]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
58 | #include <linux/kthread.h> | |
b5aadf7f | 59 | #include <linux/proc_fs.h> |
1da177e4 | 60 | #include <linux/seq_file.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
5a0e3ad6 | 74 | #include <linux/slab.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 GH |
79 | #include "sched_cpupri.h" |
80 | ||
a8d154b0 | 81 | #define CREATE_TRACE_POINTS |
ad8d75ff | 82 | #include <trace/events/sched.h> |
a8d154b0 | 83 | |
1da177e4 LT |
84 | /* |
85 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
86 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
87 | * and back. | |
88 | */ | |
89 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
90 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
91 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
92 | ||
93 | /* | |
94 | * 'User priority' is the nice value converted to something we | |
95 | * can work with better when scaling various scheduler parameters, | |
96 | * it's a [ 0 ... 39 ] range. | |
97 | */ | |
98 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
99 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
100 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
101 | ||
102 | /* | |
d7876a08 | 103 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 104 | */ |
d6322faf | 105 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 106 | |
6aa645ea IM |
107 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
108 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
109 | ||
1da177e4 LT |
110 | /* |
111 | * These are the 'tuning knobs' of the scheduler: | |
112 | * | |
a4ec24b4 | 113 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
114 | * Timeslices get refilled after they expire. |
115 | */ | |
1da177e4 | 116 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 117 | |
d0b27fa7 PZ |
118 | /* |
119 | * single value that denotes runtime == period, ie unlimited time. | |
120 | */ | |
121 | #define RUNTIME_INF ((u64)~0ULL) | |
122 | ||
e05606d3 IM |
123 | static inline int rt_policy(int policy) |
124 | { | |
3f33a7ce | 125 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
126 | return 1; |
127 | return 0; | |
128 | } | |
129 | ||
130 | static inline int task_has_rt_policy(struct task_struct *p) | |
131 | { | |
132 | return rt_policy(p->policy); | |
133 | } | |
134 | ||
1da177e4 | 135 | /* |
6aa645ea | 136 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 137 | */ |
6aa645ea IM |
138 | struct rt_prio_array { |
139 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
140 | struct list_head queue[MAX_RT_PRIO]; | |
141 | }; | |
142 | ||
d0b27fa7 | 143 | struct rt_bandwidth { |
ea736ed5 | 144 | /* nests inside the rq lock: */ |
0986b11b | 145 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
146 | ktime_t rt_period; |
147 | u64 rt_runtime; | |
148 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
149 | }; |
150 | ||
151 | static struct rt_bandwidth def_rt_bandwidth; | |
152 | ||
153 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
154 | ||
155 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
156 | { | |
157 | struct rt_bandwidth *rt_b = | |
158 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
159 | ktime_t now; | |
160 | int overrun; | |
161 | int idle = 0; | |
162 | ||
163 | for (;;) { | |
164 | now = hrtimer_cb_get_time(timer); | |
165 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
166 | ||
167 | if (!overrun) | |
168 | break; | |
169 | ||
170 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
171 | } | |
172 | ||
173 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
174 | } | |
175 | ||
176 | static | |
177 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
178 | { | |
179 | rt_b->rt_period = ns_to_ktime(period); | |
180 | rt_b->rt_runtime = runtime; | |
181 | ||
0986b11b | 182 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 183 | |
d0b27fa7 PZ |
184 | hrtimer_init(&rt_b->rt_period_timer, |
185 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
186 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
187 | } |
188 | ||
c8bfff6d KH |
189 | static inline int rt_bandwidth_enabled(void) |
190 | { | |
191 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
192 | } |
193 | ||
194 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
195 | { | |
196 | ktime_t now; | |
197 | ||
cac64d00 | 198 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
199 | return; |
200 | ||
201 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
202 | return; | |
203 | ||
0986b11b | 204 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 205 | for (;;) { |
7f1e2ca9 PZ |
206 | unsigned long delta; |
207 | ktime_t soft, hard; | |
208 | ||
d0b27fa7 PZ |
209 | if (hrtimer_active(&rt_b->rt_period_timer)) |
210 | break; | |
211 | ||
212 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
213 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
214 | |
215 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
216 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
217 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
218 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 219 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 220 | } |
0986b11b | 221 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
222 | } |
223 | ||
224 | #ifdef CONFIG_RT_GROUP_SCHED | |
225 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
226 | { | |
227 | hrtimer_cancel(&rt_b->rt_period_timer); | |
228 | } | |
229 | #endif | |
230 | ||
712555ee HC |
231 | /* |
232 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
233 | * detach_destroy_domains and partition_sched_domains. | |
234 | */ | |
235 | static DEFINE_MUTEX(sched_domains_mutex); | |
236 | ||
7c941438 | 237 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 238 | |
68318b8e SV |
239 | #include <linux/cgroup.h> |
240 | ||
29f59db3 SV |
241 | struct cfs_rq; |
242 | ||
6f505b16 PZ |
243 | static LIST_HEAD(task_groups); |
244 | ||
29f59db3 | 245 | /* task group related information */ |
4cf86d77 | 246 | struct task_group { |
68318b8e | 247 | struct cgroup_subsys_state css; |
6c415b92 | 248 | |
052f1dc7 | 249 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
250 | /* schedulable entities of this group on each cpu */ |
251 | struct sched_entity **se; | |
252 | /* runqueue "owned" by this group on each cpu */ | |
253 | struct cfs_rq **cfs_rq; | |
254 | unsigned long shares; | |
052f1dc7 PZ |
255 | #endif |
256 | ||
257 | #ifdef CONFIG_RT_GROUP_SCHED | |
258 | struct sched_rt_entity **rt_se; | |
259 | struct rt_rq **rt_rq; | |
260 | ||
d0b27fa7 | 261 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 262 | #endif |
6b2d7700 | 263 | |
ae8393e5 | 264 | struct rcu_head rcu; |
6f505b16 | 265 | struct list_head list; |
f473aa5e PZ |
266 | |
267 | struct task_group *parent; | |
268 | struct list_head siblings; | |
269 | struct list_head children; | |
29f59db3 SV |
270 | }; |
271 | ||
eff766a6 | 272 | #define root_task_group init_task_group |
6f505b16 | 273 | |
8ed36996 | 274 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
275 | * a task group's cpu shares. |
276 | */ | |
8ed36996 | 277 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 278 | |
e9036b36 CG |
279 | #ifdef CONFIG_FAIR_GROUP_SCHED |
280 | ||
57310a98 PZ |
281 | #ifdef CONFIG_SMP |
282 | static int root_task_group_empty(void) | |
283 | { | |
284 | return list_empty(&root_task_group.children); | |
285 | } | |
286 | #endif | |
287 | ||
052f1dc7 | 288 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 289 | |
cb4ad1ff | 290 | /* |
2e084786 LJ |
291 | * A weight of 0 or 1 can cause arithmetics problems. |
292 | * A weight of a cfs_rq is the sum of weights of which entities | |
293 | * are queued on this cfs_rq, so a weight of a entity should not be | |
294 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
295 | * (The default weight is 1024 - so there's no practical |
296 | * limitation from this.) | |
297 | */ | |
18d95a28 | 298 | #define MIN_SHARES 2 |
2e084786 | 299 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 300 | |
052f1dc7 PZ |
301 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
302 | #endif | |
303 | ||
29f59db3 | 304 | /* Default task group. |
3a252015 | 305 | * Every task in system belong to this group at bootup. |
29f59db3 | 306 | */ |
434d53b0 | 307 | struct task_group init_task_group; |
29f59db3 SV |
308 | |
309 | /* return group to which a task belongs */ | |
4cf86d77 | 310 | static inline struct task_group *task_group(struct task_struct *p) |
29f59db3 | 311 | { |
4cf86d77 | 312 | struct task_group *tg; |
9b5b7751 | 313 | |
7c941438 | 314 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
315 | tg = container_of(task_subsys_state(p, cpu_cgroup_subsys_id), |
316 | struct task_group, css); | |
24e377a8 | 317 | #else |
41a2d6cf | 318 | tg = &init_task_group; |
24e377a8 | 319 | #endif |
9b5b7751 | 320 | return tg; |
29f59db3 SV |
321 | } |
322 | ||
323 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
6f505b16 | 324 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) |
29f59db3 | 325 | { |
8b08ca52 PZ |
326 | /* |
327 | * Strictly speaking this rcu_read_lock() is not needed since the | |
328 | * task_group is tied to the cgroup, which in turn can never go away | |
329 | * as long as there are tasks attached to it. | |
330 | * | |
331 | * However since task_group() uses task_subsys_state() which is an | |
332 | * rcu_dereference() user, this quiets CONFIG_PROVE_RCU. | |
333 | */ | |
334 | rcu_read_lock(); | |
052f1dc7 | 335 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ce96b5ac DA |
336 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; |
337 | p->se.parent = task_group(p)->se[cpu]; | |
052f1dc7 | 338 | #endif |
6f505b16 | 339 | |
052f1dc7 | 340 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 PZ |
341 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; |
342 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
052f1dc7 | 343 | #endif |
8b08ca52 | 344 | rcu_read_unlock(); |
29f59db3 SV |
345 | } |
346 | ||
347 | #else | |
348 | ||
6f505b16 | 349 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } |
83378269 PZ |
350 | static inline struct task_group *task_group(struct task_struct *p) |
351 | { | |
352 | return NULL; | |
353 | } | |
29f59db3 | 354 | |
7c941438 | 355 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 356 | |
6aa645ea IM |
357 | /* CFS-related fields in a runqueue */ |
358 | struct cfs_rq { | |
359 | struct load_weight load; | |
360 | unsigned long nr_running; | |
361 | ||
6aa645ea | 362 | u64 exec_clock; |
e9acbff6 | 363 | u64 min_vruntime; |
6aa645ea IM |
364 | |
365 | struct rb_root tasks_timeline; | |
366 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
367 | |
368 | struct list_head tasks; | |
369 | struct list_head *balance_iterator; | |
370 | ||
371 | /* | |
372 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
373 | * It is set to NULL otherwise (i.e when none are currently running). |
374 | */ | |
4793241b | 375 | struct sched_entity *curr, *next, *last; |
ddc97297 | 376 | |
5ac5c4d6 | 377 | unsigned int nr_spread_over; |
ddc97297 | 378 | |
62160e3f | 379 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
380 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
381 | ||
41a2d6cf IM |
382 | /* |
383 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
384 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
385 | * (like users, containers etc.) | |
386 | * | |
387 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
388 | * list is used during load balance. | |
389 | */ | |
41a2d6cf IM |
390 | struct list_head leaf_cfs_rq_list; |
391 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
392 | |
393 | #ifdef CONFIG_SMP | |
c09595f6 | 394 | /* |
c8cba857 | 395 | * the part of load.weight contributed by tasks |
c09595f6 | 396 | */ |
c8cba857 | 397 | unsigned long task_weight; |
c09595f6 | 398 | |
c8cba857 PZ |
399 | /* |
400 | * h_load = weight * f(tg) | |
401 | * | |
402 | * Where f(tg) is the recursive weight fraction assigned to | |
403 | * this group. | |
404 | */ | |
405 | unsigned long h_load; | |
c09595f6 | 406 | |
c8cba857 PZ |
407 | /* |
408 | * this cpu's part of tg->shares | |
409 | */ | |
410 | unsigned long shares; | |
f1d239f7 PZ |
411 | |
412 | /* | |
413 | * load.weight at the time we set shares | |
414 | */ | |
415 | unsigned long rq_weight; | |
c09595f6 | 416 | #endif |
6aa645ea IM |
417 | #endif |
418 | }; | |
1da177e4 | 419 | |
6aa645ea IM |
420 | /* Real-Time classes' related field in a runqueue: */ |
421 | struct rt_rq { | |
422 | struct rt_prio_array active; | |
63489e45 | 423 | unsigned long rt_nr_running; |
052f1dc7 | 424 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
425 | struct { |
426 | int curr; /* highest queued rt task prio */ | |
398a153b | 427 | #ifdef CONFIG_SMP |
e864c499 | 428 | int next; /* next highest */ |
398a153b | 429 | #endif |
e864c499 | 430 | } highest_prio; |
6f505b16 | 431 | #endif |
fa85ae24 | 432 | #ifdef CONFIG_SMP |
73fe6aae | 433 | unsigned long rt_nr_migratory; |
a1ba4d8b | 434 | unsigned long rt_nr_total; |
a22d7fc1 | 435 | int overloaded; |
917b627d | 436 | struct plist_head pushable_tasks; |
fa85ae24 | 437 | #endif |
6f505b16 | 438 | int rt_throttled; |
fa85ae24 | 439 | u64 rt_time; |
ac086bc2 | 440 | u64 rt_runtime; |
ea736ed5 | 441 | /* Nests inside the rq lock: */ |
0986b11b | 442 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 443 | |
052f1dc7 | 444 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
445 | unsigned long rt_nr_boosted; |
446 | ||
6f505b16 PZ |
447 | struct rq *rq; |
448 | struct list_head leaf_rt_rq_list; | |
449 | struct task_group *tg; | |
6f505b16 | 450 | #endif |
6aa645ea IM |
451 | }; |
452 | ||
57d885fe GH |
453 | #ifdef CONFIG_SMP |
454 | ||
455 | /* | |
456 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
457 | * variables. Each exclusive cpuset essentially defines an island domain by |
458 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
459 | * exclusive cpuset is created, we also create and attach a new root-domain |
460 | * object. | |
461 | * | |
57d885fe GH |
462 | */ |
463 | struct root_domain { | |
464 | atomic_t refcount; | |
c6c4927b RR |
465 | cpumask_var_t span; |
466 | cpumask_var_t online; | |
637f5085 | 467 | |
0eab9146 | 468 | /* |
637f5085 GH |
469 | * The "RT overload" flag: it gets set if a CPU has more than |
470 | * one runnable RT task. | |
471 | */ | |
c6c4927b | 472 | cpumask_var_t rto_mask; |
0eab9146 | 473 | atomic_t rto_count; |
6e0534f2 GH |
474 | #ifdef CONFIG_SMP |
475 | struct cpupri cpupri; | |
476 | #endif | |
57d885fe GH |
477 | }; |
478 | ||
dc938520 GH |
479 | /* |
480 | * By default the system creates a single root-domain with all cpus as | |
481 | * members (mimicking the global state we have today). | |
482 | */ | |
57d885fe GH |
483 | static struct root_domain def_root_domain; |
484 | ||
485 | #endif | |
486 | ||
1da177e4 LT |
487 | /* |
488 | * This is the main, per-CPU runqueue data structure. | |
489 | * | |
490 | * Locking rule: those places that want to lock multiple runqueues | |
491 | * (such as the load balancing or the thread migration code), lock | |
492 | * acquire operations must be ordered by ascending &runqueue. | |
493 | */ | |
70b97a7f | 494 | struct rq { |
d8016491 | 495 | /* runqueue lock: */ |
05fa785c | 496 | raw_spinlock_t lock; |
1da177e4 LT |
497 | |
498 | /* | |
499 | * nr_running and cpu_load should be in the same cacheline because | |
500 | * remote CPUs use both these fields when doing load calculation. | |
501 | */ | |
502 | unsigned long nr_running; | |
6aa645ea IM |
503 | #define CPU_LOAD_IDX_MAX 5 |
504 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c SS |
505 | #ifdef CONFIG_NO_HZ |
506 | unsigned char in_nohz_recently; | |
507 | #endif | |
d8016491 IM |
508 | /* capture load from *all* tasks on this cpu: */ |
509 | struct load_weight load; | |
6aa645ea IM |
510 | unsigned long nr_load_updates; |
511 | u64 nr_switches; | |
512 | ||
513 | struct cfs_rq cfs; | |
6f505b16 | 514 | struct rt_rq rt; |
6f505b16 | 515 | |
6aa645ea | 516 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
517 | /* list of leaf cfs_rq on this cpu: */ |
518 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
519 | #endif |
520 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 521 | struct list_head leaf_rt_rq_list; |
1da177e4 | 522 | #endif |
1da177e4 LT |
523 | |
524 | /* | |
525 | * This is part of a global counter where only the total sum | |
526 | * over all CPUs matters. A task can increase this counter on | |
527 | * one CPU and if it got migrated afterwards it may decrease | |
528 | * it on another CPU. Always updated under the runqueue lock: | |
529 | */ | |
530 | unsigned long nr_uninterruptible; | |
531 | ||
36c8b586 | 532 | struct task_struct *curr, *idle; |
c9819f45 | 533 | unsigned long next_balance; |
1da177e4 | 534 | struct mm_struct *prev_mm; |
6aa645ea | 535 | |
3e51f33f | 536 | u64 clock; |
6aa645ea | 537 | |
1da177e4 LT |
538 | atomic_t nr_iowait; |
539 | ||
540 | #ifdef CONFIG_SMP | |
0eab9146 | 541 | struct root_domain *rd; |
1da177e4 LT |
542 | struct sched_domain *sd; |
543 | ||
a0a522ce | 544 | unsigned char idle_at_tick; |
1da177e4 | 545 | /* For active balancing */ |
3f029d3c | 546 | int post_schedule; |
1da177e4 LT |
547 | int active_balance; |
548 | int push_cpu; | |
d8016491 IM |
549 | /* cpu of this runqueue: */ |
550 | int cpu; | |
1f11eb6a | 551 | int online; |
1da177e4 | 552 | |
a8a51d5e | 553 | unsigned long avg_load_per_task; |
1da177e4 | 554 | |
36c8b586 | 555 | struct task_struct *migration_thread; |
1da177e4 | 556 | struct list_head migration_queue; |
e9e9250b PZ |
557 | |
558 | u64 rt_avg; | |
559 | u64 age_stamp; | |
1b9508f6 MG |
560 | u64 idle_stamp; |
561 | u64 avg_idle; | |
1da177e4 LT |
562 | #endif |
563 | ||
dce48a84 TG |
564 | /* calc_load related fields */ |
565 | unsigned long calc_load_update; | |
566 | long calc_load_active; | |
567 | ||
8f4d37ec | 568 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
569 | #ifdef CONFIG_SMP |
570 | int hrtick_csd_pending; | |
571 | struct call_single_data hrtick_csd; | |
572 | #endif | |
8f4d37ec PZ |
573 | struct hrtimer hrtick_timer; |
574 | #endif | |
575 | ||
1da177e4 LT |
576 | #ifdef CONFIG_SCHEDSTATS |
577 | /* latency stats */ | |
578 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
579 | unsigned long long rq_cpu_time; |
580 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
581 | |
582 | /* sys_sched_yield() stats */ | |
480b9434 | 583 | unsigned int yld_count; |
1da177e4 LT |
584 | |
585 | /* schedule() stats */ | |
480b9434 KC |
586 | unsigned int sched_switch; |
587 | unsigned int sched_count; | |
588 | unsigned int sched_goidle; | |
1da177e4 LT |
589 | |
590 | /* try_to_wake_up() stats */ | |
480b9434 KC |
591 | unsigned int ttwu_count; |
592 | unsigned int ttwu_local; | |
b8efb561 IM |
593 | |
594 | /* BKL stats */ | |
480b9434 | 595 | unsigned int bkl_count; |
1da177e4 LT |
596 | #endif |
597 | }; | |
598 | ||
f34e3b61 | 599 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 600 | |
7d478721 PZ |
601 | static inline |
602 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 603 | { |
7d478721 | 604 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
dd41f596 IM |
605 | } |
606 | ||
0a2966b4 CL |
607 | static inline int cpu_of(struct rq *rq) |
608 | { | |
609 | #ifdef CONFIG_SMP | |
610 | return rq->cpu; | |
611 | #else | |
612 | return 0; | |
613 | #endif | |
614 | } | |
615 | ||
497f0ab3 | 616 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
617 | rcu_dereference_check((p), \ |
618 | rcu_read_lock_sched_held() || \ | |
619 | lockdep_is_held(&sched_domains_mutex)) | |
620 | ||
674311d5 NP |
621 | /* |
622 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 623 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
624 | * |
625 | * The domain tree of any CPU may only be accessed from within | |
626 | * preempt-disabled sections. | |
627 | */ | |
48f24c4d | 628 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 629 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
630 | |
631 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
632 | #define this_rq() (&__get_cpu_var(runqueues)) | |
633 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
634 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 635 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 636 | |
aa9c4c0f | 637 | inline void update_rq_clock(struct rq *rq) |
3e51f33f PZ |
638 | { |
639 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
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 | } | |
6d6bc0ad | 1242 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1243 | |
e9e9250b PZ |
1244 | static u64 sched_avg_period(void) |
1245 | { | |
1246 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1247 | } | |
1248 | ||
1249 | static void sched_avg_update(struct rq *rq) | |
1250 | { | |
1251 | s64 period = sched_avg_period(); | |
1252 | ||
1253 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1254 | rq->age_stamp += period; | |
1255 | rq->rt_avg /= 2; | |
1256 | } | |
1257 | } | |
1258 | ||
1259 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1260 | { | |
1261 | rq->rt_avg += rt_delta; | |
1262 | sched_avg_update(rq); | |
1263 | } | |
1264 | ||
6d6bc0ad | 1265 | #else /* !CONFIG_SMP */ |
31656519 | 1266 | static void resched_task(struct task_struct *p) |
c24d20db | 1267 | { |
05fa785c | 1268 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1269 | set_tsk_need_resched(p); |
c24d20db | 1270 | } |
e9e9250b PZ |
1271 | |
1272 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1273 | { | |
1274 | } | |
6d6bc0ad | 1275 | #endif /* CONFIG_SMP */ |
c24d20db | 1276 | |
45bf76df IM |
1277 | #if BITS_PER_LONG == 32 |
1278 | # define WMULT_CONST (~0UL) | |
1279 | #else | |
1280 | # define WMULT_CONST (1UL << 32) | |
1281 | #endif | |
1282 | ||
1283 | #define WMULT_SHIFT 32 | |
1284 | ||
194081eb IM |
1285 | /* |
1286 | * Shift right and round: | |
1287 | */ | |
cf2ab469 | 1288 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1289 | |
a7be37ac PZ |
1290 | /* |
1291 | * delta *= weight / lw | |
1292 | */ | |
cb1c4fc9 | 1293 | static unsigned long |
45bf76df IM |
1294 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1295 | struct load_weight *lw) | |
1296 | { | |
1297 | u64 tmp; | |
1298 | ||
7a232e03 LJ |
1299 | if (!lw->inv_weight) { |
1300 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1301 | lw->inv_weight = 1; | |
1302 | else | |
1303 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1304 | / (lw->weight+1); | |
1305 | } | |
45bf76df IM |
1306 | |
1307 | tmp = (u64)delta_exec * weight; | |
1308 | /* | |
1309 | * Check whether we'd overflow the 64-bit multiplication: | |
1310 | */ | |
194081eb | 1311 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1312 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1313 | WMULT_SHIFT/2); |
1314 | else | |
cf2ab469 | 1315 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1316 | |
ecf691da | 1317 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1318 | } |
1319 | ||
1091985b | 1320 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1321 | { |
1322 | lw->weight += inc; | |
e89996ae | 1323 | lw->inv_weight = 0; |
45bf76df IM |
1324 | } |
1325 | ||
1091985b | 1326 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1327 | { |
1328 | lw->weight -= dec; | |
e89996ae | 1329 | lw->inv_weight = 0; |
45bf76df IM |
1330 | } |
1331 | ||
2dd73a4f PW |
1332 | /* |
1333 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1334 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1335 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1336 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1337 | * scaled version of the new time slice allocation that they receive on time |
1338 | * slice expiry etc. | |
1339 | */ | |
1340 | ||
cce7ade8 PZ |
1341 | #define WEIGHT_IDLEPRIO 3 |
1342 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1343 | |
1344 | /* | |
1345 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1346 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1347 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1348 | * that remained on nice 0. | |
1349 | * | |
1350 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1351 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1352 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1353 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1354 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1355 | */ |
1356 | static const int prio_to_weight[40] = { | |
254753dc IM |
1357 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1358 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1359 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1360 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1361 | /* 0 */ 1024, 820, 655, 526, 423, | |
1362 | /* 5 */ 335, 272, 215, 172, 137, | |
1363 | /* 10 */ 110, 87, 70, 56, 45, | |
1364 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1365 | }; |
1366 | ||
5714d2de IM |
1367 | /* |
1368 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1369 | * | |
1370 | * In cases where the weight does not change often, we can use the | |
1371 | * precalculated inverse to speed up arithmetics by turning divisions | |
1372 | * into multiplications: | |
1373 | */ | |
dd41f596 | 1374 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1375 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1376 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1377 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1378 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1379 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1380 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1381 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1382 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1383 | }; |
2dd73a4f | 1384 | |
ef12fefa BR |
1385 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1386 | enum cpuacct_stat_index { | |
1387 | CPUACCT_STAT_USER, /* ... user mode */ | |
1388 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1389 | ||
1390 | CPUACCT_STAT_NSTATS, | |
1391 | }; | |
1392 | ||
d842de87 SV |
1393 | #ifdef CONFIG_CGROUP_CPUACCT |
1394 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1395 | static void cpuacct_update_stats(struct task_struct *tsk, |
1396 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1397 | #else |
1398 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1399 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1400 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1401 | #endif |
1402 | ||
18d95a28 PZ |
1403 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1404 | { | |
1405 | update_load_add(&rq->load, load); | |
1406 | } | |
1407 | ||
1408 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1409 | { | |
1410 | update_load_sub(&rq->load, load); | |
1411 | } | |
1412 | ||
7940ca36 | 1413 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1414 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1415 | |
1416 | /* | |
1417 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1418 | * leaving it for the final time. | |
1419 | */ | |
eb755805 | 1420 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1421 | { |
1422 | struct task_group *parent, *child; | |
eb755805 | 1423 | int ret; |
c09595f6 PZ |
1424 | |
1425 | rcu_read_lock(); | |
1426 | parent = &root_task_group; | |
1427 | down: | |
eb755805 PZ |
1428 | ret = (*down)(parent, data); |
1429 | if (ret) | |
1430 | goto out_unlock; | |
c09595f6 PZ |
1431 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1432 | parent = child; | |
1433 | goto down; | |
1434 | ||
1435 | up: | |
1436 | continue; | |
1437 | } | |
eb755805 PZ |
1438 | ret = (*up)(parent, data); |
1439 | if (ret) | |
1440 | goto out_unlock; | |
c09595f6 PZ |
1441 | |
1442 | child = parent; | |
1443 | parent = parent->parent; | |
1444 | if (parent) | |
1445 | goto up; | |
eb755805 | 1446 | out_unlock: |
c09595f6 | 1447 | rcu_read_unlock(); |
eb755805 PZ |
1448 | |
1449 | return ret; | |
c09595f6 PZ |
1450 | } |
1451 | ||
eb755805 PZ |
1452 | static int tg_nop(struct task_group *tg, void *data) |
1453 | { | |
1454 | return 0; | |
c09595f6 | 1455 | } |
eb755805 PZ |
1456 | #endif |
1457 | ||
1458 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1459 | /* Used instead of source_load when we know the type == 0 */ |
1460 | static unsigned long weighted_cpuload(const int cpu) | |
1461 | { | |
1462 | return cpu_rq(cpu)->load.weight; | |
1463 | } | |
1464 | ||
1465 | /* | |
1466 | * Return a low guess at the load of a migration-source cpu weighted | |
1467 | * according to the scheduling class and "nice" value. | |
1468 | * | |
1469 | * We want to under-estimate the load of migration sources, to | |
1470 | * balance conservatively. | |
1471 | */ | |
1472 | static unsigned long source_load(int cpu, int type) | |
1473 | { | |
1474 | struct rq *rq = cpu_rq(cpu); | |
1475 | unsigned long total = weighted_cpuload(cpu); | |
1476 | ||
1477 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1478 | return total; | |
1479 | ||
1480 | return min(rq->cpu_load[type-1], total); | |
1481 | } | |
1482 | ||
1483 | /* | |
1484 | * Return a high guess at the load of a migration-target cpu weighted | |
1485 | * according to the scheduling class and "nice" value. | |
1486 | */ | |
1487 | static unsigned long target_load(int cpu, int type) | |
1488 | { | |
1489 | struct rq *rq = cpu_rq(cpu); | |
1490 | unsigned long total = weighted_cpuload(cpu); | |
1491 | ||
1492 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1493 | return total; | |
1494 | ||
1495 | return max(rq->cpu_load[type-1], total); | |
1496 | } | |
1497 | ||
ae154be1 PZ |
1498 | static struct sched_group *group_of(int cpu) |
1499 | { | |
d11c563d | 1500 | struct sched_domain *sd = rcu_dereference_sched(cpu_rq(cpu)->sd); |
ae154be1 PZ |
1501 | |
1502 | if (!sd) | |
1503 | return NULL; | |
1504 | ||
1505 | return sd->groups; | |
1506 | } | |
1507 | ||
1508 | static unsigned long power_of(int cpu) | |
1509 | { | |
1510 | struct sched_group *group = group_of(cpu); | |
1511 | ||
1512 | if (!group) | |
1513 | return SCHED_LOAD_SCALE; | |
1514 | ||
1515 | return group->cpu_power; | |
1516 | } | |
1517 | ||
eb755805 PZ |
1518 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1519 | ||
1520 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1521 | { | |
1522 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1523 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1524 | |
4cd42620 SR |
1525 | if (nr_running) |
1526 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1527 | else |
1528 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1529 | |
1530 | return rq->avg_load_per_task; | |
1531 | } | |
1532 | ||
1533 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1534 | |
43cf38eb | 1535 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1536 | |
c09595f6 PZ |
1537 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1538 | ||
1539 | /* | |
1540 | * Calculate and set the cpu's group shares. | |
1541 | */ | |
34d76c41 PZ |
1542 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1543 | unsigned long sd_shares, | |
1544 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1545 | unsigned long *usd_rq_weight) |
18d95a28 | 1546 | { |
34d76c41 | 1547 | unsigned long shares, rq_weight; |
a5004278 | 1548 | int boost = 0; |
c09595f6 | 1549 | |
4a6cc4bd | 1550 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1551 | if (!rq_weight) { |
1552 | boost = 1; | |
1553 | rq_weight = NICE_0_LOAD; | |
1554 | } | |
c8cba857 | 1555 | |
c09595f6 | 1556 | /* |
a8af7246 PZ |
1557 | * \Sum_j shares_j * rq_weight_i |
1558 | * shares_i = ----------------------------- | |
1559 | * \Sum_j rq_weight_j | |
c09595f6 | 1560 | */ |
ec4e0e2f | 1561 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1562 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1563 | |
ffda12a1 PZ |
1564 | if (abs(shares - tg->se[cpu]->load.weight) > |
1565 | sysctl_sched_shares_thresh) { | |
1566 | struct rq *rq = cpu_rq(cpu); | |
1567 | unsigned long flags; | |
c09595f6 | 1568 | |
05fa785c | 1569 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1570 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1571 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1572 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1573 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1574 | } |
18d95a28 | 1575 | } |
c09595f6 PZ |
1576 | |
1577 | /* | |
c8cba857 PZ |
1578 | * Re-compute the task group their per cpu shares over the given domain. |
1579 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1580 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1581 | */ |
eb755805 | 1582 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1583 | { |
cd8ad40d | 1584 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1585 | unsigned long *usd_rq_weight; |
eb755805 | 1586 | struct sched_domain *sd = data; |
34d76c41 | 1587 | unsigned long flags; |
c8cba857 | 1588 | int i; |
c09595f6 | 1589 | |
34d76c41 PZ |
1590 | if (!tg->se[0]) |
1591 | return 0; | |
1592 | ||
1593 | local_irq_save(flags); | |
4a6cc4bd | 1594 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1595 | |
758b2cdc | 1596 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1597 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1598 | usd_rq_weight[i] = weight; |
34d76c41 | 1599 | |
cd8ad40d | 1600 | rq_weight += weight; |
ec4e0e2f KC |
1601 | /* |
1602 | * If there are currently no tasks on the cpu pretend there | |
1603 | * is one of average load so that when a new task gets to | |
1604 | * run here it will not get delayed by group starvation. | |
1605 | */ | |
ec4e0e2f KC |
1606 | if (!weight) |
1607 | weight = NICE_0_LOAD; | |
1608 | ||
cd8ad40d | 1609 | sum_weight += weight; |
c8cba857 | 1610 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1611 | } |
c09595f6 | 1612 | |
cd8ad40d PZ |
1613 | if (!rq_weight) |
1614 | rq_weight = sum_weight; | |
1615 | ||
c8cba857 PZ |
1616 | if ((!shares && rq_weight) || shares > tg->shares) |
1617 | shares = tg->shares; | |
1618 | ||
1619 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1620 | shares = tg->shares; | |
c09595f6 | 1621 | |
758b2cdc | 1622 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1623 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1624 | |
1625 | local_irq_restore(flags); | |
eb755805 PZ |
1626 | |
1627 | return 0; | |
c09595f6 PZ |
1628 | } |
1629 | ||
1630 | /* | |
c8cba857 PZ |
1631 | * Compute the cpu's hierarchical load factor for each task group. |
1632 | * This needs to be done in a top-down fashion because the load of a child | |
1633 | * group is a fraction of its parents load. | |
c09595f6 | 1634 | */ |
eb755805 | 1635 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1636 | { |
c8cba857 | 1637 | unsigned long load; |
eb755805 | 1638 | long cpu = (long)data; |
c09595f6 | 1639 | |
c8cba857 PZ |
1640 | if (!tg->parent) { |
1641 | load = cpu_rq(cpu)->load.weight; | |
1642 | } else { | |
1643 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1644 | load *= tg->cfs_rq[cpu]->shares; | |
1645 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1646 | } | |
c09595f6 | 1647 | |
c8cba857 | 1648 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1649 | |
eb755805 | 1650 | return 0; |
c09595f6 PZ |
1651 | } |
1652 | ||
c8cba857 | 1653 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1654 | { |
e7097159 PZ |
1655 | s64 elapsed; |
1656 | u64 now; | |
1657 | ||
1658 | if (root_task_group_empty()) | |
1659 | return; | |
1660 | ||
1661 | now = cpu_clock(raw_smp_processor_id()); | |
1662 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1663 | |
1664 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1665 | sd->last_update = now; | |
eb755805 | 1666 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1667 | } |
4d8d595d PZ |
1668 | } |
1669 | ||
eb755805 | 1670 | static void update_h_load(long cpu) |
c09595f6 | 1671 | { |
e7097159 PZ |
1672 | if (root_task_group_empty()) |
1673 | return; | |
1674 | ||
eb755805 | 1675 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1676 | } |
1677 | ||
c09595f6 PZ |
1678 | #else |
1679 | ||
c8cba857 | 1680 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1681 | { |
1682 | } | |
1683 | ||
18d95a28 PZ |
1684 | #endif |
1685 | ||
8f45e2b5 GH |
1686 | #ifdef CONFIG_PREEMPT |
1687 | ||
b78bb868 PZ |
1688 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1689 | ||
70574a99 | 1690 | /* |
8f45e2b5 GH |
1691 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1692 | * way at the expense of forcing extra atomic operations in all | |
1693 | * invocations. This assures that the double_lock is acquired using the | |
1694 | * same underlying policy as the spinlock_t on this architecture, which | |
1695 | * reduces latency compared to the unfair variant below. However, it | |
1696 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1697 | */ |
8f45e2b5 GH |
1698 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1699 | __releases(this_rq->lock) | |
1700 | __acquires(busiest->lock) | |
1701 | __acquires(this_rq->lock) | |
1702 | { | |
05fa785c | 1703 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1704 | double_rq_lock(this_rq, busiest); |
1705 | ||
1706 | return 1; | |
1707 | } | |
1708 | ||
1709 | #else | |
1710 | /* | |
1711 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1712 | * latency by eliminating extra atomic operations when the locks are | |
1713 | * already in proper order on entry. This favors lower cpu-ids and will | |
1714 | * grant the double lock to lower cpus over higher ids under contention, | |
1715 | * regardless of entry order into the function. | |
1716 | */ | |
1717 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1718 | __releases(this_rq->lock) |
1719 | __acquires(busiest->lock) | |
1720 | __acquires(this_rq->lock) | |
1721 | { | |
1722 | int ret = 0; | |
1723 | ||
05fa785c | 1724 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1725 | if (busiest < this_rq) { |
05fa785c TG |
1726 | raw_spin_unlock(&this_rq->lock); |
1727 | raw_spin_lock(&busiest->lock); | |
1728 | raw_spin_lock_nested(&this_rq->lock, | |
1729 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1730 | ret = 1; |
1731 | } else | |
05fa785c TG |
1732 | raw_spin_lock_nested(&busiest->lock, |
1733 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1734 | } |
1735 | return ret; | |
1736 | } | |
1737 | ||
8f45e2b5 GH |
1738 | #endif /* CONFIG_PREEMPT */ |
1739 | ||
1740 | /* | |
1741 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1742 | */ | |
1743 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1744 | { | |
1745 | if (unlikely(!irqs_disabled())) { | |
1746 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1747 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1748 | BUG_ON(1); |
1749 | } | |
1750 | ||
1751 | return _double_lock_balance(this_rq, busiest); | |
1752 | } | |
1753 | ||
70574a99 AD |
1754 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1755 | __releases(busiest->lock) | |
1756 | { | |
05fa785c | 1757 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1758 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1759 | } | |
1e3c88bd PZ |
1760 | |
1761 | /* | |
1762 | * double_rq_lock - safely lock two runqueues | |
1763 | * | |
1764 | * Note this does not disable interrupts like task_rq_lock, | |
1765 | * you need to do so manually before calling. | |
1766 | */ | |
1767 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1768 | __acquires(rq1->lock) | |
1769 | __acquires(rq2->lock) | |
1770 | { | |
1771 | BUG_ON(!irqs_disabled()); | |
1772 | if (rq1 == rq2) { | |
1773 | raw_spin_lock(&rq1->lock); | |
1774 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1775 | } else { | |
1776 | if (rq1 < rq2) { | |
1777 | raw_spin_lock(&rq1->lock); | |
1778 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1779 | } else { | |
1780 | raw_spin_lock(&rq2->lock); | |
1781 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1782 | } | |
1783 | } | |
1784 | update_rq_clock(rq1); | |
1785 | update_rq_clock(rq2); | |
1786 | } | |
1787 | ||
1788 | /* | |
1789 | * double_rq_unlock - safely unlock two runqueues | |
1790 | * | |
1791 | * Note this does not restore interrupts like task_rq_unlock, | |
1792 | * you need to do so manually after calling. | |
1793 | */ | |
1794 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1795 | __releases(rq1->lock) | |
1796 | __releases(rq2->lock) | |
1797 | { | |
1798 | raw_spin_unlock(&rq1->lock); | |
1799 | if (rq1 != rq2) | |
1800 | raw_spin_unlock(&rq2->lock); | |
1801 | else | |
1802 | __release(rq2->lock); | |
1803 | } | |
1804 | ||
18d95a28 PZ |
1805 | #endif |
1806 | ||
30432094 | 1807 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1808 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1809 | { | |
30432094 | 1810 | #ifdef CONFIG_SMP |
34e83e85 IM |
1811 | cfs_rq->shares = shares; |
1812 | #endif | |
1813 | } | |
30432094 | 1814 | #endif |
e7693a36 | 1815 | |
dce48a84 | 1816 | static void calc_load_account_active(struct rq *this_rq); |
0bcdcf28 | 1817 | static void update_sysctl(void); |
acb4a848 | 1818 | static int get_update_sysctl_factor(void); |
dce48a84 | 1819 | |
cd29fe6f PZ |
1820 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1821 | { | |
1822 | set_task_rq(p, cpu); | |
1823 | #ifdef CONFIG_SMP | |
1824 | /* | |
1825 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1826 | * successfuly executed on another CPU. We must ensure that updates of | |
1827 | * per-task data have been completed by this moment. | |
1828 | */ | |
1829 | smp_wmb(); | |
1830 | task_thread_info(p)->cpu = cpu; | |
1831 | #endif | |
1832 | } | |
dce48a84 | 1833 | |
1e3c88bd | 1834 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1835 | |
1836 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1837 | #define for_each_class(class) \ |
1838 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1839 | |
1e3c88bd PZ |
1840 | #include "sched_stats.h" |
1841 | ||
c09595f6 | 1842 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1843 | { |
1844 | rq->nr_running++; | |
9c217245 IM |
1845 | } |
1846 | ||
c09595f6 | 1847 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1848 | { |
1849 | rq->nr_running--; | |
9c217245 IM |
1850 | } |
1851 | ||
45bf76df IM |
1852 | static void set_load_weight(struct task_struct *p) |
1853 | { | |
1854 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1855 | p->se.load.weight = prio_to_weight[0] * 2; |
1856 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1857 | return; | |
1858 | } | |
45bf76df | 1859 | |
dd41f596 IM |
1860 | /* |
1861 | * SCHED_IDLE tasks get minimal weight: | |
1862 | */ | |
1863 | if (p->policy == SCHED_IDLE) { | |
1864 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1865 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1866 | return; | |
1867 | } | |
71f8bd46 | 1868 | |
dd41f596 IM |
1869 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1870 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1871 | } |
1872 | ||
2087a1ad GH |
1873 | static void update_avg(u64 *avg, u64 sample) |
1874 | { | |
1875 | s64 diff = sample - *avg; | |
1876 | *avg += diff >> 3; | |
1877 | } | |
1878 | ||
ea87bb78 TG |
1879 | static void |
1880 | enqueue_task(struct rq *rq, struct task_struct *p, int wakeup, bool head) | |
71f8bd46 | 1881 | { |
831451ac PZ |
1882 | if (wakeup) |
1883 | p->se.start_runtime = p->se.sum_exec_runtime; | |
1884 | ||
dd41f596 | 1885 | sched_info_queued(p); |
ea87bb78 | 1886 | p->sched_class->enqueue_task(rq, p, wakeup, head); |
dd41f596 | 1887 | p->se.on_rq = 1; |
71f8bd46 IM |
1888 | } |
1889 | ||
69be72c1 | 1890 | static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) |
71f8bd46 | 1891 | { |
831451ac PZ |
1892 | if (sleep) { |
1893 | if (p->se.last_wakeup) { | |
1894 | update_avg(&p->se.avg_overlap, | |
1895 | p->se.sum_exec_runtime - p->se.last_wakeup); | |
1896 | p->se.last_wakeup = 0; | |
1897 | } else { | |
1898 | update_avg(&p->se.avg_wakeup, | |
1899 | sysctl_sched_wakeup_granularity); | |
1900 | } | |
2087a1ad GH |
1901 | } |
1902 | ||
46ac22ba | 1903 | sched_info_dequeued(p); |
f02231e5 | 1904 | p->sched_class->dequeue_task(rq, p, sleep); |
dd41f596 | 1905 | p->se.on_rq = 0; |
71f8bd46 IM |
1906 | } |
1907 | ||
1e3c88bd PZ |
1908 | /* |
1909 | * activate_task - move a task to the runqueue. | |
1910 | */ | |
1911 | static void activate_task(struct rq *rq, struct task_struct *p, int wakeup) | |
1912 | { | |
1913 | if (task_contributes_to_load(p)) | |
1914 | rq->nr_uninterruptible--; | |
1915 | ||
ea87bb78 | 1916 | enqueue_task(rq, p, wakeup, false); |
1e3c88bd PZ |
1917 | inc_nr_running(rq); |
1918 | } | |
1919 | ||
1920 | /* | |
1921 | * deactivate_task - remove a task from the runqueue. | |
1922 | */ | |
1923 | static void deactivate_task(struct rq *rq, struct task_struct *p, int sleep) | |
1924 | { | |
1925 | if (task_contributes_to_load(p)) | |
1926 | rq->nr_uninterruptible++; | |
1927 | ||
1928 | dequeue_task(rq, p, sleep); | |
1929 | dec_nr_running(rq); | |
1930 | } | |
1931 | ||
1932 | #include "sched_idletask.c" | |
1933 | #include "sched_fair.c" | |
1934 | #include "sched_rt.c" | |
1935 | #ifdef CONFIG_SCHED_DEBUG | |
1936 | # include "sched_debug.c" | |
1937 | #endif | |
1938 | ||
14531189 | 1939 | /* |
dd41f596 | 1940 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1941 | */ |
14531189 IM |
1942 | static inline int __normal_prio(struct task_struct *p) |
1943 | { | |
dd41f596 | 1944 | return p->static_prio; |
14531189 IM |
1945 | } |
1946 | ||
b29739f9 IM |
1947 | /* |
1948 | * Calculate the expected normal priority: i.e. priority | |
1949 | * without taking RT-inheritance into account. Might be | |
1950 | * boosted by interactivity modifiers. Changes upon fork, | |
1951 | * setprio syscalls, and whenever the interactivity | |
1952 | * estimator recalculates. | |
1953 | */ | |
36c8b586 | 1954 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1955 | { |
1956 | int prio; | |
1957 | ||
e05606d3 | 1958 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1959 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1960 | else | |
1961 | prio = __normal_prio(p); | |
1962 | return prio; | |
1963 | } | |
1964 | ||
1965 | /* | |
1966 | * Calculate the current priority, i.e. the priority | |
1967 | * taken into account by the scheduler. This value might | |
1968 | * be boosted by RT tasks, or might be boosted by | |
1969 | * interactivity modifiers. Will be RT if the task got | |
1970 | * RT-boosted. If not then it returns p->normal_prio. | |
1971 | */ | |
36c8b586 | 1972 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1973 | { |
1974 | p->normal_prio = normal_prio(p); | |
1975 | /* | |
1976 | * If we are RT tasks or we were boosted to RT priority, | |
1977 | * keep the priority unchanged. Otherwise, update priority | |
1978 | * to the normal priority: | |
1979 | */ | |
1980 | if (!rt_prio(p->prio)) | |
1981 | return p->normal_prio; | |
1982 | return p->prio; | |
1983 | } | |
1984 | ||
1da177e4 LT |
1985 | /** |
1986 | * task_curr - is this task currently executing on a CPU? | |
1987 | * @p: the task in question. | |
1988 | */ | |
36c8b586 | 1989 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1990 | { |
1991 | return cpu_curr(task_cpu(p)) == p; | |
1992 | } | |
1993 | ||
cb469845 SR |
1994 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1995 | const struct sched_class *prev_class, | |
1996 | int oldprio, int running) | |
1997 | { | |
1998 | if (prev_class != p->sched_class) { | |
1999 | if (prev_class->switched_from) | |
2000 | prev_class->switched_from(rq, p, running); | |
2001 | p->sched_class->switched_to(rq, p, running); | |
2002 | } else | |
2003 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
2004 | } | |
2005 | ||
1da177e4 | 2006 | #ifdef CONFIG_SMP |
cc367732 IM |
2007 | /* |
2008 | * Is this task likely cache-hot: | |
2009 | */ | |
e7693a36 | 2010 | static int |
cc367732 IM |
2011 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2012 | { | |
2013 | s64 delta; | |
2014 | ||
e6c8fba7 PZ |
2015 | if (p->sched_class != &fair_sched_class) |
2016 | return 0; | |
2017 | ||
f540a608 IM |
2018 | /* |
2019 | * Buddy candidates are cache hot: | |
2020 | */ | |
f685ceac | 2021 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2022 | (&p->se == cfs_rq_of(&p->se)->next || |
2023 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2024 | return 1; |
2025 | ||
6bc1665b IM |
2026 | if (sysctl_sched_migration_cost == -1) |
2027 | return 1; | |
2028 | if (sysctl_sched_migration_cost == 0) | |
2029 | return 0; | |
2030 | ||
cc367732 IM |
2031 | delta = now - p->se.exec_start; |
2032 | ||
2033 | return delta < (s64)sysctl_sched_migration_cost; | |
2034 | } | |
2035 | ||
dd41f596 | 2036 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2037 | { |
e2912009 PZ |
2038 | #ifdef CONFIG_SCHED_DEBUG |
2039 | /* | |
2040 | * We should never call set_task_cpu() on a blocked task, | |
2041 | * ttwu() will sort out the placement. | |
2042 | */ | |
077614ee PZ |
2043 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2044 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2045 | #endif |
2046 | ||
de1d7286 | 2047 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2048 | |
0c69774e PZ |
2049 | if (task_cpu(p) != new_cpu) { |
2050 | p->se.nr_migrations++; | |
2051 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2052 | } | |
dd41f596 IM |
2053 | |
2054 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2055 | } |
2056 | ||
70b97a7f | 2057 | struct migration_req { |
1da177e4 | 2058 | struct list_head list; |
1da177e4 | 2059 | |
36c8b586 | 2060 | struct task_struct *task; |
1da177e4 LT |
2061 | int dest_cpu; |
2062 | ||
1da177e4 | 2063 | struct completion done; |
70b97a7f | 2064 | }; |
1da177e4 LT |
2065 | |
2066 | /* | |
2067 | * The task's runqueue lock must be held. | |
2068 | * Returns true if you have to wait for migration thread. | |
2069 | */ | |
36c8b586 | 2070 | static int |
70b97a7f | 2071 | migrate_task(struct task_struct *p, int dest_cpu, struct migration_req *req) |
1da177e4 | 2072 | { |
70b97a7f | 2073 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2074 | |
2075 | /* | |
2076 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2077 | * the next wake-up will properly place the task. |
1da177e4 | 2078 | */ |
e2912009 | 2079 | if (!p->se.on_rq && !task_running(rq, p)) |
1da177e4 | 2080 | return 0; |
1da177e4 LT |
2081 | |
2082 | init_completion(&req->done); | |
1da177e4 LT |
2083 | req->task = p; |
2084 | req->dest_cpu = dest_cpu; | |
2085 | list_add(&req->list, &rq->migration_queue); | |
48f24c4d | 2086 | |
1da177e4 LT |
2087 | return 1; |
2088 | } | |
2089 | ||
a26b89f0 MM |
2090 | /* |
2091 | * wait_task_context_switch - wait for a thread to complete at least one | |
2092 | * context switch. | |
2093 | * | |
2094 | * @p must not be current. | |
2095 | */ | |
2096 | void wait_task_context_switch(struct task_struct *p) | |
2097 | { | |
2098 | unsigned long nvcsw, nivcsw, flags; | |
2099 | int running; | |
2100 | struct rq *rq; | |
2101 | ||
2102 | nvcsw = p->nvcsw; | |
2103 | nivcsw = p->nivcsw; | |
2104 | for (;;) { | |
2105 | /* | |
2106 | * The runqueue is assigned before the actual context | |
2107 | * switch. We need to take the runqueue lock. | |
2108 | * | |
2109 | * We could check initially without the lock but it is | |
2110 | * very likely that we need to take the lock in every | |
2111 | * iteration. | |
2112 | */ | |
2113 | rq = task_rq_lock(p, &flags); | |
2114 | running = task_running(rq, p); | |
2115 | task_rq_unlock(rq, &flags); | |
2116 | ||
2117 | if (likely(!running)) | |
2118 | break; | |
2119 | /* | |
2120 | * The switch count is incremented before the actual | |
2121 | * context switch. We thus wait for two switches to be | |
2122 | * sure at least one completed. | |
2123 | */ | |
2124 | if ((p->nvcsw - nvcsw) > 1) | |
2125 | break; | |
2126 | if ((p->nivcsw - nivcsw) > 1) | |
2127 | break; | |
2128 | ||
2129 | cpu_relax(); | |
2130 | } | |
2131 | } | |
2132 | ||
1da177e4 LT |
2133 | /* |
2134 | * wait_task_inactive - wait for a thread to unschedule. | |
2135 | * | |
85ba2d86 RM |
2136 | * If @match_state is nonzero, it's the @p->state value just checked and |
2137 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2138 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2139 | * we return a positive number (its total switch count). If a second call | |
2140 | * a short while later returns the same number, the caller can be sure that | |
2141 | * @p has remained unscheduled the whole time. | |
2142 | * | |
1da177e4 LT |
2143 | * The caller must ensure that the task *will* unschedule sometime soon, |
2144 | * else this function might spin for a *long* time. This function can't | |
2145 | * be called with interrupts off, or it may introduce deadlock with | |
2146 | * smp_call_function() if an IPI is sent by the same process we are | |
2147 | * waiting to become inactive. | |
2148 | */ | |
85ba2d86 | 2149 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2150 | { |
2151 | unsigned long flags; | |
dd41f596 | 2152 | int running, on_rq; |
85ba2d86 | 2153 | unsigned long ncsw; |
70b97a7f | 2154 | struct rq *rq; |
1da177e4 | 2155 | |
3a5c359a AK |
2156 | for (;;) { |
2157 | /* | |
2158 | * We do the initial early heuristics without holding | |
2159 | * any task-queue locks at all. We'll only try to get | |
2160 | * the runqueue lock when things look like they will | |
2161 | * work out! | |
2162 | */ | |
2163 | rq = task_rq(p); | |
fa490cfd | 2164 | |
3a5c359a AK |
2165 | /* |
2166 | * If the task is actively running on another CPU | |
2167 | * still, just relax and busy-wait without holding | |
2168 | * any locks. | |
2169 | * | |
2170 | * NOTE! Since we don't hold any locks, it's not | |
2171 | * even sure that "rq" stays as the right runqueue! | |
2172 | * But we don't care, since "task_running()" will | |
2173 | * return false if the runqueue has changed and p | |
2174 | * is actually now running somewhere else! | |
2175 | */ | |
85ba2d86 RM |
2176 | while (task_running(rq, p)) { |
2177 | if (match_state && unlikely(p->state != match_state)) | |
2178 | return 0; | |
3a5c359a | 2179 | cpu_relax(); |
85ba2d86 | 2180 | } |
fa490cfd | 2181 | |
3a5c359a AK |
2182 | /* |
2183 | * Ok, time to look more closely! We need the rq | |
2184 | * lock now, to be *sure*. If we're wrong, we'll | |
2185 | * just go back and repeat. | |
2186 | */ | |
2187 | rq = task_rq_lock(p, &flags); | |
0a16b607 | 2188 | trace_sched_wait_task(rq, p); |
3a5c359a AK |
2189 | running = task_running(rq, p); |
2190 | on_rq = p->se.on_rq; | |
85ba2d86 | 2191 | ncsw = 0; |
f31e11d8 | 2192 | if (!match_state || p->state == match_state) |
93dcf55f | 2193 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2194 | task_rq_unlock(rq, &flags); |
fa490cfd | 2195 | |
85ba2d86 RM |
2196 | /* |
2197 | * If it changed from the expected state, bail out now. | |
2198 | */ | |
2199 | if (unlikely(!ncsw)) | |
2200 | break; | |
2201 | ||
3a5c359a AK |
2202 | /* |
2203 | * Was it really running after all now that we | |
2204 | * checked with the proper locks actually held? | |
2205 | * | |
2206 | * Oops. Go back and try again.. | |
2207 | */ | |
2208 | if (unlikely(running)) { | |
2209 | cpu_relax(); | |
2210 | continue; | |
2211 | } | |
fa490cfd | 2212 | |
3a5c359a AK |
2213 | /* |
2214 | * It's not enough that it's not actively running, | |
2215 | * it must be off the runqueue _entirely_, and not | |
2216 | * preempted! | |
2217 | * | |
80dd99b3 | 2218 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2219 | * running right now), it's preempted, and we should |
2220 | * yield - it could be a while. | |
2221 | */ | |
2222 | if (unlikely(on_rq)) { | |
2223 | schedule_timeout_uninterruptible(1); | |
2224 | continue; | |
2225 | } | |
fa490cfd | 2226 | |
3a5c359a AK |
2227 | /* |
2228 | * Ahh, all good. It wasn't running, and it wasn't | |
2229 | * runnable, which means that it will never become | |
2230 | * running in the future either. We're all done! | |
2231 | */ | |
2232 | break; | |
2233 | } | |
85ba2d86 RM |
2234 | |
2235 | return ncsw; | |
1da177e4 LT |
2236 | } |
2237 | ||
2238 | /*** | |
2239 | * kick_process - kick a running thread to enter/exit the kernel | |
2240 | * @p: the to-be-kicked thread | |
2241 | * | |
2242 | * Cause a process which is running on another CPU to enter | |
2243 | * kernel-mode, without any delay. (to get signals handled.) | |
2244 | * | |
2245 | * NOTE: this function doesnt have to take the runqueue lock, | |
2246 | * because all it wants to ensure is that the remote task enters | |
2247 | * the kernel. If the IPI races and the task has been migrated | |
2248 | * to another CPU then no harm is done and the purpose has been | |
2249 | * achieved as well. | |
2250 | */ | |
36c8b586 | 2251 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2252 | { |
2253 | int cpu; | |
2254 | ||
2255 | preempt_disable(); | |
2256 | cpu = task_cpu(p); | |
2257 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2258 | smp_send_reschedule(cpu); | |
2259 | preempt_enable(); | |
2260 | } | |
b43e3521 | 2261 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2262 | #endif /* CONFIG_SMP */ |
1da177e4 | 2263 | |
0793a61d TG |
2264 | /** |
2265 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2266 | * @p: the task to evaluate | |
2267 | * @func: the function to be called | |
2268 | * @info: the function call argument | |
2269 | * | |
2270 | * Calls the function @func when the task is currently running. This might | |
2271 | * be on the current CPU, which just calls the function directly | |
2272 | */ | |
2273 | void task_oncpu_function_call(struct task_struct *p, | |
2274 | void (*func) (void *info), void *info) | |
2275 | { | |
2276 | int cpu; | |
2277 | ||
2278 | preempt_disable(); | |
2279 | cpu = task_cpu(p); | |
2280 | if (task_curr(p)) | |
2281 | smp_call_function_single(cpu, func, info, 1); | |
2282 | preempt_enable(); | |
2283 | } | |
2284 | ||
970b13ba | 2285 | #ifdef CONFIG_SMP |
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. */ | |
2302 | if (dest_cpu >= nr_cpu_ids) { | |
2303 | rcu_read_lock(); | |
2304 | cpuset_cpus_allowed_locked(p, &p->cpus_allowed); | |
2305 | rcu_read_unlock(); | |
2306 | dest_cpu = cpumask_any_and(cpu_active_mask, &p->cpus_allowed); | |
2307 | ||
2308 | /* | |
2309 | * Don't tell them about moving exiting tasks or | |
2310 | * kernel threads (both mm NULL), since they never | |
2311 | * leave kernel. | |
2312 | */ | |
2313 | if (p->mm && printk_ratelimit()) { | |
2314 | printk(KERN_INFO "process %d (%s) no " | |
2315 | "longer affine to cpu%d\n", | |
2316 | task_pid_nr(p), p->comm, cpu); | |
2317 | } | |
2318 | } | |
2319 | ||
2320 | return dest_cpu; | |
2321 | } | |
2322 | ||
e2912009 | 2323 | /* |
fabf318e PZ |
2324 | * Gets called from 3 sites (exec, fork, wakeup), since it is called without |
2325 | * holding rq->lock we need to ensure ->cpus_allowed is stable, this is done | |
2326 | * by: | |
e2912009 | 2327 | * |
fabf318e PZ |
2328 | * exec: is unstable, retry loop |
2329 | * fork & wake-up: serialize ->cpus_allowed against TASK_WAKING | |
e2912009 | 2330 | */ |
970b13ba PZ |
2331 | static inline |
2332 | int select_task_rq(struct task_struct *p, int sd_flags, int wake_flags) | |
2333 | { | |
e2912009 PZ |
2334 | int cpu = p->sched_class->select_task_rq(p, sd_flags, wake_flags); |
2335 | ||
2336 | /* | |
2337 | * In order not to call set_task_cpu() on a blocking task we need | |
2338 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2339 | * cpu. | |
2340 | * | |
2341 | * Since this is common to all placement strategies, this lives here. | |
2342 | * | |
2343 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2344 | * not worry about this generic constraint ] | |
2345 | */ | |
2346 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2347 | !cpu_online(cpu))) |
5da9a0fb | 2348 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2349 | |
2350 | return cpu; | |
970b13ba PZ |
2351 | } |
2352 | #endif | |
2353 | ||
1da177e4 LT |
2354 | /*** |
2355 | * try_to_wake_up - wake up a thread | |
2356 | * @p: the to-be-woken-up thread | |
2357 | * @state: the mask of task states that can be woken | |
2358 | * @sync: do a synchronous wakeup? | |
2359 | * | |
2360 | * Put it on the run-queue if it's not already there. The "current" | |
2361 | * thread is always on the run-queue (except when the actual | |
2362 | * re-schedule is in progress), and as such you're allowed to do | |
2363 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2364 | * runnable without the overhead of this. | |
2365 | * | |
2366 | * returns failure only if the task is already active. | |
2367 | */ | |
7d478721 PZ |
2368 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2369 | int wake_flags) | |
1da177e4 | 2370 | { |
cc367732 | 2371 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2372 | unsigned long flags; |
ab3b3aa5 | 2373 | struct rq *rq; |
1da177e4 | 2374 | |
b85d0667 | 2375 | if (!sched_feat(SYNC_WAKEUPS)) |
7d478721 | 2376 | wake_flags &= ~WF_SYNC; |
2398f2c6 | 2377 | |
e9c84311 | 2378 | this_cpu = get_cpu(); |
2398f2c6 | 2379 | |
04e2f174 | 2380 | smp_wmb(); |
ab3b3aa5 | 2381 | rq = task_rq_lock(p, &flags); |
03e89e45 | 2382 | update_rq_clock(rq); |
e9c84311 | 2383 | if (!(p->state & state)) |
1da177e4 LT |
2384 | goto out; |
2385 | ||
dd41f596 | 2386 | if (p->se.on_rq) |
1da177e4 LT |
2387 | goto out_running; |
2388 | ||
2389 | cpu = task_cpu(p); | |
cc367732 | 2390 | orig_cpu = cpu; |
1da177e4 LT |
2391 | |
2392 | #ifdef CONFIG_SMP | |
2393 | if (unlikely(task_running(rq, p))) | |
2394 | goto out_activate; | |
2395 | ||
e9c84311 PZ |
2396 | /* |
2397 | * In order to handle concurrent wakeups and release the rq->lock | |
2398 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2399 | * |
2400 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2401 | */ |
eb24073b IM |
2402 | if (task_contributes_to_load(p)) |
2403 | rq->nr_uninterruptible--; | |
e9c84311 | 2404 | p->state = TASK_WAKING; |
efbbd05a PZ |
2405 | |
2406 | if (p->sched_class->task_waking) | |
2407 | p->sched_class->task_waking(rq, p); | |
2408 | ||
ab19cb23 | 2409 | __task_rq_unlock(rq); |
e9c84311 | 2410 | |
970b13ba | 2411 | cpu = select_task_rq(p, SD_BALANCE_WAKE, wake_flags); |
0970d299 PZ |
2412 | if (cpu != orig_cpu) { |
2413 | /* | |
2414 | * Since we migrate the task without holding any rq->lock, | |
2415 | * we need to be careful with task_rq_lock(), since that | |
2416 | * might end up locking an invalid rq. | |
2417 | */ | |
5d2f5a61 | 2418 | set_task_cpu(p, cpu); |
0970d299 | 2419 | } |
ab19cb23 | 2420 | |
0970d299 PZ |
2421 | rq = cpu_rq(cpu); |
2422 | raw_spin_lock(&rq->lock); | |
ab19cb23 | 2423 | update_rq_clock(rq); |
f5dc3753 | 2424 | |
0970d299 PZ |
2425 | /* |
2426 | * We migrated the task without holding either rq->lock, however | |
2427 | * since the task is not on the task list itself, nobody else | |
2428 | * will try and migrate the task, hence the rq should match the | |
2429 | * cpu we just moved it to. | |
2430 | */ | |
2431 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2432 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2433 | |
e7693a36 GH |
2434 | #ifdef CONFIG_SCHEDSTATS |
2435 | schedstat_inc(rq, ttwu_count); | |
2436 | if (cpu == this_cpu) | |
2437 | schedstat_inc(rq, ttwu_local); | |
2438 | else { | |
2439 | struct sched_domain *sd; | |
2440 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2441 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2442 | schedstat_inc(sd, ttwu_wake_remote); |
2443 | break; | |
2444 | } | |
2445 | } | |
2446 | } | |
6d6bc0ad | 2447 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2448 | |
1da177e4 LT |
2449 | out_activate: |
2450 | #endif /* CONFIG_SMP */ | |
cc367732 | 2451 | schedstat_inc(p, se.nr_wakeups); |
7d478721 | 2452 | if (wake_flags & WF_SYNC) |
cc367732 IM |
2453 | schedstat_inc(p, se.nr_wakeups_sync); |
2454 | if (orig_cpu != cpu) | |
2455 | schedstat_inc(p, se.nr_wakeups_migrate); | |
2456 | if (cpu == this_cpu) | |
2457 | schedstat_inc(p, se.nr_wakeups_local); | |
2458 | else | |
2459 | schedstat_inc(p, se.nr_wakeups_remote); | |
dd41f596 | 2460 | activate_task(rq, p, 1); |
1da177e4 LT |
2461 | success = 1; |
2462 | ||
831451ac PZ |
2463 | /* |
2464 | * Only attribute actual wakeups done by this task. | |
2465 | */ | |
2466 | if (!in_interrupt()) { | |
2467 | struct sched_entity *se = ¤t->se; | |
2468 | u64 sample = se->sum_exec_runtime; | |
2469 | ||
2470 | if (se->last_wakeup) | |
2471 | sample -= se->last_wakeup; | |
2472 | else | |
2473 | sample -= se->start_runtime; | |
2474 | update_avg(&se->avg_wakeup, sample); | |
2475 | ||
2476 | se->last_wakeup = se->sum_exec_runtime; | |
2477 | } | |
2478 | ||
1da177e4 | 2479 | out_running: |
468a15bb | 2480 | trace_sched_wakeup(rq, p, success); |
7d478721 | 2481 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2482 | |
1da177e4 | 2483 | p->state = TASK_RUNNING; |
9a897c5a | 2484 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2485 | if (p->sched_class->task_woken) |
2486 | p->sched_class->task_woken(rq, p); | |
eae0c9df MG |
2487 | |
2488 | if (unlikely(rq->idle_stamp)) { | |
2489 | u64 delta = rq->clock - rq->idle_stamp; | |
2490 | u64 max = 2*sysctl_sched_migration_cost; | |
2491 | ||
2492 | if (delta > max) | |
2493 | rq->avg_idle = max; | |
2494 | else | |
2495 | update_avg(&rq->avg_idle, delta); | |
2496 | rq->idle_stamp = 0; | |
2497 | } | |
9a897c5a | 2498 | #endif |
1da177e4 LT |
2499 | out: |
2500 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2501 | put_cpu(); |
1da177e4 LT |
2502 | |
2503 | return success; | |
2504 | } | |
2505 | ||
50fa610a DH |
2506 | /** |
2507 | * wake_up_process - Wake up a specific process | |
2508 | * @p: The process to be woken up. | |
2509 | * | |
2510 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2511 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2512 | * running. | |
2513 | * | |
2514 | * It may be assumed that this function implies a write memory barrier before | |
2515 | * changing the task state if and only if any tasks are woken up. | |
2516 | */ | |
7ad5b3a5 | 2517 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2518 | { |
d9514f6c | 2519 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2520 | } |
1da177e4 LT |
2521 | EXPORT_SYMBOL(wake_up_process); |
2522 | ||
7ad5b3a5 | 2523 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2524 | { |
2525 | return try_to_wake_up(p, state, 0); | |
2526 | } | |
2527 | ||
1da177e4 LT |
2528 | /* |
2529 | * Perform scheduler related setup for a newly forked process p. | |
2530 | * p is forked by current. | |
dd41f596 IM |
2531 | * |
2532 | * __sched_fork() is basic setup used by init_idle() too: | |
2533 | */ | |
2534 | static void __sched_fork(struct task_struct *p) | |
2535 | { | |
dd41f596 IM |
2536 | p->se.exec_start = 0; |
2537 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2538 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2539 | p->se.nr_migrations = 0; |
4ae7d5ce IM |
2540 | p->se.last_wakeup = 0; |
2541 | p->se.avg_overlap = 0; | |
831451ac PZ |
2542 | p->se.start_runtime = 0; |
2543 | p->se.avg_wakeup = sysctl_sched_wakeup_granularity; | |
6cfb0d5d IM |
2544 | |
2545 | #ifdef CONFIG_SCHEDSTATS | |
7793527b LDM |
2546 | p->se.wait_start = 0; |
2547 | p->se.wait_max = 0; | |
2548 | p->se.wait_count = 0; | |
2549 | p->se.wait_sum = 0; | |
2550 | ||
2551 | p->se.sleep_start = 0; | |
2552 | p->se.sleep_max = 0; | |
2553 | p->se.sum_sleep_runtime = 0; | |
2554 | ||
2555 | p->se.block_start = 0; | |
2556 | p->se.block_max = 0; | |
2557 | p->se.exec_max = 0; | |
2558 | p->se.slice_max = 0; | |
2559 | ||
2560 | p->se.nr_migrations_cold = 0; | |
2561 | p->se.nr_failed_migrations_affine = 0; | |
2562 | p->se.nr_failed_migrations_running = 0; | |
2563 | p->se.nr_failed_migrations_hot = 0; | |
2564 | p->se.nr_forced_migrations = 0; | |
7793527b LDM |
2565 | |
2566 | p->se.nr_wakeups = 0; | |
2567 | p->se.nr_wakeups_sync = 0; | |
2568 | p->se.nr_wakeups_migrate = 0; | |
2569 | p->se.nr_wakeups_local = 0; | |
2570 | p->se.nr_wakeups_remote = 0; | |
2571 | p->se.nr_wakeups_affine = 0; | |
2572 | p->se.nr_wakeups_affine_attempts = 0; | |
2573 | p->se.nr_wakeups_passive = 0; | |
2574 | p->se.nr_wakeups_idle = 0; | |
2575 | ||
6cfb0d5d | 2576 | #endif |
476d139c | 2577 | |
fa717060 | 2578 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2579 | p->se.on_rq = 0; |
4a55bd5e | 2580 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2581 | |
e107be36 AK |
2582 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2583 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2584 | #endif | |
dd41f596 IM |
2585 | } |
2586 | ||
2587 | /* | |
2588 | * fork()/clone()-time setup: | |
2589 | */ | |
2590 | void sched_fork(struct task_struct *p, int clone_flags) | |
2591 | { | |
2592 | int cpu = get_cpu(); | |
2593 | ||
2594 | __sched_fork(p); | |
06b83b5f PZ |
2595 | /* |
2596 | * We mark the process as waking here. This guarantees that | |
2597 | * nobody will actually run it, and a signal or other external | |
2598 | * event cannot wake it up and insert it on the runqueue either. | |
2599 | */ | |
2600 | p->state = TASK_WAKING; | |
dd41f596 | 2601 | |
b9dc29e7 MG |
2602 | /* |
2603 | * Revert to default priority/policy on fork if requested. | |
2604 | */ | |
2605 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2606 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2607 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2608 | p->normal_prio = p->static_prio; |
2609 | } | |
b9dc29e7 | 2610 | |
6c697bdf MG |
2611 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2612 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2613 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2614 | set_load_weight(p); |
2615 | } | |
2616 | ||
b9dc29e7 MG |
2617 | /* |
2618 | * We don't need the reset flag anymore after the fork. It has | |
2619 | * fulfilled its duty: | |
2620 | */ | |
2621 | p->sched_reset_on_fork = 0; | |
2622 | } | |
ca94c442 | 2623 | |
f83f9ac2 PW |
2624 | /* |
2625 | * Make sure we do not leak PI boosting priority to the child. | |
2626 | */ | |
2627 | p->prio = current->normal_prio; | |
2628 | ||
2ddbf952 HS |
2629 | if (!rt_prio(p->prio)) |
2630 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2631 | |
cd29fe6f PZ |
2632 | if (p->sched_class->task_fork) |
2633 | p->sched_class->task_fork(p); | |
2634 | ||
5f3edc1b PZ |
2635 | set_task_cpu(p, cpu); |
2636 | ||
52f17b6c | 2637 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2638 | if (likely(sched_info_on())) |
52f17b6c | 2639 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2640 | #endif |
d6077cb8 | 2641 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2642 | p->oncpu = 0; |
2643 | #endif | |
1da177e4 | 2644 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2645 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2646 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2647 | #endif |
917b627d GH |
2648 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2649 | ||
476d139c | 2650 | put_cpu(); |
1da177e4 LT |
2651 | } |
2652 | ||
2653 | /* | |
2654 | * wake_up_new_task - wake up a newly created task for the first time. | |
2655 | * | |
2656 | * This function will do some initial scheduler statistics housekeeping | |
2657 | * that must be done for every newly created context, then puts the task | |
2658 | * on the runqueue and wakes it. | |
2659 | */ | |
7ad5b3a5 | 2660 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2661 | { |
2662 | unsigned long flags; | |
dd41f596 | 2663 | struct rq *rq; |
c890692b | 2664 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2665 | |
2666 | #ifdef CONFIG_SMP | |
2667 | /* | |
2668 | * Fork balancing, do it here and not earlier because: | |
2669 | * - cpus_allowed can change in the fork path | |
2670 | * - any previously selected cpu might disappear through hotplug | |
2671 | * | |
2672 | * We still have TASK_WAKING but PF_STARTING is gone now, meaning | |
2673 | * ->cpus_allowed is stable, we have preemption disabled, meaning | |
2674 | * cpu_online_mask is stable. | |
2675 | */ | |
2676 | cpu = select_task_rq(p, SD_BALANCE_FORK, 0); | |
2677 | set_task_cpu(p, cpu); | |
2678 | #endif | |
1da177e4 | 2679 | |
0970d299 PZ |
2680 | /* |
2681 | * Since the task is not on the rq and we still have TASK_WAKING set | |
2682 | * nobody else will migrate this task. | |
2683 | */ | |
2684 | rq = cpu_rq(cpu); | |
2685 | raw_spin_lock_irqsave(&rq->lock, flags); | |
2686 | ||
06b83b5f PZ |
2687 | BUG_ON(p->state != TASK_WAKING); |
2688 | p->state = TASK_RUNNING; | |
a8e504d2 | 2689 | update_rq_clock(rq); |
cd29fe6f | 2690 | activate_task(rq, p, 0); |
c71dd42d | 2691 | trace_sched_wakeup_new(rq, p, 1); |
a7558e01 | 2692 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2693 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2694 | if (p->sched_class->task_woken) |
2695 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2696 | #endif |
dd41f596 | 2697 | task_rq_unlock(rq, &flags); |
fabf318e | 2698 | put_cpu(); |
1da177e4 LT |
2699 | } |
2700 | ||
e107be36 AK |
2701 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2702 | ||
2703 | /** | |
80dd99b3 | 2704 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2705 | * @notifier: notifier struct to register |
e107be36 AK |
2706 | */ |
2707 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2708 | { | |
2709 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2710 | } | |
2711 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2712 | ||
2713 | /** | |
2714 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2715 | * @notifier: notifier struct to unregister |
e107be36 AK |
2716 | * |
2717 | * This is safe to call from within a preemption notifier. | |
2718 | */ | |
2719 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2720 | { | |
2721 | hlist_del(¬ifier->link); | |
2722 | } | |
2723 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2724 | ||
2725 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2726 | { | |
2727 | struct preempt_notifier *notifier; | |
2728 | struct hlist_node *node; | |
2729 | ||
2730 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2731 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2732 | } | |
2733 | ||
2734 | static void | |
2735 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2736 | struct task_struct *next) | |
2737 | { | |
2738 | struct preempt_notifier *notifier; | |
2739 | struct hlist_node *node; | |
2740 | ||
2741 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2742 | notifier->ops->sched_out(notifier, next); | |
2743 | } | |
2744 | ||
6d6bc0ad | 2745 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2746 | |
2747 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2748 | { | |
2749 | } | |
2750 | ||
2751 | static void | |
2752 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2753 | struct task_struct *next) | |
2754 | { | |
2755 | } | |
2756 | ||
6d6bc0ad | 2757 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2758 | |
4866cde0 NP |
2759 | /** |
2760 | * prepare_task_switch - prepare to switch tasks | |
2761 | * @rq: the runqueue preparing to switch | |
421cee29 | 2762 | * @prev: the current task that is being switched out |
4866cde0 NP |
2763 | * @next: the task we are going to switch to. |
2764 | * | |
2765 | * This is called with the rq lock held and interrupts off. It must | |
2766 | * be paired with a subsequent finish_task_switch after the context | |
2767 | * switch. | |
2768 | * | |
2769 | * prepare_task_switch sets up locking and calls architecture specific | |
2770 | * hooks. | |
2771 | */ | |
e107be36 AK |
2772 | static inline void |
2773 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2774 | struct task_struct *next) | |
4866cde0 | 2775 | { |
e107be36 | 2776 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2777 | prepare_lock_switch(rq, next); |
2778 | prepare_arch_switch(next); | |
2779 | } | |
2780 | ||
1da177e4 LT |
2781 | /** |
2782 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2783 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2784 | * @prev: the thread we just switched away from. |
2785 | * | |
4866cde0 NP |
2786 | * finish_task_switch must be called after the context switch, paired |
2787 | * with a prepare_task_switch call before the context switch. | |
2788 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2789 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2790 | * |
2791 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2792 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2793 | * with the lock held can cause deadlocks; see schedule() for |
2794 | * details.) | |
2795 | */ | |
a9957449 | 2796 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2797 | __releases(rq->lock) |
2798 | { | |
1da177e4 | 2799 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2800 | long prev_state; |
1da177e4 LT |
2801 | |
2802 | rq->prev_mm = NULL; | |
2803 | ||
2804 | /* | |
2805 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2806 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2807 | * schedule one last time. The schedule call will never return, and |
2808 | * the scheduled task must drop that reference. | |
c394cc9f | 2809 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2810 | * still held, otherwise prev could be scheduled on another cpu, die |
2811 | * there before we look at prev->state, and then the reference would | |
2812 | * be dropped twice. | |
2813 | * Manfred Spraul <manfred@colorfullife.com> | |
2814 | */ | |
55a101f8 | 2815 | prev_state = prev->state; |
4866cde0 | 2816 | finish_arch_switch(prev); |
8381f65d JI |
2817 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2818 | local_irq_disable(); | |
2819 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2820 | perf_event_task_sched_in(current); |
8381f65d JI |
2821 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2822 | local_irq_enable(); | |
2823 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2824 | finish_lock_switch(rq, prev); |
e8fa1362 | 2825 | |
e107be36 | 2826 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2827 | if (mm) |
2828 | mmdrop(mm); | |
c394cc9f | 2829 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2830 | /* |
2831 | * Remove function-return probe instances associated with this | |
2832 | * task and put them back on the free list. | |
9761eea8 | 2833 | */ |
c6fd91f0 | 2834 | kprobe_flush_task(prev); |
1da177e4 | 2835 | put_task_struct(prev); |
c6fd91f0 | 2836 | } |
1da177e4 LT |
2837 | } |
2838 | ||
3f029d3c GH |
2839 | #ifdef CONFIG_SMP |
2840 | ||
2841 | /* assumes rq->lock is held */ | |
2842 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2843 | { | |
2844 | if (prev->sched_class->pre_schedule) | |
2845 | prev->sched_class->pre_schedule(rq, prev); | |
2846 | } | |
2847 | ||
2848 | /* rq->lock is NOT held, but preemption is disabled */ | |
2849 | static inline void post_schedule(struct rq *rq) | |
2850 | { | |
2851 | if (rq->post_schedule) { | |
2852 | unsigned long flags; | |
2853 | ||
05fa785c | 2854 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2855 | if (rq->curr->sched_class->post_schedule) |
2856 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2857 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2858 | |
2859 | rq->post_schedule = 0; | |
2860 | } | |
2861 | } | |
2862 | ||
2863 | #else | |
da19ab51 | 2864 | |
3f029d3c GH |
2865 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2866 | { | |
2867 | } | |
2868 | ||
2869 | static inline void post_schedule(struct rq *rq) | |
2870 | { | |
1da177e4 LT |
2871 | } |
2872 | ||
3f029d3c GH |
2873 | #endif |
2874 | ||
1da177e4 LT |
2875 | /** |
2876 | * schedule_tail - first thing a freshly forked thread must call. | |
2877 | * @prev: the thread we just switched away from. | |
2878 | */ | |
36c8b586 | 2879 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2880 | __releases(rq->lock) |
2881 | { | |
70b97a7f IM |
2882 | struct rq *rq = this_rq(); |
2883 | ||
4866cde0 | 2884 | finish_task_switch(rq, prev); |
da19ab51 | 2885 | |
3f029d3c GH |
2886 | /* |
2887 | * FIXME: do we need to worry about rq being invalidated by the | |
2888 | * task_switch? | |
2889 | */ | |
2890 | post_schedule(rq); | |
70b97a7f | 2891 | |
4866cde0 NP |
2892 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2893 | /* In this case, finish_task_switch does not reenable preemption */ | |
2894 | preempt_enable(); | |
2895 | #endif | |
1da177e4 | 2896 | if (current->set_child_tid) |
b488893a | 2897 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2898 | } |
2899 | ||
2900 | /* | |
2901 | * context_switch - switch to the new MM and the new | |
2902 | * thread's register state. | |
2903 | */ | |
dd41f596 | 2904 | static inline void |
70b97a7f | 2905 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2906 | struct task_struct *next) |
1da177e4 | 2907 | { |
dd41f596 | 2908 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2909 | |
e107be36 | 2910 | prepare_task_switch(rq, prev, next); |
0a16b607 | 2911 | trace_sched_switch(rq, prev, next); |
dd41f596 IM |
2912 | mm = next->mm; |
2913 | oldmm = prev->active_mm; | |
9226d125 ZA |
2914 | /* |
2915 | * For paravirt, this is coupled with an exit in switch_to to | |
2916 | * combine the page table reload and the switch backend into | |
2917 | * one hypercall. | |
2918 | */ | |
224101ed | 2919 | arch_start_context_switch(prev); |
9226d125 | 2920 | |
710390d9 | 2921 | if (likely(!mm)) { |
1da177e4 LT |
2922 | next->active_mm = oldmm; |
2923 | atomic_inc(&oldmm->mm_count); | |
2924 | enter_lazy_tlb(oldmm, next); | |
2925 | } else | |
2926 | switch_mm(oldmm, mm, next); | |
2927 | ||
710390d9 | 2928 | if (likely(!prev->mm)) { |
1da177e4 | 2929 | prev->active_mm = NULL; |
1da177e4 LT |
2930 | rq->prev_mm = oldmm; |
2931 | } | |
3a5f5e48 IM |
2932 | /* |
2933 | * Since the runqueue lock will be released by the next | |
2934 | * task (which is an invalid locking op but in the case | |
2935 | * of the scheduler it's an obvious special-case), so we | |
2936 | * do an early lockdep release here: | |
2937 | */ | |
2938 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2939 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2940 | #endif |
1da177e4 LT |
2941 | |
2942 | /* Here we just switch the register state and the stack. */ | |
2943 | switch_to(prev, next, prev); | |
2944 | ||
dd41f596 IM |
2945 | barrier(); |
2946 | /* | |
2947 | * this_rq must be evaluated again because prev may have moved | |
2948 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2949 | * frame will be invalid. | |
2950 | */ | |
2951 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2952 | } |
2953 | ||
2954 | /* | |
2955 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2956 | * | |
2957 | * externally visible scheduler statistics: current number of runnable | |
2958 | * threads, current number of uninterruptible-sleeping threads, total | |
2959 | * number of context switches performed since bootup. | |
2960 | */ | |
2961 | unsigned long nr_running(void) | |
2962 | { | |
2963 | unsigned long i, sum = 0; | |
2964 | ||
2965 | for_each_online_cpu(i) | |
2966 | sum += cpu_rq(i)->nr_running; | |
2967 | ||
2968 | return sum; | |
f711f609 | 2969 | } |
1da177e4 LT |
2970 | |
2971 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2972 | { |
1da177e4 | 2973 | unsigned long i, sum = 0; |
f711f609 | 2974 | |
0a945022 | 2975 | for_each_possible_cpu(i) |
1da177e4 | 2976 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2977 | |
2978 | /* | |
1da177e4 LT |
2979 | * Since we read the counters lockless, it might be slightly |
2980 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2981 | */ |
1da177e4 LT |
2982 | if (unlikely((long)sum < 0)) |
2983 | sum = 0; | |
f711f609 | 2984 | |
1da177e4 | 2985 | return sum; |
f711f609 | 2986 | } |
f711f609 | 2987 | |
1da177e4 | 2988 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2989 | { |
cc94abfc SR |
2990 | int i; |
2991 | unsigned long long sum = 0; | |
46cb4b7c | 2992 | |
0a945022 | 2993 | for_each_possible_cpu(i) |
1da177e4 | 2994 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2995 | |
1da177e4 LT |
2996 | return sum; |
2997 | } | |
483b4ee6 | 2998 | |
1da177e4 LT |
2999 | unsigned long nr_iowait(void) |
3000 | { | |
3001 | unsigned long i, sum = 0; | |
483b4ee6 | 3002 | |
0a945022 | 3003 | for_each_possible_cpu(i) |
1da177e4 | 3004 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3005 | |
1da177e4 LT |
3006 | return sum; |
3007 | } | |
483b4ee6 | 3008 | |
69d25870 AV |
3009 | unsigned long nr_iowait_cpu(void) |
3010 | { | |
3011 | struct rq *this = this_rq(); | |
3012 | return atomic_read(&this->nr_iowait); | |
3013 | } | |
46cb4b7c | 3014 | |
69d25870 AV |
3015 | unsigned long this_cpu_load(void) |
3016 | { | |
3017 | struct rq *this = this_rq(); | |
3018 | return this->cpu_load[0]; | |
3019 | } | |
e790fb0b | 3020 | |
46cb4b7c | 3021 | |
dce48a84 TG |
3022 | /* Variables and functions for calc_load */ |
3023 | static atomic_long_t calc_load_tasks; | |
3024 | static unsigned long calc_load_update; | |
3025 | unsigned long avenrun[3]; | |
3026 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3027 | |
2d02494f TG |
3028 | /** |
3029 | * get_avenrun - get the load average array | |
3030 | * @loads: pointer to dest load array | |
3031 | * @offset: offset to add | |
3032 | * @shift: shift count to shift the result left | |
3033 | * | |
3034 | * These values are estimates at best, so no need for locking. | |
3035 | */ | |
3036 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3037 | { | |
3038 | loads[0] = (avenrun[0] + offset) << shift; | |
3039 | loads[1] = (avenrun[1] + offset) << shift; | |
3040 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3041 | } |
46cb4b7c | 3042 | |
dce48a84 TG |
3043 | static unsigned long |
3044 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3045 | { |
dce48a84 TG |
3046 | load *= exp; |
3047 | load += active * (FIXED_1 - exp); | |
3048 | return load >> FSHIFT; | |
3049 | } | |
46cb4b7c SS |
3050 | |
3051 | /* | |
dce48a84 TG |
3052 | * calc_load - update the avenrun load estimates 10 ticks after the |
3053 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3054 | */ |
dce48a84 | 3055 | void calc_global_load(void) |
7835b98b | 3056 | { |
dce48a84 TG |
3057 | unsigned long upd = calc_load_update + 10; |
3058 | long active; | |
1da177e4 | 3059 | |
dce48a84 TG |
3060 | if (time_before(jiffies, upd)) |
3061 | return; | |
1da177e4 | 3062 | |
dce48a84 TG |
3063 | active = atomic_long_read(&calc_load_tasks); |
3064 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3065 | |
dce48a84 TG |
3066 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3067 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3068 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3069 | |
dce48a84 TG |
3070 | calc_load_update += LOAD_FREQ; |
3071 | } | |
1da177e4 | 3072 | |
dce48a84 TG |
3073 | /* |
3074 | * Either called from update_cpu_load() or from a cpu going idle | |
3075 | */ | |
3076 | static void calc_load_account_active(struct rq *this_rq) | |
3077 | { | |
3078 | long nr_active, delta; | |
08c183f3 | 3079 | |
dce48a84 TG |
3080 | nr_active = this_rq->nr_running; |
3081 | nr_active += (long) this_rq->nr_uninterruptible; | |
783609c6 | 3082 | |
dce48a84 TG |
3083 | if (nr_active != this_rq->calc_load_active) { |
3084 | delta = nr_active - this_rq->calc_load_active; | |
3085 | this_rq->calc_load_active = nr_active; | |
3086 | atomic_long_add(delta, &calc_load_tasks); | |
1da177e4 | 3087 | } |
46cb4b7c SS |
3088 | } |
3089 | ||
3090 | /* | |
dd41f596 IM |
3091 | * Update rq->cpu_load[] statistics. This function is usually called every |
3092 | * scheduler tick (TICK_NSEC). | |
46cb4b7c | 3093 | */ |
dd41f596 | 3094 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3095 | { |
495eca49 | 3096 | unsigned long this_load = this_rq->load.weight; |
dd41f596 | 3097 | int i, scale; |
46cb4b7c | 3098 | |
dd41f596 | 3099 | this_rq->nr_load_updates++; |
46cb4b7c | 3100 | |
dd41f596 IM |
3101 | /* Update our load: */ |
3102 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3103 | unsigned long old_load, new_load; | |
7d1e6a9b | 3104 | |
dd41f596 | 3105 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3106 | |
dd41f596 IM |
3107 | old_load = this_rq->cpu_load[i]; |
3108 | new_load = this_load; | |
a25707f3 IM |
3109 | /* |
3110 | * Round up the averaging division if load is increasing. This | |
3111 | * prevents us from getting stuck on 9 if the load is 10, for | |
3112 | * example. | |
3113 | */ | |
3114 | if (new_load > old_load) | |
3115 | new_load += scale-1; | |
dd41f596 IM |
3116 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3117 | } | |
46cb4b7c | 3118 | |
dce48a84 TG |
3119 | if (time_after_eq(jiffies, this_rq->calc_load_update)) { |
3120 | this_rq->calc_load_update += LOAD_FREQ; | |
3121 | calc_load_account_active(this_rq); | |
46cb4b7c | 3122 | } |
46cb4b7c SS |
3123 | } |
3124 | ||
dd41f596 | 3125 | #ifdef CONFIG_SMP |
8a0be9ef | 3126 | |
46cb4b7c | 3127 | /* |
38022906 PZ |
3128 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3129 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3130 | */ |
38022906 | 3131 | void sched_exec(void) |
46cb4b7c | 3132 | { |
38022906 | 3133 | struct task_struct *p = current; |
70b97a7f | 3134 | struct migration_req req; |
38022906 | 3135 | int dest_cpu, this_cpu; |
1da177e4 | 3136 | unsigned long flags; |
70b97a7f | 3137 | struct rq *rq; |
46cb4b7c | 3138 | |
38022906 PZ |
3139 | again: |
3140 | this_cpu = get_cpu(); | |
3141 | dest_cpu = select_task_rq(p, SD_BALANCE_EXEC, 0); | |
3142 | if (dest_cpu == this_cpu) { | |
3143 | put_cpu(); | |
3144 | return; | |
46cb4b7c SS |
3145 | } |
3146 | ||
1da177e4 | 3147 | rq = task_rq_lock(p, &flags); |
38022906 PZ |
3148 | put_cpu(); |
3149 | ||
46cb4b7c | 3150 | /* |
38022906 | 3151 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3152 | */ |
96f874e2 | 3153 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed) |
38022906 PZ |
3154 | || unlikely(!cpu_active(dest_cpu))) { |
3155 | task_rq_unlock(rq, &flags); | |
3156 | goto again; | |
46cb4b7c SS |
3157 | } |
3158 | ||
1da177e4 LT |
3159 | /* force the process onto the specified CPU */ |
3160 | if (migrate_task(p, dest_cpu, &req)) { | |
3161 | /* Need to wait for migration thread (might exit: take ref). */ | |
3162 | struct task_struct *mt = rq->migration_thread; | |
dd41f596 | 3163 | |
1da177e4 LT |
3164 | get_task_struct(mt); |
3165 | task_rq_unlock(rq, &flags); | |
3166 | wake_up_process(mt); | |
3167 | put_task_struct(mt); | |
3168 | wait_for_completion(&req.done); | |
dd41f596 | 3169 | |
1da177e4 LT |
3170 | return; |
3171 | } | |
1da177e4 | 3172 | task_rq_unlock(rq, &flags); |
1da177e4 | 3173 | } |
dd41f596 | 3174 | |
1da177e4 LT |
3175 | #endif |
3176 | ||
1da177e4 LT |
3177 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3178 | ||
3179 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3180 | ||
3181 | /* | |
c5f8d995 | 3182 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3183 | * @p in case that task is currently running. |
c5f8d995 HS |
3184 | * |
3185 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3186 | */ |
c5f8d995 HS |
3187 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3188 | { | |
3189 | u64 ns = 0; | |
3190 | ||
3191 | if (task_current(rq, p)) { | |
3192 | update_rq_clock(rq); | |
3193 | ns = rq->clock - p->se.exec_start; | |
3194 | if ((s64)ns < 0) | |
3195 | ns = 0; | |
3196 | } | |
3197 | ||
3198 | return ns; | |
3199 | } | |
3200 | ||
bb34d92f | 3201 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3202 | { |
1da177e4 | 3203 | unsigned long flags; |
41b86e9c | 3204 | struct rq *rq; |
bb34d92f | 3205 | u64 ns = 0; |
48f24c4d | 3206 | |
41b86e9c | 3207 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3208 | ns = do_task_delta_exec(p, rq); |
3209 | task_rq_unlock(rq, &flags); | |
1508487e | 3210 | |
c5f8d995 HS |
3211 | return ns; |
3212 | } | |
f06febc9 | 3213 | |
c5f8d995 HS |
3214 | /* |
3215 | * Return accounted runtime for the task. | |
3216 | * In case the task is currently running, return the runtime plus current's | |
3217 | * pending runtime that have not been accounted yet. | |
3218 | */ | |
3219 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3220 | { | |
3221 | unsigned long flags; | |
3222 | struct rq *rq; | |
3223 | u64 ns = 0; | |
3224 | ||
3225 | rq = task_rq_lock(p, &flags); | |
3226 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3227 | task_rq_unlock(rq, &flags); | |
3228 | ||
3229 | return ns; | |
3230 | } | |
48f24c4d | 3231 | |
c5f8d995 HS |
3232 | /* |
3233 | * Return sum_exec_runtime for the thread group. | |
3234 | * In case the task is currently running, return the sum plus current's | |
3235 | * pending runtime that have not been accounted yet. | |
3236 | * | |
3237 | * Note that the thread group might have other running tasks as well, | |
3238 | * so the return value not includes other pending runtime that other | |
3239 | * running tasks might have. | |
3240 | */ | |
3241 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3242 | { | |
3243 | struct task_cputime totals; | |
3244 | unsigned long flags; | |
3245 | struct rq *rq; | |
3246 | u64 ns; | |
3247 | ||
3248 | rq = task_rq_lock(p, &flags); | |
3249 | thread_group_cputime(p, &totals); | |
3250 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3251 | task_rq_unlock(rq, &flags); |
48f24c4d | 3252 | |
1da177e4 LT |
3253 | return ns; |
3254 | } | |
3255 | ||
1da177e4 LT |
3256 | /* |
3257 | * Account user cpu time to a process. | |
3258 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3259 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3260 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3261 | */ |
457533a7 MS |
3262 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3263 | cputime_t cputime_scaled) | |
1da177e4 LT |
3264 | { |
3265 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3266 | cputime64_t tmp; | |
3267 | ||
457533a7 | 3268 | /* Add user time to process. */ |
1da177e4 | 3269 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3270 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3271 | account_group_user_time(p, cputime); |
1da177e4 LT |
3272 | |
3273 | /* Add user time to cpustat. */ | |
3274 | tmp = cputime_to_cputime64(cputime); | |
3275 | if (TASK_NICE(p) > 0) | |
3276 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3277 | else | |
3278 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3279 | |
3280 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3281 | /* Account for user time used */ |
3282 | acct_update_integrals(p); | |
1da177e4 LT |
3283 | } |
3284 | ||
94886b84 LV |
3285 | /* |
3286 | * Account guest cpu time to a process. | |
3287 | * @p: the process that the cpu time gets accounted to | |
3288 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3289 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3290 | */ |
457533a7 MS |
3291 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3292 | cputime_t cputime_scaled) | |
94886b84 LV |
3293 | { |
3294 | cputime64_t tmp; | |
3295 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3296 | ||
3297 | tmp = cputime_to_cputime64(cputime); | |
3298 | ||
457533a7 | 3299 | /* Add guest time to process. */ |
94886b84 | 3300 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3301 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3302 | account_group_user_time(p, cputime); |
94886b84 LV |
3303 | p->gtime = cputime_add(p->gtime, cputime); |
3304 | ||
457533a7 | 3305 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3306 | if (TASK_NICE(p) > 0) { |
3307 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3308 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3309 | } else { | |
3310 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3311 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3312 | } | |
94886b84 LV |
3313 | } |
3314 | ||
1da177e4 LT |
3315 | /* |
3316 | * Account system cpu time to a process. | |
3317 | * @p: the process that the cpu time gets accounted to | |
3318 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3319 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3320 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3321 | */ |
3322 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3323 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3324 | { |
3325 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3326 | cputime64_t tmp; |
3327 | ||
983ed7a6 | 3328 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3329 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3330 | return; |
3331 | } | |
94886b84 | 3332 | |
457533a7 | 3333 | /* Add system time to process. */ |
1da177e4 | 3334 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3335 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3336 | account_group_system_time(p, cputime); |
1da177e4 LT |
3337 | |
3338 | /* Add system time to cpustat. */ | |
3339 | tmp = cputime_to_cputime64(cputime); | |
3340 | if (hardirq_count() - hardirq_offset) | |
3341 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3342 | else if (softirq_count()) | |
3343 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3344 | else |
79741dd3 MS |
3345 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3346 | ||
ef12fefa BR |
3347 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3348 | ||
1da177e4 LT |
3349 | /* Account for system time used */ |
3350 | acct_update_integrals(p); | |
1da177e4 LT |
3351 | } |
3352 | ||
c66f08be | 3353 | /* |
1da177e4 | 3354 | * Account for involuntary wait time. |
1da177e4 | 3355 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3356 | */ |
79741dd3 | 3357 | void account_steal_time(cputime_t cputime) |
c66f08be | 3358 | { |
79741dd3 MS |
3359 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3360 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3361 | ||
3362 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3363 | } |
3364 | ||
1da177e4 | 3365 | /* |
79741dd3 MS |
3366 | * Account for idle time. |
3367 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3368 | */ |
79741dd3 | 3369 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3370 | { |
3371 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3372 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3373 | struct rq *rq = this_rq(); |
1da177e4 | 3374 | |
79741dd3 MS |
3375 | if (atomic_read(&rq->nr_iowait) > 0) |
3376 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3377 | else | |
3378 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3379 | } |
3380 | ||
79741dd3 MS |
3381 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3382 | ||
3383 | /* | |
3384 | * Account a single tick of cpu time. | |
3385 | * @p: the process that the cpu time gets accounted to | |
3386 | * @user_tick: indicates if the tick is a user or a system tick | |
3387 | */ | |
3388 | void account_process_tick(struct task_struct *p, int user_tick) | |
3389 | { | |
a42548a1 | 3390 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3391 | struct rq *rq = this_rq(); |
3392 | ||
3393 | if (user_tick) | |
a42548a1 | 3394 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3395 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3396 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3397 | one_jiffy_scaled); |
3398 | else | |
a42548a1 | 3399 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3400 | } |
3401 | ||
3402 | /* | |
3403 | * Account multiple ticks of steal time. | |
3404 | * @p: the process from which the cpu time has been stolen | |
3405 | * @ticks: number of stolen ticks | |
3406 | */ | |
3407 | void account_steal_ticks(unsigned long ticks) | |
3408 | { | |
3409 | account_steal_time(jiffies_to_cputime(ticks)); | |
3410 | } | |
3411 | ||
3412 | /* | |
3413 | * Account multiple ticks of idle time. | |
3414 | * @ticks: number of stolen ticks | |
3415 | */ | |
3416 | void account_idle_ticks(unsigned long ticks) | |
3417 | { | |
3418 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3419 | } |
3420 | ||
79741dd3 MS |
3421 | #endif |
3422 | ||
49048622 BS |
3423 | /* |
3424 | * Use precise platform statistics if available: | |
3425 | */ | |
3426 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3427 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3428 | { |
d99ca3b9 HS |
3429 | *ut = p->utime; |
3430 | *st = p->stime; | |
49048622 BS |
3431 | } |
3432 | ||
0cf55e1e | 3433 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3434 | { |
0cf55e1e HS |
3435 | struct task_cputime cputime; |
3436 | ||
3437 | thread_group_cputime(p, &cputime); | |
3438 | ||
3439 | *ut = cputime.utime; | |
3440 | *st = cputime.stime; | |
49048622 BS |
3441 | } |
3442 | #else | |
761b1d26 HS |
3443 | |
3444 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3445 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3446 | #endif |
3447 | ||
d180c5bc | 3448 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3449 | { |
d99ca3b9 | 3450 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3451 | |
3452 | /* | |
3453 | * Use CFS's precise accounting: | |
3454 | */ | |
d180c5bc | 3455 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3456 | |
3457 | if (total) { | |
d180c5bc HS |
3458 | u64 temp; |
3459 | ||
3460 | temp = (u64)(rtime * utime); | |
49048622 | 3461 | do_div(temp, total); |
d180c5bc HS |
3462 | utime = (cputime_t)temp; |
3463 | } else | |
3464 | utime = rtime; | |
49048622 | 3465 | |
d180c5bc HS |
3466 | /* |
3467 | * Compare with previous values, to keep monotonicity: | |
3468 | */ | |
761b1d26 | 3469 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3470 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3471 | |
d99ca3b9 HS |
3472 | *ut = p->prev_utime; |
3473 | *st = p->prev_stime; | |
49048622 BS |
3474 | } |
3475 | ||
0cf55e1e HS |
3476 | /* |
3477 | * Must be called with siglock held. | |
3478 | */ | |
3479 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3480 | { |
0cf55e1e HS |
3481 | struct signal_struct *sig = p->signal; |
3482 | struct task_cputime cputime; | |
3483 | cputime_t rtime, utime, total; | |
49048622 | 3484 | |
0cf55e1e | 3485 | thread_group_cputime(p, &cputime); |
49048622 | 3486 | |
0cf55e1e HS |
3487 | total = cputime_add(cputime.utime, cputime.stime); |
3488 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3489 | |
0cf55e1e HS |
3490 | if (total) { |
3491 | u64 temp; | |
49048622 | 3492 | |
0cf55e1e HS |
3493 | temp = (u64)(rtime * cputime.utime); |
3494 | do_div(temp, total); | |
3495 | utime = (cputime_t)temp; | |
3496 | } else | |
3497 | utime = rtime; | |
3498 | ||
3499 | sig->prev_utime = max(sig->prev_utime, utime); | |
3500 | sig->prev_stime = max(sig->prev_stime, | |
3501 | cputime_sub(rtime, sig->prev_utime)); | |
3502 | ||
3503 | *ut = sig->prev_utime; | |
3504 | *st = sig->prev_stime; | |
49048622 | 3505 | } |
49048622 | 3506 | #endif |
49048622 | 3507 | |
7835b98b CL |
3508 | /* |
3509 | * This function gets called by the timer code, with HZ frequency. | |
3510 | * We call it with interrupts disabled. | |
3511 | * | |
3512 | * It also gets called by the fork code, when changing the parent's | |
3513 | * timeslices. | |
3514 | */ | |
3515 | void scheduler_tick(void) | |
3516 | { | |
7835b98b CL |
3517 | int cpu = smp_processor_id(); |
3518 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3519 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3520 | |
3521 | sched_clock_tick(); | |
dd41f596 | 3522 | |
05fa785c | 3523 | raw_spin_lock(&rq->lock); |
3e51f33f | 3524 | update_rq_clock(rq); |
f1a438d8 | 3525 | update_cpu_load(rq); |
fa85ae24 | 3526 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3527 | raw_spin_unlock(&rq->lock); |
7835b98b | 3528 | |
49f47433 | 3529 | perf_event_task_tick(curr); |
e220d2dc | 3530 | |
e418e1c2 | 3531 | #ifdef CONFIG_SMP |
dd41f596 IM |
3532 | rq->idle_at_tick = idle_cpu(cpu); |
3533 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3534 | #endif |
1da177e4 LT |
3535 | } |
3536 | ||
132380a0 | 3537 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3538 | { |
3539 | if (in_lock_functions(addr)) { | |
3540 | addr = CALLER_ADDR2; | |
3541 | if (in_lock_functions(addr)) | |
3542 | addr = CALLER_ADDR3; | |
3543 | } | |
3544 | return addr; | |
3545 | } | |
1da177e4 | 3546 | |
7e49fcce SR |
3547 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3548 | defined(CONFIG_PREEMPT_TRACER)) | |
3549 | ||
43627582 | 3550 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3551 | { |
6cd8a4bb | 3552 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3553 | /* |
3554 | * Underflow? | |
3555 | */ | |
9a11b49a IM |
3556 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3557 | return; | |
6cd8a4bb | 3558 | #endif |
1da177e4 | 3559 | preempt_count() += val; |
6cd8a4bb | 3560 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3561 | /* |
3562 | * Spinlock count overflowing soon? | |
3563 | */ | |
33859f7f MOS |
3564 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3565 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3566 | #endif |
3567 | if (preempt_count() == val) | |
3568 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3569 | } |
3570 | EXPORT_SYMBOL(add_preempt_count); | |
3571 | ||
43627582 | 3572 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3573 | { |
6cd8a4bb | 3574 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3575 | /* |
3576 | * Underflow? | |
3577 | */ | |
01e3eb82 | 3578 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3579 | return; |
1da177e4 LT |
3580 | /* |
3581 | * Is the spinlock portion underflowing? | |
3582 | */ | |
9a11b49a IM |
3583 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3584 | !(preempt_count() & PREEMPT_MASK))) | |
3585 | return; | |
6cd8a4bb | 3586 | #endif |
9a11b49a | 3587 | |
6cd8a4bb SR |
3588 | if (preempt_count() == val) |
3589 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3590 | preempt_count() -= val; |
3591 | } | |
3592 | EXPORT_SYMBOL(sub_preempt_count); | |
3593 | ||
3594 | #endif | |
3595 | ||
3596 | /* | |
dd41f596 | 3597 | * Print scheduling while atomic bug: |
1da177e4 | 3598 | */ |
dd41f596 | 3599 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3600 | { |
838225b4 SS |
3601 | struct pt_regs *regs = get_irq_regs(); |
3602 | ||
3df0fc5b PZ |
3603 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3604 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3605 | |
dd41f596 | 3606 | debug_show_held_locks(prev); |
e21f5b15 | 3607 | print_modules(); |
dd41f596 IM |
3608 | if (irqs_disabled()) |
3609 | print_irqtrace_events(prev); | |
838225b4 SS |
3610 | |
3611 | if (regs) | |
3612 | show_regs(regs); | |
3613 | else | |
3614 | dump_stack(); | |
dd41f596 | 3615 | } |
1da177e4 | 3616 | |
dd41f596 IM |
3617 | /* |
3618 | * Various schedule()-time debugging checks and statistics: | |
3619 | */ | |
3620 | static inline void schedule_debug(struct task_struct *prev) | |
3621 | { | |
1da177e4 | 3622 | /* |
41a2d6cf | 3623 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3624 | * schedule() atomically, we ignore that path for now. |
3625 | * Otherwise, whine if we are scheduling when we should not be. | |
3626 | */ | |
3f33a7ce | 3627 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3628 | __schedule_bug(prev); |
3629 | ||
1da177e4 LT |
3630 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3631 | ||
2d72376b | 3632 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3633 | #ifdef CONFIG_SCHEDSTATS |
3634 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3635 | schedstat_inc(this_rq(), bkl_count); |
3636 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3637 | } |
3638 | #endif | |
dd41f596 IM |
3639 | } |
3640 | ||
6cecd084 | 3641 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3642 | { |
6cecd084 PZ |
3643 | if (prev->state == TASK_RUNNING) { |
3644 | u64 runtime = prev->se.sum_exec_runtime; | |
df1c99d4 | 3645 | |
6cecd084 PZ |
3646 | runtime -= prev->se.prev_sum_exec_runtime; |
3647 | runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); | |
df1c99d4 MG |
3648 | |
3649 | /* | |
3650 | * In order to avoid avg_overlap growing stale when we are | |
3651 | * indeed overlapping and hence not getting put to sleep, grow | |
3652 | * the avg_overlap on preemption. | |
3653 | * | |
3654 | * We use the average preemption runtime because that | |
3655 | * correlates to the amount of cache footprint a task can | |
3656 | * build up. | |
3657 | */ | |
6cecd084 | 3658 | update_avg(&prev->se.avg_overlap, runtime); |
df1c99d4 | 3659 | } |
6cecd084 | 3660 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3661 | } |
3662 | ||
dd41f596 IM |
3663 | /* |
3664 | * Pick up the highest-prio task: | |
3665 | */ | |
3666 | static inline struct task_struct * | |
b67802ea | 3667 | pick_next_task(struct rq *rq) |
dd41f596 | 3668 | { |
5522d5d5 | 3669 | const struct sched_class *class; |
dd41f596 | 3670 | struct task_struct *p; |
1da177e4 LT |
3671 | |
3672 | /* | |
dd41f596 IM |
3673 | * Optimization: we know that if all tasks are in |
3674 | * the fair class we can call that function directly: | |
1da177e4 | 3675 | */ |
dd41f596 | 3676 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3677 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3678 | if (likely(p)) |
3679 | return p; | |
1da177e4 LT |
3680 | } |
3681 | ||
dd41f596 IM |
3682 | class = sched_class_highest; |
3683 | for ( ; ; ) { | |
fb8d4724 | 3684 | p = class->pick_next_task(rq); |
dd41f596 IM |
3685 | if (p) |
3686 | return p; | |
3687 | /* | |
3688 | * Will never be NULL as the idle class always | |
3689 | * returns a non-NULL p: | |
3690 | */ | |
3691 | class = class->next; | |
3692 | } | |
3693 | } | |
1da177e4 | 3694 | |
dd41f596 IM |
3695 | /* |
3696 | * schedule() is the main scheduler function. | |
3697 | */ | |
ff743345 | 3698 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3699 | { |
3700 | struct task_struct *prev, *next; | |
67ca7bde | 3701 | unsigned long *switch_count; |
dd41f596 | 3702 | struct rq *rq; |
31656519 | 3703 | int cpu; |
dd41f596 | 3704 | |
ff743345 PZ |
3705 | need_resched: |
3706 | preempt_disable(); | |
dd41f596 IM |
3707 | cpu = smp_processor_id(); |
3708 | rq = cpu_rq(cpu); | |
d6714c22 | 3709 | rcu_sched_qs(cpu); |
dd41f596 IM |
3710 | prev = rq->curr; |
3711 | switch_count = &prev->nivcsw; | |
3712 | ||
3713 | release_kernel_lock(prev); | |
3714 | need_resched_nonpreemptible: | |
3715 | ||
3716 | schedule_debug(prev); | |
1da177e4 | 3717 | |
31656519 | 3718 | if (sched_feat(HRTICK)) |
f333fdc9 | 3719 | hrtick_clear(rq); |
8f4d37ec | 3720 | |
05fa785c | 3721 | raw_spin_lock_irq(&rq->lock); |
3e51f33f | 3722 | update_rq_clock(rq); |
1e819950 | 3723 | clear_tsk_need_resched(prev); |
1da177e4 | 3724 | |
1da177e4 | 3725 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 3726 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 3727 | prev->state = TASK_RUNNING; |
16882c1e | 3728 | else |
2e1cb74a | 3729 | deactivate_task(rq, prev, 1); |
dd41f596 | 3730 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3731 | } |
3732 | ||
3f029d3c | 3733 | pre_schedule(rq, prev); |
f65eda4f | 3734 | |
dd41f596 | 3735 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3736 | idle_balance(cpu, rq); |
1da177e4 | 3737 | |
df1c99d4 | 3738 | put_prev_task(rq, prev); |
b67802ea | 3739 | next = pick_next_task(rq); |
1da177e4 | 3740 | |
1da177e4 | 3741 | if (likely(prev != next)) { |
673a90a1 | 3742 | sched_info_switch(prev, next); |
49f47433 | 3743 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3744 | |
1da177e4 LT |
3745 | rq->nr_switches++; |
3746 | rq->curr = next; | |
3747 | ++*switch_count; | |
3748 | ||
dd41f596 | 3749 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3750 | /* |
3751 | * the context switch might have flipped the stack from under | |
3752 | * us, hence refresh the local variables. | |
3753 | */ | |
3754 | cpu = smp_processor_id(); | |
3755 | rq = cpu_rq(cpu); | |
1da177e4 | 3756 | } else |
05fa785c | 3757 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3758 | |
3f029d3c | 3759 | post_schedule(rq); |
1da177e4 | 3760 | |
6d558c3a YZ |
3761 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3762 | prev = rq->curr; | |
3763 | switch_count = &prev->nivcsw; | |
1da177e4 | 3764 | goto need_resched_nonpreemptible; |
6d558c3a | 3765 | } |
8f4d37ec | 3766 | |
1da177e4 | 3767 | preempt_enable_no_resched(); |
ff743345 | 3768 | if (need_resched()) |
1da177e4 LT |
3769 | goto need_resched; |
3770 | } | |
1da177e4 LT |
3771 | EXPORT_SYMBOL(schedule); |
3772 | ||
c08f7829 | 3773 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3774 | /* |
3775 | * Look out! "owner" is an entirely speculative pointer | |
3776 | * access and not reliable. | |
3777 | */ | |
3778 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3779 | { | |
3780 | unsigned int cpu; | |
3781 | struct rq *rq; | |
3782 | ||
3783 | if (!sched_feat(OWNER_SPIN)) | |
3784 | return 0; | |
3785 | ||
3786 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3787 | /* | |
3788 | * Need to access the cpu field knowing that | |
3789 | * DEBUG_PAGEALLOC could have unmapped it if | |
3790 | * the mutex owner just released it and exited. | |
3791 | */ | |
3792 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 3793 | return 0; |
0d66bf6d PZ |
3794 | #else |
3795 | cpu = owner->cpu; | |
3796 | #endif | |
3797 | ||
3798 | /* | |
3799 | * Even if the access succeeded (likely case), | |
3800 | * the cpu field may no longer be valid. | |
3801 | */ | |
3802 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 3803 | return 0; |
0d66bf6d PZ |
3804 | |
3805 | /* | |
3806 | * We need to validate that we can do a | |
3807 | * get_cpu() and that we have the percpu area. | |
3808 | */ | |
3809 | if (!cpu_online(cpu)) | |
4b402210 | 3810 | return 0; |
0d66bf6d PZ |
3811 | |
3812 | rq = cpu_rq(cpu); | |
3813 | ||
3814 | for (;;) { | |
3815 | /* | |
3816 | * Owner changed, break to re-assess state. | |
3817 | */ | |
3818 | if (lock->owner != owner) | |
3819 | break; | |
3820 | ||
3821 | /* | |
3822 | * Is that owner really running on that cpu? | |
3823 | */ | |
3824 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3825 | return 0; | |
3826 | ||
3827 | cpu_relax(); | |
3828 | } | |
4b402210 | 3829 | |
0d66bf6d PZ |
3830 | return 1; |
3831 | } | |
3832 | #endif | |
3833 | ||
1da177e4 LT |
3834 | #ifdef CONFIG_PREEMPT |
3835 | /* | |
2ed6e34f | 3836 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3837 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3838 | * occur there and call schedule directly. |
3839 | */ | |
3840 | asmlinkage void __sched preempt_schedule(void) | |
3841 | { | |
3842 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3843 | |
1da177e4 LT |
3844 | /* |
3845 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3846 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3847 | */ |
beed33a8 | 3848 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3849 | return; |
3850 | ||
3a5c359a AK |
3851 | do { |
3852 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3853 | schedule(); |
3a5c359a | 3854 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3855 | |
3a5c359a AK |
3856 | /* |
3857 | * Check again in case we missed a preemption opportunity | |
3858 | * between schedule and now. | |
3859 | */ | |
3860 | barrier(); | |
5ed0cec0 | 3861 | } while (need_resched()); |
1da177e4 | 3862 | } |
1da177e4 LT |
3863 | EXPORT_SYMBOL(preempt_schedule); |
3864 | ||
3865 | /* | |
2ed6e34f | 3866 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3867 | * off of irq context. |
3868 | * Note, that this is called and return with irqs disabled. This will | |
3869 | * protect us against recursive calling from irq. | |
3870 | */ | |
3871 | asmlinkage void __sched preempt_schedule_irq(void) | |
3872 | { | |
3873 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3874 | |
2ed6e34f | 3875 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3876 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3877 | ||
3a5c359a AK |
3878 | do { |
3879 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3880 | local_irq_enable(); |
3881 | schedule(); | |
3882 | local_irq_disable(); | |
3a5c359a | 3883 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3884 | |
3a5c359a AK |
3885 | /* |
3886 | * Check again in case we missed a preemption opportunity | |
3887 | * between schedule and now. | |
3888 | */ | |
3889 | barrier(); | |
5ed0cec0 | 3890 | } while (need_resched()); |
1da177e4 LT |
3891 | } |
3892 | ||
3893 | #endif /* CONFIG_PREEMPT */ | |
3894 | ||
63859d4f | 3895 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3896 | void *key) |
1da177e4 | 3897 | { |
63859d4f | 3898 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3899 | } |
1da177e4 LT |
3900 | EXPORT_SYMBOL(default_wake_function); |
3901 | ||
3902 | /* | |
41a2d6cf IM |
3903 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3904 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3905 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3906 | * | |
3907 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3908 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3909 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3910 | */ | |
78ddb08f | 3911 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3912 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3913 | { |
2e45874c | 3914 | wait_queue_t *curr, *next; |
1da177e4 | 3915 | |
2e45874c | 3916 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3917 | unsigned flags = curr->flags; |
3918 | ||
63859d4f | 3919 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3920 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3921 | break; |
3922 | } | |
3923 | } | |
3924 | ||
3925 | /** | |
3926 | * __wake_up - wake up threads blocked on a waitqueue. | |
3927 | * @q: the waitqueue | |
3928 | * @mode: which threads | |
3929 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3930 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3931 | * |
3932 | * It may be assumed that this function implies a write memory barrier before | |
3933 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3934 | */ |
7ad5b3a5 | 3935 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3936 | int nr_exclusive, void *key) |
1da177e4 LT |
3937 | { |
3938 | unsigned long flags; | |
3939 | ||
3940 | spin_lock_irqsave(&q->lock, flags); | |
3941 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3942 | spin_unlock_irqrestore(&q->lock, flags); | |
3943 | } | |
1da177e4 LT |
3944 | EXPORT_SYMBOL(__wake_up); |
3945 | ||
3946 | /* | |
3947 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3948 | */ | |
7ad5b3a5 | 3949 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3950 | { |
3951 | __wake_up_common(q, mode, 1, 0, NULL); | |
3952 | } | |
3953 | ||
4ede816a DL |
3954 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3955 | { | |
3956 | __wake_up_common(q, mode, 1, 0, key); | |
3957 | } | |
3958 | ||
1da177e4 | 3959 | /** |
4ede816a | 3960 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3961 | * @q: the waitqueue |
3962 | * @mode: which threads | |
3963 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3964 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3965 | * |
3966 | * The sync wakeup differs that the waker knows that it will schedule | |
3967 | * away soon, so while the target thread will be woken up, it will not | |
3968 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3969 | * with each other. This can prevent needless bouncing between CPUs. | |
3970 | * | |
3971 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3972 | * |
3973 | * It may be assumed that this function implies a write memory barrier before | |
3974 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3975 | */ |
4ede816a DL |
3976 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3977 | int nr_exclusive, void *key) | |
1da177e4 LT |
3978 | { |
3979 | unsigned long flags; | |
7d478721 | 3980 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3981 | |
3982 | if (unlikely(!q)) | |
3983 | return; | |
3984 | ||
3985 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3986 | wake_flags = 0; |
1da177e4 LT |
3987 | |
3988 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3989 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3990 | spin_unlock_irqrestore(&q->lock, flags); |
3991 | } | |
4ede816a DL |
3992 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3993 | ||
3994 | /* | |
3995 | * __wake_up_sync - see __wake_up_sync_key() | |
3996 | */ | |
3997 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3998 | { | |
3999 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4000 | } | |
1da177e4 LT |
4001 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4002 | ||
65eb3dc6 KD |
4003 | /** |
4004 | * complete: - signals a single thread waiting on this completion | |
4005 | * @x: holds the state of this particular completion | |
4006 | * | |
4007 | * This will wake up a single thread waiting on this completion. Threads will be | |
4008 | * awakened in the same order in which they were queued. | |
4009 | * | |
4010 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4011 | * |
4012 | * It may be assumed that this function implies a write memory barrier before | |
4013 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4014 | */ |
b15136e9 | 4015 | void complete(struct completion *x) |
1da177e4 LT |
4016 | { |
4017 | unsigned long flags; | |
4018 | ||
4019 | spin_lock_irqsave(&x->wait.lock, flags); | |
4020 | x->done++; | |
d9514f6c | 4021 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4022 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4023 | } | |
4024 | EXPORT_SYMBOL(complete); | |
4025 | ||
65eb3dc6 KD |
4026 | /** |
4027 | * complete_all: - signals all threads waiting on this completion | |
4028 | * @x: holds the state of this particular completion | |
4029 | * | |
4030 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4031 | * |
4032 | * It may be assumed that this function implies a write memory barrier before | |
4033 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4034 | */ |
b15136e9 | 4035 | void complete_all(struct completion *x) |
1da177e4 LT |
4036 | { |
4037 | unsigned long flags; | |
4038 | ||
4039 | spin_lock_irqsave(&x->wait.lock, flags); | |
4040 | x->done += UINT_MAX/2; | |
d9514f6c | 4041 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4042 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4043 | } | |
4044 | EXPORT_SYMBOL(complete_all); | |
4045 | ||
8cbbe86d AK |
4046 | static inline long __sched |
4047 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4048 | { |
1da177e4 LT |
4049 | if (!x->done) { |
4050 | DECLARE_WAITQUEUE(wait, current); | |
4051 | ||
4052 | wait.flags |= WQ_FLAG_EXCLUSIVE; | |
4053 | __add_wait_queue_tail(&x->wait, &wait); | |
4054 | do { | |
94d3d824 | 4055 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4056 | timeout = -ERESTARTSYS; |
4057 | break; | |
8cbbe86d AK |
4058 | } |
4059 | __set_current_state(state); | |
1da177e4 LT |
4060 | spin_unlock_irq(&x->wait.lock); |
4061 | timeout = schedule_timeout(timeout); | |
4062 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4063 | } while (!x->done && timeout); |
1da177e4 | 4064 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4065 | if (!x->done) |
4066 | return timeout; | |
1da177e4 LT |
4067 | } |
4068 | x->done--; | |
ea71a546 | 4069 | return timeout ?: 1; |
1da177e4 | 4070 | } |
1da177e4 | 4071 | |
8cbbe86d AK |
4072 | static long __sched |
4073 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4074 | { |
1da177e4 LT |
4075 | might_sleep(); |
4076 | ||
4077 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4078 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4079 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4080 | return timeout; |
4081 | } | |
1da177e4 | 4082 | |
65eb3dc6 KD |
4083 | /** |
4084 | * wait_for_completion: - waits for completion of a task | |
4085 | * @x: holds the state of this particular completion | |
4086 | * | |
4087 | * This waits to be signaled for completion of a specific task. It is NOT | |
4088 | * interruptible and there is no timeout. | |
4089 | * | |
4090 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4091 | * and interrupt capability. Also see complete(). | |
4092 | */ | |
b15136e9 | 4093 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4094 | { |
4095 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4096 | } |
8cbbe86d | 4097 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4098 | |
65eb3dc6 KD |
4099 | /** |
4100 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4101 | * @x: holds the state of this particular completion | |
4102 | * @timeout: timeout value in jiffies | |
4103 | * | |
4104 | * This waits for either a completion of a specific task to be signaled or for a | |
4105 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4106 | * interruptible. | |
4107 | */ | |
b15136e9 | 4108 | unsigned long __sched |
8cbbe86d | 4109 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4110 | { |
8cbbe86d | 4111 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4112 | } |
8cbbe86d | 4113 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4114 | |
65eb3dc6 KD |
4115 | /** |
4116 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4117 | * @x: holds the state of this particular completion | |
4118 | * | |
4119 | * This waits for completion of a specific task to be signaled. It is | |
4120 | * interruptible. | |
4121 | */ | |
8cbbe86d | 4122 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4123 | { |
51e97990 AK |
4124 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4125 | if (t == -ERESTARTSYS) | |
4126 | return t; | |
4127 | return 0; | |
0fec171c | 4128 | } |
8cbbe86d | 4129 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4130 | |
65eb3dc6 KD |
4131 | /** |
4132 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4133 | * @x: holds the state of this particular completion | |
4134 | * @timeout: timeout value in jiffies | |
4135 | * | |
4136 | * This waits for either a completion of a specific task to be signaled or for a | |
4137 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4138 | */ | |
b15136e9 | 4139 | unsigned long __sched |
8cbbe86d AK |
4140 | wait_for_completion_interruptible_timeout(struct completion *x, |
4141 | unsigned long timeout) | |
0fec171c | 4142 | { |
8cbbe86d | 4143 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4144 | } |
8cbbe86d | 4145 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4146 | |
65eb3dc6 KD |
4147 | /** |
4148 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4149 | * @x: holds the state of this particular completion | |
4150 | * | |
4151 | * This waits to be signaled for completion of a specific task. It can be | |
4152 | * interrupted by a kill signal. | |
4153 | */ | |
009e577e MW |
4154 | int __sched wait_for_completion_killable(struct completion *x) |
4155 | { | |
4156 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4157 | if (t == -ERESTARTSYS) | |
4158 | return t; | |
4159 | return 0; | |
4160 | } | |
4161 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4162 | ||
0aa12fb4 SW |
4163 | /** |
4164 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4165 | * @x: holds the state of this particular completion | |
4166 | * @timeout: timeout value in jiffies | |
4167 | * | |
4168 | * This waits for either a completion of a specific task to be | |
4169 | * signaled or for a specified timeout to expire. It can be | |
4170 | * interrupted by a kill signal. The timeout is in jiffies. | |
4171 | */ | |
4172 | unsigned long __sched | |
4173 | wait_for_completion_killable_timeout(struct completion *x, | |
4174 | unsigned long timeout) | |
4175 | { | |
4176 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4177 | } | |
4178 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4179 | ||
be4de352 DC |
4180 | /** |
4181 | * try_wait_for_completion - try to decrement a completion without blocking | |
4182 | * @x: completion structure | |
4183 | * | |
4184 | * Returns: 0 if a decrement cannot be done without blocking | |
4185 | * 1 if a decrement succeeded. | |
4186 | * | |
4187 | * If a completion is being used as a counting completion, | |
4188 | * attempt to decrement the counter without blocking. This | |
4189 | * enables us to avoid waiting if the resource the completion | |
4190 | * is protecting is not available. | |
4191 | */ | |
4192 | bool try_wait_for_completion(struct completion *x) | |
4193 | { | |
7539a3b3 | 4194 | unsigned long flags; |
be4de352 DC |
4195 | int ret = 1; |
4196 | ||
7539a3b3 | 4197 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4198 | if (!x->done) |
4199 | ret = 0; | |
4200 | else | |
4201 | x->done--; | |
7539a3b3 | 4202 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4203 | return ret; |
4204 | } | |
4205 | EXPORT_SYMBOL(try_wait_for_completion); | |
4206 | ||
4207 | /** | |
4208 | * completion_done - Test to see if a completion has any waiters | |
4209 | * @x: completion structure | |
4210 | * | |
4211 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4212 | * 1 if there are no waiters. | |
4213 | * | |
4214 | */ | |
4215 | bool completion_done(struct completion *x) | |
4216 | { | |
7539a3b3 | 4217 | unsigned long flags; |
be4de352 DC |
4218 | int ret = 1; |
4219 | ||
7539a3b3 | 4220 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4221 | if (!x->done) |
4222 | ret = 0; | |
7539a3b3 | 4223 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4224 | return ret; |
4225 | } | |
4226 | EXPORT_SYMBOL(completion_done); | |
4227 | ||
8cbbe86d AK |
4228 | static long __sched |
4229 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4230 | { |
0fec171c IM |
4231 | unsigned long flags; |
4232 | wait_queue_t wait; | |
4233 | ||
4234 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4235 | |
8cbbe86d | 4236 | __set_current_state(state); |
1da177e4 | 4237 | |
8cbbe86d AK |
4238 | spin_lock_irqsave(&q->lock, flags); |
4239 | __add_wait_queue(q, &wait); | |
4240 | spin_unlock(&q->lock); | |
4241 | timeout = schedule_timeout(timeout); | |
4242 | spin_lock_irq(&q->lock); | |
4243 | __remove_wait_queue(q, &wait); | |
4244 | spin_unlock_irqrestore(&q->lock, flags); | |
4245 | ||
4246 | return timeout; | |
4247 | } | |
4248 | ||
4249 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4250 | { | |
4251 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4252 | } |
1da177e4 LT |
4253 | EXPORT_SYMBOL(interruptible_sleep_on); |
4254 | ||
0fec171c | 4255 | long __sched |
95cdf3b7 | 4256 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4257 | { |
8cbbe86d | 4258 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4259 | } |
1da177e4 LT |
4260 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4261 | ||
0fec171c | 4262 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4263 | { |
8cbbe86d | 4264 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4265 | } |
1da177e4 LT |
4266 | EXPORT_SYMBOL(sleep_on); |
4267 | ||
0fec171c | 4268 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4269 | { |
8cbbe86d | 4270 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4271 | } |
1da177e4 LT |
4272 | EXPORT_SYMBOL(sleep_on_timeout); |
4273 | ||
b29739f9 IM |
4274 | #ifdef CONFIG_RT_MUTEXES |
4275 | ||
4276 | /* | |
4277 | * rt_mutex_setprio - set the current priority of a task | |
4278 | * @p: task | |
4279 | * @prio: prio value (kernel-internal form) | |
4280 | * | |
4281 | * This function changes the 'effective' priority of a task. It does | |
4282 | * not touch ->normal_prio like __setscheduler(). | |
4283 | * | |
4284 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4285 | */ | |
36c8b586 | 4286 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4287 | { |
4288 | unsigned long flags; | |
83b699ed | 4289 | int oldprio, on_rq, running; |
70b97a7f | 4290 | struct rq *rq; |
83ab0aa0 | 4291 | const struct sched_class *prev_class; |
b29739f9 IM |
4292 | |
4293 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4294 | ||
4295 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4296 | update_rq_clock(rq); |
b29739f9 | 4297 | |
d5f9f942 | 4298 | oldprio = p->prio; |
83ab0aa0 | 4299 | prev_class = p->sched_class; |
dd41f596 | 4300 | on_rq = p->se.on_rq; |
051a1d1a | 4301 | running = task_current(rq, p); |
0e1f3483 | 4302 | if (on_rq) |
69be72c1 | 4303 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4304 | if (running) |
4305 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4306 | |
4307 | if (rt_prio(prio)) | |
4308 | p->sched_class = &rt_sched_class; | |
4309 | else | |
4310 | p->sched_class = &fair_sched_class; | |
4311 | ||
b29739f9 IM |
4312 | p->prio = prio; |
4313 | ||
0e1f3483 HS |
4314 | if (running) |
4315 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4316 | if (on_rq) { |
60db48ca | 4317 | enqueue_task(rq, p, 0, oldprio < prio); |
cb469845 SR |
4318 | |
4319 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4320 | } |
4321 | task_rq_unlock(rq, &flags); | |
4322 | } | |
4323 | ||
4324 | #endif | |
4325 | ||
36c8b586 | 4326 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4327 | { |
dd41f596 | 4328 | int old_prio, delta, on_rq; |
1da177e4 | 4329 | unsigned long flags; |
70b97a7f | 4330 | struct rq *rq; |
1da177e4 LT |
4331 | |
4332 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4333 | return; | |
4334 | /* | |
4335 | * We have to be careful, if called from sys_setpriority(), | |
4336 | * the task might be in the middle of scheduling on another CPU. | |
4337 | */ | |
4338 | rq = task_rq_lock(p, &flags); | |
a8e504d2 | 4339 | update_rq_clock(rq); |
1da177e4 LT |
4340 | /* |
4341 | * The RT priorities are set via sched_setscheduler(), but we still | |
4342 | * allow the 'normal' nice value to be set - but as expected | |
4343 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4344 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4345 | */ |
e05606d3 | 4346 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4347 | p->static_prio = NICE_TO_PRIO(nice); |
4348 | goto out_unlock; | |
4349 | } | |
dd41f596 | 4350 | on_rq = p->se.on_rq; |
c09595f6 | 4351 | if (on_rq) |
69be72c1 | 4352 | dequeue_task(rq, p, 0); |
1da177e4 | 4353 | |
1da177e4 | 4354 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4355 | set_load_weight(p); |
b29739f9 IM |
4356 | old_prio = p->prio; |
4357 | p->prio = effective_prio(p); | |
4358 | delta = p->prio - old_prio; | |
1da177e4 | 4359 | |
dd41f596 | 4360 | if (on_rq) { |
ea87bb78 | 4361 | enqueue_task(rq, p, 0, false); |
1da177e4 | 4362 | /* |
d5f9f942 AM |
4363 | * If the task increased its priority or is running and |
4364 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4365 | */ |
d5f9f942 | 4366 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4367 | resched_task(rq->curr); |
4368 | } | |
4369 | out_unlock: | |
4370 | task_rq_unlock(rq, &flags); | |
4371 | } | |
1da177e4 LT |
4372 | EXPORT_SYMBOL(set_user_nice); |
4373 | ||
e43379f1 MM |
4374 | /* |
4375 | * can_nice - check if a task can reduce its nice value | |
4376 | * @p: task | |
4377 | * @nice: nice value | |
4378 | */ | |
36c8b586 | 4379 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4380 | { |
024f4747 MM |
4381 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4382 | int nice_rlim = 20 - nice; | |
48f24c4d | 4383 | |
78d7d407 | 4384 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4385 | capable(CAP_SYS_NICE)); |
4386 | } | |
4387 | ||
1da177e4 LT |
4388 | #ifdef __ARCH_WANT_SYS_NICE |
4389 | ||
4390 | /* | |
4391 | * sys_nice - change the priority of the current process. | |
4392 | * @increment: priority increment | |
4393 | * | |
4394 | * sys_setpriority is a more generic, but much slower function that | |
4395 | * does similar things. | |
4396 | */ | |
5add95d4 | 4397 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4398 | { |
48f24c4d | 4399 | long nice, retval; |
1da177e4 LT |
4400 | |
4401 | /* | |
4402 | * Setpriority might change our priority at the same moment. | |
4403 | * We don't have to worry. Conceptually one call occurs first | |
4404 | * and we have a single winner. | |
4405 | */ | |
e43379f1 MM |
4406 | if (increment < -40) |
4407 | increment = -40; | |
1da177e4 LT |
4408 | if (increment > 40) |
4409 | increment = 40; | |
4410 | ||
2b8f836f | 4411 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4412 | if (nice < -20) |
4413 | nice = -20; | |
4414 | if (nice > 19) | |
4415 | nice = 19; | |
4416 | ||
e43379f1 MM |
4417 | if (increment < 0 && !can_nice(current, nice)) |
4418 | return -EPERM; | |
4419 | ||
1da177e4 LT |
4420 | retval = security_task_setnice(current, nice); |
4421 | if (retval) | |
4422 | return retval; | |
4423 | ||
4424 | set_user_nice(current, nice); | |
4425 | return 0; | |
4426 | } | |
4427 | ||
4428 | #endif | |
4429 | ||
4430 | /** | |
4431 | * task_prio - return the priority value of a given task. | |
4432 | * @p: the task in question. | |
4433 | * | |
4434 | * This is the priority value as seen by users in /proc. | |
4435 | * RT tasks are offset by -200. Normal tasks are centered | |
4436 | * around 0, value goes from -16 to +15. | |
4437 | */ | |
36c8b586 | 4438 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4439 | { |
4440 | return p->prio - MAX_RT_PRIO; | |
4441 | } | |
4442 | ||
4443 | /** | |
4444 | * task_nice - return the nice value of a given task. | |
4445 | * @p: the task in question. | |
4446 | */ | |
36c8b586 | 4447 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4448 | { |
4449 | return TASK_NICE(p); | |
4450 | } | |
150d8bed | 4451 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4452 | |
4453 | /** | |
4454 | * idle_cpu - is a given cpu idle currently? | |
4455 | * @cpu: the processor in question. | |
4456 | */ | |
4457 | int idle_cpu(int cpu) | |
4458 | { | |
4459 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4460 | } | |
4461 | ||
1da177e4 LT |
4462 | /** |
4463 | * idle_task - return the idle task for a given cpu. | |
4464 | * @cpu: the processor in question. | |
4465 | */ | |
36c8b586 | 4466 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4467 | { |
4468 | return cpu_rq(cpu)->idle; | |
4469 | } | |
4470 | ||
4471 | /** | |
4472 | * find_process_by_pid - find a process with a matching PID value. | |
4473 | * @pid: the pid in question. | |
4474 | */ | |
a9957449 | 4475 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4476 | { |
228ebcbe | 4477 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4478 | } |
4479 | ||
4480 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4481 | static void |
4482 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4483 | { |
dd41f596 | 4484 | BUG_ON(p->se.on_rq); |
48f24c4d | 4485 | |
1da177e4 LT |
4486 | p->policy = policy; |
4487 | p->rt_priority = prio; | |
b29739f9 IM |
4488 | p->normal_prio = normal_prio(p); |
4489 | /* we are holding p->pi_lock already */ | |
4490 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4491 | if (rt_prio(p->prio)) |
4492 | p->sched_class = &rt_sched_class; | |
4493 | else | |
4494 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4495 | set_load_weight(p); |
1da177e4 LT |
4496 | } |
4497 | ||
c69e8d9c DH |
4498 | /* |
4499 | * check the target process has a UID that matches the current process's | |
4500 | */ | |
4501 | static bool check_same_owner(struct task_struct *p) | |
4502 | { | |
4503 | const struct cred *cred = current_cred(), *pcred; | |
4504 | bool match; | |
4505 | ||
4506 | rcu_read_lock(); | |
4507 | pcred = __task_cred(p); | |
4508 | match = (cred->euid == pcred->euid || | |
4509 | cred->euid == pcred->uid); | |
4510 | rcu_read_unlock(); | |
4511 | return match; | |
4512 | } | |
4513 | ||
961ccddd RR |
4514 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4515 | struct sched_param *param, bool user) | |
1da177e4 | 4516 | { |
83b699ed | 4517 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4518 | unsigned long flags; |
83ab0aa0 | 4519 | const struct sched_class *prev_class; |
70b97a7f | 4520 | struct rq *rq; |
ca94c442 | 4521 | int reset_on_fork; |
1da177e4 | 4522 | |
66e5393a SR |
4523 | /* may grab non-irq protected spin_locks */ |
4524 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4525 | recheck: |
4526 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4527 | if (policy < 0) { |
4528 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4529 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4530 | } else { |
4531 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4532 | policy &= ~SCHED_RESET_ON_FORK; | |
4533 | ||
4534 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4535 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4536 | policy != SCHED_IDLE) | |
4537 | return -EINVAL; | |
4538 | } | |
4539 | ||
1da177e4 LT |
4540 | /* |
4541 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4542 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4543 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4544 | */ |
4545 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4546 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4547 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4548 | return -EINVAL; |
e05606d3 | 4549 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4550 | return -EINVAL; |
4551 | ||
37e4ab3f OC |
4552 | /* |
4553 | * Allow unprivileged RT tasks to decrease priority: | |
4554 | */ | |
961ccddd | 4555 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4556 | if (rt_policy(policy)) { |
8dc3e909 | 4557 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4558 | |
4559 | if (!lock_task_sighand(p, &flags)) | |
4560 | return -ESRCH; | |
78d7d407 | 4561 | rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); |
8dc3e909 ON |
4562 | unlock_task_sighand(p, &flags); |
4563 | ||
4564 | /* can't set/change the rt policy */ | |
4565 | if (policy != p->policy && !rlim_rtprio) | |
4566 | return -EPERM; | |
4567 | ||
4568 | /* can't increase priority */ | |
4569 | if (param->sched_priority > p->rt_priority && | |
4570 | param->sched_priority > rlim_rtprio) | |
4571 | return -EPERM; | |
4572 | } | |
dd41f596 IM |
4573 | /* |
4574 | * Like positive nice levels, dont allow tasks to | |
4575 | * move out of SCHED_IDLE either: | |
4576 | */ | |
4577 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4578 | return -EPERM; | |
5fe1d75f | 4579 | |
37e4ab3f | 4580 | /* can't change other user's priorities */ |
c69e8d9c | 4581 | if (!check_same_owner(p)) |
37e4ab3f | 4582 | return -EPERM; |
ca94c442 LP |
4583 | |
4584 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4585 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4586 | return -EPERM; | |
37e4ab3f | 4587 | } |
1da177e4 | 4588 | |
725aad24 | 4589 | if (user) { |
b68aa230 | 4590 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
4591 | /* |
4592 | * Do not allow realtime tasks into groups that have no runtime | |
4593 | * assigned. | |
4594 | */ | |
9a7e0b18 PZ |
4595 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
4596 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 4597 | return -EPERM; |
b68aa230 PZ |
4598 | #endif |
4599 | ||
725aad24 JF |
4600 | retval = security_task_setscheduler(p, policy, param); |
4601 | if (retval) | |
4602 | return retval; | |
4603 | } | |
4604 | ||
b29739f9 IM |
4605 | /* |
4606 | * make sure no PI-waiters arrive (or leave) while we are | |
4607 | * changing the priority of the task: | |
4608 | */ | |
1d615482 | 4609 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4610 | /* |
4611 | * To be able to change p->policy safely, the apropriate | |
4612 | * runqueue lock must be held. | |
4613 | */ | |
b29739f9 | 4614 | rq = __task_rq_lock(p); |
1da177e4 LT |
4615 | /* recheck policy now with rq lock held */ |
4616 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4617 | policy = oldpolicy = -1; | |
b29739f9 | 4618 | __task_rq_unlock(rq); |
1d615482 | 4619 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4620 | goto recheck; |
4621 | } | |
2daa3577 | 4622 | update_rq_clock(rq); |
dd41f596 | 4623 | on_rq = p->se.on_rq; |
051a1d1a | 4624 | running = task_current(rq, p); |
0e1f3483 | 4625 | if (on_rq) |
2e1cb74a | 4626 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4627 | if (running) |
4628 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4629 | |
ca94c442 LP |
4630 | p->sched_reset_on_fork = reset_on_fork; |
4631 | ||
1da177e4 | 4632 | oldprio = p->prio; |
83ab0aa0 | 4633 | prev_class = p->sched_class; |
dd41f596 | 4634 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4635 | |
0e1f3483 HS |
4636 | if (running) |
4637 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4638 | if (on_rq) { |
4639 | activate_task(rq, p, 0); | |
cb469845 SR |
4640 | |
4641 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4642 | } |
b29739f9 | 4643 | __task_rq_unlock(rq); |
1d615482 | 4644 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4645 | |
95e02ca9 TG |
4646 | rt_mutex_adjust_pi(p); |
4647 | ||
1da177e4 LT |
4648 | return 0; |
4649 | } | |
961ccddd RR |
4650 | |
4651 | /** | |
4652 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4653 | * @p: the task in question. | |
4654 | * @policy: new policy. | |
4655 | * @param: structure containing the new RT priority. | |
4656 | * | |
4657 | * NOTE that the task may be already dead. | |
4658 | */ | |
4659 | int sched_setscheduler(struct task_struct *p, int policy, | |
4660 | struct sched_param *param) | |
4661 | { | |
4662 | return __sched_setscheduler(p, policy, param, true); | |
4663 | } | |
1da177e4 LT |
4664 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4665 | ||
961ccddd RR |
4666 | /** |
4667 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4668 | * @p: the task in question. | |
4669 | * @policy: new policy. | |
4670 | * @param: structure containing the new RT priority. | |
4671 | * | |
4672 | * Just like sched_setscheduler, only don't bother checking if the | |
4673 | * current context has permission. For example, this is needed in | |
4674 | * stop_machine(): we create temporary high priority worker threads, | |
4675 | * but our caller might not have that capability. | |
4676 | */ | |
4677 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4678 | struct sched_param *param) | |
4679 | { | |
4680 | return __sched_setscheduler(p, policy, param, false); | |
4681 | } | |
4682 | ||
95cdf3b7 IM |
4683 | static int |
4684 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4685 | { |
1da177e4 LT |
4686 | struct sched_param lparam; |
4687 | struct task_struct *p; | |
36c8b586 | 4688 | int retval; |
1da177e4 LT |
4689 | |
4690 | if (!param || pid < 0) | |
4691 | return -EINVAL; | |
4692 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4693 | return -EFAULT; | |
5fe1d75f ON |
4694 | |
4695 | rcu_read_lock(); | |
4696 | retval = -ESRCH; | |
1da177e4 | 4697 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4698 | if (p != NULL) |
4699 | retval = sched_setscheduler(p, policy, &lparam); | |
4700 | rcu_read_unlock(); | |
36c8b586 | 4701 | |
1da177e4 LT |
4702 | return retval; |
4703 | } | |
4704 | ||
4705 | /** | |
4706 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4707 | * @pid: the pid in question. | |
4708 | * @policy: new policy. | |
4709 | * @param: structure containing the new RT priority. | |
4710 | */ | |
5add95d4 HC |
4711 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4712 | struct sched_param __user *, param) | |
1da177e4 | 4713 | { |
c21761f1 JB |
4714 | /* negative values for policy are not valid */ |
4715 | if (policy < 0) | |
4716 | return -EINVAL; | |
4717 | ||
1da177e4 LT |
4718 | return do_sched_setscheduler(pid, policy, param); |
4719 | } | |
4720 | ||
4721 | /** | |
4722 | * sys_sched_setparam - set/change the RT priority of a thread | |
4723 | * @pid: the pid in question. | |
4724 | * @param: structure containing the new RT priority. | |
4725 | */ | |
5add95d4 | 4726 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4727 | { |
4728 | return do_sched_setscheduler(pid, -1, param); | |
4729 | } | |
4730 | ||
4731 | /** | |
4732 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4733 | * @pid: the pid in question. | |
4734 | */ | |
5add95d4 | 4735 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4736 | { |
36c8b586 | 4737 | struct task_struct *p; |
3a5c359a | 4738 | int retval; |
1da177e4 LT |
4739 | |
4740 | if (pid < 0) | |
3a5c359a | 4741 | return -EINVAL; |
1da177e4 LT |
4742 | |
4743 | retval = -ESRCH; | |
5fe85be0 | 4744 | rcu_read_lock(); |
1da177e4 LT |
4745 | p = find_process_by_pid(pid); |
4746 | if (p) { | |
4747 | retval = security_task_getscheduler(p); | |
4748 | if (!retval) | |
ca94c442 LP |
4749 | retval = p->policy |
4750 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4751 | } |
5fe85be0 | 4752 | rcu_read_unlock(); |
1da177e4 LT |
4753 | return retval; |
4754 | } | |
4755 | ||
4756 | /** | |
ca94c442 | 4757 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4758 | * @pid: the pid in question. |
4759 | * @param: structure containing the RT priority. | |
4760 | */ | |
5add95d4 | 4761 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4762 | { |
4763 | struct sched_param lp; | |
36c8b586 | 4764 | struct task_struct *p; |
3a5c359a | 4765 | int retval; |
1da177e4 LT |
4766 | |
4767 | if (!param || pid < 0) | |
3a5c359a | 4768 | return -EINVAL; |
1da177e4 | 4769 | |
5fe85be0 | 4770 | rcu_read_lock(); |
1da177e4 LT |
4771 | p = find_process_by_pid(pid); |
4772 | retval = -ESRCH; | |
4773 | if (!p) | |
4774 | goto out_unlock; | |
4775 | ||
4776 | retval = security_task_getscheduler(p); | |
4777 | if (retval) | |
4778 | goto out_unlock; | |
4779 | ||
4780 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4781 | rcu_read_unlock(); |
1da177e4 LT |
4782 | |
4783 | /* | |
4784 | * This one might sleep, we cannot do it with a spinlock held ... | |
4785 | */ | |
4786 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4787 | ||
1da177e4 LT |
4788 | return retval; |
4789 | ||
4790 | out_unlock: | |
5fe85be0 | 4791 | rcu_read_unlock(); |
1da177e4 LT |
4792 | return retval; |
4793 | } | |
4794 | ||
96f874e2 | 4795 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4796 | { |
5a16f3d3 | 4797 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4798 | struct task_struct *p; |
4799 | int retval; | |
1da177e4 | 4800 | |
95402b38 | 4801 | get_online_cpus(); |
23f5d142 | 4802 | rcu_read_lock(); |
1da177e4 LT |
4803 | |
4804 | p = find_process_by_pid(pid); | |
4805 | if (!p) { | |
23f5d142 | 4806 | rcu_read_unlock(); |
95402b38 | 4807 | put_online_cpus(); |
1da177e4 LT |
4808 | return -ESRCH; |
4809 | } | |
4810 | ||
23f5d142 | 4811 | /* Prevent p going away */ |
1da177e4 | 4812 | get_task_struct(p); |
23f5d142 | 4813 | rcu_read_unlock(); |
1da177e4 | 4814 | |
5a16f3d3 RR |
4815 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4816 | retval = -ENOMEM; | |
4817 | goto out_put_task; | |
4818 | } | |
4819 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4820 | retval = -ENOMEM; | |
4821 | goto out_free_cpus_allowed; | |
4822 | } | |
1da177e4 | 4823 | retval = -EPERM; |
c69e8d9c | 4824 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4825 | goto out_unlock; |
4826 | ||
e7834f8f DQ |
4827 | retval = security_task_setscheduler(p, 0, NULL); |
4828 | if (retval) | |
4829 | goto out_unlock; | |
4830 | ||
5a16f3d3 RR |
4831 | cpuset_cpus_allowed(p, cpus_allowed); |
4832 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4833 | again: |
5a16f3d3 | 4834 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4835 | |
8707d8b8 | 4836 | if (!retval) { |
5a16f3d3 RR |
4837 | cpuset_cpus_allowed(p, cpus_allowed); |
4838 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4839 | /* |
4840 | * We must have raced with a concurrent cpuset | |
4841 | * update. Just reset the cpus_allowed to the | |
4842 | * cpuset's cpus_allowed | |
4843 | */ | |
5a16f3d3 | 4844 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4845 | goto again; |
4846 | } | |
4847 | } | |
1da177e4 | 4848 | out_unlock: |
5a16f3d3 RR |
4849 | free_cpumask_var(new_mask); |
4850 | out_free_cpus_allowed: | |
4851 | free_cpumask_var(cpus_allowed); | |
4852 | out_put_task: | |
1da177e4 | 4853 | put_task_struct(p); |
95402b38 | 4854 | put_online_cpus(); |
1da177e4 LT |
4855 | return retval; |
4856 | } | |
4857 | ||
4858 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4859 | struct cpumask *new_mask) |
1da177e4 | 4860 | { |
96f874e2 RR |
4861 | if (len < cpumask_size()) |
4862 | cpumask_clear(new_mask); | |
4863 | else if (len > cpumask_size()) | |
4864 | len = cpumask_size(); | |
4865 | ||
1da177e4 LT |
4866 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4867 | } | |
4868 | ||
4869 | /** | |
4870 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4871 | * @pid: pid of the process | |
4872 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4873 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4874 | */ | |
5add95d4 HC |
4875 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4876 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4877 | { |
5a16f3d3 | 4878 | cpumask_var_t new_mask; |
1da177e4 LT |
4879 | int retval; |
4880 | ||
5a16f3d3 RR |
4881 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4882 | return -ENOMEM; | |
1da177e4 | 4883 | |
5a16f3d3 RR |
4884 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4885 | if (retval == 0) | |
4886 | retval = sched_setaffinity(pid, new_mask); | |
4887 | free_cpumask_var(new_mask); | |
4888 | return retval; | |
1da177e4 LT |
4889 | } |
4890 | ||
96f874e2 | 4891 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4892 | { |
36c8b586 | 4893 | struct task_struct *p; |
31605683 TG |
4894 | unsigned long flags; |
4895 | struct rq *rq; | |
1da177e4 | 4896 | int retval; |
1da177e4 | 4897 | |
95402b38 | 4898 | get_online_cpus(); |
23f5d142 | 4899 | rcu_read_lock(); |
1da177e4 LT |
4900 | |
4901 | retval = -ESRCH; | |
4902 | p = find_process_by_pid(pid); | |
4903 | if (!p) | |
4904 | goto out_unlock; | |
4905 | ||
e7834f8f DQ |
4906 | retval = security_task_getscheduler(p); |
4907 | if (retval) | |
4908 | goto out_unlock; | |
4909 | ||
31605683 | 4910 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4911 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4912 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4913 | |
4914 | out_unlock: | |
23f5d142 | 4915 | rcu_read_unlock(); |
95402b38 | 4916 | put_online_cpus(); |
1da177e4 | 4917 | |
9531b62f | 4918 | return retval; |
1da177e4 LT |
4919 | } |
4920 | ||
4921 | /** | |
4922 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4923 | * @pid: pid of the process | |
4924 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4925 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4926 | */ | |
5add95d4 HC |
4927 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4928 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4929 | { |
4930 | int ret; | |
f17c8607 | 4931 | cpumask_var_t mask; |
1da177e4 | 4932 | |
84fba5ec | 4933 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4934 | return -EINVAL; |
4935 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4936 | return -EINVAL; |
4937 | ||
f17c8607 RR |
4938 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4939 | return -ENOMEM; | |
1da177e4 | 4940 | |
f17c8607 RR |
4941 | ret = sched_getaffinity(pid, mask); |
4942 | if (ret == 0) { | |
8bc037fb | 4943 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4944 | |
4945 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4946 | ret = -EFAULT; |
4947 | else | |
cd3d8031 | 4948 | ret = retlen; |
f17c8607 RR |
4949 | } |
4950 | free_cpumask_var(mask); | |
1da177e4 | 4951 | |
f17c8607 | 4952 | return ret; |
1da177e4 LT |
4953 | } |
4954 | ||
4955 | /** | |
4956 | * sys_sched_yield - yield the current processor to other threads. | |
4957 | * | |
dd41f596 IM |
4958 | * This function yields the current CPU to other tasks. If there are no |
4959 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4960 | */ |
5add95d4 | 4961 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4962 | { |
70b97a7f | 4963 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4964 | |
2d72376b | 4965 | schedstat_inc(rq, yld_count); |
4530d7ab | 4966 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4967 | |
4968 | /* | |
4969 | * Since we are going to call schedule() anyway, there's | |
4970 | * no need to preempt or enable interrupts: | |
4971 | */ | |
4972 | __release(rq->lock); | |
8a25d5de | 4973 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4974 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4975 | preempt_enable_no_resched(); |
4976 | ||
4977 | schedule(); | |
4978 | ||
4979 | return 0; | |
4980 | } | |
4981 | ||
d86ee480 PZ |
4982 | static inline int should_resched(void) |
4983 | { | |
4984 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4985 | } | |
4986 | ||
e7b38404 | 4987 | static void __cond_resched(void) |
1da177e4 | 4988 | { |
e7aaaa69 FW |
4989 | add_preempt_count(PREEMPT_ACTIVE); |
4990 | schedule(); | |
4991 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
4992 | } |
4993 | ||
02b67cc3 | 4994 | int __sched _cond_resched(void) |
1da177e4 | 4995 | { |
d86ee480 | 4996 | if (should_resched()) { |
1da177e4 LT |
4997 | __cond_resched(); |
4998 | return 1; | |
4999 | } | |
5000 | return 0; | |
5001 | } | |
02b67cc3 | 5002 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5003 | |
5004 | /* | |
613afbf8 | 5005 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5006 | * call schedule, and on return reacquire the lock. |
5007 | * | |
41a2d6cf | 5008 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5009 | * operations here to prevent schedule() from being called twice (once via |
5010 | * spin_unlock(), once by hand). | |
5011 | */ | |
613afbf8 | 5012 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5013 | { |
d86ee480 | 5014 | int resched = should_resched(); |
6df3cecb JK |
5015 | int ret = 0; |
5016 | ||
f607c668 PZ |
5017 | lockdep_assert_held(lock); |
5018 | ||
95c354fe | 5019 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5020 | spin_unlock(lock); |
d86ee480 | 5021 | if (resched) |
95c354fe NP |
5022 | __cond_resched(); |
5023 | else | |
5024 | cpu_relax(); | |
6df3cecb | 5025 | ret = 1; |
1da177e4 | 5026 | spin_lock(lock); |
1da177e4 | 5027 | } |
6df3cecb | 5028 | return ret; |
1da177e4 | 5029 | } |
613afbf8 | 5030 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5031 | |
613afbf8 | 5032 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5033 | { |
5034 | BUG_ON(!in_softirq()); | |
5035 | ||
d86ee480 | 5036 | if (should_resched()) { |
98d82567 | 5037 | local_bh_enable(); |
1da177e4 LT |
5038 | __cond_resched(); |
5039 | local_bh_disable(); | |
5040 | return 1; | |
5041 | } | |
5042 | return 0; | |
5043 | } | |
613afbf8 | 5044 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5045 | |
1da177e4 LT |
5046 | /** |
5047 | * yield - yield the current processor to other threads. | |
5048 | * | |
72fd4a35 | 5049 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5050 | * thread runnable and calls sys_sched_yield(). |
5051 | */ | |
5052 | void __sched yield(void) | |
5053 | { | |
5054 | set_current_state(TASK_RUNNING); | |
5055 | sys_sched_yield(); | |
5056 | } | |
1da177e4 LT |
5057 | EXPORT_SYMBOL(yield); |
5058 | ||
5059 | /* | |
41a2d6cf | 5060 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5061 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5062 | */ |
5063 | void __sched io_schedule(void) | |
5064 | { | |
54d35f29 | 5065 | struct rq *rq = raw_rq(); |
1da177e4 | 5066 | |
0ff92245 | 5067 | delayacct_blkio_start(); |
1da177e4 | 5068 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5069 | current->in_iowait = 1; |
1da177e4 | 5070 | schedule(); |
8f0dfc34 | 5071 | current->in_iowait = 0; |
1da177e4 | 5072 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5073 | delayacct_blkio_end(); |
1da177e4 | 5074 | } |
1da177e4 LT |
5075 | EXPORT_SYMBOL(io_schedule); |
5076 | ||
5077 | long __sched io_schedule_timeout(long timeout) | |
5078 | { | |
54d35f29 | 5079 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5080 | long ret; |
5081 | ||
0ff92245 | 5082 | delayacct_blkio_start(); |
1da177e4 | 5083 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5084 | current->in_iowait = 1; |
1da177e4 | 5085 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5086 | current->in_iowait = 0; |
1da177e4 | 5087 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5088 | delayacct_blkio_end(); |
1da177e4 LT |
5089 | return ret; |
5090 | } | |
5091 | ||
5092 | /** | |
5093 | * sys_sched_get_priority_max - return maximum RT priority. | |
5094 | * @policy: scheduling class. | |
5095 | * | |
5096 | * this syscall returns the maximum rt_priority that can be used | |
5097 | * by a given scheduling class. | |
5098 | */ | |
5add95d4 | 5099 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5100 | { |
5101 | int ret = -EINVAL; | |
5102 | ||
5103 | switch (policy) { | |
5104 | case SCHED_FIFO: | |
5105 | case SCHED_RR: | |
5106 | ret = MAX_USER_RT_PRIO-1; | |
5107 | break; | |
5108 | case SCHED_NORMAL: | |
b0a9499c | 5109 | case SCHED_BATCH: |
dd41f596 | 5110 | case SCHED_IDLE: |
1da177e4 LT |
5111 | ret = 0; |
5112 | break; | |
5113 | } | |
5114 | return ret; | |
5115 | } | |
5116 | ||
5117 | /** | |
5118 | * sys_sched_get_priority_min - return minimum RT priority. | |
5119 | * @policy: scheduling class. | |
5120 | * | |
5121 | * this syscall returns the minimum rt_priority that can be used | |
5122 | * by a given scheduling class. | |
5123 | */ | |
5add95d4 | 5124 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5125 | { |
5126 | int ret = -EINVAL; | |
5127 | ||
5128 | switch (policy) { | |
5129 | case SCHED_FIFO: | |
5130 | case SCHED_RR: | |
5131 | ret = 1; | |
5132 | break; | |
5133 | case SCHED_NORMAL: | |
b0a9499c | 5134 | case SCHED_BATCH: |
dd41f596 | 5135 | case SCHED_IDLE: |
1da177e4 LT |
5136 | ret = 0; |
5137 | } | |
5138 | return ret; | |
5139 | } | |
5140 | ||
5141 | /** | |
5142 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5143 | * @pid: pid of the process. | |
5144 | * @interval: userspace pointer to the timeslice value. | |
5145 | * | |
5146 | * this syscall writes the default timeslice value of a given process | |
5147 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5148 | */ | |
17da2bd9 | 5149 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5150 | struct timespec __user *, interval) |
1da177e4 | 5151 | { |
36c8b586 | 5152 | struct task_struct *p; |
a4ec24b4 | 5153 | unsigned int time_slice; |
dba091b9 TG |
5154 | unsigned long flags; |
5155 | struct rq *rq; | |
3a5c359a | 5156 | int retval; |
1da177e4 | 5157 | struct timespec t; |
1da177e4 LT |
5158 | |
5159 | if (pid < 0) | |
3a5c359a | 5160 | return -EINVAL; |
1da177e4 LT |
5161 | |
5162 | retval = -ESRCH; | |
1a551ae7 | 5163 | rcu_read_lock(); |
1da177e4 LT |
5164 | p = find_process_by_pid(pid); |
5165 | if (!p) | |
5166 | goto out_unlock; | |
5167 | ||
5168 | retval = security_task_getscheduler(p); | |
5169 | if (retval) | |
5170 | goto out_unlock; | |
5171 | ||
dba091b9 TG |
5172 | rq = task_rq_lock(p, &flags); |
5173 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5174 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5175 | |
1a551ae7 | 5176 | rcu_read_unlock(); |
a4ec24b4 | 5177 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5178 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5179 | return retval; |
3a5c359a | 5180 | |
1da177e4 | 5181 | out_unlock: |
1a551ae7 | 5182 | rcu_read_unlock(); |
1da177e4 LT |
5183 | return retval; |
5184 | } | |
5185 | ||
7c731e0a | 5186 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5187 | |
82a1fcb9 | 5188 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5189 | { |
1da177e4 | 5190 | unsigned long free = 0; |
36c8b586 | 5191 | unsigned state; |
1da177e4 | 5192 | |
1da177e4 | 5193 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5194 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5195 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5196 | #if BITS_PER_LONG == 32 |
1da177e4 | 5197 | if (state == TASK_RUNNING) |
3df0fc5b | 5198 | printk(KERN_CONT " running "); |
1da177e4 | 5199 | else |
3df0fc5b | 5200 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5201 | #else |
5202 | if (state == TASK_RUNNING) | |
3df0fc5b | 5203 | printk(KERN_CONT " running task "); |
1da177e4 | 5204 | else |
3df0fc5b | 5205 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5206 | #endif |
5207 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5208 | free = stack_not_used(p); |
1da177e4 | 5209 | #endif |
3df0fc5b | 5210 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5211 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5212 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5213 | |
5fb5e6de | 5214 | show_stack(p, NULL); |
1da177e4 LT |
5215 | } |
5216 | ||
e59e2ae2 | 5217 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5218 | { |
36c8b586 | 5219 | struct task_struct *g, *p; |
1da177e4 | 5220 | |
4bd77321 | 5221 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5222 | printk(KERN_INFO |
5223 | " task PC stack pid father\n"); | |
1da177e4 | 5224 | #else |
3df0fc5b PZ |
5225 | printk(KERN_INFO |
5226 | " task PC stack pid father\n"); | |
1da177e4 LT |
5227 | #endif |
5228 | read_lock(&tasklist_lock); | |
5229 | do_each_thread(g, p) { | |
5230 | /* | |
5231 | * reset the NMI-timeout, listing all files on a slow | |
5232 | * console might take alot of time: | |
5233 | */ | |
5234 | touch_nmi_watchdog(); | |
39bc89fd | 5235 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5236 | sched_show_task(p); |
1da177e4 LT |
5237 | } while_each_thread(g, p); |
5238 | ||
04c9167f JF |
5239 | touch_all_softlockup_watchdogs(); |
5240 | ||
dd41f596 IM |
5241 | #ifdef CONFIG_SCHED_DEBUG |
5242 | sysrq_sched_debug_show(); | |
5243 | #endif | |
1da177e4 | 5244 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5245 | /* |
5246 | * Only show locks if all tasks are dumped: | |
5247 | */ | |
93335a21 | 5248 | if (!state_filter) |
e59e2ae2 | 5249 | debug_show_all_locks(); |
1da177e4 LT |
5250 | } |
5251 | ||
1df21055 IM |
5252 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5253 | { | |
dd41f596 | 5254 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5255 | } |
5256 | ||
f340c0d1 IM |
5257 | /** |
5258 | * init_idle - set up an idle thread for a given CPU | |
5259 | * @idle: task in question | |
5260 | * @cpu: cpu the idle task belongs to | |
5261 | * | |
5262 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5263 | * flag, to make booting more robust. | |
5264 | */ | |
5c1e1767 | 5265 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5266 | { |
70b97a7f | 5267 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5268 | unsigned long flags; |
5269 | ||
05fa785c | 5270 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5271 | |
dd41f596 | 5272 | __sched_fork(idle); |
06b83b5f | 5273 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5274 | idle->se.exec_start = sched_clock(); |
5275 | ||
96f874e2 | 5276 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5277 | __set_task_cpu(idle, cpu); |
1da177e4 | 5278 | |
1da177e4 | 5279 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5280 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5281 | idle->oncpu = 1; | |
5282 | #endif | |
05fa785c | 5283 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5284 | |
5285 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5286 | #if defined(CONFIG_PREEMPT) |
5287 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5288 | #else | |
a1261f54 | 5289 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5290 | #endif |
dd41f596 IM |
5291 | /* |
5292 | * The idle tasks have their own, simple scheduling class: | |
5293 | */ | |
5294 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5295 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5296 | } |
5297 | ||
5298 | /* | |
5299 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5300 | * indicates which cpus entered this state. This is used | |
5301 | * in the rcu update to wait only for active cpus. For system | |
5302 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5303 | * always be CPU_BITS_NONE. |
1da177e4 | 5304 | */ |
6a7b3dc3 | 5305 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5306 | |
19978ca6 IM |
5307 | /* |
5308 | * Increase the granularity value when there are more CPUs, | |
5309 | * because with more CPUs the 'effective latency' as visible | |
5310 | * to users decreases. But the relationship is not linear, | |
5311 | * so pick a second-best guess by going with the log2 of the | |
5312 | * number of CPUs. | |
5313 | * | |
5314 | * This idea comes from the SD scheduler of Con Kolivas: | |
5315 | */ | |
acb4a848 | 5316 | static int get_update_sysctl_factor(void) |
19978ca6 | 5317 | { |
4ca3ef71 | 5318 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5319 | unsigned int factor; |
5320 | ||
5321 | switch (sysctl_sched_tunable_scaling) { | |
5322 | case SCHED_TUNABLESCALING_NONE: | |
5323 | factor = 1; | |
5324 | break; | |
5325 | case SCHED_TUNABLESCALING_LINEAR: | |
5326 | factor = cpus; | |
5327 | break; | |
5328 | case SCHED_TUNABLESCALING_LOG: | |
5329 | default: | |
5330 | factor = 1 + ilog2(cpus); | |
5331 | break; | |
5332 | } | |
19978ca6 | 5333 | |
acb4a848 CE |
5334 | return factor; |
5335 | } | |
19978ca6 | 5336 | |
acb4a848 CE |
5337 | static void update_sysctl(void) |
5338 | { | |
5339 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5340 | |
0bcdcf28 CE |
5341 | #define SET_SYSCTL(name) \ |
5342 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5343 | SET_SYSCTL(sched_min_granularity); | |
5344 | SET_SYSCTL(sched_latency); | |
5345 | SET_SYSCTL(sched_wakeup_granularity); | |
5346 | SET_SYSCTL(sched_shares_ratelimit); | |
5347 | #undef SET_SYSCTL | |
5348 | } | |
55cd5340 | 5349 | |
0bcdcf28 CE |
5350 | static inline void sched_init_granularity(void) |
5351 | { | |
5352 | update_sysctl(); | |
19978ca6 IM |
5353 | } |
5354 | ||
1da177e4 LT |
5355 | #ifdef CONFIG_SMP |
5356 | /* | |
5357 | * This is how migration works: | |
5358 | * | |
70b97a7f | 5359 | * 1) we queue a struct migration_req structure in the source CPU's |
1da177e4 LT |
5360 | * runqueue and wake up that CPU's migration thread. |
5361 | * 2) we down() the locked semaphore => thread blocks. | |
5362 | * 3) migration thread wakes up (implicitly it forces the migrated | |
5363 | * thread off the CPU) | |
5364 | * 4) it gets the migration request and checks whether the migrated | |
5365 | * task is still in the wrong runqueue. | |
5366 | * 5) if it's in the wrong runqueue then the migration thread removes | |
5367 | * it and puts it into the right queue. | |
5368 | * 6) migration thread up()s the semaphore. | |
5369 | * 7) we wake up and the migration is done. | |
5370 | */ | |
5371 | ||
5372 | /* | |
5373 | * Change a given task's CPU affinity. Migrate the thread to a | |
5374 | * proper CPU and schedule it away if the CPU it's executing on | |
5375 | * is removed from the allowed bitmask. | |
5376 | * | |
5377 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5378 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5379 | * call is not atomic; no spinlocks may be held. |
5380 | */ | |
96f874e2 | 5381 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 | 5382 | { |
70b97a7f | 5383 | struct migration_req req; |
1da177e4 | 5384 | unsigned long flags; |
70b97a7f | 5385 | struct rq *rq; |
48f24c4d | 5386 | int ret = 0; |
1da177e4 LT |
5387 | |
5388 | rq = task_rq_lock(p, &flags); | |
e2912009 | 5389 | |
6ad4c188 | 5390 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5391 | ret = -EINVAL; |
5392 | goto out; | |
5393 | } | |
5394 | ||
9985b0ba | 5395 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5396 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5397 | ret = -EINVAL; |
5398 | goto out; | |
5399 | } | |
5400 | ||
73fe6aae | 5401 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5402 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5403 | else { |
96f874e2 RR |
5404 | cpumask_copy(&p->cpus_allowed, new_mask); |
5405 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5406 | } |
5407 | ||
1da177e4 | 5408 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5409 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5410 | goto out; |
5411 | ||
6ad4c188 | 5412 | if (migrate_task(p, cpumask_any_and(cpu_active_mask, new_mask), &req)) { |
1da177e4 | 5413 | /* Need help from migration thread: drop lock and wait. */ |
693525e3 PZ |
5414 | struct task_struct *mt = rq->migration_thread; |
5415 | ||
5416 | get_task_struct(mt); | |
1da177e4 | 5417 | task_rq_unlock(rq, &flags); |
47a70985 | 5418 | wake_up_process(mt); |
693525e3 | 5419 | put_task_struct(mt); |
1da177e4 LT |
5420 | wait_for_completion(&req.done); |
5421 | tlb_migrate_finish(p->mm); | |
5422 | return 0; | |
5423 | } | |
5424 | out: | |
5425 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5426 | |
1da177e4 LT |
5427 | return ret; |
5428 | } | |
cd8ba7cd | 5429 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5430 | |
5431 | /* | |
41a2d6cf | 5432 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5433 | * this because either it can't run here any more (set_cpus_allowed() |
5434 | * away from this CPU, or CPU going down), or because we're | |
5435 | * attempting to rebalance this task on exec (sched_exec). | |
5436 | * | |
5437 | * So we race with normal scheduler movements, but that's OK, as long | |
5438 | * as the task is no longer on this CPU. | |
efc30814 KK |
5439 | * |
5440 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5441 | */ |
efc30814 | 5442 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5443 | { |
70b97a7f | 5444 | struct rq *rq_dest, *rq_src; |
e2912009 | 5445 | int ret = 0; |
1da177e4 | 5446 | |
e761b772 | 5447 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5448 | return ret; |
1da177e4 LT |
5449 | |
5450 | rq_src = cpu_rq(src_cpu); | |
5451 | rq_dest = cpu_rq(dest_cpu); | |
5452 | ||
5453 | double_rq_lock(rq_src, rq_dest); | |
5454 | /* Already moved. */ | |
5455 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5456 | goto done; |
1da177e4 | 5457 | /* Affinity changed (again). */ |
96f874e2 | 5458 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5459 | goto fail; |
1da177e4 | 5460 | |
e2912009 PZ |
5461 | /* |
5462 | * If we're not on a rq, the next wake-up will ensure we're | |
5463 | * placed properly. | |
5464 | */ | |
5465 | if (p->se.on_rq) { | |
2e1cb74a | 5466 | deactivate_task(rq_src, p, 0); |
e2912009 | 5467 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5468 | activate_task(rq_dest, p, 0); |
15afe09b | 5469 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5470 | } |
b1e38734 | 5471 | done: |
efc30814 | 5472 | ret = 1; |
b1e38734 | 5473 | fail: |
1da177e4 | 5474 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5475 | return ret; |
1da177e4 LT |
5476 | } |
5477 | ||
03b042bf PM |
5478 | #define RCU_MIGRATION_IDLE 0 |
5479 | #define RCU_MIGRATION_NEED_QS 1 | |
5480 | #define RCU_MIGRATION_GOT_QS 2 | |
5481 | #define RCU_MIGRATION_MUST_SYNC 3 | |
5482 | ||
1da177e4 LT |
5483 | /* |
5484 | * migration_thread - this is a highprio system thread that performs | |
5485 | * thread migration by bumping thread off CPU then 'pushing' onto | |
5486 | * another runqueue. | |
5487 | */ | |
95cdf3b7 | 5488 | static int migration_thread(void *data) |
1da177e4 | 5489 | { |
03b042bf | 5490 | int badcpu; |
1da177e4 | 5491 | int cpu = (long)data; |
70b97a7f | 5492 | struct rq *rq; |
1da177e4 LT |
5493 | |
5494 | rq = cpu_rq(cpu); | |
5495 | BUG_ON(rq->migration_thread != current); | |
5496 | ||
5497 | set_current_state(TASK_INTERRUPTIBLE); | |
5498 | while (!kthread_should_stop()) { | |
70b97a7f | 5499 | struct migration_req *req; |
1da177e4 | 5500 | struct list_head *head; |
1da177e4 | 5501 | |
05fa785c | 5502 | raw_spin_lock_irq(&rq->lock); |
1da177e4 LT |
5503 | |
5504 | if (cpu_is_offline(cpu)) { | |
05fa785c | 5505 | raw_spin_unlock_irq(&rq->lock); |
371cbb38 | 5506 | break; |
1da177e4 LT |
5507 | } |
5508 | ||
5509 | if (rq->active_balance) { | |
5510 | active_load_balance(rq, cpu); | |
5511 | rq->active_balance = 0; | |
5512 | } | |
5513 | ||
5514 | head = &rq->migration_queue; | |
5515 | ||
5516 | if (list_empty(head)) { | |
05fa785c | 5517 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5518 | schedule(); |
5519 | set_current_state(TASK_INTERRUPTIBLE); | |
5520 | continue; | |
5521 | } | |
70b97a7f | 5522 | req = list_entry(head->next, struct migration_req, list); |
1da177e4 LT |
5523 | list_del_init(head->next); |
5524 | ||
03b042bf | 5525 | if (req->task != NULL) { |
05fa785c | 5526 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5527 | __migrate_task(req->task, cpu, req->dest_cpu); |
5528 | } else if (likely(cpu == (badcpu = smp_processor_id()))) { | |
5529 | req->dest_cpu = RCU_MIGRATION_GOT_QS; | |
05fa785c | 5530 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5531 | } else { |
5532 | req->dest_cpu = RCU_MIGRATION_MUST_SYNC; | |
05fa785c | 5533 | raw_spin_unlock(&rq->lock); |
03b042bf PM |
5534 | WARN_ONCE(1, "migration_thread() on CPU %d, expected %d\n", badcpu, cpu); |
5535 | } | |
674311d5 | 5536 | local_irq_enable(); |
1da177e4 LT |
5537 | |
5538 | complete(&req->done); | |
5539 | } | |
5540 | __set_current_state(TASK_RUNNING); | |
1da177e4 | 5541 | |
1da177e4 LT |
5542 | return 0; |
5543 | } | |
5544 | ||
5545 | #ifdef CONFIG_HOTPLUG_CPU | |
f7b4cddc ON |
5546 | |
5547 | static int __migrate_task_irq(struct task_struct *p, int src_cpu, int dest_cpu) | |
5548 | { | |
5549 | int ret; | |
5550 | ||
5551 | local_irq_disable(); | |
5552 | ret = __migrate_task(p, src_cpu, dest_cpu); | |
5553 | local_irq_enable(); | |
5554 | return ret; | |
5555 | } | |
5556 | ||
054b9108 | 5557 | /* |
3a4fa0a2 | 5558 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5559 | */ |
48f24c4d | 5560 | static void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5561 | { |
70b97a7f | 5562 | int dest_cpu; |
e76bd8d9 RR |
5563 | |
5564 | again: | |
5da9a0fb | 5565 | dest_cpu = select_fallback_rq(dead_cpu, p); |
e76bd8d9 | 5566 | |
e76bd8d9 RR |
5567 | /* It can have affinity changed while we were choosing. */ |
5568 | if (unlikely(!__migrate_task_irq(p, dead_cpu, dest_cpu))) | |
5569 | goto again; | |
1da177e4 LT |
5570 | } |
5571 | ||
5572 | /* | |
5573 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5574 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5575 | * for performance reasons the counter is not stricly tracking tasks to | |
5576 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5577 | * to keep the global sum constant after CPU-down: | |
5578 | */ | |
70b97a7f | 5579 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5580 | { |
6ad4c188 | 5581 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5582 | unsigned long flags; |
5583 | ||
5584 | local_irq_save(flags); | |
5585 | double_rq_lock(rq_src, rq_dest); | |
5586 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5587 | rq_src->nr_uninterruptible = 0; | |
5588 | double_rq_unlock(rq_src, rq_dest); | |
5589 | local_irq_restore(flags); | |
5590 | } | |
5591 | ||
5592 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5593 | static void migrate_live_tasks(int src_cpu) | |
5594 | { | |
48f24c4d | 5595 | struct task_struct *p, *t; |
1da177e4 | 5596 | |
f7b4cddc | 5597 | read_lock(&tasklist_lock); |
1da177e4 | 5598 | |
48f24c4d IM |
5599 | do_each_thread(t, p) { |
5600 | if (p == current) | |
1da177e4 LT |
5601 | continue; |
5602 | ||
48f24c4d IM |
5603 | if (task_cpu(p) == src_cpu) |
5604 | move_task_off_dead_cpu(src_cpu, p); | |
5605 | } while_each_thread(t, p); | |
1da177e4 | 5606 | |
f7b4cddc | 5607 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5608 | } |
5609 | ||
dd41f596 IM |
5610 | /* |
5611 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5612 | * It does so by boosting its priority to highest possible. |
5613 | * Used by CPU offline code. | |
1da177e4 LT |
5614 | */ |
5615 | void sched_idle_next(void) | |
5616 | { | |
48f24c4d | 5617 | int this_cpu = smp_processor_id(); |
70b97a7f | 5618 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5619 | struct task_struct *p = rq->idle; |
5620 | unsigned long flags; | |
5621 | ||
5622 | /* cpu has to be offline */ | |
48f24c4d | 5623 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5624 | |
48f24c4d IM |
5625 | /* |
5626 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5627 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5628 | */ |
05fa785c | 5629 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5630 | |
dd41f596 | 5631 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5632 | |
94bc9a7b DA |
5633 | update_rq_clock(rq); |
5634 | activate_task(rq, p, 0); | |
1da177e4 | 5635 | |
05fa785c | 5636 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5637 | } |
5638 | ||
48f24c4d IM |
5639 | /* |
5640 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5641 | * offline. |
5642 | */ | |
5643 | void idle_task_exit(void) | |
5644 | { | |
5645 | struct mm_struct *mm = current->active_mm; | |
5646 | ||
5647 | BUG_ON(cpu_online(smp_processor_id())); | |
5648 | ||
5649 | if (mm != &init_mm) | |
5650 | switch_mm(mm, &init_mm, current); | |
5651 | mmdrop(mm); | |
5652 | } | |
5653 | ||
054b9108 | 5654 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5655 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5656 | { |
70b97a7f | 5657 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5658 | |
5659 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5660 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5661 | |
5662 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5663 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5664 | |
48f24c4d | 5665 | get_task_struct(p); |
1da177e4 LT |
5666 | |
5667 | /* | |
5668 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5669 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5670 | * fine. |
5671 | */ | |
05fa785c | 5672 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5673 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5674 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5675 | |
48f24c4d | 5676 | put_task_struct(p); |
1da177e4 LT |
5677 | } |
5678 | ||
5679 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5680 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5681 | { | |
70b97a7f | 5682 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5683 | struct task_struct *next; |
48f24c4d | 5684 | |
dd41f596 IM |
5685 | for ( ; ; ) { |
5686 | if (!rq->nr_running) | |
5687 | break; | |
a8e504d2 | 5688 | update_rq_clock(rq); |
b67802ea | 5689 | next = pick_next_task(rq); |
dd41f596 IM |
5690 | if (!next) |
5691 | break; | |
79c53799 | 5692 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5693 | migrate_dead(dead_cpu, next); |
e692ab53 | 5694 | |
1da177e4 LT |
5695 | } |
5696 | } | |
dce48a84 TG |
5697 | |
5698 | /* | |
5699 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5700 | */ | |
5701 | static void calc_global_load_remove(struct rq *rq) | |
5702 | { | |
5703 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5704 | rq->calc_load_active = 0; |
dce48a84 | 5705 | } |
1da177e4 LT |
5706 | #endif /* CONFIG_HOTPLUG_CPU */ |
5707 | ||
e692ab53 NP |
5708 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5709 | ||
5710 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5711 | { |
5712 | .procname = "sched_domain", | |
c57baf1e | 5713 | .mode = 0555, |
e0361851 | 5714 | }, |
56992309 | 5715 | {} |
e692ab53 NP |
5716 | }; |
5717 | ||
5718 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5719 | { |
5720 | .procname = "kernel", | |
c57baf1e | 5721 | .mode = 0555, |
e0361851 AD |
5722 | .child = sd_ctl_dir, |
5723 | }, | |
56992309 | 5724 | {} |
e692ab53 NP |
5725 | }; |
5726 | ||
5727 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5728 | { | |
5729 | struct ctl_table *entry = | |
5cf9f062 | 5730 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5731 | |
e692ab53 NP |
5732 | return entry; |
5733 | } | |
5734 | ||
6382bc90 MM |
5735 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5736 | { | |
cd790076 | 5737 | struct ctl_table *entry; |
6382bc90 | 5738 | |
cd790076 MM |
5739 | /* |
5740 | * In the intermediate directories, both the child directory and | |
5741 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5742 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5743 | * static strings and all have proc handlers. |
5744 | */ | |
5745 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5746 | if (entry->child) |
5747 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5748 | if (entry->proc_handler == NULL) |
5749 | kfree(entry->procname); | |
5750 | } | |
6382bc90 MM |
5751 | |
5752 | kfree(*tablep); | |
5753 | *tablep = NULL; | |
5754 | } | |
5755 | ||
e692ab53 | 5756 | static void |
e0361851 | 5757 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5758 | const char *procname, void *data, int maxlen, |
5759 | mode_t mode, proc_handler *proc_handler) | |
5760 | { | |
e692ab53 NP |
5761 | entry->procname = procname; |
5762 | entry->data = data; | |
5763 | entry->maxlen = maxlen; | |
5764 | entry->mode = mode; | |
5765 | entry->proc_handler = proc_handler; | |
5766 | } | |
5767 | ||
5768 | static struct ctl_table * | |
5769 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5770 | { | |
a5d8c348 | 5771 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5772 | |
ad1cdc1d MM |
5773 | if (table == NULL) |
5774 | return NULL; | |
5775 | ||
e0361851 | 5776 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5777 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5778 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5779 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5780 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5781 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5782 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5783 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5784 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5785 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5786 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5787 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5788 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5789 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5790 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5791 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5792 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5793 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5794 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5795 | &sd->cache_nice_tries, |
5796 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5797 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5798 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5799 | set_table_entry(&table[11], "name", sd->name, |
5800 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5801 | /* &table[12] is terminator */ | |
e692ab53 NP |
5802 | |
5803 | return table; | |
5804 | } | |
5805 | ||
9a4e7159 | 5806 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5807 | { |
5808 | struct ctl_table *entry, *table; | |
5809 | struct sched_domain *sd; | |
5810 | int domain_num = 0, i; | |
5811 | char buf[32]; | |
5812 | ||
5813 | for_each_domain(cpu, sd) | |
5814 | domain_num++; | |
5815 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5816 | if (table == NULL) |
5817 | return NULL; | |
e692ab53 NP |
5818 | |
5819 | i = 0; | |
5820 | for_each_domain(cpu, sd) { | |
5821 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5822 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5823 | entry->mode = 0555; |
e692ab53 NP |
5824 | entry->child = sd_alloc_ctl_domain_table(sd); |
5825 | entry++; | |
5826 | i++; | |
5827 | } | |
5828 | return table; | |
5829 | } | |
5830 | ||
5831 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5832 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5833 | { |
6ad4c188 | 5834 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5835 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5836 | char buf[32]; | |
5837 | ||
7378547f MM |
5838 | WARN_ON(sd_ctl_dir[0].child); |
5839 | sd_ctl_dir[0].child = entry; | |
5840 | ||
ad1cdc1d MM |
5841 | if (entry == NULL) |
5842 | return; | |
5843 | ||
6ad4c188 | 5844 | for_each_possible_cpu(i) { |
e692ab53 | 5845 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5846 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5847 | entry->mode = 0555; |
e692ab53 | 5848 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5849 | entry++; |
e692ab53 | 5850 | } |
7378547f MM |
5851 | |
5852 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5853 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5854 | } | |
6382bc90 | 5855 | |
7378547f | 5856 | /* may be called multiple times per register */ |
6382bc90 MM |
5857 | static void unregister_sched_domain_sysctl(void) |
5858 | { | |
7378547f MM |
5859 | if (sd_sysctl_header) |
5860 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5861 | sd_sysctl_header = NULL; |
7378547f MM |
5862 | if (sd_ctl_dir[0].child) |
5863 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5864 | } |
e692ab53 | 5865 | #else |
6382bc90 MM |
5866 | static void register_sched_domain_sysctl(void) |
5867 | { | |
5868 | } | |
5869 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5870 | { |
5871 | } | |
5872 | #endif | |
5873 | ||
1f11eb6a GH |
5874 | static void set_rq_online(struct rq *rq) |
5875 | { | |
5876 | if (!rq->online) { | |
5877 | const struct sched_class *class; | |
5878 | ||
c6c4927b | 5879 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5880 | rq->online = 1; |
5881 | ||
5882 | for_each_class(class) { | |
5883 | if (class->rq_online) | |
5884 | class->rq_online(rq); | |
5885 | } | |
5886 | } | |
5887 | } | |
5888 | ||
5889 | static void set_rq_offline(struct rq *rq) | |
5890 | { | |
5891 | if (rq->online) { | |
5892 | const struct sched_class *class; | |
5893 | ||
5894 | for_each_class(class) { | |
5895 | if (class->rq_offline) | |
5896 | class->rq_offline(rq); | |
5897 | } | |
5898 | ||
c6c4927b | 5899 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5900 | rq->online = 0; |
5901 | } | |
5902 | } | |
5903 | ||
1da177e4 LT |
5904 | /* |
5905 | * migration_call - callback that gets triggered when a CPU is added. | |
5906 | * Here we can start up the necessary migration thread for the new CPU. | |
5907 | */ | |
48f24c4d IM |
5908 | static int __cpuinit |
5909 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5910 | { |
1da177e4 | 5911 | struct task_struct *p; |
48f24c4d | 5912 | int cpu = (long)hcpu; |
1da177e4 | 5913 | unsigned long flags; |
70b97a7f | 5914 | struct rq *rq; |
1da177e4 LT |
5915 | |
5916 | switch (action) { | |
5be9361c | 5917 | |
1da177e4 | 5918 | case CPU_UP_PREPARE: |
8bb78442 | 5919 | case CPU_UP_PREPARE_FROZEN: |
dd41f596 | 5920 | p = kthread_create(migration_thread, hcpu, "migration/%d", cpu); |
1da177e4 LT |
5921 | if (IS_ERR(p)) |
5922 | return NOTIFY_BAD; | |
1da177e4 LT |
5923 | kthread_bind(p, cpu); |
5924 | /* Must be high prio: stop_machine expects to yield to it. */ | |
5925 | rq = task_rq_lock(p, &flags); | |
dd41f596 | 5926 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
1da177e4 | 5927 | task_rq_unlock(rq, &flags); |
371cbb38 | 5928 | get_task_struct(p); |
1da177e4 | 5929 | cpu_rq(cpu)->migration_thread = p; |
a468d389 | 5930 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5931 | break; |
48f24c4d | 5932 | |
1da177e4 | 5933 | case CPU_ONLINE: |
8bb78442 | 5934 | case CPU_ONLINE_FROZEN: |
3a4fa0a2 | 5935 | /* Strictly unnecessary, as first user will wake it. */ |
1da177e4 | 5936 | wake_up_process(cpu_rq(cpu)->migration_thread); |
1f94ef59 GH |
5937 | |
5938 | /* Update our root-domain */ | |
5939 | rq = cpu_rq(cpu); | |
05fa785c | 5940 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5941 | if (rq->rd) { |
c6c4927b | 5942 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5943 | |
5944 | set_rq_online(rq); | |
1f94ef59 | 5945 | } |
05fa785c | 5946 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5947 | break; |
48f24c4d | 5948 | |
1da177e4 LT |
5949 | #ifdef CONFIG_HOTPLUG_CPU |
5950 | case CPU_UP_CANCELED: | |
8bb78442 | 5951 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
5952 | if (!cpu_rq(cpu)->migration_thread) |
5953 | break; | |
41a2d6cf | 5954 | /* Unbind it from offline cpu so it can run. Fall thru. */ |
a4c4af7c | 5955 | kthread_bind(cpu_rq(cpu)->migration_thread, |
1e5ce4f4 | 5956 | cpumask_any(cpu_online_mask)); |
1da177e4 | 5957 | kthread_stop(cpu_rq(cpu)->migration_thread); |
371cbb38 | 5958 | put_task_struct(cpu_rq(cpu)->migration_thread); |
1da177e4 LT |
5959 | cpu_rq(cpu)->migration_thread = NULL; |
5960 | break; | |
48f24c4d | 5961 | |
1da177e4 | 5962 | case CPU_DEAD: |
8bb78442 | 5963 | case CPU_DEAD_FROZEN: |
470fd646 | 5964 | cpuset_lock(); /* around calls to cpuset_cpus_allowed_lock() */ |
1da177e4 LT |
5965 | migrate_live_tasks(cpu); |
5966 | rq = cpu_rq(cpu); | |
5967 | kthread_stop(rq->migration_thread); | |
371cbb38 | 5968 | put_task_struct(rq->migration_thread); |
1da177e4 LT |
5969 | rq->migration_thread = NULL; |
5970 | /* Idle task back to normal (off runqueue, low prio) */ | |
05fa785c | 5971 | raw_spin_lock_irq(&rq->lock); |
a8e504d2 | 5972 | update_rq_clock(rq); |
2e1cb74a | 5973 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5974 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5975 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5976 | migrate_dead_tasks(cpu); |
05fa785c | 5977 | raw_spin_unlock_irq(&rq->lock); |
470fd646 | 5978 | cpuset_unlock(); |
1da177e4 LT |
5979 | migrate_nr_uninterruptible(rq); |
5980 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5981 | calc_global_load_remove(rq); |
41a2d6cf IM |
5982 | /* |
5983 | * No need to migrate the tasks: it was best-effort if | |
5984 | * they didn't take sched_hotcpu_mutex. Just wake up | |
5985 | * the requestors. | |
5986 | */ | |
05fa785c | 5987 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5988 | while (!list_empty(&rq->migration_queue)) { |
70b97a7f IM |
5989 | struct migration_req *req; |
5990 | ||
1da177e4 | 5991 | req = list_entry(rq->migration_queue.next, |
70b97a7f | 5992 | struct migration_req, list); |
1da177e4 | 5993 | list_del_init(&req->list); |
05fa785c | 5994 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5995 | complete(&req->done); |
05fa785c | 5996 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5997 | } |
05fa785c | 5998 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 5999 | break; |
57d885fe | 6000 | |
08f503b0 GH |
6001 | case CPU_DYING: |
6002 | case CPU_DYING_FROZEN: | |
57d885fe GH |
6003 | /* Update our root-domain */ |
6004 | rq = cpu_rq(cpu); | |
05fa785c | 6005 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6006 | if (rq->rd) { |
c6c4927b | 6007 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6008 | set_rq_offline(rq); |
57d885fe | 6009 | } |
05fa785c | 6010 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 6011 | break; |
1da177e4 LT |
6012 | #endif |
6013 | } | |
6014 | return NOTIFY_OK; | |
6015 | } | |
6016 | ||
f38b0820 PM |
6017 | /* |
6018 | * Register at high priority so that task migration (migrate_all_tasks) | |
6019 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6020 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6021 | */ |
26c2143b | 6022 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
6023 | .notifier_call = migration_call, |
6024 | .priority = 10 | |
6025 | }; | |
6026 | ||
7babe8db | 6027 | static int __init migration_init(void) |
1da177e4 LT |
6028 | { |
6029 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6030 | int err; |
48f24c4d IM |
6031 | |
6032 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
6033 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6034 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6035 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6036 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6037 | |
a004cd42 | 6038 | return 0; |
1da177e4 | 6039 | } |
7babe8db | 6040 | early_initcall(migration_init); |
1da177e4 LT |
6041 | #endif |
6042 | ||
6043 | #ifdef CONFIG_SMP | |
476f3534 | 6044 | |
3e9830dc | 6045 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6046 | |
f6630114 MT |
6047 | static __read_mostly int sched_domain_debug_enabled; |
6048 | ||
6049 | static int __init sched_domain_debug_setup(char *str) | |
6050 | { | |
6051 | sched_domain_debug_enabled = 1; | |
6052 | ||
6053 | return 0; | |
6054 | } | |
6055 | early_param("sched_debug", sched_domain_debug_setup); | |
6056 | ||
7c16ec58 | 6057 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6058 | struct cpumask *groupmask) |
1da177e4 | 6059 | { |
4dcf6aff | 6060 | struct sched_group *group = sd->groups; |
434d53b0 | 6061 | char str[256]; |
1da177e4 | 6062 | |
968ea6d8 | 6063 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6064 | cpumask_clear(groupmask); |
4dcf6aff IM |
6065 | |
6066 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6067 | ||
6068 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6069 | printk("does not load-balance\n"); |
4dcf6aff | 6070 | if (sd->parent) |
3df0fc5b PZ |
6071 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6072 | " has parent"); | |
4dcf6aff | 6073 | return -1; |
41c7ce9a NP |
6074 | } |
6075 | ||
3df0fc5b | 6076 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6077 | |
758b2cdc | 6078 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6079 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6080 | "CPU%d\n", cpu); | |
4dcf6aff | 6081 | } |
758b2cdc | 6082 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6083 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6084 | " CPU%d\n", cpu); | |
4dcf6aff | 6085 | } |
1da177e4 | 6086 | |
4dcf6aff | 6087 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6088 | do { |
4dcf6aff | 6089 | if (!group) { |
3df0fc5b PZ |
6090 | printk("\n"); |
6091 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6092 | break; |
6093 | } | |
6094 | ||
18a3885f | 6095 | if (!group->cpu_power) { |
3df0fc5b PZ |
6096 | printk(KERN_CONT "\n"); |
6097 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6098 | "set\n"); | |
4dcf6aff IM |
6099 | break; |
6100 | } | |
1da177e4 | 6101 | |
758b2cdc | 6102 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6103 | printk(KERN_CONT "\n"); |
6104 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6105 | break; |
6106 | } | |
1da177e4 | 6107 | |
758b2cdc | 6108 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6109 | printk(KERN_CONT "\n"); |
6110 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6111 | break; |
6112 | } | |
1da177e4 | 6113 | |
758b2cdc | 6114 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6115 | |
968ea6d8 | 6116 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6117 | |
3df0fc5b | 6118 | printk(KERN_CONT " %s", str); |
18a3885f | 6119 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6120 | printk(KERN_CONT " (cpu_power = %d)", |
6121 | group->cpu_power); | |
381512cf | 6122 | } |
1da177e4 | 6123 | |
4dcf6aff IM |
6124 | group = group->next; |
6125 | } while (group != sd->groups); | |
3df0fc5b | 6126 | printk(KERN_CONT "\n"); |
1da177e4 | 6127 | |
758b2cdc | 6128 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6129 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6130 | |
758b2cdc RR |
6131 | if (sd->parent && |
6132 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6133 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6134 | "of domain->span\n"); | |
4dcf6aff IM |
6135 | return 0; |
6136 | } | |
1da177e4 | 6137 | |
4dcf6aff IM |
6138 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6139 | { | |
d5dd3db1 | 6140 | cpumask_var_t groupmask; |
4dcf6aff | 6141 | int level = 0; |
1da177e4 | 6142 | |
f6630114 MT |
6143 | if (!sched_domain_debug_enabled) |
6144 | return; | |
6145 | ||
4dcf6aff IM |
6146 | if (!sd) { |
6147 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6148 | return; | |
6149 | } | |
1da177e4 | 6150 | |
4dcf6aff IM |
6151 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6152 | ||
d5dd3db1 | 6153 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6154 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6155 | return; | |
6156 | } | |
6157 | ||
4dcf6aff | 6158 | for (;;) { |
7c16ec58 | 6159 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6160 | break; |
1da177e4 LT |
6161 | level++; |
6162 | sd = sd->parent; | |
33859f7f | 6163 | if (!sd) |
4dcf6aff IM |
6164 | break; |
6165 | } | |
d5dd3db1 | 6166 | free_cpumask_var(groupmask); |
1da177e4 | 6167 | } |
6d6bc0ad | 6168 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6169 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6170 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6171 | |
1a20ff27 | 6172 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6173 | { |
758b2cdc | 6174 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6175 | return 1; |
6176 | ||
6177 | /* Following flags need at least 2 groups */ | |
6178 | if (sd->flags & (SD_LOAD_BALANCE | | |
6179 | SD_BALANCE_NEWIDLE | | |
6180 | SD_BALANCE_FORK | | |
89c4710e SS |
6181 | SD_BALANCE_EXEC | |
6182 | SD_SHARE_CPUPOWER | | |
6183 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6184 | if (sd->groups != sd->groups->next) |
6185 | return 0; | |
6186 | } | |
6187 | ||
6188 | /* Following flags don't use groups */ | |
c88d5910 | 6189 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6190 | return 0; |
6191 | ||
6192 | return 1; | |
6193 | } | |
6194 | ||
48f24c4d IM |
6195 | static int |
6196 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6197 | { |
6198 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6199 | ||
6200 | if (sd_degenerate(parent)) | |
6201 | return 1; | |
6202 | ||
758b2cdc | 6203 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6204 | return 0; |
6205 | ||
245af2c7 SS |
6206 | /* Flags needing groups don't count if only 1 group in parent */ |
6207 | if (parent->groups == parent->groups->next) { | |
6208 | pflags &= ~(SD_LOAD_BALANCE | | |
6209 | SD_BALANCE_NEWIDLE | | |
6210 | SD_BALANCE_FORK | | |
89c4710e SS |
6211 | SD_BALANCE_EXEC | |
6212 | SD_SHARE_CPUPOWER | | |
6213 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6214 | if (nr_node_ids == 1) |
6215 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6216 | } |
6217 | if (~cflags & pflags) | |
6218 | return 0; | |
6219 | ||
6220 | return 1; | |
6221 | } | |
6222 | ||
c6c4927b RR |
6223 | static void free_rootdomain(struct root_domain *rd) |
6224 | { | |
047106ad PZ |
6225 | synchronize_sched(); |
6226 | ||
68e74568 RR |
6227 | cpupri_cleanup(&rd->cpupri); |
6228 | ||
c6c4927b RR |
6229 | free_cpumask_var(rd->rto_mask); |
6230 | free_cpumask_var(rd->online); | |
6231 | free_cpumask_var(rd->span); | |
6232 | kfree(rd); | |
6233 | } | |
6234 | ||
57d885fe GH |
6235 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6236 | { | |
a0490fa3 | 6237 | struct root_domain *old_rd = NULL; |
57d885fe | 6238 | unsigned long flags; |
57d885fe | 6239 | |
05fa785c | 6240 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6241 | |
6242 | if (rq->rd) { | |
a0490fa3 | 6243 | old_rd = rq->rd; |
57d885fe | 6244 | |
c6c4927b | 6245 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6246 | set_rq_offline(rq); |
57d885fe | 6247 | |
c6c4927b | 6248 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6249 | |
a0490fa3 IM |
6250 | /* |
6251 | * If we dont want to free the old_rt yet then | |
6252 | * set old_rd to NULL to skip the freeing later | |
6253 | * in this function: | |
6254 | */ | |
6255 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6256 | old_rd = NULL; | |
57d885fe GH |
6257 | } |
6258 | ||
6259 | atomic_inc(&rd->refcount); | |
6260 | rq->rd = rd; | |
6261 | ||
c6c4927b | 6262 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6263 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6264 | set_rq_online(rq); |
57d885fe | 6265 | |
05fa785c | 6266 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6267 | |
6268 | if (old_rd) | |
6269 | free_rootdomain(old_rd); | |
57d885fe GH |
6270 | } |
6271 | ||
fd5e1b5d | 6272 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6273 | { |
36b7b6d4 PE |
6274 | gfp_t gfp = GFP_KERNEL; |
6275 | ||
57d885fe GH |
6276 | memset(rd, 0, sizeof(*rd)); |
6277 | ||
36b7b6d4 PE |
6278 | if (bootmem) |
6279 | gfp = GFP_NOWAIT; | |
c6c4927b | 6280 | |
36b7b6d4 | 6281 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6282 | goto out; |
36b7b6d4 | 6283 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6284 | goto free_span; |
36b7b6d4 | 6285 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6286 | goto free_online; |
6e0534f2 | 6287 | |
0fb53029 | 6288 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6289 | goto free_rto_mask; |
c6c4927b | 6290 | return 0; |
6e0534f2 | 6291 | |
68e74568 RR |
6292 | free_rto_mask: |
6293 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6294 | free_online: |
6295 | free_cpumask_var(rd->online); | |
6296 | free_span: | |
6297 | free_cpumask_var(rd->span); | |
0c910d28 | 6298 | out: |
c6c4927b | 6299 | return -ENOMEM; |
57d885fe GH |
6300 | } |
6301 | ||
6302 | static void init_defrootdomain(void) | |
6303 | { | |
c6c4927b RR |
6304 | init_rootdomain(&def_root_domain, true); |
6305 | ||
57d885fe GH |
6306 | atomic_set(&def_root_domain.refcount, 1); |
6307 | } | |
6308 | ||
dc938520 | 6309 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6310 | { |
6311 | struct root_domain *rd; | |
6312 | ||
6313 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6314 | if (!rd) | |
6315 | return NULL; | |
6316 | ||
c6c4927b RR |
6317 | if (init_rootdomain(rd, false) != 0) { |
6318 | kfree(rd); | |
6319 | return NULL; | |
6320 | } | |
57d885fe GH |
6321 | |
6322 | return rd; | |
6323 | } | |
6324 | ||
1da177e4 | 6325 | /* |
0eab9146 | 6326 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6327 | * hold the hotplug lock. |
6328 | */ | |
0eab9146 IM |
6329 | static void |
6330 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6331 | { |
70b97a7f | 6332 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6333 | struct sched_domain *tmp; |
6334 | ||
6335 | /* Remove the sched domains which do not contribute to scheduling. */ | |
f29c9b1c | 6336 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6337 | struct sched_domain *parent = tmp->parent; |
6338 | if (!parent) | |
6339 | break; | |
f29c9b1c | 6340 | |
1a848870 | 6341 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6342 | tmp->parent = parent->parent; |
1a848870 SS |
6343 | if (parent->parent) |
6344 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6345 | } else |
6346 | tmp = tmp->parent; | |
245af2c7 SS |
6347 | } |
6348 | ||
1a848870 | 6349 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6350 | sd = sd->parent; |
1a848870 SS |
6351 | if (sd) |
6352 | sd->child = NULL; | |
6353 | } | |
1da177e4 LT |
6354 | |
6355 | sched_domain_debug(sd, cpu); | |
6356 | ||
57d885fe | 6357 | rq_attach_root(rq, rd); |
674311d5 | 6358 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6359 | } |
6360 | ||
6361 | /* cpus with isolated domains */ | |
dcc30a35 | 6362 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6363 | |
6364 | /* Setup the mask of cpus configured for isolated domains */ | |
6365 | static int __init isolated_cpu_setup(char *str) | |
6366 | { | |
bdddd296 | 6367 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6368 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6369 | return 1; |
6370 | } | |
6371 | ||
8927f494 | 6372 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6373 | |
6374 | /* | |
6711cab4 SS |
6375 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6376 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6377 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6378 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6379 | * |
6380 | * init_sched_build_groups will build a circular linked list of the groups | |
6381 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6382 | * and ->cpu_power to 0. | |
6383 | */ | |
a616058b | 6384 | static void |
96f874e2 RR |
6385 | init_sched_build_groups(const struct cpumask *span, |
6386 | const struct cpumask *cpu_map, | |
6387 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6388 | struct sched_group **sg, |
96f874e2 RR |
6389 | struct cpumask *tmpmask), |
6390 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6391 | { |
6392 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6393 | int i; |
6394 | ||
96f874e2 | 6395 | cpumask_clear(covered); |
7c16ec58 | 6396 | |
abcd083a | 6397 | for_each_cpu(i, span) { |
6711cab4 | 6398 | struct sched_group *sg; |
7c16ec58 | 6399 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6400 | int j; |
6401 | ||
758b2cdc | 6402 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6403 | continue; |
6404 | ||
758b2cdc | 6405 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6406 | sg->cpu_power = 0; |
1da177e4 | 6407 | |
abcd083a | 6408 | for_each_cpu(j, span) { |
7c16ec58 | 6409 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6410 | continue; |
6411 | ||
96f874e2 | 6412 | cpumask_set_cpu(j, covered); |
758b2cdc | 6413 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6414 | } |
6415 | if (!first) | |
6416 | first = sg; | |
6417 | if (last) | |
6418 | last->next = sg; | |
6419 | last = sg; | |
6420 | } | |
6421 | last->next = first; | |
6422 | } | |
6423 | ||
9c1cfda2 | 6424 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6425 | |
9c1cfda2 | 6426 | #ifdef CONFIG_NUMA |
198e2f18 | 6427 | |
9c1cfda2 JH |
6428 | /** |
6429 | * find_next_best_node - find the next node to include in a sched_domain | |
6430 | * @node: node whose sched_domain we're building | |
6431 | * @used_nodes: nodes already in the sched_domain | |
6432 | * | |
41a2d6cf | 6433 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6434 | * finds the closest node not already in the @used_nodes map. |
6435 | * | |
6436 | * Should use nodemask_t. | |
6437 | */ | |
c5f59f08 | 6438 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6439 | { |
6440 | int i, n, val, min_val, best_node = 0; | |
6441 | ||
6442 | min_val = INT_MAX; | |
6443 | ||
076ac2af | 6444 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6445 | /* Start at @node */ |
076ac2af | 6446 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6447 | |
6448 | if (!nr_cpus_node(n)) | |
6449 | continue; | |
6450 | ||
6451 | /* Skip already used nodes */ | |
c5f59f08 | 6452 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6453 | continue; |
6454 | ||
6455 | /* Simple min distance search */ | |
6456 | val = node_distance(node, n); | |
6457 | ||
6458 | if (val < min_val) { | |
6459 | min_val = val; | |
6460 | best_node = n; | |
6461 | } | |
6462 | } | |
6463 | ||
c5f59f08 | 6464 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6465 | return best_node; |
6466 | } | |
6467 | ||
6468 | /** | |
6469 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6470 | * @node: node whose cpumask we're constructing | |
73486722 | 6471 | * @span: resulting cpumask |
9c1cfda2 | 6472 | * |
41a2d6cf | 6473 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6474 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6475 | * out optimally. | |
6476 | */ | |
96f874e2 | 6477 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6478 | { |
c5f59f08 | 6479 | nodemask_t used_nodes; |
48f24c4d | 6480 | int i; |
9c1cfda2 | 6481 | |
6ca09dfc | 6482 | cpumask_clear(span); |
c5f59f08 | 6483 | nodes_clear(used_nodes); |
9c1cfda2 | 6484 | |
6ca09dfc | 6485 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6486 | node_set(node, used_nodes); |
9c1cfda2 JH |
6487 | |
6488 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6489 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6490 | |
6ca09dfc | 6491 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6492 | } |
9c1cfda2 | 6493 | } |
6d6bc0ad | 6494 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6495 | |
5c45bf27 | 6496 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6497 | |
6c99e9ad RR |
6498 | /* |
6499 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6500 | * |
6501 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6502 | * and struct sched_domain. ) | |
6c99e9ad RR |
6503 | */ |
6504 | struct static_sched_group { | |
6505 | struct sched_group sg; | |
6506 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6507 | }; | |
6508 | ||
6509 | struct static_sched_domain { | |
6510 | struct sched_domain sd; | |
6511 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6512 | }; | |
6513 | ||
49a02c51 AH |
6514 | struct s_data { |
6515 | #ifdef CONFIG_NUMA | |
6516 | int sd_allnodes; | |
6517 | cpumask_var_t domainspan; | |
6518 | cpumask_var_t covered; | |
6519 | cpumask_var_t notcovered; | |
6520 | #endif | |
6521 | cpumask_var_t nodemask; | |
6522 | cpumask_var_t this_sibling_map; | |
6523 | cpumask_var_t this_core_map; | |
6524 | cpumask_var_t send_covered; | |
6525 | cpumask_var_t tmpmask; | |
6526 | struct sched_group **sched_group_nodes; | |
6527 | struct root_domain *rd; | |
6528 | }; | |
6529 | ||
2109b99e AH |
6530 | enum s_alloc { |
6531 | sa_sched_groups = 0, | |
6532 | sa_rootdomain, | |
6533 | sa_tmpmask, | |
6534 | sa_send_covered, | |
6535 | sa_this_core_map, | |
6536 | sa_this_sibling_map, | |
6537 | sa_nodemask, | |
6538 | sa_sched_group_nodes, | |
6539 | #ifdef CONFIG_NUMA | |
6540 | sa_notcovered, | |
6541 | sa_covered, | |
6542 | sa_domainspan, | |
6543 | #endif | |
6544 | sa_none, | |
6545 | }; | |
6546 | ||
9c1cfda2 | 6547 | /* |
48f24c4d | 6548 | * SMT sched-domains: |
9c1cfda2 | 6549 | */ |
1da177e4 | 6550 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6551 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6552 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6553 | |
41a2d6cf | 6554 | static int |
96f874e2 RR |
6555 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6556 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6557 | { |
6711cab4 | 6558 | if (sg) |
1871e52c | 6559 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6560 | return cpu; |
6561 | } | |
6d6bc0ad | 6562 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6563 | |
48f24c4d IM |
6564 | /* |
6565 | * multi-core sched-domains: | |
6566 | */ | |
1e9f28fa | 6567 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6568 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6569 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6570 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6571 | |
6572 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6573 | static int |
96f874e2 RR |
6574 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6575 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6576 | { |
6711cab4 | 6577 | int group; |
7c16ec58 | 6578 | |
c69fc56d | 6579 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6580 | group = cpumask_first(mask); |
6711cab4 | 6581 | if (sg) |
6c99e9ad | 6582 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6583 | return group; |
1e9f28fa SS |
6584 | } |
6585 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6586 | static int |
96f874e2 RR |
6587 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6588 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6589 | { |
6711cab4 | 6590 | if (sg) |
6c99e9ad | 6591 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6592 | return cpu; |
6593 | } | |
6594 | #endif | |
6595 | ||
6c99e9ad RR |
6596 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6597 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6598 | |
41a2d6cf | 6599 | static int |
96f874e2 RR |
6600 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6601 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6602 | { |
6711cab4 | 6603 | int group; |
48f24c4d | 6604 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6605 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6606 | group = cpumask_first(mask); |
1e9f28fa | 6607 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6608 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6609 | group = cpumask_first(mask); |
1da177e4 | 6610 | #else |
6711cab4 | 6611 | group = cpu; |
1da177e4 | 6612 | #endif |
6711cab4 | 6613 | if (sg) |
6c99e9ad | 6614 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6615 | return group; |
1da177e4 LT |
6616 | } |
6617 | ||
6618 | #ifdef CONFIG_NUMA | |
1da177e4 | 6619 | /* |
9c1cfda2 JH |
6620 | * The init_sched_build_groups can't handle what we want to do with node |
6621 | * groups, so roll our own. Now each node has its own list of groups which | |
6622 | * gets dynamically allocated. | |
1da177e4 | 6623 | */ |
62ea9ceb | 6624 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6625 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6626 | |
62ea9ceb | 6627 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6628 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6629 | |
96f874e2 RR |
6630 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6631 | struct sched_group **sg, | |
6632 | struct cpumask *nodemask) | |
9c1cfda2 | 6633 | { |
6711cab4 SS |
6634 | int group; |
6635 | ||
6ca09dfc | 6636 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6637 | group = cpumask_first(nodemask); |
6711cab4 SS |
6638 | |
6639 | if (sg) | |
6c99e9ad | 6640 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6641 | return group; |
1da177e4 | 6642 | } |
6711cab4 | 6643 | |
08069033 SS |
6644 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6645 | { | |
6646 | struct sched_group *sg = group_head; | |
6647 | int j; | |
6648 | ||
6649 | if (!sg) | |
6650 | return; | |
3a5c359a | 6651 | do { |
758b2cdc | 6652 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6653 | struct sched_domain *sd; |
08069033 | 6654 | |
6c99e9ad | 6655 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6656 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6657 | /* |
6658 | * Only add "power" once for each | |
6659 | * physical package. | |
6660 | */ | |
6661 | continue; | |
6662 | } | |
08069033 | 6663 | |
18a3885f | 6664 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6665 | } |
6666 | sg = sg->next; | |
6667 | } while (sg != group_head); | |
08069033 | 6668 | } |
0601a88d AH |
6669 | |
6670 | static int build_numa_sched_groups(struct s_data *d, | |
6671 | const struct cpumask *cpu_map, int num) | |
6672 | { | |
6673 | struct sched_domain *sd; | |
6674 | struct sched_group *sg, *prev; | |
6675 | int n, j; | |
6676 | ||
6677 | cpumask_clear(d->covered); | |
6678 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6679 | if (cpumask_empty(d->nodemask)) { | |
6680 | d->sched_group_nodes[num] = NULL; | |
6681 | goto out; | |
6682 | } | |
6683 | ||
6684 | sched_domain_node_span(num, d->domainspan); | |
6685 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6686 | ||
6687 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6688 | GFP_KERNEL, num); | |
6689 | if (!sg) { | |
3df0fc5b PZ |
6690 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6691 | num); | |
0601a88d AH |
6692 | return -ENOMEM; |
6693 | } | |
6694 | d->sched_group_nodes[num] = sg; | |
6695 | ||
6696 | for_each_cpu(j, d->nodemask) { | |
6697 | sd = &per_cpu(node_domains, j).sd; | |
6698 | sd->groups = sg; | |
6699 | } | |
6700 | ||
18a3885f | 6701 | sg->cpu_power = 0; |
0601a88d AH |
6702 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6703 | sg->next = sg; | |
6704 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6705 | ||
6706 | prev = sg; | |
6707 | for (j = 0; j < nr_node_ids; j++) { | |
6708 | n = (num + j) % nr_node_ids; | |
6709 | cpumask_complement(d->notcovered, d->covered); | |
6710 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6711 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6712 | if (cpumask_empty(d->tmpmask)) | |
6713 | break; | |
6714 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6715 | if (cpumask_empty(d->tmpmask)) | |
6716 | continue; | |
6717 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6718 | GFP_KERNEL, num); | |
6719 | if (!sg) { | |
3df0fc5b PZ |
6720 | printk(KERN_WARNING |
6721 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6722 | return -ENOMEM; |
6723 | } | |
18a3885f | 6724 | sg->cpu_power = 0; |
0601a88d AH |
6725 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6726 | sg->next = prev->next; | |
6727 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6728 | prev->next = sg; | |
6729 | prev = sg; | |
6730 | } | |
6731 | out: | |
6732 | return 0; | |
6733 | } | |
6d6bc0ad | 6734 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6735 | |
a616058b | 6736 | #ifdef CONFIG_NUMA |
51888ca2 | 6737 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6738 | static void free_sched_groups(const struct cpumask *cpu_map, |
6739 | struct cpumask *nodemask) | |
51888ca2 | 6740 | { |
a616058b | 6741 | int cpu, i; |
51888ca2 | 6742 | |
abcd083a | 6743 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6744 | struct sched_group **sched_group_nodes |
6745 | = sched_group_nodes_bycpu[cpu]; | |
6746 | ||
51888ca2 SV |
6747 | if (!sched_group_nodes) |
6748 | continue; | |
6749 | ||
076ac2af | 6750 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6751 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6752 | ||
6ca09dfc | 6753 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6754 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6755 | continue; |
6756 | ||
6757 | if (sg == NULL) | |
6758 | continue; | |
6759 | sg = sg->next; | |
6760 | next_sg: | |
6761 | oldsg = sg; | |
6762 | sg = sg->next; | |
6763 | kfree(oldsg); | |
6764 | if (oldsg != sched_group_nodes[i]) | |
6765 | goto next_sg; | |
6766 | } | |
6767 | kfree(sched_group_nodes); | |
6768 | sched_group_nodes_bycpu[cpu] = NULL; | |
6769 | } | |
51888ca2 | 6770 | } |
6d6bc0ad | 6771 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6772 | static void free_sched_groups(const struct cpumask *cpu_map, |
6773 | struct cpumask *nodemask) | |
a616058b SS |
6774 | { |
6775 | } | |
6d6bc0ad | 6776 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6777 | |
89c4710e SS |
6778 | /* |
6779 | * Initialize sched groups cpu_power. | |
6780 | * | |
6781 | * cpu_power indicates the capacity of sched group, which is used while | |
6782 | * distributing the load between different sched groups in a sched domain. | |
6783 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6784 | * there are asymmetries in the topology. If there are asymmetries, group | |
6785 | * having more cpu_power will pickup more load compared to the group having | |
6786 | * less cpu_power. | |
89c4710e SS |
6787 | */ |
6788 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6789 | { | |
6790 | struct sched_domain *child; | |
6791 | struct sched_group *group; | |
f93e65c1 PZ |
6792 | long power; |
6793 | int weight; | |
89c4710e SS |
6794 | |
6795 | WARN_ON(!sd || !sd->groups); | |
6796 | ||
13318a71 | 6797 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6798 | return; |
6799 | ||
6800 | child = sd->child; | |
6801 | ||
18a3885f | 6802 | sd->groups->cpu_power = 0; |
5517d86b | 6803 | |
f93e65c1 PZ |
6804 | if (!child) { |
6805 | power = SCHED_LOAD_SCALE; | |
6806 | weight = cpumask_weight(sched_domain_span(sd)); | |
6807 | /* | |
6808 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6809 | * Usually multiple threads get a better yield out of |
6810 | * that one core than a single thread would have, | |
6811 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6812 | */ |
a52bfd73 PZ |
6813 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6814 | power *= sd->smt_gain; | |
f93e65c1 | 6815 | power /= weight; |
a52bfd73 PZ |
6816 | power >>= SCHED_LOAD_SHIFT; |
6817 | } | |
18a3885f | 6818 | sd->groups->cpu_power += power; |
89c4710e SS |
6819 | return; |
6820 | } | |
6821 | ||
89c4710e | 6822 | /* |
f93e65c1 | 6823 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6824 | */ |
6825 | group = child->groups; | |
6826 | do { | |
18a3885f | 6827 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6828 | group = group->next; |
6829 | } while (group != child->groups); | |
6830 | } | |
6831 | ||
7c16ec58 MT |
6832 | /* |
6833 | * Initializers for schedule domains | |
6834 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6835 | */ | |
6836 | ||
a5d8c348 IM |
6837 | #ifdef CONFIG_SCHED_DEBUG |
6838 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6839 | #else | |
6840 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6841 | #endif | |
6842 | ||
7c16ec58 | 6843 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6844 | |
7c16ec58 MT |
6845 | #define SD_INIT_FUNC(type) \ |
6846 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6847 | { \ | |
6848 | memset(sd, 0, sizeof(*sd)); \ | |
6849 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6850 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6851 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6852 | } |
6853 | ||
6854 | SD_INIT_FUNC(CPU) | |
6855 | #ifdef CONFIG_NUMA | |
6856 | SD_INIT_FUNC(ALLNODES) | |
6857 | SD_INIT_FUNC(NODE) | |
6858 | #endif | |
6859 | #ifdef CONFIG_SCHED_SMT | |
6860 | SD_INIT_FUNC(SIBLING) | |
6861 | #endif | |
6862 | #ifdef CONFIG_SCHED_MC | |
6863 | SD_INIT_FUNC(MC) | |
6864 | #endif | |
6865 | ||
1d3504fc HS |
6866 | static int default_relax_domain_level = -1; |
6867 | ||
6868 | static int __init setup_relax_domain_level(char *str) | |
6869 | { | |
30e0e178 LZ |
6870 | unsigned long val; |
6871 | ||
6872 | val = simple_strtoul(str, NULL, 0); | |
6873 | if (val < SD_LV_MAX) | |
6874 | default_relax_domain_level = val; | |
6875 | ||
1d3504fc HS |
6876 | return 1; |
6877 | } | |
6878 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6879 | ||
6880 | static void set_domain_attribute(struct sched_domain *sd, | |
6881 | struct sched_domain_attr *attr) | |
6882 | { | |
6883 | int request; | |
6884 | ||
6885 | if (!attr || attr->relax_domain_level < 0) { | |
6886 | if (default_relax_domain_level < 0) | |
6887 | return; | |
6888 | else | |
6889 | request = default_relax_domain_level; | |
6890 | } else | |
6891 | request = attr->relax_domain_level; | |
6892 | if (request < sd->level) { | |
6893 | /* turn off idle balance on this domain */ | |
c88d5910 | 6894 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6895 | } else { |
6896 | /* turn on idle balance on this domain */ | |
c88d5910 | 6897 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6898 | } |
6899 | } | |
6900 | ||
2109b99e AH |
6901 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6902 | const struct cpumask *cpu_map) | |
6903 | { | |
6904 | switch (what) { | |
6905 | case sa_sched_groups: | |
6906 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6907 | d->sched_group_nodes = NULL; | |
6908 | case sa_rootdomain: | |
6909 | free_rootdomain(d->rd); /* fall through */ | |
6910 | case sa_tmpmask: | |
6911 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6912 | case sa_send_covered: | |
6913 | free_cpumask_var(d->send_covered); /* fall through */ | |
6914 | case sa_this_core_map: | |
6915 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6916 | case sa_this_sibling_map: | |
6917 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6918 | case sa_nodemask: | |
6919 | free_cpumask_var(d->nodemask); /* fall through */ | |
6920 | case sa_sched_group_nodes: | |
d1b55138 | 6921 | #ifdef CONFIG_NUMA |
2109b99e AH |
6922 | kfree(d->sched_group_nodes); /* fall through */ |
6923 | case sa_notcovered: | |
6924 | free_cpumask_var(d->notcovered); /* fall through */ | |
6925 | case sa_covered: | |
6926 | free_cpumask_var(d->covered); /* fall through */ | |
6927 | case sa_domainspan: | |
6928 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6929 | #endif |
2109b99e AH |
6930 | case sa_none: |
6931 | break; | |
6932 | } | |
6933 | } | |
3404c8d9 | 6934 | |
2109b99e AH |
6935 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6936 | const struct cpumask *cpu_map) | |
6937 | { | |
3404c8d9 | 6938 | #ifdef CONFIG_NUMA |
2109b99e AH |
6939 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6940 | return sa_none; | |
6941 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6942 | return sa_domainspan; | |
6943 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6944 | return sa_covered; | |
6945 | /* Allocate the per-node list of sched groups */ | |
6946 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6947 | sizeof(struct sched_group *), GFP_KERNEL); | |
6948 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6949 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6950 | return sa_notcovered; |
d1b55138 | 6951 | } |
2109b99e | 6952 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6953 | #endif |
2109b99e AH |
6954 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6955 | return sa_sched_group_nodes; | |
6956 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6957 | return sa_nodemask; | |
6958 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6959 | return sa_this_sibling_map; | |
6960 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6961 | return sa_this_core_map; | |
6962 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6963 | return sa_send_covered; | |
6964 | d->rd = alloc_rootdomain(); | |
6965 | if (!d->rd) { | |
3df0fc5b | 6966 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6967 | return sa_tmpmask; |
57d885fe | 6968 | } |
2109b99e AH |
6969 | return sa_rootdomain; |
6970 | } | |
57d885fe | 6971 | |
7f4588f3 AH |
6972 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6973 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6974 | { | |
6975 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6976 | #ifdef CONFIG_NUMA |
7f4588f3 | 6977 | struct sched_domain *parent; |
1da177e4 | 6978 | |
7f4588f3 AH |
6979 | d->sd_allnodes = 0; |
6980 | if (cpumask_weight(cpu_map) > | |
6981 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6982 | sd = &per_cpu(allnodes_domains, i).sd; | |
6983 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6984 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6985 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6986 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6987 | d->sd_allnodes = 1; | |
6988 | } | |
6989 | parent = sd; | |
6990 | ||
6991 | sd = &per_cpu(node_domains, i).sd; | |
6992 | SD_INIT(sd, NODE); | |
6993 | set_domain_attribute(sd, attr); | |
6994 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6995 | sd->parent = parent; | |
6996 | if (parent) | |
6997 | parent->child = sd; | |
6998 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6999 | #endif |
7f4588f3 AH |
7000 | return sd; |
7001 | } | |
1da177e4 | 7002 | |
87cce662 AH |
7003 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7004 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7005 | struct sched_domain *parent, int i) | |
7006 | { | |
7007 | struct sched_domain *sd; | |
7008 | sd = &per_cpu(phys_domains, i).sd; | |
7009 | SD_INIT(sd, CPU); | |
7010 | set_domain_attribute(sd, attr); | |
7011 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
7012 | sd->parent = parent; | |
7013 | if (parent) | |
7014 | parent->child = sd; | |
7015 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7016 | return sd; | |
7017 | } | |
1da177e4 | 7018 | |
410c4081 AH |
7019 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7020 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7021 | struct sched_domain *parent, int i) | |
7022 | { | |
7023 | struct sched_domain *sd = parent; | |
1e9f28fa | 7024 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7025 | sd = &per_cpu(core_domains, i).sd; |
7026 | SD_INIT(sd, MC); | |
7027 | set_domain_attribute(sd, attr); | |
7028 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7029 | sd->parent = parent; | |
7030 | parent->child = sd; | |
7031 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7032 | #endif |
410c4081 AH |
7033 | return sd; |
7034 | } | |
1e9f28fa | 7035 | |
d8173535 AH |
7036 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7037 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7038 | struct sched_domain *parent, int i) | |
7039 | { | |
7040 | struct sched_domain *sd = parent; | |
1da177e4 | 7041 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7042 | sd = &per_cpu(cpu_domains, i).sd; |
7043 | SD_INIT(sd, SIBLING); | |
7044 | set_domain_attribute(sd, attr); | |
7045 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7046 | sd->parent = parent; | |
7047 | parent->child = sd; | |
7048 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7049 | #endif |
d8173535 AH |
7050 | return sd; |
7051 | } | |
1da177e4 | 7052 | |
0e8e85c9 AH |
7053 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7054 | const struct cpumask *cpu_map, int cpu) | |
7055 | { | |
7056 | switch (l) { | |
1da177e4 | 7057 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7058 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7059 | cpumask_and(d->this_sibling_map, cpu_map, | |
7060 | topology_thread_cpumask(cpu)); | |
7061 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7062 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7063 | &cpu_to_cpu_group, | |
7064 | d->send_covered, d->tmpmask); | |
7065 | break; | |
1da177e4 | 7066 | #endif |
1e9f28fa | 7067 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7068 | case SD_LV_MC: /* set up multi-core groups */ |
7069 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7070 | if (cpu == cpumask_first(d->this_core_map)) | |
7071 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7072 | &cpu_to_core_group, | |
7073 | d->send_covered, d->tmpmask); | |
7074 | break; | |
1e9f28fa | 7075 | #endif |
86548096 AH |
7076 | case SD_LV_CPU: /* set up physical groups */ |
7077 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7078 | if (!cpumask_empty(d->nodemask)) | |
7079 | init_sched_build_groups(d->nodemask, cpu_map, | |
7080 | &cpu_to_phys_group, | |
7081 | d->send_covered, d->tmpmask); | |
7082 | break; | |
1da177e4 | 7083 | #ifdef CONFIG_NUMA |
de616e36 AH |
7084 | case SD_LV_ALLNODES: |
7085 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7086 | d->send_covered, d->tmpmask); | |
7087 | break; | |
7088 | #endif | |
0e8e85c9 AH |
7089 | default: |
7090 | break; | |
7c16ec58 | 7091 | } |
0e8e85c9 | 7092 | } |
9c1cfda2 | 7093 | |
2109b99e AH |
7094 | /* |
7095 | * Build sched domains for a given set of cpus and attach the sched domains | |
7096 | * to the individual cpus | |
7097 | */ | |
7098 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7099 | struct sched_domain_attr *attr) | |
7100 | { | |
7101 | enum s_alloc alloc_state = sa_none; | |
7102 | struct s_data d; | |
294b0c96 | 7103 | struct sched_domain *sd; |
2109b99e | 7104 | int i; |
7c16ec58 | 7105 | #ifdef CONFIG_NUMA |
2109b99e | 7106 | d.sd_allnodes = 0; |
7c16ec58 | 7107 | #endif |
9c1cfda2 | 7108 | |
2109b99e AH |
7109 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7110 | if (alloc_state != sa_rootdomain) | |
7111 | goto error; | |
7112 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7113 | |
1da177e4 | 7114 | /* |
1a20ff27 | 7115 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7116 | */ |
abcd083a | 7117 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7118 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7119 | cpu_map); | |
9761eea8 | 7120 | |
7f4588f3 | 7121 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7122 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7123 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7124 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7125 | } |
9c1cfda2 | 7126 | |
abcd083a | 7127 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7128 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 7129 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7130 | } |
9c1cfda2 | 7131 | |
1da177e4 | 7132 | /* Set up physical groups */ |
86548096 AH |
7133 | for (i = 0; i < nr_node_ids; i++) |
7134 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7135 | |
1da177e4 LT |
7136 | #ifdef CONFIG_NUMA |
7137 | /* Set up node groups */ | |
de616e36 AH |
7138 | if (d.sd_allnodes) |
7139 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7140 | |
0601a88d AH |
7141 | for (i = 0; i < nr_node_ids; i++) |
7142 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7143 | goto error; |
1da177e4 LT |
7144 | #endif |
7145 | ||
7146 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7147 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7148 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7149 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7150 | init_sched_groups_power(i, sd); |
5c45bf27 | 7151 | } |
1da177e4 | 7152 | #endif |
1e9f28fa | 7153 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7154 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7155 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7156 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7157 | } |
7158 | #endif | |
1e9f28fa | 7159 | |
abcd083a | 7160 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7161 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7162 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7163 | } |
7164 | ||
9c1cfda2 | 7165 | #ifdef CONFIG_NUMA |
076ac2af | 7166 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7167 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7168 | |
49a02c51 | 7169 | if (d.sd_allnodes) { |
6711cab4 | 7170 | struct sched_group *sg; |
f712c0c7 | 7171 | |
96f874e2 | 7172 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7173 | d.tmpmask); |
f712c0c7 SS |
7174 | init_numa_sched_groups_power(sg); |
7175 | } | |
9c1cfda2 JH |
7176 | #endif |
7177 | ||
1da177e4 | 7178 | /* Attach the domains */ |
abcd083a | 7179 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7180 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7181 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7182 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7183 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7184 | #else |
6c99e9ad | 7185 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7186 | #endif |
49a02c51 | 7187 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7188 | } |
51888ca2 | 7189 | |
2109b99e AH |
7190 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7191 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7192 | return 0; | |
51888ca2 | 7193 | |
51888ca2 | 7194 | error: |
2109b99e AH |
7195 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7196 | return -ENOMEM; | |
1da177e4 | 7197 | } |
029190c5 | 7198 | |
96f874e2 | 7199 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7200 | { |
7201 | return __build_sched_domains(cpu_map, NULL); | |
7202 | } | |
7203 | ||
acc3f5d7 | 7204 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7205 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7206 | static struct sched_domain_attr *dattr_cur; |
7207 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7208 | |
7209 | /* | |
7210 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7211 | * cpumask) fails, then fallback to a single sched domain, |
7212 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7213 | */ |
4212823f | 7214 | static cpumask_var_t fallback_doms; |
029190c5 | 7215 | |
ee79d1bd HC |
7216 | /* |
7217 | * arch_update_cpu_topology lets virtualized architectures update the | |
7218 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7219 | * or 0 if it stayed the same. | |
7220 | */ | |
7221 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7222 | { |
ee79d1bd | 7223 | return 0; |
22e52b07 HC |
7224 | } |
7225 | ||
acc3f5d7 RR |
7226 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7227 | { | |
7228 | int i; | |
7229 | cpumask_var_t *doms; | |
7230 | ||
7231 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7232 | if (!doms) | |
7233 | return NULL; | |
7234 | for (i = 0; i < ndoms; i++) { | |
7235 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7236 | free_sched_domains(doms, i); | |
7237 | return NULL; | |
7238 | } | |
7239 | } | |
7240 | return doms; | |
7241 | } | |
7242 | ||
7243 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7244 | { | |
7245 | unsigned int i; | |
7246 | for (i = 0; i < ndoms; i++) | |
7247 | free_cpumask_var(doms[i]); | |
7248 | kfree(doms); | |
7249 | } | |
7250 | ||
1a20ff27 | 7251 | /* |
41a2d6cf | 7252 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7253 | * For now this just excludes isolated cpus, but could be used to |
7254 | * exclude other special cases in the future. | |
1a20ff27 | 7255 | */ |
96f874e2 | 7256 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7257 | { |
7378547f MM |
7258 | int err; |
7259 | ||
22e52b07 | 7260 | arch_update_cpu_topology(); |
029190c5 | 7261 | ndoms_cur = 1; |
acc3f5d7 | 7262 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7263 | if (!doms_cur) |
acc3f5d7 RR |
7264 | doms_cur = &fallback_doms; |
7265 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7266 | dattr_cur = NULL; |
acc3f5d7 | 7267 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7268 | register_sched_domain_sysctl(); |
7378547f MM |
7269 | |
7270 | return err; | |
1a20ff27 DG |
7271 | } |
7272 | ||
96f874e2 RR |
7273 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7274 | struct cpumask *tmpmask) | |
1da177e4 | 7275 | { |
7c16ec58 | 7276 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7277 | } |
1da177e4 | 7278 | |
1a20ff27 DG |
7279 | /* |
7280 | * Detach sched domains from a group of cpus specified in cpu_map | |
7281 | * These cpus will now be attached to the NULL domain | |
7282 | */ | |
96f874e2 | 7283 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7284 | { |
96f874e2 RR |
7285 | /* Save because hotplug lock held. */ |
7286 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7287 | int i; |
7288 | ||
abcd083a | 7289 | for_each_cpu(i, cpu_map) |
57d885fe | 7290 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7291 | synchronize_sched(); |
96f874e2 | 7292 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7293 | } |
7294 | ||
1d3504fc HS |
7295 | /* handle null as "default" */ |
7296 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7297 | struct sched_domain_attr *new, int idx_new) | |
7298 | { | |
7299 | struct sched_domain_attr tmp; | |
7300 | ||
7301 | /* fast path */ | |
7302 | if (!new && !cur) | |
7303 | return 1; | |
7304 | ||
7305 | tmp = SD_ATTR_INIT; | |
7306 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7307 | new ? (new + idx_new) : &tmp, | |
7308 | sizeof(struct sched_domain_attr)); | |
7309 | } | |
7310 | ||
029190c5 PJ |
7311 | /* |
7312 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7313 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7314 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7315 | * It destroys each deleted domain and builds each new domain. | |
7316 | * | |
acc3f5d7 | 7317 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7318 | * The masks don't intersect (don't overlap.) We should setup one |
7319 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7320 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7321 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7322 | * it as it is. | |
7323 | * | |
acc3f5d7 RR |
7324 | * The passed in 'doms_new' should be allocated using |
7325 | * alloc_sched_domains. This routine takes ownership of it and will | |
7326 | * free_sched_domains it when done with it. If the caller failed the | |
7327 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7328 | * and partition_sched_domains() will fallback to the single partition | |
7329 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7330 | * |
96f874e2 | 7331 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7332 | * ndoms_new == 0 is a special case for destroying existing domains, |
7333 | * and it will not create the default domain. | |
dfb512ec | 7334 | * |
029190c5 PJ |
7335 | * Call with hotplug lock held |
7336 | */ | |
acc3f5d7 | 7337 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7338 | struct sched_domain_attr *dattr_new) |
029190c5 | 7339 | { |
dfb512ec | 7340 | int i, j, n; |
d65bd5ec | 7341 | int new_topology; |
029190c5 | 7342 | |
712555ee | 7343 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7344 | |
7378547f MM |
7345 | /* always unregister in case we don't destroy any domains */ |
7346 | unregister_sched_domain_sysctl(); | |
7347 | ||
d65bd5ec HC |
7348 | /* Let architecture update cpu core mappings. */ |
7349 | new_topology = arch_update_cpu_topology(); | |
7350 | ||
dfb512ec | 7351 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7352 | |
7353 | /* Destroy deleted domains */ | |
7354 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7355 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7356 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7357 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7358 | goto match1; |
7359 | } | |
7360 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7361 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7362 | match1: |
7363 | ; | |
7364 | } | |
7365 | ||
e761b772 MK |
7366 | if (doms_new == NULL) { |
7367 | ndoms_cur = 0; | |
acc3f5d7 | 7368 | doms_new = &fallback_doms; |
6ad4c188 | 7369 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7370 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7371 | } |
7372 | ||
029190c5 PJ |
7373 | /* Build new domains */ |
7374 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7375 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7376 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7377 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7378 | goto match2; |
7379 | } | |
7380 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7381 | __build_sched_domains(doms_new[i], |
1d3504fc | 7382 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7383 | match2: |
7384 | ; | |
7385 | } | |
7386 | ||
7387 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7388 | if (doms_cur != &fallback_doms) |
7389 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7390 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7391 | doms_cur = doms_new; |
1d3504fc | 7392 | dattr_cur = dattr_new; |
029190c5 | 7393 | ndoms_cur = ndoms_new; |
7378547f MM |
7394 | |
7395 | register_sched_domain_sysctl(); | |
a1835615 | 7396 | |
712555ee | 7397 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7398 | } |
7399 | ||
5c45bf27 | 7400 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7401 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7402 | { |
95402b38 | 7403 | get_online_cpus(); |
dfb512ec MK |
7404 | |
7405 | /* Destroy domains first to force the rebuild */ | |
7406 | partition_sched_domains(0, NULL, NULL); | |
7407 | ||
e761b772 | 7408 | rebuild_sched_domains(); |
95402b38 | 7409 | put_online_cpus(); |
5c45bf27 SS |
7410 | } |
7411 | ||
7412 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7413 | { | |
afb8a9b7 | 7414 | unsigned int level = 0; |
5c45bf27 | 7415 | |
afb8a9b7 GS |
7416 | if (sscanf(buf, "%u", &level) != 1) |
7417 | return -EINVAL; | |
7418 | ||
7419 | /* | |
7420 | * level is always be positive so don't check for | |
7421 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7422 | * What happens on 0 or 1 byte write, | |
7423 | * need to check for count as well? | |
7424 | */ | |
7425 | ||
7426 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7427 | return -EINVAL; |
7428 | ||
7429 | if (smt) | |
afb8a9b7 | 7430 | sched_smt_power_savings = level; |
5c45bf27 | 7431 | else |
afb8a9b7 | 7432 | sched_mc_power_savings = level; |
5c45bf27 | 7433 | |
c70f22d2 | 7434 | arch_reinit_sched_domains(); |
5c45bf27 | 7435 | |
c70f22d2 | 7436 | return count; |
5c45bf27 SS |
7437 | } |
7438 | ||
5c45bf27 | 7439 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7440 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7441 | struct sysdev_class_attribute *attr, |
f718cd4a | 7442 | char *page) |
5c45bf27 SS |
7443 | { |
7444 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7445 | } | |
f718cd4a | 7446 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7447 | struct sysdev_class_attribute *attr, |
48f24c4d | 7448 | const char *buf, size_t count) |
5c45bf27 SS |
7449 | { |
7450 | return sched_power_savings_store(buf, count, 0); | |
7451 | } | |
f718cd4a AK |
7452 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7453 | sched_mc_power_savings_show, | |
7454 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7455 | #endif |
7456 | ||
7457 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7458 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7459 | struct sysdev_class_attribute *attr, |
f718cd4a | 7460 | char *page) |
5c45bf27 SS |
7461 | { |
7462 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7463 | } | |
f718cd4a | 7464 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7465 | struct sysdev_class_attribute *attr, |
48f24c4d | 7466 | const char *buf, size_t count) |
5c45bf27 SS |
7467 | { |
7468 | return sched_power_savings_store(buf, count, 1); | |
7469 | } | |
f718cd4a AK |
7470 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7471 | sched_smt_power_savings_show, | |
6707de00 AB |
7472 | sched_smt_power_savings_store); |
7473 | #endif | |
7474 | ||
39aac648 | 7475 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7476 | { |
7477 | int err = 0; | |
7478 | ||
7479 | #ifdef CONFIG_SCHED_SMT | |
7480 | if (smt_capable()) | |
7481 | err = sysfs_create_file(&cls->kset.kobj, | |
7482 | &attr_sched_smt_power_savings.attr); | |
7483 | #endif | |
7484 | #ifdef CONFIG_SCHED_MC | |
7485 | if (!err && mc_capable()) | |
7486 | err = sysfs_create_file(&cls->kset.kobj, | |
7487 | &attr_sched_mc_power_savings.attr); | |
7488 | #endif | |
7489 | return err; | |
7490 | } | |
6d6bc0ad | 7491 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7492 | |
e761b772 | 7493 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7494 | /* |
e761b772 MK |
7495 | * Add online and remove offline CPUs from the scheduler domains. |
7496 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7497 | */ |
7498 | static int update_sched_domains(struct notifier_block *nfb, | |
7499 | unsigned long action, void *hcpu) | |
e761b772 MK |
7500 | { |
7501 | switch (action) { | |
7502 | case CPU_ONLINE: | |
7503 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
7504 | case CPU_DOWN_PREPARE: |
7505 | case CPU_DOWN_PREPARE_FROZEN: | |
7506 | case CPU_DOWN_FAILED: | |
7507 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 7508 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7509 | return NOTIFY_OK; |
7510 | ||
7511 | default: | |
7512 | return NOTIFY_DONE; | |
7513 | } | |
7514 | } | |
7515 | #endif | |
7516 | ||
7517 | static int update_runtime(struct notifier_block *nfb, | |
7518 | unsigned long action, void *hcpu) | |
1da177e4 | 7519 | { |
7def2be1 PZ |
7520 | int cpu = (int)(long)hcpu; |
7521 | ||
1da177e4 | 7522 | switch (action) { |
1da177e4 | 7523 | case CPU_DOWN_PREPARE: |
8bb78442 | 7524 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7525 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7526 | return NOTIFY_OK; |
7527 | ||
1da177e4 | 7528 | case CPU_DOWN_FAILED: |
8bb78442 | 7529 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7530 | case CPU_ONLINE: |
8bb78442 | 7531 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7532 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7533 | return NOTIFY_OK; |
7534 | ||
1da177e4 LT |
7535 | default: |
7536 | return NOTIFY_DONE; | |
7537 | } | |
1da177e4 | 7538 | } |
1da177e4 LT |
7539 | |
7540 | void __init sched_init_smp(void) | |
7541 | { | |
dcc30a35 RR |
7542 | cpumask_var_t non_isolated_cpus; |
7543 | ||
7544 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7545 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7546 | |
434d53b0 MT |
7547 | #if defined(CONFIG_NUMA) |
7548 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7549 | GFP_KERNEL); | |
7550 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7551 | #endif | |
95402b38 | 7552 | get_online_cpus(); |
712555ee | 7553 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7554 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7555 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7556 | if (cpumask_empty(non_isolated_cpus)) | |
7557 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7558 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7559 | put_online_cpus(); |
e761b772 MK |
7560 | |
7561 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7562 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7563 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7564 | #endif |
7565 | ||
7566 | /* RT runtime code needs to handle some hotplug events */ | |
7567 | hotcpu_notifier(update_runtime, 0); | |
7568 | ||
b328ca18 | 7569 | init_hrtick(); |
5c1e1767 NP |
7570 | |
7571 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7572 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7573 | BUG(); |
19978ca6 | 7574 | sched_init_granularity(); |
dcc30a35 | 7575 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7576 | |
0e3900e6 | 7577 | init_sched_rt_class(); |
1da177e4 LT |
7578 | } |
7579 | #else | |
7580 | void __init sched_init_smp(void) | |
7581 | { | |
19978ca6 | 7582 | sched_init_granularity(); |
1da177e4 LT |
7583 | } |
7584 | #endif /* CONFIG_SMP */ | |
7585 | ||
cd1bb94b AB |
7586 | const_debug unsigned int sysctl_timer_migration = 1; |
7587 | ||
1da177e4 LT |
7588 | int in_sched_functions(unsigned long addr) |
7589 | { | |
1da177e4 LT |
7590 | return in_lock_functions(addr) || |
7591 | (addr >= (unsigned long)__sched_text_start | |
7592 | && addr < (unsigned long)__sched_text_end); | |
7593 | } | |
7594 | ||
a9957449 | 7595 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7596 | { |
7597 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7598 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7599 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7600 | cfs_rq->rq = rq; | |
7601 | #endif | |
67e9fb2a | 7602 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7603 | } |
7604 | ||
fa85ae24 PZ |
7605 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7606 | { | |
7607 | struct rt_prio_array *array; | |
7608 | int i; | |
7609 | ||
7610 | array = &rt_rq->active; | |
7611 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7612 | INIT_LIST_HEAD(array->queue + i); | |
7613 | __clear_bit(i, array->bitmap); | |
7614 | } | |
7615 | /* delimiter for bitsearch: */ | |
7616 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7617 | ||
052f1dc7 | 7618 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7619 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7620 | #ifdef CONFIG_SMP |
e864c499 | 7621 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7622 | #endif |
48d5e258 | 7623 | #endif |
fa85ae24 PZ |
7624 | #ifdef CONFIG_SMP |
7625 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7626 | rt_rq->overloaded = 0; |
05fa785c | 7627 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7628 | #endif |
7629 | ||
7630 | rt_rq->rt_time = 0; | |
7631 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7632 | rt_rq->rt_runtime = 0; |
0986b11b | 7633 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7634 | |
052f1dc7 | 7635 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7636 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7637 | rt_rq->rq = rq; |
7638 | #endif | |
fa85ae24 PZ |
7639 | } |
7640 | ||
6f505b16 | 7641 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7642 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7643 | struct sched_entity *se, int cpu, int add, | |
7644 | struct sched_entity *parent) | |
6f505b16 | 7645 | { |
ec7dc8ac | 7646 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7647 | tg->cfs_rq[cpu] = cfs_rq; |
7648 | init_cfs_rq(cfs_rq, rq); | |
7649 | cfs_rq->tg = tg; | |
7650 | if (add) | |
7651 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7652 | ||
7653 | tg->se[cpu] = se; | |
354d60c2 DG |
7654 | /* se could be NULL for init_task_group */ |
7655 | if (!se) | |
7656 | return; | |
7657 | ||
ec7dc8ac DG |
7658 | if (!parent) |
7659 | se->cfs_rq = &rq->cfs; | |
7660 | else | |
7661 | se->cfs_rq = parent->my_q; | |
7662 | ||
6f505b16 PZ |
7663 | se->my_q = cfs_rq; |
7664 | se->load.weight = tg->shares; | |
e05510d0 | 7665 | se->load.inv_weight = 0; |
ec7dc8ac | 7666 | se->parent = parent; |
6f505b16 | 7667 | } |
052f1dc7 | 7668 | #endif |
6f505b16 | 7669 | |
052f1dc7 | 7670 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7671 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7672 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7673 | struct sched_rt_entity *parent) | |
6f505b16 | 7674 | { |
ec7dc8ac DG |
7675 | struct rq *rq = cpu_rq(cpu); |
7676 | ||
6f505b16 PZ |
7677 | tg->rt_rq[cpu] = rt_rq; |
7678 | init_rt_rq(rt_rq, rq); | |
7679 | rt_rq->tg = tg; | |
ac086bc2 | 7680 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7681 | if (add) |
7682 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7683 | ||
7684 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7685 | if (!rt_se) |
7686 | return; | |
7687 | ||
ec7dc8ac DG |
7688 | if (!parent) |
7689 | rt_se->rt_rq = &rq->rt; | |
7690 | else | |
7691 | rt_se->rt_rq = parent->my_q; | |
7692 | ||
6f505b16 | 7693 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7694 | rt_se->parent = parent; |
6f505b16 PZ |
7695 | INIT_LIST_HEAD(&rt_se->run_list); |
7696 | } | |
7697 | #endif | |
7698 | ||
1da177e4 LT |
7699 | void __init sched_init(void) |
7700 | { | |
dd41f596 | 7701 | int i, j; |
434d53b0 MT |
7702 | unsigned long alloc_size = 0, ptr; |
7703 | ||
7704 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7705 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7706 | #endif | |
7707 | #ifdef CONFIG_RT_GROUP_SCHED | |
7708 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7709 | #endif |
df7c8e84 | 7710 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7711 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7712 | #endif |
434d53b0 | 7713 | if (alloc_size) { |
36b7b6d4 | 7714 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7715 | |
7716 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7717 | init_task_group.se = (struct sched_entity **)ptr; | |
7718 | ptr += nr_cpu_ids * sizeof(void **); | |
7719 | ||
7720 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7721 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7722 | |
6d6bc0ad | 7723 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7724 | #ifdef CONFIG_RT_GROUP_SCHED |
7725 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7726 | ptr += nr_cpu_ids * sizeof(void **); | |
7727 | ||
7728 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7729 | ptr += nr_cpu_ids * sizeof(void **); |
7730 | ||
6d6bc0ad | 7731 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7732 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7733 | for_each_possible_cpu(i) { | |
7734 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7735 | ptr += cpumask_size(); | |
7736 | } | |
7737 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7738 | } |
dd41f596 | 7739 | |
57d885fe GH |
7740 | #ifdef CONFIG_SMP |
7741 | init_defrootdomain(); | |
7742 | #endif | |
7743 | ||
d0b27fa7 PZ |
7744 | init_rt_bandwidth(&def_rt_bandwidth, |
7745 | global_rt_period(), global_rt_runtime()); | |
7746 | ||
7747 | #ifdef CONFIG_RT_GROUP_SCHED | |
7748 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7749 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7750 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7751 | |
7c941438 | 7752 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7753 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7754 | INIT_LIST_HEAD(&init_task_group.children); |
7755 | ||
7c941438 | 7756 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7757 | |
4a6cc4bd JK |
7758 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7759 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7760 | __alignof__(unsigned long)); | |
7761 | #endif | |
0a945022 | 7762 | for_each_possible_cpu(i) { |
70b97a7f | 7763 | struct rq *rq; |
1da177e4 LT |
7764 | |
7765 | rq = cpu_rq(i); | |
05fa785c | 7766 | raw_spin_lock_init(&rq->lock); |
7897986b | 7767 | rq->nr_running = 0; |
dce48a84 TG |
7768 | rq->calc_load_active = 0; |
7769 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7770 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7771 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7772 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7773 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7774 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7775 | #ifdef CONFIG_CGROUP_SCHED |
7776 | /* | |
7777 | * How much cpu bandwidth does init_task_group get? | |
7778 | * | |
7779 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7780 | * gets 100% of the cpu resources in the system. This overall | |
7781 | * system cpu resource is divided among the tasks of | |
7782 | * init_task_group and its child task-groups in a fair manner, | |
7783 | * based on each entity's (task or task-group's) weight | |
7784 | * (se->load.weight). | |
7785 | * | |
7786 | * In other words, if init_task_group has 10 tasks of weight | |
7787 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7788 | * then A0's share of the cpu resource is: | |
7789 | * | |
0d905bca | 7790 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7791 | * |
7792 | * We achieve this by letting init_task_group's tasks sit | |
7793 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7794 | */ | |
ec7dc8ac | 7795 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7796 | #endif |
354d60c2 DG |
7797 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7798 | ||
7799 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7800 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7801 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7802 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7803 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7804 | #endif |
dd41f596 | 7805 | #endif |
1da177e4 | 7806 | |
dd41f596 IM |
7807 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7808 | rq->cpu_load[j] = 0; | |
1da177e4 | 7809 | #ifdef CONFIG_SMP |
41c7ce9a | 7810 | rq->sd = NULL; |
57d885fe | 7811 | rq->rd = NULL; |
3f029d3c | 7812 | rq->post_schedule = 0; |
1da177e4 | 7813 | rq->active_balance = 0; |
dd41f596 | 7814 | rq->next_balance = jiffies; |
1da177e4 | 7815 | rq->push_cpu = 0; |
0a2966b4 | 7816 | rq->cpu = i; |
1f11eb6a | 7817 | rq->online = 0; |
1da177e4 | 7818 | rq->migration_thread = NULL; |
eae0c9df MG |
7819 | rq->idle_stamp = 0; |
7820 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
1da177e4 | 7821 | INIT_LIST_HEAD(&rq->migration_queue); |
dc938520 | 7822 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7823 | #endif |
8f4d37ec | 7824 | init_rq_hrtick(rq); |
1da177e4 | 7825 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7826 | } |
7827 | ||
2dd73a4f | 7828 | set_load_weight(&init_task); |
b50f60ce | 7829 | |
e107be36 AK |
7830 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7831 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7832 | #endif | |
7833 | ||
c9819f45 | 7834 | #ifdef CONFIG_SMP |
962cf36c | 7835 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7836 | #endif |
7837 | ||
b50f60ce | 7838 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7839 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7840 | #endif |
7841 | ||
1da177e4 LT |
7842 | /* |
7843 | * The boot idle thread does lazy MMU switching as well: | |
7844 | */ | |
7845 | atomic_inc(&init_mm.mm_count); | |
7846 | enter_lazy_tlb(&init_mm, current); | |
7847 | ||
7848 | /* | |
7849 | * Make us the idle thread. Technically, schedule() should not be | |
7850 | * called from this thread, however somewhere below it might be, | |
7851 | * but because we are the idle thread, we just pick up running again | |
7852 | * when this runqueue becomes "idle". | |
7853 | */ | |
7854 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7855 | |
7856 | calc_load_update = jiffies + LOAD_FREQ; | |
7857 | ||
dd41f596 IM |
7858 | /* |
7859 | * During early bootup we pretend to be a normal task: | |
7860 | */ | |
7861 | current->sched_class = &fair_sched_class; | |
6892b75e | 7862 | |
6a7b3dc3 | 7863 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7864 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7865 | #ifdef CONFIG_SMP |
7d1e6a9b | 7866 | #ifdef CONFIG_NO_HZ |
49557e62 | 7867 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 7868 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 7869 | #endif |
bdddd296 RR |
7870 | /* May be allocated at isolcpus cmdline parse time */ |
7871 | if (cpu_isolated_map == NULL) | |
7872 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7873 | #endif /* SMP */ |
6a7b3dc3 | 7874 | |
cdd6c482 | 7875 | perf_event_init(); |
0d905bca | 7876 | |
6892b75e | 7877 | scheduler_running = 1; |
1da177e4 LT |
7878 | } |
7879 | ||
7880 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7881 | static inline int preempt_count_equals(int preempt_offset) |
7882 | { | |
234da7bc | 7883 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7884 | |
7885 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7886 | } | |
7887 | ||
d894837f | 7888 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7889 | { |
48f24c4d | 7890 | #ifdef in_atomic |
1da177e4 LT |
7891 | static unsigned long prev_jiffy; /* ratelimiting */ |
7892 | ||
e4aafea2 FW |
7893 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7894 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7895 | return; |
7896 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7897 | return; | |
7898 | prev_jiffy = jiffies; | |
7899 | ||
3df0fc5b PZ |
7900 | printk(KERN_ERR |
7901 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7902 | file, line); | |
7903 | printk(KERN_ERR | |
7904 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7905 | in_atomic(), irqs_disabled(), | |
7906 | current->pid, current->comm); | |
aef745fc IM |
7907 | |
7908 | debug_show_held_locks(current); | |
7909 | if (irqs_disabled()) | |
7910 | print_irqtrace_events(current); | |
7911 | dump_stack(); | |
1da177e4 LT |
7912 | #endif |
7913 | } | |
7914 | EXPORT_SYMBOL(__might_sleep); | |
7915 | #endif | |
7916 | ||
7917 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7918 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7919 | { | |
7920 | int on_rq; | |
3e51f33f | 7921 | |
3a5e4dc1 AK |
7922 | update_rq_clock(rq); |
7923 | on_rq = p->se.on_rq; | |
7924 | if (on_rq) | |
7925 | deactivate_task(rq, p, 0); | |
7926 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7927 | if (on_rq) { | |
7928 | activate_task(rq, p, 0); | |
7929 | resched_task(rq->curr); | |
7930 | } | |
7931 | } | |
7932 | ||
1da177e4 LT |
7933 | void normalize_rt_tasks(void) |
7934 | { | |
a0f98a1c | 7935 | struct task_struct *g, *p; |
1da177e4 | 7936 | unsigned long flags; |
70b97a7f | 7937 | struct rq *rq; |
1da177e4 | 7938 | |
4cf5d77a | 7939 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7940 | do_each_thread(g, p) { |
178be793 IM |
7941 | /* |
7942 | * Only normalize user tasks: | |
7943 | */ | |
7944 | if (!p->mm) | |
7945 | continue; | |
7946 | ||
6cfb0d5d | 7947 | p->se.exec_start = 0; |
6cfb0d5d | 7948 | #ifdef CONFIG_SCHEDSTATS |
dd41f596 | 7949 | p->se.wait_start = 0; |
dd41f596 | 7950 | p->se.sleep_start = 0; |
dd41f596 | 7951 | p->se.block_start = 0; |
6cfb0d5d | 7952 | #endif |
dd41f596 IM |
7953 | |
7954 | if (!rt_task(p)) { | |
7955 | /* | |
7956 | * Renice negative nice level userspace | |
7957 | * tasks back to 0: | |
7958 | */ | |
7959 | if (TASK_NICE(p) < 0 && p->mm) | |
7960 | set_user_nice(p, 0); | |
1da177e4 | 7961 | continue; |
dd41f596 | 7962 | } |
1da177e4 | 7963 | |
1d615482 | 7964 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7965 | rq = __task_rq_lock(p); |
1da177e4 | 7966 | |
178be793 | 7967 | normalize_task(rq, p); |
3a5e4dc1 | 7968 | |
b29739f9 | 7969 | __task_rq_unlock(rq); |
1d615482 | 7970 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7971 | } while_each_thread(g, p); |
7972 | ||
4cf5d77a | 7973 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7974 | } |
7975 | ||
7976 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7977 | |
7978 | #ifdef CONFIG_IA64 | |
7979 | /* | |
7980 | * These functions are only useful for the IA64 MCA handling. | |
7981 | * | |
7982 | * They can only be called when the whole system has been | |
7983 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7984 | * activity can take place. Using them for anything else would | |
7985 | * be a serious bug, and as a result, they aren't even visible | |
7986 | * under any other configuration. | |
7987 | */ | |
7988 | ||
7989 | /** | |
7990 | * curr_task - return the current task for a given cpu. | |
7991 | * @cpu: the processor in question. | |
7992 | * | |
7993 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7994 | */ | |
36c8b586 | 7995 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7996 | { |
7997 | return cpu_curr(cpu); | |
7998 | } | |
7999 | ||
8000 | /** | |
8001 | * set_curr_task - set the current task for a given cpu. | |
8002 | * @cpu: the processor in question. | |
8003 | * @p: the task pointer to set. | |
8004 | * | |
8005 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8006 | * are serviced on a separate stack. It allows the architecture to switch the |
8007 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8008 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8009 | * and caller must save the original value of the current task (see | |
8010 | * curr_task() above) and restore that value before reenabling interrupts and | |
8011 | * re-starting the system. | |
8012 | * | |
8013 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8014 | */ | |
36c8b586 | 8015 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8016 | { |
8017 | cpu_curr(cpu) = p; | |
8018 | } | |
8019 | ||
8020 | #endif | |
29f59db3 | 8021 | |
bccbe08a PZ |
8022 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8023 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8024 | { |
8025 | int i; | |
8026 | ||
8027 | for_each_possible_cpu(i) { | |
8028 | if (tg->cfs_rq) | |
8029 | kfree(tg->cfs_rq[i]); | |
8030 | if (tg->se) | |
8031 | kfree(tg->se[i]); | |
6f505b16 PZ |
8032 | } |
8033 | ||
8034 | kfree(tg->cfs_rq); | |
8035 | kfree(tg->se); | |
6f505b16 PZ |
8036 | } |
8037 | ||
ec7dc8ac DG |
8038 | static |
8039 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8040 | { |
29f59db3 | 8041 | struct cfs_rq *cfs_rq; |
eab17229 | 8042 | struct sched_entity *se; |
9b5b7751 | 8043 | struct rq *rq; |
29f59db3 SV |
8044 | int i; |
8045 | ||
434d53b0 | 8046 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8047 | if (!tg->cfs_rq) |
8048 | goto err; | |
434d53b0 | 8049 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8050 | if (!tg->se) |
8051 | goto err; | |
052f1dc7 PZ |
8052 | |
8053 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8054 | |
8055 | for_each_possible_cpu(i) { | |
9b5b7751 | 8056 | rq = cpu_rq(i); |
29f59db3 | 8057 | |
eab17229 LZ |
8058 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8059 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8060 | if (!cfs_rq) |
8061 | goto err; | |
8062 | ||
eab17229 LZ |
8063 | se = kzalloc_node(sizeof(struct sched_entity), |
8064 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8065 | if (!se) |
dfc12eb2 | 8066 | goto err_free_rq; |
29f59db3 | 8067 | |
eab17229 | 8068 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
8069 | } |
8070 | ||
8071 | return 1; | |
8072 | ||
dfc12eb2 PC |
8073 | err_free_rq: |
8074 | kfree(cfs_rq); | |
bccbe08a PZ |
8075 | err: |
8076 | return 0; | |
8077 | } | |
8078 | ||
8079 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8080 | { | |
8081 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8082 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8083 | } | |
8084 | ||
8085 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8086 | { | |
8087 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8088 | } | |
6d6bc0ad | 8089 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8090 | static inline void free_fair_sched_group(struct task_group *tg) |
8091 | { | |
8092 | } | |
8093 | ||
ec7dc8ac DG |
8094 | static inline |
8095 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8096 | { |
8097 | return 1; | |
8098 | } | |
8099 | ||
8100 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8101 | { | |
8102 | } | |
8103 | ||
8104 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8105 | { | |
8106 | } | |
6d6bc0ad | 8107 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8108 | |
8109 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8110 | static void free_rt_sched_group(struct task_group *tg) |
8111 | { | |
8112 | int i; | |
8113 | ||
d0b27fa7 PZ |
8114 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8115 | ||
bccbe08a PZ |
8116 | for_each_possible_cpu(i) { |
8117 | if (tg->rt_rq) | |
8118 | kfree(tg->rt_rq[i]); | |
8119 | if (tg->rt_se) | |
8120 | kfree(tg->rt_se[i]); | |
8121 | } | |
8122 | ||
8123 | kfree(tg->rt_rq); | |
8124 | kfree(tg->rt_se); | |
8125 | } | |
8126 | ||
ec7dc8ac DG |
8127 | static |
8128 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8129 | { |
8130 | struct rt_rq *rt_rq; | |
eab17229 | 8131 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8132 | struct rq *rq; |
8133 | int i; | |
8134 | ||
434d53b0 | 8135 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8136 | if (!tg->rt_rq) |
8137 | goto err; | |
434d53b0 | 8138 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8139 | if (!tg->rt_se) |
8140 | goto err; | |
8141 | ||
d0b27fa7 PZ |
8142 | init_rt_bandwidth(&tg->rt_bandwidth, |
8143 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8144 | |
8145 | for_each_possible_cpu(i) { | |
8146 | rq = cpu_rq(i); | |
8147 | ||
eab17229 LZ |
8148 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8149 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8150 | if (!rt_rq) |
8151 | goto err; | |
29f59db3 | 8152 | |
eab17229 LZ |
8153 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8154 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8155 | if (!rt_se) |
dfc12eb2 | 8156 | goto err_free_rq; |
29f59db3 | 8157 | |
eab17229 | 8158 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8159 | } |
8160 | ||
bccbe08a PZ |
8161 | return 1; |
8162 | ||
dfc12eb2 PC |
8163 | err_free_rq: |
8164 | kfree(rt_rq); | |
bccbe08a PZ |
8165 | err: |
8166 | return 0; | |
8167 | } | |
8168 | ||
8169 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8170 | { | |
8171 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8172 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8173 | } | |
8174 | ||
8175 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8176 | { | |
8177 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8178 | } | |
6d6bc0ad | 8179 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8180 | static inline void free_rt_sched_group(struct task_group *tg) |
8181 | { | |
8182 | } | |
8183 | ||
ec7dc8ac DG |
8184 | static inline |
8185 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8186 | { |
8187 | return 1; | |
8188 | } | |
8189 | ||
8190 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8191 | { | |
8192 | } | |
8193 | ||
8194 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8195 | { | |
8196 | } | |
6d6bc0ad | 8197 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8198 | |
7c941438 | 8199 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8200 | static void free_sched_group(struct task_group *tg) |
8201 | { | |
8202 | free_fair_sched_group(tg); | |
8203 | free_rt_sched_group(tg); | |
8204 | kfree(tg); | |
8205 | } | |
8206 | ||
8207 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8208 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8209 | { |
8210 | struct task_group *tg; | |
8211 | unsigned long flags; | |
8212 | int i; | |
8213 | ||
8214 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8215 | if (!tg) | |
8216 | return ERR_PTR(-ENOMEM); | |
8217 | ||
ec7dc8ac | 8218 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8219 | goto err; |
8220 | ||
ec7dc8ac | 8221 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8222 | goto err; |
8223 | ||
8ed36996 | 8224 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8225 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8226 | register_fair_sched_group(tg, i); |
8227 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8228 | } |
6f505b16 | 8229 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8230 | |
8231 | WARN_ON(!parent); /* root should already exist */ | |
8232 | ||
8233 | tg->parent = parent; | |
f473aa5e | 8234 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8235 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8236 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8237 | |
9b5b7751 | 8238 | return tg; |
29f59db3 SV |
8239 | |
8240 | err: | |
6f505b16 | 8241 | free_sched_group(tg); |
29f59db3 SV |
8242 | return ERR_PTR(-ENOMEM); |
8243 | } | |
8244 | ||
9b5b7751 | 8245 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8246 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8247 | { |
29f59db3 | 8248 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8249 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8250 | } |
8251 | ||
9b5b7751 | 8252 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8253 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8254 | { |
8ed36996 | 8255 | unsigned long flags; |
9b5b7751 | 8256 | int i; |
29f59db3 | 8257 | |
8ed36996 | 8258 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8259 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8260 | unregister_fair_sched_group(tg, i); |
8261 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8262 | } |
6f505b16 | 8263 | list_del_rcu(&tg->list); |
f473aa5e | 8264 | list_del_rcu(&tg->siblings); |
8ed36996 | 8265 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8266 | |
9b5b7751 | 8267 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8268 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8269 | } |
8270 | ||
9b5b7751 | 8271 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8272 | * The caller of this function should have put the task in its new group |
8273 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8274 | * reflect its new group. | |
9b5b7751 SV |
8275 | */ |
8276 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8277 | { |
8278 | int on_rq, running; | |
8279 | unsigned long flags; | |
8280 | struct rq *rq; | |
8281 | ||
8282 | rq = task_rq_lock(tsk, &flags); | |
8283 | ||
29f59db3 SV |
8284 | update_rq_clock(rq); |
8285 | ||
051a1d1a | 8286 | running = task_current(rq, tsk); |
29f59db3 SV |
8287 | on_rq = tsk->se.on_rq; |
8288 | ||
0e1f3483 | 8289 | if (on_rq) |
29f59db3 | 8290 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8291 | if (unlikely(running)) |
8292 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8293 | |
6f505b16 | 8294 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8295 | |
810b3817 PZ |
8296 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8297 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8298 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8299 | #endif |
8300 | ||
0e1f3483 HS |
8301 | if (unlikely(running)) |
8302 | tsk->sched_class->set_curr_task(rq); | |
8303 | if (on_rq) | |
ea87bb78 | 8304 | enqueue_task(rq, tsk, 0, false); |
29f59db3 | 8305 | |
29f59db3 SV |
8306 | task_rq_unlock(rq, &flags); |
8307 | } | |
7c941438 | 8308 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8309 | |
052f1dc7 | 8310 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8311 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8312 | { |
8313 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8314 | int on_rq; |
8315 | ||
29f59db3 | 8316 | on_rq = se->on_rq; |
62fb1851 | 8317 | if (on_rq) |
29f59db3 SV |
8318 | dequeue_entity(cfs_rq, se, 0); |
8319 | ||
8320 | se->load.weight = shares; | |
e05510d0 | 8321 | se->load.inv_weight = 0; |
29f59db3 | 8322 | |
62fb1851 | 8323 | if (on_rq) |
29f59db3 | 8324 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8325 | } |
62fb1851 | 8326 | |
c09595f6 PZ |
8327 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8328 | { | |
8329 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8330 | struct rq *rq = cfs_rq->rq; | |
8331 | unsigned long flags; | |
8332 | ||
05fa785c | 8333 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8334 | __set_se_shares(se, shares); |
05fa785c | 8335 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8336 | } |
8337 | ||
8ed36996 PZ |
8338 | static DEFINE_MUTEX(shares_mutex); |
8339 | ||
4cf86d77 | 8340 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8341 | { |
8342 | int i; | |
8ed36996 | 8343 | unsigned long flags; |
c61935fd | 8344 | |
ec7dc8ac DG |
8345 | /* |
8346 | * We can't change the weight of the root cgroup. | |
8347 | */ | |
8348 | if (!tg->se[0]) | |
8349 | return -EINVAL; | |
8350 | ||
18d95a28 PZ |
8351 | if (shares < MIN_SHARES) |
8352 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8353 | else if (shares > MAX_SHARES) |
8354 | shares = MAX_SHARES; | |
62fb1851 | 8355 | |
8ed36996 | 8356 | mutex_lock(&shares_mutex); |
9b5b7751 | 8357 | if (tg->shares == shares) |
5cb350ba | 8358 | goto done; |
29f59db3 | 8359 | |
8ed36996 | 8360 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8361 | for_each_possible_cpu(i) |
8362 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8363 | list_del_rcu(&tg->siblings); |
8ed36996 | 8364 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8365 | |
8366 | /* wait for any ongoing reference to this group to finish */ | |
8367 | synchronize_sched(); | |
8368 | ||
8369 | /* | |
8370 | * Now we are free to modify the group's share on each cpu | |
8371 | * w/o tripping rebalance_share or load_balance_fair. | |
8372 | */ | |
9b5b7751 | 8373 | tg->shares = shares; |
c09595f6 PZ |
8374 | for_each_possible_cpu(i) { |
8375 | /* | |
8376 | * force a rebalance | |
8377 | */ | |
8378 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8379 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8380 | } |
29f59db3 | 8381 | |
6b2d7700 SV |
8382 | /* |
8383 | * Enable load balance activity on this group, by inserting it back on | |
8384 | * each cpu's rq->leaf_cfs_rq_list. | |
8385 | */ | |
8ed36996 | 8386 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8387 | for_each_possible_cpu(i) |
8388 | register_fair_sched_group(tg, i); | |
f473aa5e | 8389 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8390 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8391 | done: |
8ed36996 | 8392 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8393 | return 0; |
29f59db3 SV |
8394 | } |
8395 | ||
5cb350ba DG |
8396 | unsigned long sched_group_shares(struct task_group *tg) |
8397 | { | |
8398 | return tg->shares; | |
8399 | } | |
052f1dc7 | 8400 | #endif |
5cb350ba | 8401 | |
052f1dc7 | 8402 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8403 | /* |
9f0c1e56 | 8404 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8405 | */ |
9f0c1e56 PZ |
8406 | static DEFINE_MUTEX(rt_constraints_mutex); |
8407 | ||
8408 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8409 | { | |
8410 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8411 | return 1ULL << 20; |
9f0c1e56 | 8412 | |
9a7e0b18 | 8413 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8414 | } |
8415 | ||
9a7e0b18 PZ |
8416 | /* Must be called with tasklist_lock held */ |
8417 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8418 | { |
9a7e0b18 | 8419 | struct task_struct *g, *p; |
b40b2e8e | 8420 | |
9a7e0b18 PZ |
8421 | do_each_thread(g, p) { |
8422 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8423 | return 1; | |
8424 | } while_each_thread(g, p); | |
b40b2e8e | 8425 | |
9a7e0b18 PZ |
8426 | return 0; |
8427 | } | |
b40b2e8e | 8428 | |
9a7e0b18 PZ |
8429 | struct rt_schedulable_data { |
8430 | struct task_group *tg; | |
8431 | u64 rt_period; | |
8432 | u64 rt_runtime; | |
8433 | }; | |
b40b2e8e | 8434 | |
9a7e0b18 PZ |
8435 | static int tg_schedulable(struct task_group *tg, void *data) |
8436 | { | |
8437 | struct rt_schedulable_data *d = data; | |
8438 | struct task_group *child; | |
8439 | unsigned long total, sum = 0; | |
8440 | u64 period, runtime; | |
b40b2e8e | 8441 | |
9a7e0b18 PZ |
8442 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8443 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8444 | |
9a7e0b18 PZ |
8445 | if (tg == d->tg) { |
8446 | period = d->rt_period; | |
8447 | runtime = d->rt_runtime; | |
b40b2e8e | 8448 | } |
b40b2e8e | 8449 | |
4653f803 PZ |
8450 | /* |
8451 | * Cannot have more runtime than the period. | |
8452 | */ | |
8453 | if (runtime > period && runtime != RUNTIME_INF) | |
8454 | return -EINVAL; | |
6f505b16 | 8455 | |
4653f803 PZ |
8456 | /* |
8457 | * Ensure we don't starve existing RT tasks. | |
8458 | */ | |
9a7e0b18 PZ |
8459 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8460 | return -EBUSY; | |
6f505b16 | 8461 | |
9a7e0b18 | 8462 | total = to_ratio(period, runtime); |
6f505b16 | 8463 | |
4653f803 PZ |
8464 | /* |
8465 | * Nobody can have more than the global setting allows. | |
8466 | */ | |
8467 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8468 | return -EINVAL; | |
6f505b16 | 8469 | |
4653f803 PZ |
8470 | /* |
8471 | * The sum of our children's runtime should not exceed our own. | |
8472 | */ | |
9a7e0b18 PZ |
8473 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8474 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8475 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8476 | |
9a7e0b18 PZ |
8477 | if (child == d->tg) { |
8478 | period = d->rt_period; | |
8479 | runtime = d->rt_runtime; | |
8480 | } | |
6f505b16 | 8481 | |
9a7e0b18 | 8482 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8483 | } |
6f505b16 | 8484 | |
9a7e0b18 PZ |
8485 | if (sum > total) |
8486 | return -EINVAL; | |
8487 | ||
8488 | return 0; | |
6f505b16 PZ |
8489 | } |
8490 | ||
9a7e0b18 | 8491 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8492 | { |
9a7e0b18 PZ |
8493 | struct rt_schedulable_data data = { |
8494 | .tg = tg, | |
8495 | .rt_period = period, | |
8496 | .rt_runtime = runtime, | |
8497 | }; | |
8498 | ||
8499 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8500 | } |
8501 | ||
d0b27fa7 PZ |
8502 | static int tg_set_bandwidth(struct task_group *tg, |
8503 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8504 | { |
ac086bc2 | 8505 | int i, err = 0; |
9f0c1e56 | 8506 | |
9f0c1e56 | 8507 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8508 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8509 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8510 | if (err) | |
9f0c1e56 | 8511 | goto unlock; |
ac086bc2 | 8512 | |
0986b11b | 8513 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8514 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8515 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8516 | |
8517 | for_each_possible_cpu(i) { | |
8518 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8519 | ||
0986b11b | 8520 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8521 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8522 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8523 | } |
0986b11b | 8524 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8525 | unlock: |
521f1a24 | 8526 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8527 | mutex_unlock(&rt_constraints_mutex); |
8528 | ||
8529 | return err; | |
6f505b16 PZ |
8530 | } |
8531 | ||
d0b27fa7 PZ |
8532 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8533 | { | |
8534 | u64 rt_runtime, rt_period; | |
8535 | ||
8536 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8537 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8538 | if (rt_runtime_us < 0) | |
8539 | rt_runtime = RUNTIME_INF; | |
8540 | ||
8541 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8542 | } | |
8543 | ||
9f0c1e56 PZ |
8544 | long sched_group_rt_runtime(struct task_group *tg) |
8545 | { | |
8546 | u64 rt_runtime_us; | |
8547 | ||
d0b27fa7 | 8548 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8549 | return -1; |
8550 | ||
d0b27fa7 | 8551 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8552 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8553 | return rt_runtime_us; | |
8554 | } | |
d0b27fa7 PZ |
8555 | |
8556 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8557 | { | |
8558 | u64 rt_runtime, rt_period; | |
8559 | ||
8560 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8561 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8562 | ||
619b0488 R |
8563 | if (rt_period == 0) |
8564 | return -EINVAL; | |
8565 | ||
d0b27fa7 PZ |
8566 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8567 | } | |
8568 | ||
8569 | long sched_group_rt_period(struct task_group *tg) | |
8570 | { | |
8571 | u64 rt_period_us; | |
8572 | ||
8573 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8574 | do_div(rt_period_us, NSEC_PER_USEC); | |
8575 | return rt_period_us; | |
8576 | } | |
8577 | ||
8578 | static int sched_rt_global_constraints(void) | |
8579 | { | |
4653f803 | 8580 | u64 runtime, period; |
d0b27fa7 PZ |
8581 | int ret = 0; |
8582 | ||
ec5d4989 HS |
8583 | if (sysctl_sched_rt_period <= 0) |
8584 | return -EINVAL; | |
8585 | ||
4653f803 PZ |
8586 | runtime = global_rt_runtime(); |
8587 | period = global_rt_period(); | |
8588 | ||
8589 | /* | |
8590 | * Sanity check on the sysctl variables. | |
8591 | */ | |
8592 | if (runtime > period && runtime != RUNTIME_INF) | |
8593 | return -EINVAL; | |
10b612f4 | 8594 | |
d0b27fa7 | 8595 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8596 | read_lock(&tasklist_lock); |
4653f803 | 8597 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8598 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8599 | mutex_unlock(&rt_constraints_mutex); |
8600 | ||
8601 | return ret; | |
8602 | } | |
54e99124 DG |
8603 | |
8604 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8605 | { | |
8606 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8607 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8608 | return 0; | |
8609 | ||
8610 | return 1; | |
8611 | } | |
8612 | ||
6d6bc0ad | 8613 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8614 | static int sched_rt_global_constraints(void) |
8615 | { | |
ac086bc2 PZ |
8616 | unsigned long flags; |
8617 | int i; | |
8618 | ||
ec5d4989 HS |
8619 | if (sysctl_sched_rt_period <= 0) |
8620 | return -EINVAL; | |
8621 | ||
60aa605d PZ |
8622 | /* |
8623 | * There's always some RT tasks in the root group | |
8624 | * -- migration, kstopmachine etc.. | |
8625 | */ | |
8626 | if (sysctl_sched_rt_runtime == 0) | |
8627 | return -EBUSY; | |
8628 | ||
0986b11b | 8629 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8630 | for_each_possible_cpu(i) { |
8631 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8632 | ||
0986b11b | 8633 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8634 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8635 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8636 | } |
0986b11b | 8637 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8638 | |
d0b27fa7 PZ |
8639 | return 0; |
8640 | } | |
6d6bc0ad | 8641 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8642 | |
8643 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8644 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8645 | loff_t *ppos) |
8646 | { | |
8647 | int ret; | |
8648 | int old_period, old_runtime; | |
8649 | static DEFINE_MUTEX(mutex); | |
8650 | ||
8651 | mutex_lock(&mutex); | |
8652 | old_period = sysctl_sched_rt_period; | |
8653 | old_runtime = sysctl_sched_rt_runtime; | |
8654 | ||
8d65af78 | 8655 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8656 | |
8657 | if (!ret && write) { | |
8658 | ret = sched_rt_global_constraints(); | |
8659 | if (ret) { | |
8660 | sysctl_sched_rt_period = old_period; | |
8661 | sysctl_sched_rt_runtime = old_runtime; | |
8662 | } else { | |
8663 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8664 | def_rt_bandwidth.rt_period = | |
8665 | ns_to_ktime(global_rt_period()); | |
8666 | } | |
8667 | } | |
8668 | mutex_unlock(&mutex); | |
8669 | ||
8670 | return ret; | |
8671 | } | |
68318b8e | 8672 | |
052f1dc7 | 8673 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8674 | |
8675 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8676 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8677 | { |
2b01dfe3 PM |
8678 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8679 | struct task_group, css); | |
68318b8e SV |
8680 | } |
8681 | ||
8682 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8683 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8684 | { |
ec7dc8ac | 8685 | struct task_group *tg, *parent; |
68318b8e | 8686 | |
2b01dfe3 | 8687 | if (!cgrp->parent) { |
68318b8e | 8688 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8689 | return &init_task_group.css; |
8690 | } | |
8691 | ||
ec7dc8ac DG |
8692 | parent = cgroup_tg(cgrp->parent); |
8693 | tg = sched_create_group(parent); | |
68318b8e SV |
8694 | if (IS_ERR(tg)) |
8695 | return ERR_PTR(-ENOMEM); | |
8696 | ||
68318b8e SV |
8697 | return &tg->css; |
8698 | } | |
8699 | ||
41a2d6cf IM |
8700 | static void |
8701 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8702 | { |
2b01dfe3 | 8703 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8704 | |
8705 | sched_destroy_group(tg); | |
8706 | } | |
8707 | ||
41a2d6cf | 8708 | static int |
be367d09 | 8709 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8710 | { |
b68aa230 | 8711 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8712 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8713 | return -EINVAL; |
8714 | #else | |
68318b8e SV |
8715 | /* We don't support RT-tasks being in separate groups */ |
8716 | if (tsk->sched_class != &fair_sched_class) | |
8717 | return -EINVAL; | |
b68aa230 | 8718 | #endif |
be367d09 BB |
8719 | return 0; |
8720 | } | |
68318b8e | 8721 | |
be367d09 BB |
8722 | static int |
8723 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8724 | struct task_struct *tsk, bool threadgroup) | |
8725 | { | |
8726 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8727 | if (retval) | |
8728 | return retval; | |
8729 | if (threadgroup) { | |
8730 | struct task_struct *c; | |
8731 | rcu_read_lock(); | |
8732 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8733 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8734 | if (retval) { | |
8735 | rcu_read_unlock(); | |
8736 | return retval; | |
8737 | } | |
8738 | } | |
8739 | rcu_read_unlock(); | |
8740 | } | |
68318b8e SV |
8741 | return 0; |
8742 | } | |
8743 | ||
8744 | static void | |
2b01dfe3 | 8745 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8746 | struct cgroup *old_cont, struct task_struct *tsk, |
8747 | bool threadgroup) | |
68318b8e SV |
8748 | { |
8749 | sched_move_task(tsk); | |
be367d09 BB |
8750 | if (threadgroup) { |
8751 | struct task_struct *c; | |
8752 | rcu_read_lock(); | |
8753 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8754 | sched_move_task(c); | |
8755 | } | |
8756 | rcu_read_unlock(); | |
8757 | } | |
68318b8e SV |
8758 | } |
8759 | ||
052f1dc7 | 8760 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8761 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8762 | u64 shareval) |
68318b8e | 8763 | { |
2b01dfe3 | 8764 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8765 | } |
8766 | ||
f4c753b7 | 8767 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8768 | { |
2b01dfe3 | 8769 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8770 | |
8771 | return (u64) tg->shares; | |
8772 | } | |
6d6bc0ad | 8773 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8774 | |
052f1dc7 | 8775 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8776 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8777 | s64 val) |
6f505b16 | 8778 | { |
06ecb27c | 8779 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8780 | } |
8781 | ||
06ecb27c | 8782 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8783 | { |
06ecb27c | 8784 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8785 | } |
d0b27fa7 PZ |
8786 | |
8787 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8788 | u64 rt_period_us) | |
8789 | { | |
8790 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8791 | } | |
8792 | ||
8793 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8794 | { | |
8795 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8796 | } | |
6d6bc0ad | 8797 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8798 | |
fe5c7cc2 | 8799 | static struct cftype cpu_files[] = { |
052f1dc7 | 8800 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8801 | { |
8802 | .name = "shares", | |
f4c753b7 PM |
8803 | .read_u64 = cpu_shares_read_u64, |
8804 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8805 | }, |
052f1dc7 PZ |
8806 | #endif |
8807 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8808 | { |
9f0c1e56 | 8809 | .name = "rt_runtime_us", |
06ecb27c PM |
8810 | .read_s64 = cpu_rt_runtime_read, |
8811 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8812 | }, |
d0b27fa7 PZ |
8813 | { |
8814 | .name = "rt_period_us", | |
f4c753b7 PM |
8815 | .read_u64 = cpu_rt_period_read_uint, |
8816 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8817 | }, |
052f1dc7 | 8818 | #endif |
68318b8e SV |
8819 | }; |
8820 | ||
8821 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8822 | { | |
fe5c7cc2 | 8823 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8824 | } |
8825 | ||
8826 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8827 | .name = "cpu", |
8828 | .create = cpu_cgroup_create, | |
8829 | .destroy = cpu_cgroup_destroy, | |
8830 | .can_attach = cpu_cgroup_can_attach, | |
8831 | .attach = cpu_cgroup_attach, | |
8832 | .populate = cpu_cgroup_populate, | |
8833 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8834 | .early_init = 1, |
8835 | }; | |
8836 | ||
052f1dc7 | 8837 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8838 | |
8839 | #ifdef CONFIG_CGROUP_CPUACCT | |
8840 | ||
8841 | /* | |
8842 | * CPU accounting code for task groups. | |
8843 | * | |
8844 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8845 | * (balbir@in.ibm.com). | |
8846 | */ | |
8847 | ||
934352f2 | 8848 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8849 | struct cpuacct { |
8850 | struct cgroup_subsys_state css; | |
8851 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8852 | u64 __percpu *cpuusage; |
ef12fefa | 8853 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8854 | struct cpuacct *parent; |
d842de87 SV |
8855 | }; |
8856 | ||
8857 | struct cgroup_subsys cpuacct_subsys; | |
8858 | ||
8859 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8860 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8861 | { |
32cd756a | 8862 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8863 | struct cpuacct, css); |
8864 | } | |
8865 | ||
8866 | /* return cpu accounting group to which this task belongs */ | |
8867 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8868 | { | |
8869 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8870 | struct cpuacct, css); | |
8871 | } | |
8872 | ||
8873 | /* create a new cpu accounting group */ | |
8874 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8875 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8876 | { |
8877 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8878 | int i; |
d842de87 SV |
8879 | |
8880 | if (!ca) | |
ef12fefa | 8881 | goto out; |
d842de87 SV |
8882 | |
8883 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8884 | if (!ca->cpuusage) |
8885 | goto out_free_ca; | |
8886 | ||
8887 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8888 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8889 | goto out_free_counters; | |
d842de87 | 8890 | |
934352f2 BR |
8891 | if (cgrp->parent) |
8892 | ca->parent = cgroup_ca(cgrp->parent); | |
8893 | ||
d842de87 | 8894 | return &ca->css; |
ef12fefa BR |
8895 | |
8896 | out_free_counters: | |
8897 | while (--i >= 0) | |
8898 | percpu_counter_destroy(&ca->cpustat[i]); | |
8899 | free_percpu(ca->cpuusage); | |
8900 | out_free_ca: | |
8901 | kfree(ca); | |
8902 | out: | |
8903 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8904 | } |
8905 | ||
8906 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8907 | static void |
32cd756a | 8908 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8909 | { |
32cd756a | 8910 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8911 | int i; |
d842de87 | 8912 | |
ef12fefa BR |
8913 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8914 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8915 | free_percpu(ca->cpuusage); |
8916 | kfree(ca); | |
8917 | } | |
8918 | ||
720f5498 KC |
8919 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8920 | { | |
b36128c8 | 8921 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8922 | u64 data; |
8923 | ||
8924 | #ifndef CONFIG_64BIT | |
8925 | /* | |
8926 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8927 | */ | |
05fa785c | 8928 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8929 | data = *cpuusage; |
05fa785c | 8930 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8931 | #else |
8932 | data = *cpuusage; | |
8933 | #endif | |
8934 | ||
8935 | return data; | |
8936 | } | |
8937 | ||
8938 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8939 | { | |
b36128c8 | 8940 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8941 | |
8942 | #ifndef CONFIG_64BIT | |
8943 | /* | |
8944 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8945 | */ | |
05fa785c | 8946 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8947 | *cpuusage = val; |
05fa785c | 8948 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8949 | #else |
8950 | *cpuusage = val; | |
8951 | #endif | |
8952 | } | |
8953 | ||
d842de87 | 8954 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8955 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8956 | { |
32cd756a | 8957 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8958 | u64 totalcpuusage = 0; |
8959 | int i; | |
8960 | ||
720f5498 KC |
8961 | for_each_present_cpu(i) |
8962 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8963 | |
8964 | return totalcpuusage; | |
8965 | } | |
8966 | ||
0297b803 DG |
8967 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8968 | u64 reset) | |
8969 | { | |
8970 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8971 | int err = 0; | |
8972 | int i; | |
8973 | ||
8974 | if (reset) { | |
8975 | err = -EINVAL; | |
8976 | goto out; | |
8977 | } | |
8978 | ||
720f5498 KC |
8979 | for_each_present_cpu(i) |
8980 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8981 | |
0297b803 DG |
8982 | out: |
8983 | return err; | |
8984 | } | |
8985 | ||
e9515c3c KC |
8986 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8987 | struct seq_file *m) | |
8988 | { | |
8989 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8990 | u64 percpu; | |
8991 | int i; | |
8992 | ||
8993 | for_each_present_cpu(i) { | |
8994 | percpu = cpuacct_cpuusage_read(ca, i); | |
8995 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8996 | } | |
8997 | seq_printf(m, "\n"); | |
8998 | return 0; | |
8999 | } | |
9000 | ||
ef12fefa BR |
9001 | static const char *cpuacct_stat_desc[] = { |
9002 | [CPUACCT_STAT_USER] = "user", | |
9003 | [CPUACCT_STAT_SYSTEM] = "system", | |
9004 | }; | |
9005 | ||
9006 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9007 | struct cgroup_map_cb *cb) | |
9008 | { | |
9009 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9010 | int i; | |
9011 | ||
9012 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9013 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9014 | val = cputime64_to_clock_t(val); | |
9015 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9016 | } | |
9017 | return 0; | |
9018 | } | |
9019 | ||
d842de87 SV |
9020 | static struct cftype files[] = { |
9021 | { | |
9022 | .name = "usage", | |
f4c753b7 PM |
9023 | .read_u64 = cpuusage_read, |
9024 | .write_u64 = cpuusage_write, | |
d842de87 | 9025 | }, |
e9515c3c KC |
9026 | { |
9027 | .name = "usage_percpu", | |
9028 | .read_seq_string = cpuacct_percpu_seq_read, | |
9029 | }, | |
ef12fefa BR |
9030 | { |
9031 | .name = "stat", | |
9032 | .read_map = cpuacct_stats_show, | |
9033 | }, | |
d842de87 SV |
9034 | }; |
9035 | ||
32cd756a | 9036 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9037 | { |
32cd756a | 9038 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9039 | } |
9040 | ||
9041 | /* | |
9042 | * charge this task's execution time to its accounting group. | |
9043 | * | |
9044 | * called with rq->lock held. | |
9045 | */ | |
9046 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9047 | { | |
9048 | struct cpuacct *ca; | |
934352f2 | 9049 | int cpu; |
d842de87 | 9050 | |
c40c6f85 | 9051 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9052 | return; |
9053 | ||
934352f2 | 9054 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9055 | |
9056 | rcu_read_lock(); | |
9057 | ||
d842de87 | 9058 | ca = task_ca(tsk); |
d842de87 | 9059 | |
934352f2 | 9060 | for (; ca; ca = ca->parent) { |
b36128c8 | 9061 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9062 | *cpuusage += cputime; |
9063 | } | |
a18b83b7 BR |
9064 | |
9065 | rcu_read_unlock(); | |
d842de87 SV |
9066 | } |
9067 | ||
fa535a77 AB |
9068 | /* |
9069 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9070 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9071 | * percpu_counter_add with values large enough to always overflow the | |
9072 | * per cpu batch limit causing bad SMP scalability. | |
9073 | * | |
9074 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9075 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9076 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9077 | */ | |
9078 | #ifdef CONFIG_SMP | |
9079 | #define CPUACCT_BATCH \ | |
9080 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9081 | #else | |
9082 | #define CPUACCT_BATCH 0 | |
9083 | #endif | |
9084 | ||
ef12fefa BR |
9085 | /* |
9086 | * Charge the system/user time to the task's accounting group. | |
9087 | */ | |
9088 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9089 | enum cpuacct_stat_index idx, cputime_t val) | |
9090 | { | |
9091 | struct cpuacct *ca; | |
fa535a77 | 9092 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9093 | |
9094 | if (unlikely(!cpuacct_subsys.active)) | |
9095 | return; | |
9096 | ||
9097 | rcu_read_lock(); | |
9098 | ca = task_ca(tsk); | |
9099 | ||
9100 | do { | |
fa535a77 | 9101 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9102 | ca = ca->parent; |
9103 | } while (ca); | |
9104 | rcu_read_unlock(); | |
9105 | } | |
9106 | ||
d842de87 SV |
9107 | struct cgroup_subsys cpuacct_subsys = { |
9108 | .name = "cpuacct", | |
9109 | .create = cpuacct_create, | |
9110 | .destroy = cpuacct_destroy, | |
9111 | .populate = cpuacct_populate, | |
9112 | .subsys_id = cpuacct_subsys_id, | |
9113 | }; | |
9114 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
9115 | |
9116 | #ifndef CONFIG_SMP | |
9117 | ||
9118 | int rcu_expedited_torture_stats(char *page) | |
9119 | { | |
9120 | return 0; | |
9121 | } | |
9122 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9123 | ||
9124 | void synchronize_sched_expedited(void) | |
9125 | { | |
9126 | } | |
9127 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9128 | ||
9129 | #else /* #ifndef CONFIG_SMP */ | |
9130 | ||
9131 | static DEFINE_PER_CPU(struct migration_req, rcu_migration_req); | |
9132 | static DEFINE_MUTEX(rcu_sched_expedited_mutex); | |
9133 | ||
9134 | #define RCU_EXPEDITED_STATE_POST -2 | |
9135 | #define RCU_EXPEDITED_STATE_IDLE -1 | |
9136 | ||
9137 | static int rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
9138 | ||
9139 | int rcu_expedited_torture_stats(char *page) | |
9140 | { | |
9141 | int cnt = 0; | |
9142 | int cpu; | |
9143 | ||
9144 | cnt += sprintf(&page[cnt], "state: %d /", rcu_expedited_state); | |
9145 | for_each_online_cpu(cpu) { | |
9146 | cnt += sprintf(&page[cnt], " %d:%d", | |
9147 | cpu, per_cpu(rcu_migration_req, cpu).dest_cpu); | |
9148 | } | |
9149 | cnt += sprintf(&page[cnt], "\n"); | |
9150 | return cnt; | |
9151 | } | |
9152 | EXPORT_SYMBOL_GPL(rcu_expedited_torture_stats); | |
9153 | ||
9154 | static long synchronize_sched_expedited_count; | |
9155 | ||
9156 | /* | |
9157 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9158 | * approach to force grace period to end quickly. This consumes | |
9159 | * significant time on all CPUs, and is thus not recommended for | |
9160 | * any sort of common-case code. | |
9161 | * | |
9162 | * Note that it is illegal to call this function while holding any | |
9163 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9164 | * observe this restriction will result in deadlock. | |
9165 | */ | |
9166 | void synchronize_sched_expedited(void) | |
9167 | { | |
9168 | int cpu; | |
9169 | unsigned long flags; | |
9170 | bool need_full_sync = 0; | |
9171 | struct rq *rq; | |
9172 | struct migration_req *req; | |
9173 | long snap; | |
9174 | int trycount = 0; | |
9175 | ||
9176 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
9177 | snap = ACCESS_ONCE(synchronize_sched_expedited_count) + 1; | |
9178 | get_online_cpus(); | |
9179 | while (!mutex_trylock(&rcu_sched_expedited_mutex)) { | |
9180 | put_online_cpus(); | |
9181 | if (trycount++ < 10) | |
9182 | udelay(trycount * num_online_cpus()); | |
9183 | else { | |
9184 | synchronize_sched(); | |
9185 | return; | |
9186 | } | |
9187 | if (ACCESS_ONCE(synchronize_sched_expedited_count) - snap > 0) { | |
9188 | smp_mb(); /* ensure test happens before caller kfree */ | |
9189 | return; | |
9190 | } | |
9191 | get_online_cpus(); | |
9192 | } | |
9193 | rcu_expedited_state = RCU_EXPEDITED_STATE_POST; | |
9194 | for_each_online_cpu(cpu) { | |
9195 | rq = cpu_rq(cpu); | |
9196 | req = &per_cpu(rcu_migration_req, cpu); | |
9197 | init_completion(&req->done); | |
9198 | req->task = NULL; | |
9199 | req->dest_cpu = RCU_MIGRATION_NEED_QS; | |
05fa785c | 9200 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf | 9201 | list_add(&req->list, &rq->migration_queue); |
05fa785c | 9202 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9203 | wake_up_process(rq->migration_thread); |
9204 | } | |
9205 | for_each_online_cpu(cpu) { | |
9206 | rcu_expedited_state = cpu; | |
9207 | req = &per_cpu(rcu_migration_req, cpu); | |
9208 | rq = cpu_rq(cpu); | |
9209 | wait_for_completion(&req->done); | |
05fa785c | 9210 | raw_spin_lock_irqsave(&rq->lock, flags); |
03b042bf PM |
9211 | if (unlikely(req->dest_cpu == RCU_MIGRATION_MUST_SYNC)) |
9212 | need_full_sync = 1; | |
9213 | req->dest_cpu = RCU_MIGRATION_IDLE; | |
05fa785c | 9214 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
03b042bf PM |
9215 | } |
9216 | rcu_expedited_state = RCU_EXPEDITED_STATE_IDLE; | |
956539b7 | 9217 | synchronize_sched_expedited_count++; |
03b042bf PM |
9218 | mutex_unlock(&rcu_sched_expedited_mutex); |
9219 | put_online_cpus(); | |
9220 | if (need_full_sync) | |
9221 | synchronize_sched(); | |
9222 | } | |
9223 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9224 | ||
9225 | #endif /* #else #ifndef CONFIG_SMP */ |