 | * since the task is not on the task list itself, nobody else | |
2406 | * will try and migrate the task, hence the rq should match the | |
2407 | * cpu we just moved it to. | |
2408 | */ | |
2409 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2410 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2411 | |
e7693a36 GH |
2412 | #ifdef CONFIG_SCHEDSTATS |
2413 | schedstat_inc(rq, ttwu_count); | |
2414 | if (cpu == this_cpu) | |
2415 | schedstat_inc(rq, ttwu_local); | |
2416 | else { | |
2417 | struct sched_domain *sd; | |
2418 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2419 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2420 | schedstat_inc(sd, ttwu_wake_remote); |
2421 | break; | |
2422 | } | |
2423 | } | |
2424 | } | |
6d6bc0ad | 2425 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2426 | |
1da177e4 LT |
2427 | out_activate: |
2428 | #endif /* CONFIG_SMP */ | |
41acab88 | 2429 | schedstat_inc(p, se.statistics.nr_wakeups); |
7d478721 | 2430 | if (wake_flags & WF_SYNC) |
41acab88 | 2431 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
cc367732 | 2432 | if (orig_cpu != cpu) |
41acab88 | 2433 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
cc367732 | 2434 | if (cpu == this_cpu) |
41acab88 | 2435 | schedstat_inc(p, se.statistics.nr_wakeups_local); |
cc367732 | 2436 | else |
41acab88 | 2437 | schedstat_inc(p, se.statistics.nr_wakeups_remote); |
371fd7e7 | 2438 | activate_task(rq, p, en_flags); |
1da177e4 LT |
2439 | success = 1; |
2440 | ||
2441 | out_running: | |
468a15bb | 2442 | trace_sched_wakeup(rq, p, success); |
7d478721 | 2443 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2444 | |
1da177e4 | 2445 | p->state = TASK_RUNNING; |
9a897c5a | 2446 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2447 | if (p->sched_class->task_woken) |
2448 | p->sched_class->task_woken(rq, p); | |
eae0c9df MG |
2449 | |
2450 | if (unlikely(rq->idle_stamp)) { | |
2451 | u64 delta = rq->clock - rq->idle_stamp; | |
2452 | u64 max = 2*sysctl_sched_migration_cost; | |
2453 | ||
2454 | if (delta > max) | |
2455 | rq->avg_idle = max; | |
2456 | else | |
2457 | update_avg(&rq->avg_idle, delta); | |
2458 | rq->idle_stamp = 0; | |
2459 | } | |
9a897c5a | 2460 | #endif |
1da177e4 LT |
2461 | out: |
2462 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2463 | put_cpu(); |
1da177e4 LT |
2464 | |
2465 | return success; | |
2466 | } | |
2467 | ||
50fa610a DH |
2468 | /** |
2469 | * wake_up_process - Wake up a specific process | |
2470 | * @p: The process to be woken up. | |
2471 | * | |
2472 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2473 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2474 | * running. | |
2475 | * | |
2476 | * It may be assumed that this function implies a write memory barrier before | |
2477 | * changing the task state if and only if any tasks are woken up. | |
2478 | */ | |
7ad5b3a5 | 2479 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2480 | { |
d9514f6c | 2481 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2482 | } |
1da177e4 LT |
2483 | EXPORT_SYMBOL(wake_up_process); |
2484 | ||
7ad5b3a5 | 2485 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2486 | { |
2487 | return try_to_wake_up(p, state, 0); | |
2488 | } | |
2489 | ||
1da177e4 LT |
2490 | /* |
2491 | * Perform scheduler related setup for a newly forked process p. | |
2492 | * p is forked by current. | |
dd41f596 IM |
2493 | * |
2494 | * __sched_fork() is basic setup used by init_idle() too: | |
2495 | */ | |
2496 | static void __sched_fork(struct task_struct *p) | |
2497 | { | |
dd41f596 IM |
2498 | p->se.exec_start = 0; |
2499 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2500 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2501 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2502 | |
2503 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2504 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2505 | #endif |
476d139c | 2506 | |
fa717060 | 2507 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2508 | p->se.on_rq = 0; |
4a55bd5e | 2509 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2510 | |
e107be36 AK |
2511 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2512 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2513 | #endif | |
dd41f596 IM |
2514 | } |
2515 | ||
2516 | /* | |
2517 | * fork()/clone()-time setup: | |
2518 | */ | |
2519 | void sched_fork(struct task_struct *p, int clone_flags) | |
2520 | { | |
2521 | int cpu = get_cpu(); | |
2522 | ||
2523 | __sched_fork(p); | |
06b83b5f | 2524 | /* |
0017d735 | 2525 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2526 | * nobody will actually run it, and a signal or other external |
2527 | * event cannot wake it up and insert it on the runqueue either. | |
2528 | */ | |
0017d735 | 2529 | p->state = TASK_RUNNING; |
dd41f596 | 2530 | |
b9dc29e7 MG |
2531 | /* |
2532 | * Revert to default priority/policy on fork if requested. | |
2533 | */ | |
2534 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2535 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2536 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2537 | p->normal_prio = p->static_prio; |
2538 | } | |
b9dc29e7 | 2539 | |
6c697bdf MG |
2540 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2541 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2542 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2543 | set_load_weight(p); |
2544 | } | |
2545 | ||
b9dc29e7 MG |
2546 | /* |
2547 | * We don't need the reset flag anymore after the fork. It has | |
2548 | * fulfilled its duty: | |
2549 | */ | |
2550 | p->sched_reset_on_fork = 0; | |
2551 | } | |
ca94c442 | 2552 | |
f83f9ac2 PW |
2553 | /* |
2554 | * Make sure we do not leak PI boosting priority to the child. | |
2555 | */ | |
2556 | p->prio = current->normal_prio; | |
2557 | ||
2ddbf952 HS |
2558 | if (!rt_prio(p->prio)) |
2559 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2560 | |
cd29fe6f PZ |
2561 | if (p->sched_class->task_fork) |
2562 | p->sched_class->task_fork(p); | |
2563 | ||
5f3edc1b PZ |
2564 | set_task_cpu(p, cpu); |
2565 | ||
52f17b6c | 2566 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2567 | if (likely(sched_info_on())) |
52f17b6c | 2568 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2569 | #endif |
d6077cb8 | 2570 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2571 | p->oncpu = 0; |
2572 | #endif | |
1da177e4 | 2573 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2574 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2575 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2576 | #endif |
917b627d GH |
2577 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2578 | ||
476d139c | 2579 | put_cpu(); |
1da177e4 LT |
2580 | } |
2581 | ||
2582 | /* | |
2583 | * wake_up_new_task - wake up a newly created task for the first time. | |
2584 | * | |
2585 | * This function will do some initial scheduler statistics housekeeping | |
2586 | * that must be done for every newly created context, then puts the task | |
2587 | * on the runqueue and wakes it. | |
2588 | */ | |
7ad5b3a5 | 2589 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2590 | { |
2591 | unsigned long flags; | |
dd41f596 | 2592 | struct rq *rq; |
c890692b | 2593 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2594 | |
2595 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2596 | rq = task_rq_lock(p, &flags); |
2597 | p->state = TASK_WAKING; | |
2598 | ||
fabf318e PZ |
2599 | /* |
2600 | * Fork balancing, do it here and not earlier because: | |
2601 | * - cpus_allowed can change in the fork path | |
2602 | * - any previously selected cpu might disappear through hotplug | |
2603 | * | |
0017d735 PZ |
2604 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2605 | * without people poking at ->cpus_allowed. | |
fabf318e | 2606 | */ |
0017d735 | 2607 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2608 | set_task_cpu(p, cpu); |
0970d299 | 2609 | |
06b83b5f | 2610 | p->state = TASK_RUNNING; |
0017d735 PZ |
2611 | task_rq_unlock(rq, &flags); |
2612 | #endif | |
2613 | ||
2614 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2615 | activate_task(rq, p, 0); |
c71dd42d | 2616 | trace_sched_wakeup_new(rq, p, 1); |
a7558e01 | 2617 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2618 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2619 | if (p->sched_class->task_woken) |
2620 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2621 | #endif |
dd41f596 | 2622 | task_rq_unlock(rq, &flags); |
fabf318e | 2623 | put_cpu(); |
1da177e4 LT |
2624 | } |
2625 | ||
e107be36 AK |
2626 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2627 | ||
2628 | /** | |
80dd99b3 | 2629 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2630 | * @notifier: notifier struct to register |
e107be36 AK |
2631 | */ |
2632 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2633 | { | |
2634 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2635 | } | |
2636 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2637 | ||
2638 | /** | |
2639 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2640 | * @notifier: notifier struct to unregister |
e107be36 AK |
2641 | * |
2642 | * This is safe to call from within a preemption notifier. | |
2643 | */ | |
2644 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2645 | { | |
2646 | hlist_del(¬ifier->link); | |
2647 | } | |
2648 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2649 | ||
2650 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2651 | { | |
2652 | struct preempt_notifier *notifier; | |
2653 | struct hlist_node *node; | |
2654 | ||
2655 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2656 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2657 | } | |
2658 | ||
2659 | static void | |
2660 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2661 | struct task_struct *next) | |
2662 | { | |
2663 | struct preempt_notifier *notifier; | |
2664 | struct hlist_node *node; | |
2665 | ||
2666 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2667 | notifier->ops->sched_out(notifier, next); | |
2668 | } | |
2669 | ||
6d6bc0ad | 2670 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2671 | |
2672 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2673 | { | |
2674 | } | |
2675 | ||
2676 | static void | |
2677 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2678 | struct task_struct *next) | |
2679 | { | |
2680 | } | |
2681 | ||
6d6bc0ad | 2682 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2683 | |
4866cde0 NP |
2684 | /** |
2685 | * prepare_task_switch - prepare to switch tasks | |
2686 | * @rq: the runqueue preparing to switch | |
421cee29 | 2687 | * @prev: the current task that is being switched out |
4866cde0 NP |
2688 | * @next: the task we are going to switch to. |
2689 | * | |
2690 | * This is called with the rq lock held and interrupts off. It must | |
2691 | * be paired with a subsequent finish_task_switch after the context | |
2692 | * switch. | |
2693 | * | |
2694 | * prepare_task_switch sets up locking and calls architecture specific | |
2695 | * hooks. | |
2696 | */ | |
e107be36 AK |
2697 | static inline void |
2698 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2699 | struct task_struct *next) | |
4866cde0 | 2700 | { |
e107be36 | 2701 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2702 | prepare_lock_switch(rq, next); |
2703 | prepare_arch_switch(next); | |
2704 | } | |
2705 | ||
1da177e4 LT |
2706 | /** |
2707 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2708 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2709 | * @prev: the thread we just switched away from. |
2710 | * | |
4866cde0 NP |
2711 | * finish_task_switch must be called after the context switch, paired |
2712 | * with a prepare_task_switch call before the context switch. | |
2713 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2714 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2715 | * |
2716 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2717 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2718 | * with the lock held can cause deadlocks; see schedule() for |
2719 | * details.) | |
2720 | */ | |
a9957449 | 2721 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2722 | __releases(rq->lock) |
2723 | { | |
1da177e4 | 2724 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2725 | long prev_state; |
1da177e4 LT |
2726 | |
2727 | rq->prev_mm = NULL; | |
2728 | ||
2729 | /* | |
2730 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2731 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2732 | * schedule one last time. The schedule call will never return, and |
2733 | * the scheduled task must drop that reference. | |
c394cc9f | 2734 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2735 | * still held, otherwise prev could be scheduled on another cpu, die |
2736 | * there before we look at prev->state, and then the reference would | |
2737 | * be dropped twice. | |
2738 | * Manfred Spraul <manfred@colorfullife.com> | |
2739 | */ | |
55a101f8 | 2740 | prev_state = prev->state; |
4866cde0 | 2741 | finish_arch_switch(prev); |
8381f65d JI |
2742 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2743 | local_irq_disable(); | |
2744 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2745 | perf_event_task_sched_in(current); |
8381f65d JI |
2746 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2747 | local_irq_enable(); | |
2748 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2749 | finish_lock_switch(rq, prev); |
e8fa1362 | 2750 | |
e107be36 | 2751 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2752 | if (mm) |
2753 | mmdrop(mm); | |
c394cc9f | 2754 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2755 | /* |
2756 | * Remove function-return probe instances associated with this | |
2757 | * task and put them back on the free list. | |
9761eea8 | 2758 | */ |
c6fd91f0 | 2759 | kprobe_flush_task(prev); |
1da177e4 | 2760 | put_task_struct(prev); |
c6fd91f0 | 2761 | } |
1da177e4 LT |
2762 | } |
2763 | ||
3f029d3c GH |
2764 | #ifdef CONFIG_SMP |
2765 | ||
2766 | /* assumes rq->lock is held */ | |
2767 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2768 | { | |
2769 | if (prev->sched_class->pre_schedule) | |
2770 | prev->sched_class->pre_schedule(rq, prev); | |
2771 | } | |
2772 | ||
2773 | /* rq->lock is NOT held, but preemption is disabled */ | |
2774 | static inline void post_schedule(struct rq *rq) | |
2775 | { | |
2776 | if (rq->post_schedule) { | |
2777 | unsigned long flags; | |
2778 | ||
05fa785c | 2779 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2780 | if (rq->curr->sched_class->post_schedule) |
2781 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2782 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2783 | |
2784 | rq->post_schedule = 0; | |
2785 | } | |
2786 | } | |
2787 | ||
2788 | #else | |
da19ab51 | 2789 | |
3f029d3c GH |
2790 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2791 | { | |
2792 | } | |
2793 | ||
2794 | static inline void post_schedule(struct rq *rq) | |
2795 | { | |
1da177e4 LT |
2796 | } |
2797 | ||
3f029d3c GH |
2798 | #endif |
2799 | ||
1da177e4 LT |
2800 | /** |
2801 | * schedule_tail - first thing a freshly forked thread must call. | |
2802 | * @prev: the thread we just switched away from. | |
2803 | */ | |
36c8b586 | 2804 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2805 | __releases(rq->lock) |
2806 | { | |
70b97a7f IM |
2807 | struct rq *rq = this_rq(); |
2808 | ||
4866cde0 | 2809 | finish_task_switch(rq, prev); |
da19ab51 | 2810 | |
3f029d3c GH |
2811 | /* |
2812 | * FIXME: do we need to worry about rq being invalidated by the | |
2813 | * task_switch? | |
2814 | */ | |
2815 | post_schedule(rq); | |
70b97a7f | 2816 | |
4866cde0 NP |
2817 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2818 | /* In this case, finish_task_switch does not reenable preemption */ | |
2819 | preempt_enable(); | |
2820 | #endif | |
1da177e4 | 2821 | if (current->set_child_tid) |
b488893a | 2822 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2823 | } |
2824 | ||
2825 | /* | |
2826 | * context_switch - switch to the new MM and the new | |
2827 | * thread's register state. | |
2828 | */ | |
dd41f596 | 2829 | static inline void |
70b97a7f | 2830 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2831 | struct task_struct *next) |
1da177e4 | 2832 | { |
dd41f596 | 2833 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2834 | |
e107be36 | 2835 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2836 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2837 | mm = next->mm; |
2838 | oldmm = prev->active_mm; | |
9226d125 ZA |
2839 | /* |
2840 | * For paravirt, this is coupled with an exit in switch_to to | |
2841 | * combine the page table reload and the switch backend into | |
2842 | * one hypercall. | |
2843 | */ | |
224101ed | 2844 | arch_start_context_switch(prev); |
9226d125 | 2845 | |
710390d9 | 2846 | if (likely(!mm)) { |
1da177e4 LT |
2847 | next->active_mm = oldmm; |
2848 | atomic_inc(&oldmm->mm_count); | |
2849 | enter_lazy_tlb(oldmm, next); | |
2850 | } else | |
2851 | switch_mm(oldmm, mm, next); | |
2852 | ||
710390d9 | 2853 | if (likely(!prev->mm)) { |
1da177e4 | 2854 | prev->active_mm = NULL; |
1da177e4 LT |
2855 | rq->prev_mm = oldmm; |
2856 | } | |
3a5f5e48 IM |
2857 | /* |
2858 | * Since the runqueue lock will be released by the next | |
2859 | * task (which is an invalid locking op but in the case | |
2860 | * of the scheduler it's an obvious special-case), so we | |
2861 | * do an early lockdep release here: | |
2862 | */ | |
2863 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2864 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2865 | #endif |
1da177e4 LT |
2866 | |
2867 | /* Here we just switch the register state and the stack. */ | |
2868 | switch_to(prev, next, prev); | |
2869 | ||
dd41f596 IM |
2870 | barrier(); |
2871 | /* | |
2872 | * this_rq must be evaluated again because prev may have moved | |
2873 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2874 | * frame will be invalid. | |
2875 | */ | |
2876 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2877 | } |
2878 | ||
2879 | /* | |
2880 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2881 | * | |
2882 | * externally visible scheduler statistics: current number of runnable | |
2883 | * threads, current number of uninterruptible-sleeping threads, total | |
2884 | * number of context switches performed since bootup. | |
2885 | */ | |
2886 | unsigned long nr_running(void) | |
2887 | { | |
2888 | unsigned long i, sum = 0; | |
2889 | ||
2890 | for_each_online_cpu(i) | |
2891 | sum += cpu_rq(i)->nr_running; | |
2892 | ||
2893 | return sum; | |
f711f609 | 2894 | } |
1da177e4 LT |
2895 | |
2896 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2897 | { |
1da177e4 | 2898 | unsigned long i, sum = 0; |
f711f609 | 2899 | |
0a945022 | 2900 | for_each_possible_cpu(i) |
1da177e4 | 2901 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2902 | |
2903 | /* | |
1da177e4 LT |
2904 | * Since we read the counters lockless, it might be slightly |
2905 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2906 | */ |
1da177e4 LT |
2907 | if (unlikely((long)sum < 0)) |
2908 | sum = 0; | |
f711f609 | 2909 | |
1da177e4 | 2910 | return sum; |
f711f609 | 2911 | } |
f711f609 | 2912 | |
1da177e4 | 2913 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2914 | { |
cc94abfc SR |
2915 | int i; |
2916 | unsigned long long sum = 0; | |
46cb4b7c | 2917 | |
0a945022 | 2918 | for_each_possible_cpu(i) |
1da177e4 | 2919 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2920 | |
1da177e4 LT |
2921 | return sum; |
2922 | } | |
483b4ee6 | 2923 | |
1da177e4 LT |
2924 | unsigned long nr_iowait(void) |
2925 | { | |
2926 | unsigned long i, sum = 0; | |
483b4ee6 | 2927 | |
0a945022 | 2928 | for_each_possible_cpu(i) |
1da177e4 | 2929 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2930 | |
1da177e4 LT |
2931 | return sum; |
2932 | } | |
483b4ee6 | 2933 | |
69d25870 AV |
2934 | unsigned long nr_iowait_cpu(void) |
2935 | { | |
2936 | struct rq *this = this_rq(); | |
2937 | return atomic_read(&this->nr_iowait); | |
2938 | } | |
46cb4b7c | 2939 | |
69d25870 AV |
2940 | unsigned long this_cpu_load(void) |
2941 | { | |
2942 | struct rq *this = this_rq(); | |
2943 | return this->cpu_load[0]; | |
2944 | } | |
e790fb0b | 2945 | |
46cb4b7c | 2946 | |
dce48a84 TG |
2947 | /* Variables and functions for calc_load */ |
2948 | static atomic_long_t calc_load_tasks; | |
2949 | static unsigned long calc_load_update; | |
2950 | unsigned long avenrun[3]; | |
2951 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2952 | |
2d02494f TG |
2953 | /** |
2954 | * get_avenrun - get the load average array | |
2955 | * @loads: pointer to dest load array | |
2956 | * @offset: offset to add | |
2957 | * @shift: shift count to shift the result left | |
2958 | * | |
2959 | * These values are estimates at best, so no need for locking. | |
2960 | */ | |
2961 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2962 | { | |
2963 | loads[0] = (avenrun[0] + offset) << shift; | |
2964 | loads[1] = (avenrun[1] + offset) << shift; | |
2965 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2966 | } |
46cb4b7c | 2967 | |
dce48a84 TG |
2968 | static unsigned long |
2969 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2970 | { |
dce48a84 TG |
2971 | load *= exp; |
2972 | load += active * (FIXED_1 - exp); | |
2973 | return load >> FSHIFT; | |
2974 | } | |
46cb4b7c SS |
2975 | |
2976 | /* | |
dce48a84 TG |
2977 | * calc_load - update the avenrun load estimates 10 ticks after the |
2978 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2979 | */ |
dce48a84 | 2980 | void calc_global_load(void) |
7835b98b | 2981 | { |
dce48a84 TG |
2982 | unsigned long upd = calc_load_update + 10; |
2983 | long active; | |
1da177e4 | 2984 | |
dce48a84 TG |
2985 | if (time_before(jiffies, upd)) |
2986 | return; | |
1da177e4 | 2987 | |
dce48a84 TG |
2988 | active = atomic_long_read(&calc_load_tasks); |
2989 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2990 | |
dce48a84 TG |
2991 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
2992 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
2993 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 2994 | |
dce48a84 TG |
2995 | calc_load_update += LOAD_FREQ; |
2996 | } | |
1da177e4 | 2997 | |
dce48a84 TG |
2998 | /* |
2999 | * Either called from update_cpu_load() or from a cpu going idle | |
3000 | */ | |
3001 | static void calc_load_account_active(struct rq *this_rq) | |
3002 | { | |
3003 | long nr_active, delta; | |
08c183f3 | 3004 | |
dce48a84 TG |
3005 | nr_active = this_rq->nr_running; |
3006 | nr_active += (long) this_rq->nr_uninterruptible; | |
783609c6 | 3007 | |
dce48a84 TG |
3008 | if (nr_active != this_rq->calc_load_active) { |
3009 | delta = nr_active - this_rq->calc_load_active; | |
3010 | this_rq->calc_load_active = nr_active; | |
3011 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 | 3012 | } |
46cb4b7c SS |
3013 | } |
3014 | ||
3015 | /* | |
dd41f596 IM |
3016 | * Update rq->cpu_load[] statistics. This function is usually called every |
3017 | * scheduler tick (TICK_NSEC). | |
46cb4b7c | 3018 | */ |
dd41f596 | 3019 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3020 | { |
495eca49 | 3021 | unsigned long this_load = this_rq->load.weight; |
dd41f596 | 3022 | int i, scale; |
46cb4b7c | 3023 | |
dd41f596 | 3024 | this_rq->nr_load_updates++; |
46cb4b7c | 3025 | |
dd41f596 IM |
3026 | /* Update our load: */ |
3027 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3028 | unsigned long old_load, new_load; | |
7d1e6a9b | 3029 | |
dd41f596 | 3030 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3031 | |
dd41f596 IM |
3032 | old_load = this_rq->cpu_load[i]; |
3033 | new_load = this_load; | |
a25707f3 IM |
3034 | /* |
3035 | * Round up the averaging division if load is increasing. This | |
3036 | * prevents us from getting stuck on 9 if the load is 10, for | |
3037 | * example. | |
3038 | */ | |
3039 | if (new_load > old_load) | |
3040 | new_load += scale-1; | |
dd41f596 IM |
3041 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3042 | } | |
46cb4b7c | 3043 | |
dce48a84 TG |
3044 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { |
3045 | this_rq->calc_load_update += LOAD_FREQ; | |
3046 | calc_load_account_active(this_rq); | |
46cb4b7c | 3047 | } |
46cb4b7c SS |
3048 | } |
3049 | ||
dd41f596 | 3050 | #ifdef CONFIG_SMP |
8a0be9ef | 3051 | |
46cb4b7c | 3052 | /* |
38022906 PZ |
3053 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3054 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3055 | */ |
38022906 | 3056 | void sched_exec(void) |
46cb4b7c | 3057 | { |
38022906 | 3058 | struct task_struct *p = current; |
70b97a7f | 3059 | struct migration_req req; |
1da177e4 | 3060 | unsigned long flags; |
70b97a7f | 3061 | struct rq *rq; |
0017d735 | 3062 | int dest_cpu; |
46cb4b7c | 3063 | |
1da177e4 | 3064 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3065 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3066 | if (dest_cpu == smp_processor_id()) | |
3067 | goto unlock; | |
3068 | ||
46cb4b7c | 3069 | /* |
38022906 | 3070 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3071 | */ |
30da688e ON |
3072 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
3073 | likely(cpu_active(dest_cpu)) && | |
3074 | migrate_task(p, dest_cpu, &req)) { | |
1da177e4 LT |
3075 | /* Need to wait for migration thread (might exit: take ref). */ |
3076 | struct task_struct *mt = rq->migration_thread; | |
dd41f596 | 3077 | |
1da177e4 LT |
3078 | get_task_struct(mt); |
3079 | task_rq_unlock(rq, &flags); | |
3080 | wake_up_process(mt); | |
3081 | put_task_struct(mt); | |
3082 | wait_for_completion(&req.done); | |
dd41f596 | 3083 | |
1da177e4 LT |
3084 | return; |
3085 | } | |
0017d735 | 3086 | unlock: |
1da177e4 | 3087 | task_rq_unlock(rq, &flags); |
1da177e4 | 3088 | } |
dd41f596 | 3089 | |
1da177e4 LT |
3090 | #endif |
3091 | ||
1da177e4 LT |
3092 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3093 | ||
3094 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3095 | ||
3096 | /* | |
c5f8d995 | 3097 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3098 | * @p in case that task is currently running. |
c5f8d995 HS |
3099 | * |
3100 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3101 | */ |
c5f8d995 HS |
3102 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3103 | { | |
3104 | u64 ns = 0; | |
3105 | ||
3106 | if (task_current(rq, p)) { | |
3107 | update_rq_clock(rq); | |
3108 | ns = rq->clock - p->se.exec_start; | |
3109 | if ((s64)ns < 0) | |
3110 | ns = 0; | |
3111 | } | |
3112 | ||
3113 | return ns; | |
3114 | } | |
3115 | ||
bb34d92f | 3116 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3117 | { |
1da177e4 | 3118 | unsigned long flags; |
41b86e9c | 3119 | struct rq *rq; |
bb34d92f | 3120 | u64 ns = 0; |
48f24c4d | 3121 | |
41b86e9c | 3122 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3123 | ns = do_task_delta_exec(p, rq); |
3124 | task_rq_unlock(rq, &flags); | |
1508487e | 3125 | |
c5f8d995 HS |
3126 | return ns; |
3127 | } | |
f06febc9 | 3128 | |
c5f8d995 HS |
3129 | /* |
3130 | * Return accounted runtime for the task. | |
3131 | * In case the task is currently running, return the runtime plus current's | |
3132 | * pending runtime that have not been accounted yet. | |
3133 | */ | |
3134 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3135 | { | |
3136 | unsigned long flags; | |
3137 | struct rq *rq; | |
3138 | u64 ns = 0; | |
3139 | ||
3140 | rq = task_rq_lock(p, &flags); | |
3141 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3142 | task_rq_unlock(rq, &flags); | |
3143 | ||
3144 | return ns; | |
3145 | } | |
48f24c4d | 3146 | |
c5f8d995 HS |
3147 | /* |
3148 | * Return sum_exec_runtime for the thread group. | |
3149 | * In case the task is currently running, return the sum plus current's | |
3150 | * pending runtime that have not been accounted yet. | |
3151 | * | |
3152 | * Note that the thread group might have other running tasks as well, | |
3153 | * so the return value not includes other pending runtime that other | |
3154 | * running tasks might have. | |
3155 | */ | |
3156 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3157 | { | |
3158 | struct task_cputime totals; | |
3159 | unsigned long flags; | |
3160 | struct rq *rq; | |
3161 | u64 ns; | |
3162 | ||
3163 | rq = task_rq_lock(p, &flags); | |
3164 | thread_group_cputime(p, &totals); | |
3165 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3166 | task_rq_unlock(rq, &flags); |
48f24c4d | 3167 | |
1da177e4 LT |
3168 | return ns; |
3169 | } | |
3170 | ||
1da177e4 LT |
3171 | /* |
3172 | * Account user cpu time to a process. | |
3173 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3174 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3175 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3176 | */ |
457533a7 MS |
3177 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3178 | cputime_t cputime_scaled) | |
1da177e4 LT |
3179 | { |
3180 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3181 | cputime64_t tmp; | |
3182 | ||
457533a7 | 3183 | /* Add user time to process. */ |
1da177e4 | 3184 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3185 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3186 | account_group_user_time(p, cputime); |
1da177e4 LT |
3187 | |
3188 | /* Add user time to cpustat. */ | |
3189 | tmp = cputime_to_cputime64(cputime); | |
3190 | if (TASK_NICE(p) > 0) | |
3191 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3192 | else | |
3193 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3194 | |
3195 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3196 | /* Account for user time used */ |
3197 | acct_update_integrals(p); | |
1da177e4 LT |
3198 | } |
3199 | ||
94886b84 LV |
3200 | /* |
3201 | * Account guest cpu time to a process. | |
3202 | * @p: the process that the cpu time gets accounted to | |
3203 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3204 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3205 | */ |
457533a7 MS |
3206 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3207 | cputime_t cputime_scaled) | |
94886b84 LV |
3208 | { |
3209 | cputime64_t tmp; | |
3210 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3211 | ||
3212 | tmp = cputime_to_cputime64(cputime); | |
3213 | ||
457533a7 | 3214 | /* Add guest time to process. */ |
94886b84 | 3215 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3216 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3217 | account_group_user_time(p, cputime); |
94886b84 LV |
3218 | p->gtime = cputime_add(p->gtime, cputime); |
3219 | ||
457533a7 | 3220 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3221 | if (TASK_NICE(p) > 0) { |
3222 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3223 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3224 | } else { | |
3225 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3226 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3227 | } | |
94886b84 LV |
3228 | } |
3229 | ||
1da177e4 LT |
3230 | /* |
3231 | * Account system cpu time to a process. | |
3232 | * @p: the process that the cpu time gets accounted to | |
3233 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3234 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3235 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3236 | */ |
3237 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3238 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3239 | { |
3240 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3241 | cputime64_t tmp; |
3242 | ||
983ed7a6 | 3243 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3244 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3245 | return; |
3246 | } | |
94886b84 | 3247 | |
457533a7 | 3248 | /* Add system time to process. */ |
1da177e4 | 3249 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3250 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3251 | account_group_system_time(p, cputime); |
1da177e4 LT |
3252 | |
3253 | /* Add system time to cpustat. */ | |
3254 | tmp = cputime_to_cputime64(cputime); | |
3255 | if (hardirq_count() - hardirq_offset) | |
3256 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3257 | else if (softirq_count()) | |
3258 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3259 | else |
79741dd3 MS |
3260 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3261 | ||
ef12fefa BR |
3262 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3263 | ||
1da177e4 LT |
3264 | /* Account for system time used */ |
3265 | acct_update_integrals(p); | |
1da177e4 LT |
3266 | } |
3267 | ||
c66f08be | 3268 | /* |
1da177e4 | 3269 | * Account for involuntary wait time. |
1da177e4 | 3270 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3271 | */ |
79741dd3 | 3272 | void account_steal_time(cputime_t cputime) |
c66f08be | 3273 | { |
79741dd3 MS |
3274 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3275 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3276 | ||
3277 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3278 | } |
3279 | ||
1da177e4 | 3280 | /* |
79741dd3 MS |
3281 | * Account for idle time. |
3282 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3283 | */ |
79741dd3 | 3284 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3285 | { |
3286 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3287 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3288 | struct rq *rq = this_rq(); |
1da177e4 | 3289 | |
79741dd3 MS |
3290 | if (atomic_read(&rq->nr_iowait) > 0) |
3291 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3292 | else | |
3293 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3294 | } |
3295 | ||
79741dd3 MS |
3296 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3297 | ||
3298 | /* | |
3299 | * Account a single tick of cpu time. | |
3300 | * @p: the process that the cpu time gets accounted to | |
3301 | * @user_tick: indicates if the tick is a user or a system tick | |
3302 | */ | |
3303 | void account_process_tick(struct task_struct *p, int user_tick) | |
3304 | { | |
a42548a1 | 3305 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3306 | struct rq *rq = this_rq(); |
3307 | ||
3308 | if (user_tick) | |
a42548a1 | 3309 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3310 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3311 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3312 | one_jiffy_scaled); |
3313 | else | |
a42548a1 | 3314 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3315 | } |
3316 | ||
3317 | /* | |
3318 | * Account multiple ticks of steal time. | |
3319 | * @p: the process from which the cpu time has been stolen | |
3320 | * @ticks: number of stolen ticks | |
3321 | */ | |
3322 | void account_steal_ticks(unsigned long ticks) | |
3323 | { | |
3324 | account_steal_time(jiffies_to_cputime(ticks)); | |
3325 | } | |
3326 | ||
3327 | /* | |
3328 | * Account multiple ticks of idle time. | |
3329 | * @ticks: number of stolen ticks | |
3330 | */ | |
3331 | void account_idle_ticks(unsigned long ticks) | |
3332 | { | |
3333 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3334 | } |
3335 | ||
79741dd3 MS |
3336 | #endif |
3337 | ||
49048622 BS |
3338 | /* |
3339 | * Use precise platform statistics if available: | |
3340 | */ | |
3341 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3342 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3343 | { |
d99ca3b9 HS |
3344 | *ut = p->utime; |
3345 | *st = p->stime; | |
49048622 BS |
3346 | } |
3347 | ||
0cf55e1e | 3348 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3349 | { |
0cf55e1e HS |
3350 | struct task_cputime cputime; |
3351 | ||
3352 | thread_group_cputime(p, &cputime); | |
3353 | ||
3354 | *ut = cputime.utime; | |
3355 | *st = cputime.stime; | |
49048622 BS |
3356 | } |
3357 | #else | |
761b1d26 HS |
3358 | |
3359 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3360 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3361 | #endif |
3362 | ||
d180c5bc | 3363 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3364 | { |
d99ca3b9 | 3365 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3366 | |
3367 | /* | |
3368 | * Use CFS's precise accounting: | |
3369 | */ | |
d180c5bc | 3370 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3371 | |
3372 | if (total) { | |
d180c5bc HS |
3373 | u64 temp; |
3374 | ||
3375 | temp = (u64)(rtime * utime); | |
49048622 | 3376 | do_div(temp, total); |
d180c5bc HS |
3377 | utime = (cputime_t)temp; |
3378 | } else | |
3379 | utime = rtime; | |
49048622 | 3380 | |
d180c5bc HS |
3381 | /* |
3382 | * Compare with previous values, to keep monotonicity: | |
3383 | */ | |
761b1d26 | 3384 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3385 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3386 | |
d99ca3b9 HS |
3387 | *ut = p->prev_utime; |
3388 | *st = p->prev_stime; | |
49048622 BS |
3389 | } |
3390 | ||
0cf55e1e HS |
3391 | /* |
3392 | * Must be called with siglock held. | |
3393 | */ | |
3394 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3395 | { |
0cf55e1e HS |
3396 | struct signal_struct *sig = p->signal; |
3397 | struct task_cputime cputime; | |
3398 | cputime_t rtime, utime, total; | |
49048622 | 3399 | |
0cf55e1e | 3400 | thread_group_cputime(p, &cputime); |
49048622 | 3401 | |
0cf55e1e HS |
3402 | total = cputime_add(cputime.utime, cputime.stime); |
3403 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3404 | |
0cf55e1e HS |
3405 | if (total) { |
3406 | u64 temp; | |
49048622 | 3407 | |
0cf55e1e HS |
3408 | temp = (u64)(rtime * cputime.utime); |
3409 | do_div(temp, total); | |
3410 | utime = (cputime_t)temp; | |
3411 | } else | |
3412 | utime = rtime; | |
3413 | ||
3414 | sig->prev_utime = max(sig->prev_utime, utime); | |
3415 | sig->prev_stime = max(sig->prev_stime, | |
3416 | cputime_sub(rtime, sig->prev_utime)); | |
3417 | ||
3418 | *ut = sig->prev_utime; | |
3419 | *st = sig->prev_stime; | |
49048622 | 3420 | } |
49048622 | 3421 | #endif |
49048622 | 3422 | |
7835b98b CL |
3423 | /* |
3424 | * This function gets called by the timer code, with HZ frequency. | |
3425 | * We call it with interrupts disabled. | |
3426 | * | |
3427 | * It also gets called by the fork code, when changing the parent's | |
3428 | * timeslices. | |
3429 | */ | |
3430 | void scheduler_tick(void) | |
3431 | { | |
7835b98b CL |
3432 | int cpu = smp_processor_id(); |
3433 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3434 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3435 | |
3436 | sched_clock_tick(); | |
dd41f596 | 3437 | |
05fa785c | 3438 | raw_spin_lock(&rq->lock); |
3e51f33f | 3439 | update_rq_clock(rq); |
f1a438d8 | 3440 | update_cpu_load(rq); |
fa85ae24 | 3441 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3442 | raw_spin_unlock(&rq->lock); |
7835b98b | 3443 | |
49f47433 | 3444 | perf_event_task_tick(curr); |
e220d2dc | 3445 | |
e418e1c2 | 3446 | #ifdef CONFIG_SMP |
dd41f596 IM |
3447 | rq->idle_at_tick = idle_cpu(cpu); |
3448 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3449 | #endif |
1da177e4 LT |
3450 | } |
3451 | ||
132380a0 | 3452 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3453 | { |
3454 | if (in_lock_functions(addr)) { | |
3455 | addr = CALLER_ADDR2; | |
3456 | if (in_lock_functions(addr)) | |
3457 | addr = CALLER_ADDR3; | |
3458 | } | |
3459 | return addr; | |
3460 | } | |
1da177e4 | 3461 | |
7e49fcce SR |
3462 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3463 | defined(CONFIG_PREEMPT_TRACER)) | |
3464 | ||
43627582 | 3465 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3466 | { |
6cd8a4bb | 3467 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3468 | /* |
3469 | * Underflow? | |
3470 | */ | |
9a11b49a IM |
3471 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3472 | return; | |
6cd8a4bb | 3473 | #endif |
1da177e4 | 3474 | preempt_count() += val; |
6cd8a4bb | 3475 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3476 | /* |
3477 | * Spinlock count overflowing soon? | |
3478 | */ | |
33859f7f MOS |
3479 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3480 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3481 | #endif |
3482 | if (preempt_count() == val) | |
3483 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3484 | } |
3485 | EXPORT_SYMBOL(add_preempt_count); | |
3486 | ||
43627582 | 3487 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3488 | { |
6cd8a4bb | 3489 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3490 | /* |
3491 | * Underflow? | |
3492 | */ | |
01e3eb82 | 3493 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3494 | return; |
1da177e4 LT |
3495 | /* |
3496 | * Is the spinlock portion underflowing? | |
3497 | */ | |
9a11b49a IM |
3498 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3499 | !(preempt_count() & PREEMPT_MASK))) | |
3500 | return; | |
6cd8a4bb | 3501 | #endif |
9a11b49a | 3502 | |
6cd8a4bb SR |
3503 | if (preempt_count() == val) |
3504 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3505 | preempt_count() -= val; |
3506 | } | |
3507 | EXPORT_SYMBOL(sub_preempt_count); | |
3508 | ||
3509 | #endif | |
3510 | ||
3511 | /* | |
dd41f596 | 3512 | * Print scheduling while atomic bug: |
1da177e4 | 3513 | */ |
dd41f596 | 3514 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3515 | { |
838225b4 SS |
3516 | struct pt_regs *regs = get_irq_regs(); |
3517 | ||
3df0fc5b PZ |
3518 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3519 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3520 | |
dd41f596 | 3521 | debug_show_held_locks(prev); |
e21f5b15 | 3522 | print_modules(); |
dd41f596 IM |
3523 | if (irqs_disabled()) |
3524 | print_irqtrace_events(prev); | |
838225b4 SS |
3525 | |
3526 | if (regs) | |
3527 | show_regs(regs); | |
3528 | else | |
3529 | dump_stack(); | |
dd41f596 | 3530 | } |
1da177e4 | 3531 | |
dd41f596 IM |
3532 | /* |
3533 | * Various schedule()-time debugging checks and statistics: | |
3534 | */ | |
3535 | static inline void schedule_debug(struct task_struct *prev) | |
3536 | { | |
1da177e4 | 3537 | /* |
41a2d6cf | 3538 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3539 | * schedule() atomically, we ignore that path for now. |
3540 | * Otherwise, whine if we are scheduling when we should not be. | |
3541 | */ | |
3f33a7ce | 3542 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3543 | __schedule_bug(prev); |
3544 | ||
1da177e4 LT |
3545 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3546 | ||
2d72376b | 3547 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3548 | #ifdef CONFIG_SCHEDSTATS |
3549 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3550 | schedstat_inc(this_rq(), bkl_count); |
3551 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3552 | } |
3553 | #endif | |
dd41f596 IM |
3554 | } |
3555 | ||
6cecd084 | 3556 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3557 | { |
a64692a3 MG |
3558 | if (prev->se.on_rq) |
3559 | update_rq_clock(rq); | |
3560 | rq->skip_clock_update = 0; | |
6cecd084 | 3561 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3562 | } |
3563 | ||
dd41f596 IM |
3564 | /* |
3565 | * Pick up the highest-prio task: | |
3566 | */ | |
3567 | static inline struct task_struct * | |
b67802ea | 3568 | pick_next_task(struct rq *rq) |
dd41f596 | 3569 | { |
5522d5d5 | 3570 | const struct sched_class *class; |
dd41f596 | 3571 | struct task_struct *p; |
1da177e4 LT |
3572 | |
3573 | /* | |
dd41f596 IM |
3574 | * Optimization: we know that if all tasks are in |
3575 | * the fair class we can call that function directly: | |
1da177e4 | 3576 | */ |
dd41f596 | 3577 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3578 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3579 | if (likely(p)) |
3580 | return p; | |
1da177e4 LT |
3581 | } |
3582 | ||
dd41f596 IM |
3583 | class = sched_class_highest; |
3584 | for ( ; ; ) { | |
fb8d4724 | 3585 | p = class->pick_next_task(rq); |
dd41f596 IM |
3586 | if (p) |
3587 | return p; | |
3588 | /* | |
3589 | * Will never be NULL as the idle class always | |
3590 | * returns a non-NULL p: | |
3591 | */ | |
3592 | class = class->next; | |
3593 | } | |
3594 | } | |
1da177e4 | 3595 | |
dd41f596 IM |
3596 | /* |
3597 | * schedule() is the main scheduler function. | |
3598 | */ | |
ff743345 | 3599 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3600 | { |
3601 | struct task_struct *prev, *next; | |
67ca7bde | 3602 | unsigned long *switch_count; |
dd41f596 | 3603 | struct rq *rq; |
31656519 | 3604 | int cpu; |
dd41f596 | 3605 | |
ff743345 PZ |
3606 | need_resched: |
3607 | preempt_disable(); | |
dd41f596 IM |
3608 | cpu = smp_processor_id(); |
3609 | rq = cpu_rq(cpu); | |
d6714c22 | 3610 | rcu_sched_qs(cpu); |
dd41f596 IM |
3611 | prev = rq->curr; |
3612 | switch_count = &prev->nivcsw; | |
3613 | ||
3614 | release_kernel_lock(prev); | |
3615 | need_resched_nonpreemptible: | |
3616 | ||
3617 | schedule_debug(prev); | |
1da177e4 | 3618 | |
31656519 | 3619 | if (sched_feat(HRTICK)) |
f333fdc9 | 3620 | hrtick_clear(rq); |
8f4d37ec | 3621 | |
05fa785c | 3622 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3623 | clear_tsk_need_resched(prev); |
1da177e4 | 3624 | |
1da177e4 | 3625 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 3626 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 3627 | prev->state = TASK_RUNNING; |
16882c1e | 3628 | else |
371fd7e7 | 3629 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
dd41f596 | 3630 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3631 | } |
3632 | ||
3f029d3c | 3633 | pre_schedule(rq, prev); |
f65eda4f | 3634 | |
dd41f596 | 3635 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3636 | idle_balance(cpu, rq); |
1da177e4 | 3637 | |
df1c99d4 | 3638 | put_prev_task(rq, prev); |
b67802ea | 3639 | next = pick_next_task(rq); |
1da177e4 | 3640 | |
1da177e4 | 3641 | if (likely(prev != next)) { |
673a90a1 | 3642 | sched_info_switch(prev, next); |
49f47433 | 3643 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3644 | |
1da177e4 LT |
3645 | rq->nr_switches++; |
3646 | rq->curr = next; | |
3647 | ++*switch_count; | |
3648 | ||
dd41f596 | 3649 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3650 | /* |
3651 | * the context switch might have flipped the stack from under | |
3652 | * us, hence refresh the local variables. | |
3653 | */ | |
3654 | cpu = smp_processor_id(); | |
3655 | rq = cpu_rq(cpu); | |
1da177e4 | 3656 | } else |
05fa785c | 3657 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3658 | |
3f029d3c | 3659 | post_schedule(rq); |
1da177e4 | 3660 | |
6d558c3a YZ |
3661 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3662 | prev = rq->curr; | |
3663 | switch_count = &prev->nivcsw; | |
1da177e4 | 3664 | goto need_resched_nonpreemptible; |
6d558c3a | 3665 | } |
8f4d37ec | 3666 | |
1da177e4 | 3667 | preempt_enable_no_resched(); |
ff743345 | 3668 | if (need_resched()) |
1da177e4 LT |
3669 | goto need_resched; |
3670 | } | |
1da177e4 LT |
3671 | EXPORT_SYMBOL(schedule); |
3672 | ||
c08f7829 | 3673 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3674 | /* |
3675 | * Look out! "owner" is an entirely speculative pointer | |
3676 | * access and not reliable. | |
3677 | */ | |
3678 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3679 | { | |
3680 | unsigned int cpu; | |
3681 | struct rq *rq; | |
3682 | ||
3683 | if (!sched_feat(OWNER_SPIN)) | |
3684 | return 0; | |
3685 | ||
3686 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3687 | /* | |
3688 | * Need to access the cpu field knowing that | |
3689 | * DEBUG_PAGEALLOC could have unmapped it if | |
3690 | * the mutex owner just released it and exited. | |
3691 | */ | |
3692 | if (probe_kernel_address(&owner->cpu, cpu)) | |
3693 | goto out; | |
3694 | #else | |
3695 | cpu = owner->cpu; | |
3696 | #endif | |
3697 | ||
3698 | /* | |
3699 | * Even if the access succeeded (likely case), | |
3700 | * the cpu field may no longer be valid. | |
3701 | */ | |
3702 | if (cpu >= nr_cpumask_bits) | |
3703 | goto out; | |
3704 | ||
3705 | /* | |
3706 | * We need to validate that we can do a | |
3707 | * get_cpu() and that we have the percpu area. | |
3708 | */ | |
3709 | if (!cpu_online(cpu)) | |
3710 | goto out; | |
3711 | ||
3712 | rq = cpu_rq(cpu); | |
3713 | ||
3714 | for (;;) { | |
3715 | /* | |
3716 | * Owner changed, break to re-assess state. | |
3717 | */ | |
3718 | if (lock->owner != owner) | |
3719 | break; | |
3720 | ||
3721 | /* | |
3722 | * Is that owner really running on that cpu? | |
3723 | */ | |
3724 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3725 | return 0; | |
3726 | ||
3727 | cpu_relax(); | |
3728 | } | |
3729 | out: | |
3730 | return 1; | |
3731 | } | |
3732 | #endif | |
3733 | ||
1da177e4 LT |
3734 | #ifdef CONFIG_PREEMPT |
3735 | /* | |
2ed6e34f | 3736 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3737 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3738 | * occur there and call schedule directly. |
3739 | */ | |
3740 | asmlinkage void __sched preempt_schedule(void) | |
3741 | { | |
3742 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3743 | |
1da177e4 LT |
3744 | /* |
3745 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3746 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3747 | */ |
beed33a8 | 3748 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3749 | return; |
3750 | ||
3a5c359a AK |
3751 | do { |
3752 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3753 | schedule(); |
3a5c359a | 3754 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3755 | |
3a5c359a AK |
3756 | /* |
3757 | * Check again in case we missed a preemption opportunity | |
3758 | * between schedule and now. | |
3759 | */ | |
3760 | barrier(); | |
5ed0cec0 | 3761 | } while (need_resched()); |
1da177e4 | 3762 | } |
1da177e4 LT |
3763 | EXPORT_SYMBOL(preempt_schedule); |
3764 | ||
3765 | /* | |
2ed6e34f | 3766 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3767 | * off of irq context. |
3768 | * Note, that this is called and return with irqs disabled. This will | |
3769 | * protect us against recursive calling from irq. | |
3770 | */ | |
3771 | asmlinkage void __sched preempt_schedule_irq(void) | |
3772 | { | |
3773 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3774 | |
2ed6e34f | 3775 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3776 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3777 | ||
3a5c359a AK |
3778 | do { |
3779 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3780 | local_irq_enable(); |
3781 | schedule(); | |
3782 | local_irq_disable(); | |
3a5c359a | 3783 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3784 | |
3a5c359a AK |
3785 | /* |
3786 | * Check again in case we missed a preemption opportunity | |
3787 | * between schedule and now. | |
3788 | */ | |
3789 | barrier(); | |
5ed0cec0 | 3790 | } while (need_resched()); |
1da177e4 LT |
3791 | } |
3792 | ||
3793 | #endif /* CONFIG_PREEMPT */ | |
3794 | ||
63859d4f | 3795 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3796 | void *key) |
1da177e4 | 3797 | { |
63859d4f | 3798 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3799 | } |
1da177e4 LT |
3800 | EXPORT_SYMBOL(default_wake_function); |
3801 | ||
3802 | /* | |
41a2d6cf IM |
3803 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3804 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3805 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3806 | * | |
3807 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3808 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3809 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3810 | */ | |
78ddb08f | 3811 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3812 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3813 | { |
2e45874c | 3814 | wait_queue_t *curr, *next; |
1da177e4 | 3815 | |
2e45874c | 3816 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3817 | unsigned flags = curr->flags; |
3818 | ||
63859d4f | 3819 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3820 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3821 | break; |
3822 | } | |
3823 | } | |
3824 | ||
3825 | /** | |
3826 | * __wake_up - wake up threads blocked on a waitqueue. | |
3827 | * @q: the waitqueue | |
3828 | * @mode: which threads | |
3829 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3830 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3831 | * |
3832 | * It may be assumed that this function implies a write memory barrier before | |
3833 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3834 | */ |
7ad5b3a5 | 3835 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3836 | int nr_exclusive, void *key) |
1da177e4 LT |
3837 | { |
3838 | unsigned long flags; | |
3839 | ||
3840 | spin_lock_irqsave(&q->lock, flags); | |
3841 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3842 | spin_unlock_irqrestore(&q->lock, flags); | |
3843 | } | |
1da177e4 LT |
3844 | EXPORT_SYMBOL(__wake_up); |
3845 | ||
3846 | /* | |
3847 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3848 | */ | |
7ad5b3a5 | 3849 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3850 | { |
3851 | __wake_up_common(q, mode, 1, 0, NULL); | |
3852 | } | |
3853 | ||
4ede816a DL |
3854 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3855 | { | |
3856 | __wake_up_common(q, mode, 1, 0, key); | |
3857 | } | |
3858 | ||
1da177e4 | 3859 | /** |
4ede816a | 3860 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3861 | * @q: the waitqueue |
3862 | * @mode: which threads | |
3863 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3864 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3865 | * |
3866 | * The sync wakeup differs that the waker knows that it will schedule | |
3867 | * away soon, so while the target thread will be woken up, it will not | |
3868 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3869 | * with each other. This can prevent needless bouncing between CPUs. | |
3870 | * | |
3871 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3872 | * |
3873 | * It may be assumed that this function implies a write memory barrier before | |
3874 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3875 | */ |
4ede816a DL |
3876 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3877 | int nr_exclusive, void *key) | |
1da177e4 LT |
3878 | { |
3879 | unsigned long flags; | |
7d478721 | 3880 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3881 | |
3882 | if (unlikely(!q)) | |
3883 | return; | |
3884 | ||
3885 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3886 | wake_flags = 0; |
1da177e4 LT |
3887 | |
3888 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3889 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3890 | spin_unlock_irqrestore(&q->lock, flags); |
3891 | } | |
4ede816a DL |
3892 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3893 | ||
3894 | /* | |
3895 | * __wake_up_sync - see __wake_up_sync_key() | |
3896 | */ | |
3897 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3898 | { | |
3899 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3900 | } | |
1da177e4 LT |
3901 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3902 | ||
65eb3dc6 KD |
3903 | /** |
3904 | * complete: - signals a single thread waiting on this completion | |
3905 | * @x: holds the state of this particular completion | |
3906 | * | |
3907 | * This will wake up a single thread waiting on this completion. Threads will be | |
3908 | * awakened in the same order in which they were queued. | |
3909 | * | |
3910 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3911 | * |
3912 | * It may be assumed that this function implies a write memory barrier before | |
3913 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3914 | */ |
b15136e9 | 3915 | void complete(struct completion *x) |
1da177e4 LT |
3916 | { |
3917 | unsigned long flags; | |
3918 | ||
3919 | spin_lock_irqsave(&x->wait.lock, flags); | |
3920 | x->done++; | |
d9514f6c | 3921 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3922 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3923 | } | |
3924 | EXPORT_SYMBOL(complete); | |
3925 | ||
65eb3dc6 KD |
3926 | /** |
3927 | * complete_all: - signals all threads waiting on this completion | |
3928 | * @x: holds the state of this particular completion | |
3929 | * | |
3930 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3931 | * |
3932 | * It may be assumed that this function implies a write memory barrier before | |
3933 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3934 | */ |
b15136e9 | 3935 | void complete_all(struct completion *x) |
1da177e4 LT |
3936 | { |
3937 | unsigned long flags; | |
3938 | ||
3939 | spin_lock_irqsave(&x->wait.lock, flags); | |
3940 | x->done += UINT_MAX/2; | |
d9514f6c | 3941 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3942 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3943 | } | |
3944 | EXPORT_SYMBOL(complete_all); | |
3945 | ||
8cbbe86d AK |
3946 | static inline long __sched |
3947 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3948 | { |
1da177e4 LT |
3949 | if (!x->done) { |
3950 | DECLARE_WAITQUEUE(wait, current); | |
3951 | ||
3952 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3953 | __add_wait_queue_tail(&x->wait, &wait); | |
3954 | do { | |
94d3d824 | 3955 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3956 | timeout = -ERESTARTSYS; |
3957 | break; | |
8cbbe86d AK |
3958 | } |
3959 | __set_current_state(state); | |
1da177e4 LT |
3960 | spin_unlock_irq(&x->wait.lock); |
3961 | timeout = schedule_timeout(timeout); | |
3962 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3963 | } while (!x->done && timeout); |
1da177e4 | 3964 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3965 | if (!x->done) |
3966 | return timeout; | |
1da177e4 LT |
3967 | } |
3968 | x->done--; | |
ea71a546 | 3969 | return timeout ?: 1; |
1da177e4 | 3970 | } |
1da177e4 | 3971 | |
8cbbe86d AK |
3972 | static long __sched |
3973 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3974 | { |
1da177e4 LT |
3975 | might_sleep(); |
3976 | ||
3977 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3978 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3979 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3980 | return timeout; |
3981 | } | |
1da177e4 | 3982 | |
65eb3dc6 KD |
3983 | /** |
3984 | * wait_for_completion: - waits for completion of a task | |
3985 | * @x: holds the state of this particular completion | |
3986 | * | |
3987 | * This waits to be signaled for completion of a specific task. It is NOT | |
3988 | * interruptible and there is no timeout. | |
3989 | * | |
3990 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3991 | * and interrupt capability. Also see complete(). | |
3992 | */ | |
b15136e9 | 3993 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3994 | { |
3995 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3996 | } |
8cbbe86d | 3997 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3998 | |
65eb3dc6 KD |
3999 | /** |
4000 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4001 | * @x: holds the state of this particular completion | |
4002 | * @timeout: timeout value in jiffies | |
4003 | * | |
4004 | * This waits for either a completion of a specific task to be signaled or for a | |
4005 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4006 | * interruptible. | |
4007 | */ | |
b15136e9 | 4008 | unsigned long __sched |
8cbbe86d | 4009 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4010 | { |
8cbbe86d | 4011 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4012 | } |
8cbbe86d | 4013 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4014 | |
65eb3dc6 KD |
4015 | /** |
4016 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4017 | * @x: holds the state of this particular completion | |
4018 | * | |
4019 | * This waits for completion of a specific task to be signaled. It is | |
4020 | * interruptible. | |
4021 | */ | |
8cbbe86d | 4022 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4023 | { |
51e97990 AK |
4024 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4025 | if (t == -ERESTARTSYS) | |
4026 | return t; | |
4027 | return 0; | |
0fec171c | 4028 | } |
8cbbe86d | 4029 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4030 | |
65eb3dc6 KD |
4031 | /** |
4032 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4033 | * @x: holds the state of this particular completion | |
4034 | * @timeout: timeout value in jiffies | |
4035 | * | |
4036 | * This waits for either a completion of a specific task to be signaled or for a | |
4037 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4038 | */ | |
b15136e9 | 4039 | unsigned long __sched |
8cbbe86d AK |
4040 | wait_for_completion_interruptible_timeout(struct completion *x, |
4041 | unsigned long timeout) | |
0fec171c | 4042 | { |
8cbbe86d | 4043 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4044 | } |
8cbbe86d | 4045 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4046 | |
65eb3dc6 KD |
4047 | /** |
4048 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4049 | * @x: holds the state of this particular completion | |
4050 | * | |
4051 | * This waits to be signaled for completion of a specific task. It can be | |
4052 | * interrupted by a kill signal. | |
4053 | */ | |
009e577e MW |
4054 | int __sched wait_for_completion_killable(struct completion *x) |
4055 | { | |
4056 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4057 | if (t == -ERESTARTSYS) | |
4058 | return t; | |
4059 | return 0; | |
4060 | } | |
4061 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4062 | ||
be4de352 DC |
4063 | /** |
4064 | * try_wait_for_completion - try to decrement a completion without blocking | |
4065 | * @x: completion structure | |
4066 | * | |
4067 | * Returns: 0 if a decrement cannot be done without blocking | |
4068 | * 1 if a decrement succeeded. | |
4069 | * | |
4070 | * If a completion is being used as a counting completion, | |
4071 | * attempt to decrement the counter without blocking. This | |
4072 | * enables us to avoid waiting if the resource the completion | |
4073 | * is protecting is not available. | |
4074 | */ | |
4075 | bool try_wait_for_completion(struct completion *x) | |
4076 | { | |
7539a3b3 | 4077 | unsigned long flags; |
be4de352 DC |
4078 | int ret = 1; |
4079 | ||
7539a3b3 | 4080 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4081 | if (!x->done) |
4082 | ret = 0; | |
4083 | else | |
4084 | x->done--; | |
7539a3b3 | 4085 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4086 | return ret; |
4087 | } | |
4088 | EXPORT_SYMBOL(try_wait_for_completion); | |
4089 | ||
4090 | /** | |
4091 | * completion_done - Test to see if a completion has any waiters | |
4092 | * @x: completion structure | |
4093 | * | |
4094 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4095 | * 1 if there are no waiters. | |
4096 | * | |
4097 | */ | |
4098 | bool completion_done(struct completion *x) | |
4099 | { | |
7539a3b3 | 4100 | unsigned long flags; |
be4de352 DC |
4101 | int ret = 1; |
4102 | ||
7539a3b3 | 4103 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4104 | if (!x->done) |
4105 | ret = 0; | |
7539a3b3 | 4106 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4107 | return ret; |
4108 | } | |
4109 | EXPORT_SYMBOL(completion_done); | |
4110 | ||
8cbbe86d AK |
4111 | static long __sched |
4112 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4113 | { |
0fec171c IM |
4114 | unsigned long flags; |
4115 | wait_queue_t wait; | |
4116 | ||
4117 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4118 | |
8cbbe86d | 4119 | __set_current_state(state); |
1da177e4 | 4120 | |
8cbbe86d AK |
4121 | spin_lock_irqsave(&q->lock, flags); |
4122 | __add_wait_queue(q, &wait); | |
4123 | spin_unlock(&q->lock); | |
4124 | timeout = schedule_timeout(timeout); | |
4125 | spin_lock_irq(&q->lock); | |
4126 | __remove_wait_queue(q, &wait); | |
4127 | spin_unlock_irqrestore(&q->lock, flags); | |
4128 | ||
4129 | return timeout; | |
4130 | } | |
4131 | ||
4132 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4133 | { | |
4134 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4135 | } |
1da177e4 LT |
4136 | EXPORT_SYMBOL(interruptible_sleep_on); |
4137 | ||
0fec171c | 4138 | long __sched |
95cdf3b7 | 4139 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4140 | { |
8cbbe86d | 4141 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4142 | } |
1da177e4 LT |
4143 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4144 | ||
0fec171c | 4145 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4146 | { |
8cbbe86d | 4147 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4148 | } |
1da177e4 LT |
4149 | EXPORT_SYMBOL(sleep_on); |
4150 | ||
0fec171c | 4151 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4152 | { |
8cbbe86d | 4153 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4154 | } |
1da177e4 LT |
4155 | EXPORT_SYMBOL(sleep_on_timeout); |
4156 | ||
b29739f9 IM |
4157 | #ifdef CONFIG_RT_MUTEXES |
4158 | ||
4159 | /* | |
4160 | * rt_mutex_setprio - set the current priority of a task | |
4161 | * @p: task | |
4162 | * @prio: prio value (kernel-internal form) | |
4163 | * | |
4164 | * This function changes the 'effective' priority of a task. It does | |
4165 | * not touch ->normal_prio like __setscheduler(). | |
4166 | * | |
4167 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4168 | */ | |
36c8b586 | 4169 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4170 | { |
4171 | unsigned long flags; | |
83b699ed | 4172 | int oldprio, on_rq, running; |
70b97a7f | 4173 | struct rq *rq; |
83ab0aa0 | 4174 | const struct sched_class *prev_class; |
b29739f9 IM |
4175 | |
4176 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4177 | ||
4178 | rq = task_rq_lock(p, &flags); | |
4179 | ||
d5f9f942 | 4180 | oldprio = p->prio; |
83ab0aa0 | 4181 | prev_class = p->sched_class; |
dd41f596 | 4182 | on_rq = p->se.on_rq; |
051a1d1a | 4183 | running = task_current(rq, p); |
0e1f3483 | 4184 | if (on_rq) |
69be72c1 | 4185 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4186 | if (running) |
4187 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4188 | |
4189 | if (rt_prio(prio)) | |
4190 | p->sched_class = &rt_sched_class; | |
4191 | else | |
4192 | p->sched_class = &fair_sched_class; | |
4193 | ||
b29739f9 IM |
4194 | p->prio = prio; |
4195 | ||
0e1f3483 HS |
4196 | if (running) |
4197 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4198 | if (on_rq) { |
371fd7e7 | 4199 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4200 | |
4201 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4202 | } |
4203 | task_rq_unlock(rq, &flags); | |
4204 | } | |
4205 | ||
4206 | #endif | |
4207 | ||
36c8b586 | 4208 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4209 | { |
dd41f596 | 4210 | int old_prio, delta, on_rq; |
1da177e4 | 4211 | unsigned long flags; |
70b97a7f | 4212 | struct rq *rq; |
1da177e4 LT |
4213 | |
4214 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4215 | return; | |
4216 | /* | |
4217 | * We have to be careful, if called from sys_setpriority(), | |
4218 | * the task might be in the middle of scheduling on another CPU. | |
4219 | */ | |
4220 | rq = task_rq_lock(p, &flags); | |
4221 | /* | |
4222 | * The RT priorities are set via sched_setscheduler(), but we still | |
4223 | * allow the 'normal' nice value to be set - but as expected | |
4224 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4225 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4226 | */ |
e05606d3 | 4227 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4228 | p->static_prio = NICE_TO_PRIO(nice); |
4229 | goto out_unlock; | |
4230 | } | |
dd41f596 | 4231 | on_rq = p->se.on_rq; |
c09595f6 | 4232 | if (on_rq) |
69be72c1 | 4233 | dequeue_task(rq, p, 0); |
1da177e4 | 4234 | |
1da177e4 | 4235 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4236 | set_load_weight(p); |
b29739f9 IM |
4237 | old_prio = p->prio; |
4238 | p->prio = effective_prio(p); | |
4239 | delta = p->prio - old_prio; | |
1da177e4 | 4240 | |
dd41f596 | 4241 | if (on_rq) { |
371fd7e7 | 4242 | enqueue_task(rq, p, 0); |
1da177e4 | 4243 | /* |
d5f9f942 AM |
4244 | * If the task increased its priority or is running and |
4245 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4246 | */ |
d5f9f942 | 4247 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4248 | resched_task(rq->curr); |
4249 | } | |
4250 | out_unlock: | |
4251 | task_rq_unlock(rq, &flags); | |
4252 | } | |
1da177e4 LT |
4253 | EXPORT_SYMBOL(set_user_nice); |
4254 | ||
e43379f1 MM |
4255 | /* |
4256 | * can_nice - check if a task can reduce its nice value | |
4257 | * @p: task | |
4258 | * @nice: nice value | |
4259 | */ | |
36c8b586 | 4260 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4261 | { |
024f4747 MM |
4262 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4263 | int nice_rlim = 20 - nice; | |
48f24c4d | 4264 | |
78d7d407 | 4265 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4266 | capable(CAP_SYS_NICE)); |
4267 | } | |
4268 | ||
1da177e4 LT |
4269 | #ifdef __ARCH_WANT_SYS_NICE |
4270 | ||
4271 | /* | |
4272 | * sys_nice - change the priority of the current process. | |
4273 | * @increment: priority increment | |
4274 | * | |
4275 | * sys_setpriority is a more generic, but much slower function that | |
4276 | * does similar things. | |
4277 | */ | |
5add95d4 | 4278 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4279 | { |
48f24c4d | 4280 | long nice, retval; |
1da177e4 LT |
4281 | |
4282 | /* | |
4283 | * Setpriority might change our priority at the same moment. | |
4284 | * We don't have to worry. Conceptually one call occurs first | |
4285 | * and we have a single winner. | |
4286 | */ | |
e43379f1 MM |
4287 | if (increment < -40) |
4288 | increment = -40; | |
1da177e4 LT |
4289 | if (increment > 40) |
4290 | increment = 40; | |
4291 | ||
2b8f836f | 4292 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4293 | if (nice < -20) |
4294 | nice = -20; | |
4295 | if (nice > 19) | |
4296 | nice = 19; | |
4297 | ||
e43379f1 MM |
4298 | if (increment < 0 && !can_nice(current, nice)) |
4299 | return -EPERM; | |
4300 | ||
1da177e4 LT |
4301 | retval = security_task_setnice(current, nice); |
4302 | if (retval) | |
4303 | return retval; | |
4304 | ||
4305 | set_user_nice(current, nice); | |
4306 | return 0; | |
4307 | } | |
4308 | ||
4309 | #endif | |
4310 | ||
4311 | /** | |
4312 | * task_prio - return the priority value of a given task. | |
4313 | * @p: the task in question. | |
4314 | * | |
4315 | * This is the priority value as seen by users in /proc. | |
4316 | * RT tasks are offset by -200. Normal tasks are centered | |
4317 | * around 0, value goes from -16 to +15. | |
4318 | */ | |
36c8b586 | 4319 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4320 | { |
4321 | return p->prio - MAX_RT_PRIO; | |
4322 | } | |
4323 | ||
4324 | /** | |
4325 | * task_nice - return the nice value of a given task. | |
4326 | * @p: the task in question. | |
4327 | */ | |
36c8b586 | 4328 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4329 | { |
4330 | return TASK_NICE(p); | |
4331 | } | |
150d8bed | 4332 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4333 | |
4334 | /** | |
4335 | * idle_cpu - is a given cpu idle currently? | |
4336 | * @cpu: the processor in question. | |
4337 | */ | |
4338 | int idle_cpu(int cpu) | |
4339 | { | |
4340 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4341 | } | |
4342 | ||
1da177e4 LT |
4343 | /** |
4344 | * idle_task - return the idle task for a given cpu. | |
4345 | * @cpu: the processor in question. | |
4346 | */ | |
36c8b586 | 4347 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4348 | { |
4349 | return cpu_rq(cpu)->idle; | |
4350 | } | |
4351 | ||
4352 | /** | |
4353 | * find_process_by_pid - find a process with a matching PID value. | |
4354 | * @pid: the pid in question. | |
4355 | */ | |
a9957449 | 4356 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4357 | { |
228ebcbe | 4358 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4359 | } |
4360 | ||
4361 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4362 | static void |
4363 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4364 | { |
dd41f596 | 4365 | BUG_ON(p->se.on_rq); |
48f24c4d | 4366 | |
1da177e4 LT |
4367 | p->policy = policy; |
4368 | p->rt_priority = prio; | |
b29739f9 IM |
4369 | p->normal_prio = normal_prio(p); |
4370 | /* we are holding p->pi_lock already */ | |
4371 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4372 | if (rt_prio(p->prio)) |
4373 | p->sched_class = &rt_sched_class; | |
4374 | else | |
4375 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4376 | set_load_weight(p); |
1da177e4 LT |
4377 | } |
4378 | ||
c69e8d9c DH |
4379 | /* |
4380 | * check the target process has a UID that matches the current process's | |
4381 | */ | |
4382 | static bool check_same_owner(struct task_struct *p) | |
4383 | { | |
4384 | const struct cred *cred = current_cred(), *pcred; | |
4385 | bool match; | |
4386 | ||
4387 | rcu_read_lock(); | |
4388 | pcred = __task_cred(p); | |
4389 | match = (cred->euid == pcred->euid || | |
4390 | cred->euid == pcred->uid); | |
4391 | rcu_read_unlock(); | |
4392 | return match; | |
4393 | } | |
4394 | ||
961ccddd RR |
4395 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4396 | struct sched_param *param, bool user) | |
1da177e4 | 4397 | { |
83b699ed | 4398 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4399 | unsigned long flags; |
83ab0aa0 | 4400 | const struct sched_class *prev_class; |
70b97a7f | 4401 | struct rq *rq; |
ca94c442 | 4402 | int reset_on_fork; |
1da177e4 | 4403 | |
66e5393a SR |
4404 | /* may grab non-irq protected spin_locks */ |
4405 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4406 | recheck: |
4407 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4408 | if (policy < 0) { |
4409 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4410 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4411 | } else { |
4412 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4413 | policy &= ~SCHED_RESET_ON_FORK; | |
4414 | ||
4415 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4416 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4417 | policy != SCHED_IDLE) | |
4418 | return -EINVAL; | |
4419 | } | |
4420 | ||
1da177e4 LT |
4421 | /* |
4422 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4423 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4424 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4425 | */ |
4426 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4427 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4428 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4429 | return -EINVAL; |
e05606d3 | 4430 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4431 | return -EINVAL; |
4432 | ||
37e4ab3f OC |
4433 | /* |
4434 | * Allow unprivileged RT tasks to decrease priority: | |
4435 | */ | |
961ccddd | 4436 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4437 | if (rt_policy(policy)) { |
8dc3e909 | 4438 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4439 | |
4440 | if (!lock_task_sighand(p, &flags)) | |
4441 | return -ESRCH; | |
78d7d407 | 4442 | rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); |
8dc3e909 ON |
4443 | unlock_task_sighand(p, &flags); |
4444 | ||
4445 | /* can't set/change the rt policy */ | |
4446 | if (policy != p->policy && !rlim_rtprio) | |
4447 | return -EPERM; | |
4448 | ||
4449 | /* can't increase priority */ | |
4450 | if (param->sched_priority > p->rt_priority && | |
4451 | param->sched_priority > rlim_rtprio) | |
4452 | return -EPERM; | |
4453 | } | |
dd41f596 IM |
4454 | /* |
4455 | * Like positive nice levels, dont allow tasks to | |
4456 | * move out of SCHED_IDLE either: | |
4457 | */ | |
4458 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4459 | return -EPERM; | |
5fe1d75f | 4460 | |
37e4ab3f | 4461 | /* can't change other user's priorities */ |
c69e8d9c | 4462 | if (!check_same_owner(p)) |
37e4ab3f | 4463 | return -EPERM; |
ca94c442 LP |
4464 | |
4465 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4466 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4467 | return -EPERM; | |
37e4ab3f | 4468 | } |
1da177e4 | 4469 | |
725aad24 | 4470 | if (user) { |
b68aa230 | 4471 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
4472 | /* |
4473 | * Do not allow realtime tasks into groups that have no runtime | |
4474 | * assigned. | |
4475 | */ | |
9a7e0b18 PZ |
4476 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
4477 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 4478 | return -EPERM; |
b68aa230 PZ |
4479 | #endif |
4480 | ||
725aad24 JF |
4481 | retval = security_task_setscheduler(p, policy, param); |
4482 | if (retval) | |
4483 | return retval; | |
4484 | } | |
4485 | ||
b29739f9 IM |
4486 | /* |
4487 | * make sure no PI-waiters arrive (or leave) while we are | |
4488 | * changing the priority of the task: | |
4489 | */ | |
1d615482 | 4490 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4491 | /* |
4492 | * To be able to change p->policy safely, the apropriate | |
4493 | * runqueue lock must be held. | |
4494 | */ | |
b29739f9 | 4495 | rq = __task_rq_lock(p); |
1da177e4 LT |
4496 | /* recheck policy now with rq lock held */ |
4497 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4498 | policy = oldpolicy = -1; | |
b29739f9 | 4499 | __task_rq_unlock(rq); |
1d615482 | 4500 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4501 | goto recheck; |
4502 | } | |
dd41f596 | 4503 | on_rq = p->se.on_rq; |
051a1d1a | 4504 | running = task_current(rq, p); |
0e1f3483 | 4505 | if (on_rq) |
2e1cb74a | 4506 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4507 | if (running) |
4508 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4509 | |
ca94c442 LP |
4510 | p->sched_reset_on_fork = reset_on_fork; |
4511 | ||
1da177e4 | 4512 | oldprio = p->prio; |
83ab0aa0 | 4513 | prev_class = p->sched_class; |
dd41f596 | 4514 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4515 | |
0e1f3483 HS |
4516 | if (running) |
4517 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4518 | if (on_rq) { |
4519 | activate_task(rq, p, 0); | |
cb469845 SR |
4520 | |
4521 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4522 | } |
b29739f9 | 4523 | __task_rq_unlock(rq); |
1d615482 | 4524 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4525 | |
95e02ca9 TG |
4526 | rt_mutex_adjust_pi(p); |
4527 | ||
1da177e4 LT |
4528 | return 0; |
4529 | } | |
961ccddd RR |
4530 | |
4531 | /** | |
4532 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4533 | * @p: the task in question. | |
4534 | * @policy: new policy. | |
4535 | * @param: structure containing the new RT priority. | |
4536 | * | |
4537 | * NOTE that the task may be already dead. | |
4538 | */ | |
4539 | int sched_setscheduler(struct task_struct *p, int policy, | |
4540 | struct sched_param *param) | |
4541 | { | |
4542 | return __sched_setscheduler(p, policy, param, true); | |
4543 | } | |
1da177e4 LT |
4544 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4545 | ||
961ccddd RR |
4546 | /** |
4547 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4548 | * @p: the task in question. | |
4549 | * @policy: new policy. | |
4550 | * @param: structure containing the new RT priority. | |
4551 | * | |
4552 | * Just like sched_setscheduler, only don't bother checking if the | |
4553 | * current context has permission. For example, this is needed in | |
4554 | * stop_machine(): we create temporary high priority worker threads, | |
4555 | * but our caller might not have that capability. | |
4556 | */ | |
4557 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4558 | struct sched_param *param) | |
4559 | { | |
4560 | return __sched_setscheduler(p, policy, param, false); | |
4561 | } | |
4562 | ||
95cdf3b7 IM |
4563 | static int |
4564 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4565 | { |
1da177e4 LT |
4566 | struct sched_param lparam; |
4567 | struct task_struct *p; | |
36c8b586 | 4568 | int retval; |
1da177e4 LT |
4569 | |
4570 | if (!param || pid < 0) | |
4571 | return -EINVAL; | |
4572 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4573 | return -EFAULT; | |
5fe1d75f ON |
4574 | |
4575 | rcu_read_lock(); | |
4576 | retval = -ESRCH; | |
1da177e4 | 4577 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4578 | if (p != NULL) |
4579 | retval = sched_setscheduler(p, policy, &lparam); | |
4580 | rcu_read_unlock(); | |
36c8b586 | 4581 | |
1da177e4 LT |
4582 | return retval; |
4583 | } | |
4584 | ||
4585 | /** | |
4586 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4587 | * @pid: the pid in question. | |
4588 | * @policy: new policy. | |
4589 | * @param: structure containing the new RT priority. | |
4590 | */ | |
5add95d4 HC |
4591 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4592 | struct sched_param __user *, param) | |
1da177e4 | 4593 | { |
c21761f1 JB |
4594 | /* negative values for policy are not valid */ |
4595 | if (policy < 0) | |
4596 | return -EINVAL; | |
4597 | ||
1da177e4 LT |
4598 | return do_sched_setscheduler(pid, policy, param); |
4599 | } | |
4600 | ||
4601 | /** | |
4602 | * sys_sched_setparam - set/change the RT priority of a thread | |
4603 | * @pid: the pid in question. | |
4604 | * @param: structure containing the new RT priority. | |
4605 | */ | |
5add95d4 | 4606 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4607 | { |
4608 | return do_sched_setscheduler(pid, -1, param); | |
4609 | } | |
4610 | ||
4611 | /** | |
4612 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4613 | * @pid: the pid in question. | |
4614 | */ | |
5add95d4 | 4615 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4616 | { |
36c8b586 | 4617 | struct task_struct *p; |
3a5c359a | 4618 | int retval; |
1da177e4 LT |
4619 | |
4620 | if (pid < 0) | |
3a5c359a | 4621 | return -EINVAL; |
1da177e4 LT |
4622 | |
4623 | retval = -ESRCH; | |
5fe85be0 | 4624 | rcu_read_lock(); |
1da177e4 LT |
4625 | p = find_process_by_pid(pid); |
4626 | if (p) { | |
4627 | retval = security_task_getscheduler(p); | |
4628 | if (!retval) | |
ca94c442 LP |
4629 | retval = p->policy |
4630 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4631 | } |
5fe85be0 | 4632 | rcu_read_unlock(); |
1da177e4 LT |
4633 | return retval; |
4634 | } | |
4635 | ||
4636 | /** | |
ca94c442 | 4637 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4638 | * @pid: the pid in question. |
4639 | * @param: structure containing the RT priority. | |
4640 | */ | |
5add95d4 | 4641 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4642 | { |
4643 | struct sched_param lp; | |
36c8b586 | 4644 | struct task_struct *p; |
3a5c359a | 4645 | int retval; |
1da177e4 LT |
4646 | |
4647 | if (!param || pid < 0) | |
3a5c359a | 4648 | return -EINVAL; |
1da177e4 | 4649 | |
5fe85be0 | 4650 | rcu_read_lock(); |
1da177e4 LT |
4651 | p = find_process_by_pid(pid); |
4652 | retval = -ESRCH; | |
4653 | if (!p) | |
4654 | goto out_unlock; | |
4655 | ||
4656 | retval = security_task_getscheduler(p); | |
4657 | if (retval) | |
4658 | goto out_unlock; | |
4659 | ||
4660 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4661 | rcu_read_unlock(); |
1da177e4 LT |
4662 | |
4663 | /* | |
4664 | * This one might sleep, we cannot do it with a spinlock held ... | |
4665 | */ | |
4666 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4667 | ||
1da177e4 LT |
4668 | return retval; |
4669 | ||
4670 | out_unlock: | |
5fe85be0 | 4671 | rcu_read_unlock(); |
1da177e4 LT |
4672 | return retval; |
4673 | } | |
4674 | ||
96f874e2 | 4675 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4676 | { |
5a16f3d3 | 4677 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4678 | struct task_struct *p; |
4679 | int retval; | |
1da177e4 | 4680 | |
95402b38 | 4681 | get_online_cpus(); |
23f5d142 | 4682 | rcu_read_lock(); |
1da177e4 LT |
4683 | |
4684 | p = find_process_by_pid(pid); | |
4685 | if (!p) { | |
23f5d142 | 4686 | rcu_read_unlock(); |
95402b38 | 4687 | put_online_cpus(); |
1da177e4 LT |
4688 | return -ESRCH; |
4689 | } | |
4690 | ||
23f5d142 | 4691 | /* Prevent p going away */ |
1da177e4 | 4692 | get_task_struct(p); |
23f5d142 | 4693 | rcu_read_unlock(); |
1da177e4 | 4694 | |
5a16f3d3 RR |
4695 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4696 | retval = -ENOMEM; | |
4697 | goto out_put_task; | |
4698 | } | |
4699 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4700 | retval = -ENOMEM; | |
4701 | goto out_free_cpus_allowed; | |
4702 | } | |
1da177e4 | 4703 | retval = -EPERM; |
c69e8d9c | 4704 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4705 | goto out_unlock; |
4706 | ||
e7834f8f DQ |
4707 | retval = security_task_setscheduler(p, 0, NULL); |
4708 | if (retval) | |
4709 | goto out_unlock; | |
4710 | ||
5a16f3d3 RR |
4711 | cpuset_cpus_allowed(p, cpus_allowed); |
4712 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4713 | again: |
5a16f3d3 | 4714 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4715 | |
8707d8b8 | 4716 | if (!retval) { |
5a16f3d3 RR |
4717 | cpuset_cpus_allowed(p, cpus_allowed); |
4718 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4719 | /* |
4720 | * We must have raced with a concurrent cpuset | |
4721 | * update. Just reset the cpus_allowed to the | |
4722 | * cpuset's cpus_allowed | |
4723 | */ | |
5a16f3d3 | 4724 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4725 | goto again; |
4726 | } | |
4727 | } | |
1da177e4 | 4728 | out_unlock: |
5a16f3d3 RR |
4729 | free_cpumask_var(new_mask); |
4730 | out_free_cpus_allowed: | |
4731 | free_cpumask_var(cpus_allowed); | |
4732 | out_put_task: | |
1da177e4 | 4733 | put_task_struct(p); |
95402b38 | 4734 | put_online_cpus(); |
1da177e4 LT |
4735 | return retval; |
4736 | } | |
4737 | ||
4738 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4739 | struct cpumask *new_mask) |
1da177e4 | 4740 | { |
96f874e2 RR |
4741 | if (len < cpumask_size()) |
4742 | cpumask_clear(new_mask); | |
4743 | else if (len > cpumask_size()) | |
4744 | len = cpumask_size(); | |
4745 | ||
1da177e4 LT |
4746 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4747 | } | |
4748 | ||
4749 | /** | |
4750 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4751 | * @pid: pid of the process | |
4752 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4753 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4754 | */ | |
5add95d4 HC |
4755 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4756 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4757 | { |
5a16f3d3 | 4758 | cpumask_var_t new_mask; |
1da177e4 LT |
4759 | int retval; |
4760 | ||
5a16f3d3 RR |
4761 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4762 | return -ENOMEM; | |
1da177e4 | 4763 | |
5a16f3d3 RR |
4764 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4765 | if (retval == 0) | |
4766 | retval = sched_setaffinity(pid, new_mask); | |
4767 | free_cpumask_var(new_mask); | |
4768 | return retval; | |
1da177e4 LT |
4769 | } |
4770 | ||
96f874e2 | 4771 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4772 | { |
36c8b586 | 4773 | struct task_struct *p; |
31605683 TG |
4774 | unsigned long flags; |
4775 | struct rq *rq; | |
1da177e4 | 4776 | int retval; |
1da177e4 | 4777 | |
95402b38 | 4778 | get_online_cpus(); |
23f5d142 | 4779 | rcu_read_lock(); |
1da177e4 LT |
4780 | |
4781 | retval = -ESRCH; | |
4782 | p = find_process_by_pid(pid); | |
4783 | if (!p) | |
4784 | goto out_unlock; | |
4785 | ||
e7834f8f DQ |
4786 | retval = security_task_getscheduler(p); |
4787 | if (retval) | |
4788 | goto out_unlock; | |
4789 | ||
31605683 | 4790 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4791 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4792 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4793 | |
4794 | out_unlock: | |
23f5d142 | 4795 | rcu_read_unlock(); |
95402b38 | 4796 | put_online_cpus(); |
1da177e4 | 4797 | |
9531b62f | 4798 | return retval; |
1da177e4 LT |
4799 | } |
4800 | ||
4801 | /** | |
4802 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4803 | * @pid: pid of the process | |
4804 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4805 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4806 | */ | |
5add95d4 HC |
4807 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4808 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4809 | { |
4810 | int ret; | |
f17c8607 | 4811 | cpumask_var_t mask; |
1da177e4 | 4812 | |
84fba5ec | 4813 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4814 | return -EINVAL; |
4815 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4816 | return -EINVAL; |
4817 | ||
f17c8607 RR |
4818 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4819 | return -ENOMEM; | |
1da177e4 | 4820 | |
f17c8607 RR |
4821 | ret = sched_getaffinity(pid, mask); |
4822 | if (ret == 0) { | |
8bc037fb | 4823 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4824 | |
4825 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4826 | ret = -EFAULT; |
4827 | else | |
cd3d8031 | 4828 | ret = retlen; |
f17c8607 RR |
4829 | } |
4830 | free_cpumask_var(mask); | |
1da177e4 | 4831 | |
f17c8607 | 4832 | return ret; |
1da177e4 LT |
4833 | } |
4834 | ||
4835 | /** | |
4836 | * sys_sched_yield - yield the current processor to other threads. | |
4837 | * | |
dd41f596 IM |
4838 | * This function yields the current CPU to other tasks. If there are no |
4839 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4840 | */ |
5add95d4 | 4841 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4842 | { |
70b97a7f | 4843 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4844 | |
2d72376b | 4845 | schedstat_inc(rq, yld_count); |
4530d7ab | 4846 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4847 | |
4848 | /* | |
4849 | * Since we are going to call schedule() anyway, there's | |
4850 | * no need to preempt or enable interrupts: | |
4851 | */ | |
4852 | __release(rq->lock); | |
8a25d5de | 4853 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4854 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4855 | preempt_enable_no_resched(); |
4856 | ||
4857 | schedule(); | |
4858 | ||
4859 | return 0; | |
4860 | } | |
4861 | ||
d86ee480 PZ |
4862 | static inline int should_resched(void) |
4863 | { | |
4864 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4865 | } | |
4866 | ||
e7b38404 | 4867 | static void __cond_resched(void) |
1da177e4 | 4868 | { |
e7aaaa69 FW |
4869 | add_preempt_count(PREEMPT_ACTIVE); |
4870 | schedule(); | |
4871 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
4872 | } |
4873 | ||
02b67cc3 | 4874 | int __sched _cond_resched(void) |
1da177e4 | 4875 | { |
d86ee480 | 4876 | if (should_resched()) { |
1da177e4 LT |
4877 | __cond_resched(); |
4878 | return 1; | |
4879 | } | |
4880 | return 0; | |
4881 | } | |
02b67cc3 | 4882 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4883 | |
4884 | /* | |
613afbf8 | 4885 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4886 | * call schedule, and on return reacquire the lock. |
4887 | * | |
41a2d6cf | 4888 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4889 | * operations here to prevent schedule() from being called twice (once via |
4890 | * spin_unlock(), once by hand). | |
4891 | */ | |
613afbf8 | 4892 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4893 | { |
d86ee480 | 4894 | int resched = should_resched(); |
6df3cecb JK |
4895 | int ret = 0; |
4896 | ||
f607c668 PZ |
4897 | lockdep_assert_held(lock); |
4898 | ||
95c354fe | 4899 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4900 | spin_unlock(lock); |
d86ee480 | 4901 | if (resched) |
95c354fe NP |
4902 | __cond_resched(); |
4903 | else | |
4904 | cpu_relax(); | |
6df3cecb | 4905 | ret = 1; |
1da177e4 | 4906 | spin_lock(lock); |
1da177e4 | 4907 | } |
6df3cecb | 4908 | return ret; |
1da177e4 | 4909 | } |
613afbf8 | 4910 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4911 | |
613afbf8 | 4912 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4913 | { |
4914 | BUG_ON(!in_softirq()); | |
4915 | ||
d86ee480 | 4916 | if (should_resched()) { |
98d82567 | 4917 | local_bh_enable(); |
1da177e4 LT |
4918 | __cond_resched(); |
4919 | local_bh_disable(); | |
4920 | return 1; | |
4921 | } | |
4922 | return 0; | |
4923 | } | |
613afbf8 | 4924 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4925 | |
1da177e4 LT |
4926 | /** |
4927 | * yield - yield the current processor to other threads. | |
4928 | * | |
72fd4a35 | 4929 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4930 | * thread runnable and calls sys_sched_yield(). |
4931 | */ | |
4932 | void __sched yield(void) | |
4933 | { | |
4934 | set_current_state(TASK_RUNNING); | |
4935 | sys_sched_yield(); | |
4936 | } | |
1da177e4 LT |
4937 | EXPORT_SYMBOL(yield); |
4938 | ||
4939 | /* | |
41a2d6cf | 4940 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4941 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4942 | */ |
4943 | void __sched io_schedule(void) | |
4944 | { | |
54d35f29 | 4945 | struct rq *rq = raw_rq(); |
1da177e4 | 4946 | |
0ff92245 | 4947 | delayacct_blkio_start(); |
1da177e4 | 4948 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4949 | current->in_iowait = 1; |
1da177e4 | 4950 | schedule(); |
8f0dfc34 | 4951 | current->in_iowait = 0; |
1da177e4 | 4952 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4953 | delayacct_blkio_end(); |
1da177e4 | 4954 | } |
1da177e4 LT |
4955 | EXPORT_SYMBOL(io_schedule); |
4956 | ||
4957 | long __sched io_schedule_timeout(long timeout) | |
4958 | { | |
54d35f29 | 4959 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4960 | long ret; |
4961 | ||
0ff92245 | 4962 | delayacct_blkio_start(); |
1da177e4 | 4963 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4964 | current->in_iowait = 1; |
1da177e4 | 4965 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4966 | current->in_iowait = 0; |
1da177e4 | 4967 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4968 | delayacct_blkio_end(); |
1da177e4 LT |
4969 | return ret; |
4970 | } | |
4971 | ||
4972 | /** | |
4973 | * sys_sched_get_priority_max - return maximum RT priority. | |
4974 | * @policy: scheduling class. | |
4975 | * | |
4976 | * this syscall returns the maximum rt_priority that can be used | |
4977 | * by a given scheduling class. | |
4978 | */ | |
5add95d4 | 4979 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4980 | { |
4981 | int ret = -EINVAL; | |
4982 | ||
4983 | switch (policy) { | |
4984 | case SCHED_FIFO: | |
4985 | case SCHED_RR: | |
4986 | ret = MAX_USER_RT_PRIO-1; | |
4987 | break; | |
4988 | case SCHED_NORMAL: | |
b0a9499c | 4989 | case SCHED_BATCH: |
dd41f596 | 4990 | case SCHED_IDLE: |
1da177e4 LT |
4991 | ret = 0; |
4992 | break; | |
4993 | } | |
4994 | return ret; | |
4995 | } | |
4996 | ||
4997 | /** | |
4998 | * sys_sched_get_priority_min - return minimum RT priority. | |
4999 | * @policy: scheduling class. | |
5000 | * | |
5001 | * this syscall returns the minimum rt_priority that can be used | |
5002 | * by a given scheduling class. | |
5003 | */ | |
5add95d4 | 5004 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5005 | { |
5006 | int ret = -EINVAL; | |
5007 | ||
5008 | switch (policy) { | |
5009 | case SCHED_FIFO: | |
5010 | case SCHED_RR: | |
5011 | ret = 1; | |
5012 | break; | |
5013 | case SCHED_NORMAL: | |
b0a9499c | 5014 | case SCHED_BATCH: |
dd41f596 | 5015 | case SCHED_IDLE: |
1da177e4 LT |
5016 | ret = 0; |
5017 | } | |
5018 | return ret; | |
5019 | } | |
5020 | ||
5021 | /** | |
5022 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5023 | * @pid: pid of the process. | |
5024 | * @interval: userspace pointer to the timeslice value. | |
5025 | * | |
5026 | * this syscall writes the default timeslice value of a given process | |
5027 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5028 | */ | |
17da2bd9 | 5029 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5030 | struct timespec __user *, interval) |
1da177e4 | 5031 | { |
36c8b586 | 5032 | struct task_struct *p; |
a4ec24b4 | 5033 | unsigned int time_slice; |
dba091b9 TG |
5034 | unsigned long flags; |
5035 | struct rq *rq; | |
3a5c359a | 5036 | int retval; |
1da177e4 | 5037 | struct timespec t; |
1da177e4 LT |
5038 | |
5039 | if (pid < 0) | |
3a5c359a | 5040 | return -EINVAL; |
1da177e4 LT |
5041 | |
5042 | retval = -ESRCH; | |
1a551ae7 | 5043 | rcu_read_lock(); |
1da177e4 LT |
5044 | p = find_process_by_pid(pid); |
5045 | if (!p) | |
5046 | goto out_unlock; | |
5047 | ||
5048 | retval = security_task_getscheduler(p); | |
5049 | if (retval) | |
5050 | goto out_unlock; | |
5051 | ||
dba091b9 TG |
5052 | rq = task_rq_lock(p, &flags); |
5053 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5054 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5055 | |
1a551ae7 | 5056 | rcu_read_unlock(); |
a4ec24b4 | 5057 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5058 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5059 | return retval; |
3a5c359a | 5060 | |
1da177e4 | 5061 | out_unlock: |
1a551ae7 | 5062 | rcu_read_unlock(); |
1da177e4 LT |
5063 | return retval; |
5064 | } | |
5065 | ||
7c731e0a | 5066 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5067 | |
82a1fcb9 | 5068 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5069 | { |
1da177e4 | 5070 | unsigned long free = 0; |
36c8b586 | 5071 | unsigned state; |
1da177e4 | 5072 | |
1da177e4 | 5073 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5074 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5075 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5076 | #if BITS_PER_LONG == 32 |
1da177e4 | 5077 | if (state == TASK_RUNNING) |
3df0fc5b | 5078 | printk(KERN_CONT " running "); |
1da177e4 | 5079 | else |
3df0fc5b | 5080 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5081 | #else |
5082 | if (state == TASK_RUNNING) | |
3df0fc5b | 5083 | printk(KERN_CONT " running task "); |
1da177e4 | 5084 | else |
3df0fc5b | 5085 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5086 | #endif |
5087 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5088 | free = stack_not_used(p); |
1da177e4 | 5089 | #endif |
3df0fc5b | 5090 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5091 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5092 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5093 | |
5fb5e6de | 5094 | show_stack(p, NULL); |
1da177e4 LT |
5095 | } |
5096 | ||
e59e2ae2 | 5097 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5098 | { |
36c8b586 | 5099 | struct task_struct *g, *p; |
1da177e4 | 5100 | |
4bd77321 | 5101 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5102 | printk(KERN_INFO |
5103 | " task PC stack pid father\n"); | |
1da177e4 | 5104 | #else |
3df0fc5b PZ |
5105 | printk(KERN_INFO |
5106 | " task PC stack pid father\n"); | |
1da177e4 LT |
5107 | #endif |
5108 | read_lock(&tasklist_lock); | |
5109 | do_each_thread(g, p) { | |
5110 | /* | |
5111 | * reset the NMI-timeout, listing all files on a slow | |
5112 | * console might take alot of time: | |
5113 | */ | |
5114 | touch_nmi_watchdog(); | |
39bc89fd | 5115 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5116 | sched_show_task(p); |
1da177e4 LT |
5117 | } while_each_thread(g, p); |
5118 | ||
04c9167f JF |
5119 | touch_all_softlockup_watchdogs(); |
5120 | ||
dd41f596 IM |
5121 | #ifdef CONFIG_SCHED_DEBUG |
5122 | sysrq_sched_debug_show(); | |
5123 | #endif | |
1da177e4 | 5124 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5125 | /* |
5126 | * Only show locks if all tasks are dumped: | |
5127 | */ | |
93335a21 | 5128 | if (!state_filter) |
e59e2ae2 | 5129 | debug_show_all_locks(); |
1da177e4 LT |
5130 | } |
5131 | ||
1df21055 IM |
5132 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5133 | { | |
dd41f596 | 5134 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5135 | } |
5136 | ||
f340c0d1 IM |
5137 | /** |
5138 | * init_idle - set up an idle thread for a given CPU | |
5139 | * @idle: task in question | |
5140 | * @cpu: cpu the idle task belongs to | |
5141 | * | |
5142 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5143 | * flag, to make booting more robust. | |
5144 | */ | |
5c1e1767 | 5145 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5146 | { |
70b97a7f | 5147 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5148 | unsigned long flags; |
5149 | ||
05fa785c | 5150 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5151 | |
dd41f596 | 5152 | __sched_fork(idle); |
06b83b5f | 5153 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5154 | idle->se.exec_start = sched_clock(); |
5155 | ||
96f874e2 | 5156 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5157 | __set_task_cpu(idle, cpu); |
1da177e4 | 5158 | |
1da177e4 | 5159 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5160 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5161 | idle->oncpu = 1; | |
5162 | #endif | |
05fa785c | 5163 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5164 | |
5165 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5166 | #if defined(CONFIG_PREEMPT) |
5167 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5168 | #else | |
a1261f54 | 5169 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5170 | #endif |
dd41f596 IM |
5171 | /* |
5172 | * The idle tasks have their own, simple scheduling class: | |
5173 | */ | |
5174 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5175 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5176 | } |
5177 | ||
5178 | /* | |
5179 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5180 | * indicates which cpus entered this state. This is used | |
5181 | * in the rcu update to wait only for active cpus. For system | |
5182 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5183 | * always be CPU_BITS_NONE. |
1da177e4 | 5184 | */ |
6a7b3dc3 | 5185 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5186 | |
19978ca6 IM |
5187 | /* |
5188 | * Increase the granularity value when there are more CPUs, | |
5189 | * because with more CPUs the 'effective latency' as visible | |
5190 | * to users decreases. But the relationship is not linear, | |
5191 | * so pick a second-best guess by going with the log2 of the | |
5192 | * number of CPUs. | |
5193 | * | |
5194 | * This idea comes from the SD scheduler of Con Kolivas: | |
5195 | */ | |
acb4a848 | 5196 | static int get_update_sysctl_factor(void) |
19978ca6 | 5197 | { |
4ca3ef71 | 5198 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5199 | unsigned int factor; |
5200 | ||
5201 | switch (sysctl_sched_tunable_scaling) { | |
5202 | case SCHED_TUNABLESCALING_NONE: | |
5203 | factor = 1; | |
5204 | break; | |
5205 | case SCHED_TUNABLESCALING_LINEAR: | |
5206 | factor = cpus; | |
5207 | break; | |
5208 | case SCHED_TUNABLESCALING_LOG: | |
5209 | default: | |
5210 | factor = 1 + ilog2(cpus); | |
5211 | break; | |
5212 | } | |
19978ca6 | 5213 | |
acb4a848 CE |
5214 | return factor; |
5215 | } | |
19978ca6 | 5216 | |
acb4a848 CE |
5217 | static void update_sysctl(void) |
5218 | { | |
5219 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5220 | |
0bcdcf28 CE |
5221 | #define SET_SYSCTL(name) \ |
5222 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5223 | SET_SYSCTL(sched_min_granularity); | |
5224 | SET_SYSCTL(sched_latency); | |
5225 | SET_SYSCTL(sched_wakeup_granularity); | |
5226 | SET_SYSCTL(sched_shares_ratelimit); | |
5227 | #undef SET_SYSCTL | |
5228 | } | |
55cd5340 | 5229 | |
0bcdcf28 CE |
5230 | static inline void sched_init_granularity(void) |
5231 | { | |
5232 | update_sysctl(); | |
19978ca6 IM |
5233 | } |
5234 | ||
1da177e4 LT |
5235 | #ifdef CONFIG_SMP |
5236 | /* | |
5237 | * This is how migration works: | |
5238 | * | |
70b97a7f | 5239 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5240 | * runqueue and wake up that CPU's migration thread. |
5241 | * 2) we down() the locked semaphore => thread blocks. | |
5242 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5243 | * thread off the CPU) | |
5244 | * 4) it gets the migration request and checks whether the migrated | |
5245 | * task is still in the wrong runqueue. | |
5246 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5247 | * it and puts it into the right queue. | |
5248 | * 6) migration thread up()s the semaphore. | |
5249 | * 7) we wake up and the migration is done. | |
5250 | */ | |
5251 | ||
5252 | /* | |
5253 | * Change a given task's CPU affinity. Migrate the thread to a | |
5254 | * proper CPU and schedule it away if the CPU it's executing on | |
5255 | * is removed from the allowed bitmask. | |
5256 | * | |
5257 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5258 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5259 | * call is not atomic; no spinlocks may be held. |
5260 | */ | |
96f874e2 | 5261 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 5262 | { |
70b97a7f | 5263 | struct migration_req req; |
1da177e4 | 5264 | unsigned long flags; |
70b97a7f | 5265 | struct rq *rq; |
48f24c4d | 5266 | int ret = 0; |
1da177e4 | 5267 | |
65cc8e48 PZ |
5268 | /* |
5269 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5270 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5271 | */ | |
5272 | again: | |
5273 | while (task_is_waking(p)) | |
5274 | cpu_relax(); | |
1da177e4 | 5275 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5276 | if (task_is_waking(p)) { |
5277 | task_rq_unlock(rq, &flags); | |
5278 | goto again; | |
5279 | } | |
e2912009 | 5280 | |
6ad4c188 | 5281 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5282 | ret = -EINVAL; |
5283 | goto out; | |
5284 | } | |
5285 | ||
9985b0ba | 5286 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5287 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5288 | ret = -EINVAL; |
5289 | goto out; | |
5290 | } | |
5291 | ||
73fe6aae | 5292 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5293 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5294 | else { |
96f874e2 RR |
5295 | cpumask_copy(&p->cpus_allowed, new_mask); |
5296 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5297 | } |
5298 | ||
1da177e4 | 5299 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5300 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5301 | goto out; |
5302 | ||
6ad4c188 | 5303 | if (migrate_task(p, cpumask_any_and(cpu_active_mask, new_mask), &req)) { |
1da177e4 | 5304 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
5305 | struct task_struct *mt = rq->migration_thread; |
5306 | ||
5307 | get_task_struct(mt); | |
1da177e4 | 5308 | task_rq_unlock(rq, &flags); |
47a70985 | 5309 | wake_up_process(mt); |
693525e3 | 5310 | put_task_struct(mt); |
1da177e4 LT |
5311 | wait_for_completion(&req.done); |
5312 | tlb_migrate_finish(p->mm); | |
5313 | return 0; | |
5314 | } | |
5315 | out: | |
5316 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5317 | |
1da177e4 LT |
5318 | return ret; |
5319 | } | |
cd8ba7cd | 5320 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5321 | |
5322 | /* | |
41a2d6cf | 5323 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5324 | * this because either it can't run here any more (set_cpus_allowed() |
5325 | * away from this CPU, or CPU going down), or because we're | |
5326 | * attempting to rebalance this task on exec (sched_exec). | |
5327 | * | |
5328 | * So we race with normal scheduler movements, but that's OK, as long | |
5329 | * as the task is no longer on this CPU. | |
efc30814 KK |
5330 | * |
5331 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5332 | */ |
efc30814 | 5333 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5334 | { |
70b97a7f | 5335 | struct rq *rq_dest, *rq_src; |
e2912009 | 5336 | int ret = 0; |
1da177e4 | 5337 | |
e761b772 | 5338 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5339 | return ret; |
1da177e4 LT |
5340 | |
5341 | rq_src = cpu_rq(src_cpu); | |
5342 | rq_dest = cpu_rq(dest_cpu); | |
5343 | ||
5344 | double_rq_lock(rq_src, rq_dest); | |
5345 | /* Already moved. */ | |
5346 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5347 | goto done; |
1da177e4 | 5348 | /* Affinity changed (again). */ |
96f874e2 | 5349 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5350 | goto fail; |
1da177e4 | 5351 | |
e2912009 PZ |
5352 | /* |
5353 | * If we're not on a rq, the next wake-up will ensure we're | |
5354 | * placed properly. | |
5355 | */ | |
5356 | if (p->se.on_rq) { | |
2e1cb74a | 5357 | deactivate_task(rq_src, p, 0); |
e2912009 | 5358 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5359 | activate_task(rq_dest, p, 0); |
15afe09b | 5360 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5361 | } |
b1e38734 | 5362 | done: |
efc30814 | 5363 | ret = 1; |
b1e38734 | 5364 | fail: |
1da177e4 | 5365 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5366 | return ret; |
1da177e4 LT |
5367 | } |
5368 | ||
03b042bf PM |
5369 | #define RCU_MIGRATION_IDLE 0 |
5370 | #define RCU_MIGRATION_NEED_QS 1 | |
5371 | #define RCU_MIGRATION_GOT_QS 2 | |
5372 | #define RCU_MIGRATION_MUST_SYNC 3 | |
5373 | ||
1da177e4 LT |
5374 | /* |
5375 | * migration_thread - this is a highprio system thread that performs | |
5376 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5377 | * another runqueue. | |
5378 | */ | |
95cdf3b7 | 5379 | static int migration_thread(void *data) |
1da177e4 | 5380 | { |
03b042bf | 5381 | int badcpu; |
1da177e4 | 5382 | int cpu = (long)data; |
70b97a7f | 5383 | struct rq *rq; |
1da177e4 LT |
5384 | |
5385 | rq = cpu_rq(cpu); | |
5386 | BUG_ON(rq->migration_thread != current); | |
5387 | ||
5388 | set_current_state(TASK_INTERRUPTIBLE); | |
5389 | while (!kthread_should_stop()) { | |
70b97a7f | 5390 | struct migration_req *req; |
1da177e4 | 5391 | struct list_head *head; |
1da177e4 | 5392 | |
05fa785c | 5393 | raw_spin_lock_irq(&rq->lock); |
1da177e4 LT |
5394 | |
5395 | if (cpu_is_offline(cpu)) { | |
05fa785c | 5396 | raw_spin_unlock_irq(&rq->lock); |
371cbb38 | 5397 | break; |
1da177e4 LT |
5398 | } |
5399 | ||
5400 | if (rq->active_balance) { | |
5401 | active_load_balance(rq, cpu); | |
5402 | rq->active_balance = 0; | |
5403 | } | |
5404 | ||
5405 | head = &rq->migration_queue; | |
5406 | ||
5407 | if (list_empty(head)) { | |
05fa785c | 5408 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5409 | schedule(); |
5410 | set_current_state(TASK_INTERRUPTIBLE); | |
5411 | continue; | |
5412 | } | |
70b97a7f | 5413 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5414 | list_del_init(head->next); |
5415 | ||
03b042bf | 5416 | if (req->task != NULL) { |
05fa785c | 5417 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5418 | __migrate_task(req->task, cpu, req->dest_cpu); |
5419 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
5420 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
05fa785c | 5421 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5422 | } else { |
5423 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
05fa785c | 5424 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5425 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); |
5426 | } | |
674311d5 | 5427 | local_irq_enable(); |
1da177e4 LT |
5428 | |
5429 | complete(&req->done); | |
5430 | } | |
5431 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 5432 | |
1da177e4 LT |
5433 | return 0; |
5434 | } | |
5435 | ||
5436 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 | 5437 | /* |
3a4fa0a2 | 5438 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5439 | */ |
6a1bdc1b | 5440 | void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5441 | { |
1445c08d ON |
5442 | struct rq *rq = cpu_rq(dead_cpu); |
5443 | int needs_cpu, uninitialized_var(dest_cpu); | |
5444 | unsigned long flags; | |
c1804d54 | 5445 | |
1445c08d ON |
5446 | local_irq_save(flags); |
5447 | ||
5448 | raw_spin_lock(&rq->lock); | |
5449 | needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING); | |
5450 | if (needs_cpu) | |
5451 | dest_cpu = select_fallback_rq(dead_cpu, p); | |
5452 | raw_spin_unlock(&rq->lock); | |
c1804d54 ON |
5453 | /* |
5454 | * It can only fail if we race with set_cpus_allowed(), | |
5455 | * in the racer should migrate the task anyway. | |
5456 | */ | |
1445c08d | 5457 | if (needs_cpu) |
c1804d54 | 5458 | __migrate_task(p, dead_cpu, dest_cpu); |
1445c08d | 5459 | local_irq_restore(flags); |
1da177e4 LT |
5460 | } |
5461 | ||
5462 | /* | |
5463 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5464 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5465 | * for performance reasons the counter is not stricly tracking tasks to | |
5466 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5467 | * to keep the global sum constant after CPU-down: | |
5468 | */ | |
70b97a7f | 5469 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5470 | { |
6ad4c188 | 5471 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5472 | unsigned long flags; |
5473 | ||
5474 | local_irq_save(flags); | |
5475 | double_rq_lock(rq_src, rq_dest); | |
5476 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5477 | rq_src->nr_uninterruptible = 0; | |
5478 | double_rq_unlock(rq_src, rq_dest); | |
5479 | local_irq_restore(flags); | |
5480 | } | |
5481 | ||
5482 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5483 | static void migrate_live_tasks(int src_cpu) | |
5484 | { | |
48f24c4d | 5485 | struct task_struct *p, *t; |
1da177e4 | 5486 | |
f7b4cddc | 5487 | read_lock(&tasklist_lock); |
1da177e4 | 5488 | |
48f24c4d IM |
5489 | do_each_thread(t, p) { |
5490 | if (p == current) | |
1da177e4 LT |
5491 | continue; |
5492 | ||
48f24c4d IM |
5493 | if (task_cpu(p) == src_cpu) |
5494 | move_task_off_dead_cpu(src_cpu, p); | |
5495 | } while_each_thread(t, p); | |
1da177e4 | 5496 | |
f7b4cddc | 5497 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5498 | } |
5499 | ||
dd41f596 IM |
5500 | /* |
5501 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5502 | * It does so by boosting its priority to highest possible. |
5503 | * Used by CPU offline code. | |
1da177e4 LT |
5504 | */ |
5505 | void sched_idle_next(void) | |
5506 | { | |
48f24c4d | 5507 | int this_cpu = smp_processor_id(); |
70b97a7f | 5508 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5509 | struct task_struct *p = rq->idle; |
5510 | unsigned long flags; | |
5511 | ||
5512 | /* cpu has to be offline */ | |
48f24c4d | 5513 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5514 | |
48f24c4d IM |
5515 | /* |
5516 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5517 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5518 | */ |
05fa785c | 5519 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5520 | |
dd41f596 | 5521 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5522 | |
94bc9a7b | 5523 | activate_task(rq, p, 0); |
1da177e4 | 5524 | |
05fa785c | 5525 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5526 | } |
5527 | ||
48f24c4d IM |
5528 | /* |
5529 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5530 | * offline. |
5531 | */ | |
5532 | void idle_task_exit(void) | |
5533 | { | |
5534 | struct mm_struct *mm = current->active_mm; | |
5535 | ||
5536 | BUG_ON(cpu_online(smp_processor_id())); | |
5537 | ||
5538 | if (mm != &init_mm) | |
5539 | switch_mm(mm, &init_mm, current); | |
5540 | mmdrop(mm); | |
5541 | } | |
5542 | ||
054b9108 | 5543 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5544 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5545 | { |
70b97a7f | 5546 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5547 | |
5548 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5549 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5550 | |
5551 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5552 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5553 | |
48f24c4d | 5554 | get_task_struct(p); |
1da177e4 LT |
5555 | |
5556 | /* | |
5557 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5558 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5559 | * fine. |
5560 | */ | |
05fa785c | 5561 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5562 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5563 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5564 | |
48f24c4d | 5565 | put_task_struct(p); |
1da177e4 LT |
5566 | } |
5567 | ||
5568 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5569 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5570 | { | |
70b97a7f | 5571 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5572 | struct task_struct *next; |
48f24c4d | 5573 | |
dd41f596 IM |
5574 | for ( ; ; ) { |
5575 | if (!rq->nr_running) | |
5576 | break; | |
b67802ea | 5577 | next = pick_next_task(rq); |
dd41f596 IM |
5578 | if (!next) |
5579 | break; | |
79c53799 | 5580 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5581 | migrate_dead(dead_cpu, next); |
e692ab53 | 5582 | |
1da177e4 LT |
5583 | } |
5584 | } | |
dce48a84 TG |
5585 | |
5586 | /* | |
5587 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5588 | */ | |
5589 | static void calc_global_load_remove(struct rq *rq) | |
5590 | { | |
5591 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5592 | rq->calc_load_active = 0; |
dce48a84 | 5593 | } |
1da177e4 LT |
5594 | #endif /* CONFIG_HOTPLUG_CPU */ |
5595 | ||
e692ab53 NP |
5596 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5597 | ||
5598 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5599 | { |
5600 | .procname = "sched_domain", | |
c57baf1e | 5601 | .mode = 0555, |
e0361851 | 5602 | }, |
56992309 | 5603 | {} |
e692ab53 NP |
5604 | }; |
5605 | ||
5606 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5607 | { |
5608 | .procname = "kernel", | |
c57baf1e | 5609 | .mode = 0555, |
e0361851 AD |
5610 | .child = sd_ctl_dir, |
5611 | }, | |
56992309 | 5612 | {} |
e692ab53 NP |
5613 | }; |
5614 | ||
5615 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5616 | { | |
5617 | struct ctl_table *entry = | |
5cf9f062 | 5618 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5619 | |
e692ab53 NP |
5620 | return entry; |
5621 | } | |
5622 | ||
6382bc90 MM |
5623 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5624 | { | |
cd790076 | 5625 | struct ctl_table *entry; |
6382bc90 | 5626 | |
cd790076 MM |
5627 | /* |
5628 | * In the intermediate directories, both the child directory and | |
5629 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5630 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5631 | * static strings and all have proc handlers. |
5632 | */ | |
5633 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5634 | if (entry->child) |
5635 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5636 | if (entry->proc_handler == NULL) |
5637 | kfree(entry->procname); | |
5638 | } | |
6382bc90 MM |
5639 | |
5640 | kfree(*tablep); | |
5641 | *tablep = NULL; | |
5642 | } | |
5643 | ||
e692ab53 | 5644 | static void |
e0361851 | 5645 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5646 | const char *procname, void *data, int maxlen, |
5647 | mode_t mode, proc_handler *proc_handler) | |
5648 | { | |
e692ab53 NP |
5649 | entry->procname = procname; |
5650 | entry->data = data; | |
5651 | entry->maxlen = maxlen; | |
5652 | entry->mode = mode; | |
5653 | entry->proc_handler = proc_handler; | |
5654 | } | |
5655 | ||
5656 | static struct ctl_table * | |
5657 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5658 | { | |
a5d8c348 | 5659 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5660 | |
ad1cdc1d MM |
5661 | if (table == NULL) |
5662 | return NULL; | |
5663 | ||
e0361851 | 5664 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5665 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5666 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5667 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5668 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5669 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5670 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5671 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5672 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5673 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5674 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5675 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5676 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5677 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5678 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5679 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5680 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5681 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5682 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5683 | &sd->cache_nice_tries, |
5684 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5685 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5686 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5687 | set_table_entry(&table[11], "name", sd->name, |
5688 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5689 | /* &table[12] is terminator */ | |
e692ab53 NP |
5690 | |
5691 | return table; | |
5692 | } | |
5693 | ||
9a4e7159 | 5694 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5695 | { |
5696 | struct ctl_table *entry, *table; | |
5697 | struct sched_domain *sd; | |
5698 | int domain_num = 0, i; | |
5699 | char buf[32]; | |
5700 | ||
5701 | for_each_domain(cpu, sd) | |
5702 | domain_num++; | |
5703 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5704 | if (table == NULL) |
5705 | return NULL; | |
e692ab53 NP |
5706 | |
5707 | i = 0; | |
5708 | for_each_domain(cpu, sd) { | |
5709 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5710 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5711 | entry->mode = 0555; |
e692ab53 NP |
5712 | entry->child = sd_alloc_ctl_domain_table(sd); |
5713 | entry++; | |
5714 | i++; | |
5715 | } | |
5716 | return table; | |
5717 | } | |
5718 | ||
5719 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5720 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5721 | { |
6ad4c188 | 5722 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5723 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5724 | char buf[32]; | |
5725 | ||
7378547f MM |
5726 | WARN_ON(sd_ctl_dir[0].child); |
5727 | sd_ctl_dir[0].child = entry; | |
5728 | ||
ad1cdc1d MM |
5729 | if (entry == NULL) |
5730 | return; | |
5731 | ||
6ad4c188 | 5732 | for_each_possible_cpu(i) { |
e692ab53 | 5733 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5734 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5735 | entry->mode = 0555; |
e692ab53 | 5736 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5737 | entry++; |
e692ab53 | 5738 | } |
7378547f MM |
5739 | |
5740 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5741 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5742 | } | |
6382bc90 | 5743 | |
7378547f | 5744 | /* may be called multiple times per register */ |
6382bc90 MM |
5745 | static void unregister_sched_domain_sysctl(void) |
5746 | { | |
7378547f MM |
5747 | if (sd_sysctl_header) |
5748 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5749 | sd_sysctl_header = NULL; |
7378547f MM |
5750 | if (sd_ctl_dir[0].child) |
5751 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5752 | } |
e692ab53 | 5753 | #else |
6382bc90 MM |
5754 | static void register_sched_domain_sysctl(void) |
5755 | { | |
5756 | } | |
5757 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5758 | { |
5759 | } | |
5760 | #endif | |
5761 | ||
1f11eb6a GH |
5762 | static void set_rq_online(struct rq *rq) |
5763 | { | |
5764 | if (!rq->online) { | |
5765 | const struct sched_class *class; | |
5766 | ||
c6c4927b | 5767 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5768 | rq->online = 1; |
5769 | ||
5770 | for_each_class(class) { | |
5771 | if (class->rq_online) | |
5772 | class->rq_online(rq); | |
5773 | } | |
5774 | } | |
5775 | } | |
5776 | ||
5777 | static void set_rq_offline(struct rq *rq) | |
5778 | { | |
5779 | if (rq->online) { | |
5780 | const struct sched_class *class; | |
5781 | ||
5782 | for_each_class(class) { | |
5783 | if (class->rq_offline) | |
5784 | class->rq_offline(rq); | |
5785 | } | |
5786 | ||
c6c4927b | 5787 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5788 | rq->online = 0; |
5789 | } | |
5790 | } | |
5791 | ||
1da177e4 LT |
5792 | /* |
5793 | * migration_call - callback that gets triggered when a CPU is added. | |
5794 | * Here we can start up the necessary migration thread for the new CPU. | |
5795 | */ | |
48f24c4d IM |
5796 | static int __cpuinit |
5797 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5798 | { |
1da177e4 | 5799 | struct task_struct *p; |
48f24c4d | 5800 | int cpu = (long)hcpu; |
1da177e4 | 5801 | unsigned long flags; |
70b97a7f | 5802 | struct rq *rq; |
1da177e4 LT |
5803 | |
5804 | switch (action) { | |
5be9361c | 5805 | |
1da177e4 | 5806 | case CPU_UP_PREPARE: |
8bb78442 | 5807 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5808 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5809 | if (IS_ERR(p)) |
5810 | return NOTIFY_BAD; | |
1da177e4 LT |
5811 | kthread_bind(p, cpu); |
5812 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5813 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5814 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 5815 | task_rq_unlock(rq, &flags); |
371cbb38 | 5816 | get_task_struct(p); |
1da177e4 | 5817 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 5818 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5819 | break; |
48f24c4d | 5820 | |
1da177e4 | 5821 | case CPU_ONLINE: |
8bb78442 | 5822 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5823 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 5824 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
5825 | |
5826 | /* Update our root-domain */ | |
5827 | rq = cpu_rq(cpu); | |
05fa785c | 5828 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5829 | if (rq->rd) { |
c6c4927b | 5830 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5831 | |
5832 | set_rq_online(rq); | |
1f94ef59 | 5833 | } |
05fa785c | 5834 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5835 | break; |
48f24c4d | 5836 | |
1da177e4 LT |
5837 | #ifdef CONFIG_HOTPLUG_CPU |
5838 | case CPU_UP_CANCELED: | |
8bb78442 | 5839 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5840 | if (!cpu_rq(cpu)->migration_thread) |
5841 | break; | |
41a2d6cf | 5842 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 5843 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 5844 | cpumask_any(cpu_online_mask)); |
1da177e4 | 5845 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 5846 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
5847 | cpu_rq(cpu)->migration_thread = NULL; |
5848 | break; | |
48f24c4d | 5849 | |
1da177e4 | 5850 | case CPU_DEAD: |
8bb78442 | 5851 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5852 | migrate_live_tasks(cpu); |
5853 | rq = cpu_rq(cpu); | |
5854 | kthread_stop(rq->migration_thread); | |
371cbb38 | 5855 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
5856 | rq->migration_thread = NULL; |
5857 | /* Idle task back to normal (off runqueue, low prio) */ | |
05fa785c | 5858 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 5859 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5860 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5861 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5862 | migrate_dead_tasks(cpu); |
05fa785c | 5863 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5864 | migrate_nr_uninterruptible(rq); |
5865 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5866 | calc_global_load_remove(rq); |
41a2d6cf IM |
5867 | /* |
5868 | * No need to migrate the tasks: it was best-effort if | |
5869 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5870 | * the requestors. | |
5871 | */ | |
05fa785c | 5872 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5873 | while (!list_empty(&rq->migration_queue)) { |
70b97a7f IM |
5874 | struct migration_req *req; |
5875 | ||
1da177e4 | 5876 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5877 | struct migration_req, list); |
1da177e4 | 5878 | list_del_init(&req->list); |
05fa785c | 5879 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5880 | complete(&req->done); |
05fa785c | 5881 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5882 | } |
05fa785c | 5883 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5884 | break; |
57d885fe | 5885 | |
08f503b0 GH |
5886 | case CPU_DYING: |
5887 | case CPU_DYING_FROZEN: | |
57d885fe GH |
5888 | /* Update our root-domain */ |
5889 | rq = cpu_rq(cpu); | |
05fa785c | 5890 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5891 | if (rq->rd) { |
c6c4927b | 5892 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5893 | set_rq_offline(rq); |
57d885fe | 5894 | } |
05fa785c | 5895 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 5896 | break; |
1da177e4 LT |
5897 | #endif |
5898 | } | |
5899 | return NOTIFY_OK; | |
5900 | } | |
5901 | ||
f38b0820 PM |
5902 | /* |
5903 | * Register at high priority so that task migration (migrate_all_tasks) | |
5904 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5905 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5906 | */ |
26c2143b | 5907 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5908 | .notifier_call = migration_call, |
5909 | .priority = 10 | |
5910 | }; | |
5911 | ||
7babe8db | 5912 | static int __init migration_init(void) |
1da177e4 LT |
5913 | { |
5914 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5915 | int err; |
48f24c4d IM |
5916 | |
5917 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5918 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5919 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5920 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5921 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5922 | |
a004cd42 | 5923 | return 0; |
1da177e4 | 5924 | } |
7babe8db | 5925 | early_initcall(migration_init); |
1da177e4 LT |
5926 | #endif |
5927 | ||
5928 | #ifdef CONFIG_SMP | |
476f3534 | 5929 | |
3e9830dc | 5930 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5931 | |
f6630114 MT |
5932 | static __read_mostly int sched_domain_debug_enabled; |
5933 | ||
5934 | static int __init sched_domain_debug_setup(char *str) | |
5935 | { | |
5936 | sched_domain_debug_enabled = 1; | |
5937 | ||
5938 | return 0; | |
5939 | } | |
5940 | early_param("sched_debug", sched_domain_debug_setup); | |
5941 | ||
7c16ec58 | 5942 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5943 | struct cpumask *groupmask) |
1da177e4 | 5944 | { |
4dcf6aff | 5945 | struct sched_group *group = sd->groups; |
434d53b0 | 5946 | char str[256]; |
1da177e4 | 5947 | |
968ea6d8 | 5948 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5949 | cpumask_clear(groupmask); |
4dcf6aff IM |
5950 | |
5951 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5952 | ||
5953 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5954 | printk("does not load-balance\n"); |
4dcf6aff | 5955 | if (sd->parent) |
3df0fc5b PZ |
5956 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5957 | " has parent"); | |
4dcf6aff | 5958 | return -1; |
41c7ce9a NP |
5959 | } |
5960 | ||
3df0fc5b | 5961 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5962 | |
758b2cdc | 5963 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5964 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5965 | "CPU%d\n", cpu); | |
4dcf6aff | 5966 | } |
758b2cdc | 5967 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5968 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5969 | " CPU%d\n", cpu); | |
4dcf6aff | 5970 | } |
1da177e4 | 5971 | |
4dcf6aff | 5972 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5973 | do { |
4dcf6aff | 5974 | if (!group) { |
3df0fc5b PZ |
5975 | printk("\n"); |
5976 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5977 | break; |
5978 | } | |
5979 | ||
18a3885f | 5980 | if (!group->cpu_power) { |
3df0fc5b PZ |
5981 | printk(KERN_CONT "\n"); |
5982 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5983 | "set\n"); | |
4dcf6aff IM |
5984 | break; |
5985 | } | |
1da177e4 | 5986 | |
758b2cdc | 5987 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5988 | printk(KERN_CONT "\n"); |
5989 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5990 | break; |
5991 | } | |
1da177e4 | 5992 | |
758b2cdc | 5993 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5994 | printk(KERN_CONT "\n"); |
5995 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5996 | break; |
5997 | } | |
1da177e4 | 5998 | |
758b2cdc | 5999 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6000 | |
968ea6d8 | 6001 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6002 | |
3df0fc5b | 6003 | printk(KERN_CONT " %s", str); |
18a3885f | 6004 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6005 | printk(KERN_CONT " (cpu_power = %d)", |
6006 | group->cpu_power); | |
381512cf | 6007 | } |
1da177e4 | 6008 | |
4dcf6aff IM |
6009 | group = group->next; |
6010 | } while (group != sd->groups); | |
3df0fc5b | 6011 | printk(KERN_CONT "\n"); |
1da177e4 | 6012 | |
758b2cdc | 6013 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6014 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6015 | |
758b2cdc RR |
6016 | if (sd->parent && |
6017 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6018 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6019 | "of domain->span\n"); | |
4dcf6aff IM |
6020 | return 0; |
6021 | } | |
1da177e4 | 6022 | |
4dcf6aff IM |
6023 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6024 | { | |
d5dd3db1 | 6025 | cpumask_var_t groupmask; |
4dcf6aff | 6026 | int level = 0; |
1da177e4 | 6027 | |
f6630114 MT |
6028 | if (!sched_domain_debug_enabled) |
6029 | return; | |
6030 | ||
4dcf6aff IM |
6031 | if (!sd) { |
6032 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6033 | return; | |
6034 | } | |
1da177e4 | 6035 | |
4dcf6aff IM |
6036 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6037 | ||
d5dd3db1 | 6038 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6039 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6040 | return; | |
6041 | } | |
6042 | ||
4dcf6aff | 6043 | for (;;) { |
7c16ec58 | 6044 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6045 | break; |
1da177e4 LT |
6046 | level++; |
6047 | sd = sd->parent; | |
33859f7f | 6048 | if (!sd) |
4dcf6aff IM |
6049 | break; |
6050 | } | |
d5dd3db1 | 6051 | free_cpumask_var(groupmask); |
1da177e4 | 6052 | } |
6d6bc0ad | 6053 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6054 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6055 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6056 | |
1a20ff27 | 6057 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6058 | { |
758b2cdc | 6059 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6060 | return 1; |
6061 | ||
6062 | /* Following flags need at least 2 groups */ | |
6063 | if (sd->flags & (SD_LOAD_BALANCE | | |
6064 | SD_BALANCE_NEWIDLE | | |
6065 | SD_BALANCE_FORK | | |
89c4710e SS |
6066 | SD_BALANCE_EXEC | |
6067 | SD_SHARE_CPUPOWER | | |
6068 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6069 | if (sd->groups != sd->groups->next) |
6070 | return 0; | |
6071 | } | |
6072 | ||
6073 | /* Following flags don't use groups */ | |
c88d5910 | 6074 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6075 | return 0; |
6076 | ||
6077 | return 1; | |
6078 | } | |
6079 | ||
48f24c4d IM |
6080 | static int |
6081 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6082 | { |
6083 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6084 | ||
6085 | if (sd_degenerate(parent)) | |
6086 | return 1; | |
6087 | ||
758b2cdc | 6088 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6089 | return 0; |
6090 | ||
245af2c7 SS |
6091 | /* Flags needing groups don't count if only 1 group in parent */ |
6092 | if (parent->groups == parent->groups->next) { | |
6093 | pflags &= ~(SD_LOAD_BALANCE | | |
6094 | SD_BALANCE_NEWIDLE | | |
6095 | SD_BALANCE_FORK | | |
89c4710e SS |
6096 | SD_BALANCE_EXEC | |
6097 | SD_SHARE_CPUPOWER | | |
6098 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6099 | if (nr_node_ids == 1) |
6100 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6101 | } |
6102 | if (~cflags & pflags) | |
6103 | return 0; | |
6104 | ||
6105 | return 1; | |
6106 | } | |
6107 | ||
c6c4927b RR |
6108 | static void free_rootdomain(struct root_domain *rd) |
6109 | { | |
047106ad PZ |
6110 | synchronize_sched(); |
6111 | ||
68e74568 RR |
6112 | cpupri_cleanup(&rd->cpupri); |
6113 | ||
c6c4927b RR |
6114 | free_cpumask_var(rd->rto_mask); |
6115 | free_cpumask_var(rd->online); | |
6116 | free_cpumask_var(rd->span); | |
6117 | kfree(rd); | |
6118 | } | |
6119 | ||
57d885fe GH |
6120 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6121 | { | |
a0490fa3 | 6122 | struct root_domain *old_rd = NULL; |
57d885fe | 6123 | unsigned long flags; |
57d885fe | 6124 | |
05fa785c | 6125 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6126 | |
6127 | if (rq->rd) { | |
a0490fa3 | 6128 | old_rd = rq->rd; |
57d885fe | 6129 | |
c6c4927b | 6130 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6131 | set_rq_offline(rq); |
57d885fe | 6132 | |
c6c4927b | 6133 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6134 | |
a0490fa3 IM |
6135 | /* |
6136 | * If we dont want to free the old_rt yet then | |
6137 | * set old_rd to NULL to skip the freeing later | |
6138 | * in this function: | |
6139 | */ | |
6140 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6141 | old_rd = NULL; | |
57d885fe GH |
6142 | } |
6143 | ||
6144 | atomic_inc(&rd->refcount); | |
6145 | rq->rd = rd; | |
6146 | ||
c6c4927b | 6147 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6148 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6149 | set_rq_online(rq); |
57d885fe | 6150 | |
05fa785c | 6151 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6152 | |
6153 | if (old_rd) | |
6154 | free_rootdomain(old_rd); | |
57d885fe GH |
6155 | } |
6156 | ||
fd5e1b5d | 6157 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6158 | { |
36b7b6d4 PE |
6159 | gfp_t gfp = GFP_KERNEL; |
6160 | ||
57d885fe GH |
6161 | memset(rd, 0, sizeof(*rd)); |
6162 | ||
36b7b6d4 PE |
6163 | if (bootmem) |
6164 | gfp = GFP_NOWAIT; | |
c6c4927b | 6165 | |
36b7b6d4 | 6166 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6167 | goto out; |
36b7b6d4 | 6168 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6169 | goto free_span; |
36b7b6d4 | 6170 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6171 | goto free_online; |
6e0534f2 | 6172 | |
0fb53029 | 6173 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6174 | goto free_rto_mask; |
c6c4927b | 6175 | return 0; |
6e0534f2 | 6176 | |
68e74568 RR |
6177 | free_rto_mask: |
6178 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6179 | free_online: |
6180 | free_cpumask_var(rd->online); | |
6181 | free_span: | |
6182 | free_cpumask_var(rd->span); | |
0c910d28 | 6183 | out: |
c6c4927b | 6184 | return -ENOMEM; |
57d885fe GH |
6185 | } |
6186 | ||
6187 | static void init_defrootdomain(void) | |
6188 | { | |
c6c4927b RR |
6189 | init_rootdomain(&def_root_domain, true); |
6190 | ||
57d885fe GH |
6191 | atomic_set(&def_root_domain.refcount, 1); |
6192 | } | |
6193 | ||
dc938520 | 6194 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6195 | { |
6196 | struct root_domain *rd; | |
6197 | ||
6198 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6199 | if (!rd) | |
6200 | return NULL; | |
6201 | ||
c6c4927b RR |
6202 | if (init_rootdomain(rd, false) != 0) { |
6203 | kfree(rd); | |
6204 | return NULL; | |
6205 | } | |
57d885fe GH |
6206 | |
6207 | return rd; | |
6208 | } | |
6209 | ||
1da177e4 | 6210 | /* |
0eab9146 | 6211 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6212 | * hold the hotplug lock. |
6213 | */ | |
0eab9146 IM |
6214 | static void |
6215 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6216 | { |
70b97a7f | 6217 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6218 | struct sched_domain *tmp; |
6219 | ||
6220 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6221 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6222 | struct sched_domain *parent = tmp->parent; |
6223 | if (!parent) | |
6224 | break; | |
f29c9b1c | 6225 | |
1a848870 | 6226 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6227 | tmp->parent = parent->parent; |
1a848870 SS |
6228 | if (parent->parent) |
6229 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6230 | } else |
6231 | tmp = tmp->parent; | |
245af2c7 SS |
6232 | } |
6233 | ||
1a848870 | 6234 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6235 | sd = sd->parent; |
1a848870 SS |
6236 | if (sd) |
6237 | sd->child = NULL; | |
6238 | } | |
1da177e4 LT |
6239 | |
6240 | sched_domain_debug(sd, cpu); | |
6241 | ||
57d885fe | 6242 | rq_attach_root(rq, rd); |
674311d5 | 6243 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6244 | } |
6245 | ||
6246 | /* cpus with isolated domains */ | |
dcc30a35 | 6247 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6248 | |
6249 | /* Setup the mask of cpus configured for isolated domains */ | |
6250 | static int __init isolated_cpu_setup(char *str) | |
6251 | { | |
bdddd296 | 6252 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6253 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6254 | return 1; |
6255 | } | |
6256 | ||
8927f494 | 6257 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6258 | |
6259 | /* | |
6711cab4 SS |
6260 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6261 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6262 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6263 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6264 | * |
6265 | * init_sched_build_groups will build a circular linked list of the groups | |
6266 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6267 | * and ->cpu_power to 0. | |
6268 | */ | |
a616058b | 6269 | static void |
96f874e2 RR |
6270 | init_sched_build_groups(const struct cpumask *span, |
6271 | const struct cpumask *cpu_map, | |
6272 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6273 | struct sched_group **sg, |
96f874e2 RR |
6274 | struct cpumask *tmpmask), |
6275 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6276 | { |
6277 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6278 | int i; |
6279 | ||
96f874e2 | 6280 | cpumask_clear(covered); |
7c16ec58 | 6281 | |
abcd083a | 6282 | for_each_cpu(i, span) { |
6711cab4 | 6283 | struct sched_group *sg; |
7c16ec58 | 6284 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6285 | int j; |
6286 | ||
758b2cdc | 6287 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6288 | continue; |
6289 | ||
758b2cdc | 6290 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6291 | sg->cpu_power = 0; |
1da177e4 | 6292 | |
abcd083a | 6293 | for_each_cpu(j, span) { |
7c16ec58 | 6294 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6295 | continue; |
6296 | ||
96f874e2 | 6297 | cpumask_set_cpu(j, covered); |
758b2cdc | 6298 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6299 | } |
6300 | if (!first) | |
6301 | first = sg; | |
6302 | if (last) | |
6303 | last->next = sg; | |
6304 | last = sg; | |
6305 | } | |
6306 | last->next = first; | |
6307 | } | |
6308 | ||
9c1cfda2 | 6309 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6310 | |
9c1cfda2 | 6311 | #ifdef CONFIG_NUMA |
198e2f18 | 6312 | |
9c1cfda2 JH |
6313 | /** |
6314 | * find_next_best_node - find the next node to include in a sched_domain | |
6315 | * @node: node whose sched_domain we're building | |
6316 | * @used_nodes: nodes already in the sched_domain | |
6317 | * | |
41a2d6cf | 6318 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6319 | * finds the closest node not already in the @used_nodes map. |
6320 | * | |
6321 | * Should use nodemask_t. | |
6322 | */ | |
c5f59f08 | 6323 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6324 | { |
6325 | int i, n, val, min_val, best_node = 0; | |
6326 | ||
6327 | min_val = INT_MAX; | |
6328 | ||
076ac2af | 6329 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6330 | /* Start at @node */ |
076ac2af | 6331 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6332 | |
6333 | if (!nr_cpus_node(n)) | |
6334 | continue; | |
6335 | ||
6336 | /* Skip already used nodes */ | |
c5f59f08 | 6337 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6338 | continue; |
6339 | ||
6340 | /* Simple min distance search */ | |
6341 | val = node_distance(node, n); | |
6342 | ||
6343 | if (val < min_val) { | |
6344 | min_val = val; | |
6345 | best_node = n; | |
6346 | } | |
6347 | } | |
6348 | ||
c5f59f08 | 6349 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6350 | return best_node; |
6351 | } | |
6352 | ||
6353 | /** | |
6354 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6355 | * @node: node whose cpumask we're constructing | |
73486722 | 6356 | * @span: resulting cpumask |
9c1cfda2 | 6357 | * |
41a2d6cf | 6358 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6359 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6360 | * out optimally. | |
6361 | */ | |
96f874e2 | 6362 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6363 | { |
c5f59f08 | 6364 | nodemask_t used_nodes; |
48f24c4d | 6365 | int i; |
9c1cfda2 | 6366 | |
6ca09dfc | 6367 | cpumask_clear(span); |
c5f59f08 | 6368 | nodes_clear(used_nodes); |
9c1cfda2 | 6369 | |
6ca09dfc | 6370 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6371 | node_set(node, used_nodes); |
9c1cfda2 JH |
6372 | |
6373 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6374 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6375 | |
6ca09dfc | 6376 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6377 | } |
9c1cfda2 | 6378 | } |
6d6bc0ad | 6379 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6380 | |
5c45bf27 | 6381 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6382 | |
6c99e9ad RR |
6383 | /* |
6384 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6385 | * |
6386 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6387 | * and struct sched_domain. ) | |
6c99e9ad RR |
6388 | */ |
6389 | struct static_sched_group { | |
6390 | struct sched_group sg; | |
6391 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6392 | }; | |
6393 | ||
6394 | struct static_sched_domain { | |
6395 | struct sched_domain sd; | |
6396 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6397 | }; | |
6398 | ||
49a02c51 AH |
6399 | struct s_data { |
6400 | #ifdef CONFIG_NUMA | |
6401 | int sd_allnodes; | |
6402 | cpumask_var_t domainspan; | |
6403 | cpumask_var_t covered; | |
6404 | cpumask_var_t notcovered; | |
6405 | #endif | |
6406 | cpumask_var_t nodemask; | |
6407 | cpumask_var_t this_sibling_map; | |
6408 | cpumask_var_t this_core_map; | |
6409 | cpumask_var_t send_covered; | |
6410 | cpumask_var_t tmpmask; | |
6411 | struct sched_group **sched_group_nodes; | |
6412 | struct root_domain *rd; | |
6413 | }; | |
6414 | ||
2109b99e AH |
6415 | enum s_alloc { |
6416 | sa_sched_groups = 0, | |
6417 | sa_rootdomain, | |
6418 | sa_tmpmask, | |
6419 | sa_send_covered, | |
6420 | sa_this_core_map, | |
6421 | sa_this_sibling_map, | |
6422 | sa_nodemask, | |
6423 | sa_sched_group_nodes, | |
6424 | #ifdef CONFIG_NUMA | |
6425 | sa_notcovered, | |
6426 | sa_covered, | |
6427 | sa_domainspan, | |
6428 | #endif | |
6429 | sa_none, | |
6430 | }; | |
6431 | ||
9c1cfda2 | 6432 | /* |
48f24c4d | 6433 | * SMT sched-domains: |
9c1cfda2 | 6434 | */ |
1da177e4 | 6435 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6436 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6437 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6438 | |
41a2d6cf | 6439 | static int |
96f874e2 RR |
6440 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6441 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6442 | { |
6711cab4 | 6443 | if (sg) |
1871e52c | 6444 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6445 | return cpu; |
6446 | } | |
6d6bc0ad | 6447 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6448 | |
48f24c4d IM |
6449 | /* |
6450 | * multi-core sched-domains: | |
6451 | */ | |
1e9f28fa | 6452 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6453 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6454 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6455 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6456 | |
6457 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6458 | static int |
96f874e2 RR |
6459 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6460 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6461 | { |
6711cab4 | 6462 | int group; |
7c16ec58 | 6463 | |
c69fc56d | 6464 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6465 | group = cpumask_first(mask); |
6711cab4 | 6466 | if (sg) |
6c99e9ad | 6467 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6468 | return group; |
1e9f28fa SS |
6469 | } |
6470 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6471 | static int |
96f874e2 RR |
6472 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6473 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6474 | { |
6711cab4 | 6475 | if (sg) |
6c99e9ad | 6476 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6477 | return cpu; |
6478 | } | |
6479 | #endif | |
6480 | ||
6c99e9ad RR |
6481 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6482 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6483 | |
41a2d6cf | 6484 | static int |
96f874e2 RR |
6485 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6486 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6487 | { |
6711cab4 | 6488 | int group; |
48f24c4d | 6489 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6490 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6491 | group = cpumask_first(mask); |
1e9f28fa | 6492 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6493 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6494 | group = cpumask_first(mask); |
1da177e4 | 6495 | #else |
6711cab4 | 6496 | group = cpu; |
1da177e4 | 6497 | #endif |
6711cab4 | 6498 | if (sg) |
6c99e9ad | 6499 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6500 | return group; |
1da177e4 LT |
6501 | } |
6502 | ||
6503 | #ifdef CONFIG_NUMA | |
1da177e4 | 6504 | /* |
9c1cfda2 JH |
6505 | * The init_sched_build_groups can't handle what we want to do with node |
6506 | * groups, so roll our own. Now each node has its own list of groups which | |
6507 | * gets dynamically allocated. | |
1da177e4 | 6508 | */ |
62ea9ceb | 6509 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6510 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6511 | |
62ea9ceb | 6512 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6513 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6514 | |
96f874e2 RR |
6515 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6516 | struct sched_group **sg, | |
6517 | struct cpumask *nodemask) | |
9c1cfda2 | 6518 | { |
6711cab4 SS |
6519 | int group; |
6520 | ||
6ca09dfc | 6521 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6522 | group = cpumask_first(nodemask); |
6711cab4 SS |
6523 | |
6524 | if (sg) | |
6c99e9ad | 6525 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6526 | return group; |
1da177e4 | 6527 | } |
6711cab4 | 6528 | |
08069033 SS |
6529 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6530 | { | |
6531 | struct sched_group *sg = group_head; | |
6532 | int j; | |
6533 | ||
6534 | if (!sg) | |
6535 | return; | |
3a5c359a | 6536 | do { |
758b2cdc | 6537 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6538 | struct sched_domain *sd; |
08069033 | 6539 | |
6c99e9ad | 6540 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6541 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6542 | /* |
6543 | * Only add "power" once for each | |
6544 | * physical package. | |
6545 | */ | |
6546 | continue; | |
6547 | } | |
08069033 | 6548 | |
18a3885f | 6549 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6550 | } |
6551 | sg = sg->next; | |
6552 | } while (sg != group_head); | |
08069033 | 6553 | } |
0601a88d AH |
6554 | |
6555 | static int build_numa_sched_groups(struct s_data *d, | |
6556 | const struct cpumask *cpu_map, int num) | |
6557 | { | |
6558 | struct sched_domain *sd; | |
6559 | struct sched_group *sg, *prev; | |
6560 | int n, j; | |
6561 | ||
6562 | cpumask_clear(d->covered); | |
6563 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6564 | if (cpumask_empty(d->nodemask)) { | |
6565 | d->sched_group_nodes[num] = NULL; | |
6566 | goto out; | |
6567 | } | |
6568 | ||
6569 | sched_domain_node_span(num, d->domainspan); | |
6570 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6571 | ||
6572 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6573 | GFP_KERNEL, num); | |
6574 | if (!sg) { | |
3df0fc5b PZ |
6575 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6576 | num); | |
0601a88d AH |
6577 | return -ENOMEM; |
6578 | } | |
6579 | d->sched_group_nodes[num] = sg; | |
6580 | ||
6581 | for_each_cpu(j, d->nodemask) { | |
6582 | sd = &per_cpu(node_domains, j).sd; | |
6583 | sd->groups = sg; | |
6584 | } | |
6585 | ||
18a3885f | 6586 | sg->cpu_power = 0; |
0601a88d AH |
6587 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6588 | sg->next = sg; | |
6589 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6590 | ||
6591 | prev = sg; | |
6592 | for (j = 0; j < nr_node_ids; j++) { | |
6593 | n = (num + j) % nr_node_ids; | |
6594 | cpumask_complement(d->notcovered, d->covered); | |
6595 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6596 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6597 | if (cpumask_empty(d->tmpmask)) | |
6598 | break; | |
6599 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6600 | if (cpumask_empty(d->tmpmask)) | |
6601 | continue; | |
6602 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6603 | GFP_KERNEL, num); | |
6604 | if (!sg) { | |
3df0fc5b PZ |
6605 | printk(KERN_WARNING |
6606 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6607 | return -ENOMEM; |
6608 | } | |
18a3885f | 6609 | sg->cpu_power = 0; |
0601a88d AH |
6610 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6611 | sg->next = prev->next; | |
6612 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6613 | prev->next = sg; | |
6614 | prev = sg; | |
6615 | } | |
6616 | out: | |
6617 | return 0; | |
6618 | } | |
6d6bc0ad | 6619 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6620 | |
a616058b | 6621 | #ifdef CONFIG_NUMA |
51888ca2 | 6622 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6623 | static void free_sched_groups(const struct cpumask *cpu_map, |
6624 | struct cpumask *nodemask) | |
51888ca2 | 6625 | { |
a616058b | 6626 | int cpu, i; |
51888ca2 | 6627 | |
abcd083a | 6628 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6629 | struct sched_group **sched_group_nodes |
6630 | = sched_group_nodes_bycpu[cpu]; | |
6631 | ||
51888ca2 SV |
6632 | if (!sched_group_nodes) |
6633 | continue; | |
6634 | ||
076ac2af | 6635 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6636 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6637 | ||
6ca09dfc | 6638 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6639 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6640 | continue; |
6641 | ||
6642 | if (sg == NULL) | |
6643 | continue; | |
6644 | sg = sg->next; | |
6645 | next_sg: | |
6646 | oldsg = sg; | |
6647 | sg = sg->next; | |
6648 | kfree(oldsg); | |
6649 | if (oldsg != sched_group_nodes[i]) | |
6650 | goto next_sg; | |
6651 | } | |
6652 | kfree(sched_group_nodes); | |
6653 | sched_group_nodes_bycpu[cpu] = NULL; | |
6654 | } | |
51888ca2 | 6655 | } |
6d6bc0ad | 6656 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6657 | static void free_sched_groups(const struct cpumask *cpu_map, |
6658 | struct cpumask *nodemask) | |
a616058b SS |
6659 | { |
6660 | } | |
6d6bc0ad | 6661 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6662 | |
89c4710e SS |
6663 | /* |
6664 | * Initialize sched groups cpu_power. | |
6665 | * | |
6666 | * cpu_power indicates the capacity of sched group, which is used while | |
6667 | * distributing the load between different sched groups in a sched domain. | |
6668 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6669 | * there are asymmetries in the topology. If there are asymmetries, group | |
6670 | * having more cpu_power will pickup more load compared to the group having | |
6671 | * less cpu_power. | |
89c4710e SS |
6672 | */ |
6673 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6674 | { | |
6675 | struct sched_domain *child; | |
6676 | struct sched_group *group; | |
f93e65c1 PZ |
6677 | long power; |
6678 | int weight; | |
89c4710e SS |
6679 | |
6680 | WARN_ON(!sd || !sd->groups); | |
6681 | ||
13318a71 | 6682 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6683 | return; |
6684 | ||
6685 | child = sd->child; | |
6686 | ||
18a3885f | 6687 | sd->groups->cpu_power = 0; |
5517d86b | 6688 | |
f93e65c1 PZ |
6689 | if (!child) { |
6690 | power = SCHED_LOAD_SCALE; | |
6691 | weight = cpumask_weight(sched_domain_span(sd)); | |
6692 | /* | |
6693 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6694 | * Usually multiple threads get a better yield out of |
6695 | * that one core than a single thread would have, | |
6696 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6697 | */ |
a52bfd73 PZ |
6698 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6699 | power *= sd->smt_gain; | |
f93e65c1 | 6700 | power /= weight; |
a52bfd73 PZ |
6701 | power >>= SCHED_LOAD_SHIFT; |
6702 | } | |
18a3885f | 6703 | sd->groups->cpu_power += power; |
89c4710e SS |
6704 | return; |
6705 | } | |
6706 | ||
89c4710e | 6707 | /* |
f93e65c1 | 6708 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6709 | */ |
6710 | group = child->groups; | |
6711 | do { | |
18a3885f | 6712 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6713 | group = group->next; |
6714 | } while (group != child->groups); | |
6715 | } | |
6716 | ||
7c16ec58 MT |
6717 | /* |
6718 | * Initializers for schedule domains | |
6719 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6720 | */ | |
6721 | ||
a5d8c348 IM |
6722 | #ifdef CONFIG_SCHED_DEBUG |
6723 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6724 | #else | |
6725 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6726 | #endif | |
6727 | ||
7c16ec58 | 6728 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6729 | |
7c16ec58 MT |
6730 | #define SD_INIT_FUNC(type) \ |
6731 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6732 | { \ | |
6733 | memset(sd, 0, sizeof(*sd)); \ | |
6734 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6735 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6736 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6737 | } |
6738 | ||
6739 | SD_INIT_FUNC(CPU) | |
6740 | #ifdef CONFIG_NUMA | |
6741 | SD_INIT_FUNC(ALLNODES) | |
6742 | SD_INIT_FUNC(NODE) | |
6743 | #endif | |
6744 | #ifdef CONFIG_SCHED_SMT | |
6745 | SD_INIT_FUNC(SIBLING) | |
6746 | #endif | |
6747 | #ifdef CONFIG_SCHED_MC | |
6748 | SD_INIT_FUNC(MC) | |
6749 | #endif | |
6750 | ||
1d3504fc HS |
6751 | static int default_relax_domain_level = -1; |
6752 | ||
6753 | static int __init setup_relax_domain_level(char *str) | |
6754 | { | |
30e0e178 LZ |
6755 | unsigned long val; |
6756 | ||
6757 | val = simple_strtoul(str, NULL, 0); | |
6758 | if (val < SD_LV_MAX) | |
6759 | default_relax_domain_level = val; | |
6760 | ||
1d3504fc HS |
6761 | return 1; |
6762 | } | |
6763 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6764 | ||
6765 | static void set_domain_attribute(struct sched_domain *sd, | |
6766 | struct sched_domain_attr *attr) | |
6767 | { | |
6768 | int request; | |
6769 | ||
6770 | if (!attr || attr->relax_domain_level < 0) { | |
6771 | if (default_relax_domain_level < 0) | |
6772 | return; | |
6773 | else | |
6774 | request = default_relax_domain_level; | |
6775 | } else | |
6776 | request = attr->relax_domain_level; | |
6777 | if (request < sd->level) { | |
6778 | /* turn off idle balance on this domain */ | |
c88d5910 | 6779 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6780 | } else { |
6781 | /* turn on idle balance on this domain */ | |
c88d5910 | 6782 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6783 | } |
6784 | } | |
6785 | ||
2109b99e AH |
6786 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6787 | const struct cpumask *cpu_map) | |
6788 | { | |
6789 | switch (what) { | |
6790 | case sa_sched_groups: | |
6791 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6792 | d->sched_group_nodes = NULL; | |
6793 | case sa_rootdomain: | |
6794 | free_rootdomain(d->rd); /* fall through */ | |
6795 | case sa_tmpmask: | |
6796 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6797 | case sa_send_covered: | |
6798 | free_cpumask_var(d->send_covered); /* fall through */ | |
6799 | case sa_this_core_map: | |
6800 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6801 | case sa_this_sibling_map: | |
6802 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6803 | case sa_nodemask: | |
6804 | free_cpumask_var(d->nodemask); /* fall through */ | |
6805 | case sa_sched_group_nodes: | |
d1b55138 | 6806 | #ifdef CONFIG_NUMA |
2109b99e AH |
6807 | kfree(d->sched_group_nodes); /* fall through */ |
6808 | case sa_notcovered: | |
6809 | free_cpumask_var(d->notcovered); /* fall through */ | |
6810 | case sa_covered: | |
6811 | free_cpumask_var(d->covered); /* fall through */ | |
6812 | case sa_domainspan: | |
6813 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6814 | #endif |
2109b99e AH |
6815 | case sa_none: |
6816 | break; | |
6817 | } | |
6818 | } | |
3404c8d9 | 6819 | |
2109b99e AH |
6820 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6821 | const struct cpumask *cpu_map) | |
6822 | { | |
3404c8d9 | 6823 | #ifdef CONFIG_NUMA |
2109b99e AH |
6824 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6825 | return sa_none; | |
6826 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6827 | return sa_domainspan; | |
6828 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6829 | return sa_covered; | |
6830 | /* Allocate the per-node list of sched groups */ | |
6831 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6832 | sizeof(struct sched_group *), GFP_KERNEL); | |
6833 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6834 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6835 | return sa_notcovered; |
d1b55138 | 6836 | } |
2109b99e | 6837 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6838 | #endif |
2109b99e AH |
6839 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6840 | return sa_sched_group_nodes; | |
6841 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6842 | return sa_nodemask; | |
6843 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6844 | return sa_this_sibling_map; | |
6845 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6846 | return sa_this_core_map; | |
6847 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6848 | return sa_send_covered; | |
6849 | d->rd = alloc_rootdomain(); | |
6850 | if (!d->rd) { | |
3df0fc5b | 6851 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6852 | return sa_tmpmask; |
57d885fe | 6853 | } |
2109b99e AH |
6854 | return sa_rootdomain; |
6855 | } | |
57d885fe | 6856 | |
7f4588f3 AH |
6857 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6858 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6859 | { | |
6860 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6861 | #ifdef CONFIG_NUMA |
7f4588f3 | 6862 | struct sched_domain *parent; |
1da177e4 | 6863 | |
7f4588f3 AH |
6864 | d->sd_allnodes = 0; |
6865 | if (cpumask_weight(cpu_map) > | |
6866 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6867 | sd = &per_cpu(allnodes_domains, i).sd; | |
6868 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6869 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6870 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6871 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6872 | d->sd_allnodes = 1; | |
6873 | } | |
6874 | parent = sd; | |
6875 | ||
6876 | sd = &per_cpu(node_domains, i).sd; | |
6877 | SD_INIT(sd, NODE); | |
6878 | set_domain_attribute(sd, attr); | |
6879 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6880 | sd->parent = parent; | |
6881 | if (parent) | |
6882 | parent->child = sd; | |
6883 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6884 | #endif |
7f4588f3 AH |
6885 | return sd; |
6886 | } | |
1da177e4 | 6887 | |
87cce662 AH |
6888 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6889 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6890 | struct sched_domain *parent, int i) | |
6891 | { | |
6892 | struct sched_domain *sd; | |
6893 | sd = &per_cpu(phys_domains, i).sd; | |
6894 | SD_INIT(sd, CPU); | |
6895 | set_domain_attribute(sd, attr); | |
6896 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6897 | sd->parent = parent; | |
6898 | if (parent) | |
6899 | parent->child = sd; | |
6900 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6901 | return sd; | |
6902 | } | |
1da177e4 | 6903 | |
410c4081 AH |
6904 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
6905 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6906 | struct sched_domain *parent, int i) | |
6907 | { | |
6908 | struct sched_domain *sd = parent; | |
1e9f28fa | 6909 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
6910 | sd = &per_cpu(core_domains, i).sd; |
6911 | SD_INIT(sd, MC); | |
6912 | set_domain_attribute(sd, attr); | |
6913 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
6914 | sd->parent = parent; | |
6915 | parent->child = sd; | |
6916 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 6917 | #endif |
410c4081 AH |
6918 | return sd; |
6919 | } | |
1e9f28fa | 6920 | |
d8173535 AH |
6921 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
6922 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6923 | struct sched_domain *parent, int i) | |
6924 | { | |
6925 | struct sched_domain *sd = parent; | |
1da177e4 | 6926 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
6927 | sd = &per_cpu(cpu_domains, i).sd; |
6928 | SD_INIT(sd, SIBLING); | |
6929 | set_domain_attribute(sd, attr); | |
6930 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
6931 | sd->parent = parent; | |
6932 | parent->child = sd; | |
6933 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 6934 | #endif |
d8173535 AH |
6935 | return sd; |
6936 | } | |
1da177e4 | 6937 | |
0e8e85c9 AH |
6938 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
6939 | const struct cpumask *cpu_map, int cpu) | |
6940 | { | |
6941 | switch (l) { | |
1da177e4 | 6942 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
6943 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
6944 | cpumask_and(d->this_sibling_map, cpu_map, | |
6945 | topology_thread_cpumask(cpu)); | |
6946 | if (cpu == cpumask_first(d->this_sibling_map)) | |
6947 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
6948 | &cpu_to_cpu_group, | |
6949 | d->send_covered, d->tmpmask); | |
6950 | break; | |
1da177e4 | 6951 | #endif |
1e9f28fa | 6952 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
6953 | case SD_LV_MC: /* set up multi-core groups */ |
6954 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
6955 | if (cpu == cpumask_first(d->this_core_map)) | |
6956 | init_sched_build_groups(d->this_core_map, cpu_map, | |
6957 | &cpu_to_core_group, | |
6958 | d->send_covered, d->tmpmask); | |
6959 | break; | |
1e9f28fa | 6960 | #endif |
86548096 AH |
6961 | case SD_LV_CPU: /* set up physical groups */ |
6962 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
6963 | if (!cpumask_empty(d->nodemask)) | |
6964 | init_sched_build_groups(d->nodemask, cpu_map, | |
6965 | &cpu_to_phys_group, | |
6966 | d->send_covered, d->tmpmask); | |
6967 | break; | |
1da177e4 | 6968 | #ifdef CONFIG_NUMA |
de616e36 AH |
6969 | case SD_LV_ALLNODES: |
6970 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
6971 | d->send_covered, d->tmpmask); | |
6972 | break; | |
6973 | #endif | |
0e8e85c9 AH |
6974 | default: |
6975 | break; | |
7c16ec58 | 6976 | } |
0e8e85c9 | 6977 | } |
9c1cfda2 | 6978 | |
2109b99e AH |
6979 | /* |
6980 | * Build sched domains for a given set of cpus and attach the sched domains | |
6981 | * to the individual cpus | |
6982 | */ | |
6983 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
6984 | struct sched_domain_attr *attr) | |
6985 | { | |
6986 | enum s_alloc alloc_state = sa_none; | |
6987 | struct s_data d; | |
294b0c96 | 6988 | struct sched_domain *sd; |
2109b99e | 6989 | int i; |
7c16ec58 | 6990 | #ifdef CONFIG_NUMA |
2109b99e | 6991 | d.sd_allnodes = 0; |
7c16ec58 | 6992 | #endif |
9c1cfda2 | 6993 | |
2109b99e AH |
6994 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6995 | if (alloc_state != sa_rootdomain) | |
6996 | goto error; | |
6997 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 6998 | |
1da177e4 | 6999 | /* |
1a20ff27 | 7000 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7001 | */ |
abcd083a | 7002 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7003 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7004 | cpu_map); | |
9761eea8 | 7005 | |
7f4588f3 | 7006 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7007 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7008 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7009 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7010 | } |
9c1cfda2 | 7011 | |
abcd083a | 7012 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7013 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 7014 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7015 | } |
9c1cfda2 | 7016 | |
1da177e4 | 7017 | /* Set up physical groups */ |
86548096 AH |
7018 | for (i = 0; i < nr_node_ids; i++) |
7019 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7020 | |
1da177e4 LT |
7021 | #ifdef CONFIG_NUMA |
7022 | /* Set up node groups */ | |
de616e36 AH |
7023 | if (d.sd_allnodes) |
7024 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7025 | |
0601a88d AH |
7026 | for (i = 0; i < nr_node_ids; i++) |
7027 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7028 | goto error; |
1da177e4 LT |
7029 | #endif |
7030 | ||
7031 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7032 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7033 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7034 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7035 | init_sched_groups_power(i, sd); |
5c45bf27 | 7036 | } |
1da177e4 | 7037 | #endif |
1e9f28fa | 7038 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7039 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7040 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7041 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7042 | } |
7043 | #endif | |
1e9f28fa | 7044 | |
abcd083a | 7045 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7046 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7047 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7048 | } |
7049 | ||
9c1cfda2 | 7050 | #ifdef CONFIG_NUMA |
076ac2af | 7051 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7052 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7053 | |
49a02c51 | 7054 | if (d.sd_allnodes) { |
6711cab4 | 7055 | struct sched_group *sg; |
f712c0c7 | 7056 | |
96f874e2 | 7057 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7058 | d.tmpmask); |
f712c0c7 SS |
7059 | init_numa_sched_groups_power(sg); |
7060 | } | |
9c1cfda2 JH |
7061 | #endif |
7062 | ||
1da177e4 | 7063 | /* Attach the domains */ |
abcd083a | 7064 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7065 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7066 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7067 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7068 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7069 | #else |
6c99e9ad | 7070 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7071 | #endif |
49a02c51 | 7072 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7073 | } |
51888ca2 | 7074 | |
2109b99e AH |
7075 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7076 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7077 | return 0; | |
51888ca2 | 7078 | |
51888ca2 | 7079 | error: |
2109b99e AH |
7080 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7081 | return -ENOMEM; | |
1da177e4 | 7082 | } |
029190c5 | 7083 | |
96f874e2 | 7084 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7085 | { |
7086 | return __build_sched_domains(cpu_map, NULL); | |
7087 | } | |
7088 | ||
acc3f5d7 | 7089 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7090 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7091 | static struct sched_domain_attr *dattr_cur; |
7092 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7093 | |
7094 | /* | |
7095 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7096 | * cpumask) fails, then fallback to a single sched domain, |
7097 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7098 | */ |
4212823f | 7099 | static cpumask_var_t fallback_doms; |
029190c5 | 7100 | |
ee79d1bd HC |
7101 | /* |
7102 | * arch_update_cpu_topology lets virtualized architectures update the | |
7103 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7104 | * or 0 if it stayed the same. | |
7105 | */ | |
7106 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7107 | { |
ee79d1bd | 7108 | return 0; |
22e52b07 HC |
7109 | } |
7110 | ||
acc3f5d7 RR |
7111 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7112 | { | |
7113 | int i; | |
7114 | cpumask_var_t *doms; | |
7115 | ||
7116 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7117 | if (!doms) | |
7118 | return NULL; | |
7119 | for (i = 0; i < ndoms; i++) { | |
7120 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7121 | free_sched_domains(doms, i); | |
7122 | return NULL; | |
7123 | } | |
7124 | } | |
7125 | return doms; | |
7126 | } | |
7127 | ||
7128 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7129 | { | |
7130 | unsigned int i; | |
7131 | for (i = 0; i < ndoms; i++) | |
7132 | free_cpumask_var(doms[i]); | |
7133 | kfree(doms); | |
7134 | } | |
7135 | ||
1a20ff27 | 7136 | /* |
41a2d6cf | 7137 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7138 | * For now this just excludes isolated cpus, but could be used to |
7139 | * exclude other special cases in the future. | |
1a20ff27 | 7140 | */ |
96f874e2 | 7141 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7142 | { |
7378547f MM |
7143 | int err; |
7144 | ||
22e52b07 | 7145 | arch_update_cpu_topology(); |
029190c5 | 7146 | ndoms_cur = 1; |
acc3f5d7 | 7147 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7148 | if (!doms_cur) |
acc3f5d7 RR |
7149 | doms_cur = &fallback_doms; |
7150 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7151 | dattr_cur = NULL; |
acc3f5d7 | 7152 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7153 | register_sched_domain_sysctl(); |
7378547f MM |
7154 | |
7155 | return err; | |
1a20ff27 DG |
7156 | } |
7157 | ||
96f874e2 RR |
7158 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7159 | struct cpumask *tmpmask) | |
1da177e4 | 7160 | { |
7c16ec58 | 7161 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7162 | } |
1da177e4 | 7163 | |
1a20ff27 DG |
7164 | /* |
7165 | * Detach sched domains from a group of cpus specified in cpu_map | |
7166 | * These cpus will now be attached to the NULL domain | |
7167 | */ | |
96f874e2 | 7168 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7169 | { |
96f874e2 RR |
7170 | /* Save because hotplug lock held. */ |
7171 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7172 | int i; |
7173 | ||
abcd083a | 7174 | for_each_cpu(i, cpu_map) |
57d885fe | 7175 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7176 | synchronize_sched(); |
96f874e2 | 7177 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7178 | } |
7179 | ||
1d3504fc HS |
7180 | /* handle null as "default" */ |
7181 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7182 | struct sched_domain_attr *new, int idx_new) | |
7183 | { | |
7184 | struct sched_domain_attr tmp; | |
7185 | ||
7186 | /* fast path */ | |
7187 | if (!new && !cur) | |
7188 | return 1; | |
7189 | ||
7190 | tmp = SD_ATTR_INIT; | |
7191 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7192 | new ? (new + idx_new) : &tmp, | |
7193 | sizeof(struct sched_domain_attr)); | |
7194 | } | |
7195 | ||
029190c5 PJ |
7196 | /* |
7197 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7198 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7199 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7200 | * It destroys each deleted domain and builds each new domain. | |
7201 | * | |
acc3f5d7 | 7202 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7203 | * The masks don't intersect (don't overlap.) We should setup one |
7204 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7205 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7206 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7207 | * it as it is. | |
7208 | * | |
acc3f5d7 RR |
7209 | * The passed in 'doms_new' should be allocated using |
7210 | * alloc_sched_domains. This routine takes ownership of it and will | |
7211 | * free_sched_domains it when done with it. If the caller failed the | |
7212 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7213 | * and partition_sched_domains() will fallback to the single partition | |
7214 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7215 | * |
96f874e2 | 7216 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7217 | * ndoms_new == 0 is a special case for destroying existing domains, |
7218 | * and it will not create the default domain. | |
dfb512ec | 7219 | * |
029190c5 PJ |
7220 | * Call with hotplug lock held |
7221 | */ | |
acc3f5d7 | 7222 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7223 | struct sched_domain_attr *dattr_new) |
029190c5 | 7224 | { |
dfb512ec | 7225 | int i, j, n; |
d65bd5ec | 7226 | int new_topology; |
029190c5 | 7227 | |
712555ee | 7228 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7229 | |
7378547f MM |
7230 | /* always unregister in case we don't destroy any domains */ |
7231 | unregister_sched_domain_sysctl(); | |
7232 | ||
d65bd5ec HC |
7233 | /* Let architecture update cpu core mappings. */ |
7234 | new_topology = arch_update_cpu_topology(); | |
7235 | ||
dfb512ec | 7236 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7237 | |
7238 | /* Destroy deleted domains */ | |
7239 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7240 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7241 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7242 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7243 | goto match1; |
7244 | } | |
7245 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7246 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7247 | match1: |
7248 | ; | |
7249 | } | |
7250 | ||
e761b772 MK |
7251 | if (doms_new == NULL) { |
7252 | ndoms_cur = 0; | |
acc3f5d7 | 7253 | doms_new = &fallback_doms; |
6ad4c188 | 7254 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7255 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7256 | } |
7257 | ||
029190c5 PJ |
7258 | /* Build new domains */ |
7259 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7260 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7261 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7262 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7263 | goto match2; |
7264 | } | |
7265 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7266 | __build_sched_domains(doms_new[i], |
1d3504fc | 7267 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7268 | match2: |
7269 | ; | |
7270 | } | |
7271 | ||
7272 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7273 | if (doms_cur != &fallback_doms) |
7274 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7275 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7276 | doms_cur = doms_new; |
1d3504fc | 7277 | dattr_cur = dattr_new; |
029190c5 | 7278 | ndoms_cur = ndoms_new; |
7378547f MM |
7279 | |
7280 | register_sched_domain_sysctl(); | |
a1835615 | 7281 | |
712555ee | 7282 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7283 | } |
7284 | ||
5c45bf27 | 7285 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7286 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7287 | { |
95402b38 | 7288 | get_online_cpus(); |
dfb512ec MK |
7289 | |
7290 | /* Destroy domains first to force the rebuild */ | |
7291 | partition_sched_domains(0, NULL, NULL); | |
7292 | ||
e761b772 | 7293 | rebuild_sched_domains(); |
95402b38 | 7294 | put_online_cpus(); |
5c45bf27 SS |
7295 | } |
7296 | ||
7297 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7298 | { | |
afb8a9b7 | 7299 | unsigned int level = 0; |
5c45bf27 | 7300 | |
afb8a9b7 GS |
7301 | if (sscanf(buf, "%u", &level) != 1) |
7302 | return -EINVAL; | |
7303 | ||
7304 | /* | |
7305 | * level is always be positive so don't check for | |
7306 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7307 | * What happens on 0 or 1 byte write, | |
7308 | * need to check for count as well? | |
7309 | */ | |
7310 | ||
7311 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7312 | return -EINVAL; |
7313 | ||
7314 | if (smt) | |
afb8a9b7 | 7315 | sched_smt_power_savings = level; |
5c45bf27 | 7316 | else |
afb8a9b7 | 7317 | sched_mc_power_savings = level; |
5c45bf27 | 7318 | |
c70f22d2 | 7319 | arch_reinit_sched_domains(); |
5c45bf27 | 7320 | |
c70f22d2 | 7321 | return count; |
5c45bf27 SS |
7322 | } |
7323 | ||
5c45bf27 | 7324 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7325 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7326 | struct sysdev_class_attribute *attr, |
f718cd4a | 7327 | char *page) |
5c45bf27 SS |
7328 | { |
7329 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7330 | } | |
f718cd4a | 7331 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7332 | struct sysdev_class_attribute *attr, |
48f24c4d | 7333 | const char *buf, size_t count) |
5c45bf27 SS |
7334 | { |
7335 | return sched_power_savings_store(buf, count, 0); | |
7336 | } | |
f718cd4a AK |
7337 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7338 | sched_mc_power_savings_show, | |
7339 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7340 | #endif |
7341 | ||
7342 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7343 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7344 | struct sysdev_class_attribute *attr, |
f718cd4a | 7345 | char *page) |
5c45bf27 SS |
7346 | { |
7347 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7348 | } | |
f718cd4a | 7349 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7350 | struct sysdev_class_attribute *attr, |
48f24c4d | 7351 | const char *buf, size_t count) |
5c45bf27 SS |
7352 | { |
7353 | return sched_power_savings_store(buf, count, 1); | |
7354 | } | |
f718cd4a AK |
7355 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7356 | sched_smt_power_savings_show, | |
6707de00 AB |
7357 | sched_smt_power_savings_store); |
7358 | #endif | |
7359 | ||
39aac648 | 7360 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7361 | { |
7362 | int err = 0; | |
7363 | ||
7364 | #ifdef CONFIG_SCHED_SMT | |
7365 | if (smt_capable()) | |
7366 | err = sysfs_create_file(&cls->kset.kobj, | |
7367 | &attr_sched_smt_power_savings.attr); | |
7368 | #endif | |
7369 | #ifdef CONFIG_SCHED_MC | |
7370 | if (!err && mc_capable()) | |
7371 | err = sysfs_create_file(&cls->kset.kobj, | |
7372 | &attr_sched_mc_power_savings.attr); | |
7373 | #endif | |
7374 | return err; | |
7375 | } | |
6d6bc0ad | 7376 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7377 | |
e761b772 | 7378 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7379 | /* |
e761b772 MK |
7380 | * Add online and remove offline CPUs from the scheduler domains. |
7381 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7382 | */ |
7383 | static int update_sched_domains(struct notifier_block *nfb, | |
7384 | unsigned long action, void *hcpu) | |
e761b772 MK |
7385 | { |
7386 | switch (action) { | |
7387 | case CPU_ONLINE: | |
7388 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
7389 | case CPU_DOWN_PREPARE: |
7390 | case CPU_DOWN_PREPARE_FROZEN: | |
7391 | case CPU_DOWN_FAILED: | |
7392 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 7393 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7394 | return NOTIFY_OK; |
7395 | ||
7396 | default: | |
7397 | return NOTIFY_DONE; | |
7398 | } | |
7399 | } | |
7400 | #endif | |
7401 | ||
7402 | static int update_runtime(struct notifier_block *nfb, | |
7403 | unsigned long action, void *hcpu) | |
1da177e4 | 7404 | { |
7def2be1 PZ |
7405 | int cpu = (int)(long)hcpu; |
7406 | ||
1da177e4 | 7407 | switch (action) { |
1da177e4 | 7408 | case CPU_DOWN_PREPARE: |
8bb78442 | 7409 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7410 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7411 | return NOTIFY_OK; |
7412 | ||
1da177e4 | 7413 | case CPU_DOWN_FAILED: |
8bb78442 | 7414 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7415 | case CPU_ONLINE: |
8bb78442 | 7416 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7417 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7418 | return NOTIFY_OK; |
7419 | ||
1da177e4 LT |
7420 | default: |
7421 | return NOTIFY_DONE; | |
7422 | } | |
1da177e4 | 7423 | } |
1da177e4 LT |
7424 | |
7425 | void __init sched_init_smp(void) | |
7426 | { | |
dcc30a35 RR |
7427 | cpumask_var_t non_isolated_cpus; |
7428 | ||
7429 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7430 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7431 | |
434d53b0 MT |
7432 | #if defined(CONFIG_NUMA) |
7433 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7434 | GFP_KERNEL); | |
7435 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7436 | #endif | |
95402b38 | 7437 | get_online_cpus(); |
712555ee | 7438 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7439 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7440 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7441 | if (cpumask_empty(non_isolated_cpus)) | |
7442 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7443 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7444 | put_online_cpus(); |
e761b772 MK |
7445 | |
7446 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7447 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7448 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7449 | #endif |
7450 | ||
7451 | /* RT runtime code needs to handle some hotplug events */ | |
7452 | hotcpu_notifier(update_runtime, 0); | |
7453 | ||
b328ca18 | 7454 | init_hrtick(); |
5c1e1767 NP |
7455 | |
7456 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7457 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7458 | BUG(); |
19978ca6 | 7459 | sched_init_granularity(); |
dcc30a35 | 7460 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7461 | |
0e3900e6 | 7462 | init_sched_rt_class(); |
1da177e4 LT |
7463 | } |
7464 | #else | |
7465 | void __init sched_init_smp(void) | |
7466 | { | |
19978ca6 | 7467 | sched_init_granularity(); |
1da177e4 LT |
7468 | } |
7469 | #endif /* CONFIG_SMP */ | |
7470 | ||
cd1bb94b AB |
7471 | const_debug unsigned int sysctl_timer_migration = 1; |
7472 | ||
1da177e4 LT |
7473 | int in_sched_functions(unsigned long addr) |
7474 | { | |
1da177e4 LT |
7475 | return in_lock_functions(addr) || |
7476 | (addr >= (unsigned long)__sched_text_start | |
7477 | && addr < (unsigned long)__sched_text_end); | |
7478 | } | |
7479 | ||
a9957449 | 7480 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7481 | { |
7482 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7483 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7484 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7485 | cfs_rq->rq = rq; | |
7486 | #endif | |
67e9fb2a | 7487 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7488 | } |
7489 | ||
fa85ae24 PZ |
7490 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7491 | { | |
7492 | struct rt_prio_array *array; | |
7493 | int i; | |
7494 | ||
7495 | array = &rt_rq->active; | |
7496 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7497 | INIT_LIST_HEAD(array->queue + i); | |
7498 | __clear_bit(i, array->bitmap); | |
7499 | } | |
7500 | /* delimiter for bitsearch: */ | |
7501 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7502 | ||
052f1dc7 | 7503 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7504 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7505 | #ifdef CONFIG_SMP |
e864c499 | 7506 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7507 | #endif |
48d5e258 | 7508 | #endif |
fa85ae24 PZ |
7509 | #ifdef CONFIG_SMP |
7510 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7511 | rt_rq->overloaded = 0; |
05fa785c | 7512 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7513 | #endif |
7514 | ||
7515 | rt_rq->rt_time = 0; | |
7516 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7517 | rt_rq->rt_runtime = 0; |
0986b11b | 7518 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7519 | |
052f1dc7 | 7520 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7521 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7522 | rt_rq->rq = rq; |
7523 | #endif | |
fa85ae24 PZ |
7524 | } |
7525 | ||
6f505b16 | 7526 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7527 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7528 | struct sched_entity *se, int cpu, int add, | |
7529 | struct sched_entity *parent) | |
6f505b16 | 7530 | { |
ec7dc8ac | 7531 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7532 | tg->cfs_rq[cpu] = cfs_rq; |
7533 | init_cfs_rq(cfs_rq, rq); | |
7534 | cfs_rq->tg = tg; | |
7535 | if (add) | |
7536 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7537 | ||
7538 | tg->se[cpu] = se; | |
354d60c2 DG |
7539 | /* se could be NULL for init_task_group */ |
7540 | if (!se) | |
7541 | return; | |
7542 | ||
ec7dc8ac DG |
7543 | if (!parent) |
7544 | se->cfs_rq = &rq->cfs; | |
7545 | else | |
7546 | se->cfs_rq = parent->my_q; | |
7547 | ||
6f505b16 PZ |
7548 | se->my_q = cfs_rq; |
7549 | se->load.weight = tg->shares; | |
e05510d0 | 7550 | se->load.inv_weight = 0; |
ec7dc8ac | 7551 | se->parent = parent; |
6f505b16 | 7552 | } |
052f1dc7 | 7553 | #endif |
6f505b16 | 7554 | |
052f1dc7 | 7555 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7556 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7557 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7558 | struct sched_rt_entity *parent) | |
6f505b16 | 7559 | { |
ec7dc8ac DG |
7560 | struct rq *rq = cpu_rq(cpu); |
7561 | ||
6f505b16 PZ |
7562 | tg->rt_rq[cpu] = rt_rq; |
7563 | init_rt_rq(rt_rq, rq); | |
7564 | rt_rq->tg = tg; | |
ac086bc2 | 7565 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7566 | if (add) |
7567 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7568 | ||
7569 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7570 | if (!rt_se) |
7571 | return; | |
7572 | ||
ec7dc8ac DG |
7573 | if (!parent) |
7574 | rt_se->rt_rq = &rq->rt; | |
7575 | else | |
7576 | rt_se->rt_rq = parent->my_q; | |
7577 | ||
6f505b16 | 7578 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7579 | rt_se->parent = parent; |
6f505b16 PZ |
7580 | INIT_LIST_HEAD(&rt_se->run_list); |
7581 | } | |
7582 | #endif | |
7583 | ||
1da177e4 LT |
7584 | void __init sched_init(void) |
7585 | { | |
dd41f596 | 7586 | int i, j; |
434d53b0 MT |
7587 | unsigned long alloc_size = 0, ptr; |
7588 | ||
7589 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7590 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7591 | #endif | |
7592 | #ifdef CONFIG_RT_GROUP_SCHED | |
7593 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7594 | #endif |
df7c8e84 | 7595 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7596 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7597 | #endif |
434d53b0 | 7598 | if (alloc_size) { |
36b7b6d4 | 7599 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7600 | |
7601 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7602 | init_task_group.se = (struct sched_entity **)ptr; | |
7603 | ptr += nr_cpu_ids * sizeof(void **); | |
7604 | ||
7605 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7606 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7607 | |
6d6bc0ad | 7608 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7609 | #ifdef CONFIG_RT_GROUP_SCHED |
7610 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7611 | ptr += nr_cpu_ids * sizeof(void **); | |
7612 | ||
7613 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7614 | ptr += nr_cpu_ids * sizeof(void **); |
7615 | ||
6d6bc0ad | 7616 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7617 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7618 | for_each_possible_cpu(i) { | |
7619 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7620 | ptr += cpumask_size(); | |
7621 | } | |
7622 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7623 | } |
dd41f596 | 7624 | |
57d885fe GH |
7625 | #ifdef CONFIG_SMP |
7626 | init_defrootdomain(); | |
7627 | #endif | |
7628 | ||
d0b27fa7 PZ |
7629 | init_rt_bandwidth(&def_rt_bandwidth, |
7630 | global_rt_period(), global_rt_runtime()); | |
7631 | ||
7632 | #ifdef CONFIG_RT_GROUP_SCHED | |
7633 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7634 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7635 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7636 | |
7c941438 | 7637 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7638 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7639 | INIT_LIST_HEAD(&init_task_group.children); |
7640 | ||
7c941438 | 7641 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7642 | |
4a6cc4bd JK |
7643 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7644 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7645 | __alignof__(unsigned long)); | |
7646 | #endif | |
0a945022 | 7647 | for_each_possible_cpu(i) { |
70b97a7f | 7648 | struct rq *rq; |
1da177e4 LT |
7649 | |
7650 | rq = cpu_rq(i); | |
05fa785c | 7651 | raw_spin_lock_init(&rq->lock); |
7897986b | 7652 | rq->nr_running = 0; |
dce48a84 TG |
7653 | rq->calc_load_active = 0; |
7654 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7655 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7656 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7657 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7658 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7659 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7660 | #ifdef CONFIG_CGROUP_SCHED |
7661 | /* | |
7662 | * How much cpu bandwidth does init_task_group get? | |
7663 | * | |
7664 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7665 | * gets 100% of the cpu resources in the system. This overall | |
7666 | * system cpu resource is divided among the tasks of | |
7667 | * init_task_group and its child task-groups in a fair manner, | |
7668 | * based on each entity's (task or task-group's) weight | |
7669 | * (se->load.weight). | |
7670 | * | |
7671 | * In other words, if init_task_group has 10 tasks of weight | |
7672 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7673 | * then A0's share of the cpu resource is: | |
7674 | * | |
0d905bca | 7675 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7676 | * |
7677 | * We achieve this by letting init_task_group's tasks sit | |
7678 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7679 | */ | |
ec7dc8ac | 7680 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7681 | #endif |
354d60c2 DG |
7682 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7683 | ||
7684 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7685 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7686 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7687 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7688 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7689 | #endif |
dd41f596 | 7690 | #endif |
1da177e4 | 7691 | |
dd41f596 IM |
7692 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7693 | rq->cpu_load[j] = 0; | |
1da177e4 | 7694 | #ifdef CONFIG_SMP |
41c7ce9a | 7695 | rq->sd = NULL; |
57d885fe | 7696 | rq->rd = NULL; |
3f029d3c | 7697 | rq->post_schedule = 0; |
1da177e4 | 7698 | rq->active_balance = 0; |
dd41f596 | 7699 | rq->next_balance = jiffies; |
1da177e4 | 7700 | rq->push_cpu = 0; |
0a2966b4 | 7701 | rq->cpu = i; |
1f11eb6a | 7702 | rq->online = 0; |
1da177e4 | 7703 | rq->migration_thread = NULL; |
eae0c9df MG |
7704 | rq->idle_stamp = 0; |
7705 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
1da177e4 | 7706 | INIT_LIST_HEAD(&rq->migration_queue); |
dc938520 | 7707 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7708 | #endif |
8f4d37ec | 7709 | init_rq_hrtick(rq); |
1da177e4 | 7710 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7711 | } |
7712 | ||
2dd73a4f | 7713 | set_load_weight(&init_task); |
b50f60ce | 7714 | |
e107be36 AK |
7715 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7716 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7717 | #endif | |
7718 | ||
c9819f45 | 7719 | #ifdef CONFIG_SMP |
962cf36c | 7720 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7721 | #endif |
7722 | ||
b50f60ce | 7723 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7724 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7725 | #endif |
7726 | ||
1da177e4 LT |
7727 | /* |
7728 | * The boot idle thread does lazy MMU switching as well: | |
7729 | */ | |
7730 | atomic_inc(&init_mm.mm_count); | |
7731 | enter_lazy_tlb(&init_mm, current); | |
7732 | ||
7733 | /* | |
7734 | * Make us the idle thread. Technically, schedule() should not be | |
7735 | * called from this thread, however somewhere below it might be, | |
7736 | * but because we are the idle thread, we just pick up running again | |
7737 | * when this runqueue becomes "idle". | |
7738 | */ | |
7739 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7740 | |
7741 | calc_load_update = jiffies + LOAD_FREQ; | |
7742 | ||
dd41f596 IM |
7743 | /* |
7744 | * During early bootup we pretend to be a normal task: | |
7745 | */ | |
7746 | current->sched_class = &fair_sched_class; | |
6892b75e | 7747 | |
6a7b3dc3 | 7748 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7749 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7750 | #ifdef CONFIG_SMP |
7d1e6a9b | 7751 | #ifdef CONFIG_NO_HZ |
49557e62 | 7752 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 7753 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 7754 | #endif |
bdddd296 RR |
7755 | /* May be allocated at isolcpus cmdline parse time */ |
7756 | if (cpu_isolated_map == NULL) | |
7757 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7758 | #endif /* SMP */ |
6a7b3dc3 | 7759 | |
cdd6c482 | 7760 | perf_event_init(); |
0d905bca | 7761 | |
6892b75e | 7762 | scheduler_running = 1; |
1da177e4 LT |
7763 | } |
7764 | ||
7765 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7766 | static inline int preempt_count_equals(int preempt_offset) |
7767 | { | |
234da7bc | 7768 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7769 | |
7770 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7771 | } | |
7772 | ||
d894837f | 7773 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7774 | { |
48f24c4d | 7775 | #ifdef in_atomic |
1da177e4 LT |
7776 | static unsigned long prev_jiffy; /* ratelimiting */ |
7777 | ||
e4aafea2 FW |
7778 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7779 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7780 | return; |
7781 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7782 | return; | |
7783 | prev_jiffy = jiffies; | |
7784 | ||
3df0fc5b PZ |
7785 | printk(KERN_ERR |
7786 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7787 | file, line); | |
7788 | printk(KERN_ERR | |
7789 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7790 | in_atomic(), irqs_disabled(), | |
7791 | current->pid, current->comm); | |
aef745fc IM |
7792 | |
7793 | debug_show_held_locks(current); | |
7794 | if (irqs_disabled()) | |
7795 | print_irqtrace_events(current); | |
7796 | dump_stack(); | |
1da177e4 LT |
7797 | #endif |
7798 | } | |
7799 | EXPORT_SYMBOL(__might_sleep); | |
7800 | #endif | |
7801 | ||
7802 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7803 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7804 | { | |
7805 | int on_rq; | |
3e51f33f | 7806 | |
3a5e4dc1 AK |
7807 | on_rq = p->se.on_rq; |
7808 | if (on_rq) | |
7809 | deactivate_task(rq, p, 0); | |
7810 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7811 | if (on_rq) { | |
7812 | activate_task(rq, p, 0); | |
7813 | resched_task(rq->curr); | |
7814 | } | |
7815 | } | |
7816 | ||
1da177e4 LT |
7817 | void normalize_rt_tasks(void) |
7818 | { | |
a0f98a1c | 7819 | struct task_struct *g, *p; |
1da177e4 | 7820 | unsigned long flags; |
70b97a7f | 7821 | struct rq *rq; |
1da177e4 | 7822 | |
4cf5d77a | 7823 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7824 | do_each_thread(g, p) { |
178be793 IM |
7825 | /* |
7826 | * Only normalize user tasks: | |
7827 | */ | |
7828 | if (!p->mm) | |
7829 | continue; | |
7830 | ||
6cfb0d5d | 7831 | p->se.exec_start = 0; |
6cfb0d5d | 7832 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7833 | p->se.statistics.wait_start = 0; |
7834 | p->se.statistics.sleep_start = 0; | |
7835 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7836 | #endif |
dd41f596 IM |
7837 | |
7838 | if (!rt_task(p)) { | |
7839 | /* | |
7840 | * Renice negative nice level userspace | |
7841 | * tasks back to 0: | |
7842 | */ | |
7843 | if (TASK_NICE(p) < 0 && p->mm) | |
7844 | set_user_nice(p, 0); | |
1da177e4 | 7845 | continue; |
dd41f596 | 7846 | } |
1da177e4 | 7847 | |
1d615482 | 7848 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7849 | rq = __task_rq_lock(p); |
1da177e4 | 7850 | |
178be793 | 7851 | normalize_task(rq, p); |
3a5e4dc1 | 7852 | |
b29739f9 | 7853 | __task_rq_unlock(rq); |
1d615482 | 7854 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7855 | } while_each_thread(g, p); |
7856 | ||
4cf5d77a | 7857 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7858 | } |
7859 | ||
7860 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7861 | |
7862 | #ifdef CONFIG_IA64 | |
7863 | /* | |
7864 | * These functions are only useful for the IA64 MCA handling. | |
7865 | * | |
7866 | * They can only be called when the whole system has been | |
7867 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7868 | * activity can take place. Using them for anything else would | |
7869 | * be a serious bug, and as a result, they aren't even visible | |
7870 | * under any other configuration. | |
7871 | */ | |
7872 | ||
7873 | /** | |
7874 | * curr_task - return the current task for a given cpu. | |
7875 | * @cpu: the processor in question. | |
7876 | * | |
7877 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7878 | */ | |
36c8b586 | 7879 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7880 | { |
7881 | return cpu_curr(cpu); | |
7882 | } | |
7883 | ||
7884 | /** | |
7885 | * set_curr_task - set the current task for a given cpu. | |
7886 | * @cpu: the processor in question. | |
7887 | * @p: the task pointer to set. | |
7888 | * | |
7889 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7890 | * are serviced on a separate stack. It allows the architecture to switch the |
7891 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7892 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7893 | * and caller must save the original value of the current task (see | |
7894 | * curr_task() above) and restore that value before reenabling interrupts and | |
7895 | * re-starting the system. | |
7896 | * | |
7897 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7898 | */ | |
36c8b586 | 7899 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7900 | { |
7901 | cpu_curr(cpu) = p; | |
7902 | } | |
7903 | ||
7904 | #endif | |
29f59db3 | 7905 | |
bccbe08a PZ |
7906 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7907 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7908 | { |
7909 | int i; | |
7910 | ||
7911 | for_each_possible_cpu(i) { | |
7912 | if (tg->cfs_rq) | |
7913 | kfree(tg->cfs_rq[i]); | |
7914 | if (tg->se) | |
7915 | kfree(tg->se[i]); | |
6f505b16 PZ |
7916 | } |
7917 | ||
7918 | kfree(tg->cfs_rq); | |
7919 | kfree(tg->se); | |
6f505b16 PZ |
7920 | } |
7921 | ||
ec7dc8ac DG |
7922 | static |
7923 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 7924 | { |
29f59db3 | 7925 | struct cfs_rq *cfs_rq; |
eab17229 | 7926 | struct sched_entity *se; |
9b5b7751 | 7927 | struct rq *rq; |
29f59db3 SV |
7928 | int i; |
7929 | ||
434d53b0 | 7930 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7931 | if (!tg->cfs_rq) |
7932 | goto err; | |
434d53b0 | 7933 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7934 | if (!tg->se) |
7935 | goto err; | |
052f1dc7 PZ |
7936 | |
7937 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7938 | |
7939 | for_each_possible_cpu(i) { | |
9b5b7751 | 7940 | rq = cpu_rq(i); |
29f59db3 | 7941 | |
eab17229 LZ |
7942 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
7943 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
7944 | if (!cfs_rq) |
7945 | goto err; | |
7946 | ||
eab17229 LZ |
7947 | se = kzalloc_node(sizeof(struct sched_entity), |
7948 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 7949 | if (!se) |
dfc12eb2 | 7950 | goto err_free_rq; |
29f59db3 | 7951 | |
eab17229 | 7952 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
7953 | } |
7954 | ||
7955 | return 1; | |
7956 | ||
dfc12eb2 PC |
7957 | err_free_rq: |
7958 | kfree(cfs_rq); | |
bccbe08a PZ |
7959 | err: |
7960 | return 0; | |
7961 | } | |
7962 | ||
7963 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7964 | { | |
7965 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7966 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7967 | } | |
7968 | ||
7969 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7970 | { | |
7971 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7972 | } | |
6d6bc0ad | 7973 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
7974 | static inline void free_fair_sched_group(struct task_group *tg) |
7975 | { | |
7976 | } | |
7977 | ||
ec7dc8ac DG |
7978 | static inline |
7979 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7980 | { |
7981 | return 1; | |
7982 | } | |
7983 | ||
7984 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7985 | { | |
7986 | } | |
7987 | ||
7988 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7989 | { | |
7990 | } | |
6d6bc0ad | 7991 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
7992 | |
7993 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7994 | static void free_rt_sched_group(struct task_group *tg) |
7995 | { | |
7996 | int i; | |
7997 | ||
d0b27fa7 PZ |
7998 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
7999 | ||
bccbe08a PZ |
8000 | for_each_possible_cpu(i) { |
8001 | if (tg->rt_rq) | |
8002 | kfree(tg->rt_rq[i]); | |
8003 | if (tg->rt_se) | |
8004 | kfree(tg->rt_se[i]); | |
8005 | } | |
8006 | ||
8007 | kfree(tg->rt_rq); | |
8008 | kfree(tg->rt_se); | |
8009 | } | |
8010 | ||
ec7dc8ac DG |
8011 | static |
8012 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8013 | { |
8014 | struct rt_rq *rt_rq; | |
eab17229 | 8015 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8016 | struct rq *rq; |
8017 | int i; | |
8018 | ||
434d53b0 | 8019 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8020 | if (!tg->rt_rq) |
8021 | goto err; | |
434d53b0 | 8022 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8023 | if (!tg->rt_se) |
8024 | goto err; | |
8025 | ||
d0b27fa7 PZ |
8026 | init_rt_bandwidth(&tg->rt_bandwidth, |
8027 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8028 | |
8029 | for_each_possible_cpu(i) { | |
8030 | rq = cpu_rq(i); | |
8031 | ||
eab17229 LZ |
8032 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8033 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8034 | if (!rt_rq) |
8035 | goto err; | |
29f59db3 | 8036 | |
eab17229 LZ |
8037 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8038 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8039 | if (!rt_se) |
dfc12eb2 | 8040 | goto err_free_rq; |
29f59db3 | 8041 | |
eab17229 | 8042 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8043 | } |
8044 | ||
bccbe08a PZ |
8045 | return 1; |
8046 | ||
dfc12eb2 PC |
8047 | err_free_rq: |
8048 | kfree(rt_rq); | |
bccbe08a PZ |
8049 | err: |
8050 | return 0; | |
8051 | } | |
8052 | ||
8053 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8054 | { | |
8055 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8056 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8057 | } | |
8058 | ||
8059 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8060 | { | |
8061 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8062 | } | |
6d6bc0ad | 8063 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8064 | static inline void free_rt_sched_group(struct task_group *tg) |
8065 | { | |
8066 | } | |
8067 | ||
ec7dc8ac DG |
8068 | static inline |
8069 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8070 | { |
8071 | return 1; | |
8072 | } | |
8073 | ||
8074 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8075 | { | |
8076 | } | |
8077 | ||
8078 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8079 | { | |
8080 | } | |
6d6bc0ad | 8081 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8082 | |
7c941438 | 8083 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8084 | static void free_sched_group(struct task_group *tg) |
8085 | { | |
8086 | free_fair_sched_group(tg); | |
8087 | free_rt_sched_group(tg); | |
8088 | kfree(tg); | |
8089 | } | |
8090 | ||
8091 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8092 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8093 | { |
8094 | struct task_group *tg; | |
8095 | unsigned long flags; | |
8096 | int i; | |
8097 | ||
8098 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8099 | if (!tg) | |
8100 | return ERR_PTR(-ENOMEM); | |
8101 | ||
ec7dc8ac | 8102 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8103 | goto err; |
8104 | ||
ec7dc8ac | 8105 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8106 | goto err; |
8107 | ||
8ed36996 | 8108 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8109 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8110 | register_fair_sched_group(tg, i); |
8111 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8112 | } |
6f505b16 | 8113 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8114 | |
8115 | WARN_ON(!parent); /* root should already exist */ | |
8116 | ||
8117 | tg->parent = parent; | |
f473aa5e | 8118 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8119 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8120 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8121 | |
9b5b7751 | 8122 | return tg; |
29f59db3 SV |
8123 | |
8124 | err: | |
6f505b16 | 8125 | free_sched_group(tg); |
29f59db3 SV |
8126 | return ERR_PTR(-ENOMEM); |
8127 | } | |
8128 | ||
9b5b7751 | 8129 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8130 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8131 | { |
29f59db3 | 8132 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8133 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8134 | } |
8135 | ||
9b5b7751 | 8136 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8137 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8138 | { |
8ed36996 | 8139 | unsigned long flags; |
9b5b7751 | 8140 | int i; |
29f59db3 | 8141 | |
8ed36996 | 8142 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8143 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8144 | unregister_fair_sched_group(tg, i); |
8145 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8146 | } |
6f505b16 | 8147 | list_del_rcu(&tg->list); |
f473aa5e | 8148 | list_del_rcu(&tg->siblings); |
8ed36996 | 8149 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8150 | |
9b5b7751 | 8151 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8152 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8153 | } |
8154 | ||
9b5b7751 | 8155 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8156 | * The caller of this function should have put the task in its new group |
8157 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8158 | * reflect its new group. | |
9b5b7751 SV |
8159 | */ |
8160 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8161 | { |
8162 | int on_rq, running; | |
8163 | unsigned long flags; | |
8164 | struct rq *rq; | |
8165 | ||
8166 | rq = task_rq_lock(tsk, &flags); | |
8167 | ||
051a1d1a | 8168 | running = task_current(rq, tsk); |
29f59db3 SV |
8169 | on_rq = tsk->se.on_rq; |
8170 | ||
0e1f3483 | 8171 | if (on_rq) |
29f59db3 | 8172 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8173 | if (unlikely(running)) |
8174 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8175 | |
6f505b16 | 8176 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8177 | |
810b3817 PZ |
8178 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8179 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8180 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8181 | #endif |
8182 | ||
0e1f3483 HS |
8183 | if (unlikely(running)) |
8184 | tsk->sched_class->set_curr_task(rq); | |
8185 | if (on_rq) | |
371fd7e7 | 8186 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8187 | |
29f59db3 SV |
8188 | task_rq_unlock(rq, &flags); |
8189 | } | |
7c941438 | 8190 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8191 | |
052f1dc7 | 8192 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8193 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8194 | { |
8195 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8196 | int on_rq; |
8197 | ||
29f59db3 | 8198 | on_rq = se->on_rq; |
62fb1851 | 8199 | if (on_rq) |
29f59db3 SV |
8200 | dequeue_entity(cfs_rq, se, 0); |
8201 | ||
8202 | se->load.weight = shares; | |
e05510d0 | 8203 | se->load.inv_weight = 0; |
29f59db3 | 8204 | |
62fb1851 | 8205 | if (on_rq) |
29f59db3 | 8206 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8207 | } |
62fb1851 | 8208 | |
c09595f6 PZ |
8209 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8210 | { | |
8211 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8212 | struct rq *rq = cfs_rq->rq; | |
8213 | unsigned long flags; | |
8214 | ||
05fa785c | 8215 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8216 | __set_se_shares(se, shares); |
05fa785c | 8217 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8218 | } |
8219 | ||
8ed36996 PZ |
8220 | static DEFINE_MUTEX(shares_mutex); |
8221 | ||
4cf86d77 | 8222 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8223 | { |
8224 | int i; | |
8ed36996 | 8225 | unsigned long flags; |
c61935fd | 8226 | |
ec7dc8ac DG |
8227 | /* |
8228 | * We can't change the weight of the root cgroup. | |
8229 | */ | |
8230 | if (!tg->se[0]) | |
8231 | return -EINVAL; | |
8232 | ||
18d95a28 PZ |
8233 | if (shares < MIN_SHARES) |
8234 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8235 | else if (shares > MAX_SHARES) |
8236 | shares = MAX_SHARES; | |
62fb1851 | 8237 | |
8ed36996 | 8238 | mutex_lock(&shares_mutex); |
9b5b7751 | 8239 | if (tg->shares == shares) |
5cb350ba | 8240 | goto done; |
29f59db3 | 8241 | |
8ed36996 | 8242 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8243 | for_each_possible_cpu(i) |
8244 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8245 | list_del_rcu(&tg->siblings); |
8ed36996 | 8246 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8247 | |
8248 | /* wait for any ongoing reference to this group to finish */ | |
8249 | synchronize_sched(); | |
8250 | ||
8251 | /* | |
8252 | * Now we are free to modify the group's share on each cpu | |
8253 | * w/o tripping rebalance_share or load_balance_fair. | |
8254 | */ | |
9b5b7751 | 8255 | tg->shares = shares; |
c09595f6 PZ |
8256 | for_each_possible_cpu(i) { |
8257 | /* | |
8258 | * force a rebalance | |
8259 | */ | |
8260 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8261 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8262 | } |
29f59db3 | 8263 | |
6b2d7700 SV |
8264 | /* |
8265 | * Enable load balance activity on this group, by inserting it back on | |
8266 | * each cpu's rq->leaf_cfs_rq_list. | |
8267 | */ | |
8ed36996 | 8268 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8269 | for_each_possible_cpu(i) |
8270 | register_fair_sched_group(tg, i); | |
f473aa5e | 8271 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8272 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8273 | done: |
8ed36996 | 8274 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8275 | return 0; |
29f59db3 SV |
8276 | } |
8277 | ||
5cb350ba DG |
8278 | unsigned long sched_group_shares(struct task_group *tg) |
8279 | { | |
8280 | return tg->shares; | |
8281 | } | |
052f1dc7 | 8282 | #endif |
5cb350ba | 8283 | |
052f1dc7 | 8284 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8285 | /* |
9f0c1e56 | 8286 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8287 | */ |
9f0c1e56 PZ |
8288 | static DEFINE_MUTEX(rt_constraints_mutex); |
8289 | ||
8290 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8291 | { | |
8292 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8293 | return 1ULL << 20; |
9f0c1e56 | 8294 | |
9a7e0b18 | 8295 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8296 | } |
8297 | ||
9a7e0b18 PZ |
8298 | /* Must be called with tasklist_lock held */ |
8299 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8300 | { |
9a7e0b18 | 8301 | struct task_struct *g, *p; |
b40b2e8e | 8302 | |
9a7e0b18 PZ |
8303 | do_each_thread(g, p) { |
8304 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8305 | return 1; | |
8306 | } while_each_thread(g, p); | |
b40b2e8e | 8307 | |
9a7e0b18 PZ |
8308 | return 0; |
8309 | } | |
b40b2e8e | 8310 | |
9a7e0b18 PZ |
8311 | struct rt_schedulable_data { |
8312 | struct task_group *tg; | |
8313 | u64 rt_period; | |
8314 | u64 rt_runtime; | |
8315 | }; | |
b40b2e8e | 8316 | |
9a7e0b18 PZ |
8317 | static int tg_schedulable(struct task_group *tg, void *data) |
8318 | { | |
8319 | struct rt_schedulable_data *d = data; | |
8320 | struct task_group *child; | |
8321 | unsigned long total, sum = 0; | |
8322 | u64 period, runtime; | |
b40b2e8e | 8323 | |
9a7e0b18 PZ |
8324 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8325 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8326 | |
9a7e0b18 PZ |
8327 | if (tg == d->tg) { |
8328 | period = d->rt_period; | |
8329 | runtime = d->rt_runtime; | |
b40b2e8e | 8330 | } |
b40b2e8e | 8331 | |
4653f803 PZ |
8332 | /* |
8333 | * Cannot have more runtime than the period. | |
8334 | */ | |
8335 | if (runtime > period && runtime != RUNTIME_INF) | |
8336 | return -EINVAL; | |
6f505b16 | 8337 | |
4653f803 PZ |
8338 | /* |
8339 | * Ensure we don't starve existing RT tasks. | |
8340 | */ | |
9a7e0b18 PZ |
8341 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8342 | return -EBUSY; | |
6f505b16 | 8343 | |
9a7e0b18 | 8344 | total = to_ratio(period, runtime); |
6f505b16 | 8345 | |
4653f803 PZ |
8346 | /* |
8347 | * Nobody can have more than the global setting allows. | |
8348 | */ | |
8349 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8350 | return -EINVAL; | |
6f505b16 | 8351 | |
4653f803 PZ |
8352 | /* |
8353 | * The sum of our children's runtime should not exceed our own. | |
8354 | */ | |
9a7e0b18 PZ |
8355 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8356 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8357 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8358 | |
9a7e0b18 PZ |
8359 | if (child == d->tg) { |
8360 | period = d->rt_period; | |
8361 | runtime = d->rt_runtime; | |
8362 | } | |
6f505b16 | 8363 | |
9a7e0b18 | 8364 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8365 | } |
6f505b16 | 8366 | |
9a7e0b18 PZ |
8367 | if (sum > total) |
8368 | return -EINVAL; | |
8369 | ||
8370 | return 0; | |
6f505b16 PZ |
8371 | } |
8372 | ||
9a7e0b18 | 8373 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8374 | { |
9a7e0b18 PZ |
8375 | struct rt_schedulable_data data = { |
8376 | .tg = tg, | |
8377 | .rt_period = period, | |
8378 | .rt_runtime = runtime, | |
8379 | }; | |
8380 | ||
8381 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8382 | } |
8383 | ||
d0b27fa7 PZ |
8384 | static int tg_set_bandwidth(struct task_group *tg, |
8385 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8386 | { |
ac086bc2 | 8387 | int i, err = 0; |
9f0c1e56 | 8388 | |
9f0c1e56 | 8389 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8390 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8391 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8392 | if (err) | |
9f0c1e56 | 8393 | goto unlock; |
ac086bc2 | 8394 | |
0986b11b | 8395 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8396 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8397 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8398 | |
8399 | for_each_possible_cpu(i) { | |
8400 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8401 | ||
0986b11b | 8402 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8403 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8404 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8405 | } |
0986b11b | 8406 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8407 | unlock: |
521f1a24 | 8408 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8409 | mutex_unlock(&rt_constraints_mutex); |
8410 | ||
8411 | return err; | |
6f505b16 PZ |
8412 | } |
8413 | ||
d0b27fa7 PZ |
8414 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8415 | { | |
8416 | u64 rt_runtime, rt_period; | |
8417 | ||
8418 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8419 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8420 | if (rt_runtime_us < 0) | |
8421 | rt_runtime = RUNTIME_INF; | |
8422 | ||
8423 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8424 | } | |
8425 | ||
9f0c1e56 PZ |
8426 | long sched_group_rt_runtime(struct task_group *tg) |
8427 | { | |
8428 | u64 rt_runtime_us; | |
8429 | ||
d0b27fa7 | 8430 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8431 | return -1; |
8432 | ||
d0b27fa7 | 8433 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8434 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8435 | return rt_runtime_us; | |
8436 | } | |
d0b27fa7 PZ |
8437 | |
8438 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8439 | { | |
8440 | u64 rt_runtime, rt_period; | |
8441 | ||
8442 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8443 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8444 | ||
619b0488 R |
8445 | if (rt_period == 0) |
8446 | return -EINVAL; | |
8447 | ||
d0b27fa7 PZ |
8448 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8449 | } | |
8450 | ||
8451 | long sched_group_rt_period(struct task_group *tg) | |
8452 | { | |
8453 | u64 rt_period_us; | |
8454 | ||
8455 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8456 | do_div(rt_period_us, NSEC_PER_USEC); | |
8457 | return rt_period_us; | |
8458 | } | |
8459 | ||
8460 | static int sched_rt_global_constraints(void) | |
8461 | { | |
4653f803 | 8462 | u64 runtime, period; |
d0b27fa7 PZ |
8463 | int ret = 0; |
8464 | ||
ec5d4989 HS |
8465 | if (sysctl_sched_rt_period <= 0) |
8466 | return -EINVAL; | |
8467 | ||
4653f803 PZ |
8468 | runtime = global_rt_runtime(); |
8469 | period = global_rt_period(); | |
8470 | ||
8471 | /* | |
8472 | * Sanity check on the sysctl variables. | |
8473 | */ | |
8474 | if (runtime > period && runtime != RUNTIME_INF) | |
8475 | return -EINVAL; | |
10b612f4 | 8476 | |
d0b27fa7 | 8477 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8478 | read_lock(&tasklist_lock); |
4653f803 | 8479 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8480 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8481 | mutex_unlock(&rt_constraints_mutex); |
8482 | ||
8483 | return ret; | |
8484 | } | |
54e99124 DG |
8485 | |
8486 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8487 | { | |
8488 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8489 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8490 | return 0; | |
8491 | ||
8492 | return 1; | |
8493 | } | |
8494 | ||
6d6bc0ad | 8495 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8496 | static int sched_rt_global_constraints(void) |
8497 | { | |
ac086bc2 PZ |
8498 | unsigned long flags; |
8499 | int i; | |
8500 | ||
ec5d4989 HS |
8501 | if (sysctl_sched_rt_period <= 0) |
8502 | return -EINVAL; | |
8503 | ||
60aa605d PZ |
8504 | /* |
8505 | * There's always some RT tasks in the root group | |
8506 | * -- migration, kstopmachine etc.. | |
8507 | */ | |
8508 | if (sysctl_sched_rt_runtime == 0) | |
8509 | return -EBUSY; | |
8510 | ||
0986b11b | 8511 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8512 | for_each_possible_cpu(i) { |
8513 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8514 | ||
0986b11b | 8515 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8516 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8517 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8518 | } |
0986b11b | 8519 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8520 | |
d0b27fa7 PZ |
8521 | return 0; |
8522 | } | |
6d6bc0ad | 8523 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8524 | |
8525 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8526 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8527 | loff_t *ppos) |
8528 | { | |
8529 | int ret; | |
8530 | int old_period, old_runtime; | |
8531 | static DEFINE_MUTEX(mutex); | |
8532 | ||
8533 | mutex_lock(&mutex); | |
8534 | old_period = sysctl_sched_rt_period; | |
8535 | old_runtime = sysctl_sched_rt_runtime; | |
8536 | ||
8d65af78 | 8537 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8538 | |
8539 | if (!ret && write) { | |
8540 | ret = sched_rt_global_constraints(); | |
8541 | if (ret) { | |
8542 | sysctl_sched_rt_period = old_period; | |
8543 | sysctl_sched_rt_runtime = old_runtime; | |
8544 | } else { | |
8545 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8546 | def_rt_bandwidth.rt_period = | |
8547 | ns_to_ktime(global_rt_period()); | |
8548 | } | |
8549 | } | |
8550 | mutex_unlock(&mutex); | |
8551 | ||
8552 | return ret; | |
8553 | } | |
68318b8e | 8554 | |
052f1dc7 | 8555 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8556 | |
8557 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8558 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8559 | { |
2b01dfe3 PM |
8560 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8561 | struct task_group, css); | |
68318b8e SV |
8562 | } |
8563 | ||
8564 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8565 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8566 | { |
ec7dc8ac | 8567 | struct task_group *tg, *parent; |
68318b8e | 8568 | |
2b01dfe3 | 8569 | if (!cgrp->parent) { |
68318b8e | 8570 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8571 | return &init_task_group.css; |
8572 | } | |
8573 | ||
ec7dc8ac DG |
8574 | parent = cgroup_tg(cgrp->parent); |
8575 | tg = sched_create_group(parent); | |
68318b8e SV |
8576 | if (IS_ERR(tg)) |
8577 | return ERR_PTR(-ENOMEM); | |
8578 | ||
68318b8e SV |
8579 | return &tg->css; |
8580 | } | |
8581 | ||
41a2d6cf IM |
8582 | static void |
8583 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8584 | { |
2b01dfe3 | 8585 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8586 | |
8587 | sched_destroy_group(tg); | |
8588 | } | |
8589 | ||
41a2d6cf | 8590 | static int |
be367d09 | 8591 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8592 | { |
b68aa230 | 8593 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8594 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8595 | return -EINVAL; |
8596 | #else | |
68318b8e SV |
8597 | /* We don't support RT-tasks being in separate groups */ |
8598 | if (tsk->sched_class != &fair_sched_class) | |
8599 | return -EINVAL; | |
b68aa230 | 8600 | #endif |
be367d09 BB |
8601 | return 0; |
8602 | } | |
68318b8e | 8603 | |
be367d09 BB |
8604 | static int |
8605 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8606 | struct task_struct *tsk, bool threadgroup) | |
8607 | { | |
8608 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8609 | if (retval) | |
8610 | return retval; | |
8611 | if (threadgroup) { | |
8612 | struct task_struct *c; | |
8613 | rcu_read_lock(); | |
8614 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8615 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8616 | if (retval) { | |
8617 | rcu_read_unlock(); | |
8618 | return retval; | |
8619 | } | |
8620 | } | |
8621 | rcu_read_unlock(); | |
8622 | } | |
68318b8e SV |
8623 | return 0; |
8624 | } | |
8625 | ||
8626 | static void | |
2b01dfe3 | 8627 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8628 | struct cgroup *old_cont, struct task_struct *tsk, |
8629 | bool threadgroup) | |
68318b8e SV |
8630 | { |
8631 | sched_move_task(tsk); | |
be367d09 BB |
8632 | if (threadgroup) { |
8633 | struct task_struct *c; | |
8634 | rcu_read_lock(); | |
8635 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8636 | sched_move_task(c); | |
8637 | } | |
8638 | rcu_read_unlock(); | |
8639 | } | |
68318b8e SV |
8640 | } |
8641 | ||
052f1dc7 | 8642 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8643 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8644 | u64 shareval) |
68318b8e | 8645 | { |
2b01dfe3 | 8646 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8647 | } |
8648 | ||
f4c753b7 | 8649 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8650 | { |
2b01dfe3 | 8651 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8652 | |
8653 | return (u64) tg->shares; | |
8654 | } | |
6d6bc0ad | 8655 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8656 | |
052f1dc7 | 8657 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8658 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8659 | s64 val) |
6f505b16 | 8660 | { |
06ecb27c | 8661 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8662 | } |
8663 | ||
06ecb27c | 8664 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8665 | { |
06ecb27c | 8666 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8667 | } |
d0b27fa7 PZ |
8668 | |
8669 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8670 | u64 rt_period_us) | |
8671 | { | |
8672 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8673 | } | |
8674 | ||
8675 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8676 | { | |
8677 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8678 | } | |
6d6bc0ad | 8679 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8680 | |
fe5c7cc2 | 8681 | static struct cftype cpu_files[] = { |
052f1dc7 | 8682 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8683 | { |
8684 | .name = "shares", | |
f4c753b7 PM |
8685 | .read_u64 = cpu_shares_read_u64, |
8686 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8687 | }, |
052f1dc7 PZ |
8688 | #endif |
8689 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8690 | { |
9f0c1e56 | 8691 | .name = "rt_runtime_us", |
06ecb27c PM |
8692 | .read_s64 = cpu_rt_runtime_read, |
8693 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8694 | }, |
d0b27fa7 PZ |
8695 | { |
8696 | .name = "rt_period_us", | |
f4c753b7 PM |
8697 | .read_u64 = cpu_rt_period_read_uint, |
8698 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8699 | }, |
052f1dc7 | 8700 | #endif |
68318b8e SV |
8701 | }; |
8702 | ||
8703 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8704 | { | |
fe5c7cc2 | 8705 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8706 | } |
8707 | ||
8708 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8709 | .name = "cpu", |
8710 | .create = cpu_cgroup_create, | |
8711 | .destroy = cpu_cgroup_destroy, | |
8712 | .can_attach = cpu_cgroup_can_attach, | |
8713 | .attach = cpu_cgroup_attach, | |
8714 | .populate = cpu_cgroup_populate, | |
8715 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8716 | .early_init = 1, |
8717 | }; | |
8718 | ||
052f1dc7 | 8719 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8720 | |
8721 | #ifdef CONFIG_CGROUP_CPUACCT | |
8722 | ||
8723 | /* | |
8724 | * CPU accounting code for task groups. | |
8725 | * | |
8726 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8727 | * (balbir@in.ibm.com). | |
8728 | */ | |
8729 | ||
934352f2 | 8730 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8731 | struct cpuacct { |
8732 | struct cgroup_subsys_state css; | |
8733 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8734 | u64 __percpu *cpuusage; |
ef12fefa | 8735 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8736 | struct cpuacct *parent; |
d842de87 SV |
8737 | }; |
8738 | ||
8739 | struct cgroup_subsys cpuacct_subsys; | |
8740 | ||
8741 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8742 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8743 | { |
32cd756a | 8744 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8745 | struct cpuacct, css); |
8746 | } | |
8747 | ||
8748 | /* return cpu accounting group to which this task belongs */ | |
8749 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8750 | { | |
8751 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8752 | struct cpuacct, css); | |
8753 | } | |
8754 | ||
8755 | /* create a new cpu accounting group */ | |
8756 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8757 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8758 | { |
8759 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8760 | int i; |
d842de87 SV |
8761 | |
8762 | if (!ca) | |
ef12fefa | 8763 | goto out; |
d842de87 SV |
8764 | |
8765 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8766 | if (!ca->cpuusage) |
8767 | goto out_free_ca; | |
8768 | ||
8769 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8770 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8771 | goto out_free_counters; | |
d842de87 | 8772 | |
934352f2 BR |
8773 | if (cgrp->parent) |
8774 | ca->parent = cgroup_ca(cgrp->parent); | |
8775 | ||
d842de87 | 8776 | return &ca->css; |
ef12fefa BR |
8777 | |
8778 | out_free_counters: | |
8779 | while (--i >= 0) | |
8780 | percpu_counter_destroy(&ca->cpustat[i]); | |
8781 | free_percpu(ca->cpuusage); | |
8782 | out_free_ca: | |
8783 | kfree(ca); | |
8784 | out: | |
8785 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8786 | } |
8787 | ||
8788 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8789 | static void |
32cd756a | 8790 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8791 | { |
32cd756a | 8792 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8793 | int i; |
d842de87 | 8794 | |
ef12fefa BR |
8795 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8796 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8797 | free_percpu(ca->cpuusage); |
8798 | kfree(ca); | |
8799 | } | |
8800 | ||
720f5498 KC |
8801 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8802 | { | |
b36128c8 | 8803 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8804 | u64 data; |
8805 | ||
8806 | #ifndef CONFIG_64BIT | |
8807 | /* | |
8808 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8809 | */ | |
05fa785c | 8810 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8811 | data = *cpuusage; |
05fa785c | 8812 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8813 | #else |
8814 | data = *cpuusage; | |
8815 | #endif | |
8816 | ||
8817 | return data; | |
8818 | } | |
8819 | ||
8820 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8821 | { | |
b36128c8 | 8822 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8823 | |
8824 | #ifndef CONFIG_64BIT | |
8825 | /* | |
8826 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8827 | */ | |
05fa785c | 8828 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8829 | *cpuusage = val; |
05fa785c | 8830 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8831 | #else |
8832 | *cpuusage = val; | |
8833 | #endif | |
8834 | } | |
8835 | ||
d842de87 | 8836 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8837 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8838 | { |
32cd756a | 8839 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8840 | u64 totalcpuusage = 0; |
8841 | int i; | |
8842 | ||
720f5498 KC |
8843 | for_each_present_cpu(i) |
8844 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8845 | |
8846 | return totalcpuusage; | |
8847 | } | |
8848 | ||
0297b803 DG |
8849 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8850 | u64 reset) | |
8851 | { | |
8852 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8853 | int err = 0; | |
8854 | int i; | |
8855 | ||
8856 | if (reset) { | |
8857 | err = -EINVAL; | |
8858 | goto out; | |
8859 | } | |
8860 | ||
720f5498 KC |
8861 | for_each_present_cpu(i) |
8862 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8863 | |
0297b803 DG |
8864 | out: |
8865 | return err; | |
8866 | } | |
8867 | ||
e9515c3c KC |
8868 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8869 | struct seq_file *m) | |
8870 | { | |
8871 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8872 | u64 percpu; | |
8873 | int i; | |
8874 | ||
8875 | for_each_present_cpu(i) { | |
8876 | percpu = cpuacct_cpuusage_read(ca, i); | |
8877 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8878 | } | |
8879 | seq_printf(m, "\n"); | |
8880 | return 0; | |
8881 | } | |
8882 | ||
ef12fefa BR |
8883 | static const char *cpuacct_stat_desc[] = { |
8884 | [CPUACCT_STAT_USER] = "user", | |
8885 | [CPUACCT_STAT_SYSTEM] = "system", | |
8886 | }; | |
8887 | ||
8888 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8889 | struct cgroup_map_cb *cb) | |
8890 | { | |
8891 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8892 | int i; | |
8893 | ||
8894 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8895 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8896 | val = cputime64_to_clock_t(val); | |
8897 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8898 | } | |
8899 | return 0; | |
8900 | } | |
8901 | ||
d842de87 SV |
8902 | static struct cftype files[] = { |
8903 | { | |
8904 | .name = "usage", | |
f4c753b7 PM |
8905 | .read_u64 = cpuusage_read, |
8906 | .write_u64 = cpuusage_write, | |
d842de87 | 8907 | }, |
e9515c3c KC |
8908 | { |
8909 | .name = "usage_percpu", | |
8910 | .read_seq_string = cpuacct_percpu_seq_read, | |
8911 | }, | |
ef12fefa BR |
8912 | { |
8913 | .name = "stat", | |
8914 | .read_map = cpuacct_stats_show, | |
8915 | }, | |
d842de87 SV |
8916 | }; |
8917 | ||
32cd756a | 8918 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8919 | { |
32cd756a | 8920 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8921 | } |
8922 | ||
8923 | /* | |
8924 | * charge this task's execution time to its accounting group. | |
8925 | * | |
8926 | * called with rq->lock held. | |
8927 | */ | |
8928 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8929 | { | |
8930 | struct cpuacct *ca; | |
934352f2 | 8931 | int cpu; |
d842de87 | 8932 | |
c40c6f85 | 8933 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8934 | return; |
8935 | ||
934352f2 | 8936 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8937 | |
8938 | rcu_read_lock(); | |
8939 | ||
d842de87 | 8940 | ca = task_ca(tsk); |
d842de87 | 8941 | |
934352f2 | 8942 | for (; ca; ca = ca->parent) { |
b36128c8 | 8943 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8944 | *cpuusage += cputime; |
8945 | } | |
a18b83b7 BR |
8946 | |
8947 | rcu_read_unlock(); | |
d842de87 SV |
8948 | } |
8949 | ||
fa535a77 AB |
8950 | /* |
8951 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
8952 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
8953 | * percpu_counter_add with values large enough to always overflow the | |
8954 | * per cpu batch limit causing bad SMP scalability. | |
8955 | * | |
8956 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
8957 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
8958 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
8959 | */ | |
8960 | #ifdef CONFIG_SMP | |
8961 | #define CPUACCT_BATCH \ | |
8962 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
8963 | #else | |
8964 | #define CPUACCT_BATCH 0 | |
8965 | #endif | |
8966 | ||
ef12fefa BR |
8967 | /* |
8968 | * Charge the system/user time to the task's accounting group. | |
8969 | */ | |
8970 | static void cpuacct_update_stats(struct task_struct *tsk, | |
8971 | enum cpuacct_stat_index idx, cputime_t val) | |
8972 | { | |
8973 | struct cpuacct *ca; | |
fa535a77 | 8974 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
8975 | |
8976 | if (unlikely(!cpuacct_subsys.active)) | |
8977 | return; | |
8978 | ||
8979 | rcu_read_lock(); | |
8980 | ca = task_ca(tsk); | |
8981 | ||
8982 | do { | |
fa535a77 | 8983 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
8984 | ca = ca->parent; |
8985 | } while (ca); | |
8986 | rcu_read_unlock(); | |
8987 | } | |
8988 | ||
d842de87 SV |
8989 | struct cgroup_subsys cpuacct_subsys = { |
8990 | .name = "cpuacct", | |
8991 | .create = cpuacct_create, | |
8992 | .destroy = cpuacct_destroy, | |
8993 | .populate = cpuacct_populate, | |
8994 | .subsys_id = cpuacct_subsys_id, | |
8995 | }; | |
8996 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
8997 | |
8998 | #ifndef CONFIG_SMP | |
8999 | ||
9000 | int rcu_expedited_torture_stats(char *page) | |
9001 | { | |
9002 | return 0; | |
9003 | } | |
9004 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9005 | ||
9006 | void synchronize_sched_expedited(void) | |
9007 | { | |
9008 | } | |
9009 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9010 | ||
9011 | #else /* #ifndef CONFIG_SMP */ | |
9012 | ||
9013 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
9014 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
9015 | ||
9016 | #define RCU_EXPEDITED_STATE_POST -2 | |
9017 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
9018 | ||
9019 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
9020 | ||
9021 | int rcu_expedited_torture_stats(char *page) | |
9022 | { | |
9023 | int cnt = 0; | |
9024 | int cpu; | |
9025 | ||
9026 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
9027 | for_each_online_cpu(cpu) { | |
9028 | cnt += sprintf(&page[cnt], " %d:%d", | |
9029 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
9030 | } | |
9031 | cnt += sprintf(&page[cnt], "\n"); | |
9032 | return cnt; | |
9033 | } | |
9034 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9035 | ||
9036 | static long synchronize_sched_expedited_count; | |
9037 | ||
9038 | /* | |
9039 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9040 | * approach to force grace period to end quickly. This consumes | |
9041 | * significant time on all CPUs, and is thus not recommended for | |
9042 | * any sort of common-case code. | |
9043 | * | |
9044 | * Note that it is illegal to call this function while holding any | |
9045 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9046 | * observe this restriction will result in deadlock. | |
9047 | */ | |
9048 | void synchronize_sched_expedited(void) | |
9049 | { | |
9050 | int cpu; | |
9051 | unsigned long flags; | |
9052 | bool need_full_sync = 0; | |
9053 | struct rq *rq; | |
9054 | struct migration_req *req; | |
9055 | long snap; | |
9056 | int trycount = 0; | |
9057 | ||
9058 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
9059 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
9060 | get_online_cpus(); | |
9061 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
9062 | put_online_cpus(); | |
9063 | if (trycount++ < 10) | |
9064 | udelay(trycount * num_online_cpus()); | |
9065 | else { | |
9066 | synchronize_sched(); | |
9067 | return; | |
9068 | } | |
9069 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
9070 | smp_mb(); /* ensure test happens before caller kfree */ | |
9071 | return; | |
9072 | } | |
9073 | get_online_cpus(); | |
9074 | } | |
9075 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
9076 | for_each_online_cpu(cpu) { | |
9077 | rq = cpu_rq(cpu); | |
9078 | req = &per_cpu(rcu_migration_req, cpu); | |
9079 | init_completion(&req->done); | |
9080 | req->task = NULL; | |
9081 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
05fa785c | 9082 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf | 9083 | list_add(&req->list, &rq->migration_queue); |
05fa785c | 9084 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9085 | wake_up_process(rq->migration_thread); |
9086 | } | |
9087 | for_each_online_cpu(cpu) { | |
9088 | rcu_expedited_state = cpu; | |
9089 | req = &per_cpu(rcu_migration_req, cpu); | |
9090 | rq = cpu_rq(cpu); | |
9091 | wait_for_completion(&req->done); | |
05fa785c | 9092 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf PM |
9093 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) |
9094 | need_full_sync = 1; | |
9095 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
05fa785c | 9096 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9097 | } |
9098 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
956539b7 | 9099 | synchronize_sched_expedited_count++; |
03b042bf PM |
9100 | mutex_unlock(&rcu_sched_expedited_mutex); |
9101 | put_online_cpus(); | |
9102 | if (need_full_sync) | |
9103 | synchronize_sched(); | |
9104 | } | |
9105 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9106 | ||
9107 | #endif /* #else #ifndef CONFIG_SMP */ |