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; | |
6f505b16 | 439 | #endif |
6aa645ea IM |
440 | }; |
441 | ||
57d885fe GH |
442 | #ifdef CONFIG_SMP |
443 | ||
444 | /* | |
445 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
446 | * variables. Each exclusive cpuset essentially defines an island domain by |
447 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
448 | * exclusive cpuset is created, we also create and attach a new root-domain |
449 | * object. | |
450 | * | |
57d885fe GH |
451 | */ |
452 | struct root_domain { | |
453 | atomic_t refcount; | |
c6c4927b RR |
454 | cpumask_var_t span; |
455 | cpumask_var_t online; | |
637f5085 | 456 | |
0eab9146 | 457 | /* |
637f5085 GH |
458 | * The "RT overload" flag: it gets set if a CPU has more than |
459 | * one runnable RT task. | |
460 | */ | |
c6c4927b | 461 | cpumask_var_t rto_mask; |
0eab9146 | 462 | atomic_t rto_count; |
6e0534f2 GH |
463 | #ifdef CONFIG_SMP |
464 | struct cpupri cpupri; | |
465 | #endif | |
57d885fe GH |
466 | }; |
467 | ||
dc938520 GH |
468 | /* |
469 | * By default the system creates a single root-domain with all cpus as | |
470 | * members (mimicking the global state we have today). | |
471 | */ | |
57d885fe GH |
472 | static struct root_domain def_root_domain; |
473 | ||
474 | #endif | |
475 | ||
1da177e4 LT |
476 | /* |
477 | * This is the main, per-CPU runqueue data structure. | |
478 | * | |
479 | * Locking rule: those places that want to lock multiple runqueues | |
480 | * (such as the load balancing or the thread migration code), lock | |
481 | * acquire operations must be ordered by ascending &runqueue. | |
482 | */ | |
70b97a7f | 483 | struct rq { |
d8016491 | 484 | /* runqueue lock: */ |
05fa785c | 485 | raw_spinlock_t lock; |
1da177e4 LT |
486 | |
487 | /* | |
488 | * nr_running and cpu_load should be in the same cacheline because | |
489 | * remote CPUs use both these fields when doing load calculation. | |
490 | */ | |
491 | unsigned long nr_running; | |
6aa645ea IM |
492 | #define CPU_LOAD_IDX_MAX 5 |
493 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 494 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 495 | u64 nohz_stamp; |
46cb4b7c SS |
496 | unsigned char in_nohz_recently; |
497 | #endif | |
a64692a3 MG |
498 | unsigned int skip_clock_update; |
499 | ||
d8016491 IM |
500 | /* capture load from *all* tasks on this cpu: */ |
501 | struct load_weight load; | |
6aa645ea IM |
502 | unsigned long nr_load_updates; |
503 | u64 nr_switches; | |
504 | ||
505 | struct cfs_rq cfs; | |
6f505b16 | 506 | struct rt_rq rt; |
6f505b16 | 507 | |
6aa645ea | 508 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
509 | /* list of leaf cfs_rq on this cpu: */ |
510 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
511 | #endif |
512 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 513 | struct list_head leaf_rt_rq_list; |
1da177e4 | 514 | #endif |
1da177e4 LT |
515 | |
516 | /* | |
517 | * This is part of a global counter where only the total sum | |
518 | * over all CPUs matters. A task can increase this counter on | |
519 | * one CPU and if it got migrated afterwards it may decrease | |
520 | * it on another CPU. Always updated under the runqueue lock: | |
521 | */ | |
522 | unsigned long nr_uninterruptible; | |
523 | ||
36c8b586 | 524 | struct task_struct *curr, *idle; |
c9819f45 | 525 | unsigned long next_balance; |
1da177e4 | 526 | struct mm_struct *prev_mm; |
6aa645ea | 527 | |
3e51f33f | 528 | u64 clock; |
6aa645ea | 529 | |
1da177e4 LT |
530 | atomic_t nr_iowait; |
531 | ||
532 | #ifdef CONFIG_SMP | |
0eab9146 | 533 | struct root_domain *rd; |
1da177e4 LT |
534 | struct sched_domain *sd; |
535 | ||
a0a522ce | 536 | unsigned char idle_at_tick; |
1da177e4 | 537 | /* For active balancing */ |
3f029d3c | 538 | int post_schedule; |
1da177e4 LT |
539 | int active_balance; |
540 | int push_cpu; | |
d8016491 IM |
541 | /* cpu of this runqueue: */ |
542 | int cpu; | |
1f11eb6a | 543 | int online; |
1da177e4 | 544 | |
a8a51d5e | 545 | unsigned long avg_load_per_task; |
1da177e4 | 546 | |
36c8b586 | 547 | struct task_struct *migration_thread; |
1da177e4 | 548 | struct list_head migration_queue; |
e9e9250b PZ |
549 | |
550 | u64 rt_avg; | |
551 | u64 age_stamp; | |
1b9508f6 MG |
552 | u64 idle_stamp; |
553 | u64 avg_idle; | |
1da177e4 LT |
554 | #endif |
555 | ||
dce48a84 TG |
556 | /* calc_load related fields */ |
557 | unsigned long calc_load_update; | |
558 | long calc_load_active; | |
559 | ||
8f4d37ec | 560 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
561 | #ifdef CONFIG_SMP |
562 | int hrtick_csd_pending; | |
563 | struct call_single_data hrtick_csd; | |
564 | #endif | |
8f4d37ec PZ |
565 | struct hrtimer hrtick_timer; |
566 | #endif | |
567 | ||
1da177e4 LT |
568 | #ifdef CONFIG_SCHEDSTATS |
569 | /* latency stats */ | |
570 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
571 | unsigned long long rq_cpu_time; |
572 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
573 | |
574 | /* sys_sched_yield() stats */ | |
480b9434 | 575 | unsigned int yld_count; |
1da177e4 LT |
576 | |
577 | /* schedule() stats */ | |
480b9434 KC |
578 | unsigned int sched_switch; |
579 | unsigned int sched_count; | |
580 | unsigned int sched_goidle; | |
1da177e4 LT |
581 | |
582 | /* try_to_wake_up() stats */ | |
480b9434 KC |
583 | unsigned int ttwu_count; |
584 | unsigned int ttwu_local; | |
b8efb561 IM |
585 | |
586 | /* BKL stats */ | |
480b9434 | 587 | unsigned int bkl_count; |
1da177e4 LT |
588 | #endif |
589 | }; | |
590 | ||
f34e3b61 | 591 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 592 | |
7d478721 PZ |
593 | static inline |
594 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 595 | { |
7d478721 | 596 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
a64692a3 MG |
597 | |
598 | /* | |
599 | * A queue event has occurred, and we're going to schedule. In | |
600 | * this case, we can save a useless back to back clock update. | |
601 | */ | |
602 | if (test_tsk_need_resched(p)) | |
603 | rq->skip_clock_update = 1; | |
dd41f596 IM |
604 | } |
605 | ||
0a2966b4 CL |
606 | static inline int cpu_of(struct rq *rq) |
607 | { | |
608 | #ifdef CONFIG_SMP | |
609 | return rq->cpu; | |
610 | #else | |
611 | return 0; | |
612 | #endif | |
613 | } | |
614 | ||
497f0ab3 | 615 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
616 | rcu_dereference_check((p), \ |
617 | rcu_read_lock_sched_held() || \ | |
618 | lockdep_is_held(&sched_domains_mutex)) | |
619 | ||
674311d5 NP |
620 | /* |
621 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 622 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
623 | * |
624 | * The domain tree of any CPU may only be accessed from within | |
625 | * preempt-disabled sections. | |
626 | */ | |
48f24c4d | 627 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 628 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
629 | |
630 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
631 | #define this_rq() (&__get_cpu_var(runqueues)) | |
632 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
633 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 634 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 635 | |
aa9c4c0f | 636 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 637 | { |
a64692a3 MG |
638 | if (!rq->skip_clock_update) |
639 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
3e51f33f PZ |
640 | } |
641 | ||
bf5c91ba IM |
642 | /* |
643 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
644 | */ | |
645 | #ifdef CONFIG_SCHED_DEBUG | |
646 | # define const_debug __read_mostly | |
647 | #else | |
648 | # define const_debug static const | |
649 | #endif | |
650 | ||
017730c1 IM |
651 | /** |
652 | * runqueue_is_locked | |
e17b38bf | 653 | * @cpu: the processor in question. |
017730c1 IM |
654 | * |
655 | * Returns true if the current cpu runqueue is locked. | |
656 | * This interface allows printk to be called with the runqueue lock | |
657 | * held and know whether or not it is OK to wake up the klogd. | |
658 | */ | |
89f19f04 | 659 | int runqueue_is_locked(int cpu) |
017730c1 | 660 | { |
05fa785c | 661 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
662 | } |
663 | ||
bf5c91ba IM |
664 | /* |
665 | * Debugging: various feature bits | |
666 | */ | |
f00b45c1 PZ |
667 | |
668 | #define SCHED_FEAT(name, enabled) \ | |
669 | __SCHED_FEAT_##name , | |
670 | ||
bf5c91ba | 671 | enum { |
f00b45c1 | 672 | #include "sched_features.h" |
bf5c91ba IM |
673 | }; |
674 | ||
f00b45c1 PZ |
675 | #undef SCHED_FEAT |
676 | ||
677 | #define SCHED_FEAT(name, enabled) \ | |
678 | (1UL << __SCHED_FEAT_##name) * enabled | | |
679 | ||
bf5c91ba | 680 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
681 | #include "sched_features.h" |
682 | 0; | |
683 | ||
684 | #undef SCHED_FEAT | |
685 | ||
686 | #ifdef CONFIG_SCHED_DEBUG | |
687 | #define SCHED_FEAT(name, enabled) \ | |
688 | #name , | |
689 | ||
983ed7a6 | 690 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
691 | #include "sched_features.h" |
692 | NULL | |
693 | }; | |
694 | ||
695 | #undef SCHED_FEAT | |
696 | ||
34f3a814 | 697 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 698 | { |
f00b45c1 PZ |
699 | int i; |
700 | ||
701 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
702 | if (!(sysctl_sched_features & (1UL << i))) |
703 | seq_puts(m, "NO_"); | |
704 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 705 | } |
34f3a814 | 706 | seq_puts(m, "\n"); |
f00b45c1 | 707 | |
34f3a814 | 708 | return 0; |
f00b45c1 PZ |
709 | } |
710 | ||
711 | static ssize_t | |
712 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
713 | size_t cnt, loff_t *ppos) | |
714 | { | |
715 | char buf[64]; | |
716 | char *cmp = buf; | |
717 | int neg = 0; | |
718 | int i; | |
719 | ||
720 | if (cnt > 63) | |
721 | cnt = 63; | |
722 | ||
723 | if (copy_from_user(&buf, ubuf, cnt)) | |
724 | return -EFAULT; | |
725 | ||
726 | buf[cnt] = 0; | |
727 | ||
c24b7c52 | 728 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
729 | neg = 1; |
730 | cmp += 3; | |
731 | } | |
732 | ||
733 | for (i = 0; sched_feat_names[i]; i++) { | |
734 | int len = strlen(sched_feat_names[i]); | |
735 | ||
736 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
737 | if (neg) | |
738 | sysctl_sched_features &= ~(1UL << i); | |
739 | else | |
740 | sysctl_sched_features |= (1UL << i); | |
741 | break; | |
742 | } | |
743 | } | |
744 | ||
745 | if (!sched_feat_names[i]) | |
746 | return -EINVAL; | |
747 | ||
42994724 | 748 | *ppos += cnt; |
f00b45c1 PZ |
749 | |
750 | return cnt; | |
751 | } | |
752 | ||
34f3a814 LZ |
753 | static int sched_feat_open(struct inode *inode, struct file *filp) |
754 | { | |
755 | return single_open(filp, sched_feat_show, NULL); | |
756 | } | |
757 | ||
828c0950 | 758 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
759 | .open = sched_feat_open, |
760 | .write = sched_feat_write, | |
761 | .read = seq_read, | |
762 | .llseek = seq_lseek, | |
763 | .release = single_release, | |
f00b45c1 PZ |
764 | }; |
765 | ||
766 | static __init int sched_init_debug(void) | |
767 | { | |
f00b45c1 PZ |
768 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
769 | &sched_feat_fops); | |
770 | ||
771 | return 0; | |
772 | } | |
773 | late_initcall(sched_init_debug); | |
774 | ||
775 | #endif | |
776 | ||
777 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 778 | |
b82d9fdd PZ |
779 | /* |
780 | * Number of tasks to iterate in a single balance run. | |
781 | * Limited because this is done with IRQs disabled. | |
782 | */ | |
783 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
784 | ||
2398f2c6 PZ |
785 | /* |
786 | * ratelimit for updating the group shares. | |
55cd5340 | 787 | * default: 0.25ms |
2398f2c6 | 788 | */ |
55cd5340 | 789 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 790 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 791 | |
ffda12a1 PZ |
792 | /* |
793 | * Inject some fuzzyness into changing the per-cpu group shares | |
794 | * this avoids remote rq-locks at the expense of fairness. | |
795 | * default: 4 | |
796 | */ | |
797 | unsigned int sysctl_sched_shares_thresh = 4; | |
798 | ||
e9e9250b PZ |
799 | /* |
800 | * period over which we average the RT time consumption, measured | |
801 | * in ms. | |
802 | * | |
803 | * default: 1s | |
804 | */ | |
805 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
806 | ||
fa85ae24 | 807 | /* |
9f0c1e56 | 808 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
809 | * default: 1s |
810 | */ | |
9f0c1e56 | 811 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 812 | |
6892b75e IM |
813 | static __read_mostly int scheduler_running; |
814 | ||
9f0c1e56 PZ |
815 | /* |
816 | * part of the period that we allow rt tasks to run in us. | |
817 | * default: 0.95s | |
818 | */ | |
819 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 820 | |
d0b27fa7 PZ |
821 | static inline u64 global_rt_period(void) |
822 | { | |
823 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
824 | } | |
825 | ||
826 | static inline u64 global_rt_runtime(void) | |
827 | { | |
e26873bb | 828 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
829 | return RUNTIME_INF; |
830 | ||
831 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
832 | } | |
fa85ae24 | 833 | |
1da177e4 | 834 | #ifndef prepare_arch_switch |
4866cde0 NP |
835 | # define prepare_arch_switch(next) do { } while (0) |
836 | #endif | |
837 | #ifndef finish_arch_switch | |
838 | # define finish_arch_switch(prev) do { } while (0) | |
839 | #endif | |
840 | ||
051a1d1a DA |
841 | static inline int task_current(struct rq *rq, struct task_struct *p) |
842 | { | |
843 | return rq->curr == p; | |
844 | } | |
845 | ||
4866cde0 | 846 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 847 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 848 | { |
051a1d1a | 849 | return task_current(rq, p); |
4866cde0 NP |
850 | } |
851 | ||
70b97a7f | 852 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
853 | { |
854 | } | |
855 | ||
70b97a7f | 856 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 857 | { |
da04c035 IM |
858 | #ifdef CONFIG_DEBUG_SPINLOCK |
859 | /* this is a valid case when another task releases the spinlock */ | |
860 | rq->lock.owner = current; | |
861 | #endif | |
8a25d5de IM |
862 | /* |
863 | * If we are tracking spinlock dependencies then we have to | |
864 | * fix up the runqueue lock - which gets 'carried over' from | |
865 | * prev into current: | |
866 | */ | |
867 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
868 | ||
05fa785c | 869 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
870 | } |
871 | ||
872 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 873 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
874 | { |
875 | #ifdef CONFIG_SMP | |
876 | return p->oncpu; | |
877 | #else | |
051a1d1a | 878 | return task_current(rq, p); |
4866cde0 NP |
879 | #endif |
880 | } | |
881 | ||
70b97a7f | 882 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
883 | { |
884 | #ifdef CONFIG_SMP | |
885 | /* | |
886 | * We can optimise this out completely for !SMP, because the | |
887 | * SMP rebalancing from interrupt is the only thing that cares | |
888 | * here. | |
889 | */ | |
890 | next->oncpu = 1; | |
891 | #endif | |
892 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 893 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 894 | #else |
05fa785c | 895 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
896 | #endif |
897 | } | |
898 | ||
70b97a7f | 899 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
900 | { |
901 | #ifdef CONFIG_SMP | |
902 | /* | |
903 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
904 | * We must ensure this doesn't happen until the switch is completely | |
905 | * finished. | |
906 | */ | |
907 | smp_wmb(); | |
908 | prev->oncpu = 0; | |
909 | #endif | |
910 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
911 | local_irq_enable(); | |
1da177e4 | 912 | #endif |
4866cde0 NP |
913 | } |
914 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 915 | |
0970d299 PZ |
916 | /* |
917 | * Check whether the task is waking, we use this to synchronize against | |
918 | * ttwu() so that task_cpu() reports a stable number. | |
919 | * | |
920 | * We need to make an exception for PF_STARTING tasks because the fork | |
921 | * path might require task_rq_lock() to work, eg. it can call | |
922 | * set_cpus_allowed_ptr() from the cpuset clone_ns code. | |
923 | */ | |
924 | static inline int task_is_waking(struct task_struct *p) | |
925 | { | |
926 | return unlikely((p->state == TASK_WAKING) && !(p->flags & PF_STARTING)); | |
927 | } | |
928 | ||
b29739f9 IM |
929 | /* |
930 | * __task_rq_lock - lock the runqueue a given task resides on. | |
931 | * Must be called interrupts disabled. | |
932 | */ | |
70b97a7f | 933 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
934 | __acquires(rq->lock) |
935 | { | |
0970d299 PZ |
936 | struct rq *rq; |
937 | ||
3a5c359a | 938 | for (;;) { |
0970d299 PZ |
939 | while (task_is_waking(p)) |
940 | cpu_relax(); | |
941 | rq = task_rq(p); | |
05fa785c | 942 | raw_spin_lock(&rq->lock); |
0970d299 | 943 | if (likely(rq == task_rq(p) && !task_is_waking(p))) |
3a5c359a | 944 | return rq; |
05fa785c | 945 | raw_spin_unlock(&rq->lock); |
b29739f9 | 946 | } |
b29739f9 IM |
947 | } |
948 | ||
1da177e4 LT |
949 | /* |
950 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 951 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
952 | * explicitly disabling preemption. |
953 | */ | |
70b97a7f | 954 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
955 | __acquires(rq->lock) |
956 | { | |
70b97a7f | 957 | struct rq *rq; |
1da177e4 | 958 | |
3a5c359a | 959 | for (;;) { |
0970d299 PZ |
960 | while (task_is_waking(p)) |
961 | cpu_relax(); | |
3a5c359a AK |
962 | local_irq_save(*flags); |
963 | rq = task_rq(p); | |
05fa785c | 964 | raw_spin_lock(&rq->lock); |
0970d299 | 965 | if (likely(rq == task_rq(p) && !task_is_waking(p))) |
3a5c359a | 966 | return rq; |
05fa785c | 967 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 968 | } |
1da177e4 LT |
969 | } |
970 | ||
ad474cac ON |
971 | void task_rq_unlock_wait(struct task_struct *p) |
972 | { | |
973 | struct rq *rq = task_rq(p); | |
974 | ||
975 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
05fa785c | 976 | raw_spin_unlock_wait(&rq->lock); |
ad474cac ON |
977 | } |
978 | ||
a9957449 | 979 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
980 | __releases(rq->lock) |
981 | { | |
05fa785c | 982 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
983 | } |
984 | ||
70b97a7f | 985 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
986 | __releases(rq->lock) |
987 | { | |
05fa785c | 988 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
989 | } |
990 | ||
1da177e4 | 991 | /* |
cc2a73b5 | 992 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 993 | */ |
a9957449 | 994 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
995 | __acquires(rq->lock) |
996 | { | |
70b97a7f | 997 | struct rq *rq; |
1da177e4 LT |
998 | |
999 | local_irq_disable(); | |
1000 | rq = this_rq(); | |
05fa785c | 1001 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
1002 | |
1003 | return rq; | |
1004 | } | |
1005 | ||
8f4d37ec PZ |
1006 | #ifdef CONFIG_SCHED_HRTICK |
1007 | /* | |
1008 | * Use HR-timers to deliver accurate preemption points. | |
1009 | * | |
1010 | * Its all a bit involved since we cannot program an hrt while holding the | |
1011 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1012 | * reschedule event. | |
1013 | * | |
1014 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1015 | * rq->lock. | |
1016 | */ | |
8f4d37ec PZ |
1017 | |
1018 | /* | |
1019 | * Use hrtick when: | |
1020 | * - enabled by features | |
1021 | * - hrtimer is actually high res | |
1022 | */ | |
1023 | static inline int hrtick_enabled(struct rq *rq) | |
1024 | { | |
1025 | if (!sched_feat(HRTICK)) | |
1026 | return 0; | |
ba42059f | 1027 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1028 | return 0; |
8f4d37ec PZ |
1029 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1030 | } | |
1031 | ||
8f4d37ec PZ |
1032 | static void hrtick_clear(struct rq *rq) |
1033 | { | |
1034 | if (hrtimer_active(&rq->hrtick_timer)) | |
1035 | hrtimer_cancel(&rq->hrtick_timer); | |
1036 | } | |
1037 | ||
8f4d37ec PZ |
1038 | /* |
1039 | * High-resolution timer tick. | |
1040 | * Runs from hardirq context with interrupts disabled. | |
1041 | */ | |
1042 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1043 | { | |
1044 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1045 | ||
1046 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1047 | ||
05fa785c | 1048 | raw_spin_lock(&rq->lock); |
3e51f33f | 1049 | update_rq_clock(rq); |
8f4d37ec | 1050 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1051 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1052 | |
1053 | return HRTIMER_NORESTART; | |
1054 | } | |
1055 | ||
95e904c7 | 1056 | #ifdef CONFIG_SMP |
31656519 PZ |
1057 | /* |
1058 | * called from hardirq (IPI) context | |
1059 | */ | |
1060 | static void __hrtick_start(void *arg) | |
b328ca18 | 1061 | { |
31656519 | 1062 | struct rq *rq = arg; |
b328ca18 | 1063 | |
05fa785c | 1064 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1065 | hrtimer_restart(&rq->hrtick_timer); |
1066 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1067 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1068 | } |
1069 | ||
31656519 PZ |
1070 | /* |
1071 | * Called to set the hrtick timer state. | |
1072 | * | |
1073 | * called with rq->lock held and irqs disabled | |
1074 | */ | |
1075 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1076 | { |
31656519 PZ |
1077 | struct hrtimer *timer = &rq->hrtick_timer; |
1078 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1079 | |
cc584b21 | 1080 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1081 | |
1082 | if (rq == this_rq()) { | |
1083 | hrtimer_restart(timer); | |
1084 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1085 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1086 | rq->hrtick_csd_pending = 1; |
1087 | } | |
b328ca18 PZ |
1088 | } |
1089 | ||
1090 | static int | |
1091 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1092 | { | |
1093 | int cpu = (int)(long)hcpu; | |
1094 | ||
1095 | switch (action) { | |
1096 | case CPU_UP_CANCELED: | |
1097 | case CPU_UP_CANCELED_FROZEN: | |
1098 | case CPU_DOWN_PREPARE: | |
1099 | case CPU_DOWN_PREPARE_FROZEN: | |
1100 | case CPU_DEAD: | |
1101 | case CPU_DEAD_FROZEN: | |
31656519 | 1102 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1103 | return NOTIFY_OK; |
1104 | } | |
1105 | ||
1106 | return NOTIFY_DONE; | |
1107 | } | |
1108 | ||
fa748203 | 1109 | static __init void init_hrtick(void) |
b328ca18 PZ |
1110 | { |
1111 | hotcpu_notifier(hotplug_hrtick, 0); | |
1112 | } | |
31656519 PZ |
1113 | #else |
1114 | /* | |
1115 | * Called to set the hrtick timer state. | |
1116 | * | |
1117 | * called with rq->lock held and irqs disabled | |
1118 | */ | |
1119 | static void hrtick_start(struct rq *rq, u64 delay) | |
1120 | { | |
7f1e2ca9 | 1121 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1122 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1123 | } |
b328ca18 | 1124 | |
006c75f1 | 1125 | static inline void init_hrtick(void) |
8f4d37ec | 1126 | { |
8f4d37ec | 1127 | } |
31656519 | 1128 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1129 | |
31656519 | 1130 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1131 | { |
31656519 PZ |
1132 | #ifdef CONFIG_SMP |
1133 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1134 | |
31656519 PZ |
1135 | rq->hrtick_csd.flags = 0; |
1136 | rq->hrtick_csd.func = __hrtick_start; | |
1137 | rq->hrtick_csd.info = rq; | |
1138 | #endif | |
8f4d37ec | 1139 | |
31656519 PZ |
1140 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1141 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1142 | } |
006c75f1 | 1143 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1144 | static inline void hrtick_clear(struct rq *rq) |
1145 | { | |
1146 | } | |
1147 | ||
8f4d37ec PZ |
1148 | static inline void init_rq_hrtick(struct rq *rq) |
1149 | { | |
1150 | } | |
1151 | ||
b328ca18 PZ |
1152 | static inline void init_hrtick(void) |
1153 | { | |
1154 | } | |
006c75f1 | 1155 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1156 | |
c24d20db IM |
1157 | /* |
1158 | * resched_task - mark a task 'to be rescheduled now'. | |
1159 | * | |
1160 | * On UP this means the setting of the need_resched flag, on SMP it | |
1161 | * might also involve a cross-CPU call to trigger the scheduler on | |
1162 | * the target CPU. | |
1163 | */ | |
1164 | #ifdef CONFIG_SMP | |
1165 | ||
1166 | #ifndef tsk_is_polling | |
1167 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1168 | #endif | |
1169 | ||
31656519 | 1170 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1171 | { |
1172 | int cpu; | |
1173 | ||
05fa785c | 1174 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1175 | |
5ed0cec0 | 1176 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1177 | return; |
1178 | ||
5ed0cec0 | 1179 | set_tsk_need_resched(p); |
c24d20db IM |
1180 | |
1181 | cpu = task_cpu(p); | |
1182 | if (cpu == smp_processor_id()) | |
1183 | return; | |
1184 | ||
1185 | /* NEED_RESCHED must be visible before we test polling */ | |
1186 | smp_mb(); | |
1187 | if (!tsk_is_polling(p)) | |
1188 | smp_send_reschedule(cpu); | |
1189 | } | |
1190 | ||
1191 | static void resched_cpu(int cpu) | |
1192 | { | |
1193 | struct rq *rq = cpu_rq(cpu); | |
1194 | unsigned long flags; | |
1195 | ||
05fa785c | 1196 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1197 | return; |
1198 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1199 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1200 | } |
06d8308c TG |
1201 | |
1202 | #ifdef CONFIG_NO_HZ | |
1203 | /* | |
1204 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1205 | * idle CPU then this timer might expire before the next timer event | |
1206 | * which is scheduled to wake up that CPU. In case of a completely | |
1207 | * idle system the next event might even be infinite time into the | |
1208 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1209 | * leaves the inner idle loop so the newly added timer is taken into | |
1210 | * account when the CPU goes back to idle and evaluates the timer | |
1211 | * wheel for the next timer event. | |
1212 | */ | |
1213 | void wake_up_idle_cpu(int cpu) | |
1214 | { | |
1215 | struct rq *rq = cpu_rq(cpu); | |
1216 | ||
1217 | if (cpu == smp_processor_id()) | |
1218 | return; | |
1219 | ||
1220 | /* | |
1221 | * This is safe, as this function is called with the timer | |
1222 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1223 | * to idle and has not yet set rq->curr to idle then it will | |
1224 | * be serialized on the timer wheel base lock and take the new | |
1225 | * timer into account automatically. | |
1226 | */ | |
1227 | if (rq->curr != rq->idle) | |
1228 | return; | |
1229 | ||
1230 | /* | |
1231 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1232 | * lockless. The worst case is that the other CPU runs the | |
1233 | * idle task through an additional NOOP schedule() | |
1234 | */ | |
5ed0cec0 | 1235 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1236 | |
1237 | /* NEED_RESCHED must be visible before we test polling */ | |
1238 | smp_mb(); | |
1239 | if (!tsk_is_polling(rq->idle)) | |
1240 | smp_send_reschedule(cpu); | |
1241 | } | |
39c0cbe2 MG |
1242 | |
1243 | int nohz_ratelimit(int cpu) | |
1244 | { | |
1245 | struct rq *rq = cpu_rq(cpu); | |
1246 | u64 diff = rq->clock - rq->nohz_stamp; | |
1247 | ||
1248 | rq->nohz_stamp = rq->clock; | |
1249 | ||
1250 | return diff < (NSEC_PER_SEC / HZ) >> 1; | |
1251 | } | |
1252 | ||
6d6bc0ad | 1253 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1254 | |
e9e9250b PZ |
1255 | static u64 sched_avg_period(void) |
1256 | { | |
1257 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1258 | } | |
1259 | ||
1260 | static void sched_avg_update(struct rq *rq) | |
1261 | { | |
1262 | s64 period = sched_avg_period(); | |
1263 | ||
1264 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1265 | rq->age_stamp += period; | |
1266 | rq->rt_avg /= 2; | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1271 | { | |
1272 | rq->rt_avg += rt_delta; | |
1273 | sched_avg_update(rq); | |
1274 | } | |
1275 | ||
6d6bc0ad | 1276 | #else /* !CONFIG_SMP */ |
31656519 | 1277 | static void resched_task(struct task_struct *p) |
c24d20db | 1278 | { |
05fa785c | 1279 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1280 | set_tsk_need_resched(p); |
c24d20db | 1281 | } |
e9e9250b PZ |
1282 | |
1283 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1284 | { | |
1285 | } | |
6d6bc0ad | 1286 | #endif /* CONFIG_SMP */ |
c24d20db | 1287 | |
45bf76df IM |
1288 | #if BITS_PER_LONG == 32 |
1289 | # define WMULT_CONST (~0UL) | |
1290 | #else | |
1291 | # define WMULT_CONST (1UL << 32) | |
1292 | #endif | |
1293 | ||
1294 | #define WMULT_SHIFT 32 | |
1295 | ||
194081eb IM |
1296 | /* |
1297 | * Shift right and round: | |
1298 | */ | |
cf2ab469 | 1299 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1300 | |
a7be37ac PZ |
1301 | /* |
1302 | * delta *= weight / lw | |
1303 | */ | |
cb1c4fc9 | 1304 | static unsigned long |
45bf76df IM |
1305 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1306 | struct load_weight *lw) | |
1307 | { | |
1308 | u64 tmp; | |
1309 | ||
7a232e03 LJ |
1310 | if (!lw->inv_weight) { |
1311 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1312 | lw->inv_weight = 1; | |
1313 | else | |
1314 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1315 | / (lw->weight+1); | |
1316 | } | |
45bf76df IM |
1317 | |
1318 | tmp = (u64)delta_exec * weight; | |
1319 | /* | |
1320 | * Check whether we'd overflow the 64-bit multiplication: | |
1321 | */ | |
194081eb | 1322 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1323 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1324 | WMULT_SHIFT/2); |
1325 | else | |
cf2ab469 | 1326 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1327 | |
ecf691da | 1328 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1329 | } |
1330 | ||
1091985b | 1331 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1332 | { |
1333 | lw->weight += inc; | |
e89996ae | 1334 | lw->inv_weight = 0; |
45bf76df IM |
1335 | } |
1336 | ||
1091985b | 1337 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1338 | { |
1339 | lw->weight -= dec; | |
e89996ae | 1340 | lw->inv_weight = 0; |
45bf76df IM |
1341 | } |
1342 | ||
2dd73a4f PW |
1343 | /* |
1344 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1345 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1346 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1347 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1348 | * scaled version of the new time slice allocation that they receive on time |
1349 | * slice expiry etc. | |
1350 | */ | |
1351 | ||
cce7ade8 PZ |
1352 | #define WEIGHT_IDLEPRIO 3 |
1353 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1354 | |
1355 | /* | |
1356 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1357 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1358 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1359 | * that remained on nice 0. | |
1360 | * | |
1361 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1362 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1363 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1364 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1365 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1366 | */ |
1367 | static const int prio_to_weight[40] = { | |
254753dc IM |
1368 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1369 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1370 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1371 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1372 | /* 0 */ 1024, 820, 655, 526, 423, | |
1373 | /* 5 */ 335, 272, 215, 172, 137, | |
1374 | /* 10 */ 110, 87, 70, 56, 45, | |
1375 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1376 | }; |
1377 | ||
5714d2de IM |
1378 | /* |
1379 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1380 | * | |
1381 | * In cases where the weight does not change often, we can use the | |
1382 | * precalculated inverse to speed up arithmetics by turning divisions | |
1383 | * into multiplications: | |
1384 | */ | |
dd41f596 | 1385 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1386 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1387 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1388 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1389 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1390 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1391 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1392 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1393 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1394 | }; |
2dd73a4f | 1395 | |
ef12fefa BR |
1396 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1397 | enum cpuacct_stat_index { | |
1398 | CPUACCT_STAT_USER, /* ... user mode */ | |
1399 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1400 | ||
1401 | CPUACCT_STAT_NSTATS, | |
1402 | }; | |
1403 | ||
d842de87 SV |
1404 | #ifdef CONFIG_CGROUP_CPUACCT |
1405 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1406 | static void cpuacct_update_stats(struct task_struct *tsk, |
1407 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1408 | #else |
1409 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1410 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1411 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1412 | #endif |
1413 | ||
18d95a28 PZ |
1414 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1415 | { | |
1416 | update_load_add(&rq->load, load); | |
1417 | } | |
1418 | ||
1419 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1420 | { | |
1421 | update_load_sub(&rq->load, load); | |
1422 | } | |
1423 | ||
7940ca36 | 1424 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1425 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1426 | |
1427 | /* | |
1428 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1429 | * leaving it for the final time. | |
1430 | */ | |
eb755805 | 1431 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1432 | { |
1433 | struct task_group *parent, *child; | |
eb755805 | 1434 | int ret; |
c09595f6 PZ |
1435 | |
1436 | rcu_read_lock(); | |
1437 | parent = &root_task_group; | |
1438 | down: | |
eb755805 PZ |
1439 | ret = (*down)(parent, data); |
1440 | if (ret) | |
1441 | goto out_unlock; | |
c09595f6 PZ |
1442 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1443 | parent = child; | |
1444 | goto down; | |
1445 | ||
1446 | up: | |
1447 | continue; | |
1448 | } | |
eb755805 PZ |
1449 | ret = (*up)(parent, data); |
1450 | if (ret) | |
1451 | goto out_unlock; | |
c09595f6 PZ |
1452 | |
1453 | child = parent; | |
1454 | parent = parent->parent; | |
1455 | if (parent) | |
1456 | goto up; | |
eb755805 | 1457 | out_unlock: |
c09595f6 | 1458 | rcu_read_unlock(); |
eb755805 PZ |
1459 | |
1460 | return ret; | |
c09595f6 PZ |
1461 | } |
1462 | ||
eb755805 PZ |
1463 | static int tg_nop(struct task_group *tg, void *data) |
1464 | { | |
1465 | return 0; | |
c09595f6 | 1466 | } |
eb755805 PZ |
1467 | #endif |
1468 | ||
1469 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1470 | /* Used instead of source_load when we know the type == 0 */ |
1471 | static unsigned long weighted_cpuload(const int cpu) | |
1472 | { | |
1473 | return cpu_rq(cpu)->load.weight; | |
1474 | } | |
1475 | ||
1476 | /* | |
1477 | * Return a low guess at the load of a migration-source cpu weighted | |
1478 | * according to the scheduling class and "nice" value. | |
1479 | * | |
1480 | * We want to under-estimate the load of migration sources, to | |
1481 | * balance conservatively. | |
1482 | */ | |
1483 | static unsigned long source_load(int cpu, int type) | |
1484 | { | |
1485 | struct rq *rq = cpu_rq(cpu); | |
1486 | unsigned long total = weighted_cpuload(cpu); | |
1487 | ||
1488 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1489 | return total; | |
1490 | ||
1491 | return min(rq->cpu_load[type-1], total); | |
1492 | } | |
1493 | ||
1494 | /* | |
1495 | * Return a high guess at the load of a migration-target cpu weighted | |
1496 | * according to the scheduling class and "nice" value. | |
1497 | */ | |
1498 | static unsigned long target_load(int cpu, int type) | |
1499 | { | |
1500 | struct rq *rq = cpu_rq(cpu); | |
1501 | unsigned long total = weighted_cpuload(cpu); | |
1502 | ||
1503 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1504 | return total; | |
1505 | ||
1506 | return max(rq->cpu_load[type-1], total); | |
1507 | } | |
1508 | ||
ae154be1 PZ |
1509 | static struct sched_group *group_of(int cpu) |
1510 | { | |
d11c563d | 1511 | struct sched_domain *sd = rcu_dereference_sched(cpu_rq(cpu)->sd); |
ae154be1 PZ |
1512 | |
1513 | if (!sd) | |
1514 | return NULL; | |
1515 | ||
1516 | return sd->groups; | |
1517 | } | |
1518 | ||
1519 | static unsigned long power_of(int cpu) | |
1520 | { | |
1521 | struct sched_group *group = group_of(cpu); | |
1522 | ||
1523 | if (!group) | |
1524 | return SCHED_LOAD_SCALE; | |
1525 | ||
1526 | return group->cpu_power; | |
1527 | } | |
1528 | ||
eb755805 PZ |
1529 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1530 | ||
1531 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1532 | { | |
1533 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1534 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1535 | |
4cd42620 SR |
1536 | if (nr_running) |
1537 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1538 | else |
1539 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1540 | |
1541 | return rq->avg_load_per_task; | |
1542 | } | |
1543 | ||
1544 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1545 | |
43cf38eb | 1546 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1547 | |
c09595f6 PZ |
1548 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1549 | ||
1550 | /* | |
1551 | * Calculate and set the cpu's group shares. | |
1552 | */ | |
34d76c41 PZ |
1553 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1554 | unsigned long sd_shares, | |
1555 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1556 | unsigned long *usd_rq_weight) |
18d95a28 | 1557 | { |
34d76c41 | 1558 | unsigned long shares, rq_weight; |
a5004278 | 1559 | int boost = 0; |
c09595f6 | 1560 | |
4a6cc4bd | 1561 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1562 | if (!rq_weight) { |
1563 | boost = 1; | |
1564 | rq_weight = NICE_0_LOAD; | |
1565 | } | |
c8cba857 | 1566 | |
c09595f6 | 1567 | /* |
a8af7246 PZ |
1568 | * \Sum_j shares_j * rq_weight_i |
1569 | * shares_i = ----------------------------- | |
1570 | * \Sum_j rq_weight_j | |
c09595f6 | 1571 | */ |
ec4e0e2f | 1572 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1573 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1574 | |
ffda12a1 PZ |
1575 | if (abs(shares - tg->se[cpu]->load.weight) > |
1576 | sysctl_sched_shares_thresh) { | |
1577 | struct rq *rq = cpu_rq(cpu); | |
1578 | unsigned long flags; | |
c09595f6 | 1579 | |
05fa785c | 1580 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1581 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1582 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1583 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1584 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1585 | } |
18d95a28 | 1586 | } |
c09595f6 PZ |
1587 | |
1588 | /* | |
c8cba857 PZ |
1589 | * Re-compute the task group their per cpu shares over the given domain. |
1590 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1591 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1592 | */ |
eb755805 | 1593 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1594 | { |
cd8ad40d | 1595 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1596 | unsigned long *usd_rq_weight; |
eb755805 | 1597 | struct sched_domain *sd = data; |
34d76c41 | 1598 | unsigned long flags; |
c8cba857 | 1599 | int i; |
c09595f6 | 1600 | |
34d76c41 PZ |
1601 | if (!tg->se[0]) |
1602 | return 0; | |
1603 | ||
1604 | local_irq_save(flags); | |
4a6cc4bd | 1605 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1606 | |
758b2cdc | 1607 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1608 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1609 | usd_rq_weight[i] = weight; |
34d76c41 | 1610 | |
cd8ad40d | 1611 | rq_weight += weight; |
ec4e0e2f KC |
1612 | /* |
1613 | * If there are currently no tasks on the cpu pretend there | |
1614 | * is one of average load so that when a new task gets to | |
1615 | * run here it will not get delayed by group starvation. | |
1616 | */ | |
ec4e0e2f KC |
1617 | if (!weight) |
1618 | weight = NICE_0_LOAD; | |
1619 | ||
cd8ad40d | 1620 | sum_weight += weight; |
c8cba857 | 1621 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1622 | } |
c09595f6 | 1623 | |
cd8ad40d PZ |
1624 | if (!rq_weight) |
1625 | rq_weight = sum_weight; | |
1626 | ||
c8cba857 PZ |
1627 | if ((!shares && rq_weight) || shares > tg->shares) |
1628 | shares = tg->shares; | |
1629 | ||
1630 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1631 | shares = tg->shares; | |
c09595f6 | 1632 | |
758b2cdc | 1633 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1634 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1635 | |
1636 | local_irq_restore(flags); | |
eb755805 PZ |
1637 | |
1638 | return 0; | |
c09595f6 PZ |
1639 | } |
1640 | ||
1641 | /* | |
c8cba857 PZ |
1642 | * Compute the cpu's hierarchical load factor for each task group. |
1643 | * This needs to be done in a top-down fashion because the load of a child | |
1644 | * group is a fraction of its parents load. | |
c09595f6 | 1645 | */ |
eb755805 | 1646 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1647 | { |
c8cba857 | 1648 | unsigned long load; |
eb755805 | 1649 | long cpu = (long)data; |
c09595f6 | 1650 | |
c8cba857 PZ |
1651 | if (!tg->parent) { |
1652 | load = cpu_rq(cpu)->load.weight; | |
1653 | } else { | |
1654 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1655 | load *= tg->cfs_rq[cpu]->shares; | |
1656 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1657 | } | |
c09595f6 | 1658 | |
c8cba857 | 1659 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1660 | |
eb755805 | 1661 | return 0; |
c09595f6 PZ |
1662 | } |
1663 | ||
c8cba857 | 1664 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1665 | { |
e7097159 PZ |
1666 | s64 elapsed; |
1667 | u64 now; | |
1668 | ||
1669 | if (root_task_group_empty()) | |
1670 | return; | |
1671 | ||
1672 | now = cpu_clock(raw_smp_processor_id()); | |
1673 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1674 | |
1675 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1676 | sd->last_update = now; | |
eb755805 | 1677 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1678 | } |
4d8d595d PZ |
1679 | } |
1680 | ||
eb755805 | 1681 | static void update_h_load(long cpu) |
c09595f6 | 1682 | { |
e7097159 PZ |
1683 | if (root_task_group_empty()) |
1684 | return; | |
1685 | ||
eb755805 | 1686 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1687 | } |
1688 | ||
c09595f6 PZ |
1689 | #else |
1690 | ||
c8cba857 | 1691 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1692 | { |
1693 | } | |
1694 | ||
18d95a28 PZ |
1695 | #endif |
1696 | ||
8f45e2b5 GH |
1697 | #ifdef CONFIG_PREEMPT |
1698 | ||
b78bb868 PZ |
1699 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1700 | ||
70574a99 | 1701 | /* |
8f45e2b5 GH |
1702 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1703 | * way at the expense of forcing extra atomic operations in all | |
1704 | * invocations. This assures that the double_lock is acquired using the | |
1705 | * same underlying policy as the spinlock_t on this architecture, which | |
1706 | * reduces latency compared to the unfair variant below. However, it | |
1707 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1708 | */ |
8f45e2b5 GH |
1709 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1710 | __releases(this_rq->lock) | |
1711 | __acquires(busiest->lock) | |
1712 | __acquires(this_rq->lock) | |
1713 | { | |
05fa785c | 1714 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1715 | double_rq_lock(this_rq, busiest); |
1716 | ||
1717 | return 1; | |
1718 | } | |
1719 | ||
1720 | #else | |
1721 | /* | |
1722 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1723 | * latency by eliminating extra atomic operations when the locks are | |
1724 | * already in proper order on entry. This favors lower cpu-ids and will | |
1725 | * grant the double lock to lower cpus over higher ids under contention, | |
1726 | * regardless of entry order into the function. | |
1727 | */ | |
1728 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1729 | __releases(this_rq->lock) |
1730 | __acquires(busiest->lock) | |
1731 | __acquires(this_rq->lock) | |
1732 | { | |
1733 | int ret = 0; | |
1734 | ||
05fa785c | 1735 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1736 | if (busiest < this_rq) { |
05fa785c TG |
1737 | raw_spin_unlock(&this_rq->lock); |
1738 | raw_spin_lock(&busiest->lock); | |
1739 | raw_spin_lock_nested(&this_rq->lock, | |
1740 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1741 | ret = 1; |
1742 | } else | |
05fa785c TG |
1743 | raw_spin_lock_nested(&busiest->lock, |
1744 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1745 | } |
1746 | return ret; | |
1747 | } | |
1748 | ||
8f45e2b5 GH |
1749 | #endif /* CONFIG_PREEMPT */ |
1750 | ||
1751 | /* | |
1752 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1753 | */ | |
1754 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1755 | { | |
1756 | if (unlikely(!irqs_disabled())) { | |
1757 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1758 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1759 | BUG_ON(1); |
1760 | } | |
1761 | ||
1762 | return _double_lock_balance(this_rq, busiest); | |
1763 | } | |
1764 | ||
70574a99 AD |
1765 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1766 | __releases(busiest->lock) | |
1767 | { | |
05fa785c | 1768 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1769 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1770 | } | |
1e3c88bd PZ |
1771 | |
1772 | /* | |
1773 | * double_rq_lock - safely lock two runqueues | |
1774 | * | |
1775 | * Note this does not disable interrupts like task_rq_lock, | |
1776 | * you need to do so manually before calling. | |
1777 | */ | |
1778 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1779 | __acquires(rq1->lock) | |
1780 | __acquires(rq2->lock) | |
1781 | { | |
1782 | BUG_ON(!irqs_disabled()); | |
1783 | if (rq1 == rq2) { | |
1784 | raw_spin_lock(&rq1->lock); | |
1785 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1786 | } else { | |
1787 | if (rq1 < rq2) { | |
1788 | raw_spin_lock(&rq1->lock); | |
1789 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1790 | } else { | |
1791 | raw_spin_lock(&rq2->lock); | |
1792 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1793 | } | |
1794 | } | |
1e3c88bd PZ |
1795 | } |
1796 | ||
1797 | /* | |
1798 | * double_rq_unlock - safely unlock two runqueues | |
1799 | * | |
1800 | * Note this does not restore interrupts like task_rq_unlock, | |
1801 | * you need to do so manually after calling. | |
1802 | */ | |
1803 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1804 | __releases(rq1->lock) | |
1805 | __releases(rq2->lock) | |
1806 | { | |
1807 | raw_spin_unlock(&rq1->lock); | |
1808 | if (rq1 != rq2) | |
1809 | raw_spin_unlock(&rq2->lock); | |
1810 | else | |
1811 | __release(rq2->lock); | |
1812 | } | |
1813 | ||
18d95a28 PZ |
1814 | #endif |
1815 | ||
30432094 | 1816 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1817 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1818 | { | |
30432094 | 1819 | #ifdef CONFIG_SMP |
34e83e85 IM |
1820 | cfs_rq->shares = shares; |
1821 | #endif | |
1822 | } | |
30432094 | 1823 | #endif |
e7693a36 | 1824 | |
dce48a84 | 1825 | static void calc_load_account_active(struct rq *this_rq); |
0bcdcf28 | 1826 | static void update_sysctl(void); |
acb4a848 | 1827 | static int get_update_sysctl_factor(void); |
dce48a84 | 1828 | |
cd29fe6f PZ |
1829 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1830 | { | |
1831 | set_task_rq(p, cpu); | |
1832 | #ifdef CONFIG_SMP | |
1833 | /* | |
1834 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1835 | * successfuly executed on another CPU. We must ensure that updates of | |
1836 | * per-task data have been completed by this moment. | |
1837 | */ | |
1838 | smp_wmb(); | |
1839 | task_thread_info(p)->cpu = cpu; | |
1840 | #endif | |
1841 | } | |
dce48a84 | 1842 | |
1e3c88bd | 1843 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1844 | |
1845 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1846 | #define for_each_class(class) \ |
1847 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1848 | |
1e3c88bd PZ |
1849 | #include "sched_stats.h" |
1850 | ||
c09595f6 | 1851 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1852 | { |
1853 | rq->nr_running++; | |
9c217245 IM |
1854 | } |
1855 | ||
c09595f6 | 1856 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1857 | { |
1858 | rq->nr_running--; | |
9c217245 IM |
1859 | } |
1860 | ||
45bf76df IM |
1861 | static void set_load_weight(struct task_struct *p) |
1862 | { | |
1863 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1864 | p->se.load.weight = prio_to_weight[0] * 2; |
1865 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1866 | return; | |
1867 | } | |
45bf76df | 1868 | |
dd41f596 IM |
1869 | /* |
1870 | * SCHED_IDLE tasks get minimal weight: | |
1871 | */ | |
1872 | if (p->policy == SCHED_IDLE) { | |
1873 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1874 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1875 | return; | |
1876 | } | |
71f8bd46 | 1877 | |
dd41f596 IM |
1878 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1879 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1880 | } |
1881 | ||
2087a1ad GH |
1882 | static void update_avg(u64 *avg, u64 sample) |
1883 | { | |
1884 | s64 diff = sample - *avg; | |
1885 | *avg += diff >> 3; | |
1886 | } | |
1887 | ||
ea87bb78 TG |
1888 | static void |
1889 | enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, bool head) | |
71f8bd46 | 1890 | { |
a64692a3 | 1891 | update_rq_clock(rq); |
dd41f596 | 1892 | sched_info_queued(p); |
ea87bb78 | 1893 | p->sched_class->enqueue_task(rq, p, wakeup, head); |
dd41f596 | 1894 | p->se.on_rq = 1; |
71f8bd46 IM |
1895 | } |
1896 | ||
69be72c1 | 1897 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1898 | { |
a64692a3 | 1899 | update_rq_clock(rq); |
46ac22ba | 1900 | sched_info_dequeued(p); |
f02231e5 | 1901 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1902 | p->se.on_rq = 0; |
71f8bd46 IM |
1903 | } |
1904 | ||
1e3c88bd PZ |
1905 | /* |
1906 | * activate_task - move a task to the runqueue. | |
1907 | */ | |
1908 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) | |
1909 | { | |
1910 | if (task_contributes_to_load(p)) | |
1911 | rq->nr_uninterruptible--; | |
1912 | ||
ea87bb78 | 1913 | enqueue_task(rq, p, wakeup, false); |
1e3c88bd PZ |
1914 | inc_nr_running(rq); |
1915 | } | |
1916 | ||
1917 | /* | |
1918 | * deactivate_task - remove a task from the runqueue. | |
1919 | */ | |
1920 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) | |
1921 | { | |
1922 | if (task_contributes_to_load(p)) | |
1923 | rq->nr_uninterruptible++; | |
1924 | ||
1925 | dequeue_task(rq, p, sleep); | |
1926 | dec_nr_running(rq); | |
1927 | } | |
1928 | ||
1929 | #include "sched_idletask.c" | |
1930 | #include "sched_fair.c" | |
1931 | #include "sched_rt.c" | |
1932 | #ifdef CONFIG_SCHED_DEBUG | |
1933 | # include "sched_debug.c" | |
1934 | #endif | |
1935 | ||
14531189 | 1936 | /* |
dd41f596 | 1937 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1938 | */ |
14531189 IM |
1939 | static inline int __normal_prio(struct task_struct *p) |
1940 | { | |
dd41f596 | 1941 | return p->static_prio; |
14531189 IM |
1942 | } |
1943 | ||
b29739f9 IM |
1944 | /* |
1945 | * Calculate the expected normal priority: i.e. priority | |
1946 | * without taking RT-inheritance into account. Might be | |
1947 | * boosted by interactivity modifiers. Changes upon fork, | |
1948 | * setprio syscalls, and whenever the interactivity | |
1949 | * estimator recalculates. | |
1950 | */ | |
36c8b586 | 1951 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1952 | { |
1953 | int prio; | |
1954 | ||
e05606d3 | 1955 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1956 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1957 | else | |
1958 | prio = __normal_prio(p); | |
1959 | return prio; | |
1960 | } | |
1961 | ||
1962 | /* | |
1963 | * Calculate the current priority, i.e. the priority | |
1964 | * taken into account by the scheduler. This value might | |
1965 | * be boosted by RT tasks, or might be boosted by | |
1966 | * interactivity modifiers. Will be RT if the task got | |
1967 | * RT-boosted. If not then it returns p->normal_prio. | |
1968 | */ | |
36c8b586 | 1969 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1970 | { |
1971 | p->normal_prio = normal_prio(p); | |
1972 | /* | |
1973 | * If we are RT tasks or we were boosted to RT priority, | |
1974 | * keep the priority unchanged. Otherwise, update priority | |
1975 | * to the normal priority: | |
1976 | */ | |
1977 | if (!rt_prio(p->prio)) | |
1978 | return p->normal_prio; | |
1979 | return p->prio; | |
1980 | } | |
1981 | ||
1da177e4 LT |
1982 | /** |
1983 | * task_curr - is this task currently executing on a CPU? | |
1984 | * @p: the task in question. | |
1985 | */ | |
36c8b586 | 1986 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1987 | { |
1988 | return cpu_curr(task_cpu(p)) == p; | |
1989 | } | |
1990 | ||
cb469845 SR |
1991 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1992 | const struct sched_class *prev_class, | |
1993 | int oldprio, int running) | |
1994 | { | |
1995 | if (prev_class != p->sched_class) { | |
1996 | if (prev_class->switched_from) | |
1997 | prev_class->switched_from(rq, p, running); | |
1998 | p->sched_class->switched_to(rq, p, running); | |
1999 | } else | |
2000 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
2001 | } | |
2002 | ||
1da177e4 | 2003 | #ifdef CONFIG_SMP |
cc367732 IM |
2004 | /* |
2005 | * Is this task likely cache-hot: | |
2006 | */ | |
e7693a36 | 2007 | static int |
cc367732 IM |
2008 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2009 | { | |
2010 | s64 delta; | |
2011 | ||
e6c8fba7 PZ |
2012 | if (p->sched_class != &fair_sched_class) |
2013 | return 0; | |
2014 | ||
f540a608 IM |
2015 | /* |
2016 | * Buddy candidates are cache hot: | |
2017 | */ | |
f685ceac | 2018 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2019 | (&p->se == cfs_rq_of(&p->se)->next || |
2020 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2021 | return 1; |
2022 | ||
6bc1665b IM |
2023 | if (sysctl_sched_migration_cost == -1) |
2024 | return 1; | |
2025 | if (sysctl_sched_migration_cost == 0) | |
2026 | return 0; | |
2027 | ||
cc367732 IM |
2028 | delta = now - p->se.exec_start; |
2029 | ||
2030 | return delta < (s64)sysctl_sched_migration_cost; | |
2031 | } | |
2032 | ||
dd41f596 | 2033 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2034 | { |
e2912009 PZ |
2035 | #ifdef CONFIG_SCHED_DEBUG |
2036 | /* | |
2037 | * We should never call set_task_cpu() on a blocked task, | |
2038 | * ttwu() will sort out the placement. | |
2039 | */ | |
077614ee PZ |
2040 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2041 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2042 | #endif |
2043 | ||
de1d7286 | 2044 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2045 | |
0c69774e PZ |
2046 | if (task_cpu(p) != new_cpu) { |
2047 | p->se.nr_migrations++; | |
2048 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2049 | } | |
dd41f596 IM |
2050 | |
2051 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2052 | } |
2053 | ||
70b97a7f | 2054 | struct migration_req { |
1da177e4 | 2055 | struct list_head list; |
1da177e4 | 2056 | |
36c8b586 | 2057 | struct task_struct *task; |
1da177e4 LT |
2058 | int dest_cpu; |
2059 | ||
1da177e4 | 2060 | struct completion done; |
70b97a7f | 2061 | }; |
1da177e4 LT |
2062 | |
2063 | /* | |
2064 | * The task's runqueue lock must be held. | |
2065 | * Returns true if you have to wait for migration thread. | |
2066 | */ | |
36c8b586 | 2067 | static int |
70b97a7f | 2068 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2069 | { |
70b97a7f | 2070 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2071 | |
2072 | /* | |
2073 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2074 | * the next wake-up will properly place the task. |
1da177e4 | 2075 | */ |
e2912009 | 2076 | if (!p->se.on_rq && !task_running(rq, p)) |
1da177e4 | 2077 | return 0; |
1da177e4 LT |
2078 | |
2079 | init_completion(&req->done); | |
1da177e4 LT |
2080 | req->task = p; |
2081 | req->dest_cpu = dest_cpu; | |
2082 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2083 | |
1da177e4 LT |
2084 | return 1; |
2085 | } | |
2086 | ||
a26b89f0 MM |
2087 | /* |
2088 | * wait_task_context_switch - wait for a thread to complete at least one | |
2089 | * context switch. | |
2090 | * | |
2091 | * @p must not be current. | |
2092 | */ | |
2093 | void wait_task_context_switch(struct task_struct *p) | |
2094 | { | |
2095 | unsigned long nvcsw, nivcsw, flags; | |
2096 | int running; | |
2097 | struct rq *rq; | |
2098 | ||
2099 | nvcsw = p->nvcsw; | |
2100 | nivcsw = p->nivcsw; | |
2101 | for (;;) { | |
2102 | /* | |
2103 | * The runqueue is assigned before the actual context | |
2104 | * switch. We need to take the runqueue lock. | |
2105 | * | |
2106 | * We could check initially without the lock but it is | |
2107 | * very likely that we need to take the lock in every | |
2108 | * iteration. | |
2109 | */ | |
2110 | rq = task_rq_lock(p, &flags); | |
2111 | running = task_running(rq, p); | |
2112 | task_rq_unlock(rq, &flags); | |
2113 | ||
2114 | if (likely(!running)) | |
2115 | break; | |
2116 | /* | |
2117 | * The switch count is incremented before the actual | |
2118 | * context switch. We thus wait for two switches to be | |
2119 | * sure at least one completed. | |
2120 | */ | |
2121 | if ((p->nvcsw - nvcsw) > 1) | |
2122 | break; | |
2123 | if ((p->nivcsw - nivcsw) > 1) | |
2124 | break; | |
2125 | ||
2126 | cpu_relax(); | |
2127 | } | |
2128 | } | |
2129 | ||
1da177e4 LT |
2130 | /* |
2131 | * wait_task_inactive - wait for a thread to unschedule. | |
2132 | * | |
85ba2d86 RM |
2133 | * If @match_state is nonzero, it's the @p->state value just checked and |
2134 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2135 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2136 | * we return a positive number (its total switch count). If a second call | |
2137 | * a short while later returns the same number, the caller can be sure that | |
2138 | * @p has remained unscheduled the whole time. | |
2139 | * | |
1da177e4 LT |
2140 | * The caller must ensure that the task *will* unschedule sometime soon, |
2141 | * else this function might spin for a *long* time. This function can't | |
2142 | * be called with interrupts off, or it may introduce deadlock with | |
2143 | * smp_call_function() if an IPI is sent by the same process we are | |
2144 | * waiting to become inactive. | |
2145 | */ | |
85ba2d86 | 2146 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2147 | { |
2148 | unsigned long flags; | |
dd41f596 | 2149 | int running, on_rq; |
85ba2d86 | 2150 | unsigned long ncsw; |
70b97a7f | 2151 | struct rq *rq; |
1da177e4 | 2152 | |
3a5c359a AK |
2153 | for (;;) { |
2154 | /* | |
2155 | * We do the initial early heuristics without holding | |
2156 | * any task-queue locks at all. We'll only try to get | |
2157 | * the runqueue lock when things look like they will | |
2158 | * work out! | |
2159 | */ | |
2160 | rq = task_rq(p); | |
fa490cfd | 2161 | |
3a5c359a AK |
2162 | /* |
2163 | * If the task is actively running on another CPU | |
2164 | * still, just relax and busy-wait without holding | |
2165 | * any locks. | |
2166 | * | |
2167 | * NOTE! Since we don't hold any locks, it's not | |
2168 | * even sure that "rq" stays as the right runqueue! | |
2169 | * But we don't care, since "task_running()" will | |
2170 | * return false if the runqueue has changed and p | |
2171 | * is actually now running somewhere else! | |
2172 | */ | |
85ba2d86 RM |
2173 | while (task_running(rq, p)) { |
2174 | if (match_state && unlikely(p->state != match_state)) | |
2175 | return 0; | |
3a5c359a | 2176 | cpu_relax(); |
85ba2d86 | 2177 | } |
fa490cfd | 2178 | |
3a5c359a AK |
2179 | /* |
2180 | * Ok, time to look more closely! We need the rq | |
2181 | * lock now, to be *sure*. If we're wrong, we'll | |
2182 | * just go back and repeat. | |
2183 | */ | |
2184 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2185 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2186 | running = task_running(rq, p); |
2187 | on_rq = p->se.on_rq; | |
85ba2d86 | 2188 | ncsw = 0; |
f31e11d8 | 2189 | if (!match_state || p->state == match_state) |
93dcf55f | 2190 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2191 | task_rq_unlock(rq, &flags); |
fa490cfd | 2192 | |
85ba2d86 RM |
2193 | /* |
2194 | * If it changed from the expected state, bail out now. | |
2195 | */ | |
2196 | if (unlikely(!ncsw)) | |
2197 | break; | |
2198 | ||
3a5c359a AK |
2199 | /* |
2200 | * Was it really running after all now that we | |
2201 | * checked with the proper locks actually held? | |
2202 | * | |
2203 | * Oops. Go back and try again.. | |
2204 | */ | |
2205 | if (unlikely(running)) { | |
2206 | cpu_relax(); | |
2207 | continue; | |
2208 | } | |
fa490cfd | 2209 | |
3a5c359a AK |
2210 | /* |
2211 | * It's not enough that it's not actively running, | |
2212 | * it must be off the runqueue _entirely_, and not | |
2213 | * preempted! | |
2214 | * | |
80dd99b3 | 2215 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2216 | * running right now), it's preempted, and we should |
2217 | * yield - it could be a while. | |
2218 | */ | |
2219 | if (unlikely(on_rq)) { | |
2220 | schedule_timeout_uninterruptible(1); | |
2221 | continue; | |
2222 | } | |
fa490cfd | 2223 | |
3a5c359a AK |
2224 | /* |
2225 | * Ahh, all good. It wasn't running, and it wasn't | |
2226 | * runnable, which means that it will never become | |
2227 | * running in the future either. We're all done! | |
2228 | */ | |
2229 | break; | |
2230 | } | |
85ba2d86 RM |
2231 | |
2232 | return ncsw; | |
1da177e4 LT |
2233 | } |
2234 | ||
2235 | /*** | |
2236 | * kick_process - kick a running thread to enter/exit the kernel | |
2237 | * @p: the to-be-kicked thread | |
2238 | * | |
2239 | * Cause a process which is running on another CPU to enter | |
2240 | * kernel-mode, without any delay. (to get signals handled.) | |
2241 | * | |
2242 | * NOTE: this function doesnt have to take the runqueue lock, | |
2243 | * because all it wants to ensure is that the remote task enters | |
2244 | * the kernel. If the IPI races and the task has been migrated | |
2245 | * to another CPU then no harm is done and the purpose has been | |
2246 | * achieved as well. | |
2247 | */ | |
36c8b586 | 2248 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2249 | { |
2250 | int cpu; | |
2251 | ||
2252 | preempt_disable(); | |
2253 | cpu = task_cpu(p); | |
2254 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2255 | smp_send_reschedule(cpu); | |
2256 | preempt_enable(); | |
2257 | } | |
b43e3521 | 2258 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2259 | #endif /* CONFIG_SMP */ |
1da177e4 | 2260 | |
0793a61d TG |
2261 | /** |
2262 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2263 | * @p: the task to evaluate | |
2264 | * @func: the function to be called | |
2265 | * @info: the function call argument | |
2266 | * | |
2267 | * Calls the function @func when the task is currently running. This might | |
2268 | * be on the current CPU, which just calls the function directly | |
2269 | */ | |
2270 | void task_oncpu_function_call(struct task_struct *p, | |
2271 | void (*func) (void *info), void *info) | |
2272 | { | |
2273 | int cpu; | |
2274 | ||
2275 | preempt_disable(); | |
2276 | cpu = task_cpu(p); | |
2277 | if (task_curr(p)) | |
2278 | smp_call_function_single(cpu, func, info, 1); | |
2279 | preempt_enable(); | |
2280 | } | |
2281 | ||
970b13ba | 2282 | #ifdef CONFIG_SMP |
30da688e ON |
2283 | /* |
2284 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2285 | */ | |
5da9a0fb PZ |
2286 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2287 | { | |
2288 | int dest_cpu; | |
2289 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2290 | ||
2291 | /* Look for allowed, online CPU in same node. */ | |
2292 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2293 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2294 | return dest_cpu; | |
2295 | ||
2296 | /* Any allowed, online CPU? */ | |
2297 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2298 | if (dest_cpu < nr_cpu_ids) | |
2299 | return dest_cpu; | |
2300 | ||
2301 | /* No more Mr. Nice Guy. */ | |
897f0b3c | 2302 | if (unlikely(dest_cpu >= nr_cpu_ids)) { |
9084bb82 | 2303 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
5da9a0fb PZ |
2304 | /* |
2305 | * Don't tell them about moving exiting tasks or | |
2306 | * kernel threads (both mm NULL), since they never | |
2307 | * leave kernel. | |
2308 | */ | |
2309 | if (p->mm && printk_ratelimit()) { | |
2310 | printk(KERN_INFO "process %d (%s) no " | |
2311 | "longer affine to cpu%d\n", | |
2312 | task_pid_nr(p), p->comm, cpu); | |
2313 | } | |
2314 | } | |
2315 | ||
2316 | return dest_cpu; | |
2317 | } | |
2318 | ||
e2912009 | 2319 | /* |
30da688e | 2320 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2321 | */ |
970b13ba PZ |
2322 | static inline |
2323 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) | |
2324 | { | |
e2912009 PZ |
2325 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
2326 | ||
2327 | /* | |
2328 | * In order not to call set_task_cpu() on a blocking task we need | |
2329 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2330 | * cpu. | |
2331 | * | |
2332 | * Since this is common to all placement strategies, this lives here. | |
2333 | * | |
2334 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2335 | * not worry about this generic constraint ] | |
2336 | */ | |
2337 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2338 | !cpu_online(cpu))) |
5da9a0fb | 2339 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2340 | |
2341 | return cpu; | |
970b13ba PZ |
2342 | } |
2343 | #endif | |
2344 | ||
1da177e4 LT |
2345 | /*** |
2346 | * try_to_wake_up - wake up a thread | |
2347 | * @p: the to-be-woken-up thread | |
2348 | * @state: the mask of task states that can be woken | |
2349 | * @sync: do a synchronous wakeup? | |
2350 | * | |
2351 | * Put it on the run-queue if it's not already there. The "current" | |
2352 | * thread is always on the run-queue (except when the actual | |
2353 | * re-schedule is in progress), and as such you're allowed to do | |
2354 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2355 | * runnable without the overhead of this. | |
2356 | * | |
2357 | * returns failure only if the task is already active. | |
2358 | */ | |
7d478721 PZ |
2359 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2360 | int wake_flags) | |
1da177e4 | 2361 | { |
cc367732 | 2362 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2363 | unsigned long flags; |
ab3b3aa5 | 2364 | struct rq *rq; |
1da177e4 | 2365 | |
e9c84311 | 2366 | this_cpu = get_cpu(); |
2398f2c6 | 2367 | |
04e2f174 | 2368 | smp_wmb(); |
ab3b3aa5 | 2369 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2370 | if (!(p->state & state)) |
1da177e4 LT |
2371 | goto out; |
2372 | ||
dd41f596 | 2373 | if (p->se.on_rq) |
1da177e4 LT |
2374 | goto out_running; |
2375 | ||
2376 | cpu = task_cpu(p); | |
cc367732 | 2377 | orig_cpu = cpu; |
1da177e4 LT |
2378 | |
2379 | #ifdef CONFIG_SMP | |
2380 | if (unlikely(task_running(rq, p))) | |
2381 | goto out_activate; | |
2382 | ||
e9c84311 PZ |
2383 | /* |
2384 | * In order to handle concurrent wakeups and release the rq->lock | |
2385 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2386 | * |
2387 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2388 | */ |
eb24073b IM |
2389 | if (task_contributes_to_load(p)) |
2390 | rq->nr_uninterruptible--; | |
e9c84311 | 2391 | p->state = TASK_WAKING; |
efbbd05a PZ |
2392 | |
2393 | if (p->sched_class->task_waking) | |
2394 | p->sched_class->task_waking(rq, p); | |
2395 | ||
ab19cb23 | 2396 | __task_rq_unlock(rq); |
e9c84311 | 2397 | |
970b13ba | 2398 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
0970d299 PZ |
2399 | if (cpu != orig_cpu) { |
2400 | /* | |
2401 | * Since we migrate the task without holding any rq->lock, | |
2402 | * we need to be careful with task_rq_lock(), since that | |
2403 | * might end up locking an invalid rq. | |
2404 | */ | |
5d2f5a61 | 2405 | set_task_cpu(p, cpu); |
0970d299 | 2406 | } |
ab19cb23 | 2407 | |
0970d299 PZ |
2408 | rq = cpu_rq(cpu); |
2409 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2410 | |
0970d299 PZ |
2411 | /* |
2412 | * We migrated the task without holding either rq->lock, however | |
2413 | * since the task is not on the task list itself, nobody else | |
2414 | * will try and migrate the task, hence the rq should match the | |
2415 | * cpu we just moved it to. | |
2416 | */ | |
2417 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2418 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2419 | |
e7693a36 GH |
2420 | #ifdef CONFIG_SCHEDSTATS |
2421 | schedstat_inc(rq, ttwu_count); | |
2422 | if (cpu == this_cpu) | |
2423 | schedstat_inc(rq, ttwu_local); | |
2424 | else { | |
2425 | struct sched_domain *sd; | |
2426 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2427 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2428 | schedstat_inc(sd, ttwu_wake_remote); |
2429 | break; | |
2430 | } | |
2431 | } | |
2432 | } | |
6d6bc0ad | 2433 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2434 | |
1da177e4 LT |
2435 | out_activate: |
2436 | #endif /* CONFIG_SMP */ | |
41acab88 | 2437 | schedstat_inc(p, se.statistics.nr_wakeups); |
7d478721 | 2438 | if (wake_flags & WF_SYNC) |
41acab88 | 2439 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
cc367732 | 2440 | if (orig_cpu != cpu) |
41acab88 | 2441 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
cc367732 | 2442 | if (cpu == this_cpu) |
41acab88 | 2443 | schedstat_inc(p, se.statistics.nr_wakeups_local); |
cc367732 | 2444 | else |
41acab88 | 2445 | schedstat_inc(p, se.statistics.nr_wakeups_remote); |
dd41f596 | 2446 | activate_task(rq, p, 1); |
1da177e4 LT |
2447 | success = 1; |
2448 | ||
2449 | out_running: | |
468a15bb | 2450 | trace_sched_wakeup(rq, p, success); |
7d478721 | 2451 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2452 | |
1da177e4 | 2453 | p->state = TASK_RUNNING; |
9a897c5a | 2454 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2455 | if (p->sched_class->task_woken) |
2456 | p->sched_class->task_woken(rq, p); | |
eae0c9df MG |
2457 | |
2458 | if (unlikely(rq->idle_stamp)) { | |
2459 | u64 delta = rq->clock - rq->idle_stamp; | |
2460 | u64 max = 2*sysctl_sched_migration_cost; | |
2461 | ||
2462 | if (delta > max) | |
2463 | rq->avg_idle = max; | |
2464 | else | |
2465 | update_avg(&rq->avg_idle, delta); | |
2466 | rq->idle_stamp = 0; | |
2467 | } | |
9a897c5a | 2468 | #endif |
1da177e4 LT |
2469 | out: |
2470 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2471 | put_cpu(); |
1da177e4 LT |
2472 | |
2473 | return success; | |
2474 | } | |
2475 | ||
50fa610a DH |
2476 | /** |
2477 | * wake_up_process - Wake up a specific process | |
2478 | * @p: The process to be woken up. | |
2479 | * | |
2480 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2481 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2482 | * running. | |
2483 | * | |
2484 | * It may be assumed that this function implies a write memory barrier before | |
2485 | * changing the task state if and only if any tasks are woken up. | |
2486 | */ | |
7ad5b3a5 | 2487 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2488 | { |
d9514f6c | 2489 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2490 | } |
1da177e4 LT |
2491 | EXPORT_SYMBOL(wake_up_process); |
2492 | ||
7ad5b3a5 | 2493 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2494 | { |
2495 | return try_to_wake_up(p, state, 0); | |
2496 | } | |
2497 | ||
1da177e4 LT |
2498 | /* |
2499 | * Perform scheduler related setup for a newly forked process p. | |
2500 | * p is forked by current. | |
dd41f596 IM |
2501 | * |
2502 | * __sched_fork() is basic setup used by init_idle() too: | |
2503 | */ | |
2504 | static void __sched_fork(struct task_struct *p) | |
2505 | { | |
dd41f596 IM |
2506 | p->se.exec_start = 0; |
2507 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2508 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2509 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2510 | |
2511 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2512 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2513 | #endif |
476d139c | 2514 | |
fa717060 | 2515 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2516 | p->se.on_rq = 0; |
4a55bd5e | 2517 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2518 | |
e107be36 AK |
2519 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2520 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2521 | #endif | |
dd41f596 IM |
2522 | } |
2523 | ||
2524 | /* | |
2525 | * fork()/clone()-time setup: | |
2526 | */ | |
2527 | void sched_fork(struct task_struct *p, int clone_flags) | |
2528 | { | |
2529 | int cpu = get_cpu(); | |
2530 | ||
2531 | __sched_fork(p); | |
06b83b5f PZ |
2532 | /* |
2533 | * We mark the process as waking here. This guarantees that | |
2534 | * nobody will actually run it, and a signal or other external | |
2535 | * event cannot wake it up and insert it on the runqueue either. | |
2536 | */ | |
2537 | p->state = TASK_WAKING; | |
dd41f596 | 2538 | |
b9dc29e7 MG |
2539 | /* |
2540 | * Revert to default priority/policy on fork if requested. | |
2541 | */ | |
2542 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2543 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2544 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2545 | p->normal_prio = p->static_prio; |
2546 | } | |
b9dc29e7 | 2547 | |
6c697bdf MG |
2548 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2549 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2550 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2551 | set_load_weight(p); |
2552 | } | |
2553 | ||
b9dc29e7 MG |
2554 | /* |
2555 | * We don't need the reset flag anymore after the fork. It has | |
2556 | * fulfilled its duty: | |
2557 | */ | |
2558 | p->sched_reset_on_fork = 0; | |
2559 | } | |
ca94c442 | 2560 | |
f83f9ac2 PW |
2561 | /* |
2562 | * Make sure we do not leak PI boosting priority to the child. | |
2563 | */ | |
2564 | p->prio = current->normal_prio; | |
2565 | ||
2ddbf952 HS |
2566 | if (!rt_prio(p->prio)) |
2567 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2568 | |
cd29fe6f PZ |
2569 | if (p->sched_class->task_fork) |
2570 | p->sched_class->task_fork(p); | |
2571 | ||
5f3edc1b PZ |
2572 | set_task_cpu(p, cpu); |
2573 | ||
52f17b6c | 2574 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2575 | if (likely(sched_info_on())) |
52f17b6c | 2576 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2577 | #endif |
d6077cb8 | 2578 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2579 | p->oncpu = 0; |
2580 | #endif | |
1da177e4 | 2581 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2582 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2583 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2584 | #endif |
917b627d GH |
2585 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2586 | ||
476d139c | 2587 | put_cpu(); |
1da177e4 LT |
2588 | } |
2589 | ||
2590 | /* | |
2591 | * wake_up_new_task - wake up a newly created task for the first time. | |
2592 | * | |
2593 | * This function will do some initial scheduler statistics housekeeping | |
2594 | * that must be done for every newly created context, then puts the task | |
2595 | * on the runqueue and wakes it. | |
2596 | */ | |
7ad5b3a5 | 2597 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2598 | { |
2599 | unsigned long flags; | |
dd41f596 | 2600 | struct rq *rq; |
c890692b | 2601 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2602 | |
2603 | #ifdef CONFIG_SMP | |
2604 | /* | |
2605 | * Fork balancing, do it here and not earlier because: | |
2606 | * - cpus_allowed can change in the fork path | |
2607 | * - any previously selected cpu might disappear through hotplug | |
2608 | * | |
2609 | * We still have TASK_WAKING but PF_STARTING is gone now, meaning | |
2610 | * ->cpus_allowed is stable, we have preemption disabled, meaning | |
2611 | * cpu_online_mask is stable. | |
2612 | */ | |
2613 | cpu = select_task_rq(p, SD_BALANCE_FORK, 0); | |
2614 | set_task_cpu(p, cpu); | |
2615 | #endif | |
1da177e4 | 2616 | |
0970d299 PZ |
2617 | /* |
2618 | * Since the task is not on the rq and we still have TASK_WAKING set | |
2619 | * nobody else will migrate this task. | |
2620 | */ | |
2621 | rq = cpu_rq(cpu); | |
2622 | raw_spin_lock_irqsave(&rq->lock, flags); | |
2623 | ||
06b83b5f PZ |
2624 | BUG_ON(p->state != TASK_WAKING); |
2625 | p->state = TASK_RUNNING; | |
cd29fe6f | 2626 | activate_task(rq, p, 0); |
c71dd42d | 2627 | trace_sched_wakeup_new(rq, p, 1); |
a7558e01 | 2628 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2629 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2630 | if (p->sched_class->task_woken) |
2631 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2632 | #endif |
dd41f596 | 2633 | task_rq_unlock(rq, &flags); |
fabf318e | 2634 | put_cpu(); |
1da177e4 LT |
2635 | } |
2636 | ||
e107be36 AK |
2637 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2638 | ||
2639 | /** | |
80dd99b3 | 2640 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2641 | * @notifier: notifier struct to register |
e107be36 AK |
2642 | */ |
2643 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2644 | { | |
2645 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2646 | } | |
2647 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2648 | ||
2649 | /** | |
2650 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2651 | * @notifier: notifier struct to unregister |
e107be36 AK |
2652 | * |
2653 | * This is safe to call from within a preemption notifier. | |
2654 | */ | |
2655 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2656 | { | |
2657 | hlist_del(¬ifier->link); | |
2658 | } | |
2659 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2660 | ||
2661 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2662 | { | |
2663 | struct preempt_notifier *notifier; | |
2664 | struct hlist_node *node; | |
2665 | ||
2666 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2667 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2668 | } | |
2669 | ||
2670 | static void | |
2671 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2672 | struct task_struct *next) | |
2673 | { | |
2674 | struct preempt_notifier *notifier; | |
2675 | struct hlist_node *node; | |
2676 | ||
2677 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2678 | notifier->ops->sched_out(notifier, next); | |
2679 | } | |
2680 | ||
6d6bc0ad | 2681 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2682 | |
2683 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2684 | { | |
2685 | } | |
2686 | ||
2687 | static void | |
2688 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2689 | struct task_struct *next) | |
2690 | { | |
2691 | } | |
2692 | ||
6d6bc0ad | 2693 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2694 | |
4866cde0 NP |
2695 | /** |
2696 | * prepare_task_switch - prepare to switch tasks | |
2697 | * @rq: the runqueue preparing to switch | |
421cee29 | 2698 | * @prev: the current task that is being switched out |
4866cde0 NP |
2699 | * @next: the task we are going to switch to. |
2700 | * | |
2701 | * This is called with the rq lock held and interrupts off. It must | |
2702 | * be paired with a subsequent finish_task_switch after the context | |
2703 | * switch. | |
2704 | * | |
2705 | * prepare_task_switch sets up locking and calls architecture specific | |
2706 | * hooks. | |
2707 | */ | |
e107be36 AK |
2708 | static inline void |
2709 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2710 | struct task_struct *next) | |
4866cde0 | 2711 | { |
e107be36 | 2712 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2713 | prepare_lock_switch(rq, next); |
2714 | prepare_arch_switch(next); | |
2715 | } | |
2716 | ||
1da177e4 LT |
2717 | /** |
2718 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2719 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2720 | * @prev: the thread we just switched away from. |
2721 | * | |
4866cde0 NP |
2722 | * finish_task_switch must be called after the context switch, paired |
2723 | * with a prepare_task_switch call before the context switch. | |
2724 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2725 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2726 | * |
2727 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2728 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2729 | * with the lock held can cause deadlocks; see schedule() for |
2730 | * details.) | |
2731 | */ | |
a9957449 | 2732 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2733 | __releases(rq->lock) |
2734 | { | |
1da177e4 | 2735 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2736 | long prev_state; |
1da177e4 LT |
2737 | |
2738 | rq->prev_mm = NULL; | |
2739 | ||
2740 | /* | |
2741 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2742 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2743 | * schedule one last time. The schedule call will never return, and |
2744 | * the scheduled task must drop that reference. | |
c394cc9f | 2745 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2746 | * still held, otherwise prev could be scheduled on another cpu, die |
2747 | * there before we look at prev->state, and then the reference would | |
2748 | * be dropped twice. | |
2749 | * Manfred Spraul <manfred@colorfullife.com> | |
2750 | */ | |
55a101f8 | 2751 | prev_state = prev->state; |
4866cde0 | 2752 | finish_arch_switch(prev); |
8381f65d JI |
2753 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2754 | local_irq_disable(); | |
2755 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2756 | perf_event_task_sched_in(current); |
8381f65d JI |
2757 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2758 | local_irq_enable(); | |
2759 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2760 | finish_lock_switch(rq, prev); |
e8fa1362 | 2761 | |
e107be36 | 2762 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2763 | if (mm) |
2764 | mmdrop(mm); | |
c394cc9f | 2765 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2766 | /* |
2767 | * Remove function-return probe instances associated with this | |
2768 | * task and put them back on the free list. | |
9761eea8 | 2769 | */ |
c6fd91f0 | 2770 | kprobe_flush_task(prev); |
1da177e4 | 2771 | put_task_struct(prev); |
c6fd91f0 | 2772 | } |
1da177e4 LT |
2773 | } |
2774 | ||
3f029d3c GH |
2775 | #ifdef CONFIG_SMP |
2776 | ||
2777 | /* assumes rq->lock is held */ | |
2778 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2779 | { | |
2780 | if (prev->sched_class->pre_schedule) | |
2781 | prev->sched_class->pre_schedule(rq, prev); | |
2782 | } | |
2783 | ||
2784 | /* rq->lock is NOT held, but preemption is disabled */ | |
2785 | static inline void post_schedule(struct rq *rq) | |
2786 | { | |
2787 | if (rq->post_schedule) { | |
2788 | unsigned long flags; | |
2789 | ||
05fa785c | 2790 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2791 | if (rq->curr->sched_class->post_schedule) |
2792 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2793 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2794 | |
2795 | rq->post_schedule = 0; | |
2796 | } | |
2797 | } | |
2798 | ||
2799 | #else | |
da19ab51 | 2800 | |
3f029d3c GH |
2801 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2802 | { | |
2803 | } | |
2804 | ||
2805 | static inline void post_schedule(struct rq *rq) | |
2806 | { | |
1da177e4 LT |
2807 | } |
2808 | ||
3f029d3c GH |
2809 | #endif |
2810 | ||
1da177e4 LT |
2811 | /** |
2812 | * schedule_tail - first thing a freshly forked thread must call. | |
2813 | * @prev: the thread we just switched away from. | |
2814 | */ | |
36c8b586 | 2815 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2816 | __releases(rq->lock) |
2817 | { | |
70b97a7f IM |
2818 | struct rq *rq = this_rq(); |
2819 | ||
4866cde0 | 2820 | finish_task_switch(rq, prev); |
da19ab51 | 2821 | |
3f029d3c GH |
2822 | /* |
2823 | * FIXME: do we need to worry about rq being invalidated by the | |
2824 | * task_switch? | |
2825 | */ | |
2826 | post_schedule(rq); | |
70b97a7f | 2827 | |
4866cde0 NP |
2828 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2829 | /* In this case, finish_task_switch does not reenable preemption */ | |
2830 | preempt_enable(); | |
2831 | #endif | |
1da177e4 | 2832 | if (current->set_child_tid) |
b488893a | 2833 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2834 | } |
2835 | ||
2836 | /* | |
2837 | * context_switch - switch to the new MM and the new | |
2838 | * thread's register state. | |
2839 | */ | |
dd41f596 | 2840 | static inline void |
70b97a7f | 2841 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2842 | struct task_struct *next) |
1da177e4 | 2843 | { |
dd41f596 | 2844 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2845 | |
e107be36 | 2846 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2847 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2848 | mm = next->mm; |
2849 | oldmm = prev->active_mm; | |
9226d125 ZA |
2850 | /* |
2851 | * For paravirt, this is coupled with an exit in switch_to to | |
2852 | * combine the page table reload and the switch backend into | |
2853 | * one hypercall. | |
2854 | */ | |
224101ed | 2855 | arch_start_context_switch(prev); |
9226d125 | 2856 | |
710390d9 | 2857 | if (likely(!mm)) { |
1da177e4 LT |
2858 | next->active_mm = oldmm; |
2859 | atomic_inc(&oldmm->mm_count); | |
2860 | enter_lazy_tlb(oldmm, next); | |
2861 | } else | |
2862 | switch_mm(oldmm, mm, next); | |
2863 | ||
710390d9 | 2864 | if (likely(!prev->mm)) { |
1da177e4 | 2865 | prev->active_mm = NULL; |
1da177e4 LT |
2866 | rq->prev_mm = oldmm; |
2867 | } | |
3a5f5e48 IM |
2868 | /* |
2869 | * Since the runqueue lock will be released by the next | |
2870 | * task (which is an invalid locking op but in the case | |
2871 | * of the scheduler it's an obvious special-case), so we | |
2872 | * do an early lockdep release here: | |
2873 | */ | |
2874 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2875 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2876 | #endif |
1da177e4 LT |
2877 | |
2878 | /* Here we just switch the register state and the stack. */ | |
2879 | switch_to(prev, next, prev); | |
2880 | ||
dd41f596 IM |
2881 | barrier(); |
2882 | /* | |
2883 | * this_rq must be evaluated again because prev may have moved | |
2884 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2885 | * frame will be invalid. | |
2886 | */ | |
2887 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2888 | } |
2889 | ||
2890 | /* | |
2891 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2892 | * | |
2893 | * externally visible scheduler statistics: current number of runnable | |
2894 | * threads, current number of uninterruptible-sleeping threads, total | |
2895 | * number of context switches performed since bootup. | |
2896 | */ | |
2897 | unsigned long nr_running(void) | |
2898 | { | |
2899 | unsigned long i, sum = 0; | |
2900 | ||
2901 | for_each_online_cpu(i) | |
2902 | sum += cpu_rq(i)->nr_running; | |
2903 | ||
2904 | return sum; | |
f711f609 | 2905 | } |
1da177e4 LT |
2906 | |
2907 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2908 | { |
1da177e4 | 2909 | unsigned long i, sum = 0; |
f711f609 | 2910 | |
0a945022 | 2911 | for_each_possible_cpu(i) |
1da177e4 | 2912 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2913 | |
2914 | /* | |
1da177e4 LT |
2915 | * Since we read the counters lockless, it might be slightly |
2916 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2917 | */ |
1da177e4 LT |
2918 | if (unlikely((long)sum < 0)) |
2919 | sum = 0; | |
f711f609 | 2920 | |
1da177e4 | 2921 | return sum; |
f711f609 | 2922 | } |
f711f609 | 2923 | |
1da177e4 | 2924 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2925 | { |
cc94abfc SR |
2926 | int i; |
2927 | unsigned long long sum = 0; | |
46cb4b7c | 2928 | |
0a945022 | 2929 | for_each_possible_cpu(i) |
1da177e4 | 2930 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2931 | |
1da177e4 LT |
2932 | return sum; |
2933 | } | |
483b4ee6 | 2934 | |
1da177e4 LT |
2935 | unsigned long nr_iowait(void) |
2936 | { | |
2937 | unsigned long i, sum = 0; | |
483b4ee6 | 2938 | |
0a945022 | 2939 | for_each_possible_cpu(i) |
1da177e4 | 2940 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2941 | |
1da177e4 LT |
2942 | return sum; |
2943 | } | |
483b4ee6 | 2944 | |
69d25870 AV |
2945 | unsigned long nr_iowait_cpu(void) |
2946 | { | |
2947 | struct rq *this = this_rq(); | |
2948 | return atomic_read(&this->nr_iowait); | |
2949 | } | |
46cb4b7c | 2950 | |
69d25870 AV |
2951 | unsigned long this_cpu_load(void) |
2952 | { | |
2953 | struct rq *this = this_rq(); | |
2954 | return this->cpu_load[0]; | |
2955 | } | |
e790fb0b | 2956 | |
46cb4b7c | 2957 | |
dce48a84 TG |
2958 | /* Variables and functions for calc_load */ |
2959 | static atomic_long_t calc_load_tasks; | |
2960 | static unsigned long calc_load_update; | |
2961 | unsigned long avenrun[3]; | |
2962 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2963 | |
2d02494f TG |
2964 | /** |
2965 | * get_avenrun - get the load average array | |
2966 | * @loads: pointer to dest load array | |
2967 | * @offset: offset to add | |
2968 | * @shift: shift count to shift the result left | |
2969 | * | |
2970 | * These values are estimates at best, so no need for locking. | |
2971 | */ | |
2972 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2973 | { | |
2974 | loads[0] = (avenrun[0] + offset) << shift; | |
2975 | loads[1] = (avenrun[1] + offset) << shift; | |
2976 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2977 | } |
46cb4b7c | 2978 | |
dce48a84 TG |
2979 | static unsigned long |
2980 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2981 | { |
dce48a84 TG |
2982 | load *= exp; |
2983 | load += active * (FIXED_1 - exp); | |
2984 | return load >> FSHIFT; | |
2985 | } | |
46cb4b7c SS |
2986 | |
2987 | /* | |
dce48a84 TG |
2988 | * calc_load - update the avenrun load estimates 10 ticks after the |
2989 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2990 | */ |
dce48a84 | 2991 | void calc_global_load(void) |
7835b98b | 2992 | { |
dce48a84 TG |
2993 | unsigned long upd = calc_load_update + 10; |
2994 | long active; | |
1da177e4 | 2995 | |
dce48a84 TG |
2996 | if (time_before(jiffies, upd)) |
2997 | return; | |
1da177e4 | 2998 | |
dce48a84 TG |
2999 | active = atomic_long_read(&calc_load_tasks); |
3000 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3001 | |
dce48a84 TG |
3002 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3003 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3004 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3005 | |
dce48a84 TG |
3006 | calc_load_update += LOAD_FREQ; |
3007 | } | |
1da177e4 | 3008 | |
dce48a84 TG |
3009 | /* |
3010 | * Either called from update_cpu_load() or from a cpu going idle | |
3011 | */ | |
3012 | static void calc_load_account_active(struct rq *this_rq) | |
3013 | { | |
3014 | long nr_active, delta; | |
08c183f3 | 3015 | |
dce48a84 TG |
3016 | nr_active = this_rq->nr_running; |
3017 | nr_active += (long) this_rq->nr_uninterruptible; | |
783609c6 | 3018 | |
dce48a84 TG |
3019 | if (nr_active != this_rq->calc_load_active) { |
3020 | delta = nr_active - this_rq->calc_load_active; | |
3021 | this_rq->calc_load_active = nr_active; | |
3022 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 | 3023 | } |
46cb4b7c SS |
3024 | } |
3025 | ||
3026 | /* | |
dd41f596 IM |
3027 | * Update rq->cpu_load[] statistics. This function is usually called every |
3028 | * scheduler tick (TICK_NSEC). | |
46cb4b7c | 3029 | */ |
dd41f596 | 3030 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3031 | { |
495eca49 | 3032 | unsigned long this_load = this_rq->load.weight; |
dd41f596 | 3033 | int i, scale; |
46cb4b7c | 3034 | |
dd41f596 | 3035 | this_rq->nr_load_updates++; |
46cb4b7c | 3036 | |
dd41f596 IM |
3037 | /* Update our load: */ |
3038 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3039 | unsigned long old_load, new_load; | |
7d1e6a9b | 3040 | |
dd41f596 | 3041 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3042 | |
dd41f596 IM |
3043 | old_load = this_rq->cpu_load[i]; |
3044 | new_load = this_load; | |
a25707f3 IM |
3045 | /* |
3046 | * Round up the averaging division if load is increasing. This | |
3047 | * prevents us from getting stuck on 9 if the load is 10, for | |
3048 | * example. | |
3049 | */ | |
3050 | if (new_load > old_load) | |
3051 | new_load += scale-1; | |
dd41f596 IM |
3052 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3053 | } | |
46cb4b7c | 3054 | |
dce48a84 TG |
3055 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { |
3056 | this_rq->calc_load_update += LOAD_FREQ; | |
3057 | calc_load_account_active(this_rq); | |
46cb4b7c | 3058 | } |
46cb4b7c SS |
3059 | } |
3060 | ||
dd41f596 | 3061 | #ifdef CONFIG_SMP |
8a0be9ef | 3062 | |
46cb4b7c | 3063 | /* |
38022906 PZ |
3064 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3065 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3066 | */ |
38022906 | 3067 | void sched_exec(void) |
46cb4b7c | 3068 | { |
38022906 | 3069 | struct task_struct *p = current; |
70b97a7f | 3070 | struct migration_req req; |
38022906 | 3071 | int dest_cpu, this_cpu; |
1da177e4 | 3072 | unsigned long flags; |
70b97a7f | 3073 | struct rq *rq; |
46cb4b7c | 3074 | |
38022906 | 3075 | this_cpu = get_cpu(); |
30da688e | 3076 | dest_cpu = p->sched_class->select_task_rq(p, SD_BALANCE_EXEC, 0); |
38022906 PZ |
3077 | if (dest_cpu == this_cpu) { |
3078 | put_cpu(); | |
3079 | return; | |
46cb4b7c SS |
3080 | } |
3081 | ||
1da177e4 | 3082 | rq = task_rq_lock(p, &flags); |
38022906 | 3083 | put_cpu(); |
46cb4b7c | 3084 | /* |
38022906 | 3085 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3086 | */ |
30da688e ON |
3087 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
3088 | likely(cpu_active(dest_cpu)) && | |
3089 | migrate_task(p, dest_cpu, &req)) { | |
1da177e4 LT |
3090 | /* Need to wait for migration thread (might exit: take ref). */ |
3091 | struct task_struct *mt = rq->migration_thread; | |
dd41f596 | 3092 | |
1da177e4 LT |
3093 | get_task_struct(mt); |
3094 | task_rq_unlock(rq, &flags); | |
3095 | wake_up_process(mt); | |
3096 | put_task_struct(mt); | |
3097 | wait_for_completion(&req.done); | |
dd41f596 | 3098 | |
1da177e4 LT |
3099 | return; |
3100 | } | |
1da177e4 | 3101 | task_rq_unlock(rq, &flags); |
1da177e4 | 3102 | } |
dd41f596 | 3103 | |
1da177e4 LT |
3104 | #endif |
3105 | ||
1da177e4 LT |
3106 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3107 | ||
3108 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3109 | ||
3110 | /* | |
c5f8d995 | 3111 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3112 | * @p in case that task is currently running. |
c5f8d995 HS |
3113 | * |
3114 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3115 | */ |
c5f8d995 HS |
3116 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3117 | { | |
3118 | u64 ns = 0; | |
3119 | ||
3120 | if (task_current(rq, p)) { | |
3121 | update_rq_clock(rq); | |
3122 | ns = rq->clock - p->se.exec_start; | |
3123 | if ((s64)ns < 0) | |
3124 | ns = 0; | |
3125 | } | |
3126 | ||
3127 | return ns; | |
3128 | } | |
3129 | ||
bb34d92f | 3130 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3131 | { |
1da177e4 | 3132 | unsigned long flags; |
41b86e9c | 3133 | struct rq *rq; |
bb34d92f | 3134 | u64 ns = 0; |
48f24c4d | 3135 | |
41b86e9c | 3136 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3137 | ns = do_task_delta_exec(p, rq); |
3138 | task_rq_unlock(rq, &flags); | |
1508487e | 3139 | |
c5f8d995 HS |
3140 | return ns; |
3141 | } | |
f06febc9 | 3142 | |
c5f8d995 HS |
3143 | /* |
3144 | * Return accounted runtime for the task. | |
3145 | * In case the task is currently running, return the runtime plus current's | |
3146 | * pending runtime that have not been accounted yet. | |
3147 | */ | |
3148 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3149 | { | |
3150 | unsigned long flags; | |
3151 | struct rq *rq; | |
3152 | u64 ns = 0; | |
3153 | ||
3154 | rq = task_rq_lock(p, &flags); | |
3155 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3156 | task_rq_unlock(rq, &flags); | |
3157 | ||
3158 | return ns; | |
3159 | } | |
48f24c4d | 3160 | |
c5f8d995 HS |
3161 | /* |
3162 | * Return sum_exec_runtime for the thread group. | |
3163 | * In case the task is currently running, return the sum plus current's | |
3164 | * pending runtime that have not been accounted yet. | |
3165 | * | |
3166 | * Note that the thread group might have other running tasks as well, | |
3167 | * so the return value not includes other pending runtime that other | |
3168 | * running tasks might have. | |
3169 | */ | |
3170 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3171 | { | |
3172 | struct task_cputime totals; | |
3173 | unsigned long flags; | |
3174 | struct rq *rq; | |
3175 | u64 ns; | |
3176 | ||
3177 | rq = task_rq_lock(p, &flags); | |
3178 | thread_group_cputime(p, &totals); | |
3179 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3180 | task_rq_unlock(rq, &flags); |
48f24c4d | 3181 | |
1da177e4 LT |
3182 | return ns; |
3183 | } | |
3184 | ||
1da177e4 LT |
3185 | /* |
3186 | * Account user cpu time to a process. | |
3187 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3188 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3189 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3190 | */ |
457533a7 MS |
3191 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3192 | cputime_t cputime_scaled) | |
1da177e4 LT |
3193 | { |
3194 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3195 | cputime64_t tmp; | |
3196 | ||
457533a7 | 3197 | /* Add user time to process. */ |
1da177e4 | 3198 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3199 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3200 | account_group_user_time(p, cputime); |
1da177e4 LT |
3201 | |
3202 | /* Add user time to cpustat. */ | |
3203 | tmp = cputime_to_cputime64(cputime); | |
3204 | if (TASK_NICE(p) > 0) | |
3205 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3206 | else | |
3207 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3208 | |
3209 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3210 | /* Account for user time used */ |
3211 | acct_update_integrals(p); | |
1da177e4 LT |
3212 | } |
3213 | ||
94886b84 LV |
3214 | /* |
3215 | * Account guest cpu time to a process. | |
3216 | * @p: the process that the cpu time gets accounted to | |
3217 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3218 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3219 | */ |
457533a7 MS |
3220 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3221 | cputime_t cputime_scaled) | |
94886b84 LV |
3222 | { |
3223 | cputime64_t tmp; | |
3224 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3225 | ||
3226 | tmp = cputime_to_cputime64(cputime); | |
3227 | ||
457533a7 | 3228 | /* Add guest time to process. */ |
94886b84 | 3229 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3230 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3231 | account_group_user_time(p, cputime); |
94886b84 LV |
3232 | p->gtime = cputime_add(p->gtime, cputime); |
3233 | ||
457533a7 | 3234 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3235 | if (TASK_NICE(p) > 0) { |
3236 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3237 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3238 | } else { | |
3239 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3240 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3241 | } | |
94886b84 LV |
3242 | } |
3243 | ||
1da177e4 LT |
3244 | /* |
3245 | * Account system cpu time to a process. | |
3246 | * @p: the process that the cpu time gets accounted to | |
3247 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3248 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3249 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3250 | */ |
3251 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3252 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3253 | { |
3254 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3255 | cputime64_t tmp; |
3256 | ||
983ed7a6 | 3257 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3258 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3259 | return; |
3260 | } | |
94886b84 | 3261 | |
457533a7 | 3262 | /* Add system time to process. */ |
1da177e4 | 3263 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3264 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3265 | account_group_system_time(p, cputime); |
1da177e4 LT |
3266 | |
3267 | /* Add system time to cpustat. */ | |
3268 | tmp = cputime_to_cputime64(cputime); | |
3269 | if (hardirq_count() - hardirq_offset) | |
3270 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3271 | else if (softirq_count()) | |
3272 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3273 | else |
79741dd3 MS |
3274 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3275 | ||
ef12fefa BR |
3276 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3277 | ||
1da177e4 LT |
3278 | /* Account for system time used */ |
3279 | acct_update_integrals(p); | |
1da177e4 LT |
3280 | } |
3281 | ||
c66f08be | 3282 | /* |
1da177e4 | 3283 | * Account for involuntary wait time. |
1da177e4 | 3284 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3285 | */ |
79741dd3 | 3286 | void account_steal_time(cputime_t cputime) |
c66f08be | 3287 | { |
79741dd3 MS |
3288 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3289 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3290 | ||
3291 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3292 | } |
3293 | ||
1da177e4 | 3294 | /* |
79741dd3 MS |
3295 | * Account for idle time. |
3296 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3297 | */ |
79741dd3 | 3298 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3299 | { |
3300 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3301 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3302 | struct rq *rq = this_rq(); |
1da177e4 | 3303 | |
79741dd3 MS |
3304 | if (atomic_read(&rq->nr_iowait) > 0) |
3305 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3306 | else | |
3307 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3308 | } |
3309 | ||
79741dd3 MS |
3310 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3311 | ||
3312 | /* | |
3313 | * Account a single tick of cpu time. | |
3314 | * @p: the process that the cpu time gets accounted to | |
3315 | * @user_tick: indicates if the tick is a user or a system tick | |
3316 | */ | |
3317 | void account_process_tick(struct task_struct *p, int user_tick) | |
3318 | { | |
a42548a1 | 3319 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3320 | struct rq *rq = this_rq(); |
3321 | ||
3322 | if (user_tick) | |
a42548a1 | 3323 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3324 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3325 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3326 | one_jiffy_scaled); |
3327 | else | |
a42548a1 | 3328 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3329 | } |
3330 | ||
3331 | /* | |
3332 | * Account multiple ticks of steal time. | |
3333 | * @p: the process from which the cpu time has been stolen | |
3334 | * @ticks: number of stolen ticks | |
3335 | */ | |
3336 | void account_steal_ticks(unsigned long ticks) | |
3337 | { | |
3338 | account_steal_time(jiffies_to_cputime(ticks)); | |
3339 | } | |
3340 | ||
3341 | /* | |
3342 | * Account multiple ticks of idle time. | |
3343 | * @ticks: number of stolen ticks | |
3344 | */ | |
3345 | void account_idle_ticks(unsigned long ticks) | |
3346 | { | |
3347 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3348 | } |
3349 | ||
79741dd3 MS |
3350 | #endif |
3351 | ||
49048622 BS |
3352 | /* |
3353 | * Use precise platform statistics if available: | |
3354 | */ | |
3355 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3356 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3357 | { |
d99ca3b9 HS |
3358 | *ut = p->utime; |
3359 | *st = p->stime; | |
49048622 BS |
3360 | } |
3361 | ||
0cf55e1e | 3362 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3363 | { |
0cf55e1e HS |
3364 | struct task_cputime cputime; |
3365 | ||
3366 | thread_group_cputime(p, &cputime); | |
3367 | ||
3368 | *ut = cputime.utime; | |
3369 | *st = cputime.stime; | |
49048622 BS |
3370 | } |
3371 | #else | |
761b1d26 HS |
3372 | |
3373 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3374 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3375 | #endif |
3376 | ||
d180c5bc | 3377 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3378 | { |
d99ca3b9 | 3379 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3380 | |
3381 | /* | |
3382 | * Use CFS's precise accounting: | |
3383 | */ | |
d180c5bc | 3384 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3385 | |
3386 | if (total) { | |
d180c5bc HS |
3387 | u64 temp; |
3388 | ||
3389 | temp = (u64)(rtime * utime); | |
49048622 | 3390 | do_div(temp, total); |
d180c5bc HS |
3391 | utime = (cputime_t)temp; |
3392 | } else | |
3393 | utime = rtime; | |
49048622 | 3394 | |
d180c5bc HS |
3395 | /* |
3396 | * Compare with previous values, to keep monotonicity: | |
3397 | */ | |
761b1d26 | 3398 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3399 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3400 | |
d99ca3b9 HS |
3401 | *ut = p->prev_utime; |
3402 | *st = p->prev_stime; | |
49048622 BS |
3403 | } |
3404 | ||
0cf55e1e HS |
3405 | /* |
3406 | * Must be called with siglock held. | |
3407 | */ | |
3408 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3409 | { |
0cf55e1e HS |
3410 | struct signal_struct *sig = p->signal; |
3411 | struct task_cputime cputime; | |
3412 | cputime_t rtime, utime, total; | |
49048622 | 3413 | |
0cf55e1e | 3414 | thread_group_cputime(p, &cputime); |
49048622 | 3415 | |
0cf55e1e HS |
3416 | total = cputime_add(cputime.utime, cputime.stime); |
3417 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3418 | |
0cf55e1e HS |
3419 | if (total) { |
3420 | u64 temp; | |
49048622 | 3421 | |
0cf55e1e HS |
3422 | temp = (u64)(rtime * cputime.utime); |
3423 | do_div(temp, total); | |
3424 | utime = (cputime_t)temp; | |
3425 | } else | |
3426 | utime = rtime; | |
3427 | ||
3428 | sig->prev_utime = max(sig->prev_utime, utime); | |
3429 | sig->prev_stime = max(sig->prev_stime, | |
3430 | cputime_sub(rtime, sig->prev_utime)); | |
3431 | ||
3432 | *ut = sig->prev_utime; | |
3433 | *st = sig->prev_stime; | |
49048622 | 3434 | } |
49048622 | 3435 | #endif |
49048622 | 3436 | |
7835b98b CL |
3437 | /* |
3438 | * This function gets called by the timer code, with HZ frequency. | |
3439 | * We call it with interrupts disabled. | |
3440 | * | |
3441 | * It also gets called by the fork code, when changing the parent's | |
3442 | * timeslices. | |
3443 | */ | |
3444 | void scheduler_tick(void) | |
3445 | { | |
7835b98b CL |
3446 | int cpu = smp_processor_id(); |
3447 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3448 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3449 | |
3450 | sched_clock_tick(); | |
dd41f596 | 3451 | |
05fa785c | 3452 | raw_spin_lock(&rq->lock); |
3e51f33f | 3453 | update_rq_clock(rq); |
f1a438d8 | 3454 | update_cpu_load(rq); |
fa85ae24 | 3455 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3456 | raw_spin_unlock(&rq->lock); |
7835b98b | 3457 | |
49f47433 | 3458 | perf_event_task_tick(curr); |
e220d2dc | 3459 | |
e418e1c2 | 3460 | #ifdef CONFIG_SMP |
dd41f596 IM |
3461 | rq->idle_at_tick = idle_cpu(cpu); |
3462 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3463 | #endif |
1da177e4 LT |
3464 | } |
3465 | ||
132380a0 | 3466 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3467 | { |
3468 | if (in_lock_functions(addr)) { | |
3469 | addr = CALLER_ADDR2; | |
3470 | if (in_lock_functions(addr)) | |
3471 | addr = CALLER_ADDR3; | |
3472 | } | |
3473 | return addr; | |
3474 | } | |
1da177e4 | 3475 | |
7e49fcce SR |
3476 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3477 | defined(CONFIG_PREEMPT_TRACER)) | |
3478 | ||
43627582 | 3479 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3480 | { |
6cd8a4bb | 3481 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3482 | /* |
3483 | * Underflow? | |
3484 | */ | |
9a11b49a IM |
3485 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3486 | return; | |
6cd8a4bb | 3487 | #endif |
1da177e4 | 3488 | preempt_count() += val; |
6cd8a4bb | 3489 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3490 | /* |
3491 | * Spinlock count overflowing soon? | |
3492 | */ | |
33859f7f MOS |
3493 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3494 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3495 | #endif |
3496 | if (preempt_count() == val) | |
3497 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3498 | } |
3499 | EXPORT_SYMBOL(add_preempt_count); | |
3500 | ||
43627582 | 3501 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3502 | { |
6cd8a4bb | 3503 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3504 | /* |
3505 | * Underflow? | |
3506 | */ | |
01e3eb82 | 3507 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3508 | return; |
1da177e4 LT |
3509 | /* |
3510 | * Is the spinlock portion underflowing? | |
3511 | */ | |
9a11b49a IM |
3512 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3513 | !(preempt_count() & PREEMPT_MASK))) | |
3514 | return; | |
6cd8a4bb | 3515 | #endif |
9a11b49a | 3516 | |
6cd8a4bb SR |
3517 | if (preempt_count() == val) |
3518 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3519 | preempt_count() -= val; |
3520 | } | |
3521 | EXPORT_SYMBOL(sub_preempt_count); | |
3522 | ||
3523 | #endif | |
3524 | ||
3525 | /* | |
dd41f596 | 3526 | * Print scheduling while atomic bug: |
1da177e4 | 3527 | */ |
dd41f596 | 3528 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3529 | { |
838225b4 SS |
3530 | struct pt_regs *regs = get_irq_regs(); |
3531 | ||
3df0fc5b PZ |
3532 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3533 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3534 | |
dd41f596 | 3535 | debug_show_held_locks(prev); |
e21f5b15 | 3536 | print_modules(); |
dd41f596 IM |
3537 | if (irqs_disabled()) |
3538 | print_irqtrace_events(prev); | |
838225b4 SS |
3539 | |
3540 | if (regs) | |
3541 | show_regs(regs); | |
3542 | else | |
3543 | dump_stack(); | |
dd41f596 | 3544 | } |
1da177e4 | 3545 | |
dd41f596 IM |
3546 | /* |
3547 | * Various schedule()-time debugging checks and statistics: | |
3548 | */ | |
3549 | static inline void schedule_debug(struct task_struct *prev) | |
3550 | { | |
1da177e4 | 3551 | /* |
41a2d6cf | 3552 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3553 | * schedule() atomically, we ignore that path for now. |
3554 | * Otherwise, whine if we are scheduling when we should not be. | |
3555 | */ | |
3f33a7ce | 3556 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3557 | __schedule_bug(prev); |
3558 | ||
1da177e4 LT |
3559 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3560 | ||
2d72376b | 3561 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3562 | #ifdef CONFIG_SCHEDSTATS |
3563 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3564 | schedstat_inc(this_rq(), bkl_count); |
3565 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3566 | } |
3567 | #endif | |
dd41f596 IM |
3568 | } |
3569 | ||
6cecd084 | 3570 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3571 | { |
a64692a3 MG |
3572 | if (prev->se.on_rq) |
3573 | update_rq_clock(rq); | |
3574 | rq->skip_clock_update = 0; | |
6cecd084 | 3575 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3576 | } |
3577 | ||
dd41f596 IM |
3578 | /* |
3579 | * Pick up the highest-prio task: | |
3580 | */ | |
3581 | static inline struct task_struct * | |
b67802ea | 3582 | pick_next_task(struct rq *rq) |
dd41f596 | 3583 | { |
5522d5d5 | 3584 | const struct sched_class *class; |
dd41f596 | 3585 | struct task_struct *p; |
1da177e4 LT |
3586 | |
3587 | /* | |
dd41f596 IM |
3588 | * Optimization: we know that if all tasks are in |
3589 | * the fair class we can call that function directly: | |
1da177e4 | 3590 | */ |
dd41f596 | 3591 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3592 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3593 | if (likely(p)) |
3594 | return p; | |
1da177e4 LT |
3595 | } |
3596 | ||
dd41f596 IM |
3597 | class = sched_class_highest; |
3598 | for ( ; ; ) { | |
fb8d4724 | 3599 | p = class->pick_next_task(rq); |
dd41f596 IM |
3600 | if (p) |
3601 | return p; | |
3602 | /* | |
3603 | * Will never be NULL as the idle class always | |
3604 | * returns a non-NULL p: | |
3605 | */ | |
3606 | class = class->next; | |
3607 | } | |
3608 | } | |
1da177e4 | 3609 | |
dd41f596 IM |
3610 | /* |
3611 | * schedule() is the main scheduler function. | |
3612 | */ | |
ff743345 | 3613 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3614 | { |
3615 | struct task_struct *prev, *next; | |
67ca7bde | 3616 | unsigned long *switch_count; |
dd41f596 | 3617 | struct rq *rq; |
31656519 | 3618 | int cpu; |
dd41f596 | 3619 | |
ff743345 PZ |
3620 | need_resched: |
3621 | preempt_disable(); | |
dd41f596 IM |
3622 | cpu = smp_processor_id(); |
3623 | rq = cpu_rq(cpu); | |
d6714c22 | 3624 | rcu_sched_qs(cpu); |
dd41f596 IM |
3625 | prev = rq->curr; |
3626 | switch_count = &prev->nivcsw; | |
3627 | ||
3628 | release_kernel_lock(prev); | |
3629 | need_resched_nonpreemptible: | |
3630 | ||
3631 | schedule_debug(prev); | |
1da177e4 | 3632 | |
31656519 | 3633 | if (sched_feat(HRTICK)) |
f333fdc9 | 3634 | hrtick_clear(rq); |
8f4d37ec | 3635 | |
05fa785c | 3636 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3637 | clear_tsk_need_resched(prev); |
1da177e4 | 3638 | |
1da177e4 | 3639 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 3640 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 3641 | prev->state = TASK_RUNNING; |
16882c1e | 3642 | else |
2e1cb74a | 3643 | deactivate_task(rq, prev, 1); |
dd41f596 | 3644 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3645 | } |
3646 | ||
3f029d3c | 3647 | pre_schedule(rq, prev); |
f65eda4f | 3648 | |
dd41f596 | 3649 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3650 | idle_balance(cpu, rq); |
1da177e4 | 3651 | |
df1c99d4 | 3652 | put_prev_task(rq, prev); |
b67802ea | 3653 | next = pick_next_task(rq); |
1da177e4 | 3654 | |
1da177e4 | 3655 | if (likely(prev != next)) { |
673a90a1 | 3656 | sched_info_switch(prev, next); |
49f47433 | 3657 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3658 | |
1da177e4 LT |
3659 | rq->nr_switches++; |
3660 | rq->curr = next; | |
3661 | ++*switch_count; | |
3662 | ||
dd41f596 | 3663 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3664 | /* |
3665 | * the context switch might have flipped the stack from under | |
3666 | * us, hence refresh the local variables. | |
3667 | */ | |
3668 | cpu = smp_processor_id(); | |
3669 | rq = cpu_rq(cpu); | |
1da177e4 | 3670 | } else |
05fa785c | 3671 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3672 | |
3f029d3c | 3673 | post_schedule(rq); |
1da177e4 | 3674 | |
6d558c3a YZ |
3675 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3676 | prev = rq->curr; | |
3677 | switch_count = &prev->nivcsw; | |
1da177e4 | 3678 | goto need_resched_nonpreemptible; |
6d558c3a | 3679 | } |
8f4d37ec | 3680 | |
1da177e4 | 3681 | preempt_enable_no_resched(); |
ff743345 | 3682 | if (need_resched()) |
1da177e4 LT |
3683 | goto need_resched; |
3684 | } | |
1da177e4 LT |
3685 | EXPORT_SYMBOL(schedule); |
3686 | ||
c08f7829 | 3687 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3688 | /* |
3689 | * Look out! "owner" is an entirely speculative pointer | |
3690 | * access and not reliable. | |
3691 | */ | |
3692 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3693 | { | |
3694 | unsigned int cpu; | |
3695 | struct rq *rq; | |
3696 | ||
3697 | if (!sched_feat(OWNER_SPIN)) | |
3698 | return 0; | |
3699 | ||
3700 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3701 | /* | |
3702 | * Need to access the cpu field knowing that | |
3703 | * DEBUG_PAGEALLOC could have unmapped it if | |
3704 | * the mutex owner just released it and exited. | |
3705 | */ | |
3706 | if (probe_kernel_address(&owner->cpu, cpu)) | |
3707 | goto out; | |
3708 | #else | |
3709 | cpu = owner->cpu; | |
3710 | #endif | |
3711 | ||
3712 | /* | |
3713 | * Even if the access succeeded (likely case), | |
3714 | * the cpu field may no longer be valid. | |
3715 | */ | |
3716 | if (cpu >= nr_cpumask_bits) | |
3717 | goto out; | |
3718 | ||
3719 | /* | |
3720 | * We need to validate that we can do a | |
3721 | * get_cpu() and that we have the percpu area. | |
3722 | */ | |
3723 | if (!cpu_online(cpu)) | |
3724 | goto out; | |
3725 | ||
3726 | rq = cpu_rq(cpu); | |
3727 | ||
3728 | for (;;) { | |
3729 | /* | |
3730 | * Owner changed, break to re-assess state. | |
3731 | */ | |
3732 | if (lock->owner != owner) | |
3733 | break; | |
3734 | ||
3735 | /* | |
3736 | * Is that owner really running on that cpu? | |
3737 | */ | |
3738 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3739 | return 0; | |
3740 | ||
3741 | cpu_relax(); | |
3742 | } | |
3743 | out: | |
3744 | return 1; | |
3745 | } | |
3746 | #endif | |
3747 | ||
1da177e4 LT |
3748 | #ifdef CONFIG_PREEMPT |
3749 | /* | |
2ed6e34f | 3750 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3751 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3752 | * occur there and call schedule directly. |
3753 | */ | |
3754 | asmlinkage void __sched preempt_schedule(void) | |
3755 | { | |
3756 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3757 | |
1da177e4 LT |
3758 | /* |
3759 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3760 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3761 | */ |
beed33a8 | 3762 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3763 | return; |
3764 | ||
3a5c359a AK |
3765 | do { |
3766 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3767 | schedule(); |
3a5c359a | 3768 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3769 | |
3a5c359a AK |
3770 | /* |
3771 | * Check again in case we missed a preemption opportunity | |
3772 | * between schedule and now. | |
3773 | */ | |
3774 | barrier(); | |
5ed0cec0 | 3775 | } while (need_resched()); |
1da177e4 | 3776 | } |
1da177e4 LT |
3777 | EXPORT_SYMBOL(preempt_schedule); |
3778 | ||
3779 | /* | |
2ed6e34f | 3780 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3781 | * off of irq context. |
3782 | * Note, that this is called and return with irqs disabled. This will | |
3783 | * protect us against recursive calling from irq. | |
3784 | */ | |
3785 | asmlinkage void __sched preempt_schedule_irq(void) | |
3786 | { | |
3787 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3788 | |
2ed6e34f | 3789 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3790 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3791 | ||
3a5c359a AK |
3792 | do { |
3793 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3794 | local_irq_enable(); |
3795 | schedule(); | |
3796 | local_irq_disable(); | |
3a5c359a | 3797 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3798 | |
3a5c359a AK |
3799 | /* |
3800 | * Check again in case we missed a preemption opportunity | |
3801 | * between schedule and now. | |
3802 | */ | |
3803 | barrier(); | |
5ed0cec0 | 3804 | } while (need_resched()); |
1da177e4 LT |
3805 | } |
3806 | ||
3807 | #endif /* CONFIG_PREEMPT */ | |
3808 | ||
63859d4f | 3809 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3810 | void *key) |
1da177e4 | 3811 | { |
63859d4f | 3812 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3813 | } |
1da177e4 LT |
3814 | EXPORT_SYMBOL(default_wake_function); |
3815 | ||
3816 | /* | |
41a2d6cf IM |
3817 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3818 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3819 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3820 | * | |
3821 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3822 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3823 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3824 | */ | |
78ddb08f | 3825 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3826 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3827 | { |
2e45874c | 3828 | wait_queue_t *curr, *next; |
1da177e4 | 3829 | |
2e45874c | 3830 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3831 | unsigned flags = curr->flags; |
3832 | ||
63859d4f | 3833 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3834 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3835 | break; |
3836 | } | |
3837 | } | |
3838 | ||
3839 | /** | |
3840 | * __wake_up - wake up threads blocked on a waitqueue. | |
3841 | * @q: the waitqueue | |
3842 | * @mode: which threads | |
3843 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3844 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3845 | * |
3846 | * It may be assumed that this function implies a write memory barrier before | |
3847 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3848 | */ |
7ad5b3a5 | 3849 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3850 | int nr_exclusive, void *key) |
1da177e4 LT |
3851 | { |
3852 | unsigned long flags; | |
3853 | ||
3854 | spin_lock_irqsave(&q->lock, flags); | |
3855 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3856 | spin_unlock_irqrestore(&q->lock, flags); | |
3857 | } | |
1da177e4 LT |
3858 | EXPORT_SYMBOL(__wake_up); |
3859 | ||
3860 | /* | |
3861 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3862 | */ | |
7ad5b3a5 | 3863 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3864 | { |
3865 | __wake_up_common(q, mode, 1, 0, NULL); | |
3866 | } | |
3867 | ||
4ede816a DL |
3868 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3869 | { | |
3870 | __wake_up_common(q, mode, 1, 0, key); | |
3871 | } | |
3872 | ||
1da177e4 | 3873 | /** |
4ede816a | 3874 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3875 | * @q: the waitqueue |
3876 | * @mode: which threads | |
3877 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3878 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3879 | * |
3880 | * The sync wakeup differs that the waker knows that it will schedule | |
3881 | * away soon, so while the target thread will be woken up, it will not | |
3882 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3883 | * with each other. This can prevent needless bouncing between CPUs. | |
3884 | * | |
3885 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3886 | * |
3887 | * It may be assumed that this function implies a write memory barrier before | |
3888 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3889 | */ |
4ede816a DL |
3890 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3891 | int nr_exclusive, void *key) | |
1da177e4 LT |
3892 | { |
3893 | unsigned long flags; | |
7d478721 | 3894 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3895 | |
3896 | if (unlikely(!q)) | |
3897 | return; | |
3898 | ||
3899 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3900 | wake_flags = 0; |
1da177e4 LT |
3901 | |
3902 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3903 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3904 | spin_unlock_irqrestore(&q->lock, flags); |
3905 | } | |
4ede816a DL |
3906 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3907 | ||
3908 | /* | |
3909 | * __wake_up_sync - see __wake_up_sync_key() | |
3910 | */ | |
3911 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3912 | { | |
3913 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3914 | } | |
1da177e4 LT |
3915 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3916 | ||
65eb3dc6 KD |
3917 | /** |
3918 | * complete: - signals a single thread waiting on this completion | |
3919 | * @x: holds the state of this particular completion | |
3920 | * | |
3921 | * This will wake up a single thread waiting on this completion. Threads will be | |
3922 | * awakened in the same order in which they were queued. | |
3923 | * | |
3924 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3925 | * |
3926 | * It may be assumed that this function implies a write memory barrier before | |
3927 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3928 | */ |
b15136e9 | 3929 | void complete(struct completion *x) |
1da177e4 LT |
3930 | { |
3931 | unsigned long flags; | |
3932 | ||
3933 | spin_lock_irqsave(&x->wait.lock, flags); | |
3934 | x->done++; | |
d9514f6c | 3935 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3936 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3937 | } | |
3938 | EXPORT_SYMBOL(complete); | |
3939 | ||
65eb3dc6 KD |
3940 | /** |
3941 | * complete_all: - signals all threads waiting on this completion | |
3942 | * @x: holds the state of this particular completion | |
3943 | * | |
3944 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3945 | * |
3946 | * It may be assumed that this function implies a write memory barrier before | |
3947 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3948 | */ |
b15136e9 | 3949 | void complete_all(struct completion *x) |
1da177e4 LT |
3950 | { |
3951 | unsigned long flags; | |
3952 | ||
3953 | spin_lock_irqsave(&x->wait.lock, flags); | |
3954 | x->done += UINT_MAX/2; | |
d9514f6c | 3955 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3956 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3957 | } | |
3958 | EXPORT_SYMBOL(complete_all); | |
3959 | ||
8cbbe86d AK |
3960 | static inline long __sched |
3961 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3962 | { |
1da177e4 LT |
3963 | if (!x->done) { |
3964 | DECLARE_WAITQUEUE(wait, current); | |
3965 | ||
3966 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
3967 | __add_wait_queue_tail(&x->wait, &wait); | |
3968 | do { | |
94d3d824 | 3969 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3970 | timeout = -ERESTARTSYS; |
3971 | break; | |
8cbbe86d AK |
3972 | } |
3973 | __set_current_state(state); | |
1da177e4 LT |
3974 | spin_unlock_irq(&x->wait.lock); |
3975 | timeout = schedule_timeout(timeout); | |
3976 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3977 | } while (!x->done && timeout); |
1da177e4 | 3978 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3979 | if (!x->done) |
3980 | return timeout; | |
1da177e4 LT |
3981 | } |
3982 | x->done--; | |
ea71a546 | 3983 | return timeout ?: 1; |
1da177e4 | 3984 | } |
1da177e4 | 3985 | |
8cbbe86d AK |
3986 | static long __sched |
3987 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3988 | { |
1da177e4 LT |
3989 | might_sleep(); |
3990 | ||
3991 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3992 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3993 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3994 | return timeout; |
3995 | } | |
1da177e4 | 3996 | |
65eb3dc6 KD |
3997 | /** |
3998 | * wait_for_completion: - waits for completion of a task | |
3999 | * @x: holds the state of this particular completion | |
4000 | * | |
4001 | * This waits to be signaled for completion of a specific task. It is NOT | |
4002 | * interruptible and there is no timeout. | |
4003 | * | |
4004 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4005 | * and interrupt capability. Also see complete(). | |
4006 | */ | |
b15136e9 | 4007 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4008 | { |
4009 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4010 | } |
8cbbe86d | 4011 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4012 | |
65eb3dc6 KD |
4013 | /** |
4014 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4015 | * @x: holds the state of this particular completion | |
4016 | * @timeout: timeout value in jiffies | |
4017 | * | |
4018 | * This waits for either a completion of a specific task to be signaled or for a | |
4019 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4020 | * interruptible. | |
4021 | */ | |
b15136e9 | 4022 | unsigned long __sched |
8cbbe86d | 4023 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4024 | { |
8cbbe86d | 4025 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4026 | } |
8cbbe86d | 4027 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4028 | |
65eb3dc6 KD |
4029 | /** |
4030 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4031 | * @x: holds the state of this particular completion | |
4032 | * | |
4033 | * This waits for completion of a specific task to be signaled. It is | |
4034 | * interruptible. | |
4035 | */ | |
8cbbe86d | 4036 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4037 | { |
51e97990 AK |
4038 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4039 | if (t == -ERESTARTSYS) | |
4040 | return t; | |
4041 | return 0; | |
0fec171c | 4042 | } |
8cbbe86d | 4043 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4044 | |
65eb3dc6 KD |
4045 | /** |
4046 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4047 | * @x: holds the state of this particular completion | |
4048 | * @timeout: timeout value in jiffies | |
4049 | * | |
4050 | * This waits for either a completion of a specific task to be signaled or for a | |
4051 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4052 | */ | |
b15136e9 | 4053 | unsigned long __sched |
8cbbe86d AK |
4054 | wait_for_completion_interruptible_timeout(struct completion *x, |
4055 | unsigned long timeout) | |
0fec171c | 4056 | { |
8cbbe86d | 4057 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4058 | } |
8cbbe86d | 4059 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4060 | |
65eb3dc6 KD |
4061 | /** |
4062 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4063 | * @x: holds the state of this particular completion | |
4064 | * | |
4065 | * This waits to be signaled for completion of a specific task. It can be | |
4066 | * interrupted by a kill signal. | |
4067 | */ | |
009e577e MW |
4068 | int __sched wait_for_completion_killable(struct completion *x) |
4069 | { | |
4070 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4071 | if (t == -ERESTARTSYS) | |
4072 | return t; | |
4073 | return 0; | |
4074 | } | |
4075 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4076 | ||
be4de352 DC |
4077 | /** |
4078 | * try_wait_for_completion - try to decrement a completion without blocking | |
4079 | * @x: completion structure | |
4080 | * | |
4081 | * Returns: 0 if a decrement cannot be done without blocking | |
4082 | * 1 if a decrement succeeded. | |
4083 | * | |
4084 | * If a completion is being used as a counting completion, | |
4085 | * attempt to decrement the counter without blocking. This | |
4086 | * enables us to avoid waiting if the resource the completion | |
4087 | * is protecting is not available. | |
4088 | */ | |
4089 | bool try_wait_for_completion(struct completion *x) | |
4090 | { | |
7539a3b3 | 4091 | unsigned long flags; |
be4de352 DC |
4092 | int ret = 1; |
4093 | ||
7539a3b3 | 4094 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4095 | if (!x->done) |
4096 | ret = 0; | |
4097 | else | |
4098 | x->done--; | |
7539a3b3 | 4099 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4100 | return ret; |
4101 | } | |
4102 | EXPORT_SYMBOL(try_wait_for_completion); | |
4103 | ||
4104 | /** | |
4105 | * completion_done - Test to see if a completion has any waiters | |
4106 | * @x: completion structure | |
4107 | * | |
4108 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4109 | * 1 if there are no waiters. | |
4110 | * | |
4111 | */ | |
4112 | bool completion_done(struct completion *x) | |
4113 | { | |
7539a3b3 | 4114 | unsigned long flags; |
be4de352 DC |
4115 | int ret = 1; |
4116 | ||
7539a3b3 | 4117 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4118 | if (!x->done) |
4119 | ret = 0; | |
7539a3b3 | 4120 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4121 | return ret; |
4122 | } | |
4123 | EXPORT_SYMBOL(completion_done); | |
4124 | ||
8cbbe86d AK |
4125 | static long __sched |
4126 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4127 | { |
0fec171c IM |
4128 | unsigned long flags; |
4129 | wait_queue_t wait; | |
4130 | ||
4131 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4132 | |
8cbbe86d | 4133 | __set_current_state(state); |
1da177e4 | 4134 | |
8cbbe86d AK |
4135 | spin_lock_irqsave(&q->lock, flags); |
4136 | __add_wait_queue(q, &wait); | |
4137 | spin_unlock(&q->lock); | |
4138 | timeout = schedule_timeout(timeout); | |
4139 | spin_lock_irq(&q->lock); | |
4140 | __remove_wait_queue(q, &wait); | |
4141 | spin_unlock_irqrestore(&q->lock, flags); | |
4142 | ||
4143 | return timeout; | |
4144 | } | |
4145 | ||
4146 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4147 | { | |
4148 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4149 | } |
1da177e4 LT |
4150 | EXPORT_SYMBOL(interruptible_sleep_on); |
4151 | ||
0fec171c | 4152 | long __sched |
95cdf3b7 | 4153 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4154 | { |
8cbbe86d | 4155 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4156 | } |
1da177e4 LT |
4157 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4158 | ||
0fec171c | 4159 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4160 | { |
8cbbe86d | 4161 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4162 | } |
1da177e4 LT |
4163 | EXPORT_SYMBOL(sleep_on); |
4164 | ||
0fec171c | 4165 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4166 | { |
8cbbe86d | 4167 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4168 | } |
1da177e4 LT |
4169 | EXPORT_SYMBOL(sleep_on_timeout); |
4170 | ||
b29739f9 IM |
4171 | #ifdef CONFIG_RT_MUTEXES |
4172 | ||
4173 | /* | |
4174 | * rt_mutex_setprio - set the current priority of a task | |
4175 | * @p: task | |
4176 | * @prio: prio value (kernel-internal form) | |
4177 | * | |
4178 | * This function changes the 'effective' priority of a task. It does | |
4179 | * not touch ->normal_prio like __setscheduler(). | |
4180 | * | |
4181 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4182 | */ | |
36c8b586 | 4183 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4184 | { |
4185 | unsigned long flags; | |
83b699ed | 4186 | int oldprio, on_rq, running; |
70b97a7f | 4187 | struct rq *rq; |
83ab0aa0 | 4188 | const struct sched_class *prev_class; |
b29739f9 IM |
4189 | |
4190 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4191 | ||
4192 | rq = task_rq_lock(p, &flags); | |
4193 | ||
d5f9f942 | 4194 | oldprio = p->prio; |
83ab0aa0 | 4195 | prev_class = p->sched_class; |
dd41f596 | 4196 | on_rq = p->se.on_rq; |
051a1d1a | 4197 | running = task_current(rq, p); |
0e1f3483 | 4198 | if (on_rq) |
69be72c1 | 4199 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4200 | if (running) |
4201 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4202 | |
4203 | if (rt_prio(prio)) | |
4204 | p->sched_class = &rt_sched_class; | |
4205 | else | |
4206 | p->sched_class = &fair_sched_class; | |
4207 | ||
b29739f9 IM |
4208 | p->prio = prio; |
4209 | ||
0e1f3483 HS |
4210 | if (running) |
4211 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4212 | if (on_rq) { |
60db48ca | 4213 | enqueue_task(rq, p, 0, oldprio < prio); |
cb469845 SR |
4214 | |
4215 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4216 | } |
4217 | task_rq_unlock(rq, &flags); | |
4218 | } | |
4219 | ||
4220 | #endif | |
4221 | ||
36c8b586 | 4222 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4223 | { |
dd41f596 | 4224 | int old_prio, delta, on_rq; |
1da177e4 | 4225 | unsigned long flags; |
70b97a7f | 4226 | struct rq *rq; |
1da177e4 LT |
4227 | |
4228 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4229 | return; | |
4230 | /* | |
4231 | * We have to be careful, if called from sys_setpriority(), | |
4232 | * the task might be in the middle of scheduling on another CPU. | |
4233 | */ | |
4234 | rq = task_rq_lock(p, &flags); | |
4235 | /* | |
4236 | * The RT priorities are set via sched_setscheduler(), but we still | |
4237 | * allow the 'normal' nice value to be set - but as expected | |
4238 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4239 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4240 | */ |
e05606d3 | 4241 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4242 | p->static_prio = NICE_TO_PRIO(nice); |
4243 | goto out_unlock; | |
4244 | } | |
dd41f596 | 4245 | on_rq = p->se.on_rq; |
c09595f6 | 4246 | if (on_rq) |
69be72c1 | 4247 | dequeue_task(rq, p, 0); |
1da177e4 | 4248 | |
1da177e4 | 4249 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4250 | set_load_weight(p); |
b29739f9 IM |
4251 | old_prio = p->prio; |
4252 | p->prio = effective_prio(p); | |
4253 | delta = p->prio - old_prio; | |
1da177e4 | 4254 | |
dd41f596 | 4255 | if (on_rq) { |
ea87bb78 | 4256 | enqueue_task(rq, p, 0, false); |
1da177e4 | 4257 | /* |
d5f9f942 AM |
4258 | * If the task increased its priority or is running and |
4259 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4260 | */ |
d5f9f942 | 4261 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4262 | resched_task(rq->curr); |
4263 | } | |
4264 | out_unlock: | |
4265 | task_rq_unlock(rq, &flags); | |
4266 | } | |
1da177e4 LT |
4267 | EXPORT_SYMBOL(set_user_nice); |
4268 | ||
e43379f1 MM |
4269 | /* |
4270 | * can_nice - check if a task can reduce its nice value | |
4271 | * @p: task | |
4272 | * @nice: nice value | |
4273 | */ | |
36c8b586 | 4274 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4275 | { |
024f4747 MM |
4276 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4277 | int nice_rlim = 20 - nice; | |
48f24c4d | 4278 | |
78d7d407 | 4279 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4280 | capable(CAP_SYS_NICE)); |
4281 | } | |
4282 | ||
1da177e4 LT |
4283 | #ifdef __ARCH_WANT_SYS_NICE |
4284 | ||
4285 | /* | |
4286 | * sys_nice - change the priority of the current process. | |
4287 | * @increment: priority increment | |
4288 | * | |
4289 | * sys_setpriority is a more generic, but much slower function that | |
4290 | * does similar things. | |
4291 | */ | |
5add95d4 | 4292 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4293 | { |
48f24c4d | 4294 | long nice, retval; |
1da177e4 LT |
4295 | |
4296 | /* | |
4297 | * Setpriority might change our priority at the same moment. | |
4298 | * We don't have to worry. Conceptually one call occurs first | |
4299 | * and we have a single winner. | |
4300 | */ | |
e43379f1 MM |
4301 | if (increment < -40) |
4302 | increment = -40; | |
1da177e4 LT |
4303 | if (increment > 40) |
4304 | increment = 40; | |
4305 | ||
2b8f836f | 4306 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4307 | if (nice < -20) |
4308 | nice = -20; | |
4309 | if (nice > 19) | |
4310 | nice = 19; | |
4311 | ||
e43379f1 MM |
4312 | if (increment < 0 && !can_nice(current, nice)) |
4313 | return -EPERM; | |
4314 | ||
1da177e4 LT |
4315 | retval = security_task_setnice(current, nice); |
4316 | if (retval) | |
4317 | return retval; | |
4318 | ||
4319 | set_user_nice(current, nice); | |
4320 | return 0; | |
4321 | } | |
4322 | ||
4323 | #endif | |
4324 | ||
4325 | /** | |
4326 | * task_prio - return the priority value of a given task. | |
4327 | * @p: the task in question. | |
4328 | * | |
4329 | * This is the priority value as seen by users in /proc. | |
4330 | * RT tasks are offset by -200. Normal tasks are centered | |
4331 | * around 0, value goes from -16 to +15. | |
4332 | */ | |
36c8b586 | 4333 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4334 | { |
4335 | return p->prio - MAX_RT_PRIO; | |
4336 | } | |
4337 | ||
4338 | /** | |
4339 | * task_nice - return the nice value of a given task. | |
4340 | * @p: the task in question. | |
4341 | */ | |
36c8b586 | 4342 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4343 | { |
4344 | return TASK_NICE(p); | |
4345 | } | |
150d8bed | 4346 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4347 | |
4348 | /** | |
4349 | * idle_cpu - is a given cpu idle currently? | |
4350 | * @cpu: the processor in question. | |
4351 | */ | |
4352 | int idle_cpu(int cpu) | |
4353 | { | |
4354 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4355 | } | |
4356 | ||
1da177e4 LT |
4357 | /** |
4358 | * idle_task - return the idle task for a given cpu. | |
4359 | * @cpu: the processor in question. | |
4360 | */ | |
36c8b586 | 4361 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4362 | { |
4363 | return cpu_rq(cpu)->idle; | |
4364 | } | |
4365 | ||
4366 | /** | |
4367 | * find_process_by_pid - find a process with a matching PID value. | |
4368 | * @pid: the pid in question. | |
4369 | */ | |
a9957449 | 4370 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4371 | { |
228ebcbe | 4372 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4373 | } |
4374 | ||
4375 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4376 | static void |
4377 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4378 | { |
dd41f596 | 4379 | BUG_ON(p->se.on_rq); |
48f24c4d | 4380 | |
1da177e4 LT |
4381 | p->policy = policy; |
4382 | p->rt_priority = prio; | |
b29739f9 IM |
4383 | p->normal_prio = normal_prio(p); |
4384 | /* we are holding p->pi_lock already */ | |
4385 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4386 | if (rt_prio(p->prio)) |
4387 | p->sched_class = &rt_sched_class; | |
4388 | else | |
4389 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4390 | set_load_weight(p); |
1da177e4 LT |
4391 | } |
4392 | ||
c69e8d9c DH |
4393 | /* |
4394 | * check the target process has a UID that matches the current process's | |
4395 | */ | |
4396 | static bool check_same_owner(struct task_struct *p) | |
4397 | { | |
4398 | const struct cred *cred = current_cred(), *pcred; | |
4399 | bool match; | |
4400 | ||
4401 | rcu_read_lock(); | |
4402 | pcred = __task_cred(p); | |
4403 | match = (cred->euid == pcred->euid || | |
4404 | cred->euid == pcred->uid); | |
4405 | rcu_read_unlock(); | |
4406 | return match; | |
4407 | } | |
4408 | ||
961ccddd RR |
4409 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4410 | struct sched_param *param, bool user) | |
1da177e4 | 4411 | { |
83b699ed | 4412 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4413 | unsigned long flags; |
83ab0aa0 | 4414 | const struct sched_class *prev_class; |
70b97a7f | 4415 | struct rq *rq; |
ca94c442 | 4416 | int reset_on_fork; |
1da177e4 | 4417 | |
66e5393a SR |
4418 | /* may grab non-irq protected spin_locks */ |
4419 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4420 | recheck: |
4421 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4422 | if (policy < 0) { |
4423 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4424 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4425 | } else { |
4426 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4427 | policy &= ~SCHED_RESET_ON_FORK; | |
4428 | ||
4429 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4430 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4431 | policy != SCHED_IDLE) | |
4432 | return -EINVAL; | |
4433 | } | |
4434 | ||
1da177e4 LT |
4435 | /* |
4436 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4437 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4438 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4439 | */ |
4440 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4441 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4442 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4443 | return -EINVAL; |
e05606d3 | 4444 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4445 | return -EINVAL; |
4446 | ||
37e4ab3f OC |
4447 | /* |
4448 | * Allow unprivileged RT tasks to decrease priority: | |
4449 | */ | |
961ccddd | 4450 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4451 | if (rt_policy(policy)) { |
8dc3e909 | 4452 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4453 | |
4454 | if (!lock_task_sighand(p, &flags)) | |
4455 | return -ESRCH; | |
78d7d407 | 4456 | rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); |
8dc3e909 ON |
4457 | unlock_task_sighand(p, &flags); |
4458 | ||
4459 | /* can't set/change the rt policy */ | |
4460 | if (policy != p->policy && !rlim_rtprio) | |
4461 | return -EPERM; | |
4462 | ||
4463 | /* can't increase priority */ | |
4464 | if (param->sched_priority > p->rt_priority && | |
4465 | param->sched_priority > rlim_rtprio) | |
4466 | return -EPERM; | |
4467 | } | |
dd41f596 IM |
4468 | /* |
4469 | * Like positive nice levels, dont allow tasks to | |
4470 | * move out of SCHED_IDLE either: | |
4471 | */ | |
4472 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4473 | return -EPERM; | |
5fe1d75f | 4474 | |
37e4ab3f | 4475 | /* can't change other user's priorities */ |
c69e8d9c | 4476 | if (!check_same_owner(p)) |
37e4ab3f | 4477 | return -EPERM; |
ca94c442 LP |
4478 | |
4479 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4480 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4481 | return -EPERM; | |
37e4ab3f | 4482 | } |
1da177e4 | 4483 | |
725aad24 | 4484 | if (user) { |
b68aa230 | 4485 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
4486 | /* |
4487 | * Do not allow realtime tasks into groups that have no runtime | |
4488 | * assigned. | |
4489 | */ | |
9a7e0b18 PZ |
4490 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
4491 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 4492 | return -EPERM; |
b68aa230 PZ |
4493 | #endif |
4494 | ||
725aad24 JF |
4495 | retval = security_task_setscheduler(p, policy, param); |
4496 | if (retval) | |
4497 | return retval; | |
4498 | } | |
4499 | ||
b29739f9 IM |
4500 | /* |
4501 | * make sure no PI-waiters arrive (or leave) while we are | |
4502 | * changing the priority of the task: | |
4503 | */ | |
1d615482 | 4504 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4505 | /* |
4506 | * To be able to change p->policy safely, the apropriate | |
4507 | * runqueue lock must be held. | |
4508 | */ | |
b29739f9 | 4509 | rq = __task_rq_lock(p); |
1da177e4 LT |
4510 | /* recheck policy now with rq lock held */ |
4511 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4512 | policy = oldpolicy = -1; | |
b29739f9 | 4513 | __task_rq_unlock(rq); |
1d615482 | 4514 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4515 | goto recheck; |
4516 | } | |
dd41f596 | 4517 | on_rq = p->se.on_rq; |
051a1d1a | 4518 | running = task_current(rq, p); |
0e1f3483 | 4519 | if (on_rq) |
2e1cb74a | 4520 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4521 | if (running) |
4522 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4523 | |
ca94c442 LP |
4524 | p->sched_reset_on_fork = reset_on_fork; |
4525 | ||
1da177e4 | 4526 | oldprio = p->prio; |
83ab0aa0 | 4527 | prev_class = p->sched_class; |
dd41f596 | 4528 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4529 | |
0e1f3483 HS |
4530 | if (running) |
4531 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4532 | if (on_rq) { |
4533 | activate_task(rq, p, 0); | |
cb469845 SR |
4534 | |
4535 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4536 | } |
b29739f9 | 4537 | __task_rq_unlock(rq); |
1d615482 | 4538 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4539 | |
95e02ca9 TG |
4540 | rt_mutex_adjust_pi(p); |
4541 | ||
1da177e4 LT |
4542 | return 0; |
4543 | } | |
961ccddd RR |
4544 | |
4545 | /** | |
4546 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4547 | * @p: the task in question. | |
4548 | * @policy: new policy. | |
4549 | * @param: structure containing the new RT priority. | |
4550 | * | |
4551 | * NOTE that the task may be already dead. | |
4552 | */ | |
4553 | int sched_setscheduler(struct task_struct *p, int policy, | |
4554 | struct sched_param *param) | |
4555 | { | |
4556 | return __sched_setscheduler(p, policy, param, true); | |
4557 | } | |
1da177e4 LT |
4558 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4559 | ||
961ccddd RR |
4560 | /** |
4561 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4562 | * @p: the task in question. | |
4563 | * @policy: new policy. | |
4564 | * @param: structure containing the new RT priority. | |
4565 | * | |
4566 | * Just like sched_setscheduler, only don't bother checking if the | |
4567 | * current context has permission. For example, this is needed in | |
4568 | * stop_machine(): we create temporary high priority worker threads, | |
4569 | * but our caller might not have that capability. | |
4570 | */ | |
4571 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4572 | struct sched_param *param) | |
4573 | { | |
4574 | return __sched_setscheduler(p, policy, param, false); | |
4575 | } | |
4576 | ||
95cdf3b7 IM |
4577 | static int |
4578 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4579 | { |
1da177e4 LT |
4580 | struct sched_param lparam; |
4581 | struct task_struct *p; | |
36c8b586 | 4582 | int retval; |
1da177e4 LT |
4583 | |
4584 | if (!param || pid < 0) | |
4585 | return -EINVAL; | |
4586 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4587 | return -EFAULT; | |
5fe1d75f ON |
4588 | |
4589 | rcu_read_lock(); | |
4590 | retval = -ESRCH; | |
1da177e4 | 4591 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4592 | if (p != NULL) |
4593 | retval = sched_setscheduler(p, policy, &lparam); | |
4594 | rcu_read_unlock(); | |
36c8b586 | 4595 | |
1da177e4 LT |
4596 | return retval; |
4597 | } | |
4598 | ||
4599 | /** | |
4600 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4601 | * @pid: the pid in question. | |
4602 | * @policy: new policy. | |
4603 | * @param: structure containing the new RT priority. | |
4604 | */ | |
5add95d4 HC |
4605 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4606 | struct sched_param __user *, param) | |
1da177e4 | 4607 | { |
c21761f1 JB |
4608 | /* negative values for policy are not valid */ |
4609 | if (policy < 0) | |
4610 | return -EINVAL; | |
4611 | ||
1da177e4 LT |
4612 | return do_sched_setscheduler(pid, policy, param); |
4613 | } | |
4614 | ||
4615 | /** | |
4616 | * sys_sched_setparam - set/change the RT priority of a thread | |
4617 | * @pid: the pid in question. | |
4618 | * @param: structure containing the new RT priority. | |
4619 | */ | |
5add95d4 | 4620 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4621 | { |
4622 | return do_sched_setscheduler(pid, -1, param); | |
4623 | } | |
4624 | ||
4625 | /** | |
4626 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4627 | * @pid: the pid in question. | |
4628 | */ | |
5add95d4 | 4629 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4630 | { |
36c8b586 | 4631 | struct task_struct *p; |
3a5c359a | 4632 | int retval; |
1da177e4 LT |
4633 | |
4634 | if (pid < 0) | |
3a5c359a | 4635 | return -EINVAL; |
1da177e4 LT |
4636 | |
4637 | retval = -ESRCH; | |
5fe85be0 | 4638 | rcu_read_lock(); |
1da177e4 LT |
4639 | p = find_process_by_pid(pid); |
4640 | if (p) { | |
4641 | retval = security_task_getscheduler(p); | |
4642 | if (!retval) | |
ca94c442 LP |
4643 | retval = p->policy |
4644 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4645 | } |
5fe85be0 | 4646 | rcu_read_unlock(); |
1da177e4 LT |
4647 | return retval; |
4648 | } | |
4649 | ||
4650 | /** | |
ca94c442 | 4651 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4652 | * @pid: the pid in question. |
4653 | * @param: structure containing the RT priority. | |
4654 | */ | |
5add95d4 | 4655 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4656 | { |
4657 | struct sched_param lp; | |
36c8b586 | 4658 | struct task_struct *p; |
3a5c359a | 4659 | int retval; |
1da177e4 LT |
4660 | |
4661 | if (!param || pid < 0) | |
3a5c359a | 4662 | return -EINVAL; |
1da177e4 | 4663 | |
5fe85be0 | 4664 | rcu_read_lock(); |
1da177e4 LT |
4665 | p = find_process_by_pid(pid); |
4666 | retval = -ESRCH; | |
4667 | if (!p) | |
4668 | goto out_unlock; | |
4669 | ||
4670 | retval = security_task_getscheduler(p); | |
4671 | if (retval) | |
4672 | goto out_unlock; | |
4673 | ||
4674 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4675 | rcu_read_unlock(); |
1da177e4 LT |
4676 | |
4677 | /* | |
4678 | * This one might sleep, we cannot do it with a spinlock held ... | |
4679 | */ | |
4680 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4681 | ||
1da177e4 LT |
4682 | return retval; |
4683 | ||
4684 | out_unlock: | |
5fe85be0 | 4685 | rcu_read_unlock(); |
1da177e4 LT |
4686 | return retval; |
4687 | } | |
4688 | ||
96f874e2 | 4689 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4690 | { |
5a16f3d3 | 4691 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4692 | struct task_struct *p; |
4693 | int retval; | |
1da177e4 | 4694 | |
95402b38 | 4695 | get_online_cpus(); |
23f5d142 | 4696 | rcu_read_lock(); |
1da177e4 LT |
4697 | |
4698 | p = find_process_by_pid(pid); | |
4699 | if (!p) { | |
23f5d142 | 4700 | rcu_read_unlock(); |
95402b38 | 4701 | put_online_cpus(); |
1da177e4 LT |
4702 | return -ESRCH; |
4703 | } | |
4704 | ||
23f5d142 | 4705 | /* Prevent p going away */ |
1da177e4 | 4706 | get_task_struct(p); |
23f5d142 | 4707 | rcu_read_unlock(); |
1da177e4 | 4708 | |
5a16f3d3 RR |
4709 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4710 | retval = -ENOMEM; | |
4711 | goto out_put_task; | |
4712 | } | |
4713 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4714 | retval = -ENOMEM; | |
4715 | goto out_free_cpus_allowed; | |
4716 | } | |
1da177e4 | 4717 | retval = -EPERM; |
c69e8d9c | 4718 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4719 | goto out_unlock; |
4720 | ||
e7834f8f DQ |
4721 | retval = security_task_setscheduler(p, 0, NULL); |
4722 | if (retval) | |
4723 | goto out_unlock; | |
4724 | ||
5a16f3d3 RR |
4725 | cpuset_cpus_allowed(p, cpus_allowed); |
4726 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4727 | again: |
5a16f3d3 | 4728 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4729 | |
8707d8b8 | 4730 | if (!retval) { |
5a16f3d3 RR |
4731 | cpuset_cpus_allowed(p, cpus_allowed); |
4732 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4733 | /* |
4734 | * We must have raced with a concurrent cpuset | |
4735 | * update. Just reset the cpus_allowed to the | |
4736 | * cpuset's cpus_allowed | |
4737 | */ | |
5a16f3d3 | 4738 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4739 | goto again; |
4740 | } | |
4741 | } | |
1da177e4 | 4742 | out_unlock: |
5a16f3d3 RR |
4743 | free_cpumask_var(new_mask); |
4744 | out_free_cpus_allowed: | |
4745 | free_cpumask_var(cpus_allowed); | |
4746 | out_put_task: | |
1da177e4 | 4747 | put_task_struct(p); |
95402b38 | 4748 | put_online_cpus(); |
1da177e4 LT |
4749 | return retval; |
4750 | } | |
4751 | ||
4752 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4753 | struct cpumask *new_mask) |
1da177e4 | 4754 | { |
96f874e2 RR |
4755 | if (len < cpumask_size()) |
4756 | cpumask_clear(new_mask); | |
4757 | else if (len > cpumask_size()) | |
4758 | len = cpumask_size(); | |
4759 | ||
1da177e4 LT |
4760 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4761 | } | |
4762 | ||
4763 | /** | |
4764 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4765 | * @pid: pid of the process | |
4766 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4767 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4768 | */ | |
5add95d4 HC |
4769 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4770 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4771 | { |
5a16f3d3 | 4772 | cpumask_var_t new_mask; |
1da177e4 LT |
4773 | int retval; |
4774 | ||
5a16f3d3 RR |
4775 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4776 | return -ENOMEM; | |
1da177e4 | 4777 | |
5a16f3d3 RR |
4778 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4779 | if (retval == 0) | |
4780 | retval = sched_setaffinity(pid, new_mask); | |
4781 | free_cpumask_var(new_mask); | |
4782 | return retval; | |
1da177e4 LT |
4783 | } |
4784 | ||
96f874e2 | 4785 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4786 | { |
36c8b586 | 4787 | struct task_struct *p; |
31605683 TG |
4788 | unsigned long flags; |
4789 | struct rq *rq; | |
1da177e4 | 4790 | int retval; |
1da177e4 | 4791 | |
95402b38 | 4792 | get_online_cpus(); |
23f5d142 | 4793 | rcu_read_lock(); |
1da177e4 LT |
4794 | |
4795 | retval = -ESRCH; | |
4796 | p = find_process_by_pid(pid); | |
4797 | if (!p) | |
4798 | goto out_unlock; | |
4799 | ||
e7834f8f DQ |
4800 | retval = security_task_getscheduler(p); |
4801 | if (retval) | |
4802 | goto out_unlock; | |
4803 | ||
31605683 | 4804 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4805 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4806 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4807 | |
4808 | out_unlock: | |
23f5d142 | 4809 | rcu_read_unlock(); |
95402b38 | 4810 | put_online_cpus(); |
1da177e4 | 4811 | |
9531b62f | 4812 | return retval; |
1da177e4 LT |
4813 | } |
4814 | ||
4815 | /** | |
4816 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4817 | * @pid: pid of the process | |
4818 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4819 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4820 | */ | |
5add95d4 HC |
4821 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4822 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4823 | { |
4824 | int ret; | |
f17c8607 | 4825 | cpumask_var_t mask; |
1da177e4 | 4826 | |
cd3d8031 KM |
4827 | if (len < nr_cpu_ids) |
4828 | return -EINVAL; | |
4829 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4830 | return -EINVAL; |
4831 | ||
f17c8607 RR |
4832 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4833 | return -ENOMEM; | |
1da177e4 | 4834 | |
f17c8607 RR |
4835 | ret = sched_getaffinity(pid, mask); |
4836 | if (ret == 0) { | |
8bc037fb | 4837 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4838 | |
4839 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4840 | ret = -EFAULT; |
4841 | else | |
cd3d8031 | 4842 | ret = retlen; |
f17c8607 RR |
4843 | } |
4844 | free_cpumask_var(mask); | |
1da177e4 | 4845 | |
f17c8607 | 4846 | return ret; |
1da177e4 LT |
4847 | } |
4848 | ||
4849 | /** | |
4850 | * sys_sched_yield - yield the current processor to other threads. | |
4851 | * | |
dd41f596 IM |
4852 | * This function yields the current CPU to other tasks. If there are no |
4853 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4854 | */ |
5add95d4 | 4855 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4856 | { |
70b97a7f | 4857 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4858 | |
2d72376b | 4859 | schedstat_inc(rq, yld_count); |
4530d7ab | 4860 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4861 | |
4862 | /* | |
4863 | * Since we are going to call schedule() anyway, there's | |
4864 | * no need to preempt or enable interrupts: | |
4865 | */ | |
4866 | __release(rq->lock); | |
8a25d5de | 4867 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4868 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4869 | preempt_enable_no_resched(); |
4870 | ||
4871 | schedule(); | |
4872 | ||
4873 | return 0; | |
4874 | } | |
4875 | ||
d86ee480 PZ |
4876 | static inline int should_resched(void) |
4877 | { | |
4878 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4879 | } | |
4880 | ||
e7b38404 | 4881 | static void __cond_resched(void) |
1da177e4 | 4882 | { |
e7aaaa69 FW |
4883 | add_preempt_count(PREEMPT_ACTIVE); |
4884 | schedule(); | |
4885 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
4886 | } |
4887 | ||
02b67cc3 | 4888 | int __sched _cond_resched(void) |
1da177e4 | 4889 | { |
d86ee480 | 4890 | if (should_resched()) { |
1da177e4 LT |
4891 | __cond_resched(); |
4892 | return 1; | |
4893 | } | |
4894 | return 0; | |
4895 | } | |
02b67cc3 | 4896 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4897 | |
4898 | /* | |
613afbf8 | 4899 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4900 | * call schedule, and on return reacquire the lock. |
4901 | * | |
41a2d6cf | 4902 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4903 | * operations here to prevent schedule() from being called twice (once via |
4904 | * spin_unlock(), once by hand). | |
4905 | */ | |
613afbf8 | 4906 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4907 | { |
d86ee480 | 4908 | int resched = should_resched(); |
6df3cecb JK |
4909 | int ret = 0; |
4910 | ||
f607c668 PZ |
4911 | lockdep_assert_held(lock); |
4912 | ||
95c354fe | 4913 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4914 | spin_unlock(lock); |
d86ee480 | 4915 | if (resched) |
95c354fe NP |
4916 | __cond_resched(); |
4917 | else | |
4918 | cpu_relax(); | |
6df3cecb | 4919 | ret = 1; |
1da177e4 | 4920 | spin_lock(lock); |
1da177e4 | 4921 | } |
6df3cecb | 4922 | return ret; |
1da177e4 | 4923 | } |
613afbf8 | 4924 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4925 | |
613afbf8 | 4926 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4927 | { |
4928 | BUG_ON(!in_softirq()); | |
4929 | ||
d86ee480 | 4930 | if (should_resched()) { |
98d82567 | 4931 | local_bh_enable(); |
1da177e4 LT |
4932 | __cond_resched(); |
4933 | local_bh_disable(); | |
4934 | return 1; | |
4935 | } | |
4936 | return 0; | |
4937 | } | |
613afbf8 | 4938 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4939 | |
1da177e4 LT |
4940 | /** |
4941 | * yield - yield the current processor to other threads. | |
4942 | * | |
72fd4a35 | 4943 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4944 | * thread runnable and calls sys_sched_yield(). |
4945 | */ | |
4946 | void __sched yield(void) | |
4947 | { | |
4948 | set_current_state(TASK_RUNNING); | |
4949 | sys_sched_yield(); | |
4950 | } | |
1da177e4 LT |
4951 | EXPORT_SYMBOL(yield); |
4952 | ||
4953 | /* | |
41a2d6cf | 4954 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4955 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4956 | */ |
4957 | void __sched io_schedule(void) | |
4958 | { | |
54d35f29 | 4959 | struct rq *rq = raw_rq(); |
1da177e4 | 4960 | |
0ff92245 | 4961 | delayacct_blkio_start(); |
1da177e4 | 4962 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4963 | current->in_iowait = 1; |
1da177e4 | 4964 | schedule(); |
8f0dfc34 | 4965 | current->in_iowait = 0; |
1da177e4 | 4966 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4967 | delayacct_blkio_end(); |
1da177e4 | 4968 | } |
1da177e4 LT |
4969 | EXPORT_SYMBOL(io_schedule); |
4970 | ||
4971 | long __sched io_schedule_timeout(long timeout) | |
4972 | { | |
54d35f29 | 4973 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4974 | long ret; |
4975 | ||
0ff92245 | 4976 | delayacct_blkio_start(); |
1da177e4 | 4977 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4978 | current->in_iowait = 1; |
1da177e4 | 4979 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4980 | current->in_iowait = 0; |
1da177e4 | 4981 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4982 | delayacct_blkio_end(); |
1da177e4 LT |
4983 | return ret; |
4984 | } | |
4985 | ||
4986 | /** | |
4987 | * sys_sched_get_priority_max - return maximum RT priority. | |
4988 | * @policy: scheduling class. | |
4989 | * | |
4990 | * this syscall returns the maximum rt_priority that can be used | |
4991 | * by a given scheduling class. | |
4992 | */ | |
5add95d4 | 4993 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4994 | { |
4995 | int ret = -EINVAL; | |
4996 | ||
4997 | switch (policy) { | |
4998 | case SCHED_FIFO: | |
4999 | case SCHED_RR: | |
5000 | ret = MAX_USER_RT_PRIO-1; | |
5001 | break; | |
5002 | case SCHED_NORMAL: | |
b0a9499c | 5003 | case SCHED_BATCH: |
dd41f596 | 5004 | case SCHED_IDLE: |
1da177e4 LT |
5005 | ret = 0; |
5006 | break; | |
5007 | } | |
5008 | return ret; | |
5009 | } | |
5010 | ||
5011 | /** | |
5012 | * sys_sched_get_priority_min - return minimum RT priority. | |
5013 | * @policy: scheduling class. | |
5014 | * | |
5015 | * this syscall returns the minimum rt_priority that can be used | |
5016 | * by a given scheduling class. | |
5017 | */ | |
5add95d4 | 5018 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5019 | { |
5020 | int ret = -EINVAL; | |
5021 | ||
5022 | switch (policy) { | |
5023 | case SCHED_FIFO: | |
5024 | case SCHED_RR: | |
5025 | ret = 1; | |
5026 | break; | |
5027 | case SCHED_NORMAL: | |
b0a9499c | 5028 | case SCHED_BATCH: |
dd41f596 | 5029 | case SCHED_IDLE: |
1da177e4 LT |
5030 | ret = 0; |
5031 | } | |
5032 | return ret; | |
5033 | } | |
5034 | ||
5035 | /** | |
5036 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5037 | * @pid: pid of the process. | |
5038 | * @interval: userspace pointer to the timeslice value. | |
5039 | * | |
5040 | * this syscall writes the default timeslice value of a given process | |
5041 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5042 | */ | |
17da2bd9 | 5043 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5044 | struct timespec __user *, interval) |
1da177e4 | 5045 | { |
36c8b586 | 5046 | struct task_struct *p; |
a4ec24b4 | 5047 | unsigned int time_slice; |
dba091b9 TG |
5048 | unsigned long flags; |
5049 | struct rq *rq; | |
3a5c359a | 5050 | int retval; |
1da177e4 | 5051 | struct timespec t; |
1da177e4 LT |
5052 | |
5053 | if (pid < 0) | |
3a5c359a | 5054 | return -EINVAL; |
1da177e4 LT |
5055 | |
5056 | retval = -ESRCH; | |
1a551ae7 | 5057 | rcu_read_lock(); |
1da177e4 LT |
5058 | p = find_process_by_pid(pid); |
5059 | if (!p) | |
5060 | goto out_unlock; | |
5061 | ||
5062 | retval = security_task_getscheduler(p); | |
5063 | if (retval) | |
5064 | goto out_unlock; | |
5065 | ||
dba091b9 TG |
5066 | rq = task_rq_lock(p, &flags); |
5067 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5068 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5069 | |
1a551ae7 | 5070 | rcu_read_unlock(); |
a4ec24b4 | 5071 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5072 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5073 | return retval; |
3a5c359a | 5074 | |
1da177e4 | 5075 | out_unlock: |
1a551ae7 | 5076 | rcu_read_unlock(); |
1da177e4 LT |
5077 | return retval; |
5078 | } | |
5079 | ||
7c731e0a | 5080 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5081 | |
82a1fcb9 | 5082 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5083 | { |
1da177e4 | 5084 | unsigned long free = 0; |
36c8b586 | 5085 | unsigned state; |
1da177e4 | 5086 | |
1da177e4 | 5087 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5088 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5089 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5090 | #if BITS_PER_LONG == 32 |
1da177e4 | 5091 | if (state == TASK_RUNNING) |
3df0fc5b | 5092 | printk(KERN_CONT " running "); |
1da177e4 | 5093 | else |
3df0fc5b | 5094 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5095 | #else |
5096 | if (state == TASK_RUNNING) | |
3df0fc5b | 5097 | printk(KERN_CONT " running task "); |
1da177e4 | 5098 | else |
3df0fc5b | 5099 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5100 | #endif |
5101 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5102 | free = stack_not_used(p); |
1da177e4 | 5103 | #endif |
3df0fc5b | 5104 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5105 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5106 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5107 | |
5fb5e6de | 5108 | show_stack(p, NULL); |
1da177e4 LT |
5109 | } |
5110 | ||
e59e2ae2 | 5111 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5112 | { |
36c8b586 | 5113 | struct task_struct *g, *p; |
1da177e4 | 5114 | |
4bd77321 | 5115 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5116 | printk(KERN_INFO |
5117 | " task PC stack pid father\n"); | |
1da177e4 | 5118 | #else |
3df0fc5b PZ |
5119 | printk(KERN_INFO |
5120 | " task PC stack pid father\n"); | |
1da177e4 LT |
5121 | #endif |
5122 | read_lock(&tasklist_lock); | |
5123 | do_each_thread(g, p) { | |
5124 | /* | |
5125 | * reset the NMI-timeout, listing all files on a slow | |
5126 | * console might take alot of time: | |
5127 | */ | |
5128 | touch_nmi_watchdog(); | |
39bc89fd | 5129 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5130 | sched_show_task(p); |
1da177e4 LT |
5131 | } while_each_thread(g, p); |
5132 | ||
04c9167f JF |
5133 | touch_all_softlockup_watchdogs(); |
5134 | ||
dd41f596 IM |
5135 | #ifdef CONFIG_SCHED_DEBUG |
5136 | sysrq_sched_debug_show(); | |
5137 | #endif | |
1da177e4 | 5138 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5139 | /* |
5140 | * Only show locks if all tasks are dumped: | |
5141 | */ | |
93335a21 | 5142 | if (!state_filter) |
e59e2ae2 | 5143 | debug_show_all_locks(); |
1da177e4 LT |
5144 | } |
5145 | ||
1df21055 IM |
5146 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5147 | { | |
dd41f596 | 5148 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5149 | } |
5150 | ||
f340c0d1 IM |
5151 | /** |
5152 | * init_idle - set up an idle thread for a given CPU | |
5153 | * @idle: task in question | |
5154 | * @cpu: cpu the idle task belongs to | |
5155 | * | |
5156 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5157 | * flag, to make booting more robust. | |
5158 | */ | |
5c1e1767 | 5159 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5160 | { |
70b97a7f | 5161 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5162 | unsigned long flags; |
5163 | ||
05fa785c | 5164 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5165 | |
dd41f596 | 5166 | __sched_fork(idle); |
06b83b5f | 5167 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5168 | idle->se.exec_start = sched_clock(); |
5169 | ||
96f874e2 | 5170 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5171 | __set_task_cpu(idle, cpu); |
1da177e4 | 5172 | |
1da177e4 | 5173 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5174 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5175 | idle->oncpu = 1; | |
5176 | #endif | |
05fa785c | 5177 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5178 | |
5179 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5180 | #if defined(CONFIG_PREEMPT) |
5181 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5182 | #else | |
a1261f54 | 5183 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5184 | #endif |
dd41f596 IM |
5185 | /* |
5186 | * The idle tasks have their own, simple scheduling class: | |
5187 | */ | |
5188 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5189 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5190 | } |
5191 | ||
5192 | /* | |
5193 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5194 | * indicates which cpus entered this state. This is used | |
5195 | * in the rcu update to wait only for active cpus. For system | |
5196 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5197 | * always be CPU_BITS_NONE. |
1da177e4 | 5198 | */ |
6a7b3dc3 | 5199 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5200 | |
19978ca6 IM |
5201 | /* |
5202 | * Increase the granularity value when there are more CPUs, | |
5203 | * because with more CPUs the 'effective latency' as visible | |
5204 | * to users decreases. But the relationship is not linear, | |
5205 | * so pick a second-best guess by going with the log2 of the | |
5206 | * number of CPUs. | |
5207 | * | |
5208 | * This idea comes from the SD scheduler of Con Kolivas: | |
5209 | */ | |
acb4a848 | 5210 | static int get_update_sysctl_factor(void) |
19978ca6 | 5211 | { |
4ca3ef71 | 5212 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5213 | unsigned int factor; |
5214 | ||
5215 | switch (sysctl_sched_tunable_scaling) { | |
5216 | case SCHED_TUNABLESCALING_NONE: | |
5217 | factor = 1; | |
5218 | break; | |
5219 | case SCHED_TUNABLESCALING_LINEAR: | |
5220 | factor = cpus; | |
5221 | break; | |
5222 | case SCHED_TUNABLESCALING_LOG: | |
5223 | default: | |
5224 | factor = 1 + ilog2(cpus); | |
5225 | break; | |
5226 | } | |
19978ca6 | 5227 | |
acb4a848 CE |
5228 | return factor; |
5229 | } | |
19978ca6 | 5230 | |
acb4a848 CE |
5231 | static void update_sysctl(void) |
5232 | { | |
5233 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5234 | |
0bcdcf28 CE |
5235 | #define SET_SYSCTL(name) \ |
5236 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5237 | SET_SYSCTL(sched_min_granularity); | |
5238 | SET_SYSCTL(sched_latency); | |
5239 | SET_SYSCTL(sched_wakeup_granularity); | |
5240 | SET_SYSCTL(sched_shares_ratelimit); | |
5241 | #undef SET_SYSCTL | |
5242 | } | |
55cd5340 | 5243 | |
0bcdcf28 CE |
5244 | static inline void sched_init_granularity(void) |
5245 | { | |
5246 | update_sysctl(); | |
19978ca6 IM |
5247 | } |
5248 | ||
1da177e4 LT |
5249 | #ifdef CONFIG_SMP |
5250 | /* | |
5251 | * This is how migration works: | |
5252 | * | |
70b97a7f | 5253 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5254 | * runqueue and wake up that CPU's migration thread. |
5255 | * 2) we down() the locked semaphore => thread blocks. | |
5256 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5257 | * thread off the CPU) | |
5258 | * 4) it gets the migration request and checks whether the migrated | |
5259 | * task is still in the wrong runqueue. | |
5260 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5261 | * it and puts it into the right queue. | |
5262 | * 6) migration thread up()s the semaphore. | |
5263 | * 7) we wake up and the migration is done. | |
5264 | */ | |
5265 | ||
5266 | /* | |
5267 | * Change a given task's CPU affinity. Migrate the thread to a | |
5268 | * proper CPU and schedule it away if the CPU it's executing on | |
5269 | * is removed from the allowed bitmask. | |
5270 | * | |
5271 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5272 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5273 | * call is not atomic; no spinlocks may be held. |
5274 | */ | |
96f874e2 | 5275 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 5276 | { |
70b97a7f | 5277 | struct migration_req req; |
1da177e4 | 5278 | unsigned long flags; |
70b97a7f | 5279 | struct rq *rq; |
48f24c4d | 5280 | int ret = 0; |
1da177e4 LT |
5281 | |
5282 | rq = task_rq_lock(p, &flags); | |
e2912009 | 5283 | |
6ad4c188 | 5284 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5285 | ret = -EINVAL; |
5286 | goto out; | |
5287 | } | |
5288 | ||
9985b0ba | 5289 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5290 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5291 | ret = -EINVAL; |
5292 | goto out; | |
5293 | } | |
5294 | ||
73fe6aae | 5295 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5296 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5297 | else { |
96f874e2 RR |
5298 | cpumask_copy(&p->cpus_allowed, new_mask); |
5299 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5300 | } |
5301 | ||
1da177e4 | 5302 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5303 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5304 | goto out; |
5305 | ||
6ad4c188 | 5306 | if (migrate_task(p, cpumask_any_and(cpu_active_mask, new_mask), &req)) { |
1da177e4 | 5307 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
5308 | struct task_struct *mt = rq->migration_thread; |
5309 | ||
5310 | get_task_struct(mt); | |
1da177e4 LT |
5311 | task_rq_unlock(rq, &flags); |
5312 | wake_up_process(rq->migration_thread); | |
693525e3 | 5313 | put_task_struct(mt); |
1da177e4 LT |
5314 | wait_for_completion(&req.done); |
5315 | tlb_migrate_finish(p->mm); | |
5316 | return 0; | |
5317 | } | |
5318 | out: | |
5319 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5320 | |
1da177e4 LT |
5321 | return ret; |
5322 | } | |
cd8ba7cd | 5323 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5324 | |
5325 | /* | |
41a2d6cf | 5326 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5327 | * this because either it can't run here any more (set_cpus_allowed() |
5328 | * away from this CPU, or CPU going down), or because we're | |
5329 | * attempting to rebalance this task on exec (sched_exec). | |
5330 | * | |
5331 | * So we race with normal scheduler movements, but that's OK, as long | |
5332 | * as the task is no longer on this CPU. | |
efc30814 KK |
5333 | * |
5334 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5335 | */ |
efc30814 | 5336 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5337 | { |
70b97a7f | 5338 | struct rq *rq_dest, *rq_src; |
e2912009 | 5339 | int ret = 0; |
1da177e4 | 5340 | |
e761b772 | 5341 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5342 | return ret; |
1da177e4 LT |
5343 | |
5344 | rq_src = cpu_rq(src_cpu); | |
5345 | rq_dest = cpu_rq(dest_cpu); | |
5346 | ||
5347 | double_rq_lock(rq_src, rq_dest); | |
5348 | /* Already moved. */ | |
5349 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5350 | goto done; |
1da177e4 | 5351 | /* Affinity changed (again). */ |
96f874e2 | 5352 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5353 | goto fail; |
1da177e4 | 5354 | |
e2912009 PZ |
5355 | /* |
5356 | * If we're not on a rq, the next wake-up will ensure we're | |
5357 | * placed properly. | |
5358 | */ | |
5359 | if (p->se.on_rq) { | |
2e1cb74a | 5360 | deactivate_task(rq_src, p, 0); |
e2912009 | 5361 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5362 | activate_task(rq_dest, p, 0); |
15afe09b | 5363 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5364 | } |
b1e38734 | 5365 | done: |
efc30814 | 5366 | ret = 1; |
b1e38734 | 5367 | fail: |
1da177e4 | 5368 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5369 | return ret; |
1da177e4 LT |
5370 | } |
5371 | ||
03b042bf PM |
5372 | #define RCU_MIGRATION_IDLE 0 |
5373 | #define RCU_MIGRATION_NEED_QS 1 | |
5374 | #define RCU_MIGRATION_GOT_QS 2 | |
5375 | #define RCU_MIGRATION_MUST_SYNC 3 | |
5376 | ||
1da177e4 LT |
5377 | /* |
5378 | * migration_thread - this is a highprio system thread that performs | |
5379 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5380 | * another runqueue. | |
5381 | */ | |
95cdf3b7 | 5382 | static int migration_thread(void *data) |
1da177e4 | 5383 | { |
03b042bf | 5384 | int badcpu; |
1da177e4 | 5385 | int cpu = (long)data; |
70b97a7f | 5386 | struct rq *rq; |
1da177e4 LT |
5387 | |
5388 | rq = cpu_rq(cpu); | |
5389 | BUG_ON(rq->migration_thread != current); | |
5390 | ||
5391 | set_current_state(TASK_INTERRUPTIBLE); | |
5392 | while (!kthread_should_stop()) { | |
70b97a7f | 5393 | struct migration_req *req; |
1da177e4 | 5394 | struct list_head *head; |
1da177e4 | 5395 | |
05fa785c | 5396 | raw_spin_lock_irq(&rq->lock); |
1da177e4 LT |
5397 | |
5398 | if (cpu_is_offline(cpu)) { | |
05fa785c | 5399 | raw_spin_unlock_irq(&rq->lock); |
371cbb38 | 5400 | break; |
1da177e4 LT |
5401 | } |
5402 | ||
5403 | if (rq->active_balance) { | |
5404 | active_load_balance(rq, cpu); | |
5405 | rq->active_balance = 0; | |
5406 | } | |
5407 | ||
5408 | head = &rq->migration_queue; | |
5409 | ||
5410 | if (list_empty(head)) { | |
05fa785c | 5411 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5412 | schedule(); |
5413 | set_current_state(TASK_INTERRUPTIBLE); | |
5414 | continue; | |
5415 | } | |
70b97a7f | 5416 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5417 | list_del_init(head->next); |
5418 | ||
03b042bf | 5419 | if (req->task != NULL) { |
05fa785c | 5420 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5421 | __migrate_task(req->task, cpu, req->dest_cpu); |
5422 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
5423 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
05fa785c | 5424 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5425 | } else { |
5426 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
05fa785c | 5427 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5428 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); |
5429 | } | |
674311d5 | 5430 | local_irq_enable(); |
1da177e4 LT |
5431 | |
5432 | complete(&req->done); | |
5433 | } | |
5434 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 5435 | |
1da177e4 LT |
5436 | return 0; |
5437 | } | |
5438 | ||
5439 | #ifdef CONFIG_HOTPLUG_CPU | |
054b9108 | 5440 | /* |
3a4fa0a2 | 5441 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5442 | */ |
6a1bdc1b | 5443 | void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5444 | { |
1445c08d ON |
5445 | struct rq *rq = cpu_rq(dead_cpu); |
5446 | int needs_cpu, uninitialized_var(dest_cpu); | |
5447 | unsigned long flags; | |
c1804d54 | 5448 | |
1445c08d ON |
5449 | local_irq_save(flags); |
5450 | ||
5451 | raw_spin_lock(&rq->lock); | |
5452 | needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING); | |
5453 | if (needs_cpu) | |
5454 | dest_cpu = select_fallback_rq(dead_cpu, p); | |
5455 | raw_spin_unlock(&rq->lock); | |
c1804d54 ON |
5456 | /* |
5457 | * It can only fail if we race with set_cpus_allowed(), | |
5458 | * in the racer should migrate the task anyway. | |
5459 | */ | |
1445c08d | 5460 | if (needs_cpu) |
c1804d54 | 5461 | __migrate_task(p, dead_cpu, dest_cpu); |
1445c08d | 5462 | local_irq_restore(flags); |
1da177e4 LT |
5463 | } |
5464 | ||
5465 | /* | |
5466 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5467 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5468 | * for performance reasons the counter is not stricly tracking tasks to | |
5469 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5470 | * to keep the global sum constant after CPU-down: | |
5471 | */ | |
70b97a7f | 5472 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5473 | { |
6ad4c188 | 5474 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5475 | unsigned long flags; |
5476 | ||
5477 | local_irq_save(flags); | |
5478 | double_rq_lock(rq_src, rq_dest); | |
5479 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5480 | rq_src->nr_uninterruptible = 0; | |
5481 | double_rq_unlock(rq_src, rq_dest); | |
5482 | local_irq_restore(flags); | |
5483 | } | |
5484 | ||
5485 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5486 | static void migrate_live_tasks(int src_cpu) | |
5487 | { | |
48f24c4d | 5488 | struct task_struct *p, *t; |
1da177e4 | 5489 | |
f7b4cddc | 5490 | read_lock(&tasklist_lock); |
1da177e4 | 5491 | |
48f24c4d IM |
5492 | do_each_thread(t, p) { |
5493 | if (p == current) | |
1da177e4 LT |
5494 | continue; |
5495 | ||
48f24c4d IM |
5496 | if (task_cpu(p) == src_cpu) |
5497 | move_task_off_dead_cpu(src_cpu, p); | |
5498 | } while_each_thread(t, p); | |
1da177e4 | 5499 | |
f7b4cddc | 5500 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5501 | } |
5502 | ||
dd41f596 IM |
5503 | /* |
5504 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5505 | * It does so by boosting its priority to highest possible. |
5506 | * Used by CPU offline code. | |
1da177e4 LT |
5507 | */ |
5508 | void sched_idle_next(void) | |
5509 | { | |
48f24c4d | 5510 | int this_cpu = smp_processor_id(); |
70b97a7f | 5511 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5512 | struct task_struct *p = rq->idle; |
5513 | unsigned long flags; | |
5514 | ||
5515 | /* cpu has to be offline */ | |
48f24c4d | 5516 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5517 | |
48f24c4d IM |
5518 | /* |
5519 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5520 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5521 | */ |
05fa785c | 5522 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5523 | |
dd41f596 | 5524 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5525 | |
94bc9a7b | 5526 | activate_task(rq, p, 0); |
1da177e4 | 5527 | |
05fa785c | 5528 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5529 | } |
5530 | ||
48f24c4d IM |
5531 | /* |
5532 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5533 | * offline. |
5534 | */ | |
5535 | void idle_task_exit(void) | |
5536 | { | |
5537 | struct mm_struct *mm = current->active_mm; | |
5538 | ||
5539 | BUG_ON(cpu_online(smp_processor_id())); | |
5540 | ||
5541 | if (mm != &init_mm) | |
5542 | switch_mm(mm, &init_mm, current); | |
5543 | mmdrop(mm); | |
5544 | } | |
5545 | ||
054b9108 | 5546 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5547 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5548 | { |
70b97a7f | 5549 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5550 | |
5551 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5552 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5553 | |
5554 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5555 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5556 | |
48f24c4d | 5557 | get_task_struct(p); |
1da177e4 LT |
5558 | |
5559 | /* | |
5560 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5561 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5562 | * fine. |
5563 | */ | |
05fa785c | 5564 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5565 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5566 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5567 | |
48f24c4d | 5568 | put_task_struct(p); |
1da177e4 LT |
5569 | } |
5570 | ||
5571 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5572 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5573 | { | |
70b97a7f | 5574 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5575 | struct task_struct *next; |
48f24c4d | 5576 | |
dd41f596 IM |
5577 | for ( ; ; ) { |
5578 | if (!rq->nr_running) | |
5579 | break; | |
b67802ea | 5580 | next = pick_next_task(rq); |
dd41f596 IM |
5581 | if (!next) |
5582 | break; | |
79c53799 | 5583 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5584 | migrate_dead(dead_cpu, next); |
e692ab53 | 5585 | |
1da177e4 LT |
5586 | } |
5587 | } | |
dce48a84 TG |
5588 | |
5589 | /* | |
5590 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5591 | */ | |
5592 | static void calc_global_load_remove(struct rq *rq) | |
5593 | { | |
5594 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5595 | rq->calc_load_active = 0; |
dce48a84 | 5596 | } |
1da177e4 LT |
5597 | #endif /* CONFIG_HOTPLUG_CPU */ |
5598 | ||
e692ab53 NP |
5599 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5600 | ||
5601 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5602 | { |
5603 | .procname = "sched_domain", | |
c57baf1e | 5604 | .mode = 0555, |
e0361851 | 5605 | }, |
56992309 | 5606 | {} |
e692ab53 NP |
5607 | }; |
5608 | ||
5609 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5610 | { |
5611 | .procname = "kernel", | |
c57baf1e | 5612 | .mode = 0555, |
e0361851 AD |
5613 | .child = sd_ctl_dir, |
5614 | }, | |
56992309 | 5615 | {} |
e692ab53 NP |
5616 | }; |
5617 | ||
5618 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5619 | { | |
5620 | struct ctl_table *entry = | |
5cf9f062 | 5621 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5622 | |
e692ab53 NP |
5623 | return entry; |
5624 | } | |
5625 | ||
6382bc90 MM |
5626 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5627 | { | |
cd790076 | 5628 | struct ctl_table *entry; |
6382bc90 | 5629 | |
cd790076 MM |
5630 | /* |
5631 | * In the intermediate directories, both the child directory and | |
5632 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5633 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5634 | * static strings and all have proc handlers. |
5635 | */ | |
5636 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5637 | if (entry->child) |
5638 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5639 | if (entry->proc_handler == NULL) |
5640 | kfree(entry->procname); | |
5641 | } | |
6382bc90 MM |
5642 | |
5643 | kfree(*tablep); | |
5644 | *tablep = NULL; | |
5645 | } | |
5646 | ||
e692ab53 | 5647 | static void |
e0361851 | 5648 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5649 | const char *procname, void *data, int maxlen, |
5650 | mode_t mode, proc_handler *proc_handler) | |
5651 | { | |
e692ab53 NP |
5652 | entry->procname = procname; |
5653 | entry->data = data; | |
5654 | entry->maxlen = maxlen; | |
5655 | entry->mode = mode; | |
5656 | entry->proc_handler = proc_handler; | |
5657 | } | |
5658 | ||
5659 | static struct ctl_table * | |
5660 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5661 | { | |
a5d8c348 | 5662 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5663 | |
ad1cdc1d MM |
5664 | if (table == NULL) |
5665 | return NULL; | |
5666 | ||
e0361851 | 5667 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5668 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5669 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5670 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5671 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5672 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5673 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5674 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5675 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5676 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5677 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5678 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5679 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5680 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5681 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5682 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5683 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5684 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5685 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5686 | &sd->cache_nice_tries, |
5687 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5688 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5689 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5690 | set_table_entry(&table[11], "name", sd->name, |
5691 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5692 | /* &table[12] is terminator */ | |
e692ab53 NP |
5693 | |
5694 | return table; | |
5695 | } | |
5696 | ||
9a4e7159 | 5697 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5698 | { |
5699 | struct ctl_table *entry, *table; | |
5700 | struct sched_domain *sd; | |
5701 | int domain_num = 0, i; | |
5702 | char buf[32]; | |
5703 | ||
5704 | for_each_domain(cpu, sd) | |
5705 | domain_num++; | |
5706 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5707 | if (table == NULL) |
5708 | return NULL; | |
e692ab53 NP |
5709 | |
5710 | i = 0; | |
5711 | for_each_domain(cpu, sd) { | |
5712 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5713 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5714 | entry->mode = 0555; |
e692ab53 NP |
5715 | entry->child = sd_alloc_ctl_domain_table(sd); |
5716 | entry++; | |
5717 | i++; | |
5718 | } | |
5719 | return table; | |
5720 | } | |
5721 | ||
5722 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5723 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5724 | { |
6ad4c188 | 5725 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5726 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5727 | char buf[32]; | |
5728 | ||
7378547f MM |
5729 | WARN_ON(sd_ctl_dir[0].child); |
5730 | sd_ctl_dir[0].child = entry; | |
5731 | ||
ad1cdc1d MM |
5732 | if (entry == NULL) |
5733 | return; | |
5734 | ||
6ad4c188 | 5735 | for_each_possible_cpu(i) { |
e692ab53 | 5736 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5737 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5738 | entry->mode = 0555; |
e692ab53 | 5739 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5740 | entry++; |
e692ab53 | 5741 | } |
7378547f MM |
5742 | |
5743 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5744 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5745 | } | |
6382bc90 | 5746 | |
7378547f | 5747 | /* may be called multiple times per register */ |
6382bc90 MM |
5748 | static void unregister_sched_domain_sysctl(void) |
5749 | { | |
7378547f MM |
5750 | if (sd_sysctl_header) |
5751 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5752 | sd_sysctl_header = NULL; |
7378547f MM |
5753 | if (sd_ctl_dir[0].child) |
5754 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5755 | } |
e692ab53 | 5756 | #else |
6382bc90 MM |
5757 | static void register_sched_domain_sysctl(void) |
5758 | { | |
5759 | } | |
5760 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5761 | { |
5762 | } | |
5763 | #endif | |
5764 | ||
1f11eb6a GH |
5765 | static void set_rq_online(struct rq *rq) |
5766 | { | |
5767 | if (!rq->online) { | |
5768 | const struct sched_class *class; | |
5769 | ||
c6c4927b | 5770 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5771 | rq->online = 1; |
5772 | ||
5773 | for_each_class(class) { | |
5774 | if (class->rq_online) | |
5775 | class->rq_online(rq); | |
5776 | } | |
5777 | } | |
5778 | } | |
5779 | ||
5780 | static void set_rq_offline(struct rq *rq) | |
5781 | { | |
5782 | if (rq->online) { | |
5783 | const struct sched_class *class; | |
5784 | ||
5785 | for_each_class(class) { | |
5786 | if (class->rq_offline) | |
5787 | class->rq_offline(rq); | |
5788 | } | |
5789 | ||
c6c4927b | 5790 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5791 | rq->online = 0; |
5792 | } | |
5793 | } | |
5794 | ||
1da177e4 LT |
5795 | /* |
5796 | * migration_call - callback that gets triggered when a CPU is added. | |
5797 | * Here we can start up the necessary migration thread for the new CPU. | |
5798 | */ | |
48f24c4d IM |
5799 | static int __cpuinit |
5800 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5801 | { |
1da177e4 | 5802 | struct task_struct *p; |
48f24c4d | 5803 | int cpu = (long)hcpu; |
1da177e4 | 5804 | unsigned long flags; |
70b97a7f | 5805 | struct rq *rq; |
1da177e4 LT |
5806 | |
5807 | switch (action) { | |
5be9361c | 5808 | |
1da177e4 | 5809 | case CPU_UP_PREPARE: |
8bb78442 | 5810 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5811 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5812 | if (IS_ERR(p)) |
5813 | return NOTIFY_BAD; | |
1da177e4 LT |
5814 | kthread_bind(p, cpu); |
5815 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5816 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5817 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 5818 | task_rq_unlock(rq, &flags); |
371cbb38 | 5819 | get_task_struct(p); |
1da177e4 | 5820 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 5821 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5822 | break; |
48f24c4d | 5823 | |
1da177e4 | 5824 | case CPU_ONLINE: |
8bb78442 | 5825 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5826 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 5827 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
5828 | |
5829 | /* Update our root-domain */ | |
5830 | rq = cpu_rq(cpu); | |
05fa785c | 5831 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5832 | if (rq->rd) { |
c6c4927b | 5833 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5834 | |
5835 | set_rq_online(rq); | |
1f94ef59 | 5836 | } |
05fa785c | 5837 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5838 | break; |
48f24c4d | 5839 | |
1da177e4 LT |
5840 | #ifdef CONFIG_HOTPLUG_CPU |
5841 | case CPU_UP_CANCELED: | |
8bb78442 | 5842 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5843 | if (!cpu_rq(cpu)->migration_thread) |
5844 | break; | |
41a2d6cf | 5845 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 5846 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 5847 | cpumask_any(cpu_online_mask)); |
1da177e4 | 5848 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 5849 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
5850 | cpu_rq(cpu)->migration_thread = NULL; |
5851 | break; | |
48f24c4d | 5852 | |
1da177e4 | 5853 | case CPU_DEAD: |
8bb78442 | 5854 | case CPU_DEAD_FROZEN: |
1da177e4 LT |
5855 | migrate_live_tasks(cpu); |
5856 | rq = cpu_rq(cpu); | |
5857 | kthread_stop(rq->migration_thread); | |
371cbb38 | 5858 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
5859 | rq->migration_thread = NULL; |
5860 | /* Idle task back to normal (off runqueue, low prio) */ | |
05fa785c | 5861 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 5862 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5863 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5864 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5865 | migrate_dead_tasks(cpu); |
05fa785c | 5866 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5867 | migrate_nr_uninterruptible(rq); |
5868 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5869 | calc_global_load_remove(rq); |
41a2d6cf IM |
5870 | /* |
5871 | * No need to migrate the tasks: it was best-effort if | |
5872 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5873 | * the requestors. | |
5874 | */ | |
05fa785c | 5875 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5876 | while (!list_empty(&rq->migration_queue)) { |
70b97a7f IM |
5877 | struct migration_req *req; |
5878 | ||
1da177e4 | 5879 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5880 | struct migration_req, list); |
1da177e4 | 5881 | list_del_init(&req->list); |
05fa785c | 5882 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5883 | complete(&req->done); |
05fa785c | 5884 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5885 | } |
05fa785c | 5886 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5887 | break; |
57d885fe | 5888 | |
08f503b0 GH |
5889 | case CPU_DYING: |
5890 | case CPU_DYING_FROZEN: | |
57d885fe GH |
5891 | /* Update our root-domain */ |
5892 | rq = cpu_rq(cpu); | |
05fa785c | 5893 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5894 | if (rq->rd) { |
c6c4927b | 5895 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5896 | set_rq_offline(rq); |
57d885fe | 5897 | } |
05fa785c | 5898 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 5899 | break; |
1da177e4 LT |
5900 | #endif |
5901 | } | |
5902 | return NOTIFY_OK; | |
5903 | } | |
5904 | ||
f38b0820 PM |
5905 | /* |
5906 | * Register at high priority so that task migration (migrate_all_tasks) | |
5907 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5908 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5909 | */ |
26c2143b | 5910 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5911 | .notifier_call = migration_call, |
5912 | .priority = 10 | |
5913 | }; | |
5914 | ||
7babe8db | 5915 | static int __init migration_init(void) |
1da177e4 LT |
5916 | { |
5917 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5918 | int err; |
48f24c4d IM |
5919 | |
5920 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5921 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5922 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5923 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5924 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5925 | |
a004cd42 | 5926 | return 0; |
1da177e4 | 5927 | } |
7babe8db | 5928 | early_initcall(migration_init); |
1da177e4 LT |
5929 | #endif |
5930 | ||
5931 | #ifdef CONFIG_SMP | |
476f3534 | 5932 | |
3e9830dc | 5933 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5934 | |
f6630114 MT |
5935 | static __read_mostly int sched_domain_debug_enabled; |
5936 | ||
5937 | static int __init sched_domain_debug_setup(char *str) | |
5938 | { | |
5939 | sched_domain_debug_enabled = 1; | |
5940 | ||
5941 | return 0; | |
5942 | } | |
5943 | early_param("sched_debug", sched_domain_debug_setup); | |
5944 | ||
7c16ec58 | 5945 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5946 | struct cpumask *groupmask) |
1da177e4 | 5947 | { |
4dcf6aff | 5948 | struct sched_group *group = sd->groups; |
434d53b0 | 5949 | char str[256]; |
1da177e4 | 5950 | |
968ea6d8 | 5951 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5952 | cpumask_clear(groupmask); |
4dcf6aff IM |
5953 | |
5954 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5955 | ||
5956 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5957 | printk("does not load-balance\n"); |
4dcf6aff | 5958 | if (sd->parent) |
3df0fc5b PZ |
5959 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5960 | " has parent"); | |
4dcf6aff | 5961 | return -1; |
41c7ce9a NP |
5962 | } |
5963 | ||
3df0fc5b | 5964 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5965 | |
758b2cdc | 5966 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5967 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5968 | "CPU%d\n", cpu); | |
4dcf6aff | 5969 | } |
758b2cdc | 5970 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5971 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5972 | " CPU%d\n", cpu); | |
4dcf6aff | 5973 | } |
1da177e4 | 5974 | |
4dcf6aff | 5975 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5976 | do { |
4dcf6aff | 5977 | if (!group) { |
3df0fc5b PZ |
5978 | printk("\n"); |
5979 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5980 | break; |
5981 | } | |
5982 | ||
18a3885f | 5983 | if (!group->cpu_power) { |
3df0fc5b PZ |
5984 | printk(KERN_CONT "\n"); |
5985 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5986 | "set\n"); | |
4dcf6aff IM |
5987 | break; |
5988 | } | |
1da177e4 | 5989 | |
758b2cdc | 5990 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5991 | printk(KERN_CONT "\n"); |
5992 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5993 | break; |
5994 | } | |
1da177e4 | 5995 | |
758b2cdc | 5996 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5997 | printk(KERN_CONT "\n"); |
5998 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5999 | break; |
6000 | } | |
1da177e4 | 6001 | |
758b2cdc | 6002 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6003 | |
968ea6d8 | 6004 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6005 | |
3df0fc5b | 6006 | printk(KERN_CONT " %s", str); |
18a3885f | 6007 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6008 | printk(KERN_CONT " (cpu_power = %d)", |
6009 | group->cpu_power); | |
381512cf | 6010 | } |
1da177e4 | 6011 | |
4dcf6aff IM |
6012 | group = group->next; |
6013 | } while (group != sd->groups); | |
3df0fc5b | 6014 | printk(KERN_CONT "\n"); |
1da177e4 | 6015 | |
758b2cdc | 6016 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6017 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6018 | |
758b2cdc RR |
6019 | if (sd->parent && |
6020 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6021 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6022 | "of domain->span\n"); | |
4dcf6aff IM |
6023 | return 0; |
6024 | } | |
1da177e4 | 6025 | |
4dcf6aff IM |
6026 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6027 | { | |
d5dd3db1 | 6028 | cpumask_var_t groupmask; |
4dcf6aff | 6029 | int level = 0; |
1da177e4 | 6030 | |
f6630114 MT |
6031 | if (!sched_domain_debug_enabled) |
6032 | return; | |
6033 | ||
4dcf6aff IM |
6034 | if (!sd) { |
6035 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6036 | return; | |
6037 | } | |
1da177e4 | 6038 | |
4dcf6aff IM |
6039 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6040 | ||
d5dd3db1 | 6041 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6042 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6043 | return; | |
6044 | } | |
6045 | ||
4dcf6aff | 6046 | for (;;) { |
7c16ec58 | 6047 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6048 | break; |
1da177e4 LT |
6049 | level++; |
6050 | sd = sd->parent; | |
33859f7f | 6051 | if (!sd) |
4dcf6aff IM |
6052 | break; |
6053 | } | |
d5dd3db1 | 6054 | free_cpumask_var(groupmask); |
1da177e4 | 6055 | } |
6d6bc0ad | 6056 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6057 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6058 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6059 | |
1a20ff27 | 6060 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6061 | { |
758b2cdc | 6062 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6063 | return 1; |
6064 | ||
6065 | /* Following flags need at least 2 groups */ | |
6066 | if (sd->flags & (SD_LOAD_BALANCE | | |
6067 | SD_BALANCE_NEWIDLE | | |
6068 | SD_BALANCE_FORK | | |
89c4710e SS |
6069 | SD_BALANCE_EXEC | |
6070 | SD_SHARE_CPUPOWER | | |
6071 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6072 | if (sd->groups != sd->groups->next) |
6073 | return 0; | |
6074 | } | |
6075 | ||
6076 | /* Following flags don't use groups */ | |
c88d5910 | 6077 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6078 | return 0; |
6079 | ||
6080 | return 1; | |
6081 | } | |
6082 | ||
48f24c4d IM |
6083 | static int |
6084 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6085 | { |
6086 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6087 | ||
6088 | if (sd_degenerate(parent)) | |
6089 | return 1; | |
6090 | ||
758b2cdc | 6091 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6092 | return 0; |
6093 | ||
245af2c7 SS |
6094 | /* Flags needing groups don't count if only 1 group in parent */ |
6095 | if (parent->groups == parent->groups->next) { | |
6096 | pflags &= ~(SD_LOAD_BALANCE | | |
6097 | SD_BALANCE_NEWIDLE | | |
6098 | SD_BALANCE_FORK | | |
89c4710e SS |
6099 | SD_BALANCE_EXEC | |
6100 | SD_SHARE_CPUPOWER | | |
6101 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6102 | if (nr_node_ids == 1) |
6103 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6104 | } |
6105 | if (~cflags & pflags) | |
6106 | return 0; | |
6107 | ||
6108 | return 1; | |
6109 | } | |
6110 | ||
c6c4927b RR |
6111 | static void free_rootdomain(struct root_domain *rd) |
6112 | { | |
047106ad PZ |
6113 | synchronize_sched(); |
6114 | ||
68e74568 RR |
6115 | cpupri_cleanup(&rd->cpupri); |
6116 | ||
c6c4927b RR |
6117 | free_cpumask_var(rd->rto_mask); |
6118 | free_cpumask_var(rd->online); | |
6119 | free_cpumask_var(rd->span); | |
6120 | kfree(rd); | |
6121 | } | |
6122 | ||
57d885fe GH |
6123 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6124 | { | |
a0490fa3 | 6125 | struct root_domain *old_rd = NULL; |
57d885fe | 6126 | unsigned long flags; |
57d885fe | 6127 | |
05fa785c | 6128 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6129 | |
6130 | if (rq->rd) { | |
a0490fa3 | 6131 | old_rd = rq->rd; |
57d885fe | 6132 | |
c6c4927b | 6133 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6134 | set_rq_offline(rq); |
57d885fe | 6135 | |
c6c4927b | 6136 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6137 | |
a0490fa3 IM |
6138 | /* |
6139 | * If we dont want to free the old_rt yet then | |
6140 | * set old_rd to NULL to skip the freeing later | |
6141 | * in this function: | |
6142 | */ | |
6143 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6144 | old_rd = NULL; | |
57d885fe GH |
6145 | } |
6146 | ||
6147 | atomic_inc(&rd->refcount); | |
6148 | rq->rd = rd; | |
6149 | ||
c6c4927b | 6150 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6151 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6152 | set_rq_online(rq); |
57d885fe | 6153 | |
05fa785c | 6154 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6155 | |
6156 | if (old_rd) | |
6157 | free_rootdomain(old_rd); | |
57d885fe GH |
6158 | } |
6159 | ||
fd5e1b5d | 6160 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6161 | { |
36b7b6d4 PE |
6162 | gfp_t gfp = GFP_KERNEL; |
6163 | ||
57d885fe GH |
6164 | memset(rd, 0, sizeof(*rd)); |
6165 | ||
36b7b6d4 PE |
6166 | if (bootmem) |
6167 | gfp = GFP_NOWAIT; | |
c6c4927b | 6168 | |
36b7b6d4 | 6169 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6170 | goto out; |
36b7b6d4 | 6171 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6172 | goto free_span; |
36b7b6d4 | 6173 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6174 | goto free_online; |
6e0534f2 | 6175 | |
0fb53029 | 6176 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6177 | goto free_rto_mask; |
c6c4927b | 6178 | return 0; |
6e0534f2 | 6179 | |
68e74568 RR |
6180 | free_rto_mask: |
6181 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6182 | free_online: |
6183 | free_cpumask_var(rd->online); | |
6184 | free_span: | |
6185 | free_cpumask_var(rd->span); | |
0c910d28 | 6186 | out: |
c6c4927b | 6187 | return -ENOMEM; |
57d885fe GH |
6188 | } |
6189 | ||
6190 | static void init_defrootdomain(void) | |
6191 | { | |
c6c4927b RR |
6192 | init_rootdomain(&def_root_domain, true); |
6193 | ||
57d885fe GH |
6194 | atomic_set(&def_root_domain.refcount, 1); |
6195 | } | |
6196 | ||
dc938520 | 6197 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6198 | { |
6199 | struct root_domain *rd; | |
6200 | ||
6201 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6202 | if (!rd) | |
6203 | return NULL; | |
6204 | ||
c6c4927b RR |
6205 | if (init_rootdomain(rd, false) != 0) { |
6206 | kfree(rd); | |
6207 | return NULL; | |
6208 | } | |
57d885fe GH |
6209 | |
6210 | return rd; | |
6211 | } | |
6212 | ||
1da177e4 | 6213 | /* |
0eab9146 | 6214 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6215 | * hold the hotplug lock. |
6216 | */ | |
0eab9146 IM |
6217 | static void |
6218 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6219 | { |
70b97a7f | 6220 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6221 | struct sched_domain *tmp; |
6222 | ||
6223 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6224 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6225 | struct sched_domain *parent = tmp->parent; |
6226 | if (!parent) | |
6227 | break; | |
f29c9b1c | 6228 | |
1a848870 | 6229 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6230 | tmp->parent = parent->parent; |
1a848870 SS |
6231 | if (parent->parent) |
6232 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6233 | } else |
6234 | tmp = tmp->parent; | |
245af2c7 SS |
6235 | } |
6236 | ||
1a848870 | 6237 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6238 | sd = sd->parent; |
1a848870 SS |
6239 | if (sd) |
6240 | sd->child = NULL; | |
6241 | } | |
1da177e4 LT |
6242 | |
6243 | sched_domain_debug(sd, cpu); | |
6244 | ||
57d885fe | 6245 | rq_attach_root(rq, rd); |
674311d5 | 6246 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6247 | } |
6248 | ||
6249 | /* cpus with isolated domains */ | |
dcc30a35 | 6250 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6251 | |
6252 | /* Setup the mask of cpus configured for isolated domains */ | |
6253 | static int __init isolated_cpu_setup(char *str) | |
6254 | { | |
bdddd296 | 6255 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6256 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6257 | return 1; |
6258 | } | |
6259 | ||
8927f494 | 6260 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6261 | |
6262 | /* | |
6711cab4 SS |
6263 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6264 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6265 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6266 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6267 | * |
6268 | * init_sched_build_groups will build a circular linked list of the groups | |
6269 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6270 | * and ->cpu_power to 0. | |
6271 | */ | |
a616058b | 6272 | static void |
96f874e2 RR |
6273 | init_sched_build_groups(const struct cpumask *span, |
6274 | const struct cpumask *cpu_map, | |
6275 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6276 | struct sched_group **sg, |
96f874e2 RR |
6277 | struct cpumask *tmpmask), |
6278 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6279 | { |
6280 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6281 | int i; |
6282 | ||
96f874e2 | 6283 | cpumask_clear(covered); |
7c16ec58 | 6284 | |
abcd083a | 6285 | for_each_cpu(i, span) { |
6711cab4 | 6286 | struct sched_group *sg; |
7c16ec58 | 6287 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6288 | int j; |
6289 | ||
758b2cdc | 6290 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6291 | continue; |
6292 | ||
758b2cdc | 6293 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6294 | sg->cpu_power = 0; |
1da177e4 | 6295 | |
abcd083a | 6296 | for_each_cpu(j, span) { |
7c16ec58 | 6297 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6298 | continue; |
6299 | ||
96f874e2 | 6300 | cpumask_set_cpu(j, covered); |
758b2cdc | 6301 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6302 | } |
6303 | if (!first) | |
6304 | first = sg; | |
6305 | if (last) | |
6306 | last->next = sg; | |
6307 | last = sg; | |
6308 | } | |
6309 | last->next = first; | |
6310 | } | |
6311 | ||
9c1cfda2 | 6312 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6313 | |
9c1cfda2 | 6314 | #ifdef CONFIG_NUMA |
198e2f18 | 6315 | |
9c1cfda2 JH |
6316 | /** |
6317 | * find_next_best_node - find the next node to include in a sched_domain | |
6318 | * @node: node whose sched_domain we're building | |
6319 | * @used_nodes: nodes already in the sched_domain | |
6320 | * | |
41a2d6cf | 6321 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6322 | * finds the closest node not already in the @used_nodes map. |
6323 | * | |
6324 | * Should use nodemask_t. | |
6325 | */ | |
c5f59f08 | 6326 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6327 | { |
6328 | int i, n, val, min_val, best_node = 0; | |
6329 | ||
6330 | min_val = INT_MAX; | |
6331 | ||
076ac2af | 6332 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6333 | /* Start at @node */ |
076ac2af | 6334 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6335 | |
6336 | if (!nr_cpus_node(n)) | |
6337 | continue; | |
6338 | ||
6339 | /* Skip already used nodes */ | |
c5f59f08 | 6340 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6341 | continue; |
6342 | ||
6343 | /* Simple min distance search */ | |
6344 | val = node_distance(node, n); | |
6345 | ||
6346 | if (val < min_val) { | |
6347 | min_val = val; | |
6348 | best_node = n; | |
6349 | } | |
6350 | } | |
6351 | ||
c5f59f08 | 6352 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6353 | return best_node; |
6354 | } | |
6355 | ||
6356 | /** | |
6357 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6358 | * @node: node whose cpumask we're constructing | |
73486722 | 6359 | * @span: resulting cpumask |
9c1cfda2 | 6360 | * |
41a2d6cf | 6361 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6362 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6363 | * out optimally. | |
6364 | */ | |
96f874e2 | 6365 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6366 | { |
c5f59f08 | 6367 | nodemask_t used_nodes; |
48f24c4d | 6368 | int i; |
9c1cfda2 | 6369 | |
6ca09dfc | 6370 | cpumask_clear(span); |
c5f59f08 | 6371 | nodes_clear(used_nodes); |
9c1cfda2 | 6372 | |
6ca09dfc | 6373 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6374 | node_set(node, used_nodes); |
9c1cfda2 JH |
6375 | |
6376 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6377 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6378 | |
6ca09dfc | 6379 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6380 | } |
9c1cfda2 | 6381 | } |
6d6bc0ad | 6382 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6383 | |
5c45bf27 | 6384 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6385 | |
6c99e9ad RR |
6386 | /* |
6387 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6388 | * |
6389 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6390 | * and struct sched_domain. ) | |
6c99e9ad RR |
6391 | */ |
6392 | struct static_sched_group { | |
6393 | struct sched_group sg; | |
6394 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6395 | }; | |
6396 | ||
6397 | struct static_sched_domain { | |
6398 | struct sched_domain sd; | |
6399 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6400 | }; | |
6401 | ||
49a02c51 AH |
6402 | struct s_data { |
6403 | #ifdef CONFIG_NUMA | |
6404 | int sd_allnodes; | |
6405 | cpumask_var_t domainspan; | |
6406 | cpumask_var_t covered; | |
6407 | cpumask_var_t notcovered; | |
6408 | #endif | |
6409 | cpumask_var_t nodemask; | |
6410 | cpumask_var_t this_sibling_map; | |
6411 | cpumask_var_t this_core_map; | |
6412 | cpumask_var_t send_covered; | |
6413 | cpumask_var_t tmpmask; | |
6414 | struct sched_group **sched_group_nodes; | |
6415 | struct root_domain *rd; | |
6416 | }; | |
6417 | ||
2109b99e AH |
6418 | enum s_alloc { |
6419 | sa_sched_groups = 0, | |
6420 | sa_rootdomain, | |
6421 | sa_tmpmask, | |
6422 | sa_send_covered, | |
6423 | sa_this_core_map, | |
6424 | sa_this_sibling_map, | |
6425 | sa_nodemask, | |
6426 | sa_sched_group_nodes, | |
6427 | #ifdef CONFIG_NUMA | |
6428 | sa_notcovered, | |
6429 | sa_covered, | |
6430 | sa_domainspan, | |
6431 | #endif | |
6432 | sa_none, | |
6433 | }; | |
6434 | ||
9c1cfda2 | 6435 | /* |
48f24c4d | 6436 | * SMT sched-domains: |
9c1cfda2 | 6437 | */ |
1da177e4 | 6438 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6439 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6440 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6441 | |
41a2d6cf | 6442 | static int |
96f874e2 RR |
6443 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6444 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6445 | { |
6711cab4 | 6446 | if (sg) |
1871e52c | 6447 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6448 | return cpu; |
6449 | } | |
6d6bc0ad | 6450 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6451 | |
48f24c4d IM |
6452 | /* |
6453 | * multi-core sched-domains: | |
6454 | */ | |
1e9f28fa | 6455 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6456 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6457 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6458 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6459 | |
6460 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6461 | static int |
96f874e2 RR |
6462 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6463 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6464 | { |
6711cab4 | 6465 | int group; |
7c16ec58 | 6466 | |
c69fc56d | 6467 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6468 | group = cpumask_first(mask); |
6711cab4 | 6469 | if (sg) |
6c99e9ad | 6470 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6471 | return group; |
1e9f28fa SS |
6472 | } |
6473 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6474 | static int |
96f874e2 RR |
6475 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6476 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6477 | { |
6711cab4 | 6478 | if (sg) |
6c99e9ad | 6479 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6480 | return cpu; |
6481 | } | |
6482 | #endif | |
6483 | ||
6c99e9ad RR |
6484 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6485 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6486 | |
41a2d6cf | 6487 | static int |
96f874e2 RR |
6488 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6489 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6490 | { |
6711cab4 | 6491 | int group; |
48f24c4d | 6492 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6493 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6494 | group = cpumask_first(mask); |
1e9f28fa | 6495 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6496 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6497 | group = cpumask_first(mask); |
1da177e4 | 6498 | #else |
6711cab4 | 6499 | group = cpu; |
1da177e4 | 6500 | #endif |
6711cab4 | 6501 | if (sg) |
6c99e9ad | 6502 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6503 | return group; |
1da177e4 LT |
6504 | } |
6505 | ||
6506 | #ifdef CONFIG_NUMA | |
1da177e4 | 6507 | /* |
9c1cfda2 JH |
6508 | * The init_sched_build_groups can't handle what we want to do with node |
6509 | * groups, so roll our own. Now each node has its own list of groups which | |
6510 | * gets dynamically allocated. | |
1da177e4 | 6511 | */ |
62ea9ceb | 6512 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6513 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6514 | |
62ea9ceb | 6515 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6516 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6517 | |
96f874e2 RR |
6518 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6519 | struct sched_group **sg, | |
6520 | struct cpumask *nodemask) | |
9c1cfda2 | 6521 | { |
6711cab4 SS |
6522 | int group; |
6523 | ||
6ca09dfc | 6524 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6525 | group = cpumask_first(nodemask); |
6711cab4 SS |
6526 | |
6527 | if (sg) | |
6c99e9ad | 6528 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6529 | return group; |
1da177e4 | 6530 | } |
6711cab4 | 6531 | |
08069033 SS |
6532 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6533 | { | |
6534 | struct sched_group *sg = group_head; | |
6535 | int j; | |
6536 | ||
6537 | if (!sg) | |
6538 | return; | |
3a5c359a | 6539 | do { |
758b2cdc | 6540 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6541 | struct sched_domain *sd; |
08069033 | 6542 | |
6c99e9ad | 6543 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6544 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6545 | /* |
6546 | * Only add "power" once for each | |
6547 | * physical package. | |
6548 | */ | |
6549 | continue; | |
6550 | } | |
08069033 | 6551 | |
18a3885f | 6552 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6553 | } |
6554 | sg = sg->next; | |
6555 | } while (sg != group_head); | |
08069033 | 6556 | } |
0601a88d AH |
6557 | |
6558 | static int build_numa_sched_groups(struct s_data *d, | |
6559 | const struct cpumask *cpu_map, int num) | |
6560 | { | |
6561 | struct sched_domain *sd; | |
6562 | struct sched_group *sg, *prev; | |
6563 | int n, j; | |
6564 | ||
6565 | cpumask_clear(d->covered); | |
6566 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6567 | if (cpumask_empty(d->nodemask)) { | |
6568 | d->sched_group_nodes[num] = NULL; | |
6569 | goto out; | |
6570 | } | |
6571 | ||
6572 | sched_domain_node_span(num, d->domainspan); | |
6573 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6574 | ||
6575 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6576 | GFP_KERNEL, num); | |
6577 | if (!sg) { | |
3df0fc5b PZ |
6578 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6579 | num); | |
0601a88d AH |
6580 | return -ENOMEM; |
6581 | } | |
6582 | d->sched_group_nodes[num] = sg; | |
6583 | ||
6584 | for_each_cpu(j, d->nodemask) { | |
6585 | sd = &per_cpu(node_domains, j).sd; | |
6586 | sd->groups = sg; | |
6587 | } | |
6588 | ||
18a3885f | 6589 | sg->cpu_power = 0; |
0601a88d AH |
6590 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6591 | sg->next = sg; | |
6592 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6593 | ||
6594 | prev = sg; | |
6595 | for (j = 0; j < nr_node_ids; j++) { | |
6596 | n = (num + j) % nr_node_ids; | |
6597 | cpumask_complement(d->notcovered, d->covered); | |
6598 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6599 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6600 | if (cpumask_empty(d->tmpmask)) | |
6601 | break; | |
6602 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6603 | if (cpumask_empty(d->tmpmask)) | |
6604 | continue; | |
6605 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6606 | GFP_KERNEL, num); | |
6607 | if (!sg) { | |
3df0fc5b PZ |
6608 | printk(KERN_WARNING |
6609 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6610 | return -ENOMEM; |
6611 | } | |
18a3885f | 6612 | sg->cpu_power = 0; |
0601a88d AH |
6613 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6614 | sg->next = prev->next; | |
6615 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6616 | prev->next = sg; | |
6617 | prev = sg; | |
6618 | } | |
6619 | out: | |
6620 | return 0; | |
6621 | } | |
6d6bc0ad | 6622 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6623 | |
a616058b | 6624 | #ifdef CONFIG_NUMA |
51888ca2 | 6625 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6626 | static void free_sched_groups(const struct cpumask *cpu_map, |
6627 | struct cpumask *nodemask) | |
51888ca2 | 6628 | { |
a616058b | 6629 | int cpu, i; |
51888ca2 | 6630 | |
abcd083a | 6631 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6632 | struct sched_group **sched_group_nodes |
6633 | = sched_group_nodes_bycpu[cpu]; | |
6634 | ||
51888ca2 SV |
6635 | if (!sched_group_nodes) |
6636 | continue; | |
6637 | ||
076ac2af | 6638 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6639 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6640 | ||
6ca09dfc | 6641 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6642 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6643 | continue; |
6644 | ||
6645 | if (sg == NULL) | |
6646 | continue; | |
6647 | sg = sg->next; | |
6648 | next_sg: | |
6649 | oldsg = sg; | |
6650 | sg = sg->next; | |
6651 | kfree(oldsg); | |
6652 | if (oldsg != sched_group_nodes[i]) | |
6653 | goto next_sg; | |
6654 | } | |
6655 | kfree(sched_group_nodes); | |
6656 | sched_group_nodes_bycpu[cpu] = NULL; | |
6657 | } | |
51888ca2 | 6658 | } |
6d6bc0ad | 6659 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6660 | static void free_sched_groups(const struct cpumask *cpu_map, |
6661 | struct cpumask *nodemask) | |
a616058b SS |
6662 | { |
6663 | } | |
6d6bc0ad | 6664 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6665 | |
89c4710e SS |
6666 | /* |
6667 | * Initialize sched groups cpu_power. | |
6668 | * | |
6669 | * cpu_power indicates the capacity of sched group, which is used while | |
6670 | * distributing the load between different sched groups in a sched domain. | |
6671 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6672 | * there are asymmetries in the topology. If there are asymmetries, group | |
6673 | * having more cpu_power will pickup more load compared to the group having | |
6674 | * less cpu_power. | |
89c4710e SS |
6675 | */ |
6676 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6677 | { | |
6678 | struct sched_domain *child; | |
6679 | struct sched_group *group; | |
f93e65c1 PZ |
6680 | long power; |
6681 | int weight; | |
89c4710e SS |
6682 | |
6683 | WARN_ON(!sd || !sd->groups); | |
6684 | ||
13318a71 | 6685 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6686 | return; |
6687 | ||
6688 | child = sd->child; | |
6689 | ||
18a3885f | 6690 | sd->groups->cpu_power = 0; |
5517d86b | 6691 | |
f93e65c1 PZ |
6692 | if (!child) { |
6693 | power = SCHED_LOAD_SCALE; | |
6694 | weight = cpumask_weight(sched_domain_span(sd)); | |
6695 | /* | |
6696 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6697 | * Usually multiple threads get a better yield out of |
6698 | * that one core than a single thread would have, | |
6699 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6700 | */ |
a52bfd73 PZ |
6701 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6702 | power *= sd->smt_gain; | |
f93e65c1 | 6703 | power /= weight; |
a52bfd73 PZ |
6704 | power >>= SCHED_LOAD_SHIFT; |
6705 | } | |
18a3885f | 6706 | sd->groups->cpu_power += power; |
89c4710e SS |
6707 | return; |
6708 | } | |
6709 | ||
89c4710e | 6710 | /* |
f93e65c1 | 6711 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6712 | */ |
6713 | group = child->groups; | |
6714 | do { | |
18a3885f | 6715 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6716 | group = group->next; |
6717 | } while (group != child->groups); | |
6718 | } | |
6719 | ||
7c16ec58 MT |
6720 | /* |
6721 | * Initializers for schedule domains | |
6722 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6723 | */ | |
6724 | ||
a5d8c348 IM |
6725 | #ifdef CONFIG_SCHED_DEBUG |
6726 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6727 | #else | |
6728 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6729 | #endif | |
6730 | ||
7c16ec58 | 6731 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6732 | |
7c16ec58 MT |
6733 | #define SD_INIT_FUNC(type) \ |
6734 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6735 | { \ | |
6736 | memset(sd, 0, sizeof(*sd)); \ | |
6737 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6738 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6739 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6740 | } |
6741 | ||
6742 | SD_INIT_FUNC(CPU) | |
6743 | #ifdef CONFIG_NUMA | |
6744 | SD_INIT_FUNC(ALLNODES) | |
6745 | SD_INIT_FUNC(NODE) | |
6746 | #endif | |
6747 | #ifdef CONFIG_SCHED_SMT | |
6748 | SD_INIT_FUNC(SIBLING) | |
6749 | #endif | |
6750 | #ifdef CONFIG_SCHED_MC | |
6751 | SD_INIT_FUNC(MC) | |
6752 | #endif | |
6753 | ||
1d3504fc HS |
6754 | static int default_relax_domain_level = -1; |
6755 | ||
6756 | static int __init setup_relax_domain_level(char *str) | |
6757 | { | |
30e0e178 LZ |
6758 | unsigned long val; |
6759 | ||
6760 | val = simple_strtoul(str, NULL, 0); | |
6761 | if (val < SD_LV_MAX) | |
6762 | default_relax_domain_level = val; | |
6763 | ||
1d3504fc HS |
6764 | return 1; |
6765 | } | |
6766 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6767 | ||
6768 | static void set_domain_attribute(struct sched_domain *sd, | |
6769 | struct sched_domain_attr *attr) | |
6770 | { | |
6771 | int request; | |
6772 | ||
6773 | if (!attr || attr->relax_domain_level < 0) { | |
6774 | if (default_relax_domain_level < 0) | |
6775 | return; | |
6776 | else | |
6777 | request = default_relax_domain_level; | |
6778 | } else | |
6779 | request = attr->relax_domain_level; | |
6780 | if (request < sd->level) { | |
6781 | /* turn off idle balance on this domain */ | |
c88d5910 | 6782 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6783 | } else { |
6784 | /* turn on idle balance on this domain */ | |
c88d5910 | 6785 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6786 | } |
6787 | } | |
6788 | ||
2109b99e AH |
6789 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6790 | const struct cpumask *cpu_map) | |
6791 | { | |
6792 | switch (what) { | |
6793 | case sa_sched_groups: | |
6794 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6795 | d->sched_group_nodes = NULL; | |
6796 | case sa_rootdomain: | |
6797 | free_rootdomain(d->rd); /* fall through */ | |
6798 | case sa_tmpmask: | |
6799 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6800 | case sa_send_covered: | |
6801 | free_cpumask_var(d->send_covered); /* fall through */ | |
6802 | case sa_this_core_map: | |
6803 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6804 | case sa_this_sibling_map: | |
6805 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6806 | case sa_nodemask: | |
6807 | free_cpumask_var(d->nodemask); /* fall through */ | |
6808 | case sa_sched_group_nodes: | |
d1b55138 | 6809 | #ifdef CONFIG_NUMA |
2109b99e AH |
6810 | kfree(d->sched_group_nodes); /* fall through */ |
6811 | case sa_notcovered: | |
6812 | free_cpumask_var(d->notcovered); /* fall through */ | |
6813 | case sa_covered: | |
6814 | free_cpumask_var(d->covered); /* fall through */ | |
6815 | case sa_domainspan: | |
6816 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6817 | #endif |
2109b99e AH |
6818 | case sa_none: |
6819 | break; | |
6820 | } | |
6821 | } | |
3404c8d9 | 6822 | |
2109b99e AH |
6823 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6824 | const struct cpumask *cpu_map) | |
6825 | { | |
3404c8d9 | 6826 | #ifdef CONFIG_NUMA |
2109b99e AH |
6827 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6828 | return sa_none; | |
6829 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6830 | return sa_domainspan; | |
6831 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6832 | return sa_covered; | |
6833 | /* Allocate the per-node list of sched groups */ | |
6834 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6835 | sizeof(struct sched_group *), GFP_KERNEL); | |
6836 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6837 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6838 | return sa_notcovered; |
d1b55138 | 6839 | } |
2109b99e | 6840 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6841 | #endif |
2109b99e AH |
6842 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6843 | return sa_sched_group_nodes; | |
6844 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6845 | return sa_nodemask; | |
6846 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6847 | return sa_this_sibling_map; | |
6848 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6849 | return sa_this_core_map; | |
6850 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6851 | return sa_send_covered; | |
6852 | d->rd = alloc_rootdomain(); | |
6853 | if (!d->rd) { | |
3df0fc5b | 6854 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6855 | return sa_tmpmask; |
57d885fe | 6856 | } |
2109b99e AH |
6857 | return sa_rootdomain; |
6858 | } | |
57d885fe | 6859 | |
7f4588f3 AH |
6860 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6861 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6862 | { | |
6863 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6864 | #ifdef CONFIG_NUMA |
7f4588f3 | 6865 | struct sched_domain *parent; |
1da177e4 | 6866 | |
7f4588f3 AH |
6867 | d->sd_allnodes = 0; |
6868 | if (cpumask_weight(cpu_map) > | |
6869 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6870 | sd = &per_cpu(allnodes_domains, i).sd; | |
6871 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6872 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6873 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6874 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6875 | d->sd_allnodes = 1; | |
6876 | } | |
6877 | parent = sd; | |
6878 | ||
6879 | sd = &per_cpu(node_domains, i).sd; | |
6880 | SD_INIT(sd, NODE); | |
6881 | set_domain_attribute(sd, attr); | |
6882 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6883 | sd->parent = parent; | |
6884 | if (parent) | |
6885 | parent->child = sd; | |
6886 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6887 | #endif |
7f4588f3 AH |
6888 | return sd; |
6889 | } | |
1da177e4 | 6890 | |
87cce662 AH |
6891 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6892 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6893 | struct sched_domain *parent, int i) | |
6894 | { | |
6895 | struct sched_domain *sd; | |
6896 | sd = &per_cpu(phys_domains, i).sd; | |
6897 | SD_INIT(sd, CPU); | |
6898 | set_domain_attribute(sd, attr); | |
6899 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6900 | sd->parent = parent; | |
6901 | if (parent) | |
6902 | parent->child = sd; | |
6903 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6904 | return sd; | |
6905 | } | |
1da177e4 | 6906 | |
410c4081 AH |
6907 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
6908 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6909 | struct sched_domain *parent, int i) | |
6910 | { | |
6911 | struct sched_domain *sd = parent; | |
1e9f28fa | 6912 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
6913 | sd = &per_cpu(core_domains, i).sd; |
6914 | SD_INIT(sd, MC); | |
6915 | set_domain_attribute(sd, attr); | |
6916 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
6917 | sd->parent = parent; | |
6918 | parent->child = sd; | |
6919 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 6920 | #endif |
410c4081 AH |
6921 | return sd; |
6922 | } | |
1e9f28fa | 6923 | |
d8173535 AH |
6924 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
6925 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6926 | struct sched_domain *parent, int i) | |
6927 | { | |
6928 | struct sched_domain *sd = parent; | |
1da177e4 | 6929 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
6930 | sd = &per_cpu(cpu_domains, i).sd; |
6931 | SD_INIT(sd, SIBLING); | |
6932 | set_domain_attribute(sd, attr); | |
6933 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
6934 | sd->parent = parent; | |
6935 | parent->child = sd; | |
6936 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 6937 | #endif |
d8173535 AH |
6938 | return sd; |
6939 | } | |
1da177e4 | 6940 | |
0e8e85c9 AH |
6941 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
6942 | const struct cpumask *cpu_map, int cpu) | |
6943 | { | |
6944 | switch (l) { | |
1da177e4 | 6945 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
6946 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
6947 | cpumask_and(d->this_sibling_map, cpu_map, | |
6948 | topology_thread_cpumask(cpu)); | |
6949 | if (cpu == cpumask_first(d->this_sibling_map)) | |
6950 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
6951 | &cpu_to_cpu_group, | |
6952 | d->send_covered, d->tmpmask); | |
6953 | break; | |
1da177e4 | 6954 | #endif |
1e9f28fa | 6955 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
6956 | case SD_LV_MC: /* set up multi-core groups */ |
6957 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
6958 | if (cpu == cpumask_first(d->this_core_map)) | |
6959 | init_sched_build_groups(d->this_core_map, cpu_map, | |
6960 | &cpu_to_core_group, | |
6961 | d->send_covered, d->tmpmask); | |
6962 | break; | |
1e9f28fa | 6963 | #endif |
86548096 AH |
6964 | case SD_LV_CPU: /* set up physical groups */ |
6965 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
6966 | if (!cpumask_empty(d->nodemask)) | |
6967 | init_sched_build_groups(d->nodemask, cpu_map, | |
6968 | &cpu_to_phys_group, | |
6969 | d->send_covered, d->tmpmask); | |
6970 | break; | |
1da177e4 | 6971 | #ifdef CONFIG_NUMA |
de616e36 AH |
6972 | case SD_LV_ALLNODES: |
6973 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
6974 | d->send_covered, d->tmpmask); | |
6975 | break; | |
6976 | #endif | |
0e8e85c9 AH |
6977 | default: |
6978 | break; | |
7c16ec58 | 6979 | } |
0e8e85c9 | 6980 | } |
9c1cfda2 | 6981 | |
2109b99e AH |
6982 | /* |
6983 | * Build sched domains for a given set of cpus and attach the sched domains | |
6984 | * to the individual cpus | |
6985 | */ | |
6986 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
6987 | struct sched_domain_attr *attr) | |
6988 | { | |
6989 | enum s_alloc alloc_state = sa_none; | |
6990 | struct s_data d; | |
294b0c96 | 6991 | struct sched_domain *sd; |
2109b99e | 6992 | int i; |
7c16ec58 | 6993 | #ifdef CONFIG_NUMA |
2109b99e | 6994 | d.sd_allnodes = 0; |
7c16ec58 | 6995 | #endif |
9c1cfda2 | 6996 | |
2109b99e AH |
6997 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6998 | if (alloc_state != sa_rootdomain) | |
6999 | goto error; | |
7000 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7001 | |
1da177e4 | 7002 | /* |
1a20ff27 | 7003 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7004 | */ |
abcd083a | 7005 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7006 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7007 | cpu_map); | |
9761eea8 | 7008 | |
7f4588f3 | 7009 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7010 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7011 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7012 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7013 | } |
9c1cfda2 | 7014 | |
abcd083a | 7015 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7016 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 7017 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7018 | } |
9c1cfda2 | 7019 | |
1da177e4 | 7020 | /* Set up physical groups */ |
86548096 AH |
7021 | for (i = 0; i < nr_node_ids; i++) |
7022 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7023 | |
1da177e4 LT |
7024 | #ifdef CONFIG_NUMA |
7025 | /* Set up node groups */ | |
de616e36 AH |
7026 | if (d.sd_allnodes) |
7027 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7028 | |
0601a88d AH |
7029 | for (i = 0; i < nr_node_ids; i++) |
7030 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7031 | goto error; |
1da177e4 LT |
7032 | #endif |
7033 | ||
7034 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7035 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7036 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7037 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7038 | init_sched_groups_power(i, sd); |
5c45bf27 | 7039 | } |
1da177e4 | 7040 | #endif |
1e9f28fa | 7041 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7042 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7043 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7044 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7045 | } |
7046 | #endif | |
1e9f28fa | 7047 | |
abcd083a | 7048 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7049 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7050 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7051 | } |
7052 | ||
9c1cfda2 | 7053 | #ifdef CONFIG_NUMA |
076ac2af | 7054 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7055 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7056 | |
49a02c51 | 7057 | if (d.sd_allnodes) { |
6711cab4 | 7058 | struct sched_group *sg; |
f712c0c7 | 7059 | |
96f874e2 | 7060 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7061 | d.tmpmask); |
f712c0c7 SS |
7062 | init_numa_sched_groups_power(sg); |
7063 | } | |
9c1cfda2 JH |
7064 | #endif |
7065 | ||
1da177e4 | 7066 | /* Attach the domains */ |
abcd083a | 7067 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7068 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7069 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7070 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7071 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7072 | #else |
6c99e9ad | 7073 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7074 | #endif |
49a02c51 | 7075 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7076 | } |
51888ca2 | 7077 | |
2109b99e AH |
7078 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7079 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7080 | return 0; | |
51888ca2 | 7081 | |
51888ca2 | 7082 | error: |
2109b99e AH |
7083 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7084 | return -ENOMEM; | |
1da177e4 | 7085 | } |
029190c5 | 7086 | |
96f874e2 | 7087 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7088 | { |
7089 | return __build_sched_domains(cpu_map, NULL); | |
7090 | } | |
7091 | ||
acc3f5d7 | 7092 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7093 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7094 | static struct sched_domain_attr *dattr_cur; |
7095 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7096 | |
7097 | /* | |
7098 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7099 | * cpumask) fails, then fallback to a single sched domain, |
7100 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7101 | */ |
4212823f | 7102 | static cpumask_var_t fallback_doms; |
029190c5 | 7103 | |
ee79d1bd HC |
7104 | /* |
7105 | * arch_update_cpu_topology lets virtualized architectures update the | |
7106 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7107 | * or 0 if it stayed the same. | |
7108 | */ | |
7109 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7110 | { |
ee79d1bd | 7111 | return 0; |
22e52b07 HC |
7112 | } |
7113 | ||
acc3f5d7 RR |
7114 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7115 | { | |
7116 | int i; | |
7117 | cpumask_var_t *doms; | |
7118 | ||
7119 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7120 | if (!doms) | |
7121 | return NULL; | |
7122 | for (i = 0; i < ndoms; i++) { | |
7123 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7124 | free_sched_domains(doms, i); | |
7125 | return NULL; | |
7126 | } | |
7127 | } | |
7128 | return doms; | |
7129 | } | |
7130 | ||
7131 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7132 | { | |
7133 | unsigned int i; | |
7134 | for (i = 0; i < ndoms; i++) | |
7135 | free_cpumask_var(doms[i]); | |
7136 | kfree(doms); | |
7137 | } | |
7138 | ||
1a20ff27 | 7139 | /* |
41a2d6cf | 7140 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7141 | * For now this just excludes isolated cpus, but could be used to |
7142 | * exclude other special cases in the future. | |
1a20ff27 | 7143 | */ |
96f874e2 | 7144 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7145 | { |
7378547f MM |
7146 | int err; |
7147 | ||
22e52b07 | 7148 | arch_update_cpu_topology(); |
029190c5 | 7149 | ndoms_cur = 1; |
acc3f5d7 | 7150 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7151 | if (!doms_cur) |
acc3f5d7 RR |
7152 | doms_cur = &fallback_doms; |
7153 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7154 | dattr_cur = NULL; |
acc3f5d7 | 7155 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7156 | register_sched_domain_sysctl(); |
7378547f MM |
7157 | |
7158 | return err; | |
1a20ff27 DG |
7159 | } |
7160 | ||
96f874e2 RR |
7161 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7162 | struct cpumask *tmpmask) | |
1da177e4 | 7163 | { |
7c16ec58 | 7164 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7165 | } |
1da177e4 | 7166 | |
1a20ff27 DG |
7167 | /* |
7168 | * Detach sched domains from a group of cpus specified in cpu_map | |
7169 | * These cpus will now be attached to the NULL domain | |
7170 | */ | |
96f874e2 | 7171 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7172 | { |
96f874e2 RR |
7173 | /* Save because hotplug lock held. */ |
7174 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7175 | int i; |
7176 | ||
abcd083a | 7177 | for_each_cpu(i, cpu_map) |
57d885fe | 7178 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7179 | synchronize_sched(); |
96f874e2 | 7180 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7181 | } |
7182 | ||
1d3504fc HS |
7183 | /* handle null as "default" */ |
7184 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7185 | struct sched_domain_attr *new, int idx_new) | |
7186 | { | |
7187 | struct sched_domain_attr tmp; | |
7188 | ||
7189 | /* fast path */ | |
7190 | if (!new && !cur) | |
7191 | return 1; | |
7192 | ||
7193 | tmp = SD_ATTR_INIT; | |
7194 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7195 | new ? (new + idx_new) : &tmp, | |
7196 | sizeof(struct sched_domain_attr)); | |
7197 | } | |
7198 | ||
029190c5 PJ |
7199 | /* |
7200 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7201 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7202 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7203 | * It destroys each deleted domain and builds each new domain. | |
7204 | * | |
acc3f5d7 | 7205 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7206 | * The masks don't intersect (don't overlap.) We should setup one |
7207 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7208 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7209 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7210 | * it as it is. | |
7211 | * | |
acc3f5d7 RR |
7212 | * The passed in 'doms_new' should be allocated using |
7213 | * alloc_sched_domains. This routine takes ownership of it and will | |
7214 | * free_sched_domains it when done with it. If the caller failed the | |
7215 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7216 | * and partition_sched_domains() will fallback to the single partition | |
7217 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7218 | * |
96f874e2 | 7219 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7220 | * ndoms_new == 0 is a special case for destroying existing domains, |
7221 | * and it will not create the default domain. | |
dfb512ec | 7222 | * |
029190c5 PJ |
7223 | * Call with hotplug lock held |
7224 | */ | |
acc3f5d7 | 7225 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7226 | struct sched_domain_attr *dattr_new) |
029190c5 | 7227 | { |
dfb512ec | 7228 | int i, j, n; |
d65bd5ec | 7229 | int new_topology; |
029190c5 | 7230 | |
712555ee | 7231 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7232 | |
7378547f MM |
7233 | /* always unregister in case we don't destroy any domains */ |
7234 | unregister_sched_domain_sysctl(); | |
7235 | ||
d65bd5ec HC |
7236 | /* Let architecture update cpu core mappings. */ |
7237 | new_topology = arch_update_cpu_topology(); | |
7238 | ||
dfb512ec | 7239 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7240 | |
7241 | /* Destroy deleted domains */ | |
7242 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7243 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7244 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7245 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7246 | goto match1; |
7247 | } | |
7248 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7249 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7250 | match1: |
7251 | ; | |
7252 | } | |
7253 | ||
e761b772 MK |
7254 | if (doms_new == NULL) { |
7255 | ndoms_cur = 0; | |
acc3f5d7 | 7256 | doms_new = &fallback_doms; |
6ad4c188 | 7257 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7258 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7259 | } |
7260 | ||
029190c5 PJ |
7261 | /* Build new domains */ |
7262 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7263 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7264 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7265 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7266 | goto match2; |
7267 | } | |
7268 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7269 | __build_sched_domains(doms_new[i], |
1d3504fc | 7270 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7271 | match2: |
7272 | ; | |
7273 | } | |
7274 | ||
7275 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7276 | if (doms_cur != &fallback_doms) |
7277 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7278 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7279 | doms_cur = doms_new; |
1d3504fc | 7280 | dattr_cur = dattr_new; |
029190c5 | 7281 | ndoms_cur = ndoms_new; |
7378547f MM |
7282 | |
7283 | register_sched_domain_sysctl(); | |
a1835615 | 7284 | |
712555ee | 7285 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7286 | } |
7287 | ||
5c45bf27 | 7288 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7289 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7290 | { |
95402b38 | 7291 | get_online_cpus(); |
dfb512ec MK |
7292 | |
7293 | /* Destroy domains first to force the rebuild */ | |
7294 | partition_sched_domains(0, NULL, NULL); | |
7295 | ||
e761b772 | 7296 | rebuild_sched_domains(); |
95402b38 | 7297 | put_online_cpus(); |
5c45bf27 SS |
7298 | } |
7299 | ||
7300 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7301 | { | |
afb8a9b7 | 7302 | unsigned int level = 0; |
5c45bf27 | 7303 | |
afb8a9b7 GS |
7304 | if (sscanf(buf, "%u", &level) != 1) |
7305 | return -EINVAL; | |
7306 | ||
7307 | /* | |
7308 | * level is always be positive so don't check for | |
7309 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7310 | * What happens on 0 or 1 byte write, | |
7311 | * need to check for count as well? | |
7312 | */ | |
7313 | ||
7314 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7315 | return -EINVAL; |
7316 | ||
7317 | if (smt) | |
afb8a9b7 | 7318 | sched_smt_power_savings = level; |
5c45bf27 | 7319 | else |
afb8a9b7 | 7320 | sched_mc_power_savings = level; |
5c45bf27 | 7321 | |
c70f22d2 | 7322 | arch_reinit_sched_domains(); |
5c45bf27 | 7323 | |
c70f22d2 | 7324 | return count; |
5c45bf27 SS |
7325 | } |
7326 | ||
5c45bf27 | 7327 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7328 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7329 | struct sysdev_class_attribute *attr, |
f718cd4a | 7330 | char *page) |
5c45bf27 SS |
7331 | { |
7332 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7333 | } | |
f718cd4a | 7334 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7335 | struct sysdev_class_attribute *attr, |
48f24c4d | 7336 | const char *buf, size_t count) |
5c45bf27 SS |
7337 | { |
7338 | return sched_power_savings_store(buf, count, 0); | |
7339 | } | |
f718cd4a AK |
7340 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7341 | sched_mc_power_savings_show, | |
7342 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7343 | #endif |
7344 | ||
7345 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7346 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7347 | struct sysdev_class_attribute *attr, |
f718cd4a | 7348 | char *page) |
5c45bf27 SS |
7349 | { |
7350 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7351 | } | |
f718cd4a | 7352 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7353 | struct sysdev_class_attribute *attr, |
48f24c4d | 7354 | const char *buf, size_t count) |
5c45bf27 SS |
7355 | { |
7356 | return sched_power_savings_store(buf, count, 1); | |
7357 | } | |
f718cd4a AK |
7358 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7359 | sched_smt_power_savings_show, | |
6707de00 AB |
7360 | sched_smt_power_savings_store); |
7361 | #endif | |
7362 | ||
39aac648 | 7363 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7364 | { |
7365 | int err = 0; | |
7366 | ||
7367 | #ifdef CONFIG_SCHED_SMT | |
7368 | if (smt_capable()) | |
7369 | err = sysfs_create_file(&cls->kset.kobj, | |
7370 | &attr_sched_smt_power_savings.attr); | |
7371 | #endif | |
7372 | #ifdef CONFIG_SCHED_MC | |
7373 | if (!err && mc_capable()) | |
7374 | err = sysfs_create_file(&cls->kset.kobj, | |
7375 | &attr_sched_mc_power_savings.attr); | |
7376 | #endif | |
7377 | return err; | |
7378 | } | |
6d6bc0ad | 7379 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7380 | |
e761b772 | 7381 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7382 | /* |
e761b772 MK |
7383 | * Add online and remove offline CPUs from the scheduler domains. |
7384 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7385 | */ |
7386 | static int update_sched_domains(struct notifier_block *nfb, | |
7387 | unsigned long action, void *hcpu) | |
e761b772 MK |
7388 | { |
7389 | switch (action) { | |
7390 | case CPU_ONLINE: | |
7391 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
7392 | case CPU_DOWN_PREPARE: |
7393 | case CPU_DOWN_PREPARE_FROZEN: | |
7394 | case CPU_DOWN_FAILED: | |
7395 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 7396 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7397 | return NOTIFY_OK; |
7398 | ||
7399 | default: | |
7400 | return NOTIFY_DONE; | |
7401 | } | |
7402 | } | |
7403 | #endif | |
7404 | ||
7405 | static int update_runtime(struct notifier_block *nfb, | |
7406 | unsigned long action, void *hcpu) | |
1da177e4 | 7407 | { |
7def2be1 PZ |
7408 | int cpu = (int)(long)hcpu; |
7409 | ||
1da177e4 | 7410 | switch (action) { |
1da177e4 | 7411 | case CPU_DOWN_PREPARE: |
8bb78442 | 7412 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7413 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7414 | return NOTIFY_OK; |
7415 | ||
1da177e4 | 7416 | case CPU_DOWN_FAILED: |
8bb78442 | 7417 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7418 | case CPU_ONLINE: |
8bb78442 | 7419 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7420 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7421 | return NOTIFY_OK; |
7422 | ||
1da177e4 LT |
7423 | default: |
7424 | return NOTIFY_DONE; | |
7425 | } | |
1da177e4 | 7426 | } |
1da177e4 LT |
7427 | |
7428 | void __init sched_init_smp(void) | |
7429 | { | |
dcc30a35 RR |
7430 | cpumask_var_t non_isolated_cpus; |
7431 | ||
7432 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7433 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7434 | |
434d53b0 MT |
7435 | #if defined(CONFIG_NUMA) |
7436 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7437 | GFP_KERNEL); | |
7438 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7439 | #endif | |
95402b38 | 7440 | get_online_cpus(); |
712555ee | 7441 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7442 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7443 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7444 | if (cpumask_empty(non_isolated_cpus)) | |
7445 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7446 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7447 | put_online_cpus(); |
e761b772 MK |
7448 | |
7449 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7450 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7451 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7452 | #endif |
7453 | ||
7454 | /* RT runtime code needs to handle some hotplug events */ | |
7455 | hotcpu_notifier(update_runtime, 0); | |
7456 | ||
b328ca18 | 7457 | init_hrtick(); |
5c1e1767 NP |
7458 | |
7459 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7460 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7461 | BUG(); |
19978ca6 | 7462 | sched_init_granularity(); |
dcc30a35 | 7463 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7464 | |
0e3900e6 | 7465 | init_sched_rt_class(); |
1da177e4 LT |
7466 | } |
7467 | #else | |
7468 | void __init sched_init_smp(void) | |
7469 | { | |
19978ca6 | 7470 | sched_init_granularity(); |
1da177e4 LT |
7471 | } |
7472 | #endif /* CONFIG_SMP */ | |
7473 | ||
cd1bb94b AB |
7474 | const_debug unsigned int sysctl_timer_migration = 1; |
7475 | ||
1da177e4 LT |
7476 | int in_sched_functions(unsigned long addr) |
7477 | { | |
1da177e4 LT |
7478 | return in_lock_functions(addr) || |
7479 | (addr >= (unsigned long)__sched_text_start | |
7480 | && addr < (unsigned long)__sched_text_end); | |
7481 | } | |
7482 | ||
a9957449 | 7483 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7484 | { |
7485 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7486 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7487 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7488 | cfs_rq->rq = rq; | |
7489 | #endif | |
67e9fb2a | 7490 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7491 | } |
7492 | ||
fa85ae24 PZ |
7493 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7494 | { | |
7495 | struct rt_prio_array *array; | |
7496 | int i; | |
7497 | ||
7498 | array = &rt_rq->active; | |
7499 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7500 | INIT_LIST_HEAD(array->queue + i); | |
7501 | __clear_bit(i, array->bitmap); | |
7502 | } | |
7503 | /* delimiter for bitsearch: */ | |
7504 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7505 | ||
052f1dc7 | 7506 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7507 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7508 | #ifdef CONFIG_SMP |
e864c499 | 7509 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7510 | #endif |
48d5e258 | 7511 | #endif |
fa85ae24 PZ |
7512 | #ifdef CONFIG_SMP |
7513 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7514 | rt_rq->overloaded = 0; |
05fa785c | 7515 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7516 | #endif |
7517 | ||
7518 | rt_rq->rt_time = 0; | |
7519 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7520 | rt_rq->rt_runtime = 0; |
0986b11b | 7521 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7522 | |
052f1dc7 | 7523 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7524 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7525 | rt_rq->rq = rq; |
7526 | #endif | |
fa85ae24 PZ |
7527 | } |
7528 | ||
6f505b16 | 7529 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7530 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7531 | struct sched_entity *se, int cpu, int add, | |
7532 | struct sched_entity *parent) | |
6f505b16 | 7533 | { |
ec7dc8ac | 7534 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7535 | tg->cfs_rq[cpu] = cfs_rq; |
7536 | init_cfs_rq(cfs_rq, rq); | |
7537 | cfs_rq->tg = tg; | |
7538 | if (add) | |
7539 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7540 | ||
7541 | tg->se[cpu] = se; | |
354d60c2 DG |
7542 | /* se could be NULL for init_task_group */ |
7543 | if (!se) | |
7544 | return; | |
7545 | ||
ec7dc8ac DG |
7546 | if (!parent) |
7547 | se->cfs_rq = &rq->cfs; | |
7548 | else | |
7549 | se->cfs_rq = parent->my_q; | |
7550 | ||
6f505b16 PZ |
7551 | se->my_q = cfs_rq; |
7552 | se->load.weight = tg->shares; | |
e05510d0 | 7553 | se->load.inv_weight = 0; |
ec7dc8ac | 7554 | se->parent = parent; |
6f505b16 | 7555 | } |
052f1dc7 | 7556 | #endif |
6f505b16 | 7557 | |
052f1dc7 | 7558 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7559 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7560 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7561 | struct sched_rt_entity *parent) | |
6f505b16 | 7562 | { |
ec7dc8ac DG |
7563 | struct rq *rq = cpu_rq(cpu); |
7564 | ||
6f505b16 PZ |
7565 | tg->rt_rq[cpu] = rt_rq; |
7566 | init_rt_rq(rt_rq, rq); | |
7567 | rt_rq->tg = tg; | |
ac086bc2 | 7568 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7569 | if (add) |
7570 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7571 | ||
7572 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7573 | if (!rt_se) |
7574 | return; | |
7575 | ||
ec7dc8ac DG |
7576 | if (!parent) |
7577 | rt_se->rt_rq = &rq->rt; | |
7578 | else | |
7579 | rt_se->rt_rq = parent->my_q; | |
7580 | ||
6f505b16 | 7581 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7582 | rt_se->parent = parent; |
6f505b16 PZ |
7583 | INIT_LIST_HEAD(&rt_se->run_list); |
7584 | } | |
7585 | #endif | |
7586 | ||
1da177e4 LT |
7587 | void __init sched_init(void) |
7588 | { | |
dd41f596 | 7589 | int i, j; |
434d53b0 MT |
7590 | unsigned long alloc_size = 0, ptr; |
7591 | ||
7592 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7593 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7594 | #endif | |
7595 | #ifdef CONFIG_RT_GROUP_SCHED | |
7596 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7597 | #endif |
df7c8e84 | 7598 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7599 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7600 | #endif |
434d53b0 | 7601 | if (alloc_size) { |
36b7b6d4 | 7602 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7603 | |
7604 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7605 | init_task_group.se = (struct sched_entity **)ptr; | |
7606 | ptr += nr_cpu_ids * sizeof(void **); | |
7607 | ||
7608 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7609 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7610 | |
6d6bc0ad | 7611 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7612 | #ifdef CONFIG_RT_GROUP_SCHED |
7613 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7614 | ptr += nr_cpu_ids * sizeof(void **); | |
7615 | ||
7616 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7617 | ptr += nr_cpu_ids * sizeof(void **); |
7618 | ||
6d6bc0ad | 7619 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7620 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7621 | for_each_possible_cpu(i) { | |
7622 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7623 | ptr += cpumask_size(); | |
7624 | } | |
7625 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7626 | } |
dd41f596 | 7627 | |
57d885fe GH |
7628 | #ifdef CONFIG_SMP |
7629 | init_defrootdomain(); | |
7630 | #endif | |
7631 | ||
d0b27fa7 PZ |
7632 | init_rt_bandwidth(&def_rt_bandwidth, |
7633 | global_rt_period(), global_rt_runtime()); | |
7634 | ||
7635 | #ifdef CONFIG_RT_GROUP_SCHED | |
7636 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7637 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7638 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7639 | |
7c941438 | 7640 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7641 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7642 | INIT_LIST_HEAD(&init_task_group.children); |
7643 | ||
7c941438 | 7644 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7645 | |
4a6cc4bd JK |
7646 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7647 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7648 | __alignof__(unsigned long)); | |
7649 | #endif | |
0a945022 | 7650 | for_each_possible_cpu(i) { |
70b97a7f | 7651 | struct rq *rq; |
1da177e4 LT |
7652 | |
7653 | rq = cpu_rq(i); | |
05fa785c | 7654 | raw_spin_lock_init(&rq->lock); |
7897986b | 7655 | rq->nr_running = 0; |
dce48a84 TG |
7656 | rq->calc_load_active = 0; |
7657 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7658 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7659 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7660 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7661 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7662 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7663 | #ifdef CONFIG_CGROUP_SCHED |
7664 | /* | |
7665 | * How much cpu bandwidth does init_task_group get? | |
7666 | * | |
7667 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7668 | * gets 100% of the cpu resources in the system. This overall | |
7669 | * system cpu resource is divided among the tasks of | |
7670 | * init_task_group and its child task-groups in a fair manner, | |
7671 | * based on each entity's (task or task-group's) weight | |
7672 | * (se->load.weight). | |
7673 | * | |
7674 | * In other words, if init_task_group has 10 tasks of weight | |
7675 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7676 | * then A0's share of the cpu resource is: | |
7677 | * | |
0d905bca | 7678 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7679 | * |
7680 | * We achieve this by letting init_task_group's tasks sit | |
7681 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7682 | */ | |
ec7dc8ac | 7683 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7684 | #endif |
354d60c2 DG |
7685 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7686 | ||
7687 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7688 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7689 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7690 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7691 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7692 | #endif |
dd41f596 | 7693 | #endif |
1da177e4 | 7694 | |
dd41f596 IM |
7695 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7696 | rq->cpu_load[j] = 0; | |
1da177e4 | 7697 | #ifdef CONFIG_SMP |
41c7ce9a | 7698 | rq->sd = NULL; |
57d885fe | 7699 | rq->rd = NULL; |
3f029d3c | 7700 | rq->post_schedule = 0; |
1da177e4 | 7701 | rq->active_balance = 0; |
dd41f596 | 7702 | rq->next_balance = jiffies; |
1da177e4 | 7703 | rq->push_cpu = 0; |
0a2966b4 | 7704 | rq->cpu = i; |
1f11eb6a | 7705 | rq->online = 0; |
1da177e4 | 7706 | rq->migration_thread = NULL; |
eae0c9df MG |
7707 | rq->idle_stamp = 0; |
7708 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
1da177e4 | 7709 | INIT_LIST_HEAD(&rq->migration_queue); |
dc938520 | 7710 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7711 | #endif |
8f4d37ec | 7712 | init_rq_hrtick(rq); |
1da177e4 | 7713 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7714 | } |
7715 | ||
2dd73a4f | 7716 | set_load_weight(&init_task); |
b50f60ce | 7717 | |
e107be36 AK |
7718 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7719 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7720 | #endif | |
7721 | ||
c9819f45 | 7722 | #ifdef CONFIG_SMP |
962cf36c | 7723 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7724 | #endif |
7725 | ||
b50f60ce | 7726 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7727 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7728 | #endif |
7729 | ||
1da177e4 LT |
7730 | /* |
7731 | * The boot idle thread does lazy MMU switching as well: | |
7732 | */ | |
7733 | atomic_inc(&init_mm.mm_count); | |
7734 | enter_lazy_tlb(&init_mm, current); | |
7735 | ||
7736 | /* | |
7737 | * Make us the idle thread. Technically, schedule() should not be | |
7738 | * called from this thread, however somewhere below it might be, | |
7739 | * but because we are the idle thread, we just pick up running again | |
7740 | * when this runqueue becomes "idle". | |
7741 | */ | |
7742 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7743 | |
7744 | calc_load_update = jiffies + LOAD_FREQ; | |
7745 | ||
dd41f596 IM |
7746 | /* |
7747 | * During early bootup we pretend to be a normal task: | |
7748 | */ | |
7749 | current->sched_class = &fair_sched_class; | |
6892b75e | 7750 | |
6a7b3dc3 | 7751 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7752 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7753 | #ifdef CONFIG_SMP |
7d1e6a9b | 7754 | #ifdef CONFIG_NO_HZ |
49557e62 | 7755 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 7756 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 7757 | #endif |
bdddd296 RR |
7758 | /* May be allocated at isolcpus cmdline parse time */ |
7759 | if (cpu_isolated_map == NULL) | |
7760 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7761 | #endif /* SMP */ |
6a7b3dc3 | 7762 | |
cdd6c482 | 7763 | perf_event_init(); |
0d905bca | 7764 | |
6892b75e | 7765 | scheduler_running = 1; |
1da177e4 LT |
7766 | } |
7767 | ||
7768 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7769 | static inline int preempt_count_equals(int preempt_offset) |
7770 | { | |
234da7bc | 7771 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7772 | |
7773 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7774 | } | |
7775 | ||
d894837f | 7776 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7777 | { |
48f24c4d | 7778 | #ifdef in_atomic |
1da177e4 LT |
7779 | static unsigned long prev_jiffy; /* ratelimiting */ |
7780 | ||
e4aafea2 FW |
7781 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7782 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7783 | return; |
7784 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7785 | return; | |
7786 | prev_jiffy = jiffies; | |
7787 | ||
3df0fc5b PZ |
7788 | printk(KERN_ERR |
7789 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7790 | file, line); | |
7791 | printk(KERN_ERR | |
7792 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7793 | in_atomic(), irqs_disabled(), | |
7794 | current->pid, current->comm); | |
aef745fc IM |
7795 | |
7796 | debug_show_held_locks(current); | |
7797 | if (irqs_disabled()) | |
7798 | print_irqtrace_events(current); | |
7799 | dump_stack(); | |
1da177e4 LT |
7800 | #endif |
7801 | } | |
7802 | EXPORT_SYMBOL(__might_sleep); | |
7803 | #endif | |
7804 | ||
7805 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7806 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7807 | { | |
7808 | int on_rq; | |
3e51f33f | 7809 | |
3a5e4dc1 AK |
7810 | on_rq = p->se.on_rq; |
7811 | if (on_rq) | |
7812 | deactivate_task(rq, p, 0); | |
7813 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7814 | if (on_rq) { | |
7815 | activate_task(rq, p, 0); | |
7816 | resched_task(rq->curr); | |
7817 | } | |
7818 | } | |
7819 | ||
1da177e4 LT |
7820 | void normalize_rt_tasks(void) |
7821 | { | |
a0f98a1c | 7822 | struct task_struct *g, *p; |
1da177e4 | 7823 | unsigned long flags; |
70b97a7f | 7824 | struct rq *rq; |
1da177e4 | 7825 | |
4cf5d77a | 7826 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7827 | do_each_thread(g, p) { |
178be793 IM |
7828 | /* |
7829 | * Only normalize user tasks: | |
7830 | */ | |
7831 | if (!p->mm) | |
7832 | continue; | |
7833 | ||
6cfb0d5d | 7834 | p->se.exec_start = 0; |
6cfb0d5d | 7835 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7836 | p->se.statistics.wait_start = 0; |
7837 | p->se.statistics.sleep_start = 0; | |
7838 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7839 | #endif |
dd41f596 IM |
7840 | |
7841 | if (!rt_task(p)) { | |
7842 | /* | |
7843 | * Renice negative nice level userspace | |
7844 | * tasks back to 0: | |
7845 | */ | |
7846 | if (TASK_NICE(p) < 0 && p->mm) | |
7847 | set_user_nice(p, 0); | |
1da177e4 | 7848 | continue; |
dd41f596 | 7849 | } |
1da177e4 | 7850 | |
1d615482 | 7851 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7852 | rq = __task_rq_lock(p); |
1da177e4 | 7853 | |
178be793 | 7854 | normalize_task(rq, p); |
3a5e4dc1 | 7855 | |
b29739f9 | 7856 | __task_rq_unlock(rq); |
1d615482 | 7857 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7858 | } while_each_thread(g, p); |
7859 | ||
4cf5d77a | 7860 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7861 | } |
7862 | ||
7863 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7864 | |
7865 | #ifdef CONFIG_IA64 | |
7866 | /* | |
7867 | * These functions are only useful for the IA64 MCA handling. | |
7868 | * | |
7869 | * They can only be called when the whole system has been | |
7870 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7871 | * activity can take place. Using them for anything else would | |
7872 | * be a serious bug, and as a result, they aren't even visible | |
7873 | * under any other configuration. | |
7874 | */ | |
7875 | ||
7876 | /** | |
7877 | * curr_task - return the current task for a given cpu. | |
7878 | * @cpu: the processor in question. | |
7879 | * | |
7880 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7881 | */ | |
36c8b586 | 7882 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7883 | { |
7884 | return cpu_curr(cpu); | |
7885 | } | |
7886 | ||
7887 | /** | |
7888 | * set_curr_task - set the current task for a given cpu. | |
7889 | * @cpu: the processor in question. | |
7890 | * @p: the task pointer to set. | |
7891 | * | |
7892 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7893 | * are serviced on a separate stack. It allows the architecture to switch the |
7894 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7895 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7896 | * and caller must save the original value of the current task (see | |
7897 | * curr_task() above) and restore that value before reenabling interrupts and | |
7898 | * re-starting the system. | |
7899 | * | |
7900 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7901 | */ | |
36c8b586 | 7902 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7903 | { |
7904 | cpu_curr(cpu) = p; | |
7905 | } | |
7906 | ||
7907 | #endif | |
29f59db3 | 7908 | |
bccbe08a PZ |
7909 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7910 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7911 | { |
7912 | int i; | |
7913 | ||
7914 | for_each_possible_cpu(i) { | |
7915 | if (tg->cfs_rq) | |
7916 | kfree(tg->cfs_rq[i]); | |
7917 | if (tg->se) | |
7918 | kfree(tg->se[i]); | |
6f505b16 PZ |
7919 | } |
7920 | ||
7921 | kfree(tg->cfs_rq); | |
7922 | kfree(tg->se); | |
6f505b16 PZ |
7923 | } |
7924 | ||
ec7dc8ac DG |
7925 | static |
7926 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 7927 | { |
29f59db3 | 7928 | struct cfs_rq *cfs_rq; |
eab17229 | 7929 | struct sched_entity *se; |
9b5b7751 | 7930 | struct rq *rq; |
29f59db3 SV |
7931 | int i; |
7932 | ||
434d53b0 | 7933 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7934 | if (!tg->cfs_rq) |
7935 | goto err; | |
434d53b0 | 7936 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7937 | if (!tg->se) |
7938 | goto err; | |
052f1dc7 PZ |
7939 | |
7940 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7941 | |
7942 | for_each_possible_cpu(i) { | |
9b5b7751 | 7943 | rq = cpu_rq(i); |
29f59db3 | 7944 | |
eab17229 LZ |
7945 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
7946 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
7947 | if (!cfs_rq) |
7948 | goto err; | |
7949 | ||
eab17229 LZ |
7950 | se = kzalloc_node(sizeof(struct sched_entity), |
7951 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 7952 | if (!se) |
dfc12eb2 | 7953 | goto err_free_rq; |
29f59db3 | 7954 | |
eab17229 | 7955 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
7956 | } |
7957 | ||
7958 | return 1; | |
7959 | ||
dfc12eb2 PC |
7960 | err_free_rq: |
7961 | kfree(cfs_rq); | |
bccbe08a PZ |
7962 | err: |
7963 | return 0; | |
7964 | } | |
7965 | ||
7966 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7967 | { | |
7968 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7969 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7970 | } | |
7971 | ||
7972 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7973 | { | |
7974 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7975 | } | |
6d6bc0ad | 7976 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
7977 | static inline void free_fair_sched_group(struct task_group *tg) |
7978 | { | |
7979 | } | |
7980 | ||
ec7dc8ac DG |
7981 | static inline |
7982 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7983 | { |
7984 | return 1; | |
7985 | } | |
7986 | ||
7987 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7988 | { | |
7989 | } | |
7990 | ||
7991 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7992 | { | |
7993 | } | |
6d6bc0ad | 7994 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
7995 | |
7996 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7997 | static void free_rt_sched_group(struct task_group *tg) |
7998 | { | |
7999 | int i; | |
8000 | ||
d0b27fa7 PZ |
8001 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8002 | ||
bccbe08a PZ |
8003 | for_each_possible_cpu(i) { |
8004 | if (tg->rt_rq) | |
8005 | kfree(tg->rt_rq[i]); | |
8006 | if (tg->rt_se) | |
8007 | kfree(tg->rt_se[i]); | |
8008 | } | |
8009 | ||
8010 | kfree(tg->rt_rq); | |
8011 | kfree(tg->rt_se); | |
8012 | } | |
8013 | ||
ec7dc8ac DG |
8014 | static |
8015 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8016 | { |
8017 | struct rt_rq *rt_rq; | |
eab17229 | 8018 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8019 | struct rq *rq; |
8020 | int i; | |
8021 | ||
434d53b0 | 8022 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8023 | if (!tg->rt_rq) |
8024 | goto err; | |
434d53b0 | 8025 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8026 | if (!tg->rt_se) |
8027 | goto err; | |
8028 | ||
d0b27fa7 PZ |
8029 | init_rt_bandwidth(&tg->rt_bandwidth, |
8030 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8031 | |
8032 | for_each_possible_cpu(i) { | |
8033 | rq = cpu_rq(i); | |
8034 | ||
eab17229 LZ |
8035 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8036 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8037 | if (!rt_rq) |
8038 | goto err; | |
29f59db3 | 8039 | |
eab17229 LZ |
8040 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8041 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8042 | if (!rt_se) |
dfc12eb2 | 8043 | goto err_free_rq; |
29f59db3 | 8044 | |
eab17229 | 8045 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8046 | } |
8047 | ||
bccbe08a PZ |
8048 | return 1; |
8049 | ||
dfc12eb2 PC |
8050 | err_free_rq: |
8051 | kfree(rt_rq); | |
bccbe08a PZ |
8052 | err: |
8053 | return 0; | |
8054 | } | |
8055 | ||
8056 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8057 | { | |
8058 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8059 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8060 | } | |
8061 | ||
8062 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8063 | { | |
8064 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8065 | } | |
6d6bc0ad | 8066 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8067 | static inline void free_rt_sched_group(struct task_group *tg) |
8068 | { | |
8069 | } | |
8070 | ||
ec7dc8ac DG |
8071 | static inline |
8072 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8073 | { |
8074 | return 1; | |
8075 | } | |
8076 | ||
8077 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8078 | { | |
8079 | } | |
8080 | ||
8081 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8082 | { | |
8083 | } | |
6d6bc0ad | 8084 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8085 | |
7c941438 | 8086 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8087 | static void free_sched_group(struct task_group *tg) |
8088 | { | |
8089 | free_fair_sched_group(tg); | |
8090 | free_rt_sched_group(tg); | |
8091 | kfree(tg); | |
8092 | } | |
8093 | ||
8094 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8095 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8096 | { |
8097 | struct task_group *tg; | |
8098 | unsigned long flags; | |
8099 | int i; | |
8100 | ||
8101 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8102 | if (!tg) | |
8103 | return ERR_PTR(-ENOMEM); | |
8104 | ||
ec7dc8ac | 8105 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8106 | goto err; |
8107 | ||
ec7dc8ac | 8108 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8109 | goto err; |
8110 | ||
8ed36996 | 8111 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8112 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8113 | register_fair_sched_group(tg, i); |
8114 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8115 | } |
6f505b16 | 8116 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8117 | |
8118 | WARN_ON(!parent); /* root should already exist */ | |
8119 | ||
8120 | tg->parent = parent; | |
f473aa5e | 8121 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8122 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8123 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8124 | |
9b5b7751 | 8125 | return tg; |
29f59db3 SV |
8126 | |
8127 | err: | |
6f505b16 | 8128 | free_sched_group(tg); |
29f59db3 SV |
8129 | return ERR_PTR(-ENOMEM); |
8130 | } | |
8131 | ||
9b5b7751 | 8132 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8133 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8134 | { |
29f59db3 | 8135 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8136 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8137 | } |
8138 | ||
9b5b7751 | 8139 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8140 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8141 | { |
8ed36996 | 8142 | unsigned long flags; |
9b5b7751 | 8143 | int i; |
29f59db3 | 8144 | |
8ed36996 | 8145 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8146 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8147 | unregister_fair_sched_group(tg, i); |
8148 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8149 | } |
6f505b16 | 8150 | list_del_rcu(&tg->list); |
f473aa5e | 8151 | list_del_rcu(&tg->siblings); |
8ed36996 | 8152 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8153 | |
9b5b7751 | 8154 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8155 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8156 | } |
8157 | ||
9b5b7751 | 8158 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8159 | * The caller of this function should have put the task in its new group |
8160 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8161 | * reflect its new group. | |
9b5b7751 SV |
8162 | */ |
8163 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8164 | { |
8165 | int on_rq, running; | |
8166 | unsigned long flags; | |
8167 | struct rq *rq; | |
8168 | ||
8169 | rq = task_rq_lock(tsk, &flags); | |
8170 | ||
051a1d1a | 8171 | running = task_current(rq, tsk); |
29f59db3 SV |
8172 | on_rq = tsk->se.on_rq; |
8173 | ||
0e1f3483 | 8174 | if (on_rq) |
29f59db3 | 8175 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8176 | if (unlikely(running)) |
8177 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8178 | |
6f505b16 | 8179 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8180 | |
810b3817 PZ |
8181 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8182 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8183 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8184 | #endif |
8185 | ||
0e1f3483 HS |
8186 | if (unlikely(running)) |
8187 | tsk->sched_class->set_curr_task(rq); | |
8188 | if (on_rq) | |
ea87bb78 | 8189 | enqueue_task(rq, tsk, 0, false); |
29f59db3 | 8190 | |
29f59db3 SV |
8191 | task_rq_unlock(rq, &flags); |
8192 | } | |
7c941438 | 8193 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8194 | |
052f1dc7 | 8195 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8196 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8197 | { |
8198 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8199 | int on_rq; |
8200 | ||
29f59db3 | 8201 | on_rq = se->on_rq; |
62fb1851 | 8202 | if (on_rq) |
29f59db3 SV |
8203 | dequeue_entity(cfs_rq, se, 0); |
8204 | ||
8205 | se->load.weight = shares; | |
e05510d0 | 8206 | se->load.inv_weight = 0; |
29f59db3 | 8207 | |
62fb1851 | 8208 | if (on_rq) |
29f59db3 | 8209 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8210 | } |
62fb1851 | 8211 | |
c09595f6 PZ |
8212 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8213 | { | |
8214 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8215 | struct rq *rq = cfs_rq->rq; | |
8216 | unsigned long flags; | |
8217 | ||
05fa785c | 8218 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8219 | __set_se_shares(se, shares); |
05fa785c | 8220 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8221 | } |
8222 | ||
8ed36996 PZ |
8223 | static DEFINE_MUTEX(shares_mutex); |
8224 | ||
4cf86d77 | 8225 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8226 | { |
8227 | int i; | |
8ed36996 | 8228 | unsigned long flags; |
c61935fd | 8229 | |
ec7dc8ac DG |
8230 | /* |
8231 | * We can't change the weight of the root cgroup. | |
8232 | */ | |
8233 | if (!tg->se[0]) | |
8234 | return -EINVAL; | |
8235 | ||
18d95a28 PZ |
8236 | if (shares < MIN_SHARES) |
8237 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8238 | else if (shares > MAX_SHARES) |
8239 | shares = MAX_SHARES; | |
62fb1851 | 8240 | |
8ed36996 | 8241 | mutex_lock(&shares_mutex); |
9b5b7751 | 8242 | if (tg->shares == shares) |
5cb350ba | 8243 | goto done; |
29f59db3 | 8244 | |
8ed36996 | 8245 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8246 | for_each_possible_cpu(i) |
8247 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8248 | list_del_rcu(&tg->siblings); |
8ed36996 | 8249 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8250 | |
8251 | /* wait for any ongoing reference to this group to finish */ | |
8252 | synchronize_sched(); | |
8253 | ||
8254 | /* | |
8255 | * Now we are free to modify the group's share on each cpu | |
8256 | * w/o tripping rebalance_share or load_balance_fair. | |
8257 | */ | |
9b5b7751 | 8258 | tg->shares = shares; |
c09595f6 PZ |
8259 | for_each_possible_cpu(i) { |
8260 | /* | |
8261 | * force a rebalance | |
8262 | */ | |
8263 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8264 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8265 | } |
29f59db3 | 8266 | |
6b2d7700 SV |
8267 | /* |
8268 | * Enable load balance activity on this group, by inserting it back on | |
8269 | * each cpu's rq->leaf_cfs_rq_list. | |
8270 | */ | |
8ed36996 | 8271 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8272 | for_each_possible_cpu(i) |
8273 | register_fair_sched_group(tg, i); | |
f473aa5e | 8274 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8275 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8276 | done: |
8ed36996 | 8277 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8278 | return 0; |
29f59db3 SV |
8279 | } |
8280 | ||
5cb350ba DG |
8281 | unsigned long sched_group_shares(struct task_group *tg) |
8282 | { | |
8283 | return tg->shares; | |
8284 | } | |
052f1dc7 | 8285 | #endif |
5cb350ba | 8286 | |
052f1dc7 | 8287 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8288 | /* |
9f0c1e56 | 8289 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8290 | */ |
9f0c1e56 PZ |
8291 | static DEFINE_MUTEX(rt_constraints_mutex); |
8292 | ||
8293 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8294 | { | |
8295 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8296 | return 1ULL << 20; |
9f0c1e56 | 8297 | |
9a7e0b18 | 8298 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8299 | } |
8300 | ||
9a7e0b18 PZ |
8301 | /* Must be called with tasklist_lock held */ |
8302 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8303 | { |
9a7e0b18 | 8304 | struct task_struct *g, *p; |
b40b2e8e | 8305 | |
9a7e0b18 PZ |
8306 | do_each_thread(g, p) { |
8307 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8308 | return 1; | |
8309 | } while_each_thread(g, p); | |
b40b2e8e | 8310 | |
9a7e0b18 PZ |
8311 | return 0; |
8312 | } | |
b40b2e8e | 8313 | |
9a7e0b18 PZ |
8314 | struct rt_schedulable_data { |
8315 | struct task_group *tg; | |
8316 | u64 rt_period; | |
8317 | u64 rt_runtime; | |
8318 | }; | |
b40b2e8e | 8319 | |
9a7e0b18 PZ |
8320 | static int tg_schedulable(struct task_group *tg, void *data) |
8321 | { | |
8322 | struct rt_schedulable_data *d = data; | |
8323 | struct task_group *child; | |
8324 | unsigned long total, sum = 0; | |
8325 | u64 period, runtime; | |
b40b2e8e | 8326 | |
9a7e0b18 PZ |
8327 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8328 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8329 | |
9a7e0b18 PZ |
8330 | if (tg == d->tg) { |
8331 | period = d->rt_period; | |
8332 | runtime = d->rt_runtime; | |
b40b2e8e | 8333 | } |
b40b2e8e | 8334 | |
4653f803 PZ |
8335 | /* |
8336 | * Cannot have more runtime than the period. | |
8337 | */ | |
8338 | if (runtime > period && runtime != RUNTIME_INF) | |
8339 | return -EINVAL; | |
6f505b16 | 8340 | |
4653f803 PZ |
8341 | /* |
8342 | * Ensure we don't starve existing RT tasks. | |
8343 | */ | |
9a7e0b18 PZ |
8344 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8345 | return -EBUSY; | |
6f505b16 | 8346 | |
9a7e0b18 | 8347 | total = to_ratio(period, runtime); |
6f505b16 | 8348 | |
4653f803 PZ |
8349 | /* |
8350 | * Nobody can have more than the global setting allows. | |
8351 | */ | |
8352 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8353 | return -EINVAL; | |
6f505b16 | 8354 | |
4653f803 PZ |
8355 | /* |
8356 | * The sum of our children's runtime should not exceed our own. | |
8357 | */ | |
9a7e0b18 PZ |
8358 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8359 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8360 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8361 | |
9a7e0b18 PZ |
8362 | if (child == d->tg) { |
8363 | period = d->rt_period; | |
8364 | runtime = d->rt_runtime; | |
8365 | } | |
6f505b16 | 8366 | |
9a7e0b18 | 8367 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8368 | } |
6f505b16 | 8369 | |
9a7e0b18 PZ |
8370 | if (sum > total) |
8371 | return -EINVAL; | |
8372 | ||
8373 | return 0; | |
6f505b16 PZ |
8374 | } |
8375 | ||
9a7e0b18 | 8376 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8377 | { |
9a7e0b18 PZ |
8378 | struct rt_schedulable_data data = { |
8379 | .tg = tg, | |
8380 | .rt_period = period, | |
8381 | .rt_runtime = runtime, | |
8382 | }; | |
8383 | ||
8384 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8385 | } |
8386 | ||
d0b27fa7 PZ |
8387 | static int tg_set_bandwidth(struct task_group *tg, |
8388 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8389 | { |
ac086bc2 | 8390 | int i, err = 0; |
9f0c1e56 | 8391 | |
9f0c1e56 | 8392 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8393 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8394 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8395 | if (err) | |
9f0c1e56 | 8396 | goto unlock; |
ac086bc2 | 8397 | |
0986b11b | 8398 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8399 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8400 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8401 | |
8402 | for_each_possible_cpu(i) { | |
8403 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8404 | ||
0986b11b | 8405 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8406 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8407 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8408 | } |
0986b11b | 8409 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8410 | unlock: |
521f1a24 | 8411 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8412 | mutex_unlock(&rt_constraints_mutex); |
8413 | ||
8414 | return err; | |
6f505b16 PZ |
8415 | } |
8416 | ||
d0b27fa7 PZ |
8417 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8418 | { | |
8419 | u64 rt_runtime, rt_period; | |
8420 | ||
8421 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8422 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8423 | if (rt_runtime_us < 0) | |
8424 | rt_runtime = RUNTIME_INF; | |
8425 | ||
8426 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8427 | } | |
8428 | ||
9f0c1e56 PZ |
8429 | long sched_group_rt_runtime(struct task_group *tg) |
8430 | { | |
8431 | u64 rt_runtime_us; | |
8432 | ||
d0b27fa7 | 8433 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8434 | return -1; |
8435 | ||
d0b27fa7 | 8436 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8437 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8438 | return rt_runtime_us; | |
8439 | } | |
d0b27fa7 PZ |
8440 | |
8441 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8442 | { | |
8443 | u64 rt_runtime, rt_period; | |
8444 | ||
8445 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8446 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8447 | ||
619b0488 R |
8448 | if (rt_period == 0) |
8449 | return -EINVAL; | |
8450 | ||
d0b27fa7 PZ |
8451 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8452 | } | |
8453 | ||
8454 | long sched_group_rt_period(struct task_group *tg) | |
8455 | { | |
8456 | u64 rt_period_us; | |
8457 | ||
8458 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8459 | do_div(rt_period_us, NSEC_PER_USEC); | |
8460 | return rt_period_us; | |
8461 | } | |
8462 | ||
8463 | static int sched_rt_global_constraints(void) | |
8464 | { | |
4653f803 | 8465 | u64 runtime, period; |
d0b27fa7 PZ |
8466 | int ret = 0; |
8467 | ||
ec5d4989 HS |
8468 | if (sysctl_sched_rt_period <= 0) |
8469 | return -EINVAL; | |
8470 | ||
4653f803 PZ |
8471 | runtime = global_rt_runtime(); |
8472 | period = global_rt_period(); | |
8473 | ||
8474 | /* | |
8475 | * Sanity check on the sysctl variables. | |
8476 | */ | |
8477 | if (runtime > period && runtime != RUNTIME_INF) | |
8478 | return -EINVAL; | |
10b612f4 | 8479 | |
d0b27fa7 | 8480 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8481 | read_lock(&tasklist_lock); |
4653f803 | 8482 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8483 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8484 | mutex_unlock(&rt_constraints_mutex); |
8485 | ||
8486 | return ret; | |
8487 | } | |
54e99124 DG |
8488 | |
8489 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8490 | { | |
8491 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8492 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8493 | return 0; | |
8494 | ||
8495 | return 1; | |
8496 | } | |
8497 | ||
6d6bc0ad | 8498 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8499 | static int sched_rt_global_constraints(void) |
8500 | { | |
ac086bc2 PZ |
8501 | unsigned long flags; |
8502 | int i; | |
8503 | ||
ec5d4989 HS |
8504 | if (sysctl_sched_rt_period <= 0) |
8505 | return -EINVAL; | |
8506 | ||
60aa605d PZ |
8507 | /* |
8508 | * There's always some RT tasks in the root group | |
8509 | * -- migration, kstopmachine etc.. | |
8510 | */ | |
8511 | if (sysctl_sched_rt_runtime == 0) | |
8512 | return -EBUSY; | |
8513 | ||
0986b11b | 8514 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8515 | for_each_possible_cpu(i) { |
8516 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8517 | ||
0986b11b | 8518 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8519 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8520 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8521 | } |
0986b11b | 8522 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8523 | |
d0b27fa7 PZ |
8524 | return 0; |
8525 | } | |
6d6bc0ad | 8526 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8527 | |
8528 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8529 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8530 | loff_t *ppos) |
8531 | { | |
8532 | int ret; | |
8533 | int old_period, old_runtime; | |
8534 | static DEFINE_MUTEX(mutex); | |
8535 | ||
8536 | mutex_lock(&mutex); | |
8537 | old_period = sysctl_sched_rt_period; | |
8538 | old_runtime = sysctl_sched_rt_runtime; | |
8539 | ||
8d65af78 | 8540 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8541 | |
8542 | if (!ret && write) { | |
8543 | ret = sched_rt_global_constraints(); | |
8544 | if (ret) { | |
8545 | sysctl_sched_rt_period = old_period; | |
8546 | sysctl_sched_rt_runtime = old_runtime; | |
8547 | } else { | |
8548 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8549 | def_rt_bandwidth.rt_period = | |
8550 | ns_to_ktime(global_rt_period()); | |
8551 | } | |
8552 | } | |
8553 | mutex_unlock(&mutex); | |
8554 | ||
8555 | return ret; | |
8556 | } | |
68318b8e | 8557 | |
052f1dc7 | 8558 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8559 | |
8560 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8561 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8562 | { |
2b01dfe3 PM |
8563 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8564 | struct task_group, css); | |
68318b8e SV |
8565 | } |
8566 | ||
8567 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8568 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8569 | { |
ec7dc8ac | 8570 | struct task_group *tg, *parent; |
68318b8e | 8571 | |
2b01dfe3 | 8572 | if (!cgrp->parent) { |
68318b8e | 8573 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8574 | return &init_task_group.css; |
8575 | } | |
8576 | ||
ec7dc8ac DG |
8577 | parent = cgroup_tg(cgrp->parent); |
8578 | tg = sched_create_group(parent); | |
68318b8e SV |
8579 | if (IS_ERR(tg)) |
8580 | return ERR_PTR(-ENOMEM); | |
8581 | ||
68318b8e SV |
8582 | return &tg->css; |
8583 | } | |
8584 | ||
41a2d6cf IM |
8585 | static void |
8586 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8587 | { |
2b01dfe3 | 8588 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8589 | |
8590 | sched_destroy_group(tg); | |
8591 | } | |
8592 | ||
41a2d6cf | 8593 | static int |
be367d09 | 8594 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8595 | { |
b68aa230 | 8596 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8597 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8598 | return -EINVAL; |
8599 | #else | |
68318b8e SV |
8600 | /* We don't support RT-tasks being in separate groups */ |
8601 | if (tsk->sched_class != &fair_sched_class) | |
8602 | return -EINVAL; | |
b68aa230 | 8603 | #endif |
be367d09 BB |
8604 | return 0; |
8605 | } | |
68318b8e | 8606 | |
be367d09 BB |
8607 | static int |
8608 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8609 | struct task_struct *tsk, bool threadgroup) | |
8610 | { | |
8611 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8612 | if (retval) | |
8613 | return retval; | |
8614 | if (threadgroup) { | |
8615 | struct task_struct *c; | |
8616 | rcu_read_lock(); | |
8617 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8618 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8619 | if (retval) { | |
8620 | rcu_read_unlock(); | |
8621 | return retval; | |
8622 | } | |
8623 | } | |
8624 | rcu_read_unlock(); | |
8625 | } | |
68318b8e SV |
8626 | return 0; |
8627 | } | |
8628 | ||
8629 | static void | |
2b01dfe3 | 8630 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8631 | struct cgroup *old_cont, struct task_struct *tsk, |
8632 | bool threadgroup) | |
68318b8e SV |
8633 | { |
8634 | sched_move_task(tsk); | |
be367d09 BB |
8635 | if (threadgroup) { |
8636 | struct task_struct *c; | |
8637 | rcu_read_lock(); | |
8638 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8639 | sched_move_task(c); | |
8640 | } | |
8641 | rcu_read_unlock(); | |
8642 | } | |
68318b8e SV |
8643 | } |
8644 | ||
052f1dc7 | 8645 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8646 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8647 | u64 shareval) |
68318b8e | 8648 | { |
2b01dfe3 | 8649 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8650 | } |
8651 | ||
f4c753b7 | 8652 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8653 | { |
2b01dfe3 | 8654 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8655 | |
8656 | return (u64) tg->shares; | |
8657 | } | |
6d6bc0ad | 8658 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8659 | |
052f1dc7 | 8660 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8661 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8662 | s64 val) |
6f505b16 | 8663 | { |
06ecb27c | 8664 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8665 | } |
8666 | ||
06ecb27c | 8667 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8668 | { |
06ecb27c | 8669 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8670 | } |
d0b27fa7 PZ |
8671 | |
8672 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8673 | u64 rt_period_us) | |
8674 | { | |
8675 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8676 | } | |
8677 | ||
8678 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8679 | { | |
8680 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8681 | } | |
6d6bc0ad | 8682 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8683 | |
fe5c7cc2 | 8684 | static struct cftype cpu_files[] = { |
052f1dc7 | 8685 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8686 | { |
8687 | .name = "shares", | |
f4c753b7 PM |
8688 | .read_u64 = cpu_shares_read_u64, |
8689 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8690 | }, |
052f1dc7 PZ |
8691 | #endif |
8692 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8693 | { |
9f0c1e56 | 8694 | .name = "rt_runtime_us", |
06ecb27c PM |
8695 | .read_s64 = cpu_rt_runtime_read, |
8696 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8697 | }, |
d0b27fa7 PZ |
8698 | { |
8699 | .name = "rt_period_us", | |
f4c753b7 PM |
8700 | .read_u64 = cpu_rt_period_read_uint, |
8701 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8702 | }, |
052f1dc7 | 8703 | #endif |
68318b8e SV |
8704 | }; |
8705 | ||
8706 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8707 | { | |
fe5c7cc2 | 8708 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8709 | } |
8710 | ||
8711 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8712 | .name = "cpu", |
8713 | .create = cpu_cgroup_create, | |
8714 | .destroy = cpu_cgroup_destroy, | |
8715 | .can_attach = cpu_cgroup_can_attach, | |
8716 | .attach = cpu_cgroup_attach, | |
8717 | .populate = cpu_cgroup_populate, | |
8718 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8719 | .early_init = 1, |
8720 | }; | |
8721 | ||
052f1dc7 | 8722 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8723 | |
8724 | #ifdef CONFIG_CGROUP_CPUACCT | |
8725 | ||
8726 | /* | |
8727 | * CPU accounting code for task groups. | |
8728 | * | |
8729 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8730 | * (balbir@in.ibm.com). | |
8731 | */ | |
8732 | ||
934352f2 | 8733 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8734 | struct cpuacct { |
8735 | struct cgroup_subsys_state css; | |
8736 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8737 | u64 __percpu *cpuusage; |
ef12fefa | 8738 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8739 | struct cpuacct *parent; |
d842de87 SV |
8740 | }; |
8741 | ||
8742 | struct cgroup_subsys cpuacct_subsys; | |
8743 | ||
8744 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8745 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8746 | { |
32cd756a | 8747 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8748 | struct cpuacct, css); |
8749 | } | |
8750 | ||
8751 | /* return cpu accounting group to which this task belongs */ | |
8752 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8753 | { | |
8754 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8755 | struct cpuacct, css); | |
8756 | } | |
8757 | ||
8758 | /* create a new cpu accounting group */ | |
8759 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8760 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8761 | { |
8762 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8763 | int i; |
d842de87 SV |
8764 | |
8765 | if (!ca) | |
ef12fefa | 8766 | goto out; |
d842de87 SV |
8767 | |
8768 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8769 | if (!ca->cpuusage) |
8770 | goto out_free_ca; | |
8771 | ||
8772 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8773 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8774 | goto out_free_counters; | |
d842de87 | 8775 | |
934352f2 BR |
8776 | if (cgrp->parent) |
8777 | ca->parent = cgroup_ca(cgrp->parent); | |
8778 | ||
d842de87 | 8779 | return &ca->css; |
ef12fefa BR |
8780 | |
8781 | out_free_counters: | |
8782 | while (--i >= 0) | |
8783 | percpu_counter_destroy(&ca->cpustat[i]); | |
8784 | free_percpu(ca->cpuusage); | |
8785 | out_free_ca: | |
8786 | kfree(ca); | |
8787 | out: | |
8788 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8789 | } |
8790 | ||
8791 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8792 | static void |
32cd756a | 8793 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8794 | { |
32cd756a | 8795 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8796 | int i; |
d842de87 | 8797 | |
ef12fefa BR |
8798 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8799 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8800 | free_percpu(ca->cpuusage); |
8801 | kfree(ca); | |
8802 | } | |
8803 | ||
720f5498 KC |
8804 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8805 | { | |
b36128c8 | 8806 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8807 | u64 data; |
8808 | ||
8809 | #ifndef CONFIG_64BIT | |
8810 | /* | |
8811 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8812 | */ | |
05fa785c | 8813 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8814 | data = *cpuusage; |
05fa785c | 8815 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8816 | #else |
8817 | data = *cpuusage; | |
8818 | #endif | |
8819 | ||
8820 | return data; | |
8821 | } | |
8822 | ||
8823 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8824 | { | |
b36128c8 | 8825 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8826 | |
8827 | #ifndef CONFIG_64BIT | |
8828 | /* | |
8829 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8830 | */ | |
05fa785c | 8831 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8832 | *cpuusage = val; |
05fa785c | 8833 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8834 | #else |
8835 | *cpuusage = val; | |
8836 | #endif | |
8837 | } | |
8838 | ||
d842de87 | 8839 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8840 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8841 | { |
32cd756a | 8842 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8843 | u64 totalcpuusage = 0; |
8844 | int i; | |
8845 | ||
720f5498 KC |
8846 | for_each_present_cpu(i) |
8847 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8848 | |
8849 | return totalcpuusage; | |
8850 | } | |
8851 | ||
0297b803 DG |
8852 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8853 | u64 reset) | |
8854 | { | |
8855 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8856 | int err = 0; | |
8857 | int i; | |
8858 | ||
8859 | if (reset) { | |
8860 | err = -EINVAL; | |
8861 | goto out; | |
8862 | } | |
8863 | ||
720f5498 KC |
8864 | for_each_present_cpu(i) |
8865 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8866 | |
0297b803 DG |
8867 | out: |
8868 | return err; | |
8869 | } | |
8870 | ||
e9515c3c KC |
8871 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8872 | struct seq_file *m) | |
8873 | { | |
8874 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8875 | u64 percpu; | |
8876 | int i; | |
8877 | ||
8878 | for_each_present_cpu(i) { | |
8879 | percpu = cpuacct_cpuusage_read(ca, i); | |
8880 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8881 | } | |
8882 | seq_printf(m, "\n"); | |
8883 | return 0; | |
8884 | } | |
8885 | ||
ef12fefa BR |
8886 | static const char *cpuacct_stat_desc[] = { |
8887 | [CPUACCT_STAT_USER] = "user", | |
8888 | [CPUACCT_STAT_SYSTEM] = "system", | |
8889 | }; | |
8890 | ||
8891 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8892 | struct cgroup_map_cb *cb) | |
8893 | { | |
8894 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8895 | int i; | |
8896 | ||
8897 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8898 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8899 | val = cputime64_to_clock_t(val); | |
8900 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8901 | } | |
8902 | return 0; | |
8903 | } | |
8904 | ||
d842de87 SV |
8905 | static struct cftype files[] = { |
8906 | { | |
8907 | .name = "usage", | |
f4c753b7 PM |
8908 | .read_u64 = cpuusage_read, |
8909 | .write_u64 = cpuusage_write, | |
d842de87 | 8910 | }, |
e9515c3c KC |
8911 | { |
8912 | .name = "usage_percpu", | |
8913 | .read_seq_string = cpuacct_percpu_seq_read, | |
8914 | }, | |
ef12fefa BR |
8915 | { |
8916 | .name = "stat", | |
8917 | .read_map = cpuacct_stats_show, | |
8918 | }, | |
d842de87 SV |
8919 | }; |
8920 | ||
32cd756a | 8921 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8922 | { |
32cd756a | 8923 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8924 | } |
8925 | ||
8926 | /* | |
8927 | * charge this task's execution time to its accounting group. | |
8928 | * | |
8929 | * called with rq->lock held. | |
8930 | */ | |
8931 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8932 | { | |
8933 | struct cpuacct *ca; | |
934352f2 | 8934 | int cpu; |
d842de87 | 8935 | |
c40c6f85 | 8936 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8937 | return; |
8938 | ||
934352f2 | 8939 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8940 | |
8941 | rcu_read_lock(); | |
8942 | ||
d842de87 | 8943 | ca = task_ca(tsk); |
d842de87 | 8944 | |
934352f2 | 8945 | for (; ca; ca = ca->parent) { |
b36128c8 | 8946 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8947 | *cpuusage += cputime; |
8948 | } | |
a18b83b7 BR |
8949 | |
8950 | rcu_read_unlock(); | |
d842de87 SV |
8951 | } |
8952 | ||
fa535a77 AB |
8953 | /* |
8954 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
8955 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
8956 | * percpu_counter_add with values large enough to always overflow the | |
8957 | * per cpu batch limit causing bad SMP scalability. | |
8958 | * | |
8959 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
8960 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
8961 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
8962 | */ | |
8963 | #ifdef CONFIG_SMP | |
8964 | #define CPUACCT_BATCH \ | |
8965 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
8966 | #else | |
8967 | #define CPUACCT_BATCH 0 | |
8968 | #endif | |
8969 | ||
ef12fefa BR |
8970 | /* |
8971 | * Charge the system/user time to the task's accounting group. | |
8972 | */ | |
8973 | static void cpuacct_update_stats(struct task_struct *tsk, | |
8974 | enum cpuacct_stat_index idx, cputime_t val) | |
8975 | { | |
8976 | struct cpuacct *ca; | |
fa535a77 | 8977 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
8978 | |
8979 | if (unlikely(!cpuacct_subsys.active)) | |
8980 | return; | |
8981 | ||
8982 | rcu_read_lock(); | |
8983 | ca = task_ca(tsk); | |
8984 | ||
8985 | do { | |
fa535a77 | 8986 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
8987 | ca = ca->parent; |
8988 | } while (ca); | |
8989 | rcu_read_unlock(); | |
8990 | } | |
8991 | ||
d842de87 SV |
8992 | struct cgroup_subsys cpuacct_subsys = { |
8993 | .name = "cpuacct", | |
8994 | .create = cpuacct_create, | |
8995 | .destroy = cpuacct_destroy, | |
8996 | .populate = cpuacct_populate, | |
8997 | .subsys_id = cpuacct_subsys_id, | |
8998 | }; | |
8999 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
9000 | |
9001 | #ifndef CONFIG_SMP | |
9002 | ||
9003 | int rcu_expedited_torture_stats(char *page) | |
9004 | { | |
9005 | return 0; | |
9006 | } | |
9007 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9008 | ||
9009 | void synchronize_sched_expedited(void) | |
9010 | { | |
9011 | } | |
9012 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9013 | ||
9014 | #else /* #ifndef CONFIG_SMP */ | |
9015 | ||
9016 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
9017 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
9018 | ||
9019 | #define RCU_EXPEDITED_STATE_POST -2 | |
9020 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
9021 | ||
9022 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
9023 | ||
9024 | int rcu_expedited_torture_stats(char *page) | |
9025 | { | |
9026 | int cnt = 0; | |
9027 | int cpu; | |
9028 | ||
9029 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
9030 | for_each_online_cpu(cpu) { | |
9031 | cnt += sprintf(&page[cnt], " %d:%d", | |
9032 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
9033 | } | |
9034 | cnt += sprintf(&page[cnt], "\n"); | |
9035 | return cnt; | |
9036 | } | |
9037 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9038 | ||
9039 | static long synchronize_sched_expedited_count; | |
9040 | ||
9041 | /* | |
9042 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9043 | * approach to force grace period to end quickly. This consumes | |
9044 | * significant time on all CPUs, and is thus not recommended for | |
9045 | * any sort of common-case code. | |
9046 | * | |
9047 | * Note that it is illegal to call this function while holding any | |
9048 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9049 | * observe this restriction will result in deadlock. | |
9050 | */ | |
9051 | void synchronize_sched_expedited(void) | |
9052 | { | |
9053 | int cpu; | |
9054 | unsigned long flags; | |
9055 | bool need_full_sync = 0; | |
9056 | struct rq *rq; | |
9057 | struct migration_req *req; | |
9058 | long snap; | |
9059 | int trycount = 0; | |
9060 | ||
9061 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
9062 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
9063 | get_online_cpus(); | |
9064 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
9065 | put_online_cpus(); | |
9066 | if (trycount++ < 10) | |
9067 | udelay(trycount * num_online_cpus()); | |
9068 | else { | |
9069 | synchronize_sched(); | |
9070 | return; | |
9071 | } | |
9072 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
9073 | smp_mb(); /* ensure test happens before caller kfree */ | |
9074 | return; | |
9075 | } | |
9076 | get_online_cpus(); | |
9077 | } | |
9078 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
9079 | for_each_online_cpu(cpu) { | |
9080 | rq = cpu_rq(cpu); | |
9081 | req = &per_cpu(rcu_migration_req, cpu); | |
9082 | init_completion(&req->done); | |
9083 | req->task = NULL; | |
9084 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
05fa785c | 9085 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf | 9086 | list_add(&req->list, &rq->migration_queue); |
05fa785c | 9087 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9088 | wake_up_process(rq->migration_thread); |
9089 | } | |
9090 | for_each_online_cpu(cpu) { | |
9091 | rcu_expedited_state = cpu; | |
9092 | req = &per_cpu(rcu_migration_req, cpu); | |
9093 | rq = cpu_rq(cpu); | |
9094 | wait_for_completion(&req->done); | |
05fa785c | 9095 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf PM |
9096 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) |
9097 | need_full_sync = 1; | |
9098 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
05fa785c | 9099 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9100 | } |
9101 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
956539b7 | 9102 | synchronize_sched_expedited_count++; |
03b042bf PM |
9103 | mutex_unlock(&rcu_sched_expedited_mutex); |
9104 | put_online_cpus(); | |
9105 | if (need_full_sync) | |
9106 | synchronize_sched(); | |
9107 | } | |
9108 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9109 | ||
9110 | #endif /* #else #ifndef CONFIG_SMP */ |