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> |
1da177e4 | 74 | |
5517d86b | 75 | #include <asm/tlb.h> |
838225b4 | 76 | #include <asm/irq_regs.h> |
1da177e4 | 77 | |
6e0534f2 GH |
78 | #include "sched_cpupri.h" |
79 | ||
a8d154b0 | 80 | #define CREATE_TRACE_POINTS |
ad8d75ff | 81 | #include <trace/events/sched.h> |
a8d154b0 | 82 | |
1da177e4 LT |
83 | /* |
84 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
85 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
86 | * and back. | |
87 | */ | |
88 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
89 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
90 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
91 | ||
92 | /* | |
93 | * 'User priority' is the nice value converted to something we | |
94 | * can work with better when scaling various scheduler parameters, | |
95 | * it's a [ 0 ... 39 ] range. | |
96 | */ | |
97 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
98 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
99 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
100 | ||
101 | /* | |
d7876a08 | 102 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 103 | */ |
d6322faf | 104 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 105 | |
6aa645ea IM |
106 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
107 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
108 | ||
1da177e4 LT |
109 | /* |
110 | * These are the 'tuning knobs' of the scheduler: | |
111 | * | |
a4ec24b4 | 112 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
113 | * Timeslices get refilled after they expire. |
114 | */ | |
1da177e4 | 115 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 116 | |
d0b27fa7 PZ |
117 | /* |
118 | * single value that denotes runtime == period, ie unlimited time. | |
119 | */ | |
120 | #define RUNTIME_INF ((u64)~0ULL) | |
121 | ||
e05606d3 IM |
122 | static inline int rt_policy(int policy) |
123 | { | |
3f33a7ce | 124 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
125 | return 1; |
126 | return 0; | |
127 | } | |
128 | ||
129 | static inline int task_has_rt_policy(struct task_struct *p) | |
130 | { | |
131 | return rt_policy(p->policy); | |
132 | } | |
133 | ||
1da177e4 | 134 | /* |
6aa645ea | 135 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 136 | */ |
6aa645ea IM |
137 | struct rt_prio_array { |
138 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
139 | struct list_head queue[MAX_RT_PRIO]; | |
140 | }; | |
141 | ||
d0b27fa7 | 142 | struct rt_bandwidth { |
ea736ed5 | 143 | /* nests inside the rq lock: */ |
0986b11b | 144 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
145 | ktime_t rt_period; |
146 | u64 rt_runtime; | |
147 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
148 | }; |
149 | ||
150 | static struct rt_bandwidth def_rt_bandwidth; | |
151 | ||
152 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
153 | ||
154 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
155 | { | |
156 | struct rt_bandwidth *rt_b = | |
157 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
158 | ktime_t now; | |
159 | int overrun; | |
160 | int idle = 0; | |
161 | ||
162 | for (;;) { | |
163 | now = hrtimer_cb_get_time(timer); | |
164 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
165 | ||
166 | if (!overrun) | |
167 | break; | |
168 | ||
169 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
170 | } | |
171 | ||
172 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
173 | } | |
174 | ||
175 | static | |
176 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
177 | { | |
178 | rt_b->rt_period = ns_to_ktime(period); | |
179 | rt_b->rt_runtime = runtime; | |
180 | ||
0986b11b | 181 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 182 | |
d0b27fa7 PZ |
183 | hrtimer_init(&rt_b->rt_period_timer, |
184 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
185 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
186 | } |
187 | ||
c8bfff6d KH |
188 | static inline int rt_bandwidth_enabled(void) |
189 | { | |
190 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
191 | } |
192 | ||
193 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
194 | { | |
195 | ktime_t now; | |
196 | ||
cac64d00 | 197 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
198 | return; |
199 | ||
200 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
201 | return; | |
202 | ||
0986b11b | 203 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 204 | for (;;) { |
7f1e2ca9 PZ |
205 | unsigned long delta; |
206 | ktime_t soft, hard; | |
207 | ||
d0b27fa7 PZ |
208 | if (hrtimer_active(&rt_b->rt_period_timer)) |
209 | break; | |
210 | ||
211 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
212 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
213 | |
214 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
215 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
216 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
217 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 218 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 219 | } |
0986b11b | 220 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
221 | } |
222 | ||
223 | #ifdef CONFIG_RT_GROUP_SCHED | |
224 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
225 | { | |
226 | hrtimer_cancel(&rt_b->rt_period_timer); | |
227 | } | |
228 | #endif | |
229 | ||
712555ee HC |
230 | /* |
231 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
232 | * detach_destroy_domains and partition_sched_domains. | |
233 | */ | |
234 | static DEFINE_MUTEX(sched_domains_mutex); | |
235 | ||
7c941438 | 236 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 237 | |
68318b8e SV |
238 | #include <linux/cgroup.h> |
239 | ||
29f59db3 SV |
240 | struct cfs_rq; |
241 | ||
6f505b16 PZ |
242 | static LIST_HEAD(task_groups); |
243 | ||
29f59db3 | 244 | /* task group related information */ |
4cf86d77 | 245 | struct task_group { |
68318b8e | 246 | struct cgroup_subsys_state css; |
6c415b92 | 247 | |
052f1dc7 | 248 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
249 | /* schedulable entities of this group on each cpu */ |
250 | struct sched_entity **se; | |
251 | /* runqueue "owned" by this group on each cpu */ | |
252 | struct cfs_rq **cfs_rq; | |
253 | unsigned long shares; | |
052f1dc7 PZ |
254 | #endif |
255 | ||
256 | #ifdef CONFIG_RT_GROUP_SCHED | |
257 | struct sched_rt_entity **rt_se; | |
258 | struct rt_rq **rt_rq; | |
259 | ||
d0b27fa7 | 260 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 261 | #endif |
6b2d7700 | 262 | |
ae8393e5 | 263 | struct rcu_head rcu; |
6f505b16 | 264 | struct list_head list; |
f473aa5e PZ |
265 | |
266 | struct task_group *parent; | |
267 | struct list_head siblings; | |
268 | struct list_head children; | |
29f59db3 SV |
269 | }; |
270 | ||
eff766a6 | 271 | #define root_task_group init_task_group |
6f505b16 | 272 | |
8ed36996 | 273 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
274 | * a task group's cpu shares. |
275 | */ | |
8ed36996 | 276 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 277 | |
e9036b36 CG |
278 | #ifdef CONFIG_FAIR_GROUP_SCHED |
279 | ||
57310a98 PZ |
280 | #ifdef CONFIG_SMP |
281 | static int root_task_group_empty(void) | |
282 | { | |
283 | return list_empty(&root_task_group.children); | |
284 | } | |
285 | #endif | |
286 | ||
052f1dc7 | 287 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 288 | |
cb4ad1ff | 289 | /* |
2e084786 LJ |
290 | * A weight of 0 or 1 can cause arithmetics problems. |
291 | * A weight of a cfs_rq is the sum of weights of which entities | |
292 | * are queued on this cfs_rq, so a weight of a entity should not be | |
293 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
294 | * (The default weight is 1024 - so there's no practical |
295 | * limitation from this.) | |
296 | */ | |
18d95a28 | 297 | #define MIN_SHARES 2 |
2e084786 | 298 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 299 | |
052f1dc7 PZ |
300 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
301 | #endif | |
302 | ||
29f59db3 | 303 | /* Default task group. |
3a252015 | 304 | * Every task in system belong to this group at bootup. |
29f59db3 | 305 | */ |
434d53b0 | 306 | struct task_group init_task_group; |
29f59db3 SV |
307 | |
308 | /* return group to which a task belongs */ | |
4cf86d77 | 309 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 310 | { |
4cf86d77 | 311 | struct task_group *tg; |
9b5b7751 | 312 | |
7c941438 | 313 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
314 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
315 | struct task_group, css); | |
24e377a8 | 316 | #else |
41a2d6cf | 317 | tg = &init_task_group; |
24e377a8 | 318 | #endif |
9b5b7751 | 319 | return tg; |
29f59db3 SV |
320 | } |
321 | ||
322 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 323 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 324 | { |
052f1dc7 | 325 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
326 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
327 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 328 | #endif |
6f505b16 | 329 | |
052f1dc7 | 330 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
331 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
332 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 333 | #endif |
29f59db3 SV |
334 | } |
335 | ||
336 | #else | |
337 | ||
6f505b16 | 338 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
339 | static inline struct task_group *task_group(struct task_struct *p) |
340 | { | |
341 | return NULL; | |
342 | } | |
29f59db3 | 343 | |
7c941438 | 344 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 345 | |
6aa645ea IM |
346 | /* CFS-related fields in a runqueue */ |
347 | struct cfs_rq { | |
348 | struct load_weight load; | |
349 | unsigned long nr_running; | |
350 | ||
6aa645ea | 351 | u64 exec_clock; |
e9acbff6 | 352 | u64 min_vruntime; |
6aa645ea IM |
353 | |
354 | struct rb_root tasks_timeline; | |
355 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
356 | |
357 | struct list_head tasks; | |
358 | struct list_head *balance_iterator; | |
359 | ||
360 | /* | |
361 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
362 | * It is set to NULL otherwise (i.e when none are currently running). |
363 | */ | |
4793241b | 364 | struct sched_entity *curr, *next, *last; |
ddc97297 | 365 | |
5ac5c4d6 | 366 | unsigned int nr_spread_over; |
ddc97297 | 367 | |
62160e3f | 368 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
369 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
370 | ||
41a2d6cf IM |
371 | /* |
372 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
373 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
374 | * (like users, containers etc.) | |
375 | * | |
376 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
377 | * list is used during load balance. | |
378 | */ | |
41a2d6cf IM |
379 | struct list_head leaf_cfs_rq_list; |
380 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
381 | |
382 | #ifdef CONFIG_SMP | |
c09595f6 | 383 | /* |
c8cba857 | 384 | * the part of load.weight contributed by tasks |
c09595f6 | 385 | */ |
c8cba857 | 386 | unsigned long task_weight; |
c09595f6 | 387 | |
c8cba857 PZ |
388 | /* |
389 | * h_load = weight * f(tg) | |
390 | * | |
391 | * Where f(tg) is the recursive weight fraction assigned to | |
392 | * this group. | |
393 | */ | |
394 | unsigned long h_load; | |
c09595f6 | 395 | |
c8cba857 PZ |
396 | /* |
397 | * this cpu's part of tg->shares | |
398 | */ | |
399 | unsigned long shares; | |
f1d239f7 PZ |
400 | |
401 | /* | |
402 | * load.weight at the time we set shares | |
403 | */ | |
404 | unsigned long rq_weight; | |
c09595f6 | 405 | #endif |
6aa645ea IM |
406 | #endif |
407 | }; | |
1da177e4 | 408 | |
6aa645ea IM |
409 | /* Real-Time classes' related field in a runqueue: */ |
410 | struct rt_rq { | |
411 | struct rt_prio_array active; | |
63489e45 | 412 | unsigned long rt_nr_running; |
052f1dc7 | 413 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
414 | struct { |
415 | int curr; /* highest queued rt task prio */ | |
398a153b | 416 | #ifdef CONFIG_SMP |
e864c499 | 417 | int next; /* next highest */ |
398a153b | 418 | #endif |
e864c499 | 419 | } highest_prio; |
6f505b16 | 420 | #endif |
fa85ae24 | 421 | #ifdef CONFIG_SMP |
73fe6aae | 422 | unsigned long rt_nr_migratory; |
a1ba4d8b | 423 | unsigned long rt_nr_total; |
a22d7fc1 | 424 | int overloaded; |
917b627d | 425 | struct plist_head pushable_tasks; |
fa85ae24 | 426 | #endif |
6f505b16 | 427 | int rt_throttled; |
fa85ae24 | 428 | u64 rt_time; |
ac086bc2 | 429 | u64 rt_runtime; |
ea736ed5 | 430 | /* Nests inside the rq lock: */ |
0986b11b | 431 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 432 | |
052f1dc7 | 433 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
434 | unsigned long rt_nr_boosted; |
435 | ||
6f505b16 PZ |
436 | struct rq *rq; |
437 | struct list_head leaf_rt_rq_list; | |
438 | struct task_group *tg; | |
439 | struct sched_rt_entity *rt_se; | |
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 SS |
495 | #ifdef CONFIG_NO_HZ |
496 | unsigned char in_nohz_recently; | |
497 | #endif | |
d8016491 IM |
498 | /* capture load from *all* tasks on this cpu: */ |
499 | struct load_weight load; | |
6aa645ea IM |
500 | unsigned long nr_load_updates; |
501 | u64 nr_switches; | |
502 | ||
503 | struct cfs_rq cfs; | |
6f505b16 | 504 | struct rt_rq rt; |
6f505b16 | 505 | |
6aa645ea | 506 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
507 | /* list of leaf cfs_rq on this cpu: */ |
508 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
509 | #endif |
510 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 511 | struct list_head leaf_rt_rq_list; |
1da177e4 | 512 | #endif |
1da177e4 LT |
513 | |
514 | /* | |
515 | * This is part of a global counter where only the total sum | |
516 | * over all CPUs matters. A task can increase this counter on | |
517 | * one CPU and if it got migrated afterwards it may decrease | |
518 | * it on another CPU. Always updated under the runqueue lock: | |
519 | */ | |
520 | unsigned long nr_uninterruptible; | |
521 | ||
36c8b586 | 522 | struct task_struct *curr, *idle; |
c9819f45 | 523 | unsigned long next_balance; |
1da177e4 | 524 | struct mm_struct *prev_mm; |
6aa645ea | 525 | |
3e51f33f | 526 | u64 clock; |
6aa645ea | 527 | |
1da177e4 LT |
528 | atomic_t nr_iowait; |
529 | ||
530 | #ifdef CONFIG_SMP | |
0eab9146 | 531 | struct root_domain *rd; |
1da177e4 LT |
532 | struct sched_domain *sd; |
533 | ||
a0a522ce | 534 | unsigned char idle_at_tick; |
1da177e4 | 535 | /* For active balancing */ |
3f029d3c | 536 | int post_schedule; |
1da177e4 LT |
537 | int active_balance; |
538 | int push_cpu; | |
d8016491 IM |
539 | /* cpu of this runqueue: */ |
540 | int cpu; | |
1f11eb6a | 541 | int online; |
1da177e4 | 542 | |
a8a51d5e | 543 | unsigned long avg_load_per_task; |
1da177e4 | 544 | |
36c8b586 | 545 | struct task_struct *migration_thread; |
1da177e4 | 546 | struct list_head migration_queue; |
e9e9250b PZ |
547 | |
548 | u64 rt_avg; | |
549 | u64 age_stamp; | |
1b9508f6 MG |
550 | u64 idle_stamp; |
551 | u64 avg_idle; | |
1da177e4 LT |
552 | #endif |
553 | ||
dce48a84 TG |
554 | /* calc_load related fields */ |
555 | unsigned long calc_load_update; | |
556 | long calc_load_active; | |
557 | ||
8f4d37ec | 558 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
559 | #ifdef CONFIG_SMP |
560 | int hrtick_csd_pending; | |
561 | struct call_single_data hrtick_csd; | |
562 | #endif | |
8f4d37ec PZ |
563 | struct hrtimer hrtick_timer; |
564 | #endif | |
565 | ||
1da177e4 LT |
566 | #ifdef CONFIG_SCHEDSTATS |
567 | /* latency stats */ | |
568 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
569 | unsigned long long rq_cpu_time; |
570 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
571 | |
572 | /* sys_sched_yield() stats */ | |
480b9434 | 573 | unsigned int yld_count; |
1da177e4 LT |
574 | |
575 | /* schedule() stats */ | |
480b9434 KC |
576 | unsigned int sched_switch; |
577 | unsigned int sched_count; | |
578 | unsigned int sched_goidle; | |
1da177e4 LT |
579 | |
580 | /* try_to_wake_up() stats */ | |
480b9434 KC |
581 | unsigned int ttwu_count; |
582 | unsigned int ttwu_local; | |
b8efb561 IM |
583 | |
584 | /* BKL stats */ | |
480b9434 | 585 | unsigned int bkl_count; |
1da177e4 LT |
586 | #endif |
587 | }; | |
588 | ||
f34e3b61 | 589 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 590 | |
7d478721 PZ |
591 | static inline |
592 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 593 | { |
7d478721 | 594 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
dd41f596 IM |
595 | } |
596 | ||
0a2966b4 CL |
597 | static inline int cpu_of(struct rq *rq) |
598 | { | |
599 | #ifdef CONFIG_SMP | |
600 | return rq->cpu; | |
601 | #else | |
602 | return 0; | |
603 | #endif | |
604 | } | |
605 | ||
674311d5 NP |
606 | /* |
607 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 608 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
609 | * |
610 | * The domain tree of any CPU may only be accessed from within | |
611 | * preempt-disabled sections. | |
612 | */ | |
48f24c4d IM |
613 | #define for_each_domain(cpu, __sd) \ |
614 | for (__sd = rcu_dereference(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) | |
1da177e4 LT |
615 | |
616 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
617 | #define this_rq() (&__get_cpu_var(runqueues)) | |
618 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
619 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 620 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 621 | |
aa9c4c0f | 622 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
623 | { |
624 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
625 | } | |
626 | ||
bf5c91ba IM |
627 | /* |
628 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
629 | */ | |
630 | #ifdef CONFIG_SCHED_DEBUG | |
631 | # define const_debug __read_mostly | |
632 | #else | |
633 | # define const_debug static const | |
634 | #endif | |
635 | ||
017730c1 IM |
636 | /** |
637 | * runqueue_is_locked | |
e17b38bf | 638 | * @cpu: the processor in question. |
017730c1 IM |
639 | * |
640 | * Returns true if the current cpu runqueue is locked. | |
641 | * This interface allows printk to be called with the runqueue lock | |
642 | * held and know whether or not it is OK to wake up the klogd. | |
643 | */ | |
89f19f04 | 644 | int runqueue_is_locked(int cpu) |
017730c1 | 645 | { |
05fa785c | 646 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
647 | } |
648 | ||
bf5c91ba IM |
649 | /* |
650 | * Debugging: various feature bits | |
651 | */ | |
f00b45c1 PZ |
652 | |
653 | #define SCHED_FEAT(name, enabled) \ | |
654 | __SCHED_FEAT_##name , | |
655 | ||
bf5c91ba | 656 | enum { |
f00b45c1 | 657 | #include "sched_features.h" |
bf5c91ba IM |
658 | }; |
659 | ||
f00b45c1 PZ |
660 | #undef SCHED_FEAT |
661 | ||
662 | #define SCHED_FEAT(name, enabled) \ | |
663 | (1UL << __SCHED_FEAT_##name) * enabled | | |
664 | ||
bf5c91ba | 665 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
666 | #include "sched_features.h" |
667 | 0; | |
668 | ||
669 | #undef SCHED_FEAT | |
670 | ||
671 | #ifdef CONFIG_SCHED_DEBUG | |
672 | #define SCHED_FEAT(name, enabled) \ | |
673 | #name , | |
674 | ||
983ed7a6 | 675 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
676 | #include "sched_features.h" |
677 | NULL | |
678 | }; | |
679 | ||
680 | #undef SCHED_FEAT | |
681 | ||
34f3a814 | 682 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 683 | { |
f00b45c1 PZ |
684 | int i; |
685 | ||
686 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
687 | if (!(sysctl_sched_features & (1UL << i))) |
688 | seq_puts(m, "NO_"); | |
689 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 690 | } |
34f3a814 | 691 | seq_puts(m, "\n"); |
f00b45c1 | 692 | |
34f3a814 | 693 | return 0; |
f00b45c1 PZ |
694 | } |
695 | ||
696 | static ssize_t | |
697 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
698 | size_t cnt, loff_t *ppos) | |
699 | { | |
700 | char buf[64]; | |
701 | char *cmp = buf; | |
702 | int neg = 0; | |
703 | int i; | |
704 | ||
705 | if (cnt > 63) | |
706 | cnt = 63; | |
707 | ||
708 | if (copy_from_user(&buf, ubuf, cnt)) | |
709 | return -EFAULT; | |
710 | ||
711 | buf[cnt] = 0; | |
712 | ||
c24b7c52 | 713 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
714 | neg = 1; |
715 | cmp += 3; | |
716 | } | |
717 | ||
718 | for (i = 0; sched_feat_names[i]; i++) { | |
719 | int len = strlen(sched_feat_names[i]); | |
720 | ||
721 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
722 | if (neg) | |
723 | sysctl_sched_features &= ~(1UL << i); | |
724 | else | |
725 | sysctl_sched_features |= (1UL << i); | |
726 | break; | |
727 | } | |
728 | } | |
729 | ||
730 | if (!sched_feat_names[i]) | |
731 | return -EINVAL; | |
732 | ||
42994724 | 733 | *ppos += cnt; |
f00b45c1 PZ |
734 | |
735 | return cnt; | |
736 | } | |
737 | ||
34f3a814 LZ |
738 | static int sched_feat_open(struct inode *inode, struct file *filp) |
739 | { | |
740 | return single_open(filp, sched_feat_show, NULL); | |
741 | } | |
742 | ||
828c0950 | 743 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
744 | .open = sched_feat_open, |
745 | .write = sched_feat_write, | |
746 | .read = seq_read, | |
747 | .llseek = seq_lseek, | |
748 | .release = single_release, | |
f00b45c1 PZ |
749 | }; |
750 | ||
751 | static __init int sched_init_debug(void) | |
752 | { | |
f00b45c1 PZ |
753 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
754 | &sched_feat_fops); | |
755 | ||
756 | return 0; | |
757 | } | |
758 | late_initcall(sched_init_debug); | |
759 | ||
760 | #endif | |
761 | ||
762 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 763 | |
b82d9fdd PZ |
764 | /* |
765 | * Number of tasks to iterate in a single balance run. | |
766 | * Limited because this is done with IRQs disabled. | |
767 | */ | |
768 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
769 | ||
2398f2c6 PZ |
770 | /* |
771 | * ratelimit for updating the group shares. | |
55cd5340 | 772 | * default: 0.25ms |
2398f2c6 | 773 | */ |
55cd5340 | 774 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 775 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 776 | |
ffda12a1 PZ |
777 | /* |
778 | * Inject some fuzzyness into changing the per-cpu group shares | |
779 | * this avoids remote rq-locks at the expense of fairness. | |
780 | * default: 4 | |
781 | */ | |
782 | unsigned int sysctl_sched_shares_thresh = 4; | |
783 | ||
e9e9250b PZ |
784 | /* |
785 | * period over which we average the RT time consumption, measured | |
786 | * in ms. | |
787 | * | |
788 | * default: 1s | |
789 | */ | |
790 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
791 | ||
fa85ae24 | 792 | /* |
9f0c1e56 | 793 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
794 | * default: 1s |
795 | */ | |
9f0c1e56 | 796 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 797 | |
6892b75e IM |
798 | static __read_mostly int scheduler_running; |
799 | ||
9f0c1e56 PZ |
800 | /* |
801 | * part of the period that we allow rt tasks to run in us. | |
802 | * default: 0.95s | |
803 | */ | |
804 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 805 | |
d0b27fa7 PZ |
806 | static inline u64 global_rt_period(void) |
807 | { | |
808 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
809 | } | |
810 | ||
811 | static inline u64 global_rt_runtime(void) | |
812 | { | |
e26873bb | 813 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
814 | return RUNTIME_INF; |
815 | ||
816 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
817 | } | |
fa85ae24 | 818 | |
1da177e4 | 819 | #ifndef prepare_arch_switch |
4866cde0 NP |
820 | # define prepare_arch_switch(next) do { } while (0) |
821 | #endif | |
822 | #ifndef finish_arch_switch | |
823 | # define finish_arch_switch(prev) do { } while (0) | |
824 | #endif | |
825 | ||
051a1d1a DA |
826 | static inline int task_current(struct rq *rq, struct task_struct *p) |
827 | { | |
828 | return rq->curr == p; | |
829 | } | |
830 | ||
4866cde0 | 831 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 832 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 833 | { |
051a1d1a | 834 | return task_current(rq, p); |
4866cde0 NP |
835 | } |
836 | ||
70b97a7f | 837 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
838 | { |
839 | } | |
840 | ||
70b97a7f | 841 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 842 | { |
da04c035 IM |
843 | #ifdef CONFIG_DEBUG_SPINLOCK |
844 | /* this is a valid case when another task releases the spinlock */ | |
845 | rq->lock.owner = current; | |
846 | #endif | |
8a25d5de IM |
847 | /* |
848 | * If we are tracking spinlock dependencies then we have to | |
849 | * fix up the runqueue lock - which gets 'carried over' from | |
850 | * prev into current: | |
851 | */ | |
852 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
853 | ||
05fa785c | 854 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
855 | } |
856 | ||
857 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 858 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
859 | { |
860 | #ifdef CONFIG_SMP | |
861 | return p->oncpu; | |
862 | #else | |
051a1d1a | 863 | return task_current(rq, p); |
4866cde0 NP |
864 | #endif |
865 | } | |
866 | ||
70b97a7f | 867 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
868 | { |
869 | #ifdef CONFIG_SMP | |
870 | /* | |
871 | * We can optimise this out completely for !SMP, because the | |
872 | * SMP rebalancing from interrupt is the only thing that cares | |
873 | * here. | |
874 | */ | |
875 | next->oncpu = 1; | |
876 | #endif | |
877 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 878 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 879 | #else |
05fa785c | 880 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
881 | #endif |
882 | } | |
883 | ||
70b97a7f | 884 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
885 | { |
886 | #ifdef CONFIG_SMP | |
887 | /* | |
888 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
889 | * We must ensure this doesn't happen until the switch is completely | |
890 | * finished. | |
891 | */ | |
892 | smp_wmb(); | |
893 | prev->oncpu = 0; | |
894 | #endif | |
895 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
896 | local_irq_enable(); | |
1da177e4 | 897 | #endif |
4866cde0 NP |
898 | } |
899 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 900 | |
b29739f9 IM |
901 | /* |
902 | * __task_rq_lock - lock the runqueue a given task resides on. | |
903 | * Must be called interrupts disabled. | |
904 | */ | |
70b97a7f | 905 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
906 | __acquires(rq->lock) |
907 | { | |
3a5c359a AK |
908 | for (;;) { |
909 | struct rq *rq = task_rq(p); | |
05fa785c | 910 | raw_spin_lock(&rq->lock); |
3a5c359a AK |
911 | if (likely(rq == task_rq(p))) |
912 | return rq; | |
05fa785c | 913 | raw_spin_unlock(&rq->lock); |
b29739f9 | 914 | } |
b29739f9 IM |
915 | } |
916 | ||
1da177e4 LT |
917 | /* |
918 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 919 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
920 | * explicitly disabling preemption. |
921 | */ | |
70b97a7f | 922 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
923 | __acquires(rq->lock) |
924 | { | |
70b97a7f | 925 | struct rq *rq; |
1da177e4 | 926 | |
3a5c359a AK |
927 | for (;;) { |
928 | local_irq_save(*flags); | |
929 | rq = task_rq(p); | |
05fa785c | 930 | raw_spin_lock(&rq->lock); |
3a5c359a AK |
931 | if (likely(rq == task_rq(p))) |
932 | return rq; | |
05fa785c | 933 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 934 | } |
1da177e4 LT |
935 | } |
936 | ||
ad474cac ON |
937 | void task_rq_unlock_wait(struct task_struct *p) |
938 | { | |
939 | struct rq *rq = task_rq(p); | |
940 | ||
941 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
05fa785c | 942 | raw_spin_unlock_wait(&rq->lock); |
ad474cac ON |
943 | } |
944 | ||
a9957449 | 945 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
946 | __releases(rq->lock) |
947 | { | |
05fa785c | 948 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
949 | } |
950 | ||
70b97a7f | 951 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
952 | __releases(rq->lock) |
953 | { | |
05fa785c | 954 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
955 | } |
956 | ||
1da177e4 | 957 | /* |
cc2a73b5 | 958 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 959 | */ |
a9957449 | 960 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
961 | __acquires(rq->lock) |
962 | { | |
70b97a7f | 963 | struct rq *rq; |
1da177e4 LT |
964 | |
965 | local_irq_disable(); | |
966 | rq = this_rq(); | |
05fa785c | 967 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
968 | |
969 | return rq; | |
970 | } | |
971 | ||
8f4d37ec PZ |
972 | #ifdef CONFIG_SCHED_HRTICK |
973 | /* | |
974 | * Use HR-timers to deliver accurate preemption points. | |
975 | * | |
976 | * Its all a bit involved since we cannot program an hrt while holding the | |
977 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
978 | * reschedule event. | |
979 | * | |
980 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
981 | * rq->lock. | |
982 | */ | |
8f4d37ec PZ |
983 | |
984 | /* | |
985 | * Use hrtick when: | |
986 | * - enabled by features | |
987 | * - hrtimer is actually high res | |
988 | */ | |
989 | static inline int hrtick_enabled(struct rq *rq) | |
990 | { | |
991 | if (!sched_feat(HRTICK)) | |
992 | return 0; | |
ba42059f | 993 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 994 | return 0; |
8f4d37ec PZ |
995 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
996 | } | |
997 | ||
8f4d37ec PZ |
998 | static void hrtick_clear(struct rq *rq) |
999 | { | |
1000 | if (hrtimer_active(&rq->hrtick_timer)) | |
1001 | hrtimer_cancel(&rq->hrtick_timer); | |
1002 | } | |
1003 | ||
8f4d37ec PZ |
1004 | /* |
1005 | * High-resolution timer tick. | |
1006 | * Runs from hardirq context with interrupts disabled. | |
1007 | */ | |
1008 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1009 | { | |
1010 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1011 | ||
1012 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1013 | ||
05fa785c | 1014 | raw_spin_lock(&rq->lock); |
3e51f33f | 1015 | update_rq_clock(rq); |
8f4d37ec | 1016 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1017 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1018 | |
1019 | return HRTIMER_NORESTART; | |
1020 | } | |
1021 | ||
95e904c7 | 1022 | #ifdef CONFIG_SMP |
31656519 PZ |
1023 | /* |
1024 | * called from hardirq (IPI) context | |
1025 | */ | |
1026 | static void __hrtick_start(void *arg) | |
b328ca18 | 1027 | { |
31656519 | 1028 | struct rq *rq = arg; |
b328ca18 | 1029 | |
05fa785c | 1030 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1031 | hrtimer_restart(&rq->hrtick_timer); |
1032 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1033 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1034 | } |
1035 | ||
31656519 PZ |
1036 | /* |
1037 | * Called to set the hrtick timer state. | |
1038 | * | |
1039 | * called with rq->lock held and irqs disabled | |
1040 | */ | |
1041 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1042 | { |
31656519 PZ |
1043 | struct hrtimer *timer = &rq->hrtick_timer; |
1044 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1045 | |
cc584b21 | 1046 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1047 | |
1048 | if (rq == this_rq()) { | |
1049 | hrtimer_restart(timer); | |
1050 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1051 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1052 | rq->hrtick_csd_pending = 1; |
1053 | } | |
b328ca18 PZ |
1054 | } |
1055 | ||
1056 | static int | |
1057 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1058 | { | |
1059 | int cpu = (int)(long)hcpu; | |
1060 | ||
1061 | switch (action) { | |
1062 | case CPU_UP_CANCELED: | |
1063 | case CPU_UP_CANCELED_FROZEN: | |
1064 | case CPU_DOWN_PREPARE: | |
1065 | case CPU_DOWN_PREPARE_FROZEN: | |
1066 | case CPU_DEAD: | |
1067 | case CPU_DEAD_FROZEN: | |
31656519 | 1068 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1069 | return NOTIFY_OK; |
1070 | } | |
1071 | ||
1072 | return NOTIFY_DONE; | |
1073 | } | |
1074 | ||
fa748203 | 1075 | static __init void init_hrtick(void) |
b328ca18 PZ |
1076 | { |
1077 | hotcpu_notifier(hotplug_hrtick, 0); | |
1078 | } | |
31656519 PZ |
1079 | #else |
1080 | /* | |
1081 | * Called to set the hrtick timer state. | |
1082 | * | |
1083 | * called with rq->lock held and irqs disabled | |
1084 | */ | |
1085 | static void hrtick_start(struct rq *rq, u64 delay) | |
1086 | { | |
7f1e2ca9 | 1087 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1088 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1089 | } |
b328ca18 | 1090 | |
006c75f1 | 1091 | static inline void init_hrtick(void) |
8f4d37ec | 1092 | { |
8f4d37ec | 1093 | } |
31656519 | 1094 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1095 | |
31656519 | 1096 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1097 | { |
31656519 PZ |
1098 | #ifdef CONFIG_SMP |
1099 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1100 | |
31656519 PZ |
1101 | rq->hrtick_csd.flags = 0; |
1102 | rq->hrtick_csd.func = __hrtick_start; | |
1103 | rq->hrtick_csd.info = rq; | |
1104 | #endif | |
8f4d37ec | 1105 | |
31656519 PZ |
1106 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1107 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1108 | } |
006c75f1 | 1109 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1110 | static inline void hrtick_clear(struct rq *rq) |
1111 | { | |
1112 | } | |
1113 | ||
8f4d37ec PZ |
1114 | static inline void init_rq_hrtick(struct rq *rq) |
1115 | { | |
1116 | } | |
1117 | ||
b328ca18 PZ |
1118 | static inline void init_hrtick(void) |
1119 | { | |
1120 | } | |
006c75f1 | 1121 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1122 | |
c24d20db IM |
1123 | /* |
1124 | * resched_task - mark a task 'to be rescheduled now'. | |
1125 | * | |
1126 | * On UP this means the setting of the need_resched flag, on SMP it | |
1127 | * might also involve a cross-CPU call to trigger the scheduler on | |
1128 | * the target CPU. | |
1129 | */ | |
1130 | #ifdef CONFIG_SMP | |
1131 | ||
1132 | #ifndef tsk_is_polling | |
1133 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1134 | #endif | |
1135 | ||
31656519 | 1136 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1137 | { |
1138 | int cpu; | |
1139 | ||
05fa785c | 1140 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1141 | |
5ed0cec0 | 1142 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1143 | return; |
1144 | ||
5ed0cec0 | 1145 | set_tsk_need_resched(p); |
c24d20db IM |
1146 | |
1147 | cpu = task_cpu(p); | |
1148 | if (cpu == smp_processor_id()) | |
1149 | return; | |
1150 | ||
1151 | /* NEED_RESCHED must be visible before we test polling */ | |
1152 | smp_mb(); | |
1153 | if (!tsk_is_polling(p)) | |
1154 | smp_send_reschedule(cpu); | |
1155 | } | |
1156 | ||
1157 | static void resched_cpu(int cpu) | |
1158 | { | |
1159 | struct rq *rq = cpu_rq(cpu); | |
1160 | unsigned long flags; | |
1161 | ||
05fa785c | 1162 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1163 | return; |
1164 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1165 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1166 | } |
06d8308c TG |
1167 | |
1168 | #ifdef CONFIG_NO_HZ | |
1169 | /* | |
1170 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1171 | * idle CPU then this timer might expire before the next timer event | |
1172 | * which is scheduled to wake up that CPU. In case of a completely | |
1173 | * idle system the next event might even be infinite time into the | |
1174 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1175 | * leaves the inner idle loop so the newly added timer is taken into | |
1176 | * account when the CPU goes back to idle and evaluates the timer | |
1177 | * wheel for the next timer event. | |
1178 | */ | |
1179 | void wake_up_idle_cpu(int cpu) | |
1180 | { | |
1181 | struct rq *rq = cpu_rq(cpu); | |
1182 | ||
1183 | if (cpu == smp_processor_id()) | |
1184 | return; | |
1185 | ||
1186 | /* | |
1187 | * This is safe, as this function is called with the timer | |
1188 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1189 | * to idle and has not yet set rq->curr to idle then it will | |
1190 | * be serialized on the timer wheel base lock and take the new | |
1191 | * timer into account automatically. | |
1192 | */ | |
1193 | if (rq->curr != rq->idle) | |
1194 | return; | |
1195 | ||
1196 | /* | |
1197 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1198 | * lockless. The worst case is that the other CPU runs the | |
1199 | * idle task through an additional NOOP schedule() | |
1200 | */ | |
5ed0cec0 | 1201 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1202 | |
1203 | /* NEED_RESCHED must be visible before we test polling */ | |
1204 | smp_mb(); | |
1205 | if (!tsk_is_polling(rq->idle)) | |
1206 | smp_send_reschedule(cpu); | |
1207 | } | |
6d6bc0ad | 1208 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1209 | |
e9e9250b PZ |
1210 | static u64 sched_avg_period(void) |
1211 | { | |
1212 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1213 | } | |
1214 | ||
1215 | static void sched_avg_update(struct rq *rq) | |
1216 | { | |
1217 | s64 period = sched_avg_period(); | |
1218 | ||
1219 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1220 | rq->age_stamp += period; | |
1221 | rq->rt_avg /= 2; | |
1222 | } | |
1223 | } | |
1224 | ||
1225 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1226 | { | |
1227 | rq->rt_avg += rt_delta; | |
1228 | sched_avg_update(rq); | |
1229 | } | |
1230 | ||
6d6bc0ad | 1231 | #else /* !CONFIG_SMP */ |
31656519 | 1232 | static void resched_task(struct task_struct *p) |
c24d20db | 1233 | { |
05fa785c | 1234 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1235 | set_tsk_need_resched(p); |
c24d20db | 1236 | } |
e9e9250b PZ |
1237 | |
1238 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1239 | { | |
1240 | } | |
6d6bc0ad | 1241 | #endif /* CONFIG_SMP */ |
c24d20db | 1242 | |
45bf76df IM |
1243 | #if BITS_PER_LONG == 32 |
1244 | # define WMULT_CONST (~0UL) | |
1245 | #else | |
1246 | # define WMULT_CONST (1UL << 32) | |
1247 | #endif | |
1248 | ||
1249 | #define WMULT_SHIFT 32 | |
1250 | ||
194081eb IM |
1251 | /* |
1252 | * Shift right and round: | |
1253 | */ | |
cf2ab469 | 1254 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1255 | |
a7be37ac PZ |
1256 | /* |
1257 | * delta *= weight / lw | |
1258 | */ | |
cb1c4fc9 | 1259 | static unsigned long |
45bf76df IM |
1260 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1261 | struct load_weight *lw) | |
1262 | { | |
1263 | u64 tmp; | |
1264 | ||
7a232e03 LJ |
1265 | if (!lw->inv_weight) { |
1266 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1267 | lw->inv_weight = 1; | |
1268 | else | |
1269 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1270 | / (lw->weight+1); | |
1271 | } | |
45bf76df IM |
1272 | |
1273 | tmp = (u64)delta_exec * weight; | |
1274 | /* | |
1275 | * Check whether we'd overflow the 64-bit multiplication: | |
1276 | */ | |
194081eb | 1277 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1278 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1279 | WMULT_SHIFT/2); |
1280 | else | |
cf2ab469 | 1281 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1282 | |
ecf691da | 1283 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1284 | } |
1285 | ||
1091985b | 1286 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1287 | { |
1288 | lw->weight += inc; | |
e89996ae | 1289 | lw->inv_weight = 0; |
45bf76df IM |
1290 | } |
1291 | ||
1091985b | 1292 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1293 | { |
1294 | lw->weight -= dec; | |
e89996ae | 1295 | lw->inv_weight = 0; |
45bf76df IM |
1296 | } |
1297 | ||
2dd73a4f PW |
1298 | /* |
1299 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1300 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1301 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1302 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1303 | * scaled version of the new time slice allocation that they receive on time |
1304 | * slice expiry etc. | |
1305 | */ | |
1306 | ||
cce7ade8 PZ |
1307 | #define WEIGHT_IDLEPRIO 3 |
1308 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1309 | |
1310 | /* | |
1311 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1312 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1313 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1314 | * that remained on nice 0. | |
1315 | * | |
1316 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1317 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1318 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1319 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1320 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1321 | */ |
1322 | static const int prio_to_weight[40] = { | |
254753dc IM |
1323 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1324 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1325 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1326 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1327 | /* 0 */ 1024, 820, 655, 526, 423, | |
1328 | /* 5 */ 335, 272, 215, 172, 137, | |
1329 | /* 10 */ 110, 87, 70, 56, 45, | |
1330 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1331 | }; |
1332 | ||
5714d2de IM |
1333 | /* |
1334 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1335 | * | |
1336 | * In cases where the weight does not change often, we can use the | |
1337 | * precalculated inverse to speed up arithmetics by turning divisions | |
1338 | * into multiplications: | |
1339 | */ | |
dd41f596 | 1340 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1341 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1342 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1343 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1344 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1345 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1346 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1347 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1348 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1349 | }; |
2dd73a4f | 1350 | |
ef12fefa BR |
1351 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1352 | enum cpuacct_stat_index { | |
1353 | CPUACCT_STAT_USER, /* ... user mode */ | |
1354 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1355 | ||
1356 | CPUACCT_STAT_NSTATS, | |
1357 | }; | |
1358 | ||
d842de87 SV |
1359 | #ifdef CONFIG_CGROUP_CPUACCT |
1360 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1361 | static void cpuacct_update_stats(struct task_struct *tsk, |
1362 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1363 | #else |
1364 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1365 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1366 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1367 | #endif |
1368 | ||
18d95a28 PZ |
1369 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1370 | { | |
1371 | update_load_add(&rq->load, load); | |
1372 | } | |
1373 | ||
1374 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1375 | { | |
1376 | update_load_sub(&rq->load, load); | |
1377 | } | |
1378 | ||
7940ca36 | 1379 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1380 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1381 | |
1382 | /* | |
1383 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1384 | * leaving it for the final time. | |
1385 | */ | |
eb755805 | 1386 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1387 | { |
1388 | struct task_group *parent, *child; | |
eb755805 | 1389 | int ret; |
c09595f6 PZ |
1390 | |
1391 | rcu_read_lock(); | |
1392 | parent = &root_task_group; | |
1393 | down: | |
eb755805 PZ |
1394 | ret = (*down)(parent, data); |
1395 | if (ret) | |
1396 | goto out_unlock; | |
c09595f6 PZ |
1397 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1398 | parent = child; | |
1399 | goto down; | |
1400 | ||
1401 | up: | |
1402 | continue; | |
1403 | } | |
eb755805 PZ |
1404 | ret = (*up)(parent, data); |
1405 | if (ret) | |
1406 | goto out_unlock; | |
c09595f6 PZ |
1407 | |
1408 | child = parent; | |
1409 | parent = parent->parent; | |
1410 | if (parent) | |
1411 | goto up; | |
eb755805 | 1412 | out_unlock: |
c09595f6 | 1413 | rcu_read_unlock(); |
eb755805 PZ |
1414 | |
1415 | return ret; | |
c09595f6 PZ |
1416 | } |
1417 | ||
eb755805 PZ |
1418 | static int tg_nop(struct task_group *tg, void *data) |
1419 | { | |
1420 | return 0; | |
c09595f6 | 1421 | } |
eb755805 PZ |
1422 | #endif |
1423 | ||
1424 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1425 | /* Used instead of source_load when we know the type == 0 */ |
1426 | static unsigned long weighted_cpuload(const int cpu) | |
1427 | { | |
1428 | return cpu_rq(cpu)->load.weight; | |
1429 | } | |
1430 | ||
1431 | /* | |
1432 | * Return a low guess at the load of a migration-source cpu weighted | |
1433 | * according to the scheduling class and "nice" value. | |
1434 | * | |
1435 | * We want to under-estimate the load of migration sources, to | |
1436 | * balance conservatively. | |
1437 | */ | |
1438 | static unsigned long source_load(int cpu, int type) | |
1439 | { | |
1440 | struct rq *rq = cpu_rq(cpu); | |
1441 | unsigned long total = weighted_cpuload(cpu); | |
1442 | ||
1443 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1444 | return total; | |
1445 | ||
1446 | return min(rq->cpu_load[type-1], total); | |
1447 | } | |
1448 | ||
1449 | /* | |
1450 | * Return a high guess at the load of a migration-target cpu weighted | |
1451 | * according to the scheduling class and "nice" value. | |
1452 | */ | |
1453 | static unsigned long target_load(int cpu, int type) | |
1454 | { | |
1455 | struct rq *rq = cpu_rq(cpu); | |
1456 | unsigned long total = weighted_cpuload(cpu); | |
1457 | ||
1458 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1459 | return total; | |
1460 | ||
1461 | return max(rq->cpu_load[type-1], total); | |
1462 | } | |
1463 | ||
ae154be1 PZ |
1464 | static struct sched_group *group_of(int cpu) |
1465 | { | |
1466 | struct sched_domain *sd = rcu_dereference(cpu_rq(cpu)->sd); | |
1467 | ||
1468 | if (!sd) | |
1469 | return NULL; | |
1470 | ||
1471 | return sd->groups; | |
1472 | } | |
1473 | ||
1474 | static unsigned long power_of(int cpu) | |
1475 | { | |
1476 | struct sched_group *group = group_of(cpu); | |
1477 | ||
1478 | if (!group) | |
1479 | return SCHED_LOAD_SCALE; | |
1480 | ||
1481 | return group->cpu_power; | |
1482 | } | |
1483 | ||
eb755805 PZ |
1484 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1485 | ||
1486 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1487 | { | |
1488 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1489 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1490 | |
4cd42620 SR |
1491 | if (nr_running) |
1492 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1493 | else |
1494 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1495 | |
1496 | return rq->avg_load_per_task; | |
1497 | } | |
1498 | ||
1499 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1500 | |
4a6cc4bd | 1501 | static __read_mostly unsigned long *update_shares_data; |
34d76c41 | 1502 | |
c09595f6 PZ |
1503 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1504 | ||
1505 | /* | |
1506 | * Calculate and set the cpu's group shares. | |
1507 | */ | |
34d76c41 PZ |
1508 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1509 | unsigned long sd_shares, | |
1510 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1511 | unsigned long *usd_rq_weight) |
18d95a28 | 1512 | { |
34d76c41 | 1513 | unsigned long shares, rq_weight; |
a5004278 | 1514 | int boost = 0; |
c09595f6 | 1515 | |
4a6cc4bd | 1516 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1517 | if (!rq_weight) { |
1518 | boost = 1; | |
1519 | rq_weight = NICE_0_LOAD; | |
1520 | } | |
c8cba857 | 1521 | |
c09595f6 | 1522 | /* |
a8af7246 PZ |
1523 | * \Sum_j shares_j * rq_weight_i |
1524 | * shares_i = ----------------------------- | |
1525 | * \Sum_j rq_weight_j | |
c09595f6 | 1526 | */ |
ec4e0e2f | 1527 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1528 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1529 | |
ffda12a1 PZ |
1530 | if (abs(shares - tg->se[cpu]->load.weight) > |
1531 | sysctl_sched_shares_thresh) { | |
1532 | struct rq *rq = cpu_rq(cpu); | |
1533 | unsigned long flags; | |
c09595f6 | 1534 | |
05fa785c | 1535 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1536 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1537 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1538 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1539 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1540 | } |
18d95a28 | 1541 | } |
c09595f6 PZ |
1542 | |
1543 | /* | |
c8cba857 PZ |
1544 | * Re-compute the task group their per cpu shares over the given domain. |
1545 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1546 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1547 | */ |
eb755805 | 1548 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1549 | { |
cd8ad40d | 1550 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1551 | unsigned long *usd_rq_weight; |
eb755805 | 1552 | struct sched_domain *sd = data; |
34d76c41 | 1553 | unsigned long flags; |
c8cba857 | 1554 | int i; |
c09595f6 | 1555 | |
34d76c41 PZ |
1556 | if (!tg->se[0]) |
1557 | return 0; | |
1558 | ||
1559 | local_irq_save(flags); | |
4a6cc4bd | 1560 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1561 | |
758b2cdc | 1562 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1563 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1564 | usd_rq_weight[i] = weight; |
34d76c41 | 1565 | |
cd8ad40d | 1566 | rq_weight += weight; |
ec4e0e2f KC |
1567 | /* |
1568 | * If there are currently no tasks on the cpu pretend there | |
1569 | * is one of average load so that when a new task gets to | |
1570 | * run here it will not get delayed by group starvation. | |
1571 | */ | |
ec4e0e2f KC |
1572 | if (!weight) |
1573 | weight = NICE_0_LOAD; | |
1574 | ||
cd8ad40d | 1575 | sum_weight += weight; |
c8cba857 | 1576 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1577 | } |
c09595f6 | 1578 | |
cd8ad40d PZ |
1579 | if (!rq_weight) |
1580 | rq_weight = sum_weight; | |
1581 | ||
c8cba857 PZ |
1582 | if ((!shares && rq_weight) || shares > tg->shares) |
1583 | shares = tg->shares; | |
1584 | ||
1585 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1586 | shares = tg->shares; | |
c09595f6 | 1587 | |
758b2cdc | 1588 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1589 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1590 | |
1591 | local_irq_restore(flags); | |
eb755805 PZ |
1592 | |
1593 | return 0; | |
c09595f6 PZ |
1594 | } |
1595 | ||
1596 | /* | |
c8cba857 PZ |
1597 | * Compute the cpu's hierarchical load factor for each task group. |
1598 | * This needs to be done in a top-down fashion because the load of a child | |
1599 | * group is a fraction of its parents load. | |
c09595f6 | 1600 | */ |
eb755805 | 1601 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1602 | { |
c8cba857 | 1603 | unsigned long load; |
eb755805 | 1604 | long cpu = (long)data; |
c09595f6 | 1605 | |
c8cba857 PZ |
1606 | if (!tg->parent) { |
1607 | load = cpu_rq(cpu)->load.weight; | |
1608 | } else { | |
1609 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1610 | load *= tg->cfs_rq[cpu]->shares; | |
1611 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1612 | } | |
c09595f6 | 1613 | |
c8cba857 | 1614 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1615 | |
eb755805 | 1616 | return 0; |
c09595f6 PZ |
1617 | } |
1618 | ||
c8cba857 | 1619 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1620 | { |
e7097159 PZ |
1621 | s64 elapsed; |
1622 | u64 now; | |
1623 | ||
1624 | if (root_task_group_empty()) | |
1625 | return; | |
1626 | ||
1627 | now = cpu_clock(raw_smp_processor_id()); | |
1628 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1629 | |
1630 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1631 | sd->last_update = now; | |
eb755805 | 1632 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1633 | } |
4d8d595d PZ |
1634 | } |
1635 | ||
3e5459b4 PZ |
1636 | static void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1637 | { | |
e7097159 PZ |
1638 | if (root_task_group_empty()) |
1639 | return; | |
1640 | ||
05fa785c | 1641 | raw_spin_unlock(&rq->lock); |
3e5459b4 | 1642 | update_shares(sd); |
05fa785c | 1643 | raw_spin_lock(&rq->lock); |
3e5459b4 PZ |
1644 | } |
1645 | ||
eb755805 | 1646 | static void update_h_load(long cpu) |
c09595f6 | 1647 | { |
e7097159 PZ |
1648 | if (root_task_group_empty()) |
1649 | return; | |
1650 | ||
eb755805 | 1651 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1652 | } |
1653 | ||
c09595f6 PZ |
1654 | #else |
1655 | ||
c8cba857 | 1656 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1657 | { |
1658 | } | |
1659 | ||
3e5459b4 PZ |
1660 | static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd) |
1661 | { | |
1662 | } | |
1663 | ||
18d95a28 PZ |
1664 | #endif |
1665 | ||
8f45e2b5 GH |
1666 | #ifdef CONFIG_PREEMPT |
1667 | ||
b78bb868 PZ |
1668 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1669 | ||
70574a99 | 1670 | /* |
8f45e2b5 GH |
1671 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1672 | * way at the expense of forcing extra atomic operations in all | |
1673 | * invocations. This assures that the double_lock is acquired using the | |
1674 | * same underlying policy as the spinlock_t on this architecture, which | |
1675 | * reduces latency compared to the unfair variant below. However, it | |
1676 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1677 | */ |
8f45e2b5 GH |
1678 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1679 | __releases(this_rq->lock) | |
1680 | __acquires(busiest->lock) | |
1681 | __acquires(this_rq->lock) | |
1682 | { | |
05fa785c | 1683 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1684 | double_rq_lock(this_rq, busiest); |
1685 | ||
1686 | return 1; | |
1687 | } | |
1688 | ||
1689 | #else | |
1690 | /* | |
1691 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1692 | * latency by eliminating extra atomic operations when the locks are | |
1693 | * already in proper order on entry. This favors lower cpu-ids and will | |
1694 | * grant the double lock to lower cpus over higher ids under contention, | |
1695 | * regardless of entry order into the function. | |
1696 | */ | |
1697 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1698 | __releases(this_rq->lock) |
1699 | __acquires(busiest->lock) | |
1700 | __acquires(this_rq->lock) | |
1701 | { | |
1702 | int ret = 0; | |
1703 | ||
05fa785c | 1704 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1705 | if (busiest < this_rq) { |
05fa785c TG |
1706 | raw_spin_unlock(&this_rq->lock); |
1707 | raw_spin_lock(&busiest->lock); | |
1708 | raw_spin_lock_nested(&this_rq->lock, | |
1709 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1710 | ret = 1; |
1711 | } else | |
05fa785c TG |
1712 | raw_spin_lock_nested(&busiest->lock, |
1713 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1714 | } |
1715 | return ret; | |
1716 | } | |
1717 | ||
8f45e2b5 GH |
1718 | #endif /* CONFIG_PREEMPT */ |
1719 | ||
1720 | /* | |
1721 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1722 | */ | |
1723 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1724 | { | |
1725 | if (unlikely(!irqs_disabled())) { | |
1726 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1727 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1728 | BUG_ON(1); |
1729 | } | |
1730 | ||
1731 | return _double_lock_balance(this_rq, busiest); | |
1732 | } | |
1733 | ||
70574a99 AD |
1734 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1735 | __releases(busiest->lock) | |
1736 | { | |
05fa785c | 1737 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1738 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1739 | } | |
1e3c88bd PZ |
1740 | |
1741 | /* | |
1742 | * double_rq_lock - safely lock two runqueues | |
1743 | * | |
1744 | * Note this does not disable interrupts like task_rq_lock, | |
1745 | * you need to do so manually before calling. | |
1746 | */ | |
1747 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1748 | __acquires(rq1->lock) | |
1749 | __acquires(rq2->lock) | |
1750 | { | |
1751 | BUG_ON(!irqs_disabled()); | |
1752 | if (rq1 == rq2) { | |
1753 | raw_spin_lock(&rq1->lock); | |
1754 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1755 | } else { | |
1756 | if (rq1 < rq2) { | |
1757 | raw_spin_lock(&rq1->lock); | |
1758 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1759 | } else { | |
1760 | raw_spin_lock(&rq2->lock); | |
1761 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1762 | } | |
1763 | } | |
1764 | update_rq_clock(rq1); | |
1765 | update_rq_clock(rq2); | |
1766 | } | |
1767 | ||
1768 | /* | |
1769 | * double_rq_unlock - safely unlock two runqueues | |
1770 | * | |
1771 | * Note this does not restore interrupts like task_rq_unlock, | |
1772 | * you need to do so manually after calling. | |
1773 | */ | |
1774 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1775 | __releases(rq1->lock) | |
1776 | __releases(rq2->lock) | |
1777 | { | |
1778 | raw_spin_unlock(&rq1->lock); | |
1779 | if (rq1 != rq2) | |
1780 | raw_spin_unlock(&rq2->lock); | |
1781 | else | |
1782 | __release(rq2->lock); | |
1783 | } | |
1784 | ||
18d95a28 PZ |
1785 | #endif |
1786 | ||
30432094 | 1787 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1788 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1789 | { | |
30432094 | 1790 | #ifdef CONFIG_SMP |
34e83e85 IM |
1791 | cfs_rq->shares = shares; |
1792 | #endif | |
1793 | } | |
30432094 | 1794 | #endif |
e7693a36 | 1795 | |
dce48a84 | 1796 | static void calc_load_account_active(struct rq *this_rq); |
0bcdcf28 | 1797 | static void update_sysctl(void); |
acb4a848 | 1798 | static int get_update_sysctl_factor(void); |
dce48a84 | 1799 | |
cd29fe6f PZ |
1800 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1801 | { | |
1802 | set_task_rq(p, cpu); | |
1803 | #ifdef CONFIG_SMP | |
1804 | /* | |
1805 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1806 | * successfuly executed on another CPU. We must ensure that updates of | |
1807 | * per-task data have been completed by this moment. | |
1808 | */ | |
1809 | smp_wmb(); | |
1810 | task_thread_info(p)->cpu = cpu; | |
1811 | #endif | |
1812 | } | |
dce48a84 | 1813 | |
1e3c88bd | 1814 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1815 | |
1816 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1817 | #define for_each_class(class) \ |
1818 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1819 | |
1e3c88bd PZ |
1820 | #include "sched_stats.h" |
1821 | ||
c09595f6 | 1822 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1823 | { |
1824 | rq->nr_running++; | |
9c217245 IM |
1825 | } |
1826 | ||
c09595f6 | 1827 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1828 | { |
1829 | rq->nr_running--; | |
9c217245 IM |
1830 | } |
1831 | ||
45bf76df IM |
1832 | static void set_load_weight(struct task_struct *p) |
1833 | { | |
1834 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1835 | p->se.load.weight = prio_to_weight[0] * 2; |
1836 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1837 | return; | |
1838 | } | |
45bf76df | 1839 | |
dd41f596 IM |
1840 | /* |
1841 | * SCHED_IDLE tasks get minimal weight: | |
1842 | */ | |
1843 | if (p->policy == SCHED_IDLE) { | |
1844 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1845 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1846 | return; | |
1847 | } | |
71f8bd46 | 1848 | |
dd41f596 IM |
1849 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1850 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1851 | } |
1852 | ||
2087a1ad GH |
1853 | static void update_avg(u64 *avg, u64 sample) |
1854 | { | |
1855 | s64 diff = sample - *avg; | |
1856 | *avg += diff >> 3; | |
1857 | } | |
1858 | ||
8159f87e | 1859 | static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) |
71f8bd46 | 1860 | { |
831451ac PZ |
1861 | if (wakeup) |
1862 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1863 | ||
dd41f596 | 1864 | sched_info_queued(p); |
fd390f6a | 1865 | p->sched_class->enqueue_task(rq, p, wakeup); |
dd41f596 | 1866 | p->se.on_rq = 1; |
71f8bd46 IM |
1867 | } |
1868 | ||
69be72c1 | 1869 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1870 | { |
831451ac PZ |
1871 | if (sleep) { |
1872 | if (p->se.last_wakeup) { | |
1873 | update_avg(&p->se.avg_overlap, | |
1874 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1875 | p->se.last_wakeup = 0; | |
1876 | } else { | |
1877 | update_avg(&p->se.avg_wakeup, | |
1878 | sysctl_sched_wakeup_granularity); | |
1879 | } | |
2087a1ad GH |
1880 | } |
1881 | ||
46ac22ba | 1882 | sched_info_dequeued(p); |
f02231e5 | 1883 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1884 | p->se.on_rq = 0; |
71f8bd46 IM |
1885 | } |
1886 | ||
1e3c88bd PZ |
1887 | /* |
1888 | * activate_task - move a task to the runqueue. | |
1889 | */ | |
1890 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) | |
1891 | { | |
1892 | if (task_contributes_to_load(p)) | |
1893 | rq->nr_uninterruptible--; | |
1894 | ||
1895 | enqueue_task(rq, p, wakeup); | |
1896 | inc_nr_running(rq); | |
1897 | } | |
1898 | ||
1899 | /* | |
1900 | * deactivate_task - remove a task from the runqueue. | |
1901 | */ | |
1902 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) | |
1903 | { | |
1904 | if (task_contributes_to_load(p)) | |
1905 | rq->nr_uninterruptible++; | |
1906 | ||
1907 | dequeue_task(rq, p, sleep); | |
1908 | dec_nr_running(rq); | |
1909 | } | |
1910 | ||
1911 | #include "sched_idletask.c" | |
1912 | #include "sched_fair.c" | |
1913 | #include "sched_rt.c" | |
1914 | #ifdef CONFIG_SCHED_DEBUG | |
1915 | # include "sched_debug.c" | |
1916 | #endif | |
1917 | ||
14531189 | 1918 | /* |
dd41f596 | 1919 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1920 | */ |
14531189 IM |
1921 | static inline int __normal_prio(struct task_struct *p) |
1922 | { | |
dd41f596 | 1923 | return p->static_prio; |
14531189 IM |
1924 | } |
1925 | ||
b29739f9 IM |
1926 | /* |
1927 | * Calculate the expected normal priority: i.e. priority | |
1928 | * without taking RT-inheritance into account. Might be | |
1929 | * boosted by interactivity modifiers. Changes upon fork, | |
1930 | * setprio syscalls, and whenever the interactivity | |
1931 | * estimator recalculates. | |
1932 | */ | |
36c8b586 | 1933 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1934 | { |
1935 | int prio; | |
1936 | ||
e05606d3 | 1937 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1938 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1939 | else | |
1940 | prio = __normal_prio(p); | |
1941 | return prio; | |
1942 | } | |
1943 | ||
1944 | /* | |
1945 | * Calculate the current priority, i.e. the priority | |
1946 | * taken into account by the scheduler. This value might | |
1947 | * be boosted by RT tasks, or might be boosted by | |
1948 | * interactivity modifiers. Will be RT if the task got | |
1949 | * RT-boosted. If not then it returns p->normal_prio. | |
1950 | */ | |
36c8b586 | 1951 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1952 | { |
1953 | p->normal_prio = normal_prio(p); | |
1954 | /* | |
1955 | * If we are RT tasks or we were boosted to RT priority, | |
1956 | * keep the priority unchanged. Otherwise, update priority | |
1957 | * to the normal priority: | |
1958 | */ | |
1959 | if (!rt_prio(p->prio)) | |
1960 | return p->normal_prio; | |
1961 | return p->prio; | |
1962 | } | |
1963 | ||
1da177e4 LT |
1964 | /** |
1965 | * task_curr - is this task currently executing on a CPU? | |
1966 | * @p: the task in question. | |
1967 | */ | |
36c8b586 | 1968 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1969 | { |
1970 | return cpu_curr(task_cpu(p)) == p; | |
1971 | } | |
1972 | ||
cb469845 SR |
1973 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1974 | const struct sched_class *prev_class, | |
1975 | int oldprio, int running) | |
1976 | { | |
1977 | if (prev_class != p->sched_class) { | |
1978 | if (prev_class->switched_from) | |
1979 | prev_class->switched_from(rq, p, running); | |
1980 | p->sched_class->switched_to(rq, p, running); | |
1981 | } else | |
1982 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1983 | } | |
1984 | ||
1da177e4 | 1985 | #ifdef CONFIG_SMP |
cc367732 IM |
1986 | /* |
1987 | * Is this task likely cache-hot: | |
1988 | */ | |
e7693a36 | 1989 | static int |
cc367732 IM |
1990 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1991 | { | |
1992 | s64 delta; | |
1993 | ||
e6c8fba7 PZ |
1994 | if (p->sched_class != &fair_sched_class) |
1995 | return 0; | |
1996 | ||
f540a608 IM |
1997 | /* |
1998 | * Buddy candidates are cache hot: | |
1999 | */ | |
f685ceac | 2000 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2001 | (&p->se == cfs_rq_of(&p->se)->next || |
2002 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2003 | return 1; |
2004 | ||
6bc1665b IM |
2005 | if (sysctl_sched_migration_cost == -1) |
2006 | return 1; | |
2007 | if (sysctl_sched_migration_cost == 0) | |
2008 | return 0; | |
2009 | ||
cc367732 IM |
2010 | delta = now - p->se.exec_start; |
2011 | ||
2012 | return delta < (s64)sysctl_sched_migration_cost; | |
2013 | } | |
2014 | ||
dd41f596 | 2015 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2016 | { |
e2912009 PZ |
2017 | #ifdef CONFIG_SCHED_DEBUG |
2018 | /* | |
2019 | * We should never call set_task_cpu() on a blocked task, | |
2020 | * ttwu() will sort out the placement. | |
2021 | */ | |
077614ee PZ |
2022 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2023 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2024 | #endif |
2025 | ||
de1d7286 | 2026 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2027 | |
0c69774e PZ |
2028 | if (task_cpu(p) != new_cpu) { |
2029 | p->se.nr_migrations++; | |
2030 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2031 | } | |
dd41f596 IM |
2032 | |
2033 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2034 | } |
2035 | ||
70b97a7f | 2036 | struct migration_req { |
1da177e4 | 2037 | struct list_head list; |
1da177e4 | 2038 | |
36c8b586 | 2039 | struct task_struct *task; |
1da177e4 LT |
2040 | int dest_cpu; |
2041 | ||
1da177e4 | 2042 | struct completion done; |
70b97a7f | 2043 | }; |
1da177e4 LT |
2044 | |
2045 | /* | |
2046 | * The task's runqueue lock must be held. | |
2047 | * Returns true if you have to wait for migration thread. | |
2048 | */ | |
36c8b586 | 2049 | static int |
70b97a7f | 2050 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2051 | { |
70b97a7f | 2052 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2053 | |
2054 | /* | |
2055 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2056 | * the next wake-up will properly place the task. |
1da177e4 | 2057 | */ |
e2912009 | 2058 | if (!p->se.on_rq && !task_running(rq, p)) |
1da177e4 | 2059 | return 0; |
1da177e4 LT |
2060 | |
2061 | init_completion(&req->done); | |
1da177e4 LT |
2062 | req->task = p; |
2063 | req->dest_cpu = dest_cpu; | |
2064 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2065 | |
1da177e4 LT |
2066 | return 1; |
2067 | } | |
2068 | ||
a26b89f0 MM |
2069 | /* |
2070 | * wait_task_context_switch - wait for a thread to complete at least one | |
2071 | * context switch. | |
2072 | * | |
2073 | * @p must not be current. | |
2074 | */ | |
2075 | void wait_task_context_switch(struct task_struct *p) | |
2076 | { | |
2077 | unsigned long nvcsw, nivcsw, flags; | |
2078 | int running; | |
2079 | struct rq *rq; | |
2080 | ||
2081 | nvcsw = p->nvcsw; | |
2082 | nivcsw = p->nivcsw; | |
2083 | for (;;) { | |
2084 | /* | |
2085 | * The runqueue is assigned before the actual context | |
2086 | * switch. We need to take the runqueue lock. | |
2087 | * | |
2088 | * We could check initially without the lock but it is | |
2089 | * very likely that we need to take the lock in every | |
2090 | * iteration. | |
2091 | */ | |
2092 | rq = task_rq_lock(p, &flags); | |
2093 | running = task_running(rq, p); | |
2094 | task_rq_unlock(rq, &flags); | |
2095 | ||
2096 | if (likely(!running)) | |
2097 | break; | |
2098 | /* | |
2099 | * The switch count is incremented before the actual | |
2100 | * context switch. We thus wait for two switches to be | |
2101 | * sure at least one completed. | |
2102 | */ | |
2103 | if ((p->nvcsw - nvcsw) > 1) | |
2104 | break; | |
2105 | if ((p->nivcsw - nivcsw) > 1) | |
2106 | break; | |
2107 | ||
2108 | cpu_relax(); | |
2109 | } | |
2110 | } | |
2111 | ||
1da177e4 LT |
2112 | /* |
2113 | * wait_task_inactive - wait for a thread to unschedule. | |
2114 | * | |
85ba2d86 RM |
2115 | * If @match_state is nonzero, it's the @p->state value just checked and |
2116 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2117 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2118 | * we return a positive number (its total switch count). If a second call | |
2119 | * a short while later returns the same number, the caller can be sure that | |
2120 | * @p has remained unscheduled the whole time. | |
2121 | * | |
1da177e4 LT |
2122 | * The caller must ensure that the task *will* unschedule sometime soon, |
2123 | * else this function might spin for a *long* time. This function can't | |
2124 | * be called with interrupts off, or it may introduce deadlock with | |
2125 | * smp_call_function() if an IPI is sent by the same process we are | |
2126 | * waiting to become inactive. | |
2127 | */ | |
85ba2d86 | 2128 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2129 | { |
2130 | unsigned long flags; | |
dd41f596 | 2131 | int running, on_rq; |
85ba2d86 | 2132 | unsigned long ncsw; |
70b97a7f | 2133 | struct rq *rq; |
1da177e4 | 2134 | |
3a5c359a AK |
2135 | for (;;) { |
2136 | /* | |
2137 | * We do the initial early heuristics without holding | |
2138 | * any task-queue locks at all. We'll only try to get | |
2139 | * the runqueue lock when things look like they will | |
2140 | * work out! | |
2141 | */ | |
2142 | rq = task_rq(p); | |
fa490cfd | 2143 | |
3a5c359a AK |
2144 | /* |
2145 | * If the task is actively running on another CPU | |
2146 | * still, just relax and busy-wait without holding | |
2147 | * any locks. | |
2148 | * | |
2149 | * NOTE! Since we don't hold any locks, it's not | |
2150 | * even sure that "rq" stays as the right runqueue! | |
2151 | * But we don't care, since "task_running()" will | |
2152 | * return false if the runqueue has changed and p | |
2153 | * is actually now running somewhere else! | |
2154 | */ | |
85ba2d86 RM |
2155 | while (task_running(rq, p)) { |
2156 | if (match_state && unlikely(p->state != match_state)) | |
2157 | return 0; | |
3a5c359a | 2158 | cpu_relax(); |
85ba2d86 | 2159 | } |
fa490cfd | 2160 | |
3a5c359a AK |
2161 | /* |
2162 | * Ok, time to look more closely! We need the rq | |
2163 | * lock now, to be *sure*. If we're wrong, we'll | |
2164 | * just go back and repeat. | |
2165 | */ | |
2166 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2167 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2168 | running = task_running(rq, p); |
2169 | on_rq = p->se.on_rq; | |
85ba2d86 | 2170 | ncsw = 0; |
f31e11d8 | 2171 | if (!match_state || p->state == match_state) |
93dcf55f | 2172 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2173 | task_rq_unlock(rq, &flags); |
fa490cfd | 2174 | |
85ba2d86 RM |
2175 | /* |
2176 | * If it changed from the expected state, bail out now. | |
2177 | */ | |
2178 | if (unlikely(!ncsw)) | |
2179 | break; | |
2180 | ||
3a5c359a AK |
2181 | /* |
2182 | * Was it really running after all now that we | |
2183 | * checked with the proper locks actually held? | |
2184 | * | |
2185 | * Oops. Go back and try again.. | |
2186 | */ | |
2187 | if (unlikely(running)) { | |
2188 | cpu_relax(); | |
2189 | continue; | |
2190 | } | |
fa490cfd | 2191 | |
3a5c359a AK |
2192 | /* |
2193 | * It's not enough that it's not actively running, | |
2194 | * it must be off the runqueue _entirely_, and not | |
2195 | * preempted! | |
2196 | * | |
80dd99b3 | 2197 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2198 | * running right now), it's preempted, and we should |
2199 | * yield - it could be a while. | |
2200 | */ | |
2201 | if (unlikely(on_rq)) { | |
2202 | schedule_timeout_uninterruptible(1); | |
2203 | continue; | |
2204 | } | |
fa490cfd | 2205 | |
3a5c359a AK |
2206 | /* |
2207 | * Ahh, all good. It wasn't running, and it wasn't | |
2208 | * runnable, which means that it will never become | |
2209 | * running in the future either. We're all done! | |
2210 | */ | |
2211 | break; | |
2212 | } | |
85ba2d86 RM |
2213 | |
2214 | return ncsw; | |
1da177e4 LT |
2215 | } |
2216 | ||
2217 | /*** | |
2218 | * kick_process - kick a running thread to enter/exit the kernel | |
2219 | * @p: the to-be-kicked thread | |
2220 | * | |
2221 | * Cause a process which is running on another CPU to enter | |
2222 | * kernel-mode, without any delay. (to get signals handled.) | |
2223 | * | |
2224 | * NOTE: this function doesnt have to take the runqueue lock, | |
2225 | * because all it wants to ensure is that the remote task enters | |
2226 | * the kernel. If the IPI races and the task has been migrated | |
2227 | * to another CPU then no harm is done and the purpose has been | |
2228 | * achieved as well. | |
2229 | */ | |
36c8b586 | 2230 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2231 | { |
2232 | int cpu; | |
2233 | ||
2234 | preempt_disable(); | |
2235 | cpu = task_cpu(p); | |
2236 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2237 | smp_send_reschedule(cpu); | |
2238 | preempt_enable(); | |
2239 | } | |
b43e3521 | 2240 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2241 | #endif /* CONFIG_SMP */ |
1da177e4 | 2242 | |
0793a61d TG |
2243 | /** |
2244 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2245 | * @p: the task to evaluate | |
2246 | * @func: the function to be called | |
2247 | * @info: the function call argument | |
2248 | * | |
2249 | * Calls the function @func when the task is currently running. This might | |
2250 | * be on the current CPU, which just calls the function directly | |
2251 | */ | |
2252 | void task_oncpu_function_call(struct task_struct *p, | |
2253 | void (*func) (void *info), void *info) | |
2254 | { | |
2255 | int cpu; | |
2256 | ||
2257 | preempt_disable(); | |
2258 | cpu = task_cpu(p); | |
2259 | if (task_curr(p)) | |
2260 | smp_call_function_single(cpu, func, info, 1); | |
2261 | preempt_enable(); | |
2262 | } | |
2263 | ||
970b13ba | 2264 | #ifdef CONFIG_SMP |
5da9a0fb PZ |
2265 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2266 | { | |
2267 | int dest_cpu; | |
2268 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2269 | ||
2270 | /* Look for allowed, online CPU in same node. */ | |
2271 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2272 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2273 | return dest_cpu; | |
2274 | ||
2275 | /* Any allowed, online CPU? */ | |
2276 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2277 | if (dest_cpu < nr_cpu_ids) | |
2278 | return dest_cpu; | |
2279 | ||
2280 | /* No more Mr. Nice Guy. */ | |
2281 | if (dest_cpu >= nr_cpu_ids) { | |
2282 | rcu_read_lock(); | |
2283 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); | |
2284 | rcu_read_unlock(); | |
2285 | dest_cpu = cpumask_any_and(cpu_active_mask, &p->cpus_allowed); | |
2286 | ||
2287 | /* | |
2288 | * Don't tell them about moving exiting tasks or | |
2289 | * kernel threads (both mm NULL), since they never | |
2290 | * leave kernel. | |
2291 | */ | |
2292 | if (p->mm && printk_ratelimit()) { | |
2293 | printk(KERN_INFO "process %d (%s) no " | |
2294 | "longer affine to cpu%d\n", | |
2295 | task_pid_nr(p), p->comm, cpu); | |
2296 | } | |
2297 | } | |
2298 | ||
2299 | return dest_cpu; | |
2300 | } | |
2301 | ||
e2912009 PZ |
2302 | /* |
2303 | * Called from: | |
2304 | * | |
2305 | * - fork, @p is stable because it isn't on the tasklist yet | |
2306 | * | |
38022906 | 2307 | * - exec, @p is unstable, retry loop |
e2912009 PZ |
2308 | * |
2309 | * - wake-up, we serialize ->cpus_allowed against TASK_WAKING so | |
2310 | * we should be good. | |
2311 | */ | |
970b13ba PZ |
2312 | static inline |
2313 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) | |
2314 | { | |
e2912009 PZ |
2315 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
2316 | ||
2317 | /* | |
2318 | * In order not to call set_task_cpu() on a blocking task we need | |
2319 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2320 | * cpu. | |
2321 | * | |
2322 | * Since this is common to all placement strategies, this lives here. | |
2323 | * | |
2324 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2325 | * not worry about this generic constraint ] | |
2326 | */ | |
2327 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2328 | !cpu_online(cpu))) |
5da9a0fb | 2329 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2330 | |
2331 | return cpu; | |
970b13ba PZ |
2332 | } |
2333 | #endif | |
2334 | ||
1da177e4 LT |
2335 | /*** |
2336 | * try_to_wake_up - wake up a thread | |
2337 | * @p: the to-be-woken-up thread | |
2338 | * @state: the mask of task states that can be woken | |
2339 | * @sync: do a synchronous wakeup? | |
2340 | * | |
2341 | * Put it on the run-queue if it's not already there. The "current" | |
2342 | * thread is always on the run-queue (except when the actual | |
2343 | * re-schedule is in progress), and as such you're allowed to do | |
2344 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2345 | * runnable without the overhead of this. | |
2346 | * | |
2347 | * returns failure only if the task is already active. | |
2348 | */ | |
7d478721 PZ |
2349 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2350 | int wake_flags) | |
1da177e4 | 2351 | { |
cc367732 | 2352 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2353 | unsigned long flags; |
f5dc3753 | 2354 | struct rq *rq, *orig_rq; |
1da177e4 | 2355 | |
b85d0667 | 2356 | if (!sched_feat(SYNC_WAKEUPS)) |
7d478721 | 2357 | wake_flags &= ~WF_SYNC; |
2398f2c6 | 2358 | |
e9c84311 | 2359 | this_cpu = get_cpu(); |
2398f2c6 | 2360 | |
04e2f174 | 2361 | smp_wmb(); |
f5dc3753 | 2362 | rq = orig_rq = task_rq_lock(p, &flags); |
03e89e45 | 2363 | update_rq_clock(rq); |
e9c84311 | 2364 | if (!(p->state & state)) |
1da177e4 LT |
2365 | goto out; |
2366 | ||
dd41f596 | 2367 | if (p->se.on_rq) |
1da177e4 LT |
2368 | goto out_running; |
2369 | ||
2370 | cpu = task_cpu(p); | |
cc367732 | 2371 | orig_cpu = cpu; |
1da177e4 LT |
2372 | |
2373 | #ifdef CONFIG_SMP | |
2374 | if (unlikely(task_running(rq, p))) | |
2375 | goto out_activate; | |
2376 | ||
e9c84311 PZ |
2377 | /* |
2378 | * In order to handle concurrent wakeups and release the rq->lock | |
2379 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2380 | * |
2381 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2382 | */ |
eb24073b IM |
2383 | if (task_contributes_to_load(p)) |
2384 | rq->nr_uninterruptible--; | |
e9c84311 | 2385 | p->state = TASK_WAKING; |
efbbd05a PZ |
2386 | |
2387 | if (p->sched_class->task_waking) | |
2388 | p->sched_class->task_waking(rq, p); | |
2389 | ||
ab19cb23 | 2390 | __task_rq_unlock(rq); |
e9c84311 | 2391 | |
970b13ba | 2392 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
ab19cb23 | 2393 | if (cpu != orig_cpu) |
5d2f5a61 | 2394 | set_task_cpu(p, cpu); |
ab19cb23 PZ |
2395 | |
2396 | rq = __task_rq_lock(p); | |
2397 | update_rq_clock(rq); | |
f5dc3753 | 2398 | |
e9c84311 PZ |
2399 | WARN_ON(p->state != TASK_WAKING); |
2400 | cpu = task_cpu(p); | |
1da177e4 | 2401 | |
e7693a36 GH |
2402 | #ifdef CONFIG_SCHEDSTATS |
2403 | schedstat_inc(rq, ttwu_count); | |
2404 | if (cpu == this_cpu) | |
2405 | schedstat_inc(rq, ttwu_local); | |
2406 | else { | |
2407 | struct sched_domain *sd; | |
2408 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2409 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2410 | schedstat_inc(sd, ttwu_wake_remote); |
2411 | break; | |
2412 | } | |
2413 | } | |
2414 | } | |
6d6bc0ad | 2415 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2416 | |
1da177e4 LT |
2417 | out_activate: |
2418 | #endif /* CONFIG_SMP */ | |
cc367732 | 2419 | schedstat_inc(p, se.nr_wakeups); |
7d478721 | 2420 | if (wake_flags & WF_SYNC) |
cc367732 IM |
2421 | schedstat_inc(p, se.nr_wakeups_sync); |
2422 | if (orig_cpu != cpu) | |
2423 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2424 | if (cpu == this_cpu) | |
2425 | schedstat_inc(p, se.nr_wakeups_local); | |
2426 | else | |
2427 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2428 | activate_task(rq, p, 1); |
1da177e4 LT |
2429 | success = 1; |
2430 | ||
831451ac PZ |
2431 | /* |
2432 | * Only attribute actual wakeups done by this task. | |
2433 | */ | |
2434 | if (!in_interrupt()) { | |
2435 | struct sched_entity *se = ¤t->se; | |
2436 | u64 sample = se->sum_exec_runtime; | |
2437 | ||
2438 | if (se->last_wakeup) | |
2439 | sample -= se->last_wakeup; | |
2440 | else | |
2441 | sample -= se->start_runtime; | |
2442 | update_avg(&se->avg_wakeup, sample); | |
2443 | ||
2444 | se->last_wakeup = se->sum_exec_runtime; | |
2445 | } | |
2446 | ||
1da177e4 | 2447 | out_running: |
468a15bb | 2448 | trace_sched_wakeup(rq, p, success); |
7d478721 | 2449 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2450 | |
1da177e4 | 2451 | p->state = TASK_RUNNING; |
9a897c5a | 2452 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2453 | if (p->sched_class->task_woken) |
2454 | p->sched_class->task_woken(rq, p); | |
eae0c9df MG |
2455 | |
2456 | if (unlikely(rq->idle_stamp)) { | |
2457 | u64 delta = rq->clock - rq->idle_stamp; | |
2458 | u64 max = 2*sysctl_sched_migration_cost; | |
2459 | ||
2460 | if (delta > max) | |
2461 | rq->avg_idle = max; | |
2462 | else | |
2463 | update_avg(&rq->avg_idle, delta); | |
2464 | rq->idle_stamp = 0; | |
2465 | } | |
9a897c5a | 2466 | #endif |
1da177e4 LT |
2467 | out: |
2468 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2469 | put_cpu(); |
1da177e4 LT |
2470 | |
2471 | return success; | |
2472 | } | |
2473 | ||
50fa610a DH |
2474 | /** |
2475 | * wake_up_process - Wake up a specific process | |
2476 | * @p: The process to be woken up. | |
2477 | * | |
2478 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2479 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2480 | * running. | |
2481 | * | |
2482 | * It may be assumed that this function implies a write memory barrier before | |
2483 | * changing the task state if and only if any tasks are woken up. | |
2484 | */ | |
7ad5b3a5 | 2485 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2486 | { |
d9514f6c | 2487 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2488 | } |
1da177e4 LT |
2489 | EXPORT_SYMBOL(wake_up_process); |
2490 | ||
7ad5b3a5 | 2491 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2492 | { |
2493 | return try_to_wake_up(p, state, 0); | |
2494 | } | |
2495 | ||
1da177e4 LT |
2496 | /* |
2497 | * Perform scheduler related setup for a newly forked process p. | |
2498 | * p is forked by current. | |
dd41f596 IM |
2499 | * |
2500 | * __sched_fork() is basic setup used by init_idle() too: | |
2501 | */ | |
2502 | static void __sched_fork(struct task_struct *p) | |
2503 | { | |
dd41f596 IM |
2504 | p->se.exec_start = 0; |
2505 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2506 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2507 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2508 | p->se.last_wakeup = 0; |
2509 | p->se.avg_overlap = 0; | |
831451ac PZ |
2510 | p->se.start_runtime = 0; |
2511 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2512 | |
2513 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2514 | p->se.wait_start = 0; |
2515 | p->se.wait_max = 0; | |
2516 | p->se.wait_count = 0; | |
2517 | p->se.wait_sum = 0; | |
2518 | ||
2519 | p->se.sleep_start = 0; | |
2520 | p->se.sleep_max = 0; | |
2521 | p->se.sum_sleep_runtime = 0; | |
2522 | ||
2523 | p->se.block_start = 0; | |
2524 | p->se.block_max = 0; | |
2525 | p->se.exec_max = 0; | |
2526 | p->se.slice_max = 0; | |
2527 | ||
2528 | p->se.nr_migrations_cold = 0; | |
2529 | p->se.nr_failed_migrations_affine = 0; | |
2530 | p->se.nr_failed_migrations_running = 0; | |
2531 | p->se.nr_failed_migrations_hot = 0; | |
2532 | p->se.nr_forced_migrations = 0; | |
7793527b LDM |
2533 | |
2534 | p->se.nr_wakeups = 0; | |
2535 | p->se.nr_wakeups_sync = 0; | |
2536 | p->se.nr_wakeups_migrate = 0; | |
2537 | p->se.nr_wakeups_local = 0; | |
2538 | p->se.nr_wakeups_remote = 0; | |
2539 | p->se.nr_wakeups_affine = 0; | |
2540 | p->se.nr_wakeups_affine_attempts = 0; | |
2541 | p->se.nr_wakeups_passive = 0; | |
2542 | p->se.nr_wakeups_idle = 0; | |
2543 | ||
6cfb0d5d | 2544 | #endif |
476d139c | 2545 | |
fa717060 | 2546 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2547 | p->se.on_rq = 0; |
4a55bd5e | 2548 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2549 | |
e107be36 AK |
2550 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2551 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2552 | #endif | |
dd41f596 IM |
2553 | } |
2554 | ||
2555 | /* | |
2556 | * fork()/clone()-time setup: | |
2557 | */ | |
2558 | void sched_fork(struct task_struct *p, int clone_flags) | |
2559 | { | |
2560 | int cpu = get_cpu(); | |
2561 | ||
2562 | __sched_fork(p); | |
06b83b5f PZ |
2563 | /* |
2564 | * We mark the process as waking here. This guarantees that | |
2565 | * nobody will actually run it, and a signal or other external | |
2566 | * event cannot wake it up and insert it on the runqueue either. | |
2567 | */ | |
2568 | p->state = TASK_WAKING; | |
dd41f596 | 2569 | |
b9dc29e7 MG |
2570 | /* |
2571 | * Revert to default priority/policy on fork if requested. | |
2572 | */ | |
2573 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2574 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2575 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2576 | p->normal_prio = p->static_prio; |
2577 | } | |
b9dc29e7 | 2578 | |
6c697bdf MG |
2579 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2580 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2581 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2582 | set_load_weight(p); |
2583 | } | |
2584 | ||
b9dc29e7 MG |
2585 | /* |
2586 | * We don't need the reset flag anymore after the fork. It has | |
2587 | * fulfilled its duty: | |
2588 | */ | |
2589 | p->sched_reset_on_fork = 0; | |
2590 | } | |
ca94c442 | 2591 | |
f83f9ac2 PW |
2592 | /* |
2593 | * Make sure we do not leak PI boosting priority to the child. | |
2594 | */ | |
2595 | p->prio = current->normal_prio; | |
2596 | ||
2ddbf952 HS |
2597 | if (!rt_prio(p->prio)) |
2598 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2599 | |
cd29fe6f PZ |
2600 | if (p->sched_class->task_fork) |
2601 | p->sched_class->task_fork(p); | |
2602 | ||
5f3edc1b | 2603 | #ifdef CONFIG_SMP |
970b13ba | 2604 | cpu = select_task_rq(p, SD_BALANCE_FORK, 0); |
5f3edc1b PZ |
2605 | #endif |
2606 | set_task_cpu(p, cpu); | |
2607 | ||
52f17b6c | 2608 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2609 | if (likely(sched_info_on())) |
52f17b6c | 2610 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2611 | #endif |
d6077cb8 | 2612 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2613 | p->oncpu = 0; |
2614 | #endif | |
1da177e4 | 2615 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2616 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2617 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2618 | #endif |
917b627d GH |
2619 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2620 | ||
476d139c | 2621 | put_cpu(); |
1da177e4 LT |
2622 | } |
2623 | ||
2624 | /* | |
2625 | * wake_up_new_task - wake up a newly created task for the first time. | |
2626 | * | |
2627 | * This function will do some initial scheduler statistics housekeeping | |
2628 | * that must be done for every newly created context, then puts the task | |
2629 | * on the runqueue and wakes it. | |
2630 | */ | |
7ad5b3a5 | 2631 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2632 | { |
2633 | unsigned long flags; | |
dd41f596 | 2634 | struct rq *rq; |
1da177e4 LT |
2635 | |
2636 | rq = task_rq_lock(p, &flags); | |
06b83b5f PZ |
2637 | BUG_ON(p->state != TASK_WAKING); |
2638 | p->state = TASK_RUNNING; | |
a8e504d2 | 2639 | update_rq_clock(rq); |
cd29fe6f | 2640 | activate_task(rq, p, 0); |
c71dd42d | 2641 | trace_sched_wakeup_new(rq, p, 1); |
a7558e01 | 2642 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2643 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2644 | if (p->sched_class->task_woken) |
2645 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2646 | #endif |
dd41f596 | 2647 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
2648 | } |
2649 | ||
e107be36 AK |
2650 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2651 | ||
2652 | /** | |
80dd99b3 | 2653 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2654 | * @notifier: notifier struct to register |
e107be36 AK |
2655 | */ |
2656 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2657 | { | |
2658 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2659 | } | |
2660 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2661 | ||
2662 | /** | |
2663 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2664 | * @notifier: notifier struct to unregister |
e107be36 AK |
2665 | * |
2666 | * This is safe to call from within a preemption notifier. | |
2667 | */ | |
2668 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2669 | { | |
2670 | hlist_del(¬ifier->link); | |
2671 | } | |
2672 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2673 | ||
2674 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2675 | { | |
2676 | struct preempt_notifier *notifier; | |
2677 | struct hlist_node *node; | |
2678 | ||
2679 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2680 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2681 | } | |
2682 | ||
2683 | static void | |
2684 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2685 | struct task_struct *next) | |
2686 | { | |
2687 | struct preempt_notifier *notifier; | |
2688 | struct hlist_node *node; | |
2689 | ||
2690 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2691 | notifier->ops->sched_out(notifier, next); | |
2692 | } | |
2693 | ||
6d6bc0ad | 2694 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2695 | |
2696 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2697 | { | |
2698 | } | |
2699 | ||
2700 | static void | |
2701 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2702 | struct task_struct *next) | |
2703 | { | |
2704 | } | |
2705 | ||
6d6bc0ad | 2706 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2707 | |
4866cde0 NP |
2708 | /** |
2709 | * prepare_task_switch - prepare to switch tasks | |
2710 | * @rq: the runqueue preparing to switch | |
421cee29 | 2711 | * @prev: the current task that is being switched out |
4866cde0 NP |
2712 | * @next: the task we are going to switch to. |
2713 | * | |
2714 | * This is called with the rq lock held and interrupts off. It must | |
2715 | * be paired with a subsequent finish_task_switch after the context | |
2716 | * switch. | |
2717 | * | |
2718 | * prepare_task_switch sets up locking and calls architecture specific | |
2719 | * hooks. | |
2720 | */ | |
e107be36 AK |
2721 | static inline void |
2722 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2723 | struct task_struct *next) | |
4866cde0 | 2724 | { |
e107be36 | 2725 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2726 | prepare_lock_switch(rq, next); |
2727 | prepare_arch_switch(next); | |
2728 | } | |
2729 | ||
1da177e4 LT |
2730 | /** |
2731 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2732 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2733 | * @prev: the thread we just switched away from. |
2734 | * | |
4866cde0 NP |
2735 | * finish_task_switch must be called after the context switch, paired |
2736 | * with a prepare_task_switch call before the context switch. | |
2737 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2738 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2739 | * |
2740 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2741 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2742 | * with the lock held can cause deadlocks; see schedule() for |
2743 | * details.) | |
2744 | */ | |
a9957449 | 2745 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2746 | __releases(rq->lock) |
2747 | { | |
1da177e4 | 2748 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2749 | long prev_state; |
1da177e4 LT |
2750 | |
2751 | rq->prev_mm = NULL; | |
2752 | ||
2753 | /* | |
2754 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2755 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2756 | * schedule one last time. The schedule call will never return, and |
2757 | * the scheduled task must drop that reference. | |
c394cc9f | 2758 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2759 | * still held, otherwise prev could be scheduled on another cpu, die |
2760 | * there before we look at prev->state, and then the reference would | |
2761 | * be dropped twice. | |
2762 | * Manfred Spraul <manfred@colorfullife.com> | |
2763 | */ | |
55a101f8 | 2764 | prev_state = prev->state; |
4866cde0 | 2765 | finish_arch_switch(prev); |
cdd6c482 | 2766 | perf_event_task_sched_in(current, cpu_of(rq)); |
4866cde0 | 2767 | finish_lock_switch(rq, prev); |
e8fa1362 | 2768 | |
e107be36 | 2769 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2770 | if (mm) |
2771 | mmdrop(mm); | |
c394cc9f | 2772 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2773 | /* |
2774 | * Remove function-return probe instances associated with this | |
2775 | * task and put them back on the free list. | |
9761eea8 | 2776 | */ |
c6fd91f0 | 2777 | kprobe_flush_task(prev); |
1da177e4 | 2778 | put_task_struct(prev); |
c6fd91f0 | 2779 | } |
1da177e4 LT |
2780 | } |
2781 | ||
3f029d3c GH |
2782 | #ifdef CONFIG_SMP |
2783 | ||
2784 | /* assumes rq->lock is held */ | |
2785 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2786 | { | |
2787 | if (prev->sched_class->pre_schedule) | |
2788 | prev->sched_class->pre_schedule(rq, prev); | |
2789 | } | |
2790 | ||
2791 | /* rq->lock is NOT held, but preemption is disabled */ | |
2792 | static inline void post_schedule(struct rq *rq) | |
2793 | { | |
2794 | if (rq->post_schedule) { | |
2795 | unsigned long flags; | |
2796 | ||
05fa785c | 2797 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2798 | if (rq->curr->sched_class->post_schedule) |
2799 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2800 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2801 | |
2802 | rq->post_schedule = 0; | |
2803 | } | |
2804 | } | |
2805 | ||
2806 | #else | |
da19ab51 | 2807 | |
3f029d3c GH |
2808 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2809 | { | |
2810 | } | |
2811 | ||
2812 | static inline void post_schedule(struct rq *rq) | |
2813 | { | |
1da177e4 LT |
2814 | } |
2815 | ||
3f029d3c GH |
2816 | #endif |
2817 | ||
1da177e4 LT |
2818 | /** |
2819 | * schedule_tail - first thing a freshly forked thread must call. | |
2820 | * @prev: the thread we just switched away from. | |
2821 | */ | |
36c8b586 | 2822 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2823 | __releases(rq->lock) |
2824 | { | |
70b97a7f IM |
2825 | struct rq *rq = this_rq(); |
2826 | ||
4866cde0 | 2827 | finish_task_switch(rq, prev); |
da19ab51 | 2828 | |
3f029d3c GH |
2829 | /* |
2830 | * FIXME: do we need to worry about rq being invalidated by the | |
2831 | * task_switch? | |
2832 | */ | |
2833 | post_schedule(rq); | |
70b97a7f | 2834 | |
4866cde0 NP |
2835 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2836 | /* In this case, finish_task_switch does not reenable preemption */ | |
2837 | preempt_enable(); | |
2838 | #endif | |
1da177e4 | 2839 | if (current->set_child_tid) |
b488893a | 2840 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2841 | } |
2842 | ||
2843 | /* | |
2844 | * context_switch - switch to the new MM and the new | |
2845 | * thread's register state. | |
2846 | */ | |
dd41f596 | 2847 | static inline void |
70b97a7f | 2848 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2849 | struct task_struct *next) |
1da177e4 | 2850 | { |
dd41f596 | 2851 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2852 | |
e107be36 | 2853 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2854 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2855 | mm = next->mm; |
2856 | oldmm = prev->active_mm; | |
9226d125 ZA |
2857 | /* |
2858 | * For paravirt, this is coupled with an exit in switch_to to | |
2859 | * combine the page table reload and the switch backend into | |
2860 | * one hypercall. | |
2861 | */ | |
224101ed | 2862 | arch_start_context_switch(prev); |
9226d125 | 2863 | |
710390d9 | 2864 | if (likely(!mm)) { |
1da177e4 LT |
2865 | next->active_mm = oldmm; |
2866 | atomic_inc(&oldmm->mm_count); | |
2867 | enter_lazy_tlb(oldmm, next); | |
2868 | } else | |
2869 | switch_mm(oldmm, mm, next); | |
2870 | ||
710390d9 | 2871 | if (likely(!prev->mm)) { |
1da177e4 | 2872 | prev->active_mm = NULL; |
1da177e4 LT |
2873 | rq->prev_mm = oldmm; |
2874 | } | |
3a5f5e48 IM |
2875 | /* |
2876 | * Since the runqueue lock will be released by the next | |
2877 | * task (which is an invalid locking op but in the case | |
2878 | * of the scheduler it's an obvious special-case), so we | |
2879 | * do an early lockdep release here: | |
2880 | */ | |
2881 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2882 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2883 | #endif |
1da177e4 LT |
2884 | |
2885 | /* Here we just switch the register state and the stack. */ | |
2886 | switch_to(prev, next, prev); | |
2887 | ||
dd41f596 IM |
2888 | barrier(); |
2889 | /* | |
2890 | * this_rq must be evaluated again because prev may have moved | |
2891 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2892 | * frame will be invalid. | |
2893 | */ | |
2894 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2895 | } |
2896 | ||
2897 | /* | |
2898 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2899 | * | |
2900 | * externally visible scheduler statistics: current number of runnable | |
2901 | * threads, current number of uninterruptible-sleeping threads, total | |
2902 | * number of context switches performed since bootup. | |
2903 | */ | |
2904 | unsigned long nr_running(void) | |
2905 | { | |
2906 | unsigned long i, sum = 0; | |
2907 | ||
2908 | for_each_online_cpu(i) | |
2909 | sum += cpu_rq(i)->nr_running; | |
2910 | ||
2911 | return sum; | |
2912 | } | |
2913 | ||
2914 | unsigned long nr_uninterruptible(void) | |
2915 | { | |
2916 | unsigned long i, sum = 0; | |
2917 | ||
0a945022 | 2918 | for_each_possible_cpu(i) |
1da177e4 LT |
2919 | sum += cpu_rq(i)->nr_uninterruptible; |
2920 | ||
2921 | /* | |
2922 | * Since we read the counters lockless, it might be slightly | |
2923 | * inaccurate. Do not allow it to go below zero though: | |
2924 | */ | |
2925 | if (unlikely((long)sum < 0)) | |
2926 | sum = 0; | |
2927 | ||
2928 | return sum; | |
2929 | } | |
2930 | ||
2931 | unsigned long long nr_context_switches(void) | |
2932 | { | |
cc94abfc SR |
2933 | int i; |
2934 | unsigned long long sum = 0; | |
1da177e4 | 2935 | |
0a945022 | 2936 | for_each_possible_cpu(i) |
1da177e4 LT |
2937 | sum += cpu_rq(i)->nr_switches; |
2938 | ||
2939 | return sum; | |
2940 | } | |
2941 | ||
2942 | unsigned long nr_iowait(void) | |
2943 | { | |
2944 | unsigned long i, sum = 0; | |
2945 | ||
0a945022 | 2946 | for_each_possible_cpu(i) |
1da177e4 LT |
2947 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
2948 | ||
2949 | return sum; | |
2950 | } | |
2951 | ||
69d25870 AV |
2952 | unsigned long nr_iowait_cpu(void) |
2953 | { | |
2954 | struct rq *this = this_rq(); | |
2955 | return atomic_read(&this->nr_iowait); | |
2956 | } | |
2957 | ||
2958 | unsigned long this_cpu_load(void) | |
2959 | { | |
2960 | struct rq *this = this_rq(); | |
2961 | return this->cpu_load[0]; | |
2962 | } | |
2963 | ||
2964 | ||
dce48a84 TG |
2965 | /* Variables and functions for calc_load */ |
2966 | static atomic_long_t calc_load_tasks; | |
2967 | static unsigned long calc_load_update; | |
2968 | unsigned long avenrun[3]; | |
2969 | EXPORT_SYMBOL(avenrun); | |
2970 | ||
2d02494f TG |
2971 | /** |
2972 | * get_avenrun - get the load average array | |
2973 | * @loads: pointer to dest load array | |
2974 | * @offset: offset to add | |
2975 | * @shift: shift count to shift the result left | |
2976 | * | |
2977 | * These values are estimates at best, so no need for locking. | |
2978 | */ | |
2979 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2980 | { | |
2981 | loads[0] = (avenrun[0] + offset) << shift; | |
2982 | loads[1] = (avenrun[1] + offset) << shift; | |
2983 | loads[2] = (avenrun[2] + offset) << shift; | |
2984 | } | |
2985 | ||
dce48a84 TG |
2986 | static unsigned long |
2987 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2988 | { |
dce48a84 TG |
2989 | load *= exp; |
2990 | load += active * (FIXED_1 - exp); | |
2991 | return load >> FSHIFT; | |
2992 | } | |
db1b1fef | 2993 | |
dce48a84 TG |
2994 | /* |
2995 | * calc_load - update the avenrun load estimates 10 ticks after the | |
2996 | * CPUs have updated calc_load_tasks. | |
2997 | */ | |
2998 | void calc_global_load(void) | |
2999 | { | |
3000 | unsigned long upd = calc_load_update + 10; | |
3001 | long active; | |
3002 | ||
3003 | if (time_before(jiffies, upd)) | |
3004 | return; | |
db1b1fef | 3005 | |
dce48a84 TG |
3006 | active = atomic_long_read(&calc_load_tasks); |
3007 | active = active > 0 ? active * FIXED_1 : 0; | |
db1b1fef | 3008 | |
dce48a84 TG |
3009 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3010 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3011 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
3012 | ||
3013 | calc_load_update += LOAD_FREQ; | |
3014 | } | |
3015 | ||
3016 | /* | |
3017 | * Either called from update_cpu_load() or from a cpu going idle | |
3018 | */ | |
3019 | static void calc_load_account_active(struct rq *this_rq) | |
3020 | { | |
3021 | long nr_active, delta; | |
3022 | ||
3023 | nr_active = this_rq->nr_running; | |
3024 | nr_active += (long) this_rq->nr_uninterruptible; | |
3025 | ||
3026 | if (nr_active != this_rq->calc_load_active) { | |
3027 | delta = nr_active - this_rq->calc_load_active; | |
3028 | this_rq->calc_load_active = nr_active; | |
3029 | atomic_long_add(delta, &calc_load_tasks); | |
3030 | } | |
db1b1fef JS |
3031 | } |
3032 | ||
48f24c4d | 3033 | /* |
dd41f596 IM |
3034 | * Update rq->cpu_load[] statistics. This function is usually called every |
3035 | * scheduler tick (TICK_NSEC). | |
48f24c4d | 3036 | */ |
dd41f596 | 3037 | static void update_cpu_load(struct rq *this_rq) |
48f24c4d | 3038 | { |
495eca49 | 3039 | unsigned long this_load = this_rq->load.weight; |
dd41f596 IM |
3040 | int i, scale; |
3041 | ||
3042 | this_rq->nr_load_updates++; | |
dd41f596 IM |
3043 | |
3044 | /* Update our load: */ | |
3045 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3046 | unsigned long old_load, new_load; | |
3047 | ||
3048 | /* scale is effectively 1 << i now, and >> i divides by scale */ | |
3049 | ||
3050 | old_load = this_rq->cpu_load[i]; | |
3051 | new_load = this_load; | |
a25707f3 IM |
3052 | /* |
3053 | * Round up the averaging division if load is increasing. This | |
3054 | * prevents us from getting stuck on 9 if the load is 10, for | |
3055 | * example. | |
3056 | */ | |
3057 | if (new_load > old_load) | |
3058 | new_load += scale-1; | |
dd41f596 IM |
3059 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3060 | } | |
dce48a84 TG |
3061 | |
3062 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { | |
3063 | this_rq->calc_load_update += LOAD_FREQ; | |
3064 | calc_load_account_active(this_rq); | |
3065 | } | |
48f24c4d IM |
3066 | } |
3067 | ||
dd41f596 IM |
3068 | #ifdef CONFIG_SMP |
3069 | ||
1da177e4 | 3070 | /* |
38022906 PZ |
3071 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3072 | * this point the task has the smallest effective memory and cache footprint. | |
1da177e4 | 3073 | */ |
38022906 | 3074 | void sched_exec(void) |
1da177e4 | 3075 | { |
38022906 | 3076 | struct task_struct *p = current; |
70b97a7f | 3077 | struct migration_req req; |
38022906 | 3078 | int dest_cpu, this_cpu; |
1da177e4 | 3079 | unsigned long flags; |
70b97a7f | 3080 | struct rq *rq; |
1da177e4 | 3081 | |
38022906 PZ |
3082 | again: |
3083 | this_cpu = get_cpu(); | |
3084 | dest_cpu = select_task_rq(p, SD_BALANCE_EXEC, 0); | |
3085 | if (dest_cpu == this_cpu) { | |
3086 | put_cpu(); | |
3087 | return; | |
3088 | } | |
3089 | ||
1da177e4 | 3090 | rq = task_rq_lock(p, &flags); |
38022906 PZ |
3091 | put_cpu(); |
3092 | ||
3093 | /* | |
3094 | * select_task_rq() can race against ->cpus_allowed | |
3095 | */ | |
96f874e2 | 3096 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
38022906 PZ |
3097 | || unlikely(!cpu_active(dest_cpu))) { |
3098 | task_rq_unlock(rq, &flags); | |
3099 | goto again; | |
3100 | } | |
1da177e4 LT |
3101 | |
3102 | /* force the process onto the specified CPU */ | |
3103 | if (migrate_task(p, dest_cpu, &req)) { | |
3104 | /* Need to wait for migration thread (might exit: take ref). */ | |
3105 | struct task_struct *mt = rq->migration_thread; | |
36c8b586 | 3106 | |
1da177e4 LT |
3107 | get_task_struct(mt); |
3108 | task_rq_unlock(rq, &flags); | |
3109 | wake_up_process(mt); | |
3110 | put_task_struct(mt); | |
3111 | wait_for_completion(&req.done); | |
36c8b586 | 3112 | |
1da177e4 LT |
3113 | return; |
3114 | } | |
1da177e4 LT |
3115 | task_rq_unlock(rq, &flags); |
3116 | } | |
3117 | ||
1da177e4 LT |
3118 | #endif |
3119 | ||
1da177e4 LT |
3120 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3121 | ||
3122 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3123 | ||
3124 | /* | |
c5f8d995 | 3125 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3126 | * @p in case that task is currently running. |
c5f8d995 HS |
3127 | * |
3128 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3129 | */ |
c5f8d995 HS |
3130 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3131 | { | |
3132 | u64 ns = 0; | |
3133 | ||
3134 | if (task_current(rq, p)) { | |
3135 | update_rq_clock(rq); | |
3136 | ns = rq->clock - p->se.exec_start; | |
3137 | if ((s64)ns < 0) | |
3138 | ns = 0; | |
3139 | } | |
3140 | ||
3141 | return ns; | |
3142 | } | |
3143 | ||
bb34d92f | 3144 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3145 | { |
1da177e4 | 3146 | unsigned long flags; |
41b86e9c | 3147 | struct rq *rq; |
bb34d92f | 3148 | u64 ns = 0; |
48f24c4d | 3149 | |
41b86e9c | 3150 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3151 | ns = do_task_delta_exec(p, rq); |
3152 | task_rq_unlock(rq, &flags); | |
1508487e | 3153 | |
c5f8d995 HS |
3154 | return ns; |
3155 | } | |
f06febc9 | 3156 | |
c5f8d995 HS |
3157 | /* |
3158 | * Return accounted runtime for the task. | |
3159 | * In case the task is currently running, return the runtime plus current's | |
3160 | * pending runtime that have not been accounted yet. | |
3161 | */ | |
3162 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3163 | { | |
3164 | unsigned long flags; | |
3165 | struct rq *rq; | |
3166 | u64 ns = 0; | |
3167 | ||
3168 | rq = task_rq_lock(p, &flags); | |
3169 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3170 | task_rq_unlock(rq, &flags); | |
3171 | ||
3172 | return ns; | |
3173 | } | |
48f24c4d | 3174 | |
c5f8d995 HS |
3175 | /* |
3176 | * Return sum_exec_runtime for the thread group. | |
3177 | * In case the task is currently running, return the sum plus current's | |
3178 | * pending runtime that have not been accounted yet. | |
3179 | * | |
3180 | * Note that the thread group might have other running tasks as well, | |
3181 | * so the return value not includes other pending runtime that other | |
3182 | * running tasks might have. | |
3183 | */ | |
3184 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3185 | { | |
3186 | struct task_cputime totals; | |
3187 | unsigned long flags; | |
3188 | struct rq *rq; | |
3189 | u64 ns; | |
3190 | ||
3191 | rq = task_rq_lock(p, &flags); | |
3192 | thread_group_cputime(p, &totals); | |
3193 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3194 | task_rq_unlock(rq, &flags); |
48f24c4d | 3195 | |
1da177e4 LT |
3196 | return ns; |
3197 | } | |
3198 | ||
1da177e4 LT |
3199 | /* |
3200 | * Account user cpu time to a process. | |
3201 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3202 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3203 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3204 | */ |
457533a7 MS |
3205 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3206 | cputime_t cputime_scaled) | |
1da177e4 LT |
3207 | { |
3208 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3209 | cputime64_t tmp; | |
3210 | ||
457533a7 | 3211 | /* Add user time to process. */ |
1da177e4 | 3212 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3213 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3214 | account_group_user_time(p, cputime); |
1da177e4 LT |
3215 | |
3216 | /* Add user time to cpustat. */ | |
3217 | tmp = cputime_to_cputime64(cputime); | |
3218 | if (TASK_NICE(p) > 0) | |
3219 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3220 | else | |
3221 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3222 | |
3223 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3224 | /* Account for user time used */ |
3225 | acct_update_integrals(p); | |
1da177e4 LT |
3226 | } |
3227 | ||
94886b84 LV |
3228 | /* |
3229 | * Account guest cpu time to a process. | |
3230 | * @p: the process that the cpu time gets accounted to | |
3231 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3232 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3233 | */ |
457533a7 MS |
3234 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3235 | cputime_t cputime_scaled) | |
94886b84 LV |
3236 | { |
3237 | cputime64_t tmp; | |
3238 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3239 | ||
3240 | tmp = cputime_to_cputime64(cputime); | |
3241 | ||
457533a7 | 3242 | /* Add guest time to process. */ |
94886b84 | 3243 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3244 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3245 | account_group_user_time(p, cputime); |
94886b84 LV |
3246 | p->gtime = cputime_add(p->gtime, cputime); |
3247 | ||
457533a7 | 3248 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3249 | if (TASK_NICE(p) > 0) { |
3250 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3251 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3252 | } else { | |
3253 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3254 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3255 | } | |
94886b84 LV |
3256 | } |
3257 | ||
1da177e4 LT |
3258 | /* |
3259 | * Account system cpu time to a process. | |
3260 | * @p: the process that the cpu time gets accounted to | |
3261 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3262 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3263 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3264 | */ |
3265 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3266 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3267 | { |
3268 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3269 | cputime64_t tmp; |
3270 | ||
983ed7a6 | 3271 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3272 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3273 | return; |
3274 | } | |
94886b84 | 3275 | |
457533a7 | 3276 | /* Add system time to process. */ |
1da177e4 | 3277 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3278 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3279 | account_group_system_time(p, cputime); |
1da177e4 LT |
3280 | |
3281 | /* Add system time to cpustat. */ | |
3282 | tmp = cputime_to_cputime64(cputime); | |
3283 | if (hardirq_count() - hardirq_offset) | |
3284 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3285 | else if (softirq_count()) | |
3286 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3287 | else |
79741dd3 MS |
3288 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3289 | ||
ef12fefa BR |
3290 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3291 | ||
1da177e4 LT |
3292 | /* Account for system time used */ |
3293 | acct_update_integrals(p); | |
1da177e4 LT |
3294 | } |
3295 | ||
c66f08be | 3296 | /* |
1da177e4 | 3297 | * Account for involuntary wait time. |
1da177e4 | 3298 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3299 | */ |
79741dd3 | 3300 | void account_steal_time(cputime_t cputime) |
c66f08be | 3301 | { |
79741dd3 MS |
3302 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3303 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3304 | ||
3305 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3306 | } |
3307 | ||
1da177e4 | 3308 | /* |
79741dd3 MS |
3309 | * Account for idle time. |
3310 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3311 | */ |
79741dd3 | 3312 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3313 | { |
3314 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3315 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3316 | struct rq *rq = this_rq(); |
1da177e4 | 3317 | |
79741dd3 MS |
3318 | if (atomic_read(&rq->nr_iowait) > 0) |
3319 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3320 | else | |
3321 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3322 | } |
3323 | ||
79741dd3 MS |
3324 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3325 | ||
3326 | /* | |
3327 | * Account a single tick of cpu time. | |
3328 | * @p: the process that the cpu time gets accounted to | |
3329 | * @user_tick: indicates if the tick is a user or a system tick | |
3330 | */ | |
3331 | void account_process_tick(struct task_struct *p, int user_tick) | |
3332 | { | |
a42548a1 | 3333 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3334 | struct rq *rq = this_rq(); |
3335 | ||
3336 | if (user_tick) | |
a42548a1 | 3337 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3338 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3339 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3340 | one_jiffy_scaled); |
3341 | else | |
a42548a1 | 3342 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3343 | } |
3344 | ||
3345 | /* | |
3346 | * Account multiple ticks of steal time. | |
3347 | * @p: the process from which the cpu time has been stolen | |
3348 | * @ticks: number of stolen ticks | |
3349 | */ | |
3350 | void account_steal_ticks(unsigned long ticks) | |
3351 | { | |
3352 | account_steal_time(jiffies_to_cputime(ticks)); | |
3353 | } | |
3354 | ||
3355 | /* | |
3356 | * Account multiple ticks of idle time. | |
3357 | * @ticks: number of stolen ticks | |
3358 | */ | |
3359 | void account_idle_ticks(unsigned long ticks) | |
3360 | { | |
3361 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3362 | } |
3363 | ||
79741dd3 MS |
3364 | #endif |
3365 | ||
49048622 BS |
3366 | /* |
3367 | * Use precise platform statistics if available: | |
3368 | */ | |
3369 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3370 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3371 | { |
d99ca3b9 HS |
3372 | *ut = p->utime; |
3373 | *st = p->stime; | |
49048622 BS |
3374 | } |
3375 | ||
0cf55e1e | 3376 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3377 | { |
0cf55e1e HS |
3378 | struct task_cputime cputime; |
3379 | ||
3380 | thread_group_cputime(p, &cputime); | |
3381 | ||
3382 | *ut = cputime.utime; | |
3383 | *st = cputime.stime; | |
49048622 BS |
3384 | } |
3385 | #else | |
761b1d26 HS |
3386 | |
3387 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3388 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3389 | #endif |
3390 | ||
d180c5bc | 3391 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3392 | { |
d99ca3b9 | 3393 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3394 | |
3395 | /* | |
3396 | * Use CFS's precise accounting: | |
3397 | */ | |
d180c5bc | 3398 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3399 | |
3400 | if (total) { | |
d180c5bc HS |
3401 | u64 temp; |
3402 | ||
3403 | temp = (u64)(rtime * utime); | |
49048622 | 3404 | do_div(temp, total); |
d180c5bc HS |
3405 | utime = (cputime_t)temp; |
3406 | } else | |
3407 | utime = rtime; | |
49048622 | 3408 | |
d180c5bc HS |
3409 | /* |
3410 | * Compare with previous values, to keep monotonicity: | |
3411 | */ | |
761b1d26 | 3412 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3413 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3414 | |
d99ca3b9 HS |
3415 | *ut = p->prev_utime; |
3416 | *st = p->prev_stime; | |
49048622 BS |
3417 | } |
3418 | ||
0cf55e1e HS |
3419 | /* |
3420 | * Must be called with siglock held. | |
3421 | */ | |
3422 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3423 | { |
0cf55e1e HS |
3424 | struct signal_struct *sig = p->signal; |
3425 | struct task_cputime cputime; | |
3426 | cputime_t rtime, utime, total; | |
49048622 | 3427 | |
0cf55e1e | 3428 | thread_group_cputime(p, &cputime); |
49048622 | 3429 | |
0cf55e1e HS |
3430 | total = cputime_add(cputime.utime, cputime.stime); |
3431 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3432 | |
0cf55e1e HS |
3433 | if (total) { |
3434 | u64 temp; | |
49048622 | 3435 | |
0cf55e1e HS |
3436 | temp = (u64)(rtime * cputime.utime); |
3437 | do_div(temp, total); | |
3438 | utime = (cputime_t)temp; | |
3439 | } else | |
3440 | utime = rtime; | |
3441 | ||
3442 | sig->prev_utime = max(sig->prev_utime, utime); | |
3443 | sig->prev_stime = max(sig->prev_stime, | |
3444 | cputime_sub(rtime, sig->prev_utime)); | |
3445 | ||
3446 | *ut = sig->prev_utime; | |
3447 | *st = sig->prev_stime; | |
49048622 | 3448 | } |
49048622 | 3449 | #endif |
49048622 | 3450 | |
7835b98b CL |
3451 | /* |
3452 | * This function gets called by the timer code, with HZ frequency. | |
3453 | * We call it with interrupts disabled. | |
3454 | * | |
3455 | * It also gets called by the fork code, when changing the parent's | |
3456 | * timeslices. | |
3457 | */ | |
3458 | void scheduler_tick(void) | |
3459 | { | |
7835b98b CL |
3460 | int cpu = smp_processor_id(); |
3461 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3462 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3463 | |
3464 | sched_clock_tick(); | |
dd41f596 | 3465 | |
05fa785c | 3466 | raw_spin_lock(&rq->lock); |
3e51f33f | 3467 | update_rq_clock(rq); |
f1a438d8 | 3468 | update_cpu_load(rq); |
fa85ae24 | 3469 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3470 | raw_spin_unlock(&rq->lock); |
7835b98b | 3471 | |
cdd6c482 | 3472 | perf_event_task_tick(curr, cpu); |
e220d2dc | 3473 | |
e418e1c2 | 3474 | #ifdef CONFIG_SMP |
dd41f596 IM |
3475 | rq->idle_at_tick = idle_cpu(cpu); |
3476 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3477 | #endif |
1da177e4 LT |
3478 | } |
3479 | ||
132380a0 | 3480 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3481 | { |
3482 | if (in_lock_functions(addr)) { | |
3483 | addr = CALLER_ADDR2; | |
3484 | if (in_lock_functions(addr)) | |
3485 | addr = CALLER_ADDR3; | |
3486 | } | |
3487 | return addr; | |
3488 | } | |
1da177e4 | 3489 | |
7e49fcce SR |
3490 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3491 | defined(CONFIG_PREEMPT_TRACER)) | |
3492 | ||
43627582 | 3493 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3494 | { |
6cd8a4bb | 3495 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3496 | /* |
3497 | * Underflow? | |
3498 | */ | |
9a11b49a IM |
3499 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3500 | return; | |
6cd8a4bb | 3501 | #endif |
1da177e4 | 3502 | preempt_count() += val; |
6cd8a4bb | 3503 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3504 | /* |
3505 | * Spinlock count overflowing soon? | |
3506 | */ | |
33859f7f MOS |
3507 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3508 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3509 | #endif |
3510 | if (preempt_count() == val) | |
3511 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3512 | } |
3513 | EXPORT_SYMBOL(add_preempt_count); | |
3514 | ||
43627582 | 3515 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3516 | { |
6cd8a4bb | 3517 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3518 | /* |
3519 | * Underflow? | |
3520 | */ | |
01e3eb82 | 3521 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3522 | return; |
1da177e4 LT |
3523 | /* |
3524 | * Is the spinlock portion underflowing? | |
3525 | */ | |
9a11b49a IM |
3526 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3527 | !(preempt_count() & PREEMPT_MASK))) | |
3528 | return; | |
6cd8a4bb | 3529 | #endif |
9a11b49a | 3530 | |
6cd8a4bb SR |
3531 | if (preempt_count() == val) |
3532 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3533 | preempt_count() -= val; |
3534 | } | |
3535 | EXPORT_SYMBOL(sub_preempt_count); | |
3536 | ||
3537 | #endif | |
3538 | ||
3539 | /* | |
dd41f596 | 3540 | * Print scheduling while atomic bug: |
1da177e4 | 3541 | */ |
dd41f596 | 3542 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3543 | { |
838225b4 SS |
3544 | struct pt_regs *regs = get_irq_regs(); |
3545 | ||
3df0fc5b PZ |
3546 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3547 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3548 | |
dd41f596 | 3549 | debug_show_held_locks(prev); |
e21f5b15 | 3550 | print_modules(); |
dd41f596 IM |
3551 | if (irqs_disabled()) |
3552 | print_irqtrace_events(prev); | |
838225b4 SS |
3553 | |
3554 | if (regs) | |
3555 | show_regs(regs); | |
3556 | else | |
3557 | dump_stack(); | |
dd41f596 | 3558 | } |
1da177e4 | 3559 | |
dd41f596 IM |
3560 | /* |
3561 | * Various schedule()-time debugging checks and statistics: | |
3562 | */ | |
3563 | static inline void schedule_debug(struct task_struct *prev) | |
3564 | { | |
1da177e4 | 3565 | /* |
41a2d6cf | 3566 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3567 | * schedule() atomically, we ignore that path for now. |
3568 | * Otherwise, whine if we are scheduling when we should not be. | |
3569 | */ | |
3f33a7ce | 3570 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3571 | __schedule_bug(prev); |
3572 | ||
1da177e4 LT |
3573 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3574 | ||
2d72376b | 3575 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3576 | #ifdef CONFIG_SCHEDSTATS |
3577 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3578 | schedstat_inc(this_rq(), bkl_count); |
3579 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3580 | } |
3581 | #endif | |
dd41f596 IM |
3582 | } |
3583 | ||
6cecd084 | 3584 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3585 | { |
6cecd084 PZ |
3586 | if (prev->state == TASK_RUNNING) { |
3587 | u64 runtime = prev->se.sum_exec_runtime; | |
df1c99d4 | 3588 | |
6cecd084 PZ |
3589 | runtime -= prev->se.prev_sum_exec_runtime; |
3590 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
df1c99d4 MG |
3591 | |
3592 | /* | |
3593 | * In order to avoid avg_overlap growing stale when we are | |
3594 | * indeed overlapping and hence not getting put to sleep, grow | |
3595 | * the avg_overlap on preemption. | |
3596 | * | |
3597 | * We use the average preemption runtime because that | |
3598 | * correlates to the amount of cache footprint a task can | |
3599 | * build up. | |
3600 | */ | |
6cecd084 | 3601 | update_avg(&prev->se.avg_overlap, runtime); |
df1c99d4 | 3602 | } |
6cecd084 | 3603 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3604 | } |
3605 | ||
dd41f596 IM |
3606 | /* |
3607 | * Pick up the highest-prio task: | |
3608 | */ | |
3609 | static inline struct task_struct * | |
b67802ea | 3610 | pick_next_task(struct rq *rq) |
dd41f596 | 3611 | { |
5522d5d5 | 3612 | const struct sched_class *class; |
dd41f596 | 3613 | struct task_struct *p; |
1da177e4 LT |
3614 | |
3615 | /* | |
dd41f596 IM |
3616 | * Optimization: we know that if all tasks are in |
3617 | * the fair class we can call that function directly: | |
1da177e4 | 3618 | */ |
dd41f596 | 3619 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3620 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3621 | if (likely(p)) |
3622 | return p; | |
1da177e4 LT |
3623 | } |
3624 | ||
dd41f596 IM |
3625 | class = sched_class_highest; |
3626 | for ( ; ; ) { | |
fb8d4724 | 3627 | p = class->pick_next_task(rq); |
dd41f596 IM |
3628 | if (p) |
3629 | return p; | |
3630 | /* | |
3631 | * Will never be NULL as the idle class always | |
3632 | * returns a non-NULL p: | |
3633 | */ | |
3634 | class = class->next; | |
3635 | } | |
3636 | } | |
1da177e4 | 3637 | |
dd41f596 IM |
3638 | /* |
3639 | * schedule() is the main scheduler function. | |
3640 | */ | |
ff743345 | 3641 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3642 | { |
3643 | struct task_struct *prev, *next; | |
67ca7bde | 3644 | unsigned long *switch_count; |
dd41f596 | 3645 | struct rq *rq; |
31656519 | 3646 | int cpu; |
dd41f596 | 3647 | |
ff743345 PZ |
3648 | need_resched: |
3649 | preempt_disable(); | |
dd41f596 IM |
3650 | cpu = smp_processor_id(); |
3651 | rq = cpu_rq(cpu); | |
d6714c22 | 3652 | rcu_sched_qs(cpu); |
dd41f596 IM |
3653 | prev = rq->curr; |
3654 | switch_count = &prev->nivcsw; | |
3655 | ||
3656 | release_kernel_lock(prev); | |
3657 | need_resched_nonpreemptible: | |
3658 | ||
3659 | schedule_debug(prev); | |
1da177e4 | 3660 | |
31656519 | 3661 | if (sched_feat(HRTICK)) |
f333fdc9 | 3662 | hrtick_clear(rq); |
8f4d37ec | 3663 | |
05fa785c | 3664 | raw_spin_lock_irq(&rq->lock); |
3e51f33f | 3665 | update_rq_clock(rq); |
1e819950 | 3666 | clear_tsk_need_resched(prev); |
1da177e4 | 3667 | |
1da177e4 | 3668 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 3669 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 3670 | prev->state = TASK_RUNNING; |
16882c1e | 3671 | else |
2e1cb74a | 3672 | deactivate_task(rq, prev, 1); |
dd41f596 | 3673 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3674 | } |
3675 | ||
3f029d3c | 3676 | pre_schedule(rq, prev); |
f65eda4f | 3677 | |
dd41f596 | 3678 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3679 | idle_balance(cpu, rq); |
1da177e4 | 3680 | |
df1c99d4 | 3681 | put_prev_task(rq, prev); |
b67802ea | 3682 | next = pick_next_task(rq); |
1da177e4 | 3683 | |
1da177e4 | 3684 | if (likely(prev != next)) { |
673a90a1 | 3685 | sched_info_switch(prev, next); |
cdd6c482 | 3686 | perf_event_task_sched_out(prev, next, cpu); |
673a90a1 | 3687 | |
1da177e4 LT |
3688 | rq->nr_switches++; |
3689 | rq->curr = next; | |
3690 | ++*switch_count; | |
3691 | ||
dd41f596 | 3692 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3693 | /* |
3694 | * the context switch might have flipped the stack from under | |
3695 | * us, hence refresh the local variables. | |
3696 | */ | |
3697 | cpu = smp_processor_id(); | |
3698 | rq = cpu_rq(cpu); | |
1da177e4 | 3699 | } else |
05fa785c | 3700 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3701 | |
3f029d3c | 3702 | post_schedule(rq); |
1da177e4 | 3703 | |
8f4d37ec | 3704 | if (unlikely(reacquire_kernel_lock(current) < 0)) |
1da177e4 | 3705 | goto need_resched_nonpreemptible; |
8f4d37ec | 3706 | |
1da177e4 | 3707 | preempt_enable_no_resched(); |
ff743345 | 3708 | if (need_resched()) |
1da177e4 LT |
3709 | goto need_resched; |
3710 | } | |
1da177e4 LT |
3711 | EXPORT_SYMBOL(schedule); |
3712 | ||
c08f7829 | 3713 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3714 | /* |
3715 | * Look out! "owner" is an entirely speculative pointer | |
3716 | * access and not reliable. | |
3717 | */ | |
3718 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3719 | { | |
3720 | unsigned int cpu; | |
3721 | struct rq *rq; | |
3722 | ||
3723 | if (!sched_feat(OWNER_SPIN)) | |
3724 | return 0; | |
3725 | ||
3726 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3727 | /* | |
3728 | * Need to access the cpu field knowing that | |
3729 | * DEBUG_PAGEALLOC could have unmapped it if | |
3730 | * the mutex owner just released it and exited. | |
3731 | */ | |
3732 | if (probe_kernel_address(&owner->cpu, cpu)) | |
3733 | goto out; | |
3734 | #else | |
3735 | cpu = owner->cpu; | |
3736 | #endif | |
3737 | ||
3738 | /* | |
3739 | * Even if the access succeeded (likely case), | |
3740 | * the cpu field may no longer be valid. | |
3741 | */ | |
3742 | if (cpu >= nr_cpumask_bits) | |
3743 | goto out; | |
3744 | ||
3745 | /* | |
3746 | * We need to validate that we can do a | |
3747 | * get_cpu() and that we have the percpu area. | |
3748 | */ | |
3749 | if (!cpu_online(cpu)) | |
3750 | goto out; | |
3751 | ||
3752 | rq = cpu_rq(cpu); | |
3753 | ||
3754 | for (;;) { | |
3755 | /* | |
3756 | * Owner changed, break to re-assess state. | |
3757 | */ | |
3758 | if (lock->owner != owner) | |
3759 | break; | |
3760 | ||
3761 | /* | |
3762 | * Is that owner really running on that cpu? | |
3763 | */ | |
3764 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3765 | return 0; | |
3766 | ||
3767 | cpu_relax(); | |
3768 | } | |
3769 | out: | |
3770 | return 1; | |
3771 | } | |
3772 | #endif | |
3773 | ||
1da177e4 LT |
3774 | #ifdef CONFIG_PREEMPT |
3775 | /* | |
2ed6e34f | 3776 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3777 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3778 | * occur there and call schedule directly. |
3779 | */ | |
3780 | asmlinkage void __sched preempt_schedule(void) | |
3781 | { | |
3782 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3783 | |
1da177e4 LT |
3784 | /* |
3785 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3786 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3787 | */ |
beed33a8 | 3788 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3789 | return; |
3790 | ||
3a5c359a AK |
3791 | do { |
3792 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3793 | schedule(); |
3a5c359a | 3794 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3795 | |
3a5c359a AK |
3796 | /* |
3797 | * Check again in case we missed a preemption opportunity | |
3798 | * between schedule and now. | |
3799 | */ | |
3800 | barrier(); | |
5ed0cec0 | 3801 | } while (need_resched()); |
1da177e4 | 3802 | } |
1da177e4 LT |
3803 | EXPORT_SYMBOL(preempt_schedule); |
3804 | ||
3805 | /* | |
2ed6e34f | 3806 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3807 | * off of irq context. |
3808 | * Note, that this is called and return with irqs disabled. This will | |
3809 | * protect us against recursive calling from irq. | |
3810 | */ | |
3811 | asmlinkage void __sched preempt_schedule_irq(void) | |
3812 | { | |
3813 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3814 | |
2ed6e34f | 3815 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3816 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3817 | ||
3a5c359a AK |
3818 | do { |
3819 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3820 | local_irq_enable(); |
3821 | schedule(); | |
3822 | local_irq_disable(); | |
3a5c359a | 3823 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3824 | |
3a5c359a AK |
3825 | /* |
3826 | * Check again in case we missed a preemption opportunity | |
3827 | * between schedule and now. | |
3828 | */ | |
3829 | barrier(); | |
5ed0cec0 | 3830 | } while (need_resched()); |
1da177e4 LT |
3831 | } |
3832 | ||
3833 | #endif /* CONFIG_PREEMPT */ | |
3834 | ||
63859d4f | 3835 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3836 | void *key) |
1da177e4 | 3837 | { |
63859d4f | 3838 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3839 | } |
1da177e4 LT |
3840 | EXPORT_SYMBOL(default_wake_function); |
3841 | ||
3842 | /* | |
41a2d6cf IM |
3843 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3844 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3845 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3846 | * | |
3847 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3848 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3849 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3850 | */ | |
78ddb08f | 3851 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3852 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3853 | { |
2e45874c | 3854 | wait_queue_t *curr, *next; |
1da177e4 | 3855 | |
2e45874c | 3856 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3857 | unsigned flags = curr->flags; |
3858 | ||
63859d4f | 3859 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3860 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3861 | break; |
3862 | } | |
3863 | } | |
3864 | ||
3865 | /** | |
3866 | * __wake_up - wake up threads blocked on a waitqueue. | |
3867 | * @q: the waitqueue | |
3868 | * @mode: which threads | |
3869 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3870 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3871 | * |
3872 | * It may be assumed that this function implies a write memory barrier before | |
3873 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3874 | */ |
7ad5b3a5 | 3875 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3876 | int nr_exclusive, void *key) |
1da177e4 LT |
3877 | { |
3878 | unsigned long flags; | |
3879 | ||
3880 | spin_lock_irqsave(&q->lock, flags); | |
3881 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3882 | spin_unlock_irqrestore(&q->lock, flags); | |
3883 | } | |
1da177e4 LT |
3884 | EXPORT_SYMBOL(__wake_up); |
3885 | ||
3886 | /* | |
3887 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3888 | */ | |
7ad5b3a5 | 3889 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3890 | { |
3891 | __wake_up_common(q, mode, 1, 0, NULL); | |
3892 | } | |
3893 | ||
4ede816a DL |
3894 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3895 | { | |
3896 | __wake_up_common(q, mode, 1, 0, key); | |
3897 | } | |
3898 | ||
1da177e4 | 3899 | /** |
4ede816a | 3900 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3901 | * @q: the waitqueue |
3902 | * @mode: which threads | |
3903 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3904 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3905 | * |
3906 | * The sync wakeup differs that the waker knows that it will schedule | |
3907 | * away soon, so while the target thread will be woken up, it will not | |
3908 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3909 | * with each other. This can prevent needless bouncing between CPUs. | |
3910 | * | |
3911 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3912 | * |
3913 | * It may be assumed that this function implies a write memory barrier before | |
3914 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3915 | */ |
4ede816a DL |
3916 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3917 | int nr_exclusive, void *key) | |
1da177e4 LT |
3918 | { |
3919 | unsigned long flags; | |
7d478721 | 3920 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3921 | |
3922 | if (unlikely(!q)) | |
3923 | return; | |
3924 | ||
3925 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3926 | wake_flags = 0; |
1da177e4 LT |
3927 | |
3928 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3929 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3930 | spin_unlock_irqrestore(&q->lock, flags); |
3931 | } | |
4ede816a DL |
3932 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3933 | ||
3934 | /* | |
3935 | * __wake_up_sync - see __wake_up_sync_key() | |
3936 | */ | |
3937 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3938 | { | |
3939 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3940 | } | |
1da177e4 LT |
3941 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3942 | ||
65eb3dc6 KD |
3943 | /** |
3944 | * complete: - signals a single thread waiting on this completion | |
3945 | * @x: holds the state of this particular completion | |
3946 | * | |
3947 | * This will wake up a single thread waiting on this completion. Threads will be | |
3948 | * awakened in the same order in which they were queued. | |
3949 | * | |
3950 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3951 | * |
3952 | * It may be assumed that this function implies a write memory barrier before | |
3953 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3954 | */ |
b15136e9 | 3955 | void complete(struct completion *x) |
1da177e4 LT |
3956 | { |
3957 | unsigned long flags; | |
3958 | ||
3959 | spin_lock_irqsave(&x->wait.lock, flags); | |
3960 | x->done++; | |
d9514f6c | 3961 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3962 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3963 | } | |
3964 | EXPORT_SYMBOL(complete); | |
3965 | ||
65eb3dc6 KD |
3966 | /** |
3967 | * complete_all: - signals all threads waiting on this completion | |
3968 | * @x: holds the state of this particular completion | |
3969 | * | |
3970 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3971 | * |
3972 | * It may be assumed that this function implies a write memory barrier before | |
3973 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3974 | */ |
b15136e9 | 3975 | void complete_all(struct completion *x) |
1da177e4 LT |
3976 | { |
3977 | unsigned long flags; | |
3978 | ||
3979 | spin_lock_irqsave(&x->wait.lock, flags); | |
3980 | x->done += UINT_MAX/2; | |
d9514f6c | 3981 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3982 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3983 | } | |
3984 | EXPORT_SYMBOL(complete_all); | |
3985 | ||
8cbbe86d AK |
3986 | static inline long __sched |
3987 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3988 | { |
1da177e4 LT |
3989 | if (!x->done) { |
3990 | DECLARE_WAITQUEUE(wait, current); | |
3991 | ||
3992 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3993 | __add_wait_queue_tail(&x->wait, &wait); | |
3994 | do { | |
94d3d824 | 3995 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3996 | timeout = -ERESTARTSYS; |
3997 | break; | |
8cbbe86d AK |
3998 | } |
3999 | __set_current_state(state); | |
1da177e4 LT |
4000 | spin_unlock_irq(&x->wait.lock); |
4001 | timeout = schedule_timeout(timeout); | |
4002 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4003 | } while (!x->done && timeout); |
1da177e4 | 4004 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4005 | if (!x->done) |
4006 | return timeout; | |
1da177e4 LT |
4007 | } |
4008 | x->done--; | |
ea71a546 | 4009 | return timeout ?: 1; |
1da177e4 | 4010 | } |
1da177e4 | 4011 | |
8cbbe86d AK |
4012 | static long __sched |
4013 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4014 | { |
1da177e4 LT |
4015 | might_sleep(); |
4016 | ||
4017 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4018 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4019 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4020 | return timeout; |
4021 | } | |
1da177e4 | 4022 | |
65eb3dc6 KD |
4023 | /** |
4024 | * wait_for_completion: - waits for completion of a task | |
4025 | * @x: holds the state of this particular completion | |
4026 | * | |
4027 | * This waits to be signaled for completion of a specific task. It is NOT | |
4028 | * interruptible and there is no timeout. | |
4029 | * | |
4030 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4031 | * and interrupt capability. Also see complete(). | |
4032 | */ | |
b15136e9 | 4033 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4034 | { |
4035 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4036 | } |
8cbbe86d | 4037 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4038 | |
65eb3dc6 KD |
4039 | /** |
4040 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4041 | * @x: holds the state of this particular completion | |
4042 | * @timeout: timeout value in jiffies | |
4043 | * | |
4044 | * This waits for either a completion of a specific task to be signaled or for a | |
4045 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4046 | * interruptible. | |
4047 | */ | |
b15136e9 | 4048 | unsigned long __sched |
8cbbe86d | 4049 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4050 | { |
8cbbe86d | 4051 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4052 | } |
8cbbe86d | 4053 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4054 | |
65eb3dc6 KD |
4055 | /** |
4056 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4057 | * @x: holds the state of this particular completion | |
4058 | * | |
4059 | * This waits for completion of a specific task to be signaled. It is | |
4060 | * interruptible. | |
4061 | */ | |
8cbbe86d | 4062 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4063 | { |
51e97990 AK |
4064 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4065 | if (t == -ERESTARTSYS) | |
4066 | return t; | |
4067 | return 0; | |
0fec171c | 4068 | } |
8cbbe86d | 4069 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4070 | |
65eb3dc6 KD |
4071 | /** |
4072 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4073 | * @x: holds the state of this particular completion | |
4074 | * @timeout: timeout value in jiffies | |
4075 | * | |
4076 | * This waits for either a completion of a specific task to be signaled or for a | |
4077 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4078 | */ | |
b15136e9 | 4079 | unsigned long __sched |
8cbbe86d AK |
4080 | wait_for_completion_interruptible_timeout(struct completion *x, |
4081 | unsigned long timeout) | |
0fec171c | 4082 | { |
8cbbe86d | 4083 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4084 | } |
8cbbe86d | 4085 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4086 | |
65eb3dc6 KD |
4087 | /** |
4088 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4089 | * @x: holds the state of this particular completion | |
4090 | * | |
4091 | * This waits to be signaled for completion of a specific task. It can be | |
4092 | * interrupted by a kill signal. | |
4093 | */ | |
009e577e MW |
4094 | int __sched wait_for_completion_killable(struct completion *x) |
4095 | { | |
4096 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4097 | if (t == -ERESTARTSYS) | |
4098 | return t; | |
4099 | return 0; | |
4100 | } | |
4101 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4102 | ||
be4de352 DC |
4103 | /** |
4104 | * try_wait_for_completion - try to decrement a completion without blocking | |
4105 | * @x: completion structure | |
4106 | * | |
4107 | * Returns: 0 if a decrement cannot be done without blocking | |
4108 | * 1 if a decrement succeeded. | |
4109 | * | |
4110 | * If a completion is being used as a counting completion, | |
4111 | * attempt to decrement the counter without blocking. This | |
4112 | * enables us to avoid waiting if the resource the completion | |
4113 | * is protecting is not available. | |
4114 | */ | |
4115 | bool try_wait_for_completion(struct completion *x) | |
4116 | { | |
7539a3b3 | 4117 | unsigned long flags; |
be4de352 DC |
4118 | int ret = 1; |
4119 | ||
7539a3b3 | 4120 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4121 | if (!x->done) |
4122 | ret = 0; | |
4123 | else | |
4124 | x->done--; | |
7539a3b3 | 4125 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4126 | return ret; |
4127 | } | |
4128 | EXPORT_SYMBOL(try_wait_for_completion); | |
4129 | ||
4130 | /** | |
4131 | * completion_done - Test to see if a completion has any waiters | |
4132 | * @x: completion structure | |
4133 | * | |
4134 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4135 | * 1 if there are no waiters. | |
4136 | * | |
4137 | */ | |
4138 | bool completion_done(struct completion *x) | |
4139 | { | |
7539a3b3 | 4140 | unsigned long flags; |
be4de352 DC |
4141 | int ret = 1; |
4142 | ||
7539a3b3 | 4143 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4144 | if (!x->done) |
4145 | ret = 0; | |
7539a3b3 | 4146 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4147 | return ret; |
4148 | } | |
4149 | EXPORT_SYMBOL(completion_done); | |
4150 | ||
8cbbe86d AK |
4151 | static long __sched |
4152 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4153 | { |
0fec171c IM |
4154 | unsigned long flags; |
4155 | wait_queue_t wait; | |
4156 | ||
4157 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4158 | |
8cbbe86d | 4159 | __set_current_state(state); |
1da177e4 | 4160 | |
8cbbe86d AK |
4161 | spin_lock_irqsave(&q->lock, flags); |
4162 | __add_wait_queue(q, &wait); | |
4163 | spin_unlock(&q->lock); | |
4164 | timeout = schedule_timeout(timeout); | |
4165 | spin_lock_irq(&q->lock); | |
4166 | __remove_wait_queue(q, &wait); | |
4167 | spin_unlock_irqrestore(&q->lock, flags); | |
4168 | ||
4169 | return timeout; | |
4170 | } | |
4171 | ||
4172 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4173 | { | |
4174 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4175 | } |
1da177e4 LT |
4176 | EXPORT_SYMBOL(interruptible_sleep_on); |
4177 | ||
0fec171c | 4178 | long __sched |
95cdf3b7 | 4179 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4180 | { |
8cbbe86d | 4181 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4182 | } |
1da177e4 LT |
4183 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4184 | ||
0fec171c | 4185 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4186 | { |
8cbbe86d | 4187 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4188 | } |
1da177e4 LT |
4189 | EXPORT_SYMBOL(sleep_on); |
4190 | ||
0fec171c | 4191 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4192 | { |
8cbbe86d | 4193 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4194 | } |
1da177e4 LT |
4195 | EXPORT_SYMBOL(sleep_on_timeout); |
4196 | ||
b29739f9 IM |
4197 | #ifdef CONFIG_RT_MUTEXES |
4198 | ||
4199 | /* | |
4200 | * rt_mutex_setprio - set the current priority of a task | |
4201 | * @p: task | |
4202 | * @prio: prio value (kernel-internal form) | |
4203 | * | |
4204 | * This function changes the 'effective' priority of a task. It does | |
4205 | * not touch ->normal_prio like __setscheduler(). | |
4206 | * | |
4207 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4208 | */ | |
36c8b586 | 4209 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4210 | { |
4211 | unsigned long flags; | |
83b699ed | 4212 | int oldprio, on_rq, running; |
70b97a7f | 4213 | struct rq *rq; |
cb469845 | 4214 | const struct sched_class *prev_class = p->sched_class; |
b29739f9 IM |
4215 | |
4216 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4217 | ||
4218 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4219 | update_rq_clock(rq); |
b29739f9 | 4220 | |
d5f9f942 | 4221 | oldprio = p->prio; |
dd41f596 | 4222 | on_rq = p->se.on_rq; |
051a1d1a | 4223 | running = task_current(rq, p); |
0e1f3483 | 4224 | if (on_rq) |
69be72c1 | 4225 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4226 | if (running) |
4227 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4228 | |
4229 | if (rt_prio(prio)) | |
4230 | p->sched_class = &rt_sched_class; | |
4231 | else | |
4232 | p->sched_class = &fair_sched_class; | |
4233 | ||
b29739f9 IM |
4234 | p->prio = prio; |
4235 | ||
0e1f3483 HS |
4236 | if (running) |
4237 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4238 | if (on_rq) { |
8159f87e | 4239 | enqueue_task(rq, p, 0); |
cb469845 SR |
4240 | |
4241 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4242 | } |
4243 | task_rq_unlock(rq, &flags); | |
4244 | } | |
4245 | ||
4246 | #endif | |
4247 | ||
36c8b586 | 4248 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4249 | { |
dd41f596 | 4250 | int old_prio, delta, on_rq; |
1da177e4 | 4251 | unsigned long flags; |
70b97a7f | 4252 | struct rq *rq; |
1da177e4 LT |
4253 | |
4254 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4255 | return; | |
4256 | /* | |
4257 | * We have to be careful, if called from sys_setpriority(), | |
4258 | * the task might be in the middle of scheduling on another CPU. | |
4259 | */ | |
4260 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4261 | update_rq_clock(rq); |
1da177e4 LT |
4262 | /* |
4263 | * The RT priorities are set via sched_setscheduler(), but we still | |
4264 | * allow the 'normal' nice value to be set - but as expected | |
4265 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4266 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4267 | */ |
e05606d3 | 4268 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4269 | p->static_prio = NICE_TO_PRIO(nice); |
4270 | goto out_unlock; | |
4271 | } | |
dd41f596 | 4272 | on_rq = p->se.on_rq; |
c09595f6 | 4273 | if (on_rq) |
69be72c1 | 4274 | dequeue_task(rq, p, 0); |
1da177e4 | 4275 | |
1da177e4 | 4276 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4277 | set_load_weight(p); |
b29739f9 IM |
4278 | old_prio = p->prio; |
4279 | p->prio = effective_prio(p); | |
4280 | delta = p->prio - old_prio; | |
1da177e4 | 4281 | |
dd41f596 | 4282 | if (on_rq) { |
8159f87e | 4283 | enqueue_task(rq, p, 0); |
1da177e4 | 4284 | /* |
d5f9f942 AM |
4285 | * If the task increased its priority or is running and |
4286 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4287 | */ |
d5f9f942 | 4288 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4289 | resched_task(rq->curr); |
4290 | } | |
4291 | out_unlock: | |
4292 | task_rq_unlock(rq, &flags); | |
4293 | } | |
1da177e4 LT |
4294 | EXPORT_SYMBOL(set_user_nice); |
4295 | ||
e43379f1 MM |
4296 | /* |
4297 | * can_nice - check if a task can reduce its nice value | |
4298 | * @p: task | |
4299 | * @nice: nice value | |
4300 | */ | |
36c8b586 | 4301 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4302 | { |
024f4747 MM |
4303 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4304 | int nice_rlim = 20 - nice; | |
48f24c4d | 4305 | |
e43379f1 MM |
4306 | return (nice_rlim <= p->signal->rlim[RLIMIT_NICE].rlim_cur || |
4307 | capable(CAP_SYS_NICE)); | |
4308 | } | |
4309 | ||
1da177e4 LT |
4310 | #ifdef __ARCH_WANT_SYS_NICE |
4311 | ||
4312 | /* | |
4313 | * sys_nice - change the priority of the current process. | |
4314 | * @increment: priority increment | |
4315 | * | |
4316 | * sys_setpriority is a more generic, but much slower function that | |
4317 | * does similar things. | |
4318 | */ | |
5add95d4 | 4319 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4320 | { |
48f24c4d | 4321 | long nice, retval; |
1da177e4 LT |
4322 | |
4323 | /* | |
4324 | * Setpriority might change our priority at the same moment. | |
4325 | * We don't have to worry. Conceptually one call occurs first | |
4326 | * and we have a single winner. | |
4327 | */ | |
e43379f1 MM |
4328 | if (increment < -40) |
4329 | increment = -40; | |
1da177e4 LT |
4330 | if (increment > 40) |
4331 | increment = 40; | |
4332 | ||
2b8f836f | 4333 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4334 | if (nice < -20) |
4335 | nice = -20; | |
4336 | if (nice > 19) | |
4337 | nice = 19; | |
4338 | ||
e43379f1 MM |
4339 | if (increment < 0 && !can_nice(current, nice)) |
4340 | return -EPERM; | |
4341 | ||
1da177e4 LT |
4342 | retval = security_task_setnice(current, nice); |
4343 | if (retval) | |
4344 | return retval; | |
4345 | ||
4346 | set_user_nice(current, nice); | |
4347 | return 0; | |
4348 | } | |
4349 | ||
4350 | #endif | |
4351 | ||
4352 | /** | |
4353 | * task_prio - return the priority value of a given task. | |
4354 | * @p: the task in question. | |
4355 | * | |
4356 | * This is the priority value as seen by users in /proc. | |
4357 | * RT tasks are offset by -200. Normal tasks are centered | |
4358 | * around 0, value goes from -16 to +15. | |
4359 | */ | |
36c8b586 | 4360 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4361 | { |
4362 | return p->prio - MAX_RT_PRIO; | |
4363 | } | |
4364 | ||
4365 | /** | |
4366 | * task_nice - return the nice value of a given task. | |
4367 | * @p: the task in question. | |
4368 | */ | |
36c8b586 | 4369 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4370 | { |
4371 | return TASK_NICE(p); | |
4372 | } | |
150d8bed | 4373 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4374 | |
4375 | /** | |
4376 | * idle_cpu - is a given cpu idle currently? | |
4377 | * @cpu: the processor in question. | |
4378 | */ | |
4379 | int idle_cpu(int cpu) | |
4380 | { | |
4381 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4382 | } | |
4383 | ||
1da177e4 LT |
4384 | /** |
4385 | * idle_task - return the idle task for a given cpu. | |
4386 | * @cpu: the processor in question. | |
4387 | */ | |
36c8b586 | 4388 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4389 | { |
4390 | return cpu_rq(cpu)->idle; | |
4391 | } | |
4392 | ||
4393 | /** | |
4394 | * find_process_by_pid - find a process with a matching PID value. | |
4395 | * @pid: the pid in question. | |
4396 | */ | |
a9957449 | 4397 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4398 | { |
228ebcbe | 4399 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4400 | } |
4401 | ||
4402 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4403 | static void |
4404 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4405 | { |
dd41f596 | 4406 | BUG_ON(p->se.on_rq); |
48f24c4d | 4407 | |
1da177e4 LT |
4408 | p->policy = policy; |
4409 | p->rt_priority = prio; | |
b29739f9 IM |
4410 | p->normal_prio = normal_prio(p); |
4411 | /* we are holding p->pi_lock already */ | |
4412 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4413 | if (rt_prio(p->prio)) |
4414 | p->sched_class = &rt_sched_class; | |
4415 | else | |
4416 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4417 | set_load_weight(p); |
1da177e4 LT |
4418 | } |
4419 | ||
c69e8d9c DH |
4420 | /* |
4421 | * check the target process has a UID that matches the current process's | |
4422 | */ | |
4423 | static bool check_same_owner(struct task_struct *p) | |
4424 | { | |
4425 | const struct cred *cred = current_cred(), *pcred; | |
4426 | bool match; | |
4427 | ||
4428 | rcu_read_lock(); | |
4429 | pcred = __task_cred(p); | |
4430 | match = (cred->euid == pcred->euid || | |
4431 | cred->euid == pcred->uid); | |
4432 | rcu_read_unlock(); | |
4433 | return match; | |
4434 | } | |
4435 | ||
961ccddd RR |
4436 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4437 | struct sched_param *param, bool user) | |
1da177e4 | 4438 | { |
83b699ed | 4439 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4440 | unsigned long flags; |
cb469845 | 4441 | const struct sched_class *prev_class = p->sched_class; |
70b97a7f | 4442 | struct rq *rq; |
ca94c442 | 4443 | int reset_on_fork; |
1da177e4 | 4444 | |
66e5393a SR |
4445 | /* may grab non-irq protected spin_locks */ |
4446 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4447 | recheck: |
4448 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4449 | if (policy < 0) { |
4450 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4451 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4452 | } else { |
4453 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4454 | policy &= ~SCHED_RESET_ON_FORK; | |
4455 | ||
4456 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4457 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4458 | policy != SCHED_IDLE) | |
4459 | return -EINVAL; | |
4460 | } | |
4461 | ||
1da177e4 LT |
4462 | /* |
4463 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4464 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4465 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4466 | */ |
4467 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4468 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4469 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4470 | return -EINVAL; |
e05606d3 | 4471 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4472 | return -EINVAL; |
4473 | ||
37e4ab3f OC |
4474 | /* |
4475 | * Allow unprivileged RT tasks to decrease priority: | |
4476 | */ | |
961ccddd | 4477 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4478 | if (rt_policy(policy)) { |
8dc3e909 | 4479 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4480 | |
4481 | if (!lock_task_sighand(p, &flags)) | |
4482 | return -ESRCH; | |
4483 | rlim_rtprio = p->signal->rlim[RLIMIT_RTPRIO].rlim_cur; | |
4484 | unlock_task_sighand(p, &flags); | |
4485 | ||
4486 | /* can't set/change the rt policy */ | |
4487 | if (policy != p->policy && !rlim_rtprio) | |
4488 | return -EPERM; | |
4489 | ||
4490 | /* can't increase priority */ | |
4491 | if (param->sched_priority > p->rt_priority && | |
4492 | param->sched_priority > rlim_rtprio) | |
4493 | return -EPERM; | |
4494 | } | |
dd41f596 IM |
4495 | /* |
4496 | * Like positive nice levels, dont allow tasks to | |
4497 | * move out of SCHED_IDLE either: | |
4498 | */ | |
4499 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4500 | return -EPERM; | |
5fe1d75f | 4501 | |
37e4ab3f | 4502 | /* can't change other user's priorities */ |
c69e8d9c | 4503 | if (!check_same_owner(p)) |
37e4ab3f | 4504 | return -EPERM; |
ca94c442 LP |
4505 | |
4506 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4507 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4508 | return -EPERM; | |
37e4ab3f | 4509 | } |
1da177e4 | 4510 | |
725aad24 | 4511 | if (user) { |
b68aa230 | 4512 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
4513 | /* |
4514 | * Do not allow realtime tasks into groups that have no runtime | |
4515 | * assigned. | |
4516 | */ | |
9a7e0b18 PZ |
4517 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
4518 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 4519 | return -EPERM; |
b68aa230 PZ |
4520 | #endif |
4521 | ||
725aad24 JF |
4522 | retval = security_task_setscheduler(p, policy, param); |
4523 | if (retval) | |
4524 | return retval; | |
4525 | } | |
4526 | ||
b29739f9 IM |
4527 | /* |
4528 | * make sure no PI-waiters arrive (or leave) while we are | |
4529 | * changing the priority of the task: | |
4530 | */ | |
1d615482 | 4531 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4532 | /* |
4533 | * To be able to change p->policy safely, the apropriate | |
4534 | * runqueue lock must be held. | |
4535 | */ | |
b29739f9 | 4536 | rq = __task_rq_lock(p); |
1da177e4 LT |
4537 | /* recheck policy now with rq lock held */ |
4538 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4539 | policy = oldpolicy = -1; | |
b29739f9 | 4540 | __task_rq_unlock(rq); |
1d615482 | 4541 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4542 | goto recheck; |
4543 | } | |
2daa3577 | 4544 | update_rq_clock(rq); |
dd41f596 | 4545 | on_rq = p->se.on_rq; |
051a1d1a | 4546 | running = task_current(rq, p); |
0e1f3483 | 4547 | if (on_rq) |
2e1cb74a | 4548 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4549 | if (running) |
4550 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4551 | |
ca94c442 LP |
4552 | p->sched_reset_on_fork = reset_on_fork; |
4553 | ||
1da177e4 | 4554 | oldprio = p->prio; |
dd41f596 | 4555 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4556 | |
0e1f3483 HS |
4557 | if (running) |
4558 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4559 | if (on_rq) { |
4560 | activate_task(rq, p, 0); | |
cb469845 SR |
4561 | |
4562 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4563 | } |
b29739f9 | 4564 | __task_rq_unlock(rq); |
1d615482 | 4565 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4566 | |
95e02ca9 TG |
4567 | rt_mutex_adjust_pi(p); |
4568 | ||
1da177e4 LT |
4569 | return 0; |
4570 | } | |
961ccddd RR |
4571 | |
4572 | /** | |
4573 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4574 | * @p: the task in question. | |
4575 | * @policy: new policy. | |
4576 | * @param: structure containing the new RT priority. | |
4577 | * | |
4578 | * NOTE that the task may be already dead. | |
4579 | */ | |
4580 | int sched_setscheduler(struct task_struct *p, int policy, | |
4581 | struct sched_param *param) | |
4582 | { | |
4583 | return __sched_setscheduler(p, policy, param, true); | |
4584 | } | |
1da177e4 LT |
4585 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4586 | ||
961ccddd RR |
4587 | /** |
4588 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4589 | * @p: the task in question. | |
4590 | * @policy: new policy. | |
4591 | * @param: structure containing the new RT priority. | |
4592 | * | |
4593 | * Just like sched_setscheduler, only don't bother checking if the | |
4594 | * current context has permission. For example, this is needed in | |
4595 | * stop_machine(): we create temporary high priority worker threads, | |
4596 | * but our caller might not have that capability. | |
4597 | */ | |
4598 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4599 | struct sched_param *param) | |
4600 | { | |
4601 | return __sched_setscheduler(p, policy, param, false); | |
4602 | } | |
4603 | ||
95cdf3b7 IM |
4604 | static int |
4605 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4606 | { |
1da177e4 LT |
4607 | struct sched_param lparam; |
4608 | struct task_struct *p; | |
36c8b586 | 4609 | int retval; |
1da177e4 LT |
4610 | |
4611 | if (!param || pid < 0) | |
4612 | return -EINVAL; | |
4613 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4614 | return -EFAULT; | |
5fe1d75f ON |
4615 | |
4616 | rcu_read_lock(); | |
4617 | retval = -ESRCH; | |
1da177e4 | 4618 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4619 | if (p != NULL) |
4620 | retval = sched_setscheduler(p, policy, &lparam); | |
4621 | rcu_read_unlock(); | |
36c8b586 | 4622 | |
1da177e4 LT |
4623 | return retval; |
4624 | } | |
4625 | ||
4626 | /** | |
4627 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4628 | * @pid: the pid in question. | |
4629 | * @policy: new policy. | |
4630 | * @param: structure containing the new RT priority. | |
4631 | */ | |
5add95d4 HC |
4632 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4633 | struct sched_param __user *, param) | |
1da177e4 | 4634 | { |
c21761f1 JB |
4635 | /* negative values for policy are not valid */ |
4636 | if (policy < 0) | |
4637 | return -EINVAL; | |
4638 | ||
1da177e4 LT |
4639 | return do_sched_setscheduler(pid, policy, param); |
4640 | } | |
4641 | ||
4642 | /** | |
4643 | * sys_sched_setparam - set/change the RT priority of a thread | |
4644 | * @pid: the pid in question. | |
4645 | * @param: structure containing the new RT priority. | |
4646 | */ | |
5add95d4 | 4647 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4648 | { |
4649 | return do_sched_setscheduler(pid, -1, param); | |
4650 | } | |
4651 | ||
4652 | /** | |
4653 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4654 | * @pid: the pid in question. | |
4655 | */ | |
5add95d4 | 4656 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4657 | { |
36c8b586 | 4658 | struct task_struct *p; |
3a5c359a | 4659 | int retval; |
1da177e4 LT |
4660 | |
4661 | if (pid < 0) | |
3a5c359a | 4662 | return -EINVAL; |
1da177e4 LT |
4663 | |
4664 | retval = -ESRCH; | |
5fe85be0 | 4665 | rcu_read_lock(); |
1da177e4 LT |
4666 | p = find_process_by_pid(pid); |
4667 | if (p) { | |
4668 | retval = security_task_getscheduler(p); | |
4669 | if (!retval) | |
ca94c442 LP |
4670 | retval = p->policy |
4671 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4672 | } |
5fe85be0 | 4673 | rcu_read_unlock(); |
1da177e4 LT |
4674 | return retval; |
4675 | } | |
4676 | ||
4677 | /** | |
ca94c442 | 4678 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4679 | * @pid: the pid in question. |
4680 | * @param: structure containing the RT priority. | |
4681 | */ | |
5add95d4 | 4682 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4683 | { |
4684 | struct sched_param lp; | |
36c8b586 | 4685 | struct task_struct *p; |
3a5c359a | 4686 | int retval; |
1da177e4 LT |
4687 | |
4688 | if (!param || pid < 0) | |
3a5c359a | 4689 | return -EINVAL; |
1da177e4 | 4690 | |
5fe85be0 | 4691 | rcu_read_lock(); |
1da177e4 LT |
4692 | p = find_process_by_pid(pid); |
4693 | retval = -ESRCH; | |
4694 | if (!p) | |
4695 | goto out_unlock; | |
4696 | ||
4697 | retval = security_task_getscheduler(p); | |
4698 | if (retval) | |
4699 | goto out_unlock; | |
4700 | ||
4701 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4702 | rcu_read_unlock(); |
1da177e4 LT |
4703 | |
4704 | /* | |
4705 | * This one might sleep, we cannot do it with a spinlock held ... | |
4706 | */ | |
4707 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4708 | ||
1da177e4 LT |
4709 | return retval; |
4710 | ||
4711 | out_unlock: | |
5fe85be0 | 4712 | rcu_read_unlock(); |
1da177e4 LT |
4713 | return retval; |
4714 | } | |
4715 | ||
96f874e2 | 4716 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4717 | { |
5a16f3d3 | 4718 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4719 | struct task_struct *p; |
4720 | int retval; | |
1da177e4 | 4721 | |
95402b38 | 4722 | get_online_cpus(); |
23f5d142 | 4723 | rcu_read_lock(); |
1da177e4 LT |
4724 | |
4725 | p = find_process_by_pid(pid); | |
4726 | if (!p) { | |
23f5d142 | 4727 | rcu_read_unlock(); |
95402b38 | 4728 | put_online_cpus(); |
1da177e4 LT |
4729 | return -ESRCH; |
4730 | } | |
4731 | ||
23f5d142 | 4732 | /* Prevent p going away */ |
1da177e4 | 4733 | get_task_struct(p); |
23f5d142 | 4734 | rcu_read_unlock(); |
1da177e4 | 4735 | |
5a16f3d3 RR |
4736 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4737 | retval = -ENOMEM; | |
4738 | goto out_put_task; | |
4739 | } | |
4740 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4741 | retval = -ENOMEM; | |
4742 | goto out_free_cpus_allowed; | |
4743 | } | |
1da177e4 | 4744 | retval = -EPERM; |
c69e8d9c | 4745 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4746 | goto out_unlock; |
4747 | ||
e7834f8f DQ |
4748 | retval = security_task_setscheduler(p, 0, NULL); |
4749 | if (retval) | |
4750 | goto out_unlock; | |
4751 | ||
5a16f3d3 RR |
4752 | cpuset_cpus_allowed(p, cpus_allowed); |
4753 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4754 | again: |
5a16f3d3 | 4755 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4756 | |
8707d8b8 | 4757 | if (!retval) { |
5a16f3d3 RR |
4758 | cpuset_cpus_allowed(p, cpus_allowed); |
4759 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4760 | /* |
4761 | * We must have raced with a concurrent cpuset | |
4762 | * update. Just reset the cpus_allowed to the | |
4763 | * cpuset's cpus_allowed | |
4764 | */ | |
5a16f3d3 | 4765 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4766 | goto again; |
4767 | } | |
4768 | } | |
1da177e4 | 4769 | out_unlock: |
5a16f3d3 RR |
4770 | free_cpumask_var(new_mask); |
4771 | out_free_cpus_allowed: | |
4772 | free_cpumask_var(cpus_allowed); | |
4773 | out_put_task: | |
1da177e4 | 4774 | put_task_struct(p); |
95402b38 | 4775 | put_online_cpus(); |
1da177e4 LT |
4776 | return retval; |
4777 | } | |
4778 | ||
4779 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4780 | struct cpumask *new_mask) |
1da177e4 | 4781 | { |
96f874e2 RR |
4782 | if (len < cpumask_size()) |
4783 | cpumask_clear(new_mask); | |
4784 | else if (len > cpumask_size()) | |
4785 | len = cpumask_size(); | |
4786 | ||
1da177e4 LT |
4787 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4788 | } | |
4789 | ||
4790 | /** | |
4791 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4792 | * @pid: pid of the process | |
4793 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4794 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4795 | */ | |
5add95d4 HC |
4796 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4797 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4798 | { |
5a16f3d3 | 4799 | cpumask_var_t new_mask; |
1da177e4 LT |
4800 | int retval; |
4801 | ||
5a16f3d3 RR |
4802 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4803 | return -ENOMEM; | |
1da177e4 | 4804 | |
5a16f3d3 RR |
4805 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4806 | if (retval == 0) | |
4807 | retval = sched_setaffinity(pid, new_mask); | |
4808 | free_cpumask_var(new_mask); | |
4809 | return retval; | |
1da177e4 LT |
4810 | } |
4811 | ||
96f874e2 | 4812 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4813 | { |
36c8b586 | 4814 | struct task_struct *p; |
31605683 TG |
4815 | unsigned long flags; |
4816 | struct rq *rq; | |
1da177e4 | 4817 | int retval; |
1da177e4 | 4818 | |
95402b38 | 4819 | get_online_cpus(); |
23f5d142 | 4820 | rcu_read_lock(); |
1da177e4 LT |
4821 | |
4822 | retval = -ESRCH; | |
4823 | p = find_process_by_pid(pid); | |
4824 | if (!p) | |
4825 | goto out_unlock; | |
4826 | ||
e7834f8f DQ |
4827 | retval = security_task_getscheduler(p); |
4828 | if (retval) | |
4829 | goto out_unlock; | |
4830 | ||
31605683 | 4831 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4832 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4833 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4834 | |
4835 | out_unlock: | |
23f5d142 | 4836 | rcu_read_unlock(); |
95402b38 | 4837 | put_online_cpus(); |
1da177e4 | 4838 | |
9531b62f | 4839 | return retval; |
1da177e4 LT |
4840 | } |
4841 | ||
4842 | /** | |
4843 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4844 | * @pid: pid of the process | |
4845 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4846 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4847 | */ | |
5add95d4 HC |
4848 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4849 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4850 | { |
4851 | int ret; | |
f17c8607 | 4852 | cpumask_var_t mask; |
1da177e4 | 4853 | |
f17c8607 | 4854 | if (len < cpumask_size()) |
1da177e4 LT |
4855 | return -EINVAL; |
4856 | ||
f17c8607 RR |
4857 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4858 | return -ENOMEM; | |
1da177e4 | 4859 | |
f17c8607 RR |
4860 | ret = sched_getaffinity(pid, mask); |
4861 | if (ret == 0) { | |
4862 | if (copy_to_user(user_mask_ptr, mask, cpumask_size())) | |
4863 | ret = -EFAULT; | |
4864 | else | |
4865 | ret = cpumask_size(); | |
4866 | } | |
4867 | free_cpumask_var(mask); | |
1da177e4 | 4868 | |
f17c8607 | 4869 | return ret; |
1da177e4 LT |
4870 | } |
4871 | ||
4872 | /** | |
4873 | * sys_sched_yield - yield the current processor to other threads. | |
4874 | * | |
dd41f596 IM |
4875 | * This function yields the current CPU to other tasks. If there are no |
4876 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4877 | */ |
5add95d4 | 4878 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4879 | { |
70b97a7f | 4880 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4881 | |
2d72376b | 4882 | schedstat_inc(rq, yld_count); |
4530d7ab | 4883 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4884 | |
4885 | /* | |
4886 | * Since we are going to call schedule() anyway, there's | |
4887 | * no need to preempt or enable interrupts: | |
4888 | */ | |
4889 | __release(rq->lock); | |
8a25d5de | 4890 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4891 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4892 | preempt_enable_no_resched(); |
4893 | ||
4894 | schedule(); | |
4895 | ||
4896 | return 0; | |
4897 | } | |
4898 | ||
d86ee480 PZ |
4899 | static inline int should_resched(void) |
4900 | { | |
4901 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4902 | } | |
4903 | ||
e7b38404 | 4904 | static void __cond_resched(void) |
1da177e4 | 4905 | { |
e7aaaa69 FW |
4906 | add_preempt_count(PREEMPT_ACTIVE); |
4907 | schedule(); | |
4908 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
4909 | } |
4910 | ||
02b67cc3 | 4911 | int __sched _cond_resched(void) |
1da177e4 | 4912 | { |
d86ee480 | 4913 | if (should_resched()) { |
1da177e4 LT |
4914 | __cond_resched(); |
4915 | return 1; | |
4916 | } | |
4917 | return 0; | |
4918 | } | |
02b67cc3 | 4919 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4920 | |
4921 | /* | |
613afbf8 | 4922 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4923 | * call schedule, and on return reacquire the lock. |
4924 | * | |
41a2d6cf | 4925 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4926 | * operations here to prevent schedule() from being called twice (once via |
4927 | * spin_unlock(), once by hand). | |
4928 | */ | |
613afbf8 | 4929 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4930 | { |
d86ee480 | 4931 | int resched = should_resched(); |
6df3cecb JK |
4932 | int ret = 0; |
4933 | ||
f607c668 PZ |
4934 | lockdep_assert_held(lock); |
4935 | ||
95c354fe | 4936 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4937 | spin_unlock(lock); |
d86ee480 | 4938 | if (resched) |
95c354fe NP |
4939 | __cond_resched(); |
4940 | else | |
4941 | cpu_relax(); | |
6df3cecb | 4942 | ret = 1; |
1da177e4 | 4943 | spin_lock(lock); |
1da177e4 | 4944 | } |
6df3cecb | 4945 | return ret; |
1da177e4 | 4946 | } |
613afbf8 | 4947 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4948 | |
613afbf8 | 4949 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4950 | { |
4951 | BUG_ON(!in_softirq()); | |
4952 | ||
d86ee480 | 4953 | if (should_resched()) { |
98d82567 | 4954 | local_bh_enable(); |
1da177e4 LT |
4955 | __cond_resched(); |
4956 | local_bh_disable(); | |
4957 | return 1; | |
4958 | } | |
4959 | return 0; | |
4960 | } | |
613afbf8 | 4961 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4962 | |
1da177e4 LT |
4963 | /** |
4964 | * yield - yield the current processor to other threads. | |
4965 | * | |
72fd4a35 | 4966 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4967 | * thread runnable and calls sys_sched_yield(). |
4968 | */ | |
4969 | void __sched yield(void) | |
4970 | { | |
4971 | set_current_state(TASK_RUNNING); | |
4972 | sys_sched_yield(); | |
4973 | } | |
1da177e4 LT |
4974 | EXPORT_SYMBOL(yield); |
4975 | ||
4976 | /* | |
41a2d6cf | 4977 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4978 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4979 | */ |
4980 | void __sched io_schedule(void) | |
4981 | { | |
54d35f29 | 4982 | struct rq *rq = raw_rq(); |
1da177e4 | 4983 | |
0ff92245 | 4984 | delayacct_blkio_start(); |
1da177e4 | 4985 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4986 | current->in_iowait = 1; |
1da177e4 | 4987 | schedule(); |
8f0dfc34 | 4988 | current->in_iowait = 0; |
1da177e4 | 4989 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4990 | delayacct_blkio_end(); |
1da177e4 | 4991 | } |
1da177e4 LT |
4992 | EXPORT_SYMBOL(io_schedule); |
4993 | ||
4994 | long __sched io_schedule_timeout(long timeout) | |
4995 | { | |
54d35f29 | 4996 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4997 | long ret; |
4998 | ||
0ff92245 | 4999 | delayacct_blkio_start(); |
1da177e4 | 5000 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5001 | current->in_iowait = 1; |
1da177e4 | 5002 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5003 | current->in_iowait = 0; |
1da177e4 | 5004 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5005 | delayacct_blkio_end(); |
1da177e4 LT |
5006 | return ret; |
5007 | } | |
5008 | ||
5009 | /** | |
5010 | * sys_sched_get_priority_max - return maximum RT priority. | |
5011 | * @policy: scheduling class. | |
5012 | * | |
5013 | * this syscall returns the maximum rt_priority that can be used | |
5014 | * by a given scheduling class. | |
5015 | */ | |
5add95d4 | 5016 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5017 | { |
5018 | int ret = -EINVAL; | |
5019 | ||
5020 | switch (policy) { | |
5021 | case SCHED_FIFO: | |
5022 | case SCHED_RR: | |
5023 | ret = MAX_USER_RT_PRIO-1; | |
5024 | break; | |
5025 | case SCHED_NORMAL: | |
b0a9499c | 5026 | case SCHED_BATCH: |
dd41f596 | 5027 | case SCHED_IDLE: |
1da177e4 LT |
5028 | ret = 0; |
5029 | break; | |
5030 | } | |
5031 | return ret; | |
5032 | } | |
5033 | ||
5034 | /** | |
5035 | * sys_sched_get_priority_min - return minimum RT priority. | |
5036 | * @policy: scheduling class. | |
5037 | * | |
5038 | * this syscall returns the minimum rt_priority that can be used | |
5039 | * by a given scheduling class. | |
5040 | */ | |
5add95d4 | 5041 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5042 | { |
5043 | int ret = -EINVAL; | |
5044 | ||
5045 | switch (policy) { | |
5046 | case SCHED_FIFO: | |
5047 | case SCHED_RR: | |
5048 | ret = 1; | |
5049 | break; | |
5050 | case SCHED_NORMAL: | |
b0a9499c | 5051 | case SCHED_BATCH: |
dd41f596 | 5052 | case SCHED_IDLE: |
1da177e4 LT |
5053 | ret = 0; |
5054 | } | |
5055 | return ret; | |
5056 | } | |
5057 | ||
5058 | /** | |
5059 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5060 | * @pid: pid of the process. | |
5061 | * @interval: userspace pointer to the timeslice value. | |
5062 | * | |
5063 | * this syscall writes the default timeslice value of a given process | |
5064 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5065 | */ | |
17da2bd9 | 5066 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5067 | struct timespec __user *, interval) |
1da177e4 | 5068 | { |
36c8b586 | 5069 | struct task_struct *p; |
a4ec24b4 | 5070 | unsigned int time_slice; |
dba091b9 TG |
5071 | unsigned long flags; |
5072 | struct rq *rq; | |
3a5c359a | 5073 | int retval; |
1da177e4 | 5074 | struct timespec t; |
1da177e4 LT |
5075 | |
5076 | if (pid < 0) | |
3a5c359a | 5077 | return -EINVAL; |
1da177e4 LT |
5078 | |
5079 | retval = -ESRCH; | |
1a551ae7 | 5080 | rcu_read_lock(); |
1da177e4 LT |
5081 | p = find_process_by_pid(pid); |
5082 | if (!p) | |
5083 | goto out_unlock; | |
5084 | ||
5085 | retval = security_task_getscheduler(p); | |
5086 | if (retval) | |
5087 | goto out_unlock; | |
5088 | ||
dba091b9 TG |
5089 | rq = task_rq_lock(p, &flags); |
5090 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5091 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5092 | |
1a551ae7 | 5093 | rcu_read_unlock(); |
a4ec24b4 | 5094 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5095 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5096 | return retval; |
3a5c359a | 5097 | |
1da177e4 | 5098 | out_unlock: |
1a551ae7 | 5099 | rcu_read_unlock(); |
1da177e4 LT |
5100 | return retval; |
5101 | } | |
5102 | ||
7c731e0a | 5103 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5104 | |
82a1fcb9 | 5105 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5106 | { |
1da177e4 | 5107 | unsigned long free = 0; |
36c8b586 | 5108 | unsigned state; |
1da177e4 | 5109 | |
1da177e4 | 5110 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5111 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5112 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5113 | #if BITS_PER_LONG == 32 |
1da177e4 | 5114 | if (state == TASK_RUNNING) |
3df0fc5b | 5115 | printk(KERN_CONT " running "); |
1da177e4 | 5116 | else |
3df0fc5b | 5117 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5118 | #else |
5119 | if (state == TASK_RUNNING) | |
3df0fc5b | 5120 | printk(KERN_CONT " running task "); |
1da177e4 | 5121 | else |
3df0fc5b | 5122 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5123 | #endif |
5124 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5125 | free = stack_not_used(p); |
1da177e4 | 5126 | #endif |
3df0fc5b | 5127 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5128 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5129 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5130 | |
5fb5e6de | 5131 | show_stack(p, NULL); |
1da177e4 LT |
5132 | } |
5133 | ||
e59e2ae2 | 5134 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5135 | { |
36c8b586 | 5136 | struct task_struct *g, *p; |
1da177e4 | 5137 | |
4bd77321 | 5138 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5139 | printk(KERN_INFO |
5140 | " task PC stack pid father\n"); | |
1da177e4 | 5141 | #else |
3df0fc5b PZ |
5142 | printk(KERN_INFO |
5143 | " task PC stack pid father\n"); | |
1da177e4 LT |
5144 | #endif |
5145 | read_lock(&tasklist_lock); | |
5146 | do_each_thread(g, p) { | |
5147 | /* | |
5148 | * reset the NMI-timeout, listing all files on a slow | |
5149 | * console might take alot of time: | |
5150 | */ | |
5151 | touch_nmi_watchdog(); | |
39bc89fd | 5152 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5153 | sched_show_task(p); |
1da177e4 LT |
5154 | } while_each_thread(g, p); |
5155 | ||
04c9167f JF |
5156 | touch_all_softlockup_watchdogs(); |
5157 | ||
dd41f596 IM |
5158 | #ifdef CONFIG_SCHED_DEBUG |
5159 | sysrq_sched_debug_show(); | |
5160 | #endif | |
1da177e4 | 5161 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5162 | /* |
5163 | * Only show locks if all tasks are dumped: | |
5164 | */ | |
93335a21 | 5165 | if (!state_filter) |
e59e2ae2 | 5166 | debug_show_all_locks(); |
1da177e4 LT |
5167 | } |
5168 | ||
1df21055 IM |
5169 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5170 | { | |
dd41f596 | 5171 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5172 | } |
5173 | ||
f340c0d1 IM |
5174 | /** |
5175 | * init_idle - set up an idle thread for a given CPU | |
5176 | * @idle: task in question | |
5177 | * @cpu: cpu the idle task belongs to | |
5178 | * | |
5179 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5180 | * flag, to make booting more robust. | |
5181 | */ | |
5c1e1767 | 5182 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5183 | { |
70b97a7f | 5184 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5185 | unsigned long flags; |
5186 | ||
05fa785c | 5187 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5188 | |
dd41f596 | 5189 | __sched_fork(idle); |
06b83b5f | 5190 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5191 | idle->se.exec_start = sched_clock(); |
5192 | ||
96f874e2 | 5193 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5194 | __set_task_cpu(idle, cpu); |
1da177e4 | 5195 | |
1da177e4 | 5196 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5197 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5198 | idle->oncpu = 1; | |
5199 | #endif | |
05fa785c | 5200 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5201 | |
5202 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5203 | #if defined(CONFIG_PREEMPT) |
5204 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5205 | #else | |
a1261f54 | 5206 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5207 | #endif |
dd41f596 IM |
5208 | /* |
5209 | * The idle tasks have their own, simple scheduling class: | |
5210 | */ | |
5211 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5212 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5213 | } |
5214 | ||
5215 | /* | |
5216 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5217 | * indicates which cpus entered this state. This is used | |
5218 | * in the rcu update to wait only for active cpus. For system | |
5219 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5220 | * always be CPU_BITS_NONE. |
1da177e4 | 5221 | */ |
6a7b3dc3 | 5222 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5223 | |
19978ca6 IM |
5224 | /* |
5225 | * Increase the granularity value when there are more CPUs, | |
5226 | * because with more CPUs the 'effective latency' as visible | |
5227 | * to users decreases. But the relationship is not linear, | |
5228 | * so pick a second-best guess by going with the log2 of the | |
5229 | * number of CPUs. | |
5230 | * | |
5231 | * This idea comes from the SD scheduler of Con Kolivas: | |
5232 | */ | |
acb4a848 | 5233 | static int get_update_sysctl_factor(void) |
19978ca6 | 5234 | { |
4ca3ef71 | 5235 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5236 | unsigned int factor; |
5237 | ||
5238 | switch (sysctl_sched_tunable_scaling) { | |
5239 | case SCHED_TUNABLESCALING_NONE: | |
5240 | factor = 1; | |
5241 | break; | |
5242 | case SCHED_TUNABLESCALING_LINEAR: | |
5243 | factor = cpus; | |
5244 | break; | |
5245 | case SCHED_TUNABLESCALING_LOG: | |
5246 | default: | |
5247 | factor = 1 + ilog2(cpus); | |
5248 | break; | |
5249 | } | |
19978ca6 | 5250 | |
acb4a848 CE |
5251 | return factor; |
5252 | } | |
19978ca6 | 5253 | |
acb4a848 CE |
5254 | static void update_sysctl(void) |
5255 | { | |
5256 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5257 | |
0bcdcf28 CE |
5258 | #define SET_SYSCTL(name) \ |
5259 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5260 | SET_SYSCTL(sched_min_granularity); | |
5261 | SET_SYSCTL(sched_latency); | |
5262 | SET_SYSCTL(sched_wakeup_granularity); | |
5263 | SET_SYSCTL(sched_shares_ratelimit); | |
5264 | #undef SET_SYSCTL | |
5265 | } | |
55cd5340 | 5266 | |
0bcdcf28 CE |
5267 | static inline void sched_init_granularity(void) |
5268 | { | |
5269 | update_sysctl(); | |
19978ca6 IM |
5270 | } |
5271 | ||
1da177e4 LT |
5272 | #ifdef CONFIG_SMP |
5273 | /* | |
5274 | * This is how migration works: | |
5275 | * | |
70b97a7f | 5276 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5277 | * runqueue and wake up that CPU's migration thread. |
5278 | * 2) we down() the locked semaphore => thread blocks. | |
5279 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5280 | * thread off the CPU) | |
5281 | * 4) it gets the migration request and checks whether the migrated | |
5282 | * task is still in the wrong runqueue. | |
5283 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5284 | * it and puts it into the right queue. | |
5285 | * 6) migration thread up()s the semaphore. | |
5286 | * 7) we wake up and the migration is done. | |
5287 | */ | |
5288 | ||
5289 | /* | |
5290 | * Change a given task's CPU affinity. Migrate the thread to a | |
5291 | * proper CPU and schedule it away if the CPU it's executing on | |
5292 | * is removed from the allowed bitmask. | |
5293 | * | |
5294 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5295 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5296 | * call is not atomic; no spinlocks may be held. |
5297 | */ | |
96f874e2 | 5298 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 5299 | { |
70b97a7f | 5300 | struct migration_req req; |
1da177e4 | 5301 | unsigned long flags; |
70b97a7f | 5302 | struct rq *rq; |
48f24c4d | 5303 | int ret = 0; |
1da177e4 | 5304 | |
e2912009 PZ |
5305 | /* |
5306 | * Since we rely on wake-ups to migrate sleeping tasks, don't change | |
5307 | * the ->cpus_allowed mask from under waking tasks, which would be | |
5308 | * possible when we change rq->lock in ttwu(), so synchronize against | |
5309 | * TASK_WAKING to avoid that. | |
5310 | */ | |
5311 | again: | |
5312 | while (p->state == TASK_WAKING) | |
5313 | cpu_relax(); | |
5314 | ||
1da177e4 | 5315 | rq = task_rq_lock(p, &flags); |
e2912009 PZ |
5316 | |
5317 | if (p->state == TASK_WAKING) { | |
5318 | task_rq_unlock(rq, &flags); | |
5319 | goto again; | |
5320 | } | |
5321 | ||
6ad4c188 | 5322 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5323 | ret = -EINVAL; |
5324 | goto out; | |
5325 | } | |
5326 | ||
9985b0ba | 5327 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5328 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5329 | ret = -EINVAL; |
5330 | goto out; | |
5331 | } | |
5332 | ||
73fe6aae | 5333 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5334 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5335 | else { |
96f874e2 RR |
5336 | cpumask_copy(&p->cpus_allowed, new_mask); |
5337 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5338 | } |
5339 | ||
1da177e4 | 5340 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5341 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5342 | goto out; |
5343 | ||
6ad4c188 | 5344 | if (migrate_task(p, cpumask_any_and(cpu_active_mask, new_mask), &req)) { |
1da177e4 | 5345 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
5346 | struct task_struct *mt = rq->migration_thread; |
5347 | ||
5348 | get_task_struct(mt); | |
1da177e4 LT |
5349 | task_rq_unlock(rq, &flags); |
5350 | wake_up_process(rq->migration_thread); | |
693525e3 | 5351 | put_task_struct(mt); |
1da177e4 LT |
5352 | wait_for_completion(&req.done); |
5353 | tlb_migrate_finish(p->mm); | |
5354 | return 0; | |
5355 | } | |
5356 | out: | |
5357 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5358 | |
1da177e4 LT |
5359 | return ret; |
5360 | } | |
cd8ba7cd | 5361 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5362 | |
5363 | /* | |
41a2d6cf | 5364 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5365 | * this because either it can't run here any more (set_cpus_allowed() |
5366 | * away from this CPU, or CPU going down), or because we're | |
5367 | * attempting to rebalance this task on exec (sched_exec). | |
5368 | * | |
5369 | * So we race with normal scheduler movements, but that's OK, as long | |
5370 | * as the task is no longer on this CPU. | |
efc30814 KK |
5371 | * |
5372 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5373 | */ |
efc30814 | 5374 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5375 | { |
70b97a7f | 5376 | struct rq *rq_dest, *rq_src; |
e2912009 | 5377 | int ret = 0; |
1da177e4 | 5378 | |
e761b772 | 5379 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5380 | return ret; |
1da177e4 LT |
5381 | |
5382 | rq_src = cpu_rq(src_cpu); | |
5383 | rq_dest = cpu_rq(dest_cpu); | |
5384 | ||
5385 | double_rq_lock(rq_src, rq_dest); | |
5386 | /* Already moved. */ | |
5387 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5388 | goto done; |
1da177e4 | 5389 | /* Affinity changed (again). */ |
96f874e2 | 5390 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5391 | goto fail; |
1da177e4 | 5392 | |
e2912009 PZ |
5393 | /* |
5394 | * If we're not on a rq, the next wake-up will ensure we're | |
5395 | * placed properly. | |
5396 | */ | |
5397 | if (p->se.on_rq) { | |
2e1cb74a | 5398 | deactivate_task(rq_src, p, 0); |
e2912009 | 5399 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5400 | activate_task(rq_dest, p, 0); |
15afe09b | 5401 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5402 | } |
b1e38734 | 5403 | done: |
efc30814 | 5404 | ret = 1; |
b1e38734 | 5405 | fail: |
1da177e4 | 5406 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5407 | return ret; |
1da177e4 LT |
5408 | } |
5409 | ||
03b042bf PM |
5410 | #define RCU_MIGRATION_IDLE 0 |
5411 | #define RCU_MIGRATION_NEED_QS 1 | |
5412 | #define RCU_MIGRATION_GOT_QS 2 | |
5413 | #define RCU_MIGRATION_MUST_SYNC 3 | |
5414 | ||
1da177e4 LT |
5415 | /* |
5416 | * migration_thread - this is a highprio system thread that performs | |
5417 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5418 | * another runqueue. | |
5419 | */ | |
95cdf3b7 | 5420 | static int migration_thread(void *data) |
1da177e4 | 5421 | { |
03b042bf | 5422 | int badcpu; |
1da177e4 | 5423 | int cpu = (long)data; |
70b97a7f | 5424 | struct rq *rq; |
1da177e4 LT |
5425 | |
5426 | rq = cpu_rq(cpu); | |
5427 | BUG_ON(rq->migration_thread != current); | |
5428 | ||
5429 | set_current_state(TASK_INTERRUPTIBLE); | |
5430 | while (!kthread_should_stop()) { | |
70b97a7f | 5431 | struct migration_req *req; |
1da177e4 | 5432 | struct list_head *head; |
1da177e4 | 5433 | |
05fa785c | 5434 | raw_spin_lock_irq(&rq->lock); |
1da177e4 LT |
5435 | |
5436 | if (cpu_is_offline(cpu)) { | |
05fa785c | 5437 | raw_spin_unlock_irq(&rq->lock); |
371cbb38 | 5438 | break; |
1da177e4 LT |
5439 | } |
5440 | ||
5441 | if (rq->active_balance) { | |
5442 | active_load_balance(rq, cpu); | |
5443 | rq->active_balance = 0; | |
5444 | } | |
5445 | ||
5446 | head = &rq->migration_queue; | |
5447 | ||
5448 | if (list_empty(head)) { | |
05fa785c | 5449 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5450 | schedule(); |
5451 | set_current_state(TASK_INTERRUPTIBLE); | |
5452 | continue; | |
5453 | } | |
70b97a7f | 5454 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5455 | list_del_init(head->next); |
5456 | ||
03b042bf | 5457 | if (req->task != NULL) { |
05fa785c | 5458 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5459 | __migrate_task(req->task, cpu, req->dest_cpu); |
5460 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
5461 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
05fa785c | 5462 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5463 | } else { |
5464 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
05fa785c | 5465 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5466 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); |
5467 | } | |
674311d5 | 5468 | local_irq_enable(); |
1da177e4 LT |
5469 | |
5470 | complete(&req->done); | |
5471 | } | |
5472 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 5473 | |
1da177e4 LT |
5474 | return 0; |
5475 | } | |
5476 | ||
5477 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5478 | |
5479 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5480 | { | |
5481 | int ret; | |
5482 | ||
5483 | local_irq_disable(); | |
5484 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5485 | local_irq_enable(); | |
5486 | return ret; | |
5487 | } | |
5488 | ||
054b9108 | 5489 | /* |
3a4fa0a2 | 5490 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5491 | */ |
48f24c4d | 5492 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5493 | { |
70b97a7f | 5494 | int dest_cpu; |
e76bd8d9 RR |
5495 | |
5496 | again: | |
5da9a0fb | 5497 | dest_cpu = select_fallback_rq(dead_cpu, p); |
e76bd8d9 | 5498 | |
e76bd8d9 RR |
5499 | /* It can have affinity changed while we were choosing. */ |
5500 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
5501 | goto again; | |
1da177e4 LT |
5502 | } |
5503 | ||
5504 | /* | |
5505 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5506 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5507 | * for performance reasons the counter is not stricly tracking tasks to | |
5508 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5509 | * to keep the global sum constant after CPU-down: | |
5510 | */ | |
70b97a7f | 5511 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5512 | { |
6ad4c188 | 5513 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5514 | unsigned long flags; |
5515 | ||
5516 | local_irq_save(flags); | |
5517 | double_rq_lock(rq_src, rq_dest); | |
5518 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5519 | rq_src->nr_uninterruptible = 0; | |
5520 | double_rq_unlock(rq_src, rq_dest); | |
5521 | local_irq_restore(flags); | |
5522 | } | |
5523 | ||
5524 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5525 | static void migrate_live_tasks(int src_cpu) | |
5526 | { | |
48f24c4d | 5527 | struct task_struct *p, *t; |
1da177e4 | 5528 | |
f7b4cddc | 5529 | read_lock(&tasklist_lock); |
1da177e4 | 5530 | |
48f24c4d IM |
5531 | do_each_thread(t, p) { |
5532 | if (p == current) | |
1da177e4 LT |
5533 | continue; |
5534 | ||
48f24c4d IM |
5535 | if (task_cpu(p) == src_cpu) |
5536 | move_task_off_dead_cpu(src_cpu, p); | |
5537 | } while_each_thread(t, p); | |
1da177e4 | 5538 | |
f7b4cddc | 5539 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5540 | } |
5541 | ||
dd41f596 IM |
5542 | /* |
5543 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5544 | * It does so by boosting its priority to highest possible. |
5545 | * Used by CPU offline code. | |
1da177e4 LT |
5546 | */ |
5547 | void sched_idle_next(void) | |
5548 | { | |
48f24c4d | 5549 | int this_cpu = smp_processor_id(); |
70b97a7f | 5550 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5551 | struct task_struct *p = rq->idle; |
5552 | unsigned long flags; | |
5553 | ||
5554 | /* cpu has to be offline */ | |
48f24c4d | 5555 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5556 | |
48f24c4d IM |
5557 | /* |
5558 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5559 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5560 | */ |
05fa785c | 5561 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5562 | |
dd41f596 | 5563 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5564 | |
94bc9a7b DA |
5565 | update_rq_clock(rq); |
5566 | activate_task(rq, p, 0); | |
1da177e4 | 5567 | |
05fa785c | 5568 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5569 | } |
5570 | ||
48f24c4d IM |
5571 | /* |
5572 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5573 | * offline. |
5574 | */ | |
5575 | void idle_task_exit(void) | |
5576 | { | |
5577 | struct mm_struct *mm = current->active_mm; | |
5578 | ||
5579 | BUG_ON(cpu_online(smp_processor_id())); | |
5580 | ||
5581 | if (mm != &init_mm) | |
5582 | switch_mm(mm, &init_mm, current); | |
5583 | mmdrop(mm); | |
5584 | } | |
5585 | ||
054b9108 | 5586 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5587 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5588 | { |
70b97a7f | 5589 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5590 | |
5591 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5592 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5593 | |
5594 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5595 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5596 | |
48f24c4d | 5597 | get_task_struct(p); |
1da177e4 LT |
5598 | |
5599 | /* | |
5600 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5601 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5602 | * fine. |
5603 | */ | |
05fa785c | 5604 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5605 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5606 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5607 | |
48f24c4d | 5608 | put_task_struct(p); |
1da177e4 LT |
5609 | } |
5610 | ||
5611 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5612 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5613 | { | |
70b97a7f | 5614 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5615 | struct task_struct *next; |
48f24c4d | 5616 | |
dd41f596 IM |
5617 | for ( ; ; ) { |
5618 | if (!rq->nr_running) | |
5619 | break; | |
a8e504d2 | 5620 | update_rq_clock(rq); |
b67802ea | 5621 | next = pick_next_task(rq); |
dd41f596 IM |
5622 | if (!next) |
5623 | break; | |
79c53799 | 5624 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5625 | migrate_dead(dead_cpu, next); |
e692ab53 | 5626 | |
1da177e4 LT |
5627 | } |
5628 | } | |
dce48a84 TG |
5629 | |
5630 | /* | |
5631 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5632 | */ | |
5633 | static void calc_global_load_remove(struct rq *rq) | |
5634 | { | |
5635 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5636 | rq->calc_load_active = 0; |
dce48a84 | 5637 | } |
1da177e4 LT |
5638 | #endif /* CONFIG_HOTPLUG_CPU */ |
5639 | ||
e692ab53 NP |
5640 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5641 | ||
5642 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5643 | { |
5644 | .procname = "sched_domain", | |
c57baf1e | 5645 | .mode = 0555, |
e0361851 | 5646 | }, |
56992309 | 5647 | {} |
e692ab53 NP |
5648 | }; |
5649 | ||
5650 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5651 | { |
5652 | .procname = "kernel", | |
c57baf1e | 5653 | .mode = 0555, |
e0361851 AD |
5654 | .child = sd_ctl_dir, |
5655 | }, | |
56992309 | 5656 | {} |
e692ab53 NP |
5657 | }; |
5658 | ||
5659 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5660 | { | |
5661 | struct ctl_table *entry = | |
5cf9f062 | 5662 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5663 | |
e692ab53 NP |
5664 | return entry; |
5665 | } | |
5666 | ||
6382bc90 MM |
5667 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5668 | { | |
cd790076 | 5669 | struct ctl_table *entry; |
6382bc90 | 5670 | |
cd790076 MM |
5671 | /* |
5672 | * In the intermediate directories, both the child directory and | |
5673 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5674 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5675 | * static strings and all have proc handlers. |
5676 | */ | |
5677 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5678 | if (entry->child) |
5679 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5680 | if (entry->proc_handler == NULL) |
5681 | kfree(entry->procname); | |
5682 | } | |
6382bc90 MM |
5683 | |
5684 | kfree(*tablep); | |
5685 | *tablep = NULL; | |
5686 | } | |
5687 | ||
e692ab53 | 5688 | static void |
e0361851 | 5689 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5690 | const char *procname, void *data, int maxlen, |
5691 | mode_t mode, proc_handler *proc_handler) | |
5692 | { | |
e692ab53 NP |
5693 | entry->procname = procname; |
5694 | entry->data = data; | |
5695 | entry->maxlen = maxlen; | |
5696 | entry->mode = mode; | |
5697 | entry->proc_handler = proc_handler; | |
5698 | } | |
5699 | ||
5700 | static struct ctl_table * | |
5701 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5702 | { | |
a5d8c348 | 5703 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5704 | |
ad1cdc1d MM |
5705 | if (table == NULL) |
5706 | return NULL; | |
5707 | ||
e0361851 | 5708 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5709 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5710 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5711 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5712 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5713 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5714 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5715 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5716 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5717 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5718 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5719 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5720 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5721 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5722 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5723 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5724 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5725 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5726 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5727 | &sd->cache_nice_tries, |
5728 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5729 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5730 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5731 | set_table_entry(&table[11], "name", sd->name, |
5732 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5733 | /* &table[12] is terminator */ | |
e692ab53 NP |
5734 | |
5735 | return table; | |
5736 | } | |
5737 | ||
9a4e7159 | 5738 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5739 | { |
5740 | struct ctl_table *entry, *table; | |
5741 | struct sched_domain *sd; | |
5742 | int domain_num = 0, i; | |
5743 | char buf[32]; | |
5744 | ||
5745 | for_each_domain(cpu, sd) | |
5746 | domain_num++; | |
5747 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5748 | if (table == NULL) |
5749 | return NULL; | |
e692ab53 NP |
5750 | |
5751 | i = 0; | |
5752 | for_each_domain(cpu, sd) { | |
5753 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5754 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5755 | entry->mode = 0555; |
e692ab53 NP |
5756 | entry->child = sd_alloc_ctl_domain_table(sd); |
5757 | entry++; | |
5758 | i++; | |
5759 | } | |
5760 | return table; | |
5761 | } | |
5762 | ||
5763 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5764 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5765 | { |
6ad4c188 | 5766 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5767 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5768 | char buf[32]; | |
5769 | ||
7378547f MM |
5770 | WARN_ON(sd_ctl_dir[0].child); |
5771 | sd_ctl_dir[0].child = entry; | |
5772 | ||
ad1cdc1d MM |
5773 | if (entry == NULL) |
5774 | return; | |
5775 | ||
6ad4c188 | 5776 | for_each_possible_cpu(i) { |
e692ab53 | 5777 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5778 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5779 | entry->mode = 0555; |
e692ab53 | 5780 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5781 | entry++; |
e692ab53 | 5782 | } |
7378547f MM |
5783 | |
5784 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5785 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5786 | } | |
6382bc90 | 5787 | |
7378547f | 5788 | /* may be called multiple times per register */ |
6382bc90 MM |
5789 | static void unregister_sched_domain_sysctl(void) |
5790 | { | |
7378547f MM |
5791 | if (sd_sysctl_header) |
5792 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5793 | sd_sysctl_header = NULL; |
7378547f MM |
5794 | if (sd_ctl_dir[0].child) |
5795 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5796 | } |
e692ab53 | 5797 | #else |
6382bc90 MM |
5798 | static void register_sched_domain_sysctl(void) |
5799 | { | |
5800 | } | |
5801 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5802 | { |
5803 | } | |
5804 | #endif | |
5805 | ||
1f11eb6a GH |
5806 | static void set_rq_online(struct rq *rq) |
5807 | { | |
5808 | if (!rq->online) { | |
5809 | const struct sched_class *class; | |
5810 | ||
c6c4927b | 5811 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5812 | rq->online = 1; |
5813 | ||
5814 | for_each_class(class) { | |
5815 | if (class->rq_online) | |
5816 | class->rq_online(rq); | |
5817 | } | |
5818 | } | |
5819 | } | |
5820 | ||
5821 | static void set_rq_offline(struct rq *rq) | |
5822 | { | |
5823 | if (rq->online) { | |
5824 | const struct sched_class *class; | |
5825 | ||
5826 | for_each_class(class) { | |
5827 | if (class->rq_offline) | |
5828 | class->rq_offline(rq); | |
5829 | } | |
5830 | ||
c6c4927b | 5831 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5832 | rq->online = 0; |
5833 | } | |
5834 | } | |
5835 | ||
1da177e4 LT |
5836 | /* |
5837 | * migration_call - callback that gets triggered when a CPU is added. | |
5838 | * Here we can start up the necessary migration thread for the new CPU. | |
5839 | */ | |
48f24c4d IM |
5840 | static int __cpuinit |
5841 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5842 | { |
1da177e4 | 5843 | struct task_struct *p; |
48f24c4d | 5844 | int cpu = (long)hcpu; |
1da177e4 | 5845 | unsigned long flags; |
70b97a7f | 5846 | struct rq *rq; |
1da177e4 LT |
5847 | |
5848 | switch (action) { | |
5be9361c | 5849 | |
1da177e4 | 5850 | case CPU_UP_PREPARE: |
8bb78442 | 5851 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5852 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5853 | if (IS_ERR(p)) |
5854 | return NOTIFY_BAD; | |
1da177e4 LT |
5855 | kthread_bind(p, cpu); |
5856 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5857 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5858 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 5859 | task_rq_unlock(rq, &flags); |
371cbb38 | 5860 | get_task_struct(p); |
1da177e4 | 5861 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 5862 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5863 | break; |
48f24c4d | 5864 | |
1da177e4 | 5865 | case CPU_ONLINE: |
8bb78442 | 5866 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5867 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 5868 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
5869 | |
5870 | /* Update our root-domain */ | |
5871 | rq = cpu_rq(cpu); | |
05fa785c | 5872 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5873 | if (rq->rd) { |
c6c4927b | 5874 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5875 | |
5876 | set_rq_online(rq); | |
1f94ef59 | 5877 | } |
05fa785c | 5878 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5879 | break; |
48f24c4d | 5880 | |
1da177e4 LT |
5881 | #ifdef CONFIG_HOTPLUG_CPU |
5882 | case CPU_UP_CANCELED: | |
8bb78442 | 5883 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5884 | if (!cpu_rq(cpu)->migration_thread) |
5885 | break; | |
41a2d6cf | 5886 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 5887 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 5888 | cpumask_any(cpu_online_mask)); |
1da177e4 | 5889 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 5890 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
5891 | cpu_rq(cpu)->migration_thread = NULL; |
5892 | break; | |
48f24c4d | 5893 | |
1da177e4 | 5894 | case CPU_DEAD: |
8bb78442 | 5895 | case CPU_DEAD_FROZEN: |
470fd646 | 5896 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5897 | migrate_live_tasks(cpu); |
5898 | rq = cpu_rq(cpu); | |
5899 | kthread_stop(rq->migration_thread); | |
371cbb38 | 5900 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
5901 | rq->migration_thread = NULL; |
5902 | /* Idle task back to normal (off runqueue, low prio) */ | |
05fa785c | 5903 | raw_spin_lock_irq(&rq->lock); |
a8e504d2 | 5904 | update_rq_clock(rq); |
2e1cb74a | 5905 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5906 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5907 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5908 | migrate_dead_tasks(cpu); |
05fa785c | 5909 | raw_spin_unlock_irq(&rq->lock); |
470fd646 | 5910 | cpuset_unlock(); |
1da177e4 LT |
5911 | migrate_nr_uninterruptible(rq); |
5912 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5913 | calc_global_load_remove(rq); |
41a2d6cf IM |
5914 | /* |
5915 | * No need to migrate the tasks: it was best-effort if | |
5916 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5917 | * the requestors. | |
5918 | */ | |
05fa785c | 5919 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5920 | while (!list_empty(&rq->migration_queue)) { |
70b97a7f IM |
5921 | struct migration_req *req; |
5922 | ||
1da177e4 | 5923 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5924 | struct migration_req, list); |
1da177e4 | 5925 | list_del_init(&req->list); |
05fa785c | 5926 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5927 | complete(&req->done); |
05fa785c | 5928 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5929 | } |
05fa785c | 5930 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5931 | break; |
57d885fe | 5932 | |
08f503b0 GH |
5933 | case CPU_DYING: |
5934 | case CPU_DYING_FROZEN: | |
57d885fe GH |
5935 | /* Update our root-domain */ |
5936 | rq = cpu_rq(cpu); | |
05fa785c | 5937 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5938 | if (rq->rd) { |
c6c4927b | 5939 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5940 | set_rq_offline(rq); |
57d885fe | 5941 | } |
05fa785c | 5942 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 5943 | break; |
1da177e4 LT |
5944 | #endif |
5945 | } | |
5946 | return NOTIFY_OK; | |
5947 | } | |
5948 | ||
f38b0820 PM |
5949 | /* |
5950 | * Register at high priority so that task migration (migrate_all_tasks) | |
5951 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5952 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5953 | */ |
26c2143b | 5954 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5955 | .notifier_call = migration_call, |
5956 | .priority = 10 | |
5957 | }; | |
5958 | ||
7babe8db | 5959 | static int __init migration_init(void) |
1da177e4 LT |
5960 | { |
5961 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5962 | int err; |
48f24c4d IM |
5963 | |
5964 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5965 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5966 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5967 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5968 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5969 | |
a004cd42 | 5970 | return 0; |
1da177e4 | 5971 | } |
7babe8db | 5972 | early_initcall(migration_init); |
1da177e4 LT |
5973 | #endif |
5974 | ||
5975 | #ifdef CONFIG_SMP | |
476f3534 | 5976 | |
3e9830dc | 5977 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5978 | |
f6630114 MT |
5979 | static __read_mostly int sched_domain_debug_enabled; |
5980 | ||
5981 | static int __init sched_domain_debug_setup(char *str) | |
5982 | { | |
5983 | sched_domain_debug_enabled = 1; | |
5984 | ||
5985 | return 0; | |
5986 | } | |
5987 | early_param("sched_debug", sched_domain_debug_setup); | |
5988 | ||
7c16ec58 | 5989 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5990 | struct cpumask *groupmask) |
1da177e4 | 5991 | { |
4dcf6aff | 5992 | struct sched_group *group = sd->groups; |
434d53b0 | 5993 | char str[256]; |
1da177e4 | 5994 | |
968ea6d8 | 5995 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5996 | cpumask_clear(groupmask); |
4dcf6aff IM |
5997 | |
5998 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5999 | ||
6000 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6001 | printk("does not load-balance\n"); |
4dcf6aff | 6002 | if (sd->parent) |
3df0fc5b PZ |
6003 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6004 | " has parent"); | |
4dcf6aff | 6005 | return -1; |
41c7ce9a NP |
6006 | } |
6007 | ||
3df0fc5b | 6008 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6009 | |
758b2cdc | 6010 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6011 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6012 | "CPU%d\n", cpu); | |
4dcf6aff | 6013 | } |
758b2cdc | 6014 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6015 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6016 | " CPU%d\n", cpu); | |
4dcf6aff | 6017 | } |
1da177e4 | 6018 | |
4dcf6aff | 6019 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6020 | do { |
4dcf6aff | 6021 | if (!group) { |
3df0fc5b PZ |
6022 | printk("\n"); |
6023 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6024 | break; |
6025 | } | |
6026 | ||
18a3885f | 6027 | if (!group->cpu_power) { |
3df0fc5b PZ |
6028 | printk(KERN_CONT "\n"); |
6029 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6030 | "set\n"); | |
4dcf6aff IM |
6031 | break; |
6032 | } | |
1da177e4 | 6033 | |
758b2cdc | 6034 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6035 | printk(KERN_CONT "\n"); |
6036 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6037 | break; |
6038 | } | |
1da177e4 | 6039 | |
758b2cdc | 6040 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6041 | printk(KERN_CONT "\n"); |
6042 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6043 | break; |
6044 | } | |
1da177e4 | 6045 | |
758b2cdc | 6046 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6047 | |
968ea6d8 | 6048 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6049 | |
3df0fc5b | 6050 | printk(KERN_CONT " %s", str); |
18a3885f | 6051 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6052 | printk(KERN_CONT " (cpu_power = %d)", |
6053 | group->cpu_power); | |
381512cf | 6054 | } |
1da177e4 | 6055 | |
4dcf6aff IM |
6056 | group = group->next; |
6057 | } while (group != sd->groups); | |
3df0fc5b | 6058 | printk(KERN_CONT "\n"); |
1da177e4 | 6059 | |
758b2cdc | 6060 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6061 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6062 | |
758b2cdc RR |
6063 | if (sd->parent && |
6064 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6065 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6066 | "of domain->span\n"); | |
4dcf6aff IM |
6067 | return 0; |
6068 | } | |
1da177e4 | 6069 | |
4dcf6aff IM |
6070 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6071 | { | |
d5dd3db1 | 6072 | cpumask_var_t groupmask; |
4dcf6aff | 6073 | int level = 0; |
1da177e4 | 6074 | |
f6630114 MT |
6075 | if (!sched_domain_debug_enabled) |
6076 | return; | |
6077 | ||
4dcf6aff IM |
6078 | if (!sd) { |
6079 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6080 | return; | |
6081 | } | |
1da177e4 | 6082 | |
4dcf6aff IM |
6083 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6084 | ||
d5dd3db1 | 6085 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6086 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6087 | return; | |
6088 | } | |
6089 | ||
4dcf6aff | 6090 | for (;;) { |
7c16ec58 | 6091 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6092 | break; |
1da177e4 LT |
6093 | level++; |
6094 | sd = sd->parent; | |
33859f7f | 6095 | if (!sd) |
4dcf6aff IM |
6096 | break; |
6097 | } | |
d5dd3db1 | 6098 | free_cpumask_var(groupmask); |
1da177e4 | 6099 | } |
6d6bc0ad | 6100 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6101 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6102 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6103 | |
1a20ff27 | 6104 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6105 | { |
758b2cdc | 6106 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6107 | return 1; |
6108 | ||
6109 | /* Following flags need at least 2 groups */ | |
6110 | if (sd->flags & (SD_LOAD_BALANCE | | |
6111 | SD_BALANCE_NEWIDLE | | |
6112 | SD_BALANCE_FORK | | |
89c4710e SS |
6113 | SD_BALANCE_EXEC | |
6114 | SD_SHARE_CPUPOWER | | |
6115 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6116 | if (sd->groups != sd->groups->next) |
6117 | return 0; | |
6118 | } | |
6119 | ||
6120 | /* Following flags don't use groups */ | |
c88d5910 | 6121 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6122 | return 0; |
6123 | ||
6124 | return 1; | |
6125 | } | |
6126 | ||
48f24c4d IM |
6127 | static int |
6128 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6129 | { |
6130 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6131 | ||
6132 | if (sd_degenerate(parent)) | |
6133 | return 1; | |
6134 | ||
758b2cdc | 6135 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6136 | return 0; |
6137 | ||
245af2c7 SS |
6138 | /* Flags needing groups don't count if only 1 group in parent */ |
6139 | if (parent->groups == parent->groups->next) { | |
6140 | pflags &= ~(SD_LOAD_BALANCE | | |
6141 | SD_BALANCE_NEWIDLE | | |
6142 | SD_BALANCE_FORK | | |
89c4710e SS |
6143 | SD_BALANCE_EXEC | |
6144 | SD_SHARE_CPUPOWER | | |
6145 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6146 | if (nr_node_ids == 1) |
6147 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6148 | } |
6149 | if (~cflags & pflags) | |
6150 | return 0; | |
6151 | ||
6152 | return 1; | |
6153 | } | |
6154 | ||
c6c4927b RR |
6155 | static void free_rootdomain(struct root_domain *rd) |
6156 | { | |
047106ad PZ |
6157 | synchronize_sched(); |
6158 | ||
68e74568 RR |
6159 | cpupri_cleanup(&rd->cpupri); |
6160 | ||
c6c4927b RR |
6161 | free_cpumask_var(rd->rto_mask); |
6162 | free_cpumask_var(rd->online); | |
6163 | free_cpumask_var(rd->span); | |
6164 | kfree(rd); | |
6165 | } | |
6166 | ||
57d885fe GH |
6167 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6168 | { | |
a0490fa3 | 6169 | struct root_domain *old_rd = NULL; |
57d885fe | 6170 | unsigned long flags; |
57d885fe | 6171 | |
05fa785c | 6172 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6173 | |
6174 | if (rq->rd) { | |
a0490fa3 | 6175 | old_rd = rq->rd; |
57d885fe | 6176 | |
c6c4927b | 6177 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6178 | set_rq_offline(rq); |
57d885fe | 6179 | |
c6c4927b | 6180 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6181 | |
a0490fa3 IM |
6182 | /* |
6183 | * If we dont want to free the old_rt yet then | |
6184 | * set old_rd to NULL to skip the freeing later | |
6185 | * in this function: | |
6186 | */ | |
6187 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6188 | old_rd = NULL; | |
57d885fe GH |
6189 | } |
6190 | ||
6191 | atomic_inc(&rd->refcount); | |
6192 | rq->rd = rd; | |
6193 | ||
c6c4927b | 6194 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6195 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6196 | set_rq_online(rq); |
57d885fe | 6197 | |
05fa785c | 6198 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6199 | |
6200 | if (old_rd) | |
6201 | free_rootdomain(old_rd); | |
57d885fe GH |
6202 | } |
6203 | ||
fd5e1b5d | 6204 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6205 | { |
36b7b6d4 PE |
6206 | gfp_t gfp = GFP_KERNEL; |
6207 | ||
57d885fe GH |
6208 | memset(rd, 0, sizeof(*rd)); |
6209 | ||
36b7b6d4 PE |
6210 | if (bootmem) |
6211 | gfp = GFP_NOWAIT; | |
c6c4927b | 6212 | |
36b7b6d4 | 6213 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6214 | goto out; |
36b7b6d4 | 6215 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6216 | goto free_span; |
36b7b6d4 | 6217 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6218 | goto free_online; |
6e0534f2 | 6219 | |
0fb53029 | 6220 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6221 | goto free_rto_mask; |
c6c4927b | 6222 | return 0; |
6e0534f2 | 6223 | |
68e74568 RR |
6224 | free_rto_mask: |
6225 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6226 | free_online: |
6227 | free_cpumask_var(rd->online); | |
6228 | free_span: | |
6229 | free_cpumask_var(rd->span); | |
0c910d28 | 6230 | out: |
c6c4927b | 6231 | return -ENOMEM; |
57d885fe GH |
6232 | } |
6233 | ||
6234 | static void init_defrootdomain(void) | |
6235 | { | |
c6c4927b RR |
6236 | init_rootdomain(&def_root_domain, true); |
6237 | ||
57d885fe GH |
6238 | atomic_set(&def_root_domain.refcount, 1); |
6239 | } | |
6240 | ||
dc938520 | 6241 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6242 | { |
6243 | struct root_domain *rd; | |
6244 | ||
6245 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6246 | if (!rd) | |
6247 | return NULL; | |
6248 | ||
c6c4927b RR |
6249 | if (init_rootdomain(rd, false) != 0) { |
6250 | kfree(rd); | |
6251 | return NULL; | |
6252 | } | |
57d885fe GH |
6253 | |
6254 | return rd; | |
6255 | } | |
6256 | ||
1da177e4 | 6257 | /* |
0eab9146 | 6258 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6259 | * hold the hotplug lock. |
6260 | */ | |
0eab9146 IM |
6261 | static void |
6262 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6263 | { |
70b97a7f | 6264 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6265 | struct sched_domain *tmp; |
6266 | ||
6267 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6268 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6269 | struct sched_domain *parent = tmp->parent; |
6270 | if (!parent) | |
6271 | break; | |
f29c9b1c | 6272 | |
1a848870 | 6273 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6274 | tmp->parent = parent->parent; |
1a848870 SS |
6275 | if (parent->parent) |
6276 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6277 | } else |
6278 | tmp = tmp->parent; | |
245af2c7 SS |
6279 | } |
6280 | ||
1a848870 | 6281 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6282 | sd = sd->parent; |
1a848870 SS |
6283 | if (sd) |
6284 | sd->child = NULL; | |
6285 | } | |
1da177e4 LT |
6286 | |
6287 | sched_domain_debug(sd, cpu); | |
6288 | ||
57d885fe | 6289 | rq_attach_root(rq, rd); |
674311d5 | 6290 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6291 | } |
6292 | ||
6293 | /* cpus with isolated domains */ | |
dcc30a35 | 6294 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6295 | |
6296 | /* Setup the mask of cpus configured for isolated domains */ | |
6297 | static int __init isolated_cpu_setup(char *str) | |
6298 | { | |
bdddd296 | 6299 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6300 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6301 | return 1; |
6302 | } | |
6303 | ||
8927f494 | 6304 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6305 | |
6306 | /* | |
6711cab4 SS |
6307 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6308 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6309 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6310 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6311 | * |
6312 | * init_sched_build_groups will build a circular linked list of the groups | |
6313 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6314 | * and ->cpu_power to 0. | |
6315 | */ | |
a616058b | 6316 | static void |
96f874e2 RR |
6317 | init_sched_build_groups(const struct cpumask *span, |
6318 | const struct cpumask *cpu_map, | |
6319 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6320 | struct sched_group **sg, |
96f874e2 RR |
6321 | struct cpumask *tmpmask), |
6322 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6323 | { |
6324 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6325 | int i; |
6326 | ||
96f874e2 | 6327 | cpumask_clear(covered); |
7c16ec58 | 6328 | |
abcd083a | 6329 | for_each_cpu(i, span) { |
6711cab4 | 6330 | struct sched_group *sg; |
7c16ec58 | 6331 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6332 | int j; |
6333 | ||
758b2cdc | 6334 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6335 | continue; |
6336 | ||
758b2cdc | 6337 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6338 | sg->cpu_power = 0; |
1da177e4 | 6339 | |
abcd083a | 6340 | for_each_cpu(j, span) { |
7c16ec58 | 6341 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6342 | continue; |
6343 | ||
96f874e2 | 6344 | cpumask_set_cpu(j, covered); |
758b2cdc | 6345 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6346 | } |
6347 | if (!first) | |
6348 | first = sg; | |
6349 | if (last) | |
6350 | last->next = sg; | |
6351 | last = sg; | |
6352 | } | |
6353 | last->next = first; | |
6354 | } | |
6355 | ||
9c1cfda2 | 6356 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6357 | |
9c1cfda2 | 6358 | #ifdef CONFIG_NUMA |
198e2f18 | 6359 | |
9c1cfda2 JH |
6360 | /** |
6361 | * find_next_best_node - find the next node to include in a sched_domain | |
6362 | * @node: node whose sched_domain we're building | |
6363 | * @used_nodes: nodes already in the sched_domain | |
6364 | * | |
41a2d6cf | 6365 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6366 | * finds the closest node not already in the @used_nodes map. |
6367 | * | |
6368 | * Should use nodemask_t. | |
6369 | */ | |
c5f59f08 | 6370 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6371 | { |
6372 | int i, n, val, min_val, best_node = 0; | |
6373 | ||
6374 | min_val = INT_MAX; | |
6375 | ||
076ac2af | 6376 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6377 | /* Start at @node */ |
076ac2af | 6378 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6379 | |
6380 | if (!nr_cpus_node(n)) | |
6381 | continue; | |
6382 | ||
6383 | /* Skip already used nodes */ | |
c5f59f08 | 6384 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6385 | continue; |
6386 | ||
6387 | /* Simple min distance search */ | |
6388 | val = node_distance(node, n); | |
6389 | ||
6390 | if (val < min_val) { | |
6391 | min_val = val; | |
6392 | best_node = n; | |
6393 | } | |
6394 | } | |
6395 | ||
c5f59f08 | 6396 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6397 | return best_node; |
6398 | } | |
6399 | ||
6400 | /** | |
6401 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6402 | * @node: node whose cpumask we're constructing | |
73486722 | 6403 | * @span: resulting cpumask |
9c1cfda2 | 6404 | * |
41a2d6cf | 6405 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6406 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6407 | * out optimally. | |
6408 | */ | |
96f874e2 | 6409 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6410 | { |
c5f59f08 | 6411 | nodemask_t used_nodes; |
48f24c4d | 6412 | int i; |
9c1cfda2 | 6413 | |
6ca09dfc | 6414 | cpumask_clear(span); |
c5f59f08 | 6415 | nodes_clear(used_nodes); |
9c1cfda2 | 6416 | |
6ca09dfc | 6417 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6418 | node_set(node, used_nodes); |
9c1cfda2 JH |
6419 | |
6420 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6421 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6422 | |
6ca09dfc | 6423 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6424 | } |
9c1cfda2 | 6425 | } |
6d6bc0ad | 6426 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6427 | |
5c45bf27 | 6428 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6429 | |
6c99e9ad RR |
6430 | /* |
6431 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6432 | * |
6433 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6434 | * and struct sched_domain. ) | |
6c99e9ad RR |
6435 | */ |
6436 | struct static_sched_group { | |
6437 | struct sched_group sg; | |
6438 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6439 | }; | |
6440 | ||
6441 | struct static_sched_domain { | |
6442 | struct sched_domain sd; | |
6443 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6444 | }; | |
6445 | ||
49a02c51 AH |
6446 | struct s_data { |
6447 | #ifdef CONFIG_NUMA | |
6448 | int sd_allnodes; | |
6449 | cpumask_var_t domainspan; | |
6450 | cpumask_var_t covered; | |
6451 | cpumask_var_t notcovered; | |
6452 | #endif | |
6453 | cpumask_var_t nodemask; | |
6454 | cpumask_var_t this_sibling_map; | |
6455 | cpumask_var_t this_core_map; | |
6456 | cpumask_var_t send_covered; | |
6457 | cpumask_var_t tmpmask; | |
6458 | struct sched_group **sched_group_nodes; | |
6459 | struct root_domain *rd; | |
6460 | }; | |
6461 | ||
2109b99e AH |
6462 | enum s_alloc { |
6463 | sa_sched_groups = 0, | |
6464 | sa_rootdomain, | |
6465 | sa_tmpmask, | |
6466 | sa_send_covered, | |
6467 | sa_this_core_map, | |
6468 | sa_this_sibling_map, | |
6469 | sa_nodemask, | |
6470 | sa_sched_group_nodes, | |
6471 | #ifdef CONFIG_NUMA | |
6472 | sa_notcovered, | |
6473 | sa_covered, | |
6474 | sa_domainspan, | |
6475 | #endif | |
6476 | sa_none, | |
6477 | }; | |
6478 | ||
9c1cfda2 | 6479 | /* |
48f24c4d | 6480 | * SMT sched-domains: |
9c1cfda2 | 6481 | */ |
1da177e4 | 6482 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6483 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6484 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6485 | |
41a2d6cf | 6486 | static int |
96f874e2 RR |
6487 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6488 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6489 | { |
6711cab4 | 6490 | if (sg) |
1871e52c | 6491 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6492 | return cpu; |
6493 | } | |
6d6bc0ad | 6494 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6495 | |
48f24c4d IM |
6496 | /* |
6497 | * multi-core sched-domains: | |
6498 | */ | |
1e9f28fa | 6499 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6500 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6501 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6502 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6503 | |
6504 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6505 | static int |
96f874e2 RR |
6506 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6507 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6508 | { |
6711cab4 | 6509 | int group; |
7c16ec58 | 6510 | |
c69fc56d | 6511 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6512 | group = cpumask_first(mask); |
6711cab4 | 6513 | if (sg) |
6c99e9ad | 6514 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6515 | return group; |
1e9f28fa SS |
6516 | } |
6517 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6518 | static int |
96f874e2 RR |
6519 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6520 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6521 | { |
6711cab4 | 6522 | if (sg) |
6c99e9ad | 6523 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6524 | return cpu; |
6525 | } | |
6526 | #endif | |
6527 | ||
6c99e9ad RR |
6528 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6529 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6530 | |
41a2d6cf | 6531 | static int |
96f874e2 RR |
6532 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6533 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6534 | { |
6711cab4 | 6535 | int group; |
48f24c4d | 6536 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6537 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6538 | group = cpumask_first(mask); |
1e9f28fa | 6539 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6540 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6541 | group = cpumask_first(mask); |
1da177e4 | 6542 | #else |
6711cab4 | 6543 | group = cpu; |
1da177e4 | 6544 | #endif |
6711cab4 | 6545 | if (sg) |
6c99e9ad | 6546 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6547 | return group; |
1da177e4 LT |
6548 | } |
6549 | ||
6550 | #ifdef CONFIG_NUMA | |
1da177e4 | 6551 | /* |
9c1cfda2 JH |
6552 | * The init_sched_build_groups can't handle what we want to do with node |
6553 | * groups, so roll our own. Now each node has its own list of groups which | |
6554 | * gets dynamically allocated. | |
1da177e4 | 6555 | */ |
62ea9ceb | 6556 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6557 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6558 | |
62ea9ceb | 6559 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6560 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6561 | |
96f874e2 RR |
6562 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6563 | struct sched_group **sg, | |
6564 | struct cpumask *nodemask) | |
9c1cfda2 | 6565 | { |
6711cab4 SS |
6566 | int group; |
6567 | ||
6ca09dfc | 6568 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6569 | group = cpumask_first(nodemask); |
6711cab4 SS |
6570 | |
6571 | if (sg) | |
6c99e9ad | 6572 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6573 | return group; |
1da177e4 | 6574 | } |
6711cab4 | 6575 | |
08069033 SS |
6576 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6577 | { | |
6578 | struct sched_group *sg = group_head; | |
6579 | int j; | |
6580 | ||
6581 | if (!sg) | |
6582 | return; | |
3a5c359a | 6583 | do { |
758b2cdc | 6584 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6585 | struct sched_domain *sd; |
08069033 | 6586 | |
6c99e9ad | 6587 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6588 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6589 | /* |
6590 | * Only add "power" once for each | |
6591 | * physical package. | |
6592 | */ | |
6593 | continue; | |
6594 | } | |
08069033 | 6595 | |
18a3885f | 6596 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6597 | } |
6598 | sg = sg->next; | |
6599 | } while (sg != group_head); | |
08069033 | 6600 | } |
0601a88d AH |
6601 | |
6602 | static int build_numa_sched_groups(struct s_data *d, | |
6603 | const struct cpumask *cpu_map, int num) | |
6604 | { | |
6605 | struct sched_domain *sd; | |
6606 | struct sched_group *sg, *prev; | |
6607 | int n, j; | |
6608 | ||
6609 | cpumask_clear(d->covered); | |
6610 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6611 | if (cpumask_empty(d->nodemask)) { | |
6612 | d->sched_group_nodes[num] = NULL; | |
6613 | goto out; | |
6614 | } | |
6615 | ||
6616 | sched_domain_node_span(num, d->domainspan); | |
6617 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6618 | ||
6619 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6620 | GFP_KERNEL, num); | |
6621 | if (!sg) { | |
3df0fc5b PZ |
6622 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6623 | num); | |
0601a88d AH |
6624 | return -ENOMEM; |
6625 | } | |
6626 | d->sched_group_nodes[num] = sg; | |
6627 | ||
6628 | for_each_cpu(j, d->nodemask) { | |
6629 | sd = &per_cpu(node_domains, j).sd; | |
6630 | sd->groups = sg; | |
6631 | } | |
6632 | ||
18a3885f | 6633 | sg->cpu_power = 0; |
0601a88d AH |
6634 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6635 | sg->next = sg; | |
6636 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6637 | ||
6638 | prev = sg; | |
6639 | for (j = 0; j < nr_node_ids; j++) { | |
6640 | n = (num + j) % nr_node_ids; | |
6641 | cpumask_complement(d->notcovered, d->covered); | |
6642 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6643 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6644 | if (cpumask_empty(d->tmpmask)) | |
6645 | break; | |
6646 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6647 | if (cpumask_empty(d->tmpmask)) | |
6648 | continue; | |
6649 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6650 | GFP_KERNEL, num); | |
6651 | if (!sg) { | |
3df0fc5b PZ |
6652 | printk(KERN_WARNING |
6653 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6654 | return -ENOMEM; |
6655 | } | |
18a3885f | 6656 | sg->cpu_power = 0; |
0601a88d AH |
6657 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6658 | sg->next = prev->next; | |
6659 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6660 | prev->next = sg; | |
6661 | prev = sg; | |
6662 | } | |
6663 | out: | |
6664 | return 0; | |
6665 | } | |
6d6bc0ad | 6666 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6667 | |
a616058b | 6668 | #ifdef CONFIG_NUMA |
51888ca2 | 6669 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6670 | static void free_sched_groups(const struct cpumask *cpu_map, |
6671 | struct cpumask *nodemask) | |
51888ca2 | 6672 | { |
a616058b | 6673 | int cpu, i; |
51888ca2 | 6674 | |
abcd083a | 6675 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6676 | struct sched_group **sched_group_nodes |
6677 | = sched_group_nodes_bycpu[cpu]; | |
6678 | ||
51888ca2 SV |
6679 | if (!sched_group_nodes) |
6680 | continue; | |
6681 | ||
076ac2af | 6682 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6683 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6684 | ||
6ca09dfc | 6685 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6686 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6687 | continue; |
6688 | ||
6689 | if (sg == NULL) | |
6690 | continue; | |
6691 | sg = sg->next; | |
6692 | next_sg: | |
6693 | oldsg = sg; | |
6694 | sg = sg->next; | |
6695 | kfree(oldsg); | |
6696 | if (oldsg != sched_group_nodes[i]) | |
6697 | goto next_sg; | |
6698 | } | |
6699 | kfree(sched_group_nodes); | |
6700 | sched_group_nodes_bycpu[cpu] = NULL; | |
6701 | } | |
51888ca2 | 6702 | } |
6d6bc0ad | 6703 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6704 | static void free_sched_groups(const struct cpumask *cpu_map, |
6705 | struct cpumask *nodemask) | |
a616058b SS |
6706 | { |
6707 | } | |
6d6bc0ad | 6708 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6709 | |
89c4710e SS |
6710 | /* |
6711 | * Initialize sched groups cpu_power. | |
6712 | * | |
6713 | * cpu_power indicates the capacity of sched group, which is used while | |
6714 | * distributing the load between different sched groups in a sched domain. | |
6715 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6716 | * there are asymmetries in the topology. If there are asymmetries, group | |
6717 | * having more cpu_power will pickup more load compared to the group having | |
6718 | * less cpu_power. | |
89c4710e SS |
6719 | */ |
6720 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6721 | { | |
6722 | struct sched_domain *child; | |
6723 | struct sched_group *group; | |
f93e65c1 PZ |
6724 | long power; |
6725 | int weight; | |
89c4710e SS |
6726 | |
6727 | WARN_ON(!sd || !sd->groups); | |
6728 | ||
13318a71 | 6729 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6730 | return; |
6731 | ||
6732 | child = sd->child; | |
6733 | ||
18a3885f | 6734 | sd->groups->cpu_power = 0; |
5517d86b | 6735 | |
f93e65c1 PZ |
6736 | if (!child) { |
6737 | power = SCHED_LOAD_SCALE; | |
6738 | weight = cpumask_weight(sched_domain_span(sd)); | |
6739 | /* | |
6740 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6741 | * Usually multiple threads get a better yield out of |
6742 | * that one core than a single thread would have, | |
6743 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6744 | */ |
a52bfd73 PZ |
6745 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6746 | power *= sd->smt_gain; | |
f93e65c1 | 6747 | power /= weight; |
a52bfd73 PZ |
6748 | power >>= SCHED_LOAD_SHIFT; |
6749 | } | |
18a3885f | 6750 | sd->groups->cpu_power += power; |
89c4710e SS |
6751 | return; |
6752 | } | |
6753 | ||
89c4710e | 6754 | /* |
f93e65c1 | 6755 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6756 | */ |
6757 | group = child->groups; | |
6758 | do { | |
18a3885f | 6759 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6760 | group = group->next; |
6761 | } while (group != child->groups); | |
6762 | } | |
6763 | ||
7c16ec58 MT |
6764 | /* |
6765 | * Initializers for schedule domains | |
6766 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6767 | */ | |
6768 | ||
a5d8c348 IM |
6769 | #ifdef CONFIG_SCHED_DEBUG |
6770 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6771 | #else | |
6772 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6773 | #endif | |
6774 | ||
7c16ec58 | 6775 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6776 | |
7c16ec58 MT |
6777 | #define SD_INIT_FUNC(type) \ |
6778 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6779 | { \ | |
6780 | memset(sd, 0, sizeof(*sd)); \ | |
6781 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6782 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6783 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6784 | } |
6785 | ||
6786 | SD_INIT_FUNC(CPU) | |
6787 | #ifdef CONFIG_NUMA | |
6788 | SD_INIT_FUNC(ALLNODES) | |
6789 | SD_INIT_FUNC(NODE) | |
6790 | #endif | |
6791 | #ifdef CONFIG_SCHED_SMT | |
6792 | SD_INIT_FUNC(SIBLING) | |
6793 | #endif | |
6794 | #ifdef CONFIG_SCHED_MC | |
6795 | SD_INIT_FUNC(MC) | |
6796 | #endif | |
6797 | ||
1d3504fc HS |
6798 | static int default_relax_domain_level = -1; |
6799 | ||
6800 | static int __init setup_relax_domain_level(char *str) | |
6801 | { | |
30e0e178 LZ |
6802 | unsigned long val; |
6803 | ||
6804 | val = simple_strtoul(str, NULL, 0); | |
6805 | if (val < SD_LV_MAX) | |
6806 | default_relax_domain_level = val; | |
6807 | ||
1d3504fc HS |
6808 | return 1; |
6809 | } | |
6810 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6811 | ||
6812 | static void set_domain_attribute(struct sched_domain *sd, | |
6813 | struct sched_domain_attr *attr) | |
6814 | { | |
6815 | int request; | |
6816 | ||
6817 | if (!attr || attr->relax_domain_level < 0) { | |
6818 | if (default_relax_domain_level < 0) | |
6819 | return; | |
6820 | else | |
6821 | request = default_relax_domain_level; | |
6822 | } else | |
6823 | request = attr->relax_domain_level; | |
6824 | if (request < sd->level) { | |
6825 | /* turn off idle balance on this domain */ | |
c88d5910 | 6826 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6827 | } else { |
6828 | /* turn on idle balance on this domain */ | |
c88d5910 | 6829 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6830 | } |
6831 | } | |
6832 | ||
2109b99e AH |
6833 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6834 | const struct cpumask *cpu_map) | |
6835 | { | |
6836 | switch (what) { | |
6837 | case sa_sched_groups: | |
6838 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6839 | d->sched_group_nodes = NULL; | |
6840 | case sa_rootdomain: | |
6841 | free_rootdomain(d->rd); /* fall through */ | |
6842 | case sa_tmpmask: | |
6843 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6844 | case sa_send_covered: | |
6845 | free_cpumask_var(d->send_covered); /* fall through */ | |
6846 | case sa_this_core_map: | |
6847 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6848 | case sa_this_sibling_map: | |
6849 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6850 | case sa_nodemask: | |
6851 | free_cpumask_var(d->nodemask); /* fall through */ | |
6852 | case sa_sched_group_nodes: | |
d1b55138 | 6853 | #ifdef CONFIG_NUMA |
2109b99e AH |
6854 | kfree(d->sched_group_nodes); /* fall through */ |
6855 | case sa_notcovered: | |
6856 | free_cpumask_var(d->notcovered); /* fall through */ | |
6857 | case sa_covered: | |
6858 | free_cpumask_var(d->covered); /* fall through */ | |
6859 | case sa_domainspan: | |
6860 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6861 | #endif |
2109b99e AH |
6862 | case sa_none: |
6863 | break; | |
6864 | } | |
6865 | } | |
3404c8d9 | 6866 | |
2109b99e AH |
6867 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6868 | const struct cpumask *cpu_map) | |
6869 | { | |
3404c8d9 | 6870 | #ifdef CONFIG_NUMA |
2109b99e AH |
6871 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6872 | return sa_none; | |
6873 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6874 | return sa_domainspan; | |
6875 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6876 | return sa_covered; | |
6877 | /* Allocate the per-node list of sched groups */ | |
6878 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6879 | sizeof(struct sched_group *), GFP_KERNEL); | |
6880 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6881 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6882 | return sa_notcovered; |
d1b55138 | 6883 | } |
2109b99e | 6884 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6885 | #endif |
2109b99e AH |
6886 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6887 | return sa_sched_group_nodes; | |
6888 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6889 | return sa_nodemask; | |
6890 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6891 | return sa_this_sibling_map; | |
6892 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6893 | return sa_this_core_map; | |
6894 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6895 | return sa_send_covered; | |
6896 | d->rd = alloc_rootdomain(); | |
6897 | if (!d->rd) { | |
3df0fc5b | 6898 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6899 | return sa_tmpmask; |
57d885fe | 6900 | } |
2109b99e AH |
6901 | return sa_rootdomain; |
6902 | } | |
57d885fe | 6903 | |
7f4588f3 AH |
6904 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6905 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6906 | { | |
6907 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6908 | #ifdef CONFIG_NUMA |
7f4588f3 | 6909 | struct sched_domain *parent; |
1da177e4 | 6910 | |
7f4588f3 AH |
6911 | d->sd_allnodes = 0; |
6912 | if (cpumask_weight(cpu_map) > | |
6913 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6914 | sd = &per_cpu(allnodes_domains, i).sd; | |
6915 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6916 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6917 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6918 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6919 | d->sd_allnodes = 1; | |
6920 | } | |
6921 | parent = sd; | |
6922 | ||
6923 | sd = &per_cpu(node_domains, i).sd; | |
6924 | SD_INIT(sd, NODE); | |
6925 | set_domain_attribute(sd, attr); | |
6926 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6927 | sd->parent = parent; | |
6928 | if (parent) | |
6929 | parent->child = sd; | |
6930 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6931 | #endif |
7f4588f3 AH |
6932 | return sd; |
6933 | } | |
1da177e4 | 6934 | |
87cce662 AH |
6935 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6936 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6937 | struct sched_domain *parent, int i) | |
6938 | { | |
6939 | struct sched_domain *sd; | |
6940 | sd = &per_cpu(phys_domains, i).sd; | |
6941 | SD_INIT(sd, CPU); | |
6942 | set_domain_attribute(sd, attr); | |
6943 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6944 | sd->parent = parent; | |
6945 | if (parent) | |
6946 | parent->child = sd; | |
6947 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6948 | return sd; | |
6949 | } | |
1da177e4 | 6950 | |
410c4081 AH |
6951 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
6952 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6953 | struct sched_domain *parent, int i) | |
6954 | { | |
6955 | struct sched_domain *sd = parent; | |
1e9f28fa | 6956 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
6957 | sd = &per_cpu(core_domains, i).sd; |
6958 | SD_INIT(sd, MC); | |
6959 | set_domain_attribute(sd, attr); | |
6960 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
6961 | sd->parent = parent; | |
6962 | parent->child = sd; | |
6963 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 6964 | #endif |
410c4081 AH |
6965 | return sd; |
6966 | } | |
1e9f28fa | 6967 | |
d8173535 AH |
6968 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
6969 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6970 | struct sched_domain *parent, int i) | |
6971 | { | |
6972 | struct sched_domain *sd = parent; | |
1da177e4 | 6973 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
6974 | sd = &per_cpu(cpu_domains, i).sd; |
6975 | SD_INIT(sd, SIBLING); | |
6976 | set_domain_attribute(sd, attr); | |
6977 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
6978 | sd->parent = parent; | |
6979 | parent->child = sd; | |
6980 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 6981 | #endif |
d8173535 AH |
6982 | return sd; |
6983 | } | |
1da177e4 | 6984 | |
0e8e85c9 AH |
6985 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
6986 | const struct cpumask *cpu_map, int cpu) | |
6987 | { | |
6988 | switch (l) { | |
1da177e4 | 6989 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
6990 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
6991 | cpumask_and(d->this_sibling_map, cpu_map, | |
6992 | topology_thread_cpumask(cpu)); | |
6993 | if (cpu == cpumask_first(d->this_sibling_map)) | |
6994 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
6995 | &cpu_to_cpu_group, | |
6996 | d->send_covered, d->tmpmask); | |
6997 | break; | |
1da177e4 | 6998 | #endif |
1e9f28fa | 6999 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7000 | case SD_LV_MC: /* set up multi-core groups */ |
7001 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7002 | if (cpu == cpumask_first(d->this_core_map)) | |
7003 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7004 | &cpu_to_core_group, | |
7005 | d->send_covered, d->tmpmask); | |
7006 | break; | |
1e9f28fa | 7007 | #endif |
86548096 AH |
7008 | case SD_LV_CPU: /* set up physical groups */ |
7009 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7010 | if (!cpumask_empty(d->nodemask)) | |
7011 | init_sched_build_groups(d->nodemask, cpu_map, | |
7012 | &cpu_to_phys_group, | |
7013 | d->send_covered, d->tmpmask); | |
7014 | break; | |
1da177e4 | 7015 | #ifdef CONFIG_NUMA |
de616e36 AH |
7016 | case SD_LV_ALLNODES: |
7017 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7018 | d->send_covered, d->tmpmask); | |
7019 | break; | |
7020 | #endif | |
0e8e85c9 AH |
7021 | default: |
7022 | break; | |
7c16ec58 | 7023 | } |
0e8e85c9 | 7024 | } |
9c1cfda2 | 7025 | |
2109b99e AH |
7026 | /* |
7027 | * Build sched domains for a given set of cpus and attach the sched domains | |
7028 | * to the individual cpus | |
7029 | */ | |
7030 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7031 | struct sched_domain_attr *attr) | |
7032 | { | |
7033 | enum s_alloc alloc_state = sa_none; | |
7034 | struct s_data d; | |
294b0c96 | 7035 | struct sched_domain *sd; |
2109b99e | 7036 | int i; |
7c16ec58 | 7037 | #ifdef CONFIG_NUMA |
2109b99e | 7038 | d.sd_allnodes = 0; |
7c16ec58 | 7039 | #endif |
9c1cfda2 | 7040 | |
2109b99e AH |
7041 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7042 | if (alloc_state != sa_rootdomain) | |
7043 | goto error; | |
7044 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7045 | |
1da177e4 | 7046 | /* |
1a20ff27 | 7047 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7048 | */ |
abcd083a | 7049 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7050 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7051 | cpu_map); | |
9761eea8 | 7052 | |
7f4588f3 | 7053 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7054 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7055 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7056 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7057 | } |
9c1cfda2 | 7058 | |
abcd083a | 7059 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7060 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 7061 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7062 | } |
9c1cfda2 | 7063 | |
1da177e4 | 7064 | /* Set up physical groups */ |
86548096 AH |
7065 | for (i = 0; i < nr_node_ids; i++) |
7066 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7067 | |
1da177e4 LT |
7068 | #ifdef CONFIG_NUMA |
7069 | /* Set up node groups */ | |
de616e36 AH |
7070 | if (d.sd_allnodes) |
7071 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7072 | |
0601a88d AH |
7073 | for (i = 0; i < nr_node_ids; i++) |
7074 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7075 | goto error; |
1da177e4 LT |
7076 | #endif |
7077 | ||
7078 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7079 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7080 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7081 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7082 | init_sched_groups_power(i, sd); |
5c45bf27 | 7083 | } |
1da177e4 | 7084 | #endif |
1e9f28fa | 7085 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7086 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7087 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7088 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7089 | } |
7090 | #endif | |
1e9f28fa | 7091 | |
abcd083a | 7092 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7093 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7094 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7095 | } |
7096 | ||
9c1cfda2 | 7097 | #ifdef CONFIG_NUMA |
076ac2af | 7098 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7099 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7100 | |
49a02c51 | 7101 | if (d.sd_allnodes) { |
6711cab4 | 7102 | struct sched_group *sg; |
f712c0c7 | 7103 | |
96f874e2 | 7104 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7105 | d.tmpmask); |
f712c0c7 SS |
7106 | init_numa_sched_groups_power(sg); |
7107 | } | |
9c1cfda2 JH |
7108 | #endif |
7109 | ||
1da177e4 | 7110 | /* Attach the domains */ |
abcd083a | 7111 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7112 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7113 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7114 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7115 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7116 | #else |
6c99e9ad | 7117 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7118 | #endif |
49a02c51 | 7119 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7120 | } |
51888ca2 | 7121 | |
2109b99e AH |
7122 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7123 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7124 | return 0; | |
51888ca2 | 7125 | |
51888ca2 | 7126 | error: |
2109b99e AH |
7127 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7128 | return -ENOMEM; | |
1da177e4 | 7129 | } |
029190c5 | 7130 | |
96f874e2 | 7131 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7132 | { |
7133 | return __build_sched_domains(cpu_map, NULL); | |
7134 | } | |
7135 | ||
acc3f5d7 | 7136 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7137 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7138 | static struct sched_domain_attr *dattr_cur; |
7139 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7140 | |
7141 | /* | |
7142 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7143 | * cpumask) fails, then fallback to a single sched domain, |
7144 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7145 | */ |
4212823f | 7146 | static cpumask_var_t fallback_doms; |
029190c5 | 7147 | |
ee79d1bd HC |
7148 | /* |
7149 | * arch_update_cpu_topology lets virtualized architectures update the | |
7150 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7151 | * or 0 if it stayed the same. | |
7152 | */ | |
7153 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7154 | { |
ee79d1bd | 7155 | return 0; |
22e52b07 HC |
7156 | } |
7157 | ||
acc3f5d7 RR |
7158 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7159 | { | |
7160 | int i; | |
7161 | cpumask_var_t *doms; | |
7162 | ||
7163 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7164 | if (!doms) | |
7165 | return NULL; | |
7166 | for (i = 0; i < ndoms; i++) { | |
7167 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7168 | free_sched_domains(doms, i); | |
7169 | return NULL; | |
7170 | } | |
7171 | } | |
7172 | return doms; | |
7173 | } | |
7174 | ||
7175 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7176 | { | |
7177 | unsigned int i; | |
7178 | for (i = 0; i < ndoms; i++) | |
7179 | free_cpumask_var(doms[i]); | |
7180 | kfree(doms); | |
7181 | } | |
7182 | ||
1a20ff27 | 7183 | /* |
41a2d6cf | 7184 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7185 | * For now this just excludes isolated cpus, but could be used to |
7186 | * exclude other special cases in the future. | |
1a20ff27 | 7187 | */ |
96f874e2 | 7188 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7189 | { |
7378547f MM |
7190 | int err; |
7191 | ||
22e52b07 | 7192 | arch_update_cpu_topology(); |
029190c5 | 7193 | ndoms_cur = 1; |
acc3f5d7 | 7194 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7195 | if (!doms_cur) |
acc3f5d7 RR |
7196 | doms_cur = &fallback_doms; |
7197 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7198 | dattr_cur = NULL; |
acc3f5d7 | 7199 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7200 | register_sched_domain_sysctl(); |
7378547f MM |
7201 | |
7202 | return err; | |
1a20ff27 DG |
7203 | } |
7204 | ||
96f874e2 RR |
7205 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7206 | struct cpumask *tmpmask) | |
1da177e4 | 7207 | { |
7c16ec58 | 7208 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7209 | } |
1da177e4 | 7210 | |
1a20ff27 DG |
7211 | /* |
7212 | * Detach sched domains from a group of cpus specified in cpu_map | |
7213 | * These cpus will now be attached to the NULL domain | |
7214 | */ | |
96f874e2 | 7215 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7216 | { |
96f874e2 RR |
7217 | /* Save because hotplug lock held. */ |
7218 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7219 | int i; |
7220 | ||
abcd083a | 7221 | for_each_cpu(i, cpu_map) |
57d885fe | 7222 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7223 | synchronize_sched(); |
96f874e2 | 7224 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7225 | } |
7226 | ||
1d3504fc HS |
7227 | /* handle null as "default" */ |
7228 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7229 | struct sched_domain_attr *new, int idx_new) | |
7230 | { | |
7231 | struct sched_domain_attr tmp; | |
7232 | ||
7233 | /* fast path */ | |
7234 | if (!new && !cur) | |
7235 | return 1; | |
7236 | ||
7237 | tmp = SD_ATTR_INIT; | |
7238 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7239 | new ? (new + idx_new) : &tmp, | |
7240 | sizeof(struct sched_domain_attr)); | |
7241 | } | |
7242 | ||
029190c5 PJ |
7243 | /* |
7244 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7245 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7246 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7247 | * It destroys each deleted domain and builds each new domain. | |
7248 | * | |
acc3f5d7 | 7249 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7250 | * The masks don't intersect (don't overlap.) We should setup one |
7251 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7252 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7253 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7254 | * it as it is. | |
7255 | * | |
acc3f5d7 RR |
7256 | * The passed in 'doms_new' should be allocated using |
7257 | * alloc_sched_domains. This routine takes ownership of it and will | |
7258 | * free_sched_domains it when done with it. If the caller failed the | |
7259 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7260 | * and partition_sched_domains() will fallback to the single partition | |
7261 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7262 | * |
96f874e2 | 7263 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7264 | * ndoms_new == 0 is a special case for destroying existing domains, |
7265 | * and it will not create the default domain. | |
dfb512ec | 7266 | * |
029190c5 PJ |
7267 | * Call with hotplug lock held |
7268 | */ | |
acc3f5d7 | 7269 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7270 | struct sched_domain_attr *dattr_new) |
029190c5 | 7271 | { |
dfb512ec | 7272 | int i, j, n; |
d65bd5ec | 7273 | int new_topology; |
029190c5 | 7274 | |
712555ee | 7275 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7276 | |
7378547f MM |
7277 | /* always unregister in case we don't destroy any domains */ |
7278 | unregister_sched_domain_sysctl(); | |
7279 | ||
d65bd5ec HC |
7280 | /* Let architecture update cpu core mappings. */ |
7281 | new_topology = arch_update_cpu_topology(); | |
7282 | ||
dfb512ec | 7283 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7284 | |
7285 | /* Destroy deleted domains */ | |
7286 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7287 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7288 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7289 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7290 | goto match1; |
7291 | } | |
7292 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7293 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7294 | match1: |
7295 | ; | |
7296 | } | |
7297 | ||
e761b772 MK |
7298 | if (doms_new == NULL) { |
7299 | ndoms_cur = 0; | |
acc3f5d7 | 7300 | doms_new = &fallback_doms; |
6ad4c188 | 7301 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7302 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7303 | } |
7304 | ||
029190c5 PJ |
7305 | /* Build new domains */ |
7306 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7307 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7308 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7309 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7310 | goto match2; |
7311 | } | |
7312 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7313 | __build_sched_domains(doms_new[i], |
1d3504fc | 7314 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7315 | match2: |
7316 | ; | |
7317 | } | |
7318 | ||
7319 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7320 | if (doms_cur != &fallback_doms) |
7321 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7322 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7323 | doms_cur = doms_new; |
1d3504fc | 7324 | dattr_cur = dattr_new; |
029190c5 | 7325 | ndoms_cur = ndoms_new; |
7378547f MM |
7326 | |
7327 | register_sched_domain_sysctl(); | |
a1835615 | 7328 | |
712555ee | 7329 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7330 | } |
7331 | ||
5c45bf27 | 7332 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7333 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7334 | { |
95402b38 | 7335 | get_online_cpus(); |
dfb512ec MK |
7336 | |
7337 | /* Destroy domains first to force the rebuild */ | |
7338 | partition_sched_domains(0, NULL, NULL); | |
7339 | ||
e761b772 | 7340 | rebuild_sched_domains(); |
95402b38 | 7341 | put_online_cpus(); |
5c45bf27 SS |
7342 | } |
7343 | ||
7344 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7345 | { | |
afb8a9b7 | 7346 | unsigned int level = 0; |
5c45bf27 | 7347 | |
afb8a9b7 GS |
7348 | if (sscanf(buf, "%u", &level) != 1) |
7349 | return -EINVAL; | |
7350 | ||
7351 | /* | |
7352 | * level is always be positive so don't check for | |
7353 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7354 | * What happens on 0 or 1 byte write, | |
7355 | * need to check for count as well? | |
7356 | */ | |
7357 | ||
7358 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7359 | return -EINVAL; |
7360 | ||
7361 | if (smt) | |
afb8a9b7 | 7362 | sched_smt_power_savings = level; |
5c45bf27 | 7363 | else |
afb8a9b7 | 7364 | sched_mc_power_savings = level; |
5c45bf27 | 7365 | |
c70f22d2 | 7366 | arch_reinit_sched_domains(); |
5c45bf27 | 7367 | |
c70f22d2 | 7368 | return count; |
5c45bf27 SS |
7369 | } |
7370 | ||
5c45bf27 | 7371 | #ifdef CONFIG_SCHED_MC |
f718cd4a AK |
7372 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
7373 | char *page) | |
5c45bf27 SS |
7374 | { |
7375 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7376 | } | |
f718cd4a | 7377 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
48f24c4d | 7378 | const char *buf, size_t count) |
5c45bf27 SS |
7379 | { |
7380 | return sched_power_savings_store(buf, count, 0); | |
7381 | } | |
f718cd4a AK |
7382 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7383 | sched_mc_power_savings_show, | |
7384 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7385 | #endif |
7386 | ||
7387 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a AK |
7388 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
7389 | char *page) | |
5c45bf27 SS |
7390 | { |
7391 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7392 | } | |
f718cd4a | 7393 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
48f24c4d | 7394 | const char *buf, size_t count) |
5c45bf27 SS |
7395 | { |
7396 | return sched_power_savings_store(buf, count, 1); | |
7397 | } | |
f718cd4a AK |
7398 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7399 | sched_smt_power_savings_show, | |
6707de00 AB |
7400 | sched_smt_power_savings_store); |
7401 | #endif | |
7402 | ||
39aac648 | 7403 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7404 | { |
7405 | int err = 0; | |
7406 | ||
7407 | #ifdef CONFIG_SCHED_SMT | |
7408 | if (smt_capable()) | |
7409 | err = sysfs_create_file(&cls->kset.kobj, | |
7410 | &attr_sched_smt_power_savings.attr); | |
7411 | #endif | |
7412 | #ifdef CONFIG_SCHED_MC | |
7413 | if (!err && mc_capable()) | |
7414 | err = sysfs_create_file(&cls->kset.kobj, | |
7415 | &attr_sched_mc_power_savings.attr); | |
7416 | #endif | |
7417 | return err; | |
7418 | } | |
6d6bc0ad | 7419 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7420 | |
e761b772 | 7421 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7422 | /* |
e761b772 MK |
7423 | * Add online and remove offline CPUs from the scheduler domains. |
7424 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7425 | */ |
7426 | static int update_sched_domains(struct notifier_block *nfb, | |
7427 | unsigned long action, void *hcpu) | |
e761b772 MK |
7428 | { |
7429 | switch (action) { | |
7430 | case CPU_ONLINE: | |
7431 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
7432 | case CPU_DOWN_PREPARE: |
7433 | case CPU_DOWN_PREPARE_FROZEN: | |
7434 | case CPU_DOWN_FAILED: | |
7435 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 7436 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7437 | return NOTIFY_OK; |
7438 | ||
7439 | default: | |
7440 | return NOTIFY_DONE; | |
7441 | } | |
7442 | } | |
7443 | #endif | |
7444 | ||
7445 | static int update_runtime(struct notifier_block *nfb, | |
7446 | unsigned long action, void *hcpu) | |
1da177e4 | 7447 | { |
7def2be1 PZ |
7448 | int cpu = (int)(long)hcpu; |
7449 | ||
1da177e4 | 7450 | switch (action) { |
1da177e4 | 7451 | case CPU_DOWN_PREPARE: |
8bb78442 | 7452 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7453 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7454 | return NOTIFY_OK; |
7455 | ||
1da177e4 | 7456 | case CPU_DOWN_FAILED: |
8bb78442 | 7457 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7458 | case CPU_ONLINE: |
8bb78442 | 7459 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7460 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7461 | return NOTIFY_OK; |
7462 | ||
1da177e4 LT |
7463 | default: |
7464 | return NOTIFY_DONE; | |
7465 | } | |
1da177e4 | 7466 | } |
1da177e4 LT |
7467 | |
7468 | void __init sched_init_smp(void) | |
7469 | { | |
dcc30a35 RR |
7470 | cpumask_var_t non_isolated_cpus; |
7471 | ||
7472 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7473 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7474 | |
434d53b0 MT |
7475 | #if defined(CONFIG_NUMA) |
7476 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7477 | GFP_KERNEL); | |
7478 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7479 | #endif | |
95402b38 | 7480 | get_online_cpus(); |
712555ee | 7481 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7482 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7483 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7484 | if (cpumask_empty(non_isolated_cpus)) | |
7485 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7486 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7487 | put_online_cpus(); |
e761b772 MK |
7488 | |
7489 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7490 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7491 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7492 | #endif |
7493 | ||
7494 | /* RT runtime code needs to handle some hotplug events */ | |
7495 | hotcpu_notifier(update_runtime, 0); | |
7496 | ||
b328ca18 | 7497 | init_hrtick(); |
5c1e1767 NP |
7498 | |
7499 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7500 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7501 | BUG(); |
19978ca6 | 7502 | sched_init_granularity(); |
dcc30a35 | 7503 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7504 | |
0e3900e6 | 7505 | init_sched_rt_class(); |
1da177e4 LT |
7506 | } |
7507 | #else | |
7508 | void __init sched_init_smp(void) | |
7509 | { | |
19978ca6 | 7510 | sched_init_granularity(); |
1da177e4 LT |
7511 | } |
7512 | #endif /* CONFIG_SMP */ | |
7513 | ||
cd1bb94b AB |
7514 | const_debug unsigned int sysctl_timer_migration = 1; |
7515 | ||
1da177e4 LT |
7516 | int in_sched_functions(unsigned long addr) |
7517 | { | |
1da177e4 LT |
7518 | return in_lock_functions(addr) || |
7519 | (addr >= (unsigned long)__sched_text_start | |
7520 | && addr < (unsigned long)__sched_text_end); | |
7521 | } | |
7522 | ||
a9957449 | 7523 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7524 | { |
7525 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7526 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7527 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7528 | cfs_rq->rq = rq; | |
7529 | #endif | |
67e9fb2a | 7530 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7531 | } |
7532 | ||
fa85ae24 PZ |
7533 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7534 | { | |
7535 | struct rt_prio_array *array; | |
7536 | int i; | |
7537 | ||
7538 | array = &rt_rq->active; | |
7539 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7540 | INIT_LIST_HEAD(array->queue + i); | |
7541 | __clear_bit(i, array->bitmap); | |
7542 | } | |
7543 | /* delimiter for bitsearch: */ | |
7544 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7545 | ||
052f1dc7 | 7546 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7547 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7548 | #ifdef CONFIG_SMP |
e864c499 | 7549 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7550 | #endif |
48d5e258 | 7551 | #endif |
fa85ae24 PZ |
7552 | #ifdef CONFIG_SMP |
7553 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7554 | rt_rq->overloaded = 0; |
05fa785c | 7555 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7556 | #endif |
7557 | ||
7558 | rt_rq->rt_time = 0; | |
7559 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7560 | rt_rq->rt_runtime = 0; |
0986b11b | 7561 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7562 | |
052f1dc7 | 7563 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7564 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7565 | rt_rq->rq = rq; |
7566 | #endif | |
fa85ae24 PZ |
7567 | } |
7568 | ||
6f505b16 | 7569 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7570 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7571 | struct sched_entity *se, int cpu, int add, | |
7572 | struct sched_entity *parent) | |
6f505b16 | 7573 | { |
ec7dc8ac | 7574 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7575 | tg->cfs_rq[cpu] = cfs_rq; |
7576 | init_cfs_rq(cfs_rq, rq); | |
7577 | cfs_rq->tg = tg; | |
7578 | if (add) | |
7579 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7580 | ||
7581 | tg->se[cpu] = se; | |
354d60c2 DG |
7582 | /* se could be NULL for init_task_group */ |
7583 | if (!se) | |
7584 | return; | |
7585 | ||
ec7dc8ac DG |
7586 | if (!parent) |
7587 | se->cfs_rq = &rq->cfs; | |
7588 | else | |
7589 | se->cfs_rq = parent->my_q; | |
7590 | ||
6f505b16 PZ |
7591 | se->my_q = cfs_rq; |
7592 | se->load.weight = tg->shares; | |
e05510d0 | 7593 | se->load.inv_weight = 0; |
ec7dc8ac | 7594 | se->parent = parent; |
6f505b16 | 7595 | } |
052f1dc7 | 7596 | #endif |
6f505b16 | 7597 | |
052f1dc7 | 7598 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7599 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7600 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7601 | struct sched_rt_entity *parent) | |
6f505b16 | 7602 | { |
ec7dc8ac DG |
7603 | struct rq *rq = cpu_rq(cpu); |
7604 | ||
6f505b16 PZ |
7605 | tg->rt_rq[cpu] = rt_rq; |
7606 | init_rt_rq(rt_rq, rq); | |
7607 | rt_rq->tg = tg; | |
7608 | rt_rq->rt_se = rt_se; | |
ac086bc2 | 7609 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7610 | if (add) |
7611 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7612 | ||
7613 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7614 | if (!rt_se) |
7615 | return; | |
7616 | ||
ec7dc8ac DG |
7617 | if (!parent) |
7618 | rt_se->rt_rq = &rq->rt; | |
7619 | else | |
7620 | rt_se->rt_rq = parent->my_q; | |
7621 | ||
6f505b16 | 7622 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7623 | rt_se->parent = parent; |
6f505b16 PZ |
7624 | INIT_LIST_HEAD(&rt_se->run_list); |
7625 | } | |
7626 | #endif | |
7627 | ||
1da177e4 LT |
7628 | void __init sched_init(void) |
7629 | { | |
dd41f596 | 7630 | int i, j; |
434d53b0 MT |
7631 | unsigned long alloc_size = 0, ptr; |
7632 | ||
7633 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7634 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7635 | #endif | |
7636 | #ifdef CONFIG_RT_GROUP_SCHED | |
7637 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7638 | #endif |
df7c8e84 | 7639 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7640 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7641 | #endif |
434d53b0 | 7642 | if (alloc_size) { |
36b7b6d4 | 7643 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7644 | |
7645 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7646 | init_task_group.se = (struct sched_entity **)ptr; | |
7647 | ptr += nr_cpu_ids * sizeof(void **); | |
7648 | ||
7649 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7650 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7651 | |
6d6bc0ad | 7652 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7653 | #ifdef CONFIG_RT_GROUP_SCHED |
7654 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7655 | ptr += nr_cpu_ids * sizeof(void **); | |
7656 | ||
7657 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7658 | ptr += nr_cpu_ids * sizeof(void **); |
7659 | ||
6d6bc0ad | 7660 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7661 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7662 | for_each_possible_cpu(i) { | |
7663 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7664 | ptr += cpumask_size(); | |
7665 | } | |
7666 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7667 | } |
dd41f596 | 7668 | |
57d885fe GH |
7669 | #ifdef CONFIG_SMP |
7670 | init_defrootdomain(); | |
7671 | #endif | |
7672 | ||
d0b27fa7 PZ |
7673 | init_rt_bandwidth(&def_rt_bandwidth, |
7674 | global_rt_period(), global_rt_runtime()); | |
7675 | ||
7676 | #ifdef CONFIG_RT_GROUP_SCHED | |
7677 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7678 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7679 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7680 | |
7c941438 | 7681 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7682 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7683 | INIT_LIST_HEAD(&init_task_group.children); |
7684 | ||
7c941438 | 7685 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7686 | |
4a6cc4bd JK |
7687 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7688 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7689 | __alignof__(unsigned long)); | |
7690 | #endif | |
0a945022 | 7691 | for_each_possible_cpu(i) { |
70b97a7f | 7692 | struct rq *rq; |
1da177e4 LT |
7693 | |
7694 | rq = cpu_rq(i); | |
05fa785c | 7695 | raw_spin_lock_init(&rq->lock); |
7897986b | 7696 | rq->nr_running = 0; |
dce48a84 TG |
7697 | rq->calc_load_active = 0; |
7698 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7699 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7700 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7701 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7702 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7703 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7704 | #ifdef CONFIG_CGROUP_SCHED |
7705 | /* | |
7706 | * How much cpu bandwidth does init_task_group get? | |
7707 | * | |
7708 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7709 | * gets 100% of the cpu resources in the system. This overall | |
7710 | * system cpu resource is divided among the tasks of | |
7711 | * init_task_group and its child task-groups in a fair manner, | |
7712 | * based on each entity's (task or task-group's) weight | |
7713 | * (se->load.weight). | |
7714 | * | |
7715 | * In other words, if init_task_group has 10 tasks of weight | |
7716 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7717 | * then A0's share of the cpu resource is: | |
7718 | * | |
0d905bca | 7719 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7720 | * |
7721 | * We achieve this by letting init_task_group's tasks sit | |
7722 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7723 | */ | |
ec7dc8ac | 7724 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7725 | #endif |
354d60c2 DG |
7726 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7727 | ||
7728 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7729 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7730 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7731 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7732 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7733 | #endif |
dd41f596 | 7734 | #endif |
1da177e4 | 7735 | |
dd41f596 IM |
7736 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7737 | rq->cpu_load[j] = 0; | |
1da177e4 | 7738 | #ifdef CONFIG_SMP |
41c7ce9a | 7739 | rq->sd = NULL; |
57d885fe | 7740 | rq->rd = NULL; |
3f029d3c | 7741 | rq->post_schedule = 0; |
1da177e4 | 7742 | rq->active_balance = 0; |
dd41f596 | 7743 | rq->next_balance = jiffies; |
1da177e4 | 7744 | rq->push_cpu = 0; |
0a2966b4 | 7745 | rq->cpu = i; |
1f11eb6a | 7746 | rq->online = 0; |
1da177e4 | 7747 | rq->migration_thread = NULL; |
eae0c9df MG |
7748 | rq->idle_stamp = 0; |
7749 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
1da177e4 | 7750 | INIT_LIST_HEAD(&rq->migration_queue); |
dc938520 | 7751 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7752 | #endif |
8f4d37ec | 7753 | init_rq_hrtick(rq); |
1da177e4 | 7754 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7755 | } |
7756 | ||
2dd73a4f | 7757 | set_load_weight(&init_task); |
b50f60ce | 7758 | |
e107be36 AK |
7759 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7760 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7761 | #endif | |
7762 | ||
c9819f45 | 7763 | #ifdef CONFIG_SMP |
962cf36c | 7764 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7765 | #endif |
7766 | ||
b50f60ce | 7767 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7768 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7769 | #endif |
7770 | ||
1da177e4 LT |
7771 | /* |
7772 | * The boot idle thread does lazy MMU switching as well: | |
7773 | */ | |
7774 | atomic_inc(&init_mm.mm_count); | |
7775 | enter_lazy_tlb(&init_mm, current); | |
7776 | ||
7777 | /* | |
7778 | * Make us the idle thread. Technically, schedule() should not be | |
7779 | * called from this thread, however somewhere below it might be, | |
7780 | * but because we are the idle thread, we just pick up running again | |
7781 | * when this runqueue becomes "idle". | |
7782 | */ | |
7783 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7784 | |
7785 | calc_load_update = jiffies + LOAD_FREQ; | |
7786 | ||
dd41f596 IM |
7787 | /* |
7788 | * During early bootup we pretend to be a normal task: | |
7789 | */ | |
7790 | current->sched_class = &fair_sched_class; | |
6892b75e | 7791 | |
6a7b3dc3 | 7792 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7793 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7794 | #ifdef CONFIG_SMP |
7d1e6a9b | 7795 | #ifdef CONFIG_NO_HZ |
49557e62 | 7796 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 7797 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 7798 | #endif |
bdddd296 RR |
7799 | /* May be allocated at isolcpus cmdline parse time */ |
7800 | if (cpu_isolated_map == NULL) | |
7801 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7802 | #endif /* SMP */ |
6a7b3dc3 | 7803 | |
cdd6c482 | 7804 | perf_event_init(); |
0d905bca | 7805 | |
6892b75e | 7806 | scheduler_running = 1; |
1da177e4 LT |
7807 | } |
7808 | ||
7809 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7810 | static inline int preempt_count_equals(int preempt_offset) |
7811 | { | |
234da7bc | 7812 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7813 | |
7814 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7815 | } | |
7816 | ||
d894837f | 7817 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7818 | { |
48f24c4d | 7819 | #ifdef in_atomic |
1da177e4 LT |
7820 | static unsigned long prev_jiffy; /* ratelimiting */ |
7821 | ||
e4aafea2 FW |
7822 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7823 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7824 | return; |
7825 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7826 | return; | |
7827 | prev_jiffy = jiffies; | |
7828 | ||
3df0fc5b PZ |
7829 | printk(KERN_ERR |
7830 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7831 | file, line); | |
7832 | printk(KERN_ERR | |
7833 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7834 | in_atomic(), irqs_disabled(), | |
7835 | current->pid, current->comm); | |
aef745fc IM |
7836 | |
7837 | debug_show_held_locks(current); | |
7838 | if (irqs_disabled()) | |
7839 | print_irqtrace_events(current); | |
7840 | dump_stack(); | |
1da177e4 LT |
7841 | #endif |
7842 | } | |
7843 | EXPORT_SYMBOL(__might_sleep); | |
7844 | #endif | |
7845 | ||
7846 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7847 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7848 | { | |
7849 | int on_rq; | |
3e51f33f | 7850 | |
3a5e4dc1 AK |
7851 | update_rq_clock(rq); |
7852 | on_rq = p->se.on_rq; | |
7853 | if (on_rq) | |
7854 | deactivate_task(rq, p, 0); | |
7855 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7856 | if (on_rq) { | |
7857 | activate_task(rq, p, 0); | |
7858 | resched_task(rq->curr); | |
7859 | } | |
7860 | } | |
7861 | ||
1da177e4 LT |
7862 | void normalize_rt_tasks(void) |
7863 | { | |
a0f98a1c | 7864 | struct task_struct *g, *p; |
1da177e4 | 7865 | unsigned long flags; |
70b97a7f | 7866 | struct rq *rq; |
1da177e4 | 7867 | |
4cf5d77a | 7868 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7869 | do_each_thread(g, p) { |
178be793 IM |
7870 | /* |
7871 | * Only normalize user tasks: | |
7872 | */ | |
7873 | if (!p->mm) | |
7874 | continue; | |
7875 | ||
6cfb0d5d | 7876 | p->se.exec_start = 0; |
6cfb0d5d | 7877 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7878 | p->se.wait_start = 0; |
dd41f596 | 7879 | p->se.sleep_start = 0; |
dd41f596 | 7880 | p->se.block_start = 0; |
6cfb0d5d | 7881 | #endif |
dd41f596 IM |
7882 | |
7883 | if (!rt_task(p)) { | |
7884 | /* | |
7885 | * Renice negative nice level userspace | |
7886 | * tasks back to 0: | |
7887 | */ | |
7888 | if (TASK_NICE(p) < 0 && p->mm) | |
7889 | set_user_nice(p, 0); | |
1da177e4 | 7890 | continue; |
dd41f596 | 7891 | } |
1da177e4 | 7892 | |
1d615482 | 7893 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7894 | rq = __task_rq_lock(p); |
1da177e4 | 7895 | |
178be793 | 7896 | normalize_task(rq, p); |
3a5e4dc1 | 7897 | |
b29739f9 | 7898 | __task_rq_unlock(rq); |
1d615482 | 7899 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7900 | } while_each_thread(g, p); |
7901 | ||
4cf5d77a | 7902 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7903 | } |
7904 | ||
7905 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7906 | |
7907 | #ifdef CONFIG_IA64 | |
7908 | /* | |
7909 | * These functions are only useful for the IA64 MCA handling. | |
7910 | * | |
7911 | * They can only be called when the whole system has been | |
7912 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7913 | * activity can take place. Using them for anything else would | |
7914 | * be a serious bug, and as a result, they aren't even visible | |
7915 | * under any other configuration. | |
7916 | */ | |
7917 | ||
7918 | /** | |
7919 | * curr_task - return the current task for a given cpu. | |
7920 | * @cpu: the processor in question. | |
7921 | * | |
7922 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7923 | */ | |
36c8b586 | 7924 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7925 | { |
7926 | return cpu_curr(cpu); | |
7927 | } | |
7928 | ||
7929 | /** | |
7930 | * set_curr_task - set the current task for a given cpu. | |
7931 | * @cpu: the processor in question. | |
7932 | * @p: the task pointer to set. | |
7933 | * | |
7934 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7935 | * are serviced on a separate stack. It allows the architecture to switch the |
7936 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7937 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7938 | * and caller must save the original value of the current task (see | |
7939 | * curr_task() above) and restore that value before reenabling interrupts and | |
7940 | * re-starting the system. | |
7941 | * | |
7942 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7943 | */ | |
36c8b586 | 7944 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7945 | { |
7946 | cpu_curr(cpu) = p; | |
7947 | } | |
7948 | ||
7949 | #endif | |
29f59db3 | 7950 | |
bccbe08a PZ |
7951 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7952 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7953 | { |
7954 | int i; | |
7955 | ||
7956 | for_each_possible_cpu(i) { | |
7957 | if (tg->cfs_rq) | |
7958 | kfree(tg->cfs_rq[i]); | |
7959 | if (tg->se) | |
7960 | kfree(tg->se[i]); | |
6f505b16 PZ |
7961 | } |
7962 | ||
7963 | kfree(tg->cfs_rq); | |
7964 | kfree(tg->se); | |
6f505b16 PZ |
7965 | } |
7966 | ||
ec7dc8ac DG |
7967 | static |
7968 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 7969 | { |
29f59db3 | 7970 | struct cfs_rq *cfs_rq; |
eab17229 | 7971 | struct sched_entity *se; |
9b5b7751 | 7972 | struct rq *rq; |
29f59db3 SV |
7973 | int i; |
7974 | ||
434d53b0 | 7975 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7976 | if (!tg->cfs_rq) |
7977 | goto err; | |
434d53b0 | 7978 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7979 | if (!tg->se) |
7980 | goto err; | |
052f1dc7 PZ |
7981 | |
7982 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7983 | |
7984 | for_each_possible_cpu(i) { | |
9b5b7751 | 7985 | rq = cpu_rq(i); |
29f59db3 | 7986 | |
eab17229 LZ |
7987 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
7988 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
7989 | if (!cfs_rq) |
7990 | goto err; | |
7991 | ||
eab17229 LZ |
7992 | se = kzalloc_node(sizeof(struct sched_entity), |
7993 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 7994 | if (!se) |
dfc12eb2 | 7995 | goto err_free_rq; |
29f59db3 | 7996 | |
eab17229 | 7997 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
7998 | } |
7999 | ||
8000 | return 1; | |
8001 | ||
dfc12eb2 PC |
8002 | err_free_rq: |
8003 | kfree(cfs_rq); | |
bccbe08a PZ |
8004 | err: |
8005 | return 0; | |
8006 | } | |
8007 | ||
8008 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8009 | { | |
8010 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8011 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8012 | } | |
8013 | ||
8014 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8015 | { | |
8016 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8017 | } | |
6d6bc0ad | 8018 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8019 | static inline void free_fair_sched_group(struct task_group *tg) |
8020 | { | |
8021 | } | |
8022 | ||
ec7dc8ac DG |
8023 | static inline |
8024 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8025 | { |
8026 | return 1; | |
8027 | } | |
8028 | ||
8029 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8030 | { | |
8031 | } | |
8032 | ||
8033 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8034 | { | |
8035 | } | |
6d6bc0ad | 8036 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8037 | |
8038 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8039 | static void free_rt_sched_group(struct task_group *tg) |
8040 | { | |
8041 | int i; | |
8042 | ||
d0b27fa7 PZ |
8043 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8044 | ||
bccbe08a PZ |
8045 | for_each_possible_cpu(i) { |
8046 | if (tg->rt_rq) | |
8047 | kfree(tg->rt_rq[i]); | |
8048 | if (tg->rt_se) | |
8049 | kfree(tg->rt_se[i]); | |
8050 | } | |
8051 | ||
8052 | kfree(tg->rt_rq); | |
8053 | kfree(tg->rt_se); | |
8054 | } | |
8055 | ||
ec7dc8ac DG |
8056 | static |
8057 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8058 | { |
8059 | struct rt_rq *rt_rq; | |
eab17229 | 8060 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8061 | struct rq *rq; |
8062 | int i; | |
8063 | ||
434d53b0 | 8064 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8065 | if (!tg->rt_rq) |
8066 | goto err; | |
434d53b0 | 8067 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8068 | if (!tg->rt_se) |
8069 | goto err; | |
8070 | ||
d0b27fa7 PZ |
8071 | init_rt_bandwidth(&tg->rt_bandwidth, |
8072 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8073 | |
8074 | for_each_possible_cpu(i) { | |
8075 | rq = cpu_rq(i); | |
8076 | ||
eab17229 LZ |
8077 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8078 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8079 | if (!rt_rq) |
8080 | goto err; | |
29f59db3 | 8081 | |
eab17229 LZ |
8082 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8083 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8084 | if (!rt_se) |
dfc12eb2 | 8085 | goto err_free_rq; |
29f59db3 | 8086 | |
eab17229 | 8087 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8088 | } |
8089 | ||
bccbe08a PZ |
8090 | return 1; |
8091 | ||
dfc12eb2 PC |
8092 | err_free_rq: |
8093 | kfree(rt_rq); | |
bccbe08a PZ |
8094 | err: |
8095 | return 0; | |
8096 | } | |
8097 | ||
8098 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8099 | { | |
8100 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8101 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8102 | } | |
8103 | ||
8104 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8105 | { | |
8106 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8107 | } | |
6d6bc0ad | 8108 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8109 | static inline void free_rt_sched_group(struct task_group *tg) |
8110 | { | |
8111 | } | |
8112 | ||
ec7dc8ac DG |
8113 | static inline |
8114 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8115 | { |
8116 | return 1; | |
8117 | } | |
8118 | ||
8119 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8120 | { | |
8121 | } | |
8122 | ||
8123 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8124 | { | |
8125 | } | |
6d6bc0ad | 8126 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8127 | |
7c941438 | 8128 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8129 | static void free_sched_group(struct task_group *tg) |
8130 | { | |
8131 | free_fair_sched_group(tg); | |
8132 | free_rt_sched_group(tg); | |
8133 | kfree(tg); | |
8134 | } | |
8135 | ||
8136 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8137 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8138 | { |
8139 | struct task_group *tg; | |
8140 | unsigned long flags; | |
8141 | int i; | |
8142 | ||
8143 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8144 | if (!tg) | |
8145 | return ERR_PTR(-ENOMEM); | |
8146 | ||
ec7dc8ac | 8147 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8148 | goto err; |
8149 | ||
ec7dc8ac | 8150 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8151 | goto err; |
8152 | ||
8ed36996 | 8153 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8154 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8155 | register_fair_sched_group(tg, i); |
8156 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8157 | } |
6f505b16 | 8158 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8159 | |
8160 | WARN_ON(!parent); /* root should already exist */ | |
8161 | ||
8162 | tg->parent = parent; | |
f473aa5e | 8163 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8164 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8165 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8166 | |
9b5b7751 | 8167 | return tg; |
29f59db3 SV |
8168 | |
8169 | err: | |
6f505b16 | 8170 | free_sched_group(tg); |
29f59db3 SV |
8171 | return ERR_PTR(-ENOMEM); |
8172 | } | |
8173 | ||
9b5b7751 | 8174 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8175 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8176 | { |
29f59db3 | 8177 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8178 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8179 | } |
8180 | ||
9b5b7751 | 8181 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8182 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8183 | { |
8ed36996 | 8184 | unsigned long flags; |
9b5b7751 | 8185 | int i; |
29f59db3 | 8186 | |
8ed36996 | 8187 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8188 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8189 | unregister_fair_sched_group(tg, i); |
8190 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8191 | } |
6f505b16 | 8192 | list_del_rcu(&tg->list); |
f473aa5e | 8193 | list_del_rcu(&tg->siblings); |
8ed36996 | 8194 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8195 | |
9b5b7751 | 8196 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8197 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8198 | } |
8199 | ||
9b5b7751 | 8200 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8201 | * The caller of this function should have put the task in its new group |
8202 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8203 | * reflect its new group. | |
9b5b7751 SV |
8204 | */ |
8205 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8206 | { |
8207 | int on_rq, running; | |
8208 | unsigned long flags; | |
8209 | struct rq *rq; | |
8210 | ||
8211 | rq = task_rq_lock(tsk, &flags); | |
8212 | ||
29f59db3 SV |
8213 | update_rq_clock(rq); |
8214 | ||
051a1d1a | 8215 | running = task_current(rq, tsk); |
29f59db3 SV |
8216 | on_rq = tsk->se.on_rq; |
8217 | ||
0e1f3483 | 8218 | if (on_rq) |
29f59db3 | 8219 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8220 | if (unlikely(running)) |
8221 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8222 | |
6f505b16 | 8223 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8224 | |
810b3817 PZ |
8225 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8226 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8227 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8228 | #endif |
8229 | ||
0e1f3483 HS |
8230 | if (unlikely(running)) |
8231 | tsk->sched_class->set_curr_task(rq); | |
8232 | if (on_rq) | |
7074badb | 8233 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8234 | |
29f59db3 SV |
8235 | task_rq_unlock(rq, &flags); |
8236 | } | |
7c941438 | 8237 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8238 | |
052f1dc7 | 8239 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8240 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8241 | { |
8242 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8243 | int on_rq; |
8244 | ||
29f59db3 | 8245 | on_rq = se->on_rq; |
62fb1851 | 8246 | if (on_rq) |
29f59db3 SV |
8247 | dequeue_entity(cfs_rq, se, 0); |
8248 | ||
8249 | se->load.weight = shares; | |
e05510d0 | 8250 | se->load.inv_weight = 0; |
29f59db3 | 8251 | |
62fb1851 | 8252 | if (on_rq) |
29f59db3 | 8253 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8254 | } |
62fb1851 | 8255 | |
c09595f6 PZ |
8256 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8257 | { | |
8258 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8259 | struct rq *rq = cfs_rq->rq; | |
8260 | unsigned long flags; | |
8261 | ||
05fa785c | 8262 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8263 | __set_se_shares(se, shares); |
05fa785c | 8264 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8265 | } |
8266 | ||
8ed36996 PZ |
8267 | static DEFINE_MUTEX(shares_mutex); |
8268 | ||
4cf86d77 | 8269 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8270 | { |
8271 | int i; | |
8ed36996 | 8272 | unsigned long flags; |
c61935fd | 8273 | |
ec7dc8ac DG |
8274 | /* |
8275 | * We can't change the weight of the root cgroup. | |
8276 | */ | |
8277 | if (!tg->se[0]) | |
8278 | return -EINVAL; | |
8279 | ||
18d95a28 PZ |
8280 | if (shares < MIN_SHARES) |
8281 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8282 | else if (shares > MAX_SHARES) |
8283 | shares = MAX_SHARES; | |
62fb1851 | 8284 | |
8ed36996 | 8285 | mutex_lock(&shares_mutex); |
9b5b7751 | 8286 | if (tg->shares == shares) |
5cb350ba | 8287 | goto done; |
29f59db3 | 8288 | |
8ed36996 | 8289 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8290 | for_each_possible_cpu(i) |
8291 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8292 | list_del_rcu(&tg->siblings); |
8ed36996 | 8293 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8294 | |
8295 | /* wait for any ongoing reference to this group to finish */ | |
8296 | synchronize_sched(); | |
8297 | ||
8298 | /* | |
8299 | * Now we are free to modify the group's share on each cpu | |
8300 | * w/o tripping rebalance_share or load_balance_fair. | |
8301 | */ | |
9b5b7751 | 8302 | tg->shares = shares; |
c09595f6 PZ |
8303 | for_each_possible_cpu(i) { |
8304 | /* | |
8305 | * force a rebalance | |
8306 | */ | |
8307 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8308 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8309 | } |
29f59db3 | 8310 | |
6b2d7700 SV |
8311 | /* |
8312 | * Enable load balance activity on this group, by inserting it back on | |
8313 | * each cpu's rq->leaf_cfs_rq_list. | |
8314 | */ | |
8ed36996 | 8315 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8316 | for_each_possible_cpu(i) |
8317 | register_fair_sched_group(tg, i); | |
f473aa5e | 8318 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8319 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8320 | done: |
8ed36996 | 8321 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8322 | return 0; |
29f59db3 SV |
8323 | } |
8324 | ||
5cb350ba DG |
8325 | unsigned long sched_group_shares(struct task_group *tg) |
8326 | { | |
8327 | return tg->shares; | |
8328 | } | |
052f1dc7 | 8329 | #endif |
5cb350ba | 8330 | |
052f1dc7 | 8331 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8332 | /* |
9f0c1e56 | 8333 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8334 | */ |
9f0c1e56 PZ |
8335 | static DEFINE_MUTEX(rt_constraints_mutex); |
8336 | ||
8337 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8338 | { | |
8339 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8340 | return 1ULL << 20; |
9f0c1e56 | 8341 | |
9a7e0b18 | 8342 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8343 | } |
8344 | ||
9a7e0b18 PZ |
8345 | /* Must be called with tasklist_lock held */ |
8346 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8347 | { |
9a7e0b18 | 8348 | struct task_struct *g, *p; |
b40b2e8e | 8349 | |
9a7e0b18 PZ |
8350 | do_each_thread(g, p) { |
8351 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8352 | return 1; | |
8353 | } while_each_thread(g, p); | |
b40b2e8e | 8354 | |
9a7e0b18 PZ |
8355 | return 0; |
8356 | } | |
b40b2e8e | 8357 | |
9a7e0b18 PZ |
8358 | struct rt_schedulable_data { |
8359 | struct task_group *tg; | |
8360 | u64 rt_period; | |
8361 | u64 rt_runtime; | |
8362 | }; | |
b40b2e8e | 8363 | |
9a7e0b18 PZ |
8364 | static int tg_schedulable(struct task_group *tg, void *data) |
8365 | { | |
8366 | struct rt_schedulable_data *d = data; | |
8367 | struct task_group *child; | |
8368 | unsigned long total, sum = 0; | |
8369 | u64 period, runtime; | |
b40b2e8e | 8370 | |
9a7e0b18 PZ |
8371 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8372 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8373 | |
9a7e0b18 PZ |
8374 | if (tg == d->tg) { |
8375 | period = d->rt_period; | |
8376 | runtime = d->rt_runtime; | |
b40b2e8e | 8377 | } |
b40b2e8e | 8378 | |
4653f803 PZ |
8379 | /* |
8380 | * Cannot have more runtime than the period. | |
8381 | */ | |
8382 | if (runtime > period && runtime != RUNTIME_INF) | |
8383 | return -EINVAL; | |
6f505b16 | 8384 | |
4653f803 PZ |
8385 | /* |
8386 | * Ensure we don't starve existing RT tasks. | |
8387 | */ | |
9a7e0b18 PZ |
8388 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8389 | return -EBUSY; | |
6f505b16 | 8390 | |
9a7e0b18 | 8391 | total = to_ratio(period, runtime); |
6f505b16 | 8392 | |
4653f803 PZ |
8393 | /* |
8394 | * Nobody can have more than the global setting allows. | |
8395 | */ | |
8396 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8397 | return -EINVAL; | |
6f505b16 | 8398 | |
4653f803 PZ |
8399 | /* |
8400 | * The sum of our children's runtime should not exceed our own. | |
8401 | */ | |
9a7e0b18 PZ |
8402 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8403 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8404 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8405 | |
9a7e0b18 PZ |
8406 | if (child == d->tg) { |
8407 | period = d->rt_period; | |
8408 | runtime = d->rt_runtime; | |
8409 | } | |
6f505b16 | 8410 | |
9a7e0b18 | 8411 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8412 | } |
6f505b16 | 8413 | |
9a7e0b18 PZ |
8414 | if (sum > total) |
8415 | return -EINVAL; | |
8416 | ||
8417 | return 0; | |
6f505b16 PZ |
8418 | } |
8419 | ||
9a7e0b18 | 8420 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8421 | { |
9a7e0b18 PZ |
8422 | struct rt_schedulable_data data = { |
8423 | .tg = tg, | |
8424 | .rt_period = period, | |
8425 | .rt_runtime = runtime, | |
8426 | }; | |
8427 | ||
8428 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8429 | } |
8430 | ||
d0b27fa7 PZ |
8431 | static int tg_set_bandwidth(struct task_group *tg, |
8432 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8433 | { |
ac086bc2 | 8434 | int i, err = 0; |
9f0c1e56 | 8435 | |
9f0c1e56 | 8436 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8437 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8438 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8439 | if (err) | |
9f0c1e56 | 8440 | goto unlock; |
ac086bc2 | 8441 | |
0986b11b | 8442 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8443 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8444 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8445 | |
8446 | for_each_possible_cpu(i) { | |
8447 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8448 | ||
0986b11b | 8449 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8450 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8451 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8452 | } |
0986b11b | 8453 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8454 | unlock: |
521f1a24 | 8455 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8456 | mutex_unlock(&rt_constraints_mutex); |
8457 | ||
8458 | return err; | |
6f505b16 PZ |
8459 | } |
8460 | ||
d0b27fa7 PZ |
8461 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8462 | { | |
8463 | u64 rt_runtime, rt_period; | |
8464 | ||
8465 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8466 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8467 | if (rt_runtime_us < 0) | |
8468 | rt_runtime = RUNTIME_INF; | |
8469 | ||
8470 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8471 | } | |
8472 | ||
9f0c1e56 PZ |
8473 | long sched_group_rt_runtime(struct task_group *tg) |
8474 | { | |
8475 | u64 rt_runtime_us; | |
8476 | ||
d0b27fa7 | 8477 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8478 | return -1; |
8479 | ||
d0b27fa7 | 8480 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8481 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8482 | return rt_runtime_us; | |
8483 | } | |
d0b27fa7 PZ |
8484 | |
8485 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8486 | { | |
8487 | u64 rt_runtime, rt_period; | |
8488 | ||
8489 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8490 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8491 | ||
619b0488 R |
8492 | if (rt_period == 0) |
8493 | return -EINVAL; | |
8494 | ||
d0b27fa7 PZ |
8495 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8496 | } | |
8497 | ||
8498 | long sched_group_rt_period(struct task_group *tg) | |
8499 | { | |
8500 | u64 rt_period_us; | |
8501 | ||
8502 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8503 | do_div(rt_period_us, NSEC_PER_USEC); | |
8504 | return rt_period_us; | |
8505 | } | |
8506 | ||
8507 | static int sched_rt_global_constraints(void) | |
8508 | { | |
4653f803 | 8509 | u64 runtime, period; |
d0b27fa7 PZ |
8510 | int ret = 0; |
8511 | ||
ec5d4989 HS |
8512 | if (sysctl_sched_rt_period <= 0) |
8513 | return -EINVAL; | |
8514 | ||
4653f803 PZ |
8515 | runtime = global_rt_runtime(); |
8516 | period = global_rt_period(); | |
8517 | ||
8518 | /* | |
8519 | * Sanity check on the sysctl variables. | |
8520 | */ | |
8521 | if (runtime > period && runtime != RUNTIME_INF) | |
8522 | return -EINVAL; | |
10b612f4 | 8523 | |
d0b27fa7 | 8524 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8525 | read_lock(&tasklist_lock); |
4653f803 | 8526 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8527 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8528 | mutex_unlock(&rt_constraints_mutex); |
8529 | ||
8530 | return ret; | |
8531 | } | |
54e99124 DG |
8532 | |
8533 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8534 | { | |
8535 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8536 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8537 | return 0; | |
8538 | ||
8539 | return 1; | |
8540 | } | |
8541 | ||
6d6bc0ad | 8542 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8543 | static int sched_rt_global_constraints(void) |
8544 | { | |
ac086bc2 PZ |
8545 | unsigned long flags; |
8546 | int i; | |
8547 | ||
ec5d4989 HS |
8548 | if (sysctl_sched_rt_period <= 0) |
8549 | return -EINVAL; | |
8550 | ||
60aa605d PZ |
8551 | /* |
8552 | * There's always some RT tasks in the root group | |
8553 | * -- migration, kstopmachine etc.. | |
8554 | */ | |
8555 | if (sysctl_sched_rt_runtime == 0) | |
8556 | return -EBUSY; | |
8557 | ||
0986b11b | 8558 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8559 | for_each_possible_cpu(i) { |
8560 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8561 | ||
0986b11b | 8562 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8563 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8564 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8565 | } |
0986b11b | 8566 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8567 | |
d0b27fa7 PZ |
8568 | return 0; |
8569 | } | |
6d6bc0ad | 8570 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8571 | |
8572 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8573 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8574 | loff_t *ppos) |
8575 | { | |
8576 | int ret; | |
8577 | int old_period, old_runtime; | |
8578 | static DEFINE_MUTEX(mutex); | |
8579 | ||
8580 | mutex_lock(&mutex); | |
8581 | old_period = sysctl_sched_rt_period; | |
8582 | old_runtime = sysctl_sched_rt_runtime; | |
8583 | ||
8d65af78 | 8584 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8585 | |
8586 | if (!ret && write) { | |
8587 | ret = sched_rt_global_constraints(); | |
8588 | if (ret) { | |
8589 | sysctl_sched_rt_period = old_period; | |
8590 | sysctl_sched_rt_runtime = old_runtime; | |
8591 | } else { | |
8592 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8593 | def_rt_bandwidth.rt_period = | |
8594 | ns_to_ktime(global_rt_period()); | |
8595 | } | |
8596 | } | |
8597 | mutex_unlock(&mutex); | |
8598 | ||
8599 | return ret; | |
8600 | } | |
68318b8e | 8601 | |
052f1dc7 | 8602 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8603 | |
8604 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8605 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8606 | { |
2b01dfe3 PM |
8607 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8608 | struct task_group, css); | |
68318b8e SV |
8609 | } |
8610 | ||
8611 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8612 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8613 | { |
ec7dc8ac | 8614 | struct task_group *tg, *parent; |
68318b8e | 8615 | |
2b01dfe3 | 8616 | if (!cgrp->parent) { |
68318b8e | 8617 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8618 | return &init_task_group.css; |
8619 | } | |
8620 | ||
ec7dc8ac DG |
8621 | parent = cgroup_tg(cgrp->parent); |
8622 | tg = sched_create_group(parent); | |
68318b8e SV |
8623 | if (IS_ERR(tg)) |
8624 | return ERR_PTR(-ENOMEM); | |
8625 | ||
68318b8e SV |
8626 | return &tg->css; |
8627 | } | |
8628 | ||
41a2d6cf IM |
8629 | static void |
8630 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8631 | { |
2b01dfe3 | 8632 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8633 | |
8634 | sched_destroy_group(tg); | |
8635 | } | |
8636 | ||
41a2d6cf | 8637 | static int |
be367d09 | 8638 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8639 | { |
b68aa230 | 8640 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8641 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8642 | return -EINVAL; |
8643 | #else | |
68318b8e SV |
8644 | /* We don't support RT-tasks being in separate groups */ |
8645 | if (tsk->sched_class != &fair_sched_class) | |
8646 | return -EINVAL; | |
b68aa230 | 8647 | #endif |
be367d09 BB |
8648 | return 0; |
8649 | } | |
68318b8e | 8650 | |
be367d09 BB |
8651 | static int |
8652 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8653 | struct task_struct *tsk, bool threadgroup) | |
8654 | { | |
8655 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8656 | if (retval) | |
8657 | return retval; | |
8658 | if (threadgroup) { | |
8659 | struct task_struct *c; | |
8660 | rcu_read_lock(); | |
8661 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8662 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8663 | if (retval) { | |
8664 | rcu_read_unlock(); | |
8665 | return retval; | |
8666 | } | |
8667 | } | |
8668 | rcu_read_unlock(); | |
8669 | } | |
68318b8e SV |
8670 | return 0; |
8671 | } | |
8672 | ||
8673 | static void | |
2b01dfe3 | 8674 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8675 | struct cgroup *old_cont, struct task_struct *tsk, |
8676 | bool threadgroup) | |
68318b8e SV |
8677 | { |
8678 | sched_move_task(tsk); | |
be367d09 BB |
8679 | if (threadgroup) { |
8680 | struct task_struct *c; | |
8681 | rcu_read_lock(); | |
8682 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8683 | sched_move_task(c); | |
8684 | } | |
8685 | rcu_read_unlock(); | |
8686 | } | |
68318b8e SV |
8687 | } |
8688 | ||
052f1dc7 | 8689 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8690 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8691 | u64 shareval) |
68318b8e | 8692 | { |
2b01dfe3 | 8693 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8694 | } |
8695 | ||
f4c753b7 | 8696 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8697 | { |
2b01dfe3 | 8698 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8699 | |
8700 | return (u64) tg->shares; | |
8701 | } | |
6d6bc0ad | 8702 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8703 | |
052f1dc7 | 8704 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8705 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8706 | s64 val) |
6f505b16 | 8707 | { |
06ecb27c | 8708 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8709 | } |
8710 | ||
06ecb27c | 8711 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8712 | { |
06ecb27c | 8713 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8714 | } |
d0b27fa7 PZ |
8715 | |
8716 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8717 | u64 rt_period_us) | |
8718 | { | |
8719 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8720 | } | |
8721 | ||
8722 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8723 | { | |
8724 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8725 | } | |
6d6bc0ad | 8726 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8727 | |
fe5c7cc2 | 8728 | static struct cftype cpu_files[] = { |
052f1dc7 | 8729 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8730 | { |
8731 | .name = "shares", | |
f4c753b7 PM |
8732 | .read_u64 = cpu_shares_read_u64, |
8733 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8734 | }, |
052f1dc7 PZ |
8735 | #endif |
8736 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8737 | { |
9f0c1e56 | 8738 | .name = "rt_runtime_us", |
06ecb27c PM |
8739 | .read_s64 = cpu_rt_runtime_read, |
8740 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8741 | }, |
d0b27fa7 PZ |
8742 | { |
8743 | .name = "rt_period_us", | |
f4c753b7 PM |
8744 | .read_u64 = cpu_rt_period_read_uint, |
8745 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8746 | }, |
052f1dc7 | 8747 | #endif |
68318b8e SV |
8748 | }; |
8749 | ||
8750 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8751 | { | |
fe5c7cc2 | 8752 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8753 | } |
8754 | ||
8755 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8756 | .name = "cpu", |
8757 | .create = cpu_cgroup_create, | |
8758 | .destroy = cpu_cgroup_destroy, | |
8759 | .can_attach = cpu_cgroup_can_attach, | |
8760 | .attach = cpu_cgroup_attach, | |
8761 | .populate = cpu_cgroup_populate, | |
8762 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8763 | .early_init = 1, |
8764 | }; | |
8765 | ||
052f1dc7 | 8766 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8767 | |
8768 | #ifdef CONFIG_CGROUP_CPUACCT | |
8769 | ||
8770 | /* | |
8771 | * CPU accounting code for task groups. | |
8772 | * | |
8773 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8774 | * (balbir@in.ibm.com). | |
8775 | */ | |
8776 | ||
934352f2 | 8777 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8778 | struct cpuacct { |
8779 | struct cgroup_subsys_state css; | |
8780 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
8781 | u64 *cpuusage; | |
ef12fefa | 8782 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8783 | struct cpuacct *parent; |
d842de87 SV |
8784 | }; |
8785 | ||
8786 | struct cgroup_subsys cpuacct_subsys; | |
8787 | ||
8788 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8789 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8790 | { |
32cd756a | 8791 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8792 | struct cpuacct, css); |
8793 | } | |
8794 | ||
8795 | /* return cpu accounting group to which this task belongs */ | |
8796 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8797 | { | |
8798 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8799 | struct cpuacct, css); | |
8800 | } | |
8801 | ||
8802 | /* create a new cpu accounting group */ | |
8803 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8804 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8805 | { |
8806 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8807 | int i; |
d842de87 SV |
8808 | |
8809 | if (!ca) | |
ef12fefa | 8810 | goto out; |
d842de87 SV |
8811 | |
8812 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8813 | if (!ca->cpuusage) |
8814 | goto out_free_ca; | |
8815 | ||
8816 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8817 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8818 | goto out_free_counters; | |
d842de87 | 8819 | |
934352f2 BR |
8820 | if (cgrp->parent) |
8821 | ca->parent = cgroup_ca(cgrp->parent); | |
8822 | ||
d842de87 | 8823 | return &ca->css; |
ef12fefa BR |
8824 | |
8825 | out_free_counters: | |
8826 | while (--i >= 0) | |
8827 | percpu_counter_destroy(&ca->cpustat[i]); | |
8828 | free_percpu(ca->cpuusage); | |
8829 | out_free_ca: | |
8830 | kfree(ca); | |
8831 | out: | |
8832 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8833 | } |
8834 | ||
8835 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8836 | static void |
32cd756a | 8837 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8838 | { |
32cd756a | 8839 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8840 | int i; |
d842de87 | 8841 | |
ef12fefa BR |
8842 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8843 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8844 | free_percpu(ca->cpuusage); |
8845 | kfree(ca); | |
8846 | } | |
8847 | ||
720f5498 KC |
8848 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8849 | { | |
b36128c8 | 8850 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8851 | u64 data; |
8852 | ||
8853 | #ifndef CONFIG_64BIT | |
8854 | /* | |
8855 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8856 | */ | |
05fa785c | 8857 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8858 | data = *cpuusage; |
05fa785c | 8859 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8860 | #else |
8861 | data = *cpuusage; | |
8862 | #endif | |
8863 | ||
8864 | return data; | |
8865 | } | |
8866 | ||
8867 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8868 | { | |
b36128c8 | 8869 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8870 | |
8871 | #ifndef CONFIG_64BIT | |
8872 | /* | |
8873 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8874 | */ | |
05fa785c | 8875 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8876 | *cpuusage = val; |
05fa785c | 8877 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8878 | #else |
8879 | *cpuusage = val; | |
8880 | #endif | |
8881 | } | |
8882 | ||
d842de87 | 8883 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8884 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8885 | { |
32cd756a | 8886 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8887 | u64 totalcpuusage = 0; |
8888 | int i; | |
8889 | ||
720f5498 KC |
8890 | for_each_present_cpu(i) |
8891 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8892 | |
8893 | return totalcpuusage; | |
8894 | } | |
8895 | ||
0297b803 DG |
8896 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8897 | u64 reset) | |
8898 | { | |
8899 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8900 | int err = 0; | |
8901 | int i; | |
8902 | ||
8903 | if (reset) { | |
8904 | err = -EINVAL; | |
8905 | goto out; | |
8906 | } | |
8907 | ||
720f5498 KC |
8908 | for_each_present_cpu(i) |
8909 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8910 | |
0297b803 DG |
8911 | out: |
8912 | return err; | |
8913 | } | |
8914 | ||
e9515c3c KC |
8915 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8916 | struct seq_file *m) | |
8917 | { | |
8918 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8919 | u64 percpu; | |
8920 | int i; | |
8921 | ||
8922 | for_each_present_cpu(i) { | |
8923 | percpu = cpuacct_cpuusage_read(ca, i); | |
8924 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8925 | } | |
8926 | seq_printf(m, "\n"); | |
8927 | return 0; | |
8928 | } | |
8929 | ||
ef12fefa BR |
8930 | static const char *cpuacct_stat_desc[] = { |
8931 | [CPUACCT_STAT_USER] = "user", | |
8932 | [CPUACCT_STAT_SYSTEM] = "system", | |
8933 | }; | |
8934 | ||
8935 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8936 | struct cgroup_map_cb *cb) | |
8937 | { | |
8938 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8939 | int i; | |
8940 | ||
8941 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8942 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8943 | val = cputime64_to_clock_t(val); | |
8944 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8945 | } | |
8946 | return 0; | |
8947 | } | |
8948 | ||
d842de87 SV |
8949 | static struct cftype files[] = { |
8950 | { | |
8951 | .name = "usage", | |
f4c753b7 PM |
8952 | .read_u64 = cpuusage_read, |
8953 | .write_u64 = cpuusage_write, | |
d842de87 | 8954 | }, |
e9515c3c KC |
8955 | { |
8956 | .name = "usage_percpu", | |
8957 | .read_seq_string = cpuacct_percpu_seq_read, | |
8958 | }, | |
ef12fefa BR |
8959 | { |
8960 | .name = "stat", | |
8961 | .read_map = cpuacct_stats_show, | |
8962 | }, | |
d842de87 SV |
8963 | }; |
8964 | ||
32cd756a | 8965 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8966 | { |
32cd756a | 8967 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8968 | } |
8969 | ||
8970 | /* | |
8971 | * charge this task's execution time to its accounting group. | |
8972 | * | |
8973 | * called with rq->lock held. | |
8974 | */ | |
8975 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8976 | { | |
8977 | struct cpuacct *ca; | |
934352f2 | 8978 | int cpu; |
d842de87 | 8979 | |
c40c6f85 | 8980 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8981 | return; |
8982 | ||
934352f2 | 8983 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8984 | |
8985 | rcu_read_lock(); | |
8986 | ||
d842de87 | 8987 | ca = task_ca(tsk); |
d842de87 | 8988 | |
934352f2 | 8989 | for (; ca; ca = ca->parent) { |
b36128c8 | 8990 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8991 | *cpuusage += cputime; |
8992 | } | |
a18b83b7 BR |
8993 | |
8994 | rcu_read_unlock(); | |
d842de87 SV |
8995 | } |
8996 | ||
ef12fefa BR |
8997 | /* |
8998 | * Charge the system/user time to the task's accounting group. | |
8999 | */ | |
9000 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9001 | enum cpuacct_stat_index idx, cputime_t val) | |
9002 | { | |
9003 | struct cpuacct *ca; | |
9004 | ||
9005 | if (unlikely(!cpuacct_subsys.active)) | |
9006 | return; | |
9007 | ||
9008 | rcu_read_lock(); | |
9009 | ca = task_ca(tsk); | |
9010 | ||
9011 | do { | |
9012 | percpu_counter_add(&ca->cpustat[idx], val); | |
9013 | ca = ca->parent; | |
9014 | } while (ca); | |
9015 | rcu_read_unlock(); | |
9016 | } | |
9017 | ||
d842de87 SV |
9018 | struct cgroup_subsys cpuacct_subsys = { |
9019 | .name = "cpuacct", | |
9020 | .create = cpuacct_create, | |
9021 | .destroy = cpuacct_destroy, | |
9022 | .populate = cpuacct_populate, | |
9023 | .subsys_id = cpuacct_subsys_id, | |
9024 | }; | |
9025 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
9026 | |
9027 | #ifndef CONFIG_SMP | |
9028 | ||
9029 | int rcu_expedited_torture_stats(char *page) | |
9030 | { | |
9031 | return 0; | |
9032 | } | |
9033 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9034 | ||
9035 | void synchronize_sched_expedited(void) | |
9036 | { | |
9037 | } | |
9038 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9039 | ||
9040 | #else /* #ifndef CONFIG_SMP */ | |
9041 | ||
9042 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
9043 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
9044 | ||
9045 | #define RCU_EXPEDITED_STATE_POST -2 | |
9046 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
9047 | ||
9048 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
9049 | ||
9050 | int rcu_expedited_torture_stats(char *page) | |
9051 | { | |
9052 | int cnt = 0; | |
9053 | int cpu; | |
9054 | ||
9055 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
9056 | for_each_online_cpu(cpu) { | |
9057 | cnt += sprintf(&page[cnt], " %d:%d", | |
9058 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
9059 | } | |
9060 | cnt += sprintf(&page[cnt], "\n"); | |
9061 | return cnt; | |
9062 | } | |
9063 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9064 | ||
9065 | static long synchronize_sched_expedited_count; | |
9066 | ||
9067 | /* | |
9068 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9069 | * approach to force grace period to end quickly. This consumes | |
9070 | * significant time on all CPUs, and is thus not recommended for | |
9071 | * any sort of common-case code. | |
9072 | * | |
9073 | * Note that it is illegal to call this function while holding any | |
9074 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9075 | * observe this restriction will result in deadlock. | |
9076 | */ | |
9077 | void synchronize_sched_expedited(void) | |
9078 | { | |
9079 | int cpu; | |
9080 | unsigned long flags; | |
9081 | bool need_full_sync = 0; | |
9082 | struct rq *rq; | |
9083 | struct migration_req *req; | |
9084 | long snap; | |
9085 | int trycount = 0; | |
9086 | ||
9087 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
9088 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
9089 | get_online_cpus(); | |
9090 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
9091 | put_online_cpus(); | |
9092 | if (trycount++ < 10) | |
9093 | udelay(trycount * num_online_cpus()); | |
9094 | else { | |
9095 | synchronize_sched(); | |
9096 | return; | |
9097 | } | |
9098 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
9099 | smp_mb(); /* ensure test happens before caller kfree */ | |
9100 | return; | |
9101 | } | |
9102 | get_online_cpus(); | |
9103 | } | |
9104 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
9105 | for_each_online_cpu(cpu) { | |
9106 | rq = cpu_rq(cpu); | |
9107 | req = &per_cpu(rcu_migration_req, cpu); | |
9108 | init_completion(&req->done); | |
9109 | req->task = NULL; | |
9110 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
05fa785c | 9111 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf | 9112 | list_add(&req->list, &rq->migration_queue); |
05fa785c | 9113 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9114 | wake_up_process(rq->migration_thread); |
9115 | } | |
9116 | for_each_online_cpu(cpu) { | |
9117 | rcu_expedited_state = cpu; | |
9118 | req = &per_cpu(rcu_migration_req, cpu); | |
9119 | rq = cpu_rq(cpu); | |
9120 | wait_for_completion(&req->done); | |
05fa785c | 9121 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf PM |
9122 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) |
9123 | need_full_sync = 1; | |
9124 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
05fa785c | 9125 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9126 | } |
9127 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
956539b7 | 9128 | synchronize_sched_expedited_count++; |
03b042bf PM |
9129 | mutex_unlock(&rcu_sched_expedited_mutex); |
9130 | put_online_cpus(); | |
9131 | if (need_full_sync) | |
9132 | synchronize_sched(); | |
9133 | } | |
9134 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9135 | ||
9136 | #endif /* #else #ifndef CONFIG_SMP */ |