]>
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> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
969c7921 | 60 | #include <linux/stop_machine.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 | 505 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 506 | u64 nohz_stamp; |
46cb4b7c SS |
507 | unsigned char in_nohz_recently; |
508 | #endif | |
a64692a3 MG |
509 | unsigned int skip_clock_update; |
510 | ||
d8016491 IM |
511 | /* capture load from *all* tasks on this cpu: */ |
512 | struct load_weight load; | |
6aa645ea IM |
513 | unsigned long nr_load_updates; |
514 | u64 nr_switches; | |
515 | ||
516 | struct cfs_rq cfs; | |
6f505b16 | 517 | struct rt_rq rt; |
6f505b16 | 518 | |
6aa645ea | 519 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
520 | /* list of leaf cfs_rq on this cpu: */ |
521 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
522 | #endif |
523 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 524 | struct list_head leaf_rt_rq_list; |
1da177e4 | 525 | #endif |
1da177e4 LT |
526 | |
527 | /* | |
528 | * This is part of a global counter where only the total sum | |
529 | * over all CPUs matters. A task can increase this counter on | |
530 | * one CPU and if it got migrated afterwards it may decrease | |
531 | * it on another CPU. Always updated under the runqueue lock: | |
532 | */ | |
533 | unsigned long nr_uninterruptible; | |
534 | ||
36c8b586 | 535 | struct task_struct *curr, *idle; |
c9819f45 | 536 | unsigned long next_balance; |
1da177e4 | 537 | struct mm_struct *prev_mm; |
6aa645ea | 538 | |
3e51f33f | 539 | u64 clock; |
6aa645ea | 540 | |
1da177e4 LT |
541 | atomic_t nr_iowait; |
542 | ||
543 | #ifdef CONFIG_SMP | |
0eab9146 | 544 | struct root_domain *rd; |
1da177e4 LT |
545 | struct sched_domain *sd; |
546 | ||
a0a522ce | 547 | unsigned char idle_at_tick; |
1da177e4 | 548 | /* For active balancing */ |
3f029d3c | 549 | int post_schedule; |
1da177e4 LT |
550 | int active_balance; |
551 | int push_cpu; | |
969c7921 | 552 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
553 | /* cpu of this runqueue: */ |
554 | int cpu; | |
1f11eb6a | 555 | int online; |
1da177e4 | 556 | |
a8a51d5e | 557 | unsigned long avg_load_per_task; |
1da177e4 | 558 | |
e9e9250b PZ |
559 | u64 rt_avg; |
560 | u64 age_stamp; | |
1b9508f6 MG |
561 | u64 idle_stamp; |
562 | u64 avg_idle; | |
1da177e4 LT |
563 | #endif |
564 | ||
dce48a84 TG |
565 | /* calc_load related fields */ |
566 | unsigned long calc_load_update; | |
567 | long calc_load_active; | |
568 | ||
8f4d37ec | 569 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
570 | #ifdef CONFIG_SMP |
571 | int hrtick_csd_pending; | |
572 | struct call_single_data hrtick_csd; | |
573 | #endif | |
8f4d37ec PZ |
574 | struct hrtimer hrtick_timer; |
575 | #endif | |
576 | ||
1da177e4 LT |
577 | #ifdef CONFIG_SCHEDSTATS |
578 | /* latency stats */ | |
579 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
580 | unsigned long long rq_cpu_time; |
581 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
582 | |
583 | /* sys_sched_yield() stats */ | |
480b9434 | 584 | unsigned int yld_count; |
1da177e4 LT |
585 | |
586 | /* schedule() stats */ | |
480b9434 KC |
587 | unsigned int sched_switch; |
588 | unsigned int sched_count; | |
589 | unsigned int sched_goidle; | |
1da177e4 LT |
590 | |
591 | /* try_to_wake_up() stats */ | |
480b9434 KC |
592 | unsigned int ttwu_count; |
593 | unsigned int ttwu_local; | |
b8efb561 IM |
594 | |
595 | /* BKL stats */ | |
480b9434 | 596 | unsigned int bkl_count; |
1da177e4 LT |
597 | #endif |
598 | }; | |
599 | ||
f34e3b61 | 600 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 601 | |
7d478721 PZ |
602 | static inline |
603 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 604 | { |
7d478721 | 605 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
a64692a3 MG |
606 | |
607 | /* | |
608 | * A queue event has occurred, and we're going to schedule. In | |
609 | * this case, we can save a useless back to back clock update. | |
610 | */ | |
611 | if (test_tsk_need_resched(p)) | |
612 | rq->skip_clock_update = 1; | |
dd41f596 IM |
613 | } |
614 | ||
0a2966b4 CL |
615 | static inline int cpu_of(struct rq *rq) |
616 | { | |
617 | #ifdef CONFIG_SMP | |
618 | return rq->cpu; | |
619 | #else | |
620 | return 0; | |
621 | #endif | |
622 | } | |
623 | ||
497f0ab3 | 624 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
625 | rcu_dereference_check((p), \ |
626 | rcu_read_lock_sched_held() || \ | |
627 | lockdep_is_held(&sched_domains_mutex)) | |
628 | ||
674311d5 NP |
629 | /* |
630 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 631 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
632 | * |
633 | * The domain tree of any CPU may only be accessed from within | |
634 | * preempt-disabled sections. | |
635 | */ | |
48f24c4d | 636 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 637 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
638 | |
639 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
640 | #define this_rq() (&__get_cpu_var(runqueues)) | |
641 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
642 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 643 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 644 | |
aa9c4c0f | 645 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 646 | { |
a64692a3 MG |
647 | if (!rq->skip_clock_update) |
648 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
3e51f33f PZ |
649 | } |
650 | ||
bf5c91ba IM |
651 | /* |
652 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
653 | */ | |
654 | #ifdef CONFIG_SCHED_DEBUG | |
655 | # define const_debug __read_mostly | |
656 | #else | |
657 | # define const_debug static const | |
658 | #endif | |
659 | ||
017730c1 IM |
660 | /** |
661 | * runqueue_is_locked | |
e17b38bf | 662 | * @cpu: the processor in question. |
017730c1 IM |
663 | * |
664 | * Returns true if the current cpu runqueue is locked. | |
665 | * This interface allows printk to be called with the runqueue lock | |
666 | * held and know whether or not it is OK to wake up the klogd. | |
667 | */ | |
89f19f04 | 668 | int runqueue_is_locked(int cpu) |
017730c1 | 669 | { |
05fa785c | 670 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
671 | } |
672 | ||
bf5c91ba IM |
673 | /* |
674 | * Debugging: various feature bits | |
675 | */ | |
f00b45c1 PZ |
676 | |
677 | #define SCHED_FEAT(name, enabled) \ | |
678 | __SCHED_FEAT_##name , | |
679 | ||
bf5c91ba | 680 | enum { |
f00b45c1 | 681 | #include "sched_features.h" |
bf5c91ba IM |
682 | }; |
683 | ||
f00b45c1 PZ |
684 | #undef SCHED_FEAT |
685 | ||
686 | #define SCHED_FEAT(name, enabled) \ | |
687 | (1UL << __SCHED_FEAT_##name) * enabled | | |
688 | ||
bf5c91ba | 689 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
690 | #include "sched_features.h" |
691 | 0; | |
692 | ||
693 | #undef SCHED_FEAT | |
694 | ||
695 | #ifdef CONFIG_SCHED_DEBUG | |
696 | #define SCHED_FEAT(name, enabled) \ | |
697 | #name , | |
698 | ||
983ed7a6 | 699 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
700 | #include "sched_features.h" |
701 | NULL | |
702 | }; | |
703 | ||
704 | #undef SCHED_FEAT | |
705 | ||
34f3a814 | 706 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 707 | { |
f00b45c1 PZ |
708 | int i; |
709 | ||
710 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
711 | if (!(sysctl_sched_features & (1UL << i))) |
712 | seq_puts(m, "NO_"); | |
713 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 714 | } |
34f3a814 | 715 | seq_puts(m, "\n"); |
f00b45c1 | 716 | |
34f3a814 | 717 | return 0; |
f00b45c1 PZ |
718 | } |
719 | ||
720 | static ssize_t | |
721 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
722 | size_t cnt, loff_t *ppos) | |
723 | { | |
724 | char buf[64]; | |
725 | char *cmp = buf; | |
726 | int neg = 0; | |
727 | int i; | |
728 | ||
729 | if (cnt > 63) | |
730 | cnt = 63; | |
731 | ||
732 | if (copy_from_user(&buf, ubuf, cnt)) | |
733 | return -EFAULT; | |
734 | ||
735 | buf[cnt] = 0; | |
736 | ||
c24b7c52 | 737 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
738 | neg = 1; |
739 | cmp += 3; | |
740 | } | |
741 | ||
742 | for (i = 0; sched_feat_names[i]; i++) { | |
743 | int len = strlen(sched_feat_names[i]); | |
744 | ||
745 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
746 | if (neg) | |
747 | sysctl_sched_features &= ~(1UL << i); | |
748 | else | |
749 | sysctl_sched_features |= (1UL << i); | |
750 | break; | |
751 | } | |
752 | } | |
753 | ||
754 | if (!sched_feat_names[i]) | |
755 | return -EINVAL; | |
756 | ||
42994724 | 757 | *ppos += cnt; |
f00b45c1 PZ |
758 | |
759 | return cnt; | |
760 | } | |
761 | ||
34f3a814 LZ |
762 | static int sched_feat_open(struct inode *inode, struct file *filp) |
763 | { | |
764 | return single_open(filp, sched_feat_show, NULL); | |
765 | } | |
766 | ||
828c0950 | 767 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
768 | .open = sched_feat_open, |
769 | .write = sched_feat_write, | |
770 | .read = seq_read, | |
771 | .llseek = seq_lseek, | |
772 | .release = single_release, | |
f00b45c1 PZ |
773 | }; |
774 | ||
775 | static __init int sched_init_debug(void) | |
776 | { | |
f00b45c1 PZ |
777 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
778 | &sched_feat_fops); | |
779 | ||
780 | return 0; | |
781 | } | |
782 | late_initcall(sched_init_debug); | |
783 | ||
784 | #endif | |
785 | ||
786 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 787 | |
b82d9fdd PZ |
788 | /* |
789 | * Number of tasks to iterate in a single balance run. | |
790 | * Limited because this is done with IRQs disabled. | |
791 | */ | |
792 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
793 | ||
2398f2c6 PZ |
794 | /* |
795 | * ratelimit for updating the group shares. | |
55cd5340 | 796 | * default: 0.25ms |
2398f2c6 | 797 | */ |
55cd5340 | 798 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 799 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 800 | |
ffda12a1 PZ |
801 | /* |
802 | * Inject some fuzzyness into changing the per-cpu group shares | |
803 | * this avoids remote rq-locks at the expense of fairness. | |
804 | * default: 4 | |
805 | */ | |
806 | unsigned int sysctl_sched_shares_thresh = 4; | |
807 | ||
e9e9250b PZ |
808 | /* |
809 | * period over which we average the RT time consumption, measured | |
810 | * in ms. | |
811 | * | |
812 | * default: 1s | |
813 | */ | |
814 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
815 | ||
fa85ae24 | 816 | /* |
9f0c1e56 | 817 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
818 | * default: 1s |
819 | */ | |
9f0c1e56 | 820 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 821 | |
6892b75e IM |
822 | static __read_mostly int scheduler_running; |
823 | ||
9f0c1e56 PZ |
824 | /* |
825 | * part of the period that we allow rt tasks to run in us. | |
826 | * default: 0.95s | |
827 | */ | |
828 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 829 | |
d0b27fa7 PZ |
830 | static inline u64 global_rt_period(void) |
831 | { | |
832 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
833 | } | |
834 | ||
835 | static inline u64 global_rt_runtime(void) | |
836 | { | |
e26873bb | 837 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
838 | return RUNTIME_INF; |
839 | ||
840 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
841 | } | |
fa85ae24 | 842 | |
1da177e4 | 843 | #ifndef prepare_arch_switch |
4866cde0 NP |
844 | # define prepare_arch_switch(next) do { } while (0) |
845 | #endif | |
846 | #ifndef finish_arch_switch | |
847 | # define finish_arch_switch(prev) do { } while (0) | |
848 | #endif | |
849 | ||
051a1d1a DA |
850 | static inline int task_current(struct rq *rq, struct task_struct *p) |
851 | { | |
852 | return rq->curr == p; | |
853 | } | |
854 | ||
4866cde0 | 855 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 856 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 857 | { |
051a1d1a | 858 | return task_current(rq, p); |
4866cde0 NP |
859 | } |
860 | ||
70b97a7f | 861 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
862 | { |
863 | } | |
864 | ||
70b97a7f | 865 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 866 | { |
da04c035 IM |
867 | #ifdef CONFIG_DEBUG_SPINLOCK |
868 | /* this is a valid case when another task releases the spinlock */ | |
869 | rq->lock.owner = current; | |
870 | #endif | |
8a25d5de IM |
871 | /* |
872 | * If we are tracking spinlock dependencies then we have to | |
873 | * fix up the runqueue lock - which gets 'carried over' from | |
874 | * prev into current: | |
875 | */ | |
876 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
877 | ||
05fa785c | 878 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
879 | } |
880 | ||
881 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 882 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
883 | { |
884 | #ifdef CONFIG_SMP | |
885 | return p->oncpu; | |
886 | #else | |
051a1d1a | 887 | return task_current(rq, p); |
4866cde0 NP |
888 | #endif |
889 | } | |
890 | ||
70b97a7f | 891 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
892 | { |
893 | #ifdef CONFIG_SMP | |
894 | /* | |
895 | * We can optimise this out completely for !SMP, because the | |
896 | * SMP rebalancing from interrupt is the only thing that cares | |
897 | * here. | |
898 | */ | |
899 | next->oncpu = 1; | |
900 | #endif | |
901 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 902 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 903 | #else |
05fa785c | 904 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
905 | #endif |
906 | } | |
907 | ||
70b97a7f | 908 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
909 | { |
910 | #ifdef CONFIG_SMP | |
911 | /* | |
912 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
913 | * We must ensure this doesn't happen until the switch is completely | |
914 | * finished. | |
915 | */ | |
916 | smp_wmb(); | |
917 | prev->oncpu = 0; | |
918 | #endif | |
919 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
920 | local_irq_enable(); | |
1da177e4 | 921 | #endif |
4866cde0 NP |
922 | } |
923 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 924 | |
0970d299 | 925 | /* |
65cc8e48 PZ |
926 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
927 | * against ttwu(). | |
0970d299 PZ |
928 | */ |
929 | static inline int task_is_waking(struct task_struct *p) | |
930 | { | |
0017d735 | 931 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
932 | } |
933 | ||
b29739f9 IM |
934 | /* |
935 | * __task_rq_lock - lock the runqueue a given task resides on. | |
936 | * Must be called interrupts disabled. | |
937 | */ | |
70b97a7f | 938 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
939 | __acquires(rq->lock) |
940 | { | |
0970d299 PZ |
941 | struct rq *rq; |
942 | ||
3a5c359a | 943 | for (;;) { |
0970d299 | 944 | rq = task_rq(p); |
05fa785c | 945 | raw_spin_lock(&rq->lock); |
65cc8e48 | 946 | if (likely(rq == task_rq(p))) |
3a5c359a | 947 | return rq; |
05fa785c | 948 | raw_spin_unlock(&rq->lock); |
b29739f9 | 949 | } |
b29739f9 IM |
950 | } |
951 | ||
1da177e4 LT |
952 | /* |
953 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 954 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
955 | * explicitly disabling preemption. |
956 | */ | |
70b97a7f | 957 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
958 | __acquires(rq->lock) |
959 | { | |
70b97a7f | 960 | struct rq *rq; |
1da177e4 | 961 | |
3a5c359a AK |
962 | for (;;) { |
963 | local_irq_save(*flags); | |
964 | rq = task_rq(p); | |
05fa785c | 965 | raw_spin_lock(&rq->lock); |
65cc8e48 | 966 | if (likely(rq == task_rq(p))) |
3a5c359a | 967 | return rq; |
05fa785c | 968 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 969 | } |
1da177e4 LT |
970 | } |
971 | ||
ad474cac ON |
972 | void task_rq_unlock_wait(struct task_struct *p) |
973 | { | |
974 | struct rq *rq = task_rq(p); | |
975 | ||
976 | smp_mb(); /* spin-unlock-wait is not a full memory barrier */ | |
05fa785c | 977 | raw_spin_unlock_wait(&rq->lock); |
ad474cac ON |
978 | } |
979 | ||
a9957449 | 980 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
981 | __releases(rq->lock) |
982 | { | |
05fa785c | 983 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
984 | } |
985 | ||
70b97a7f | 986 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
987 | __releases(rq->lock) |
988 | { | |
05fa785c | 989 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
990 | } |
991 | ||
1da177e4 | 992 | /* |
cc2a73b5 | 993 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 994 | */ |
a9957449 | 995 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
996 | __acquires(rq->lock) |
997 | { | |
70b97a7f | 998 | struct rq *rq; |
1da177e4 LT |
999 | |
1000 | local_irq_disable(); | |
1001 | rq = this_rq(); | |
05fa785c | 1002 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
1003 | |
1004 | return rq; | |
1005 | } | |
1006 | ||
8f4d37ec PZ |
1007 | #ifdef CONFIG_SCHED_HRTICK |
1008 | /* | |
1009 | * Use HR-timers to deliver accurate preemption points. | |
1010 | * | |
1011 | * Its all a bit involved since we cannot program an hrt while holding the | |
1012 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1013 | * reschedule event. | |
1014 | * | |
1015 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1016 | * rq->lock. | |
1017 | */ | |
8f4d37ec PZ |
1018 | |
1019 | /* | |
1020 | * Use hrtick when: | |
1021 | * - enabled by features | |
1022 | * - hrtimer is actually high res | |
1023 | */ | |
1024 | static inline int hrtick_enabled(struct rq *rq) | |
1025 | { | |
1026 | if (!sched_feat(HRTICK)) | |
1027 | return 0; | |
ba42059f | 1028 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1029 | return 0; |
8f4d37ec PZ |
1030 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1031 | } | |
1032 | ||
8f4d37ec PZ |
1033 | static void hrtick_clear(struct rq *rq) |
1034 | { | |
1035 | if (hrtimer_active(&rq->hrtick_timer)) | |
1036 | hrtimer_cancel(&rq->hrtick_timer); | |
1037 | } | |
1038 | ||
8f4d37ec PZ |
1039 | /* |
1040 | * High-resolution timer tick. | |
1041 | * Runs from hardirq context with interrupts disabled. | |
1042 | */ | |
1043 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1044 | { | |
1045 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1046 | ||
1047 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1048 | ||
05fa785c | 1049 | raw_spin_lock(&rq->lock); |
3e51f33f | 1050 | update_rq_clock(rq); |
8f4d37ec | 1051 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1052 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1053 | |
1054 | return HRTIMER_NORESTART; | |
1055 | } | |
1056 | ||
95e904c7 | 1057 | #ifdef CONFIG_SMP |
31656519 PZ |
1058 | /* |
1059 | * called from hardirq (IPI) context | |
1060 | */ | |
1061 | static void __hrtick_start(void *arg) | |
b328ca18 | 1062 | { |
31656519 | 1063 | struct rq *rq = arg; |
b328ca18 | 1064 | |
05fa785c | 1065 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1066 | hrtimer_restart(&rq->hrtick_timer); |
1067 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1068 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1069 | } |
1070 | ||
31656519 PZ |
1071 | /* |
1072 | * Called to set the hrtick timer state. | |
1073 | * | |
1074 | * called with rq->lock held and irqs disabled | |
1075 | */ | |
1076 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1077 | { |
31656519 PZ |
1078 | struct hrtimer *timer = &rq->hrtick_timer; |
1079 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1080 | |
cc584b21 | 1081 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1082 | |
1083 | if (rq == this_rq()) { | |
1084 | hrtimer_restart(timer); | |
1085 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1086 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1087 | rq->hrtick_csd_pending = 1; |
1088 | } | |
b328ca18 PZ |
1089 | } |
1090 | ||
1091 | static int | |
1092 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1093 | { | |
1094 | int cpu = (int)(long)hcpu; | |
1095 | ||
1096 | switch (action) { | |
1097 | case CPU_UP_CANCELED: | |
1098 | case CPU_UP_CANCELED_FROZEN: | |
1099 | case CPU_DOWN_PREPARE: | |
1100 | case CPU_DOWN_PREPARE_FROZEN: | |
1101 | case CPU_DEAD: | |
1102 | case CPU_DEAD_FROZEN: | |
31656519 | 1103 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1104 | return NOTIFY_OK; |
1105 | } | |
1106 | ||
1107 | return NOTIFY_DONE; | |
1108 | } | |
1109 | ||
fa748203 | 1110 | static __init void init_hrtick(void) |
b328ca18 PZ |
1111 | { |
1112 | hotcpu_notifier(hotplug_hrtick, 0); | |
1113 | } | |
31656519 PZ |
1114 | #else |
1115 | /* | |
1116 | * Called to set the hrtick timer state. | |
1117 | * | |
1118 | * called with rq->lock held and irqs disabled | |
1119 | */ | |
1120 | static void hrtick_start(struct rq *rq, u64 delay) | |
1121 | { | |
7f1e2ca9 | 1122 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1123 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1124 | } |
b328ca18 | 1125 | |
006c75f1 | 1126 | static inline void init_hrtick(void) |
8f4d37ec | 1127 | { |
8f4d37ec | 1128 | } |
31656519 | 1129 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1130 | |
31656519 | 1131 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1132 | { |
31656519 PZ |
1133 | #ifdef CONFIG_SMP |
1134 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1135 | |
31656519 PZ |
1136 | rq->hrtick_csd.flags = 0; |
1137 | rq->hrtick_csd.func = __hrtick_start; | |
1138 | rq->hrtick_csd.info = rq; | |
1139 | #endif | |
8f4d37ec | 1140 | |
31656519 PZ |
1141 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1142 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1143 | } |
006c75f1 | 1144 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1145 | static inline void hrtick_clear(struct rq *rq) |
1146 | { | |
1147 | } | |
1148 | ||
8f4d37ec PZ |
1149 | static inline void init_rq_hrtick(struct rq *rq) |
1150 | { | |
1151 | } | |
1152 | ||
b328ca18 PZ |
1153 | static inline void init_hrtick(void) |
1154 | { | |
1155 | } | |
006c75f1 | 1156 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1157 | |
c24d20db IM |
1158 | /* |
1159 | * resched_task - mark a task 'to be rescheduled now'. | |
1160 | * | |
1161 | * On UP this means the setting of the need_resched flag, on SMP it | |
1162 | * might also involve a cross-CPU call to trigger the scheduler on | |
1163 | * the target CPU. | |
1164 | */ | |
1165 | #ifdef CONFIG_SMP | |
1166 | ||
1167 | #ifndef tsk_is_polling | |
1168 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1169 | #endif | |
1170 | ||
31656519 | 1171 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1172 | { |
1173 | int cpu; | |
1174 | ||
05fa785c | 1175 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1176 | |
5ed0cec0 | 1177 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1178 | return; |
1179 | ||
5ed0cec0 | 1180 | set_tsk_need_resched(p); |
c24d20db IM |
1181 | |
1182 | cpu = task_cpu(p); | |
1183 | if (cpu == smp_processor_id()) | |
1184 | return; | |
1185 | ||
1186 | /* NEED_RESCHED must be visible before we test polling */ | |
1187 | smp_mb(); | |
1188 | if (!tsk_is_polling(p)) | |
1189 | smp_send_reschedule(cpu); | |
1190 | } | |
1191 | ||
1192 | static void resched_cpu(int cpu) | |
1193 | { | |
1194 | struct rq *rq = cpu_rq(cpu); | |
1195 | unsigned long flags; | |
1196 | ||
05fa785c | 1197 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1198 | return; |
1199 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1200 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1201 | } |
06d8308c TG |
1202 | |
1203 | #ifdef CONFIG_NO_HZ | |
1204 | /* | |
1205 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1206 | * idle CPU then this timer might expire before the next timer event | |
1207 | * which is scheduled to wake up that CPU. In case of a completely | |
1208 | * idle system the next event might even be infinite time into the | |
1209 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1210 | * leaves the inner idle loop so the newly added timer is taken into | |
1211 | * account when the CPU goes back to idle and evaluates the timer | |
1212 | * wheel for the next timer event. | |
1213 | */ | |
1214 | void wake_up_idle_cpu(int cpu) | |
1215 | { | |
1216 | struct rq *rq = cpu_rq(cpu); | |
1217 | ||
1218 | if (cpu == smp_processor_id()) | |
1219 | return; | |
1220 | ||
1221 | /* | |
1222 | * This is safe, as this function is called with the timer | |
1223 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1224 | * to idle and has not yet set rq->curr to idle then it will | |
1225 | * be serialized on the timer wheel base lock and take the new | |
1226 | * timer into account automatically. | |
1227 | */ | |
1228 | if (rq->curr != rq->idle) | |
1229 | return; | |
1230 | ||
1231 | /* | |
1232 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1233 | * lockless. The worst case is that the other CPU runs the | |
1234 | * idle task through an additional NOOP schedule() | |
1235 | */ | |
5ed0cec0 | 1236 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1237 | |
1238 | /* NEED_RESCHED must be visible before we test polling */ | |
1239 | smp_mb(); | |
1240 | if (!tsk_is_polling(rq->idle)) | |
1241 | smp_send_reschedule(cpu); | |
1242 | } | |
39c0cbe2 MG |
1243 | |
1244 | int nohz_ratelimit(int cpu) | |
1245 | { | |
1246 | struct rq *rq = cpu_rq(cpu); | |
1247 | u64 diff = rq->clock - rq->nohz_stamp; | |
1248 | ||
1249 | rq->nohz_stamp = rq->clock; | |
1250 | ||
1251 | return diff < (NSEC_PER_SEC / HZ) >> 1; | |
1252 | } | |
1253 | ||
6d6bc0ad | 1254 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1255 | |
e9e9250b PZ |
1256 | static u64 sched_avg_period(void) |
1257 | { | |
1258 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1259 | } | |
1260 | ||
1261 | static void sched_avg_update(struct rq *rq) | |
1262 | { | |
1263 | s64 period = sched_avg_period(); | |
1264 | ||
1265 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1266 | rq->age_stamp += period; | |
1267 | rq->rt_avg /= 2; | |
1268 | } | |
1269 | } | |
1270 | ||
1271 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1272 | { | |
1273 | rq->rt_avg += rt_delta; | |
1274 | sched_avg_update(rq); | |
1275 | } | |
1276 | ||
6d6bc0ad | 1277 | #else /* !CONFIG_SMP */ |
31656519 | 1278 | static void resched_task(struct task_struct *p) |
c24d20db | 1279 | { |
05fa785c | 1280 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1281 | set_tsk_need_resched(p); |
c24d20db | 1282 | } |
e9e9250b PZ |
1283 | |
1284 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1285 | { | |
1286 | } | |
6d6bc0ad | 1287 | #endif /* CONFIG_SMP */ |
c24d20db | 1288 | |
45bf76df IM |
1289 | #if BITS_PER_LONG == 32 |
1290 | # define WMULT_CONST (~0UL) | |
1291 | #else | |
1292 | # define WMULT_CONST (1UL << 32) | |
1293 | #endif | |
1294 | ||
1295 | #define WMULT_SHIFT 32 | |
1296 | ||
194081eb IM |
1297 | /* |
1298 | * Shift right and round: | |
1299 | */ | |
cf2ab469 | 1300 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1301 | |
a7be37ac PZ |
1302 | /* |
1303 | * delta *= weight / lw | |
1304 | */ | |
cb1c4fc9 | 1305 | static unsigned long |
45bf76df IM |
1306 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1307 | struct load_weight *lw) | |
1308 | { | |
1309 | u64 tmp; | |
1310 | ||
7a232e03 LJ |
1311 | if (!lw->inv_weight) { |
1312 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1313 | lw->inv_weight = 1; | |
1314 | else | |
1315 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1316 | / (lw->weight+1); | |
1317 | } | |
45bf76df IM |
1318 | |
1319 | tmp = (u64)delta_exec * weight; | |
1320 | /* | |
1321 | * Check whether we'd overflow the 64-bit multiplication: | |
1322 | */ | |
194081eb | 1323 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1324 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1325 | WMULT_SHIFT/2); |
1326 | else | |
cf2ab469 | 1327 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1328 | |
ecf691da | 1329 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1330 | } |
1331 | ||
1091985b | 1332 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1333 | { |
1334 | lw->weight += inc; | |
e89996ae | 1335 | lw->inv_weight = 0; |
45bf76df IM |
1336 | } |
1337 | ||
1091985b | 1338 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1339 | { |
1340 | lw->weight -= dec; | |
e89996ae | 1341 | lw->inv_weight = 0; |
45bf76df IM |
1342 | } |
1343 | ||
2dd73a4f PW |
1344 | /* |
1345 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1346 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1347 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1348 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1349 | * scaled version of the new time slice allocation that they receive on time |
1350 | * slice expiry etc. | |
1351 | */ | |
1352 | ||
cce7ade8 PZ |
1353 | #define WEIGHT_IDLEPRIO 3 |
1354 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1355 | |
1356 | /* | |
1357 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1358 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1359 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1360 | * that remained on nice 0. | |
1361 | * | |
1362 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1363 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1364 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1365 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1366 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1367 | */ |
1368 | static const int prio_to_weight[40] = { | |
254753dc IM |
1369 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1370 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1371 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1372 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1373 | /* 0 */ 1024, 820, 655, 526, 423, | |
1374 | /* 5 */ 335, 272, 215, 172, 137, | |
1375 | /* 10 */ 110, 87, 70, 56, 45, | |
1376 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1377 | }; |
1378 | ||
5714d2de IM |
1379 | /* |
1380 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1381 | * | |
1382 | * In cases where the weight does not change often, we can use the | |
1383 | * precalculated inverse to speed up arithmetics by turning divisions | |
1384 | * into multiplications: | |
1385 | */ | |
dd41f596 | 1386 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1387 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1388 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1389 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1390 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1391 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1392 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1393 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1394 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1395 | }; |
2dd73a4f | 1396 | |
ef12fefa BR |
1397 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1398 | enum cpuacct_stat_index { | |
1399 | CPUACCT_STAT_USER, /* ... user mode */ | |
1400 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1401 | ||
1402 | CPUACCT_STAT_NSTATS, | |
1403 | }; | |
1404 | ||
d842de87 SV |
1405 | #ifdef CONFIG_CGROUP_CPUACCT |
1406 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1407 | static void cpuacct_update_stats(struct task_struct *tsk, |
1408 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1409 | #else |
1410 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1411 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1412 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1413 | #endif |
1414 | ||
18d95a28 PZ |
1415 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1416 | { | |
1417 | update_load_add(&rq->load, load); | |
1418 | } | |
1419 | ||
1420 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1421 | { | |
1422 | update_load_sub(&rq->load, load); | |
1423 | } | |
1424 | ||
7940ca36 | 1425 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1426 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1427 | |
1428 | /* | |
1429 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1430 | * leaving it for the final time. | |
1431 | */ | |
eb755805 | 1432 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1433 | { |
1434 | struct task_group *parent, *child; | |
eb755805 | 1435 | int ret; |
c09595f6 PZ |
1436 | |
1437 | rcu_read_lock(); | |
1438 | parent = &root_task_group; | |
1439 | down: | |
eb755805 PZ |
1440 | ret = (*down)(parent, data); |
1441 | if (ret) | |
1442 | goto out_unlock; | |
c09595f6 PZ |
1443 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1444 | parent = child; | |
1445 | goto down; | |
1446 | ||
1447 | up: | |
1448 | continue; | |
1449 | } | |
eb755805 PZ |
1450 | ret = (*up)(parent, data); |
1451 | if (ret) | |
1452 | goto out_unlock; | |
c09595f6 PZ |
1453 | |
1454 | child = parent; | |
1455 | parent = parent->parent; | |
1456 | if (parent) | |
1457 | goto up; | |
eb755805 | 1458 | out_unlock: |
c09595f6 | 1459 | rcu_read_unlock(); |
eb755805 PZ |
1460 | |
1461 | return ret; | |
c09595f6 PZ |
1462 | } |
1463 | ||
eb755805 PZ |
1464 | static int tg_nop(struct task_group *tg, void *data) |
1465 | { | |
1466 | return 0; | |
c09595f6 | 1467 | } |
eb755805 PZ |
1468 | #endif |
1469 | ||
1470 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1471 | /* Used instead of source_load when we know the type == 0 */ |
1472 | static unsigned long weighted_cpuload(const int cpu) | |
1473 | { | |
1474 | return cpu_rq(cpu)->load.weight; | |
1475 | } | |
1476 | ||
1477 | /* | |
1478 | * Return a low guess at the load of a migration-source cpu weighted | |
1479 | * according to the scheduling class and "nice" value. | |
1480 | * | |
1481 | * We want to under-estimate the load of migration sources, to | |
1482 | * balance conservatively. | |
1483 | */ | |
1484 | static unsigned long source_load(int cpu, int type) | |
1485 | { | |
1486 | struct rq *rq = cpu_rq(cpu); | |
1487 | unsigned long total = weighted_cpuload(cpu); | |
1488 | ||
1489 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1490 | return total; | |
1491 | ||
1492 | return min(rq->cpu_load[type-1], total); | |
1493 | } | |
1494 | ||
1495 | /* | |
1496 | * Return a high guess at the load of a migration-target cpu weighted | |
1497 | * according to the scheduling class and "nice" value. | |
1498 | */ | |
1499 | static unsigned long target_load(int cpu, int type) | |
1500 | { | |
1501 | struct rq *rq = cpu_rq(cpu); | |
1502 | unsigned long total = weighted_cpuload(cpu); | |
1503 | ||
1504 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1505 | return total; | |
1506 | ||
1507 | return max(rq->cpu_load[type-1], total); | |
1508 | } | |
1509 | ||
ae154be1 PZ |
1510 | static struct sched_group *group_of(int cpu) |
1511 | { | |
d11c563d | 1512 | struct sched_domain *sd = rcu_dereference_sched(cpu_rq(cpu)->sd); |
ae154be1 PZ |
1513 | |
1514 | if (!sd) | |
1515 | return NULL; | |
1516 | ||
1517 | return sd->groups; | |
1518 | } | |
1519 | ||
1520 | static unsigned long power_of(int cpu) | |
1521 | { | |
1522 | struct sched_group *group = group_of(cpu); | |
1523 | ||
1524 | if (!group) | |
1525 | return SCHED_LOAD_SCALE; | |
1526 | ||
1527 | return group->cpu_power; | |
1528 | } | |
1529 | ||
eb755805 PZ |
1530 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1531 | ||
1532 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1533 | { | |
1534 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1535 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1536 | |
4cd42620 SR |
1537 | if (nr_running) |
1538 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1539 | else |
1540 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1541 | |
1542 | return rq->avg_load_per_task; | |
1543 | } | |
1544 | ||
1545 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1546 | |
43cf38eb | 1547 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1548 | |
c09595f6 PZ |
1549 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1550 | ||
1551 | /* | |
1552 | * Calculate and set the cpu's group shares. | |
1553 | */ | |
34d76c41 PZ |
1554 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1555 | unsigned long sd_shares, | |
1556 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1557 | unsigned long *usd_rq_weight) |
18d95a28 | 1558 | { |
34d76c41 | 1559 | unsigned long shares, rq_weight; |
a5004278 | 1560 | int boost = 0; |
c09595f6 | 1561 | |
4a6cc4bd | 1562 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1563 | if (!rq_weight) { |
1564 | boost = 1; | |
1565 | rq_weight = NICE_0_LOAD; | |
1566 | } | |
c8cba857 | 1567 | |
c09595f6 | 1568 | /* |
a8af7246 PZ |
1569 | * \Sum_j shares_j * rq_weight_i |
1570 | * shares_i = ----------------------------- | |
1571 | * \Sum_j rq_weight_j | |
c09595f6 | 1572 | */ |
ec4e0e2f | 1573 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1574 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1575 | |
ffda12a1 PZ |
1576 | if (abs(shares - tg->se[cpu]->load.weight) > |
1577 | sysctl_sched_shares_thresh) { | |
1578 | struct rq *rq = cpu_rq(cpu); | |
1579 | unsigned long flags; | |
c09595f6 | 1580 | |
05fa785c | 1581 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1582 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1583 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1584 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1585 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1586 | } |
18d95a28 | 1587 | } |
c09595f6 PZ |
1588 | |
1589 | /* | |
c8cba857 PZ |
1590 | * Re-compute the task group their per cpu shares over the given domain. |
1591 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1592 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1593 | */ |
eb755805 | 1594 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1595 | { |
cd8ad40d | 1596 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1597 | unsigned long *usd_rq_weight; |
eb755805 | 1598 | struct sched_domain *sd = data; |
34d76c41 | 1599 | unsigned long flags; |
c8cba857 | 1600 | int i; |
c09595f6 | 1601 | |
34d76c41 PZ |
1602 | if (!tg->se[0]) |
1603 | return 0; | |
1604 | ||
1605 | local_irq_save(flags); | |
4a6cc4bd | 1606 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1607 | |
758b2cdc | 1608 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1609 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1610 | usd_rq_weight[i] = weight; |
34d76c41 | 1611 | |
cd8ad40d | 1612 | rq_weight += weight; |
ec4e0e2f KC |
1613 | /* |
1614 | * If there are currently no tasks on the cpu pretend there | |
1615 | * is one of average load so that when a new task gets to | |
1616 | * run here it will not get delayed by group starvation. | |
1617 | */ | |
ec4e0e2f KC |
1618 | if (!weight) |
1619 | weight = NICE_0_LOAD; | |
1620 | ||
cd8ad40d | 1621 | sum_weight += weight; |
c8cba857 | 1622 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1623 | } |
c09595f6 | 1624 | |
cd8ad40d PZ |
1625 | if (!rq_weight) |
1626 | rq_weight = sum_weight; | |
1627 | ||
c8cba857 PZ |
1628 | if ((!shares && rq_weight) || shares > tg->shares) |
1629 | shares = tg->shares; | |
1630 | ||
1631 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1632 | shares = tg->shares; | |
c09595f6 | 1633 | |
758b2cdc | 1634 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1635 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1636 | |
1637 | local_irq_restore(flags); | |
eb755805 PZ |
1638 | |
1639 | return 0; | |
c09595f6 PZ |
1640 | } |
1641 | ||
1642 | /* | |
c8cba857 PZ |
1643 | * Compute the cpu's hierarchical load factor for each task group. |
1644 | * This needs to be done in a top-down fashion because the load of a child | |
1645 | * group is a fraction of its parents load. | |
c09595f6 | 1646 | */ |
eb755805 | 1647 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1648 | { |
c8cba857 | 1649 | unsigned long load; |
eb755805 | 1650 | long cpu = (long)data; |
c09595f6 | 1651 | |
c8cba857 PZ |
1652 | if (!tg->parent) { |
1653 | load = cpu_rq(cpu)->load.weight; | |
1654 | } else { | |
1655 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1656 | load *= tg->cfs_rq[cpu]->shares; | |
1657 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1658 | } | |
c09595f6 | 1659 | |
c8cba857 | 1660 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1661 | |
eb755805 | 1662 | return 0; |
c09595f6 PZ |
1663 | } |
1664 | ||
c8cba857 | 1665 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1666 | { |
e7097159 PZ |
1667 | s64 elapsed; |
1668 | u64 now; | |
1669 | ||
1670 | if (root_task_group_empty()) | |
1671 | return; | |
1672 | ||
1673 | now = cpu_clock(raw_smp_processor_id()); | |
1674 | elapsed = now - sd->last_update; | |
2398f2c6 PZ |
1675 | |
1676 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1677 | sd->last_update = now; | |
eb755805 | 1678 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1679 | } |
4d8d595d PZ |
1680 | } |
1681 | ||
eb755805 | 1682 | static void update_h_load(long cpu) |
c09595f6 | 1683 | { |
e7097159 PZ |
1684 | if (root_task_group_empty()) |
1685 | return; | |
1686 | ||
eb755805 | 1687 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1688 | } |
1689 | ||
c09595f6 PZ |
1690 | #else |
1691 | ||
c8cba857 | 1692 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1693 | { |
1694 | } | |
1695 | ||
18d95a28 PZ |
1696 | #endif |
1697 | ||
8f45e2b5 GH |
1698 | #ifdef CONFIG_PREEMPT |
1699 | ||
b78bb868 PZ |
1700 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1701 | ||
70574a99 | 1702 | /* |
8f45e2b5 GH |
1703 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1704 | * way at the expense of forcing extra atomic operations in all | |
1705 | * invocations. This assures that the double_lock is acquired using the | |
1706 | * same underlying policy as the spinlock_t on this architecture, which | |
1707 | * reduces latency compared to the unfair variant below. However, it | |
1708 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1709 | */ |
8f45e2b5 GH |
1710 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1711 | __releases(this_rq->lock) | |
1712 | __acquires(busiest->lock) | |
1713 | __acquires(this_rq->lock) | |
1714 | { | |
05fa785c | 1715 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1716 | double_rq_lock(this_rq, busiest); |
1717 | ||
1718 | return 1; | |
1719 | } | |
1720 | ||
1721 | #else | |
1722 | /* | |
1723 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1724 | * latency by eliminating extra atomic operations when the locks are | |
1725 | * already in proper order on entry. This favors lower cpu-ids and will | |
1726 | * grant the double lock to lower cpus over higher ids under contention, | |
1727 | * regardless of entry order into the function. | |
1728 | */ | |
1729 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1730 | __releases(this_rq->lock) |
1731 | __acquires(busiest->lock) | |
1732 | __acquires(this_rq->lock) | |
1733 | { | |
1734 | int ret = 0; | |
1735 | ||
05fa785c | 1736 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1737 | if (busiest < this_rq) { |
05fa785c TG |
1738 | raw_spin_unlock(&this_rq->lock); |
1739 | raw_spin_lock(&busiest->lock); | |
1740 | raw_spin_lock_nested(&this_rq->lock, | |
1741 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1742 | ret = 1; |
1743 | } else | |
05fa785c TG |
1744 | raw_spin_lock_nested(&busiest->lock, |
1745 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1746 | } |
1747 | return ret; | |
1748 | } | |
1749 | ||
8f45e2b5 GH |
1750 | #endif /* CONFIG_PREEMPT */ |
1751 | ||
1752 | /* | |
1753 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1754 | */ | |
1755 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1756 | { | |
1757 | if (unlikely(!irqs_disabled())) { | |
1758 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1759 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1760 | BUG_ON(1); |
1761 | } | |
1762 | ||
1763 | return _double_lock_balance(this_rq, busiest); | |
1764 | } | |
1765 | ||
70574a99 AD |
1766 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1767 | __releases(busiest->lock) | |
1768 | { | |
05fa785c | 1769 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1770 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1771 | } | |
1e3c88bd PZ |
1772 | |
1773 | /* | |
1774 | * double_rq_lock - safely lock two runqueues | |
1775 | * | |
1776 | * Note this does not disable interrupts like task_rq_lock, | |
1777 | * you need to do so manually before calling. | |
1778 | */ | |
1779 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1780 | __acquires(rq1->lock) | |
1781 | __acquires(rq2->lock) | |
1782 | { | |
1783 | BUG_ON(!irqs_disabled()); | |
1784 | if (rq1 == rq2) { | |
1785 | raw_spin_lock(&rq1->lock); | |
1786 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1787 | } else { | |
1788 | if (rq1 < rq2) { | |
1789 | raw_spin_lock(&rq1->lock); | |
1790 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1791 | } else { | |
1792 | raw_spin_lock(&rq2->lock); | |
1793 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1794 | } | |
1795 | } | |
1e3c88bd PZ |
1796 | } |
1797 | ||
1798 | /* | |
1799 | * double_rq_unlock - safely unlock two runqueues | |
1800 | * | |
1801 | * Note this does not restore interrupts like task_rq_unlock, | |
1802 | * you need to do so manually after calling. | |
1803 | */ | |
1804 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1805 | __releases(rq1->lock) | |
1806 | __releases(rq2->lock) | |
1807 | { | |
1808 | raw_spin_unlock(&rq1->lock); | |
1809 | if (rq1 != rq2) | |
1810 | raw_spin_unlock(&rq2->lock); | |
1811 | else | |
1812 | __release(rq2->lock); | |
1813 | } | |
1814 | ||
18d95a28 PZ |
1815 | #endif |
1816 | ||
30432094 | 1817 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1818 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1819 | { | |
30432094 | 1820 | #ifdef CONFIG_SMP |
34e83e85 IM |
1821 | cfs_rq->shares = shares; |
1822 | #endif | |
1823 | } | |
30432094 | 1824 | #endif |
e7693a36 | 1825 | |
74f5187a | 1826 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1827 | static void update_sysctl(void); |
acb4a848 | 1828 | static int get_update_sysctl_factor(void); |
dce48a84 | 1829 | |
cd29fe6f PZ |
1830 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1831 | { | |
1832 | set_task_rq(p, cpu); | |
1833 | #ifdef CONFIG_SMP | |
1834 | /* | |
1835 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1836 | * successfuly executed on another CPU. We must ensure that updates of | |
1837 | * per-task data have been completed by this moment. | |
1838 | */ | |
1839 | smp_wmb(); | |
1840 | task_thread_info(p)->cpu = cpu; | |
1841 | #endif | |
1842 | } | |
dce48a84 | 1843 | |
1e3c88bd | 1844 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1845 | |
1846 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1847 | #define for_each_class(class) \ |
1848 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1849 | |
1e3c88bd PZ |
1850 | #include "sched_stats.h" |
1851 | ||
c09595f6 | 1852 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1853 | { |
1854 | rq->nr_running++; | |
9c217245 IM |
1855 | } |
1856 | ||
c09595f6 | 1857 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1858 | { |
1859 | rq->nr_running--; | |
9c217245 IM |
1860 | } |
1861 | ||
45bf76df IM |
1862 | static void set_load_weight(struct task_struct *p) |
1863 | { | |
1864 | if (task_has_rt_policy(p)) { | |
dd41f596 IM |
1865 | p->se.load.weight = prio_to_weight[0] * 2; |
1866 | p->se.load.inv_weight = prio_to_wmult[0] >> 1; | |
1867 | return; | |
1868 | } | |
45bf76df | 1869 | |
dd41f596 IM |
1870 | /* |
1871 | * SCHED_IDLE tasks get minimal weight: | |
1872 | */ | |
1873 | if (p->policy == SCHED_IDLE) { | |
1874 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1875 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1876 | return; | |
1877 | } | |
71f8bd46 | 1878 | |
dd41f596 IM |
1879 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1880 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1881 | } |
1882 | ||
371fd7e7 | 1883 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1884 | { |
a64692a3 | 1885 | update_rq_clock(rq); |
dd41f596 | 1886 | sched_info_queued(p); |
371fd7e7 | 1887 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1888 | p->se.on_rq = 1; |
71f8bd46 IM |
1889 | } |
1890 | ||
371fd7e7 | 1891 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1892 | { |
a64692a3 | 1893 | update_rq_clock(rq); |
46ac22ba | 1894 | sched_info_dequeued(p); |
371fd7e7 | 1895 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1896 | p->se.on_rq = 0; |
71f8bd46 IM |
1897 | } |
1898 | ||
1e3c88bd PZ |
1899 | /* |
1900 | * activate_task - move a task to the runqueue. | |
1901 | */ | |
371fd7e7 | 1902 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1903 | { |
1904 | if (task_contributes_to_load(p)) | |
1905 | rq->nr_uninterruptible--; | |
1906 | ||
371fd7e7 | 1907 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1908 | inc_nr_running(rq); |
1909 | } | |
1910 | ||
1911 | /* | |
1912 | * deactivate_task - remove a task from the runqueue. | |
1913 | */ | |
371fd7e7 | 1914 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1915 | { |
1916 | if (task_contributes_to_load(p)) | |
1917 | rq->nr_uninterruptible++; | |
1918 | ||
371fd7e7 | 1919 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1920 | dec_nr_running(rq); |
1921 | } | |
1922 | ||
1923 | #include "sched_idletask.c" | |
1924 | #include "sched_fair.c" | |
1925 | #include "sched_rt.c" | |
1926 | #ifdef CONFIG_SCHED_DEBUG | |
1927 | # include "sched_debug.c" | |
1928 | #endif | |
1929 | ||
14531189 | 1930 | /* |
dd41f596 | 1931 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1932 | */ |
14531189 IM |
1933 | static inline int __normal_prio(struct task_struct *p) |
1934 | { | |
dd41f596 | 1935 | return p->static_prio; |
14531189 IM |
1936 | } |
1937 | ||
b29739f9 IM |
1938 | /* |
1939 | * Calculate the expected normal priority: i.e. priority | |
1940 | * without taking RT-inheritance into account. Might be | |
1941 | * boosted by interactivity modifiers. Changes upon fork, | |
1942 | * setprio syscalls, and whenever the interactivity | |
1943 | * estimator recalculates. | |
1944 | */ | |
36c8b586 | 1945 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1946 | { |
1947 | int prio; | |
1948 | ||
e05606d3 | 1949 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1950 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1951 | else | |
1952 | prio = __normal_prio(p); | |
1953 | return prio; | |
1954 | } | |
1955 | ||
1956 | /* | |
1957 | * Calculate the current priority, i.e. the priority | |
1958 | * taken into account by the scheduler. This value might | |
1959 | * be boosted by RT tasks, or might be boosted by | |
1960 | * interactivity modifiers. Will be RT if the task got | |
1961 | * RT-boosted. If not then it returns p->normal_prio. | |
1962 | */ | |
36c8b586 | 1963 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1964 | { |
1965 | p->normal_prio = normal_prio(p); | |
1966 | /* | |
1967 | * If we are RT tasks or we were boosted to RT priority, | |
1968 | * keep the priority unchanged. Otherwise, update priority | |
1969 | * to the normal priority: | |
1970 | */ | |
1971 | if (!rt_prio(p->prio)) | |
1972 | return p->normal_prio; | |
1973 | return p->prio; | |
1974 | } | |
1975 | ||
1da177e4 LT |
1976 | /** |
1977 | * task_curr - is this task currently executing on a CPU? | |
1978 | * @p: the task in question. | |
1979 | */ | |
36c8b586 | 1980 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1981 | { |
1982 | return cpu_curr(task_cpu(p)) == p; | |
1983 | } | |
1984 | ||
cb469845 SR |
1985 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1986 | const struct sched_class *prev_class, | |
1987 | int oldprio, int running) | |
1988 | { | |
1989 | if (prev_class != p->sched_class) { | |
1990 | if (prev_class->switched_from) | |
1991 | prev_class->switched_from(rq, p, running); | |
1992 | p->sched_class->switched_to(rq, p, running); | |
1993 | } else | |
1994 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1995 | } | |
1996 | ||
1da177e4 | 1997 | #ifdef CONFIG_SMP |
cc367732 IM |
1998 | /* |
1999 | * Is this task likely cache-hot: | |
2000 | */ | |
e7693a36 | 2001 | static int |
cc367732 IM |
2002 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2003 | { | |
2004 | s64 delta; | |
2005 | ||
e6c8fba7 PZ |
2006 | if (p->sched_class != &fair_sched_class) |
2007 | return 0; | |
2008 | ||
f540a608 IM |
2009 | /* |
2010 | * Buddy candidates are cache hot: | |
2011 | */ | |
f685ceac | 2012 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2013 | (&p->se == cfs_rq_of(&p->se)->next || |
2014 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2015 | return 1; |
2016 | ||
6bc1665b IM |
2017 | if (sysctl_sched_migration_cost == -1) |
2018 | return 1; | |
2019 | if (sysctl_sched_migration_cost == 0) | |
2020 | return 0; | |
2021 | ||
cc367732 IM |
2022 | delta = now - p->se.exec_start; |
2023 | ||
2024 | return delta < (s64)sysctl_sched_migration_cost; | |
2025 | } | |
2026 | ||
dd41f596 | 2027 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2028 | { |
e2912009 PZ |
2029 | #ifdef CONFIG_SCHED_DEBUG |
2030 | /* | |
2031 | * We should never call set_task_cpu() on a blocked task, | |
2032 | * ttwu() will sort out the placement. | |
2033 | */ | |
077614ee PZ |
2034 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2035 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2036 | #endif |
2037 | ||
de1d7286 | 2038 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2039 | |
0c69774e PZ |
2040 | if (task_cpu(p) != new_cpu) { |
2041 | p->se.nr_migrations++; | |
2042 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2043 | } | |
dd41f596 IM |
2044 | |
2045 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2046 | } |
2047 | ||
969c7921 | 2048 | struct migration_arg { |
36c8b586 | 2049 | struct task_struct *task; |
1da177e4 | 2050 | int dest_cpu; |
70b97a7f | 2051 | }; |
1da177e4 | 2052 | |
969c7921 TH |
2053 | static int migration_cpu_stop(void *data); |
2054 | ||
1da177e4 LT |
2055 | /* |
2056 | * The task's runqueue lock must be held. | |
2057 | * Returns true if you have to wait for migration thread. | |
2058 | */ | |
969c7921 | 2059 | static bool migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2060 | { |
70b97a7f | 2061 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2062 | |
2063 | /* | |
2064 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2065 | * the next wake-up will properly place the task. |
1da177e4 | 2066 | */ |
969c7921 | 2067 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2068 | } |
2069 | ||
2070 | /* | |
2071 | * wait_task_inactive - wait for a thread to unschedule. | |
2072 | * | |
85ba2d86 RM |
2073 | * If @match_state is nonzero, it's the @p->state value just checked and |
2074 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2075 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2076 | * we return a positive number (its total switch count). If a second call | |
2077 | * a short while later returns the same number, the caller can be sure that | |
2078 | * @p has remained unscheduled the whole time. | |
2079 | * | |
1da177e4 LT |
2080 | * The caller must ensure that the task *will* unschedule sometime soon, |
2081 | * else this function might spin for a *long* time. This function can't | |
2082 | * be called with interrupts off, or it may introduce deadlock with | |
2083 | * smp_call_function() if an IPI is sent by the same process we are | |
2084 | * waiting to become inactive. | |
2085 | */ | |
85ba2d86 | 2086 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2087 | { |
2088 | unsigned long flags; | |
dd41f596 | 2089 | int running, on_rq; |
85ba2d86 | 2090 | unsigned long ncsw; |
70b97a7f | 2091 | struct rq *rq; |
1da177e4 | 2092 | |
3a5c359a AK |
2093 | for (;;) { |
2094 | /* | |
2095 | * We do the initial early heuristics without holding | |
2096 | * any task-queue locks at all. We'll only try to get | |
2097 | * the runqueue lock when things look like they will | |
2098 | * work out! | |
2099 | */ | |
2100 | rq = task_rq(p); | |
fa490cfd | 2101 | |
3a5c359a AK |
2102 | /* |
2103 | * If the task is actively running on another CPU | |
2104 | * still, just relax and busy-wait without holding | |
2105 | * any locks. | |
2106 | * | |
2107 | * NOTE! Since we don't hold any locks, it's not | |
2108 | * even sure that "rq" stays as the right runqueue! | |
2109 | * But we don't care, since "task_running()" will | |
2110 | * return false if the runqueue has changed and p | |
2111 | * is actually now running somewhere else! | |
2112 | */ | |
85ba2d86 RM |
2113 | while (task_running(rq, p)) { |
2114 | if (match_state && unlikely(p->state != match_state)) | |
2115 | return 0; | |
3a5c359a | 2116 | cpu_relax(); |
85ba2d86 | 2117 | } |
fa490cfd | 2118 | |
3a5c359a AK |
2119 | /* |
2120 | * Ok, time to look more closely! We need the rq | |
2121 | * lock now, to be *sure*. If we're wrong, we'll | |
2122 | * just go back and repeat. | |
2123 | */ | |
2124 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2125 | trace_sched_wait_task(p); |
3a5c359a AK |
2126 | running = task_running(rq, p); |
2127 | on_rq = p->se.on_rq; | |
85ba2d86 | 2128 | ncsw = 0; |
f31e11d8 | 2129 | if (!match_state || p->state == match_state) |
93dcf55f | 2130 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2131 | task_rq_unlock(rq, &flags); |
fa490cfd | 2132 | |
85ba2d86 RM |
2133 | /* |
2134 | * If it changed from the expected state, bail out now. | |
2135 | */ | |
2136 | if (unlikely(!ncsw)) | |
2137 | break; | |
2138 | ||
3a5c359a AK |
2139 | /* |
2140 | * Was it really running after all now that we | |
2141 | * checked with the proper locks actually held? | |
2142 | * | |
2143 | * Oops. Go back and try again.. | |
2144 | */ | |
2145 | if (unlikely(running)) { | |
2146 | cpu_relax(); | |
2147 | continue; | |
2148 | } | |
fa490cfd | 2149 | |
3a5c359a AK |
2150 | /* |
2151 | * It's not enough that it's not actively running, | |
2152 | * it must be off the runqueue _entirely_, and not | |
2153 | * preempted! | |
2154 | * | |
80dd99b3 | 2155 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2156 | * running right now), it's preempted, and we should |
2157 | * yield - it could be a while. | |
2158 | */ | |
2159 | if (unlikely(on_rq)) { | |
2160 | schedule_timeout_uninterruptible(1); | |
2161 | continue; | |
2162 | } | |
fa490cfd | 2163 | |
3a5c359a AK |
2164 | /* |
2165 | * Ahh, all good. It wasn't running, and it wasn't | |
2166 | * runnable, which means that it will never become | |
2167 | * running in the future either. We're all done! | |
2168 | */ | |
2169 | break; | |
2170 | } | |
85ba2d86 RM |
2171 | |
2172 | return ncsw; | |
1da177e4 LT |
2173 | } |
2174 | ||
2175 | /*** | |
2176 | * kick_process - kick a running thread to enter/exit the kernel | |
2177 | * @p: the to-be-kicked thread | |
2178 | * | |
2179 | * Cause a process which is running on another CPU to enter | |
2180 | * kernel-mode, without any delay. (to get signals handled.) | |
2181 | * | |
2182 | * NOTE: this function doesnt have to take the runqueue lock, | |
2183 | * because all it wants to ensure is that the remote task enters | |
2184 | * the kernel. If the IPI races and the task has been migrated | |
2185 | * to another CPU then no harm is done and the purpose has been | |
2186 | * achieved as well. | |
2187 | */ | |
36c8b586 | 2188 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2189 | { |
2190 | int cpu; | |
2191 | ||
2192 | preempt_disable(); | |
2193 | cpu = task_cpu(p); | |
2194 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2195 | smp_send_reschedule(cpu); | |
2196 | preempt_enable(); | |
2197 | } | |
b43e3521 | 2198 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2199 | #endif /* CONFIG_SMP */ |
1da177e4 | 2200 | |
0793a61d TG |
2201 | /** |
2202 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2203 | * @p: the task to evaluate | |
2204 | * @func: the function to be called | |
2205 | * @info: the function call argument | |
2206 | * | |
2207 | * Calls the function @func when the task is currently running. This might | |
2208 | * be on the current CPU, which just calls the function directly | |
2209 | */ | |
2210 | void task_oncpu_function_call(struct task_struct *p, | |
2211 | void (*func) (void *info), void *info) | |
2212 | { | |
2213 | int cpu; | |
2214 | ||
2215 | preempt_disable(); | |
2216 | cpu = task_cpu(p); | |
2217 | if (task_curr(p)) | |
2218 | smp_call_function_single(cpu, func, info, 1); | |
2219 | preempt_enable(); | |
2220 | } | |
2221 | ||
970b13ba | 2222 | #ifdef CONFIG_SMP |
30da688e ON |
2223 | /* |
2224 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2225 | */ | |
5da9a0fb PZ |
2226 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2227 | { | |
2228 | int dest_cpu; | |
2229 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2230 | ||
2231 | /* Look for allowed, online CPU in same node. */ | |
2232 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2233 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2234 | return dest_cpu; | |
2235 | ||
2236 | /* Any allowed, online CPU? */ | |
2237 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2238 | if (dest_cpu < nr_cpu_ids) | |
2239 | return dest_cpu; | |
2240 | ||
2241 | /* No more Mr. Nice Guy. */ | |
897f0b3c | 2242 | if (unlikely(dest_cpu >= nr_cpu_ids)) { |
9084bb82 | 2243 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
5da9a0fb PZ |
2244 | /* |
2245 | * Don't tell them about moving exiting tasks or | |
2246 | * kernel threads (both mm NULL), since they never | |
2247 | * leave kernel. | |
2248 | */ | |
2249 | if (p->mm && printk_ratelimit()) { | |
2250 | printk(KERN_INFO "process %d (%s) no " | |
2251 | "longer affine to cpu%d\n", | |
2252 | task_pid_nr(p), p->comm, cpu); | |
2253 | } | |
2254 | } | |
2255 | ||
2256 | return dest_cpu; | |
2257 | } | |
2258 | ||
e2912009 | 2259 | /* |
30da688e | 2260 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2261 | */ |
970b13ba | 2262 | static inline |
0017d735 | 2263 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2264 | { |
0017d735 | 2265 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2266 | |
2267 | /* | |
2268 | * In order not to call set_task_cpu() on a blocking task we need | |
2269 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2270 | * cpu. | |
2271 | * | |
2272 | * Since this is common to all placement strategies, this lives here. | |
2273 | * | |
2274 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2275 | * not worry about this generic constraint ] | |
2276 | */ | |
2277 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2278 | !cpu_online(cpu))) |
5da9a0fb | 2279 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2280 | |
2281 | return cpu; | |
970b13ba | 2282 | } |
09a40af5 MG |
2283 | |
2284 | static void update_avg(u64 *avg, u64 sample) | |
2285 | { | |
2286 | s64 diff = sample - *avg; | |
2287 | *avg += diff >> 3; | |
2288 | } | |
970b13ba PZ |
2289 | #endif |
2290 | ||
1da177e4 LT |
2291 | /*** |
2292 | * try_to_wake_up - wake up a thread | |
2293 | * @p: the to-be-woken-up thread | |
2294 | * @state: the mask of task states that can be woken | |
2295 | * @sync: do a synchronous wakeup? | |
2296 | * | |
2297 | * Put it on the run-queue if it's not already there. The "current" | |
2298 | * thread is always on the run-queue (except when the actual | |
2299 | * re-schedule is in progress), and as such you're allowed to do | |
2300 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2301 | * runnable without the overhead of this. | |
2302 | * | |
2303 | * returns failure only if the task is already active. | |
2304 | */ | |
7d478721 PZ |
2305 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2306 | int wake_flags) | |
1da177e4 | 2307 | { |
cc367732 | 2308 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2309 | unsigned long flags; |
371fd7e7 | 2310 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2311 | struct rq *rq; |
1da177e4 | 2312 | |
e9c84311 | 2313 | this_cpu = get_cpu(); |
2398f2c6 | 2314 | |
04e2f174 | 2315 | smp_wmb(); |
ab3b3aa5 | 2316 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2317 | if (!(p->state & state)) |
1da177e4 LT |
2318 | goto out; |
2319 | ||
dd41f596 | 2320 | if (p->se.on_rq) |
1da177e4 LT |
2321 | goto out_running; |
2322 | ||
2323 | cpu = task_cpu(p); | |
cc367732 | 2324 | orig_cpu = cpu; |
1da177e4 LT |
2325 | |
2326 | #ifdef CONFIG_SMP | |
2327 | if (unlikely(task_running(rq, p))) | |
2328 | goto out_activate; | |
2329 | ||
e9c84311 PZ |
2330 | /* |
2331 | * In order to handle concurrent wakeups and release the rq->lock | |
2332 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2333 | * |
2334 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2335 | */ |
cc87f76a PZ |
2336 | if (task_contributes_to_load(p)) { |
2337 | if (likely(cpu_online(orig_cpu))) | |
2338 | rq->nr_uninterruptible--; | |
2339 | else | |
2340 | this_rq()->nr_uninterruptible--; | |
2341 | } | |
e9c84311 | 2342 | p->state = TASK_WAKING; |
efbbd05a | 2343 | |
371fd7e7 | 2344 | if (p->sched_class->task_waking) { |
efbbd05a | 2345 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2346 | en_flags |= ENQUEUE_WAKING; |
2347 | } | |
efbbd05a | 2348 | |
0017d735 PZ |
2349 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2350 | if (cpu != orig_cpu) | |
5d2f5a61 | 2351 | set_task_cpu(p, cpu); |
0017d735 | 2352 | __task_rq_unlock(rq); |
ab19cb23 | 2353 | |
0970d299 PZ |
2354 | rq = cpu_rq(cpu); |
2355 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2356 | |
0970d299 PZ |
2357 | /* |
2358 | * We migrated the task without holding either rq->lock, however | |
2359 | * since the task is not on the task list itself, nobody else | |
2360 | * will try and migrate the task, hence the rq should match the | |
2361 | * cpu we just moved it to. | |
2362 | */ | |
2363 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2364 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2365 | |
e7693a36 GH |
2366 | #ifdef CONFIG_SCHEDSTATS |
2367 | schedstat_inc(rq, ttwu_count); | |
2368 | if (cpu == this_cpu) | |
2369 | schedstat_inc(rq, ttwu_local); | |
2370 | else { | |
2371 | struct sched_domain *sd; | |
2372 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2373 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2374 | schedstat_inc(sd, ttwu_wake_remote); |
2375 | break; | |
2376 | } | |
2377 | } | |
2378 | } | |
6d6bc0ad | 2379 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2380 | |
1da177e4 LT |
2381 | out_activate: |
2382 | #endif /* CONFIG_SMP */ | |
41acab88 | 2383 | schedstat_inc(p, se.statistics.nr_wakeups); |
7d478721 | 2384 | if (wake_flags & WF_SYNC) |
41acab88 | 2385 | schedstat_inc(p, se.statistics.nr_wakeups_sync); |
cc367732 | 2386 | if (orig_cpu != cpu) |
41acab88 | 2387 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); |
cc367732 | 2388 | if (cpu == this_cpu) |
41acab88 | 2389 | schedstat_inc(p, se.statistics.nr_wakeups_local); |
cc367732 | 2390 | else |
41acab88 | 2391 | schedstat_inc(p, se.statistics.nr_wakeups_remote); |
371fd7e7 | 2392 | activate_task(rq, p, en_flags); |
1da177e4 LT |
2393 | success = 1; |
2394 | ||
2395 | out_running: | |
27a9da65 | 2396 | trace_sched_wakeup(p, success); |
7d478721 | 2397 | check_preempt_curr(rq, p, wake_flags); |
4ae7d5ce | 2398 | |
1da177e4 | 2399 | p->state = TASK_RUNNING; |
9a897c5a | 2400 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2401 | if (p->sched_class->task_woken) |
2402 | p->sched_class->task_woken(rq, p); | |
eae0c9df MG |
2403 | |
2404 | if (unlikely(rq->idle_stamp)) { | |
2405 | u64 delta = rq->clock - rq->idle_stamp; | |
2406 | u64 max = 2*sysctl_sched_migration_cost; | |
2407 | ||
2408 | if (delta > max) | |
2409 | rq->avg_idle = max; | |
2410 | else | |
2411 | update_avg(&rq->avg_idle, delta); | |
2412 | rq->idle_stamp = 0; | |
2413 | } | |
9a897c5a | 2414 | #endif |
1da177e4 LT |
2415 | out: |
2416 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2417 | put_cpu(); |
1da177e4 LT |
2418 | |
2419 | return success; | |
2420 | } | |
2421 | ||
50fa610a DH |
2422 | /** |
2423 | * wake_up_process - Wake up a specific process | |
2424 | * @p: The process to be woken up. | |
2425 | * | |
2426 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2427 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2428 | * running. | |
2429 | * | |
2430 | * It may be assumed that this function implies a write memory barrier before | |
2431 | * changing the task state if and only if any tasks are woken up. | |
2432 | */ | |
7ad5b3a5 | 2433 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2434 | { |
d9514f6c | 2435 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2436 | } |
1da177e4 LT |
2437 | EXPORT_SYMBOL(wake_up_process); |
2438 | ||
7ad5b3a5 | 2439 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2440 | { |
2441 | return try_to_wake_up(p, state, 0); | |
2442 | } | |
2443 | ||
1da177e4 LT |
2444 | /* |
2445 | * Perform scheduler related setup for a newly forked process p. | |
2446 | * p is forked by current. | |
dd41f596 IM |
2447 | * |
2448 | * __sched_fork() is basic setup used by init_idle() too: | |
2449 | */ | |
2450 | static void __sched_fork(struct task_struct *p) | |
2451 | { | |
dd41f596 IM |
2452 | p->se.exec_start = 0; |
2453 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2454 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2455 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2456 | |
2457 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2458 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2459 | #endif |
476d139c | 2460 | |
fa717060 | 2461 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2462 | p->se.on_rq = 0; |
4a55bd5e | 2463 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2464 | |
e107be36 AK |
2465 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2466 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2467 | #endif | |
dd41f596 IM |
2468 | } |
2469 | ||
2470 | /* | |
2471 | * fork()/clone()-time setup: | |
2472 | */ | |
2473 | void sched_fork(struct task_struct *p, int clone_flags) | |
2474 | { | |
2475 | int cpu = get_cpu(); | |
2476 | ||
2477 | __sched_fork(p); | |
06b83b5f | 2478 | /* |
0017d735 | 2479 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2480 | * nobody will actually run it, and a signal or other external |
2481 | * event cannot wake it up and insert it on the runqueue either. | |
2482 | */ | |
0017d735 | 2483 | p->state = TASK_RUNNING; |
dd41f596 | 2484 | |
b9dc29e7 MG |
2485 | /* |
2486 | * Revert to default priority/policy on fork if requested. | |
2487 | */ | |
2488 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2489 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2490 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2491 | p->normal_prio = p->static_prio; |
2492 | } | |
b9dc29e7 | 2493 | |
6c697bdf MG |
2494 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2495 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2496 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2497 | set_load_weight(p); |
2498 | } | |
2499 | ||
b9dc29e7 MG |
2500 | /* |
2501 | * We don't need the reset flag anymore after the fork. It has | |
2502 | * fulfilled its duty: | |
2503 | */ | |
2504 | p->sched_reset_on_fork = 0; | |
2505 | } | |
ca94c442 | 2506 | |
f83f9ac2 PW |
2507 | /* |
2508 | * Make sure we do not leak PI boosting priority to the child. | |
2509 | */ | |
2510 | p->prio = current->normal_prio; | |
2511 | ||
2ddbf952 HS |
2512 | if (!rt_prio(p->prio)) |
2513 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2514 | |
cd29fe6f PZ |
2515 | if (p->sched_class->task_fork) |
2516 | p->sched_class->task_fork(p); | |
2517 | ||
5f3edc1b PZ |
2518 | set_task_cpu(p, cpu); |
2519 | ||
52f17b6c | 2520 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2521 | if (likely(sched_info_on())) |
52f17b6c | 2522 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2523 | #endif |
d6077cb8 | 2524 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2525 | p->oncpu = 0; |
2526 | #endif | |
1da177e4 | 2527 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2528 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2529 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2530 | #endif |
917b627d GH |
2531 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2532 | ||
476d139c | 2533 | put_cpu(); |
1da177e4 LT |
2534 | } |
2535 | ||
2536 | /* | |
2537 | * wake_up_new_task - wake up a newly created task for the first time. | |
2538 | * | |
2539 | * This function will do some initial scheduler statistics housekeeping | |
2540 | * that must be done for every newly created context, then puts the task | |
2541 | * on the runqueue and wakes it. | |
2542 | */ | |
7ad5b3a5 | 2543 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2544 | { |
2545 | unsigned long flags; | |
dd41f596 | 2546 | struct rq *rq; |
c890692b | 2547 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2548 | |
2549 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2550 | rq = task_rq_lock(p, &flags); |
2551 | p->state = TASK_WAKING; | |
2552 | ||
fabf318e PZ |
2553 | /* |
2554 | * Fork balancing, do it here and not earlier because: | |
2555 | * - cpus_allowed can change in the fork path | |
2556 | * - any previously selected cpu might disappear through hotplug | |
2557 | * | |
0017d735 PZ |
2558 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2559 | * without people poking at ->cpus_allowed. | |
fabf318e | 2560 | */ |
0017d735 | 2561 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2562 | set_task_cpu(p, cpu); |
1da177e4 | 2563 | |
06b83b5f | 2564 | p->state = TASK_RUNNING; |
0017d735 PZ |
2565 | task_rq_unlock(rq, &flags); |
2566 | #endif | |
2567 | ||
2568 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2569 | activate_task(rq, p, 0); |
27a9da65 | 2570 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2571 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2572 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2573 | if (p->sched_class->task_woken) |
2574 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2575 | #endif |
dd41f596 | 2576 | task_rq_unlock(rq, &flags); |
fabf318e | 2577 | put_cpu(); |
1da177e4 LT |
2578 | } |
2579 | ||
e107be36 AK |
2580 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2581 | ||
2582 | /** | |
80dd99b3 | 2583 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2584 | * @notifier: notifier struct to register |
e107be36 AK |
2585 | */ |
2586 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2587 | { | |
2588 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2589 | } | |
2590 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2591 | ||
2592 | /** | |
2593 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2594 | * @notifier: notifier struct to unregister |
e107be36 AK |
2595 | * |
2596 | * This is safe to call from within a preemption notifier. | |
2597 | */ | |
2598 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2599 | { | |
2600 | hlist_del(¬ifier->link); | |
2601 | } | |
2602 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2603 | ||
2604 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2605 | { | |
2606 | struct preempt_notifier *notifier; | |
2607 | struct hlist_node *node; | |
2608 | ||
2609 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2610 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2611 | } | |
2612 | ||
2613 | static void | |
2614 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2615 | struct task_struct *next) | |
2616 | { | |
2617 | struct preempt_notifier *notifier; | |
2618 | struct hlist_node *node; | |
2619 | ||
2620 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2621 | notifier->ops->sched_out(notifier, next); | |
2622 | } | |
2623 | ||
6d6bc0ad | 2624 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2625 | |
2626 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2627 | { | |
2628 | } | |
2629 | ||
2630 | static void | |
2631 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2632 | struct task_struct *next) | |
2633 | { | |
2634 | } | |
2635 | ||
6d6bc0ad | 2636 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2637 | |
4866cde0 NP |
2638 | /** |
2639 | * prepare_task_switch - prepare to switch tasks | |
2640 | * @rq: the runqueue preparing to switch | |
421cee29 | 2641 | * @prev: the current task that is being switched out |
4866cde0 NP |
2642 | * @next: the task we are going to switch to. |
2643 | * | |
2644 | * This is called with the rq lock held and interrupts off. It must | |
2645 | * be paired with a subsequent finish_task_switch after the context | |
2646 | * switch. | |
2647 | * | |
2648 | * prepare_task_switch sets up locking and calls architecture specific | |
2649 | * hooks. | |
2650 | */ | |
e107be36 AK |
2651 | static inline void |
2652 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2653 | struct task_struct *next) | |
4866cde0 | 2654 | { |
e107be36 | 2655 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2656 | prepare_lock_switch(rq, next); |
2657 | prepare_arch_switch(next); | |
2658 | } | |
2659 | ||
1da177e4 LT |
2660 | /** |
2661 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2662 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2663 | * @prev: the thread we just switched away from. |
2664 | * | |
4866cde0 NP |
2665 | * finish_task_switch must be called after the context switch, paired |
2666 | * with a prepare_task_switch call before the context switch. | |
2667 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2668 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2669 | * |
2670 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2671 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2672 | * with the lock held can cause deadlocks; see schedule() for |
2673 | * details.) | |
2674 | */ | |
a9957449 | 2675 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2676 | __releases(rq->lock) |
2677 | { | |
1da177e4 | 2678 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2679 | long prev_state; |
1da177e4 LT |
2680 | |
2681 | rq->prev_mm = NULL; | |
2682 | ||
2683 | /* | |
2684 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2685 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2686 | * schedule one last time. The schedule call will never return, and |
2687 | * the scheduled task must drop that reference. | |
c394cc9f | 2688 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2689 | * still held, otherwise prev could be scheduled on another cpu, die |
2690 | * there before we look at prev->state, and then the reference would | |
2691 | * be dropped twice. | |
2692 | * Manfred Spraul <manfred@colorfullife.com> | |
2693 | */ | |
55a101f8 | 2694 | prev_state = prev->state; |
4866cde0 | 2695 | finish_arch_switch(prev); |
8381f65d JI |
2696 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2697 | local_irq_disable(); | |
2698 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2699 | perf_event_task_sched_in(current); |
8381f65d JI |
2700 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2701 | local_irq_enable(); | |
2702 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2703 | finish_lock_switch(rq, prev); |
e8fa1362 | 2704 | |
e107be36 | 2705 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2706 | if (mm) |
2707 | mmdrop(mm); | |
c394cc9f | 2708 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2709 | /* |
2710 | * Remove function-return probe instances associated with this | |
2711 | * task and put them back on the free list. | |
9761eea8 | 2712 | */ |
c6fd91f0 | 2713 | kprobe_flush_task(prev); |
1da177e4 | 2714 | put_task_struct(prev); |
c6fd91f0 | 2715 | } |
1da177e4 LT |
2716 | } |
2717 | ||
3f029d3c GH |
2718 | #ifdef CONFIG_SMP |
2719 | ||
2720 | /* assumes rq->lock is held */ | |
2721 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2722 | { | |
2723 | if (prev->sched_class->pre_schedule) | |
2724 | prev->sched_class->pre_schedule(rq, prev); | |
2725 | } | |
2726 | ||
2727 | /* rq->lock is NOT held, but preemption is disabled */ | |
2728 | static inline void post_schedule(struct rq *rq) | |
2729 | { | |
2730 | if (rq->post_schedule) { | |
2731 | unsigned long flags; | |
2732 | ||
05fa785c | 2733 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2734 | if (rq->curr->sched_class->post_schedule) |
2735 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2736 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2737 | |
2738 | rq->post_schedule = 0; | |
2739 | } | |
2740 | } | |
2741 | ||
2742 | #else | |
da19ab51 | 2743 | |
3f029d3c GH |
2744 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2745 | { | |
2746 | } | |
2747 | ||
2748 | static inline void post_schedule(struct rq *rq) | |
2749 | { | |
1da177e4 LT |
2750 | } |
2751 | ||
3f029d3c GH |
2752 | #endif |
2753 | ||
1da177e4 LT |
2754 | /** |
2755 | * schedule_tail - first thing a freshly forked thread must call. | |
2756 | * @prev: the thread we just switched away from. | |
2757 | */ | |
36c8b586 | 2758 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2759 | __releases(rq->lock) |
2760 | { | |
70b97a7f IM |
2761 | struct rq *rq = this_rq(); |
2762 | ||
4866cde0 | 2763 | finish_task_switch(rq, prev); |
da19ab51 | 2764 | |
3f029d3c GH |
2765 | /* |
2766 | * FIXME: do we need to worry about rq being invalidated by the | |
2767 | * task_switch? | |
2768 | */ | |
2769 | post_schedule(rq); | |
70b97a7f | 2770 | |
4866cde0 NP |
2771 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2772 | /* In this case, finish_task_switch does not reenable preemption */ | |
2773 | preempt_enable(); | |
2774 | #endif | |
1da177e4 | 2775 | if (current->set_child_tid) |
b488893a | 2776 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2777 | } |
2778 | ||
2779 | /* | |
2780 | * context_switch - switch to the new MM and the new | |
2781 | * thread's register state. | |
2782 | */ | |
dd41f596 | 2783 | static inline void |
70b97a7f | 2784 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2785 | struct task_struct *next) |
1da177e4 | 2786 | { |
dd41f596 | 2787 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2788 | |
e107be36 | 2789 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2790 | trace_sched_switch(prev, next); |
dd41f596 IM |
2791 | mm = next->mm; |
2792 | oldmm = prev->active_mm; | |
9226d125 ZA |
2793 | /* |
2794 | * For paravirt, this is coupled with an exit in switch_to to | |
2795 | * combine the page table reload and the switch backend into | |
2796 | * one hypercall. | |
2797 | */ | |
224101ed | 2798 | arch_start_context_switch(prev); |
9226d125 | 2799 | |
710390d9 | 2800 | if (likely(!mm)) { |
1da177e4 LT |
2801 | next->active_mm = oldmm; |
2802 | atomic_inc(&oldmm->mm_count); | |
2803 | enter_lazy_tlb(oldmm, next); | |
2804 | } else | |
2805 | switch_mm(oldmm, mm, next); | |
2806 | ||
710390d9 | 2807 | if (likely(!prev->mm)) { |
1da177e4 | 2808 | prev->active_mm = NULL; |
1da177e4 LT |
2809 | rq->prev_mm = oldmm; |
2810 | } | |
3a5f5e48 IM |
2811 | /* |
2812 | * Since the runqueue lock will be released by the next | |
2813 | * task (which is an invalid locking op but in the case | |
2814 | * of the scheduler it's an obvious special-case), so we | |
2815 | * do an early lockdep release here: | |
2816 | */ | |
2817 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2818 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2819 | #endif |
1da177e4 LT |
2820 | |
2821 | /* Here we just switch the register state and the stack. */ | |
2822 | switch_to(prev, next, prev); | |
2823 | ||
dd41f596 IM |
2824 | barrier(); |
2825 | /* | |
2826 | * this_rq must be evaluated again because prev may have moved | |
2827 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2828 | * frame will be invalid. | |
2829 | */ | |
2830 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2831 | } |
2832 | ||
2833 | /* | |
2834 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2835 | * | |
2836 | * externally visible scheduler statistics: current number of runnable | |
2837 | * threads, current number of uninterruptible-sleeping threads, total | |
2838 | * number of context switches performed since bootup. | |
2839 | */ | |
2840 | unsigned long nr_running(void) | |
2841 | { | |
2842 | unsigned long i, sum = 0; | |
2843 | ||
2844 | for_each_online_cpu(i) | |
2845 | sum += cpu_rq(i)->nr_running; | |
2846 | ||
2847 | return sum; | |
f711f609 | 2848 | } |
1da177e4 LT |
2849 | |
2850 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2851 | { |
1da177e4 | 2852 | unsigned long i, sum = 0; |
f711f609 | 2853 | |
0a945022 | 2854 | for_each_possible_cpu(i) |
1da177e4 | 2855 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2856 | |
2857 | /* | |
1da177e4 LT |
2858 | * Since we read the counters lockless, it might be slightly |
2859 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2860 | */ |
1da177e4 LT |
2861 | if (unlikely((long)sum < 0)) |
2862 | sum = 0; | |
f711f609 | 2863 | |
1da177e4 | 2864 | return sum; |
f711f609 | 2865 | } |
f711f609 | 2866 | |
1da177e4 | 2867 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2868 | { |
cc94abfc SR |
2869 | int i; |
2870 | unsigned long long sum = 0; | |
46cb4b7c | 2871 | |
0a945022 | 2872 | for_each_possible_cpu(i) |
1da177e4 | 2873 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2874 | |
1da177e4 LT |
2875 | return sum; |
2876 | } | |
483b4ee6 | 2877 | |
1da177e4 LT |
2878 | unsigned long nr_iowait(void) |
2879 | { | |
2880 | unsigned long i, sum = 0; | |
483b4ee6 | 2881 | |
0a945022 | 2882 | for_each_possible_cpu(i) |
1da177e4 | 2883 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2884 | |
1da177e4 LT |
2885 | return sum; |
2886 | } | |
483b4ee6 | 2887 | |
69d25870 AV |
2888 | unsigned long nr_iowait_cpu(void) |
2889 | { | |
2890 | struct rq *this = this_rq(); | |
2891 | return atomic_read(&this->nr_iowait); | |
2892 | } | |
46cb4b7c | 2893 | |
69d25870 AV |
2894 | unsigned long this_cpu_load(void) |
2895 | { | |
2896 | struct rq *this = this_rq(); | |
2897 | return this->cpu_load[0]; | |
2898 | } | |
e790fb0b | 2899 | |
46cb4b7c | 2900 | |
dce48a84 TG |
2901 | /* Variables and functions for calc_load */ |
2902 | static atomic_long_t calc_load_tasks; | |
2903 | static unsigned long calc_load_update; | |
2904 | unsigned long avenrun[3]; | |
2905 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2906 | |
74f5187a PZ |
2907 | static long calc_load_fold_active(struct rq *this_rq) |
2908 | { | |
2909 | long nr_active, delta = 0; | |
2910 | ||
2911 | nr_active = this_rq->nr_running; | |
2912 | nr_active += (long) this_rq->nr_uninterruptible; | |
2913 | ||
2914 | if (nr_active != this_rq->calc_load_active) { | |
2915 | delta = nr_active - this_rq->calc_load_active; | |
2916 | this_rq->calc_load_active = nr_active; | |
2917 | } | |
2918 | ||
2919 | return delta; | |
2920 | } | |
2921 | ||
2922 | #ifdef CONFIG_NO_HZ | |
2923 | /* | |
2924 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2925 | * | |
2926 | * When making the ILB scale, we should try to pull this in as well. | |
2927 | */ | |
2928 | static atomic_long_t calc_load_tasks_idle; | |
2929 | ||
2930 | static void calc_load_account_idle(struct rq *this_rq) | |
2931 | { | |
2932 | long delta; | |
2933 | ||
2934 | delta = calc_load_fold_active(this_rq); | |
2935 | if (delta) | |
2936 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2937 | } | |
2938 | ||
2939 | static long calc_load_fold_idle(void) | |
2940 | { | |
2941 | long delta = 0; | |
2942 | ||
2943 | /* | |
2944 | * Its got a race, we don't care... | |
2945 | */ | |
2946 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2947 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2948 | ||
2949 | return delta; | |
2950 | } | |
2951 | #else | |
2952 | static void calc_load_account_idle(struct rq *this_rq) | |
2953 | { | |
2954 | } | |
2955 | ||
2956 | static inline long calc_load_fold_idle(void) | |
2957 | { | |
2958 | return 0; | |
2959 | } | |
2960 | #endif | |
2961 | ||
2d02494f TG |
2962 | /** |
2963 | * get_avenrun - get the load average array | |
2964 | * @loads: pointer to dest load array | |
2965 | * @offset: offset to add | |
2966 | * @shift: shift count to shift the result left | |
2967 | * | |
2968 | * These values are estimates at best, so no need for locking. | |
2969 | */ | |
2970 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2971 | { | |
2972 | loads[0] = (avenrun[0] + offset) << shift; | |
2973 | loads[1] = (avenrun[1] + offset) << shift; | |
2974 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 2975 | } |
46cb4b7c | 2976 | |
dce48a84 TG |
2977 | static unsigned long |
2978 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 2979 | { |
dce48a84 TG |
2980 | load *= exp; |
2981 | load += active * (FIXED_1 - exp); | |
2982 | return load >> FSHIFT; | |
2983 | } | |
46cb4b7c SS |
2984 | |
2985 | /* | |
dce48a84 TG |
2986 | * calc_load - update the avenrun load estimates 10 ticks after the |
2987 | * CPUs have updated calc_load_tasks. | |
7835b98b | 2988 | */ |
dce48a84 | 2989 | void calc_global_load(void) |
7835b98b | 2990 | { |
dce48a84 TG |
2991 | unsigned long upd = calc_load_update + 10; |
2992 | long active; | |
1da177e4 | 2993 | |
dce48a84 TG |
2994 | if (time_before(jiffies, upd)) |
2995 | return; | |
1da177e4 | 2996 | |
dce48a84 TG |
2997 | active = atomic_long_read(&calc_load_tasks); |
2998 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 2999 | |
dce48a84 TG |
3000 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3001 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3002 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3003 | |
dce48a84 TG |
3004 | calc_load_update += LOAD_FREQ; |
3005 | } | |
1da177e4 | 3006 | |
dce48a84 | 3007 | /* |
74f5187a PZ |
3008 | * Called from update_cpu_load() to periodically update this CPU's |
3009 | * active count. | |
dce48a84 TG |
3010 | */ |
3011 | static void calc_load_account_active(struct rq *this_rq) | |
3012 | { | |
74f5187a | 3013 | long delta; |
08c183f3 | 3014 | |
74f5187a PZ |
3015 | if (time_before(jiffies, this_rq->calc_load_update)) |
3016 | return; | |
783609c6 | 3017 | |
74f5187a PZ |
3018 | delta = calc_load_fold_active(this_rq); |
3019 | delta += calc_load_fold_idle(); | |
3020 | if (delta) | |
dce48a84 | 3021 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3022 | |
3023 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3024 | } |
3025 | ||
3026 | /* | |
dd41f596 IM |
3027 | * Update rq->cpu_load[] statistics. This function is usually called every |
3028 | * scheduler tick (TICK_NSEC). | |
46cb4b7c | 3029 | */ |
dd41f596 | 3030 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3031 | { |
495eca49 | 3032 | unsigned long this_load = this_rq->load.weight; |
dd41f596 | 3033 | int i, scale; |
46cb4b7c | 3034 | |
dd41f596 | 3035 | this_rq->nr_load_updates++; |
46cb4b7c | 3036 | |
dd41f596 IM |
3037 | /* Update our load: */ |
3038 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3039 | unsigned long old_load, new_load; | |
7d1e6a9b | 3040 | |
dd41f596 | 3041 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3042 | |
dd41f596 IM |
3043 | old_load = this_rq->cpu_load[i]; |
3044 | new_load = this_load; | |
a25707f3 IM |
3045 | /* |
3046 | * Round up the averaging division if load is increasing. This | |
3047 | * prevents us from getting stuck on 9 if the load is 10, for | |
3048 | * example. | |
3049 | */ | |
3050 | if (new_load > old_load) | |
3051 | new_load += scale-1; | |
dd41f596 IM |
3052 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3053 | } | |
46cb4b7c | 3054 | |
74f5187a | 3055 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3056 | } |
3057 | ||
dd41f596 | 3058 | #ifdef CONFIG_SMP |
8a0be9ef | 3059 | |
46cb4b7c | 3060 | /* |
38022906 PZ |
3061 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3062 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3063 | */ |
38022906 | 3064 | void sched_exec(void) |
46cb4b7c | 3065 | { |
38022906 | 3066 | struct task_struct *p = current; |
1da177e4 | 3067 | unsigned long flags; |
70b97a7f | 3068 | struct rq *rq; |
0017d735 | 3069 | int dest_cpu; |
46cb4b7c | 3070 | |
1da177e4 | 3071 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3072 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3073 | if (dest_cpu == smp_processor_id()) | |
3074 | goto unlock; | |
38022906 | 3075 | |
46cb4b7c | 3076 | /* |
38022906 | 3077 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3078 | */ |
30da688e | 3079 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
969c7921 TH |
3080 | likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) { |
3081 | struct migration_arg arg = { p, dest_cpu }; | |
46cb4b7c | 3082 | |
1da177e4 | 3083 | task_rq_unlock(rq, &flags); |
969c7921 | 3084 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3085 | return; |
3086 | } | |
0017d735 | 3087 | unlock: |
1da177e4 | 3088 | task_rq_unlock(rq, &flags); |
1da177e4 | 3089 | } |
dd41f596 | 3090 | |
1da177e4 LT |
3091 | #endif |
3092 | ||
1da177e4 LT |
3093 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3094 | ||
3095 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3096 | ||
3097 | /* | |
c5f8d995 | 3098 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3099 | * @p in case that task is currently running. |
c5f8d995 HS |
3100 | * |
3101 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3102 | */ |
c5f8d995 HS |
3103 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3104 | { | |
3105 | u64 ns = 0; | |
3106 | ||
3107 | if (task_current(rq, p)) { | |
3108 | update_rq_clock(rq); | |
3109 | ns = rq->clock - p->se.exec_start; | |
3110 | if ((s64)ns < 0) | |
3111 | ns = 0; | |
3112 | } | |
3113 | ||
3114 | return ns; | |
3115 | } | |
3116 | ||
bb34d92f | 3117 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3118 | { |
1da177e4 | 3119 | unsigned long flags; |
41b86e9c | 3120 | struct rq *rq; |
bb34d92f | 3121 | u64 ns = 0; |
48f24c4d | 3122 | |
41b86e9c | 3123 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3124 | ns = do_task_delta_exec(p, rq); |
3125 | task_rq_unlock(rq, &flags); | |
1508487e | 3126 | |
c5f8d995 HS |
3127 | return ns; |
3128 | } | |
f06febc9 | 3129 | |
c5f8d995 HS |
3130 | /* |
3131 | * Return accounted runtime for the task. | |
3132 | * In case the task is currently running, return the runtime plus current's | |
3133 | * pending runtime that have not been accounted yet. | |
3134 | */ | |
3135 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3136 | { | |
3137 | unsigned long flags; | |
3138 | struct rq *rq; | |
3139 | u64 ns = 0; | |
3140 | ||
3141 | rq = task_rq_lock(p, &flags); | |
3142 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3143 | task_rq_unlock(rq, &flags); | |
3144 | ||
3145 | return ns; | |
3146 | } | |
48f24c4d | 3147 | |
c5f8d995 HS |
3148 | /* |
3149 | * Return sum_exec_runtime for the thread group. | |
3150 | * In case the task is currently running, return the sum plus current's | |
3151 | * pending runtime that have not been accounted yet. | |
3152 | * | |
3153 | * Note that the thread group might have other running tasks as well, | |
3154 | * so the return value not includes other pending runtime that other | |
3155 | * running tasks might have. | |
3156 | */ | |
3157 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3158 | { | |
3159 | struct task_cputime totals; | |
3160 | unsigned long flags; | |
3161 | struct rq *rq; | |
3162 | u64 ns; | |
3163 | ||
3164 | rq = task_rq_lock(p, &flags); | |
3165 | thread_group_cputime(p, &totals); | |
3166 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3167 | task_rq_unlock(rq, &flags); |
48f24c4d | 3168 | |
1da177e4 LT |
3169 | return ns; |
3170 | } | |
3171 | ||
1da177e4 LT |
3172 | /* |
3173 | * Account user cpu time to a process. | |
3174 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3175 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3176 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3177 | */ |
457533a7 MS |
3178 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3179 | cputime_t cputime_scaled) | |
1da177e4 LT |
3180 | { |
3181 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3182 | cputime64_t tmp; | |
3183 | ||
457533a7 | 3184 | /* Add user time to process. */ |
1da177e4 | 3185 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3186 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3187 | account_group_user_time(p, cputime); |
1da177e4 LT |
3188 | |
3189 | /* Add user time to cpustat. */ | |
3190 | tmp = cputime_to_cputime64(cputime); | |
3191 | if (TASK_NICE(p) > 0) | |
3192 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3193 | else | |
3194 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3195 | |
3196 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3197 | /* Account for user time used */ |
3198 | acct_update_integrals(p); | |
1da177e4 LT |
3199 | } |
3200 | ||
94886b84 LV |
3201 | /* |
3202 | * Account guest cpu time to a process. | |
3203 | * @p: the process that the cpu time gets accounted to | |
3204 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3205 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3206 | */ |
457533a7 MS |
3207 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3208 | cputime_t cputime_scaled) | |
94886b84 LV |
3209 | { |
3210 | cputime64_t tmp; | |
3211 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3212 | ||
3213 | tmp = cputime_to_cputime64(cputime); | |
3214 | ||
457533a7 | 3215 | /* Add guest time to process. */ |
94886b84 | 3216 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3217 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3218 | account_group_user_time(p, cputime); |
94886b84 LV |
3219 | p->gtime = cputime_add(p->gtime, cputime); |
3220 | ||
457533a7 | 3221 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3222 | if (TASK_NICE(p) > 0) { |
3223 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3224 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3225 | } else { | |
3226 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3227 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3228 | } | |
94886b84 LV |
3229 | } |
3230 | ||
1da177e4 LT |
3231 | /* |
3232 | * Account system cpu time to a process. | |
3233 | * @p: the process that the cpu time gets accounted to | |
3234 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3235 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3236 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3237 | */ |
3238 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3239 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3240 | { |
3241 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3242 | cputime64_t tmp; |
3243 | ||
983ed7a6 | 3244 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3245 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3246 | return; |
3247 | } | |
94886b84 | 3248 | |
457533a7 | 3249 | /* Add system time to process. */ |
1da177e4 | 3250 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3251 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3252 | account_group_system_time(p, cputime); |
1da177e4 LT |
3253 | |
3254 | /* Add system time to cpustat. */ | |
3255 | tmp = cputime_to_cputime64(cputime); | |
3256 | if (hardirq_count() - hardirq_offset) | |
3257 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3258 | else if (softirq_count()) | |
3259 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3260 | else |
79741dd3 MS |
3261 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3262 | ||
ef12fefa BR |
3263 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3264 | ||
1da177e4 LT |
3265 | /* Account for system time used */ |
3266 | acct_update_integrals(p); | |
1da177e4 LT |
3267 | } |
3268 | ||
c66f08be | 3269 | /* |
1da177e4 | 3270 | * Account for involuntary wait time. |
1da177e4 | 3271 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3272 | */ |
79741dd3 | 3273 | void account_steal_time(cputime_t cputime) |
c66f08be | 3274 | { |
79741dd3 MS |
3275 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3276 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3277 | ||
3278 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3279 | } |
3280 | ||
1da177e4 | 3281 | /* |
79741dd3 MS |
3282 | * Account for idle time. |
3283 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3284 | */ |
79741dd3 | 3285 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3286 | { |
3287 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3288 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3289 | struct rq *rq = this_rq(); |
1da177e4 | 3290 | |
79741dd3 MS |
3291 | if (atomic_read(&rq->nr_iowait) > 0) |
3292 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3293 | else | |
3294 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3295 | } |
3296 | ||
79741dd3 MS |
3297 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3298 | ||
3299 | /* | |
3300 | * Account a single tick of cpu time. | |
3301 | * @p: the process that the cpu time gets accounted to | |
3302 | * @user_tick: indicates if the tick is a user or a system tick | |
3303 | */ | |
3304 | void account_process_tick(struct task_struct *p, int user_tick) | |
3305 | { | |
a42548a1 | 3306 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3307 | struct rq *rq = this_rq(); |
3308 | ||
3309 | if (user_tick) | |
a42548a1 | 3310 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3311 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3312 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3313 | one_jiffy_scaled); |
3314 | else | |
a42548a1 | 3315 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3316 | } |
3317 | ||
3318 | /* | |
3319 | * Account multiple ticks of steal time. | |
3320 | * @p: the process from which the cpu time has been stolen | |
3321 | * @ticks: number of stolen ticks | |
3322 | */ | |
3323 | void account_steal_ticks(unsigned long ticks) | |
3324 | { | |
3325 | account_steal_time(jiffies_to_cputime(ticks)); | |
3326 | } | |
3327 | ||
3328 | /* | |
3329 | * Account multiple ticks of idle time. | |
3330 | * @ticks: number of stolen ticks | |
3331 | */ | |
3332 | void account_idle_ticks(unsigned long ticks) | |
3333 | { | |
3334 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3335 | } |
3336 | ||
79741dd3 MS |
3337 | #endif |
3338 | ||
49048622 BS |
3339 | /* |
3340 | * Use precise platform statistics if available: | |
3341 | */ | |
3342 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3343 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3344 | { |
d99ca3b9 HS |
3345 | *ut = p->utime; |
3346 | *st = p->stime; | |
49048622 BS |
3347 | } |
3348 | ||
0cf55e1e | 3349 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3350 | { |
0cf55e1e HS |
3351 | struct task_cputime cputime; |
3352 | ||
3353 | thread_group_cputime(p, &cputime); | |
3354 | ||
3355 | *ut = cputime.utime; | |
3356 | *st = cputime.stime; | |
49048622 BS |
3357 | } |
3358 | #else | |
761b1d26 HS |
3359 | |
3360 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3361 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3362 | #endif |
3363 | ||
d180c5bc | 3364 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3365 | { |
d99ca3b9 | 3366 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3367 | |
3368 | /* | |
3369 | * Use CFS's precise accounting: | |
3370 | */ | |
d180c5bc | 3371 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3372 | |
3373 | if (total) { | |
d180c5bc HS |
3374 | u64 temp; |
3375 | ||
3376 | temp = (u64)(rtime * utime); | |
49048622 | 3377 | do_div(temp, total); |
d180c5bc HS |
3378 | utime = (cputime_t)temp; |
3379 | } else | |
3380 | utime = rtime; | |
49048622 | 3381 | |
d180c5bc HS |
3382 | /* |
3383 | * Compare with previous values, to keep monotonicity: | |
3384 | */ | |
761b1d26 | 3385 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3386 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3387 | |
d99ca3b9 HS |
3388 | *ut = p->prev_utime; |
3389 | *st = p->prev_stime; | |
49048622 BS |
3390 | } |
3391 | ||
0cf55e1e HS |
3392 | /* |
3393 | * Must be called with siglock held. | |
3394 | */ | |
3395 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3396 | { |
0cf55e1e HS |
3397 | struct signal_struct *sig = p->signal; |
3398 | struct task_cputime cputime; | |
3399 | cputime_t rtime, utime, total; | |
49048622 | 3400 | |
0cf55e1e | 3401 | thread_group_cputime(p, &cputime); |
49048622 | 3402 | |
0cf55e1e HS |
3403 | total = cputime_add(cputime.utime, cputime.stime); |
3404 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3405 | |
0cf55e1e HS |
3406 | if (total) { |
3407 | u64 temp; | |
49048622 | 3408 | |
0cf55e1e HS |
3409 | temp = (u64)(rtime * cputime.utime); |
3410 | do_div(temp, total); | |
3411 | utime = (cputime_t)temp; | |
3412 | } else | |
3413 | utime = rtime; | |
3414 | ||
3415 | sig->prev_utime = max(sig->prev_utime, utime); | |
3416 | sig->prev_stime = max(sig->prev_stime, | |
3417 | cputime_sub(rtime, sig->prev_utime)); | |
3418 | ||
3419 | *ut = sig->prev_utime; | |
3420 | *st = sig->prev_stime; | |
49048622 | 3421 | } |
49048622 | 3422 | #endif |
49048622 | 3423 | |
7835b98b CL |
3424 | /* |
3425 | * This function gets called by the timer code, with HZ frequency. | |
3426 | * We call it with interrupts disabled. | |
3427 | * | |
3428 | * It also gets called by the fork code, when changing the parent's | |
3429 | * timeslices. | |
3430 | */ | |
3431 | void scheduler_tick(void) | |
3432 | { | |
7835b98b CL |
3433 | int cpu = smp_processor_id(); |
3434 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3435 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3436 | |
3437 | sched_clock_tick(); | |
dd41f596 | 3438 | |
05fa785c | 3439 | raw_spin_lock(&rq->lock); |
3e51f33f | 3440 | update_rq_clock(rq); |
f1a438d8 | 3441 | update_cpu_load(rq); |
fa85ae24 | 3442 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3443 | raw_spin_unlock(&rq->lock); |
7835b98b | 3444 | |
49f47433 | 3445 | perf_event_task_tick(curr); |
e220d2dc | 3446 | |
e418e1c2 | 3447 | #ifdef CONFIG_SMP |
dd41f596 IM |
3448 | rq->idle_at_tick = idle_cpu(cpu); |
3449 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3450 | #endif |
1da177e4 LT |
3451 | } |
3452 | ||
132380a0 | 3453 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3454 | { |
3455 | if (in_lock_functions(addr)) { | |
3456 | addr = CALLER_ADDR2; | |
3457 | if (in_lock_functions(addr)) | |
3458 | addr = CALLER_ADDR3; | |
3459 | } | |
3460 | return addr; | |
3461 | } | |
1da177e4 | 3462 | |
7e49fcce SR |
3463 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3464 | defined(CONFIG_PREEMPT_TRACER)) | |
3465 | ||
43627582 | 3466 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3467 | { |
6cd8a4bb | 3468 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3469 | /* |
3470 | * Underflow? | |
3471 | */ | |
9a11b49a IM |
3472 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3473 | return; | |
6cd8a4bb | 3474 | #endif |
1da177e4 | 3475 | preempt_count() += val; |
6cd8a4bb | 3476 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3477 | /* |
3478 | * Spinlock count overflowing soon? | |
3479 | */ | |
33859f7f MOS |
3480 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3481 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3482 | #endif |
3483 | if (preempt_count() == val) | |
3484 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3485 | } |
3486 | EXPORT_SYMBOL(add_preempt_count); | |
3487 | ||
43627582 | 3488 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3489 | { |
6cd8a4bb | 3490 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3491 | /* |
3492 | * Underflow? | |
3493 | */ | |
01e3eb82 | 3494 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3495 | return; |
1da177e4 LT |
3496 | /* |
3497 | * Is the spinlock portion underflowing? | |
3498 | */ | |
9a11b49a IM |
3499 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3500 | !(preempt_count() & PREEMPT_MASK))) | |
3501 | return; | |
6cd8a4bb | 3502 | #endif |
9a11b49a | 3503 | |
6cd8a4bb SR |
3504 | if (preempt_count() == val) |
3505 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3506 | preempt_count() -= val; |
3507 | } | |
3508 | EXPORT_SYMBOL(sub_preempt_count); | |
3509 | ||
3510 | #endif | |
3511 | ||
3512 | /* | |
dd41f596 | 3513 | * Print scheduling while atomic bug: |
1da177e4 | 3514 | */ |
dd41f596 | 3515 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3516 | { |
838225b4 SS |
3517 | struct pt_regs *regs = get_irq_regs(); |
3518 | ||
3df0fc5b PZ |
3519 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3520 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3521 | |
dd41f596 | 3522 | debug_show_held_locks(prev); |
e21f5b15 | 3523 | print_modules(); |
dd41f596 IM |
3524 | if (irqs_disabled()) |
3525 | print_irqtrace_events(prev); | |
838225b4 SS |
3526 | |
3527 | if (regs) | |
3528 | show_regs(regs); | |
3529 | else | |
3530 | dump_stack(); | |
dd41f596 | 3531 | } |
1da177e4 | 3532 | |
dd41f596 IM |
3533 | /* |
3534 | * Various schedule()-time debugging checks and statistics: | |
3535 | */ | |
3536 | static inline void schedule_debug(struct task_struct *prev) | |
3537 | { | |
1da177e4 | 3538 | /* |
41a2d6cf | 3539 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3540 | * schedule() atomically, we ignore that path for now. |
3541 | * Otherwise, whine if we are scheduling when we should not be. | |
3542 | */ | |
3f33a7ce | 3543 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3544 | __schedule_bug(prev); |
3545 | ||
1da177e4 LT |
3546 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3547 | ||
2d72376b | 3548 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3549 | #ifdef CONFIG_SCHEDSTATS |
3550 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3551 | schedstat_inc(this_rq(), bkl_count); |
3552 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3553 | } |
3554 | #endif | |
dd41f596 IM |
3555 | } |
3556 | ||
6cecd084 | 3557 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3558 | { |
a64692a3 MG |
3559 | if (prev->se.on_rq) |
3560 | update_rq_clock(rq); | |
3561 | rq->skip_clock_update = 0; | |
6cecd084 | 3562 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3563 | } |
3564 | ||
dd41f596 IM |
3565 | /* |
3566 | * Pick up the highest-prio task: | |
3567 | */ | |
3568 | static inline struct task_struct * | |
b67802ea | 3569 | pick_next_task(struct rq *rq) |
dd41f596 | 3570 | { |
5522d5d5 | 3571 | const struct sched_class *class; |
dd41f596 | 3572 | struct task_struct *p; |
1da177e4 LT |
3573 | |
3574 | /* | |
dd41f596 IM |
3575 | * Optimization: we know that if all tasks are in |
3576 | * the fair class we can call that function directly: | |
1da177e4 | 3577 | */ |
dd41f596 | 3578 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3579 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3580 | if (likely(p)) |
3581 | return p; | |
1da177e4 LT |
3582 | } |
3583 | ||
dd41f596 IM |
3584 | class = sched_class_highest; |
3585 | for ( ; ; ) { | |
fb8d4724 | 3586 | p = class->pick_next_task(rq); |
dd41f596 IM |
3587 | if (p) |
3588 | return p; | |
3589 | /* | |
3590 | * Will never be NULL as the idle class always | |
3591 | * returns a non-NULL p: | |
3592 | */ | |
3593 | class = class->next; | |
3594 | } | |
3595 | } | |
1da177e4 | 3596 | |
dd41f596 IM |
3597 | /* |
3598 | * schedule() is the main scheduler function. | |
3599 | */ | |
ff743345 | 3600 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3601 | { |
3602 | struct task_struct *prev, *next; | |
67ca7bde | 3603 | unsigned long *switch_count; |
dd41f596 | 3604 | struct rq *rq; |
31656519 | 3605 | int cpu; |
dd41f596 | 3606 | |
ff743345 PZ |
3607 | need_resched: |
3608 | preempt_disable(); | |
dd41f596 IM |
3609 | cpu = smp_processor_id(); |
3610 | rq = cpu_rq(cpu); | |
25502a6c | 3611 | rcu_note_context_switch(cpu); |
dd41f596 IM |
3612 | prev = rq->curr; |
3613 | switch_count = &prev->nivcsw; | |
3614 | ||
3615 | release_kernel_lock(prev); | |
3616 | need_resched_nonpreemptible: | |
3617 | ||
3618 | schedule_debug(prev); | |
1da177e4 | 3619 | |
31656519 | 3620 | if (sched_feat(HRTICK)) |
f333fdc9 | 3621 | hrtick_clear(rq); |
8f4d37ec | 3622 | |
05fa785c | 3623 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3624 | clear_tsk_need_resched(prev); |
1da177e4 | 3625 | |
1da177e4 | 3626 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
16882c1e | 3627 | if (unlikely(signal_pending_state(prev->state, prev))) |
1da177e4 | 3628 | prev->state = TASK_RUNNING; |
16882c1e | 3629 | else |
371fd7e7 | 3630 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
dd41f596 | 3631 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3632 | } |
3633 | ||
3f029d3c | 3634 | pre_schedule(rq, prev); |
f65eda4f | 3635 | |
dd41f596 | 3636 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3637 | idle_balance(cpu, rq); |
1da177e4 | 3638 | |
df1c99d4 | 3639 | put_prev_task(rq, prev); |
b67802ea | 3640 | next = pick_next_task(rq); |
1da177e4 | 3641 | |
1da177e4 | 3642 | if (likely(prev != next)) { |
673a90a1 | 3643 | sched_info_switch(prev, next); |
49f47433 | 3644 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3645 | |
1da177e4 LT |
3646 | rq->nr_switches++; |
3647 | rq->curr = next; | |
3648 | ++*switch_count; | |
3649 | ||
dd41f596 | 3650 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3651 | /* |
3652 | * the context switch might have flipped the stack from under | |
3653 | * us, hence refresh the local variables. | |
3654 | */ | |
3655 | cpu = smp_processor_id(); | |
3656 | rq = cpu_rq(cpu); | |
1da177e4 | 3657 | } else |
05fa785c | 3658 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3659 | |
3f029d3c | 3660 | post_schedule(rq); |
1da177e4 | 3661 | |
6d558c3a YZ |
3662 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3663 | prev = rq->curr; | |
3664 | switch_count = &prev->nivcsw; | |
1da177e4 | 3665 | goto need_resched_nonpreemptible; |
6d558c3a | 3666 | } |
8f4d37ec | 3667 | |
1da177e4 | 3668 | preempt_enable_no_resched(); |
ff743345 | 3669 | if (need_resched()) |
1da177e4 LT |
3670 | goto need_resched; |
3671 | } | |
1da177e4 LT |
3672 | EXPORT_SYMBOL(schedule); |
3673 | ||
c08f7829 | 3674 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3675 | /* |
3676 | * Look out! "owner" is an entirely speculative pointer | |
3677 | * access and not reliable. | |
3678 | */ | |
3679 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3680 | { | |
3681 | unsigned int cpu; | |
3682 | struct rq *rq; | |
3683 | ||
3684 | if (!sched_feat(OWNER_SPIN)) | |
3685 | return 0; | |
3686 | ||
3687 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3688 | /* | |
3689 | * Need to access the cpu field knowing that | |
3690 | * DEBUG_PAGEALLOC could have unmapped it if | |
3691 | * the mutex owner just released it and exited. | |
3692 | */ | |
3693 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 3694 | return 0; |
0d66bf6d PZ |
3695 | #else |
3696 | cpu = owner->cpu; | |
3697 | #endif | |
3698 | ||
3699 | /* | |
3700 | * Even if the access succeeded (likely case), | |
3701 | * the cpu field may no longer be valid. | |
3702 | */ | |
3703 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 3704 | return 0; |
0d66bf6d PZ |
3705 | |
3706 | /* | |
3707 | * We need to validate that we can do a | |
3708 | * get_cpu() and that we have the percpu area. | |
3709 | */ | |
3710 | if (!cpu_online(cpu)) | |
4b402210 | 3711 | return 0; |
0d66bf6d PZ |
3712 | |
3713 | rq = cpu_rq(cpu); | |
3714 | ||
3715 | for (;;) { | |
3716 | /* | |
3717 | * Owner changed, break to re-assess state. | |
3718 | */ | |
3719 | if (lock->owner != owner) | |
3720 | break; | |
3721 | ||
3722 | /* | |
3723 | * Is that owner really running on that cpu? | |
3724 | */ | |
3725 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3726 | return 0; | |
3727 | ||
3728 | cpu_relax(); | |
3729 | } | |
4b402210 | 3730 | |
0d66bf6d PZ |
3731 | return 1; |
3732 | } | |
3733 | #endif | |
3734 | ||
1da177e4 LT |
3735 | #ifdef CONFIG_PREEMPT |
3736 | /* | |
2ed6e34f | 3737 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3738 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3739 | * occur there and call schedule directly. |
3740 | */ | |
3741 | asmlinkage void __sched preempt_schedule(void) | |
3742 | { | |
3743 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3744 | |
1da177e4 LT |
3745 | /* |
3746 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3747 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3748 | */ |
beed33a8 | 3749 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3750 | return; |
3751 | ||
3a5c359a AK |
3752 | do { |
3753 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3754 | schedule(); |
3a5c359a | 3755 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3756 | |
3a5c359a AK |
3757 | /* |
3758 | * Check again in case we missed a preemption opportunity | |
3759 | * between schedule and now. | |
3760 | */ | |
3761 | barrier(); | |
5ed0cec0 | 3762 | } while (need_resched()); |
1da177e4 | 3763 | } |
1da177e4 LT |
3764 | EXPORT_SYMBOL(preempt_schedule); |
3765 | ||
3766 | /* | |
2ed6e34f | 3767 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3768 | * off of irq context. |
3769 | * Note, that this is called and return with irqs disabled. This will | |
3770 | * protect us against recursive calling from irq. | |
3771 | */ | |
3772 | asmlinkage void __sched preempt_schedule_irq(void) | |
3773 | { | |
3774 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3775 | |
2ed6e34f | 3776 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3777 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3778 | ||
3a5c359a AK |
3779 | do { |
3780 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3781 | local_irq_enable(); |
3782 | schedule(); | |
3783 | local_irq_disable(); | |
3a5c359a | 3784 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3785 | |
3a5c359a AK |
3786 | /* |
3787 | * Check again in case we missed a preemption opportunity | |
3788 | * between schedule and now. | |
3789 | */ | |
3790 | barrier(); | |
5ed0cec0 | 3791 | } while (need_resched()); |
1da177e4 LT |
3792 | } |
3793 | ||
3794 | #endif /* CONFIG_PREEMPT */ | |
3795 | ||
63859d4f | 3796 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3797 | void *key) |
1da177e4 | 3798 | { |
63859d4f | 3799 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3800 | } |
1da177e4 LT |
3801 | EXPORT_SYMBOL(default_wake_function); |
3802 | ||
3803 | /* | |
41a2d6cf IM |
3804 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3805 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3806 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3807 | * | |
3808 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3809 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3810 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3811 | */ | |
78ddb08f | 3812 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3813 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3814 | { |
2e45874c | 3815 | wait_queue_t *curr, *next; |
1da177e4 | 3816 | |
2e45874c | 3817 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3818 | unsigned flags = curr->flags; |
3819 | ||
63859d4f | 3820 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3821 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3822 | break; |
3823 | } | |
3824 | } | |
3825 | ||
3826 | /** | |
3827 | * __wake_up - wake up threads blocked on a waitqueue. | |
3828 | * @q: the waitqueue | |
3829 | * @mode: which threads | |
3830 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3831 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3832 | * |
3833 | * It may be assumed that this function implies a write memory barrier before | |
3834 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3835 | */ |
7ad5b3a5 | 3836 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3837 | int nr_exclusive, void *key) |
1da177e4 LT |
3838 | { |
3839 | unsigned long flags; | |
3840 | ||
3841 | spin_lock_irqsave(&q->lock, flags); | |
3842 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3843 | spin_unlock_irqrestore(&q->lock, flags); | |
3844 | } | |
1da177e4 LT |
3845 | EXPORT_SYMBOL(__wake_up); |
3846 | ||
3847 | /* | |
3848 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3849 | */ | |
7ad5b3a5 | 3850 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3851 | { |
3852 | __wake_up_common(q, mode, 1, 0, NULL); | |
3853 | } | |
22c43c81 | 3854 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3855 | |
4ede816a DL |
3856 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3857 | { | |
3858 | __wake_up_common(q, mode, 1, 0, key); | |
3859 | } | |
3860 | ||
1da177e4 | 3861 | /** |
4ede816a | 3862 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3863 | * @q: the waitqueue |
3864 | * @mode: which threads | |
3865 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3866 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3867 | * |
3868 | * The sync wakeup differs that the waker knows that it will schedule | |
3869 | * away soon, so while the target thread will be woken up, it will not | |
3870 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3871 | * with each other. This can prevent needless bouncing between CPUs. | |
3872 | * | |
3873 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3874 | * |
3875 | * It may be assumed that this function implies a write memory barrier before | |
3876 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3877 | */ |
4ede816a DL |
3878 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3879 | int nr_exclusive, void *key) | |
1da177e4 LT |
3880 | { |
3881 | unsigned long flags; | |
7d478721 | 3882 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3883 | |
3884 | if (unlikely(!q)) | |
3885 | return; | |
3886 | ||
3887 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3888 | wake_flags = 0; |
1da177e4 LT |
3889 | |
3890 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3891 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3892 | spin_unlock_irqrestore(&q->lock, flags); |
3893 | } | |
4ede816a DL |
3894 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3895 | ||
3896 | /* | |
3897 | * __wake_up_sync - see __wake_up_sync_key() | |
3898 | */ | |
3899 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3900 | { | |
3901 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3902 | } | |
1da177e4 LT |
3903 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3904 | ||
65eb3dc6 KD |
3905 | /** |
3906 | * complete: - signals a single thread waiting on this completion | |
3907 | * @x: holds the state of this particular completion | |
3908 | * | |
3909 | * This will wake up a single thread waiting on this completion. Threads will be | |
3910 | * awakened in the same order in which they were queued. | |
3911 | * | |
3912 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3913 | * |
3914 | * It may be assumed that this function implies a write memory barrier before | |
3915 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3916 | */ |
b15136e9 | 3917 | void complete(struct completion *x) |
1da177e4 LT |
3918 | { |
3919 | unsigned long flags; | |
3920 | ||
3921 | spin_lock_irqsave(&x->wait.lock, flags); | |
3922 | x->done++; | |
d9514f6c | 3923 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3924 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3925 | } | |
3926 | EXPORT_SYMBOL(complete); | |
3927 | ||
65eb3dc6 KD |
3928 | /** |
3929 | * complete_all: - signals all threads waiting on this completion | |
3930 | * @x: holds the state of this particular completion | |
3931 | * | |
3932 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3933 | * |
3934 | * It may be assumed that this function implies a write memory barrier before | |
3935 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3936 | */ |
b15136e9 | 3937 | void complete_all(struct completion *x) |
1da177e4 LT |
3938 | { |
3939 | unsigned long flags; | |
3940 | ||
3941 | spin_lock_irqsave(&x->wait.lock, flags); | |
3942 | x->done += UINT_MAX/2; | |
d9514f6c | 3943 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3944 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3945 | } | |
3946 | EXPORT_SYMBOL(complete_all); | |
3947 | ||
8cbbe86d AK |
3948 | static inline long __sched |
3949 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3950 | { |
1da177e4 LT |
3951 | if (!x->done) { |
3952 | DECLARE_WAITQUEUE(wait, current); | |
3953 | ||
a93d2f17 | 3954 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3955 | do { |
94d3d824 | 3956 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3957 | timeout = -ERESTARTSYS; |
3958 | break; | |
8cbbe86d AK |
3959 | } |
3960 | __set_current_state(state); | |
1da177e4 LT |
3961 | spin_unlock_irq(&x->wait.lock); |
3962 | timeout = schedule_timeout(timeout); | |
3963 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 3964 | } while (!x->done && timeout); |
1da177e4 | 3965 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
3966 | if (!x->done) |
3967 | return timeout; | |
1da177e4 LT |
3968 | } |
3969 | x->done--; | |
ea71a546 | 3970 | return timeout ?: 1; |
1da177e4 | 3971 | } |
1da177e4 | 3972 | |
8cbbe86d AK |
3973 | static long __sched |
3974 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3975 | { |
1da177e4 LT |
3976 | might_sleep(); |
3977 | ||
3978 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 3979 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 3980 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
3981 | return timeout; |
3982 | } | |
1da177e4 | 3983 | |
65eb3dc6 KD |
3984 | /** |
3985 | * wait_for_completion: - waits for completion of a task | |
3986 | * @x: holds the state of this particular completion | |
3987 | * | |
3988 | * This waits to be signaled for completion of a specific task. It is NOT | |
3989 | * interruptible and there is no timeout. | |
3990 | * | |
3991 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
3992 | * and interrupt capability. Also see complete(). | |
3993 | */ | |
b15136e9 | 3994 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
3995 | { |
3996 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 3997 | } |
8cbbe86d | 3998 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 3999 | |
65eb3dc6 KD |
4000 | /** |
4001 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4002 | * @x: holds the state of this particular completion | |
4003 | * @timeout: timeout value in jiffies | |
4004 | * | |
4005 | * This waits for either a completion of a specific task to be signaled or for a | |
4006 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4007 | * interruptible. | |
4008 | */ | |
b15136e9 | 4009 | unsigned long __sched |
8cbbe86d | 4010 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4011 | { |
8cbbe86d | 4012 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4013 | } |
8cbbe86d | 4014 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4015 | |
65eb3dc6 KD |
4016 | /** |
4017 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4018 | * @x: holds the state of this particular completion | |
4019 | * | |
4020 | * This waits for completion of a specific task to be signaled. It is | |
4021 | * interruptible. | |
4022 | */ | |
8cbbe86d | 4023 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4024 | { |
51e97990 AK |
4025 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4026 | if (t == -ERESTARTSYS) | |
4027 | return t; | |
4028 | return 0; | |
0fec171c | 4029 | } |
8cbbe86d | 4030 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4031 | |
65eb3dc6 KD |
4032 | /** |
4033 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4034 | * @x: holds the state of this particular completion | |
4035 | * @timeout: timeout value in jiffies | |
4036 | * | |
4037 | * This waits for either a completion of a specific task to be signaled or for a | |
4038 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4039 | */ | |
b15136e9 | 4040 | unsigned long __sched |
8cbbe86d AK |
4041 | wait_for_completion_interruptible_timeout(struct completion *x, |
4042 | unsigned long timeout) | |
0fec171c | 4043 | { |
8cbbe86d | 4044 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4045 | } |
8cbbe86d | 4046 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4047 | |
65eb3dc6 KD |
4048 | /** |
4049 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4050 | * @x: holds the state of this particular completion | |
4051 | * | |
4052 | * This waits to be signaled for completion of a specific task. It can be | |
4053 | * interrupted by a kill signal. | |
4054 | */ | |
009e577e MW |
4055 | int __sched wait_for_completion_killable(struct completion *x) |
4056 | { | |
4057 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4058 | if (t == -ERESTARTSYS) | |
4059 | return t; | |
4060 | return 0; | |
4061 | } | |
4062 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4063 | ||
be4de352 DC |
4064 | /** |
4065 | * try_wait_for_completion - try to decrement a completion without blocking | |
4066 | * @x: completion structure | |
4067 | * | |
4068 | * Returns: 0 if a decrement cannot be done without blocking | |
4069 | * 1 if a decrement succeeded. | |
4070 | * | |
4071 | * If a completion is being used as a counting completion, | |
4072 | * attempt to decrement the counter without blocking. This | |
4073 | * enables us to avoid waiting if the resource the completion | |
4074 | * is protecting is not available. | |
4075 | */ | |
4076 | bool try_wait_for_completion(struct completion *x) | |
4077 | { | |
7539a3b3 | 4078 | unsigned long flags; |
be4de352 DC |
4079 | int ret = 1; |
4080 | ||
7539a3b3 | 4081 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4082 | if (!x->done) |
4083 | ret = 0; | |
4084 | else | |
4085 | x->done--; | |
7539a3b3 | 4086 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4087 | return ret; |
4088 | } | |
4089 | EXPORT_SYMBOL(try_wait_for_completion); | |
4090 | ||
4091 | /** | |
4092 | * completion_done - Test to see if a completion has any waiters | |
4093 | * @x: completion structure | |
4094 | * | |
4095 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4096 | * 1 if there are no waiters. | |
4097 | * | |
4098 | */ | |
4099 | bool completion_done(struct completion *x) | |
4100 | { | |
7539a3b3 | 4101 | unsigned long flags; |
be4de352 DC |
4102 | int ret = 1; |
4103 | ||
7539a3b3 | 4104 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4105 | if (!x->done) |
4106 | ret = 0; | |
7539a3b3 | 4107 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4108 | return ret; |
4109 | } | |
4110 | EXPORT_SYMBOL(completion_done); | |
4111 | ||
8cbbe86d AK |
4112 | static long __sched |
4113 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4114 | { |
0fec171c IM |
4115 | unsigned long flags; |
4116 | wait_queue_t wait; | |
4117 | ||
4118 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4119 | |
8cbbe86d | 4120 | __set_current_state(state); |
1da177e4 | 4121 | |
8cbbe86d AK |
4122 | spin_lock_irqsave(&q->lock, flags); |
4123 | __add_wait_queue(q, &wait); | |
4124 | spin_unlock(&q->lock); | |
4125 | timeout = schedule_timeout(timeout); | |
4126 | spin_lock_irq(&q->lock); | |
4127 | __remove_wait_queue(q, &wait); | |
4128 | spin_unlock_irqrestore(&q->lock, flags); | |
4129 | ||
4130 | return timeout; | |
4131 | } | |
4132 | ||
4133 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4134 | { | |
4135 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4136 | } |
1da177e4 LT |
4137 | EXPORT_SYMBOL(interruptible_sleep_on); |
4138 | ||
0fec171c | 4139 | long __sched |
95cdf3b7 | 4140 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4141 | { |
8cbbe86d | 4142 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4143 | } |
1da177e4 LT |
4144 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4145 | ||
0fec171c | 4146 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4147 | { |
8cbbe86d | 4148 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4149 | } |
1da177e4 LT |
4150 | EXPORT_SYMBOL(sleep_on); |
4151 | ||
0fec171c | 4152 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4153 | { |
8cbbe86d | 4154 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4155 | } |
1da177e4 LT |
4156 | EXPORT_SYMBOL(sleep_on_timeout); |
4157 | ||
b29739f9 IM |
4158 | #ifdef CONFIG_RT_MUTEXES |
4159 | ||
4160 | /* | |
4161 | * rt_mutex_setprio - set the current priority of a task | |
4162 | * @p: task | |
4163 | * @prio: prio value (kernel-internal form) | |
4164 | * | |
4165 | * This function changes the 'effective' priority of a task. It does | |
4166 | * not touch ->normal_prio like __setscheduler(). | |
4167 | * | |
4168 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4169 | */ | |
36c8b586 | 4170 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4171 | { |
4172 | unsigned long flags; | |
83b699ed | 4173 | int oldprio, on_rq, running; |
70b97a7f | 4174 | struct rq *rq; |
83ab0aa0 | 4175 | const struct sched_class *prev_class; |
b29739f9 IM |
4176 | |
4177 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4178 | ||
4179 | rq = task_rq_lock(p, &flags); | |
4180 | ||
d5f9f942 | 4181 | oldprio = p->prio; |
83ab0aa0 | 4182 | prev_class = p->sched_class; |
dd41f596 | 4183 | on_rq = p->se.on_rq; |
051a1d1a | 4184 | running = task_current(rq, p); |
0e1f3483 | 4185 | if (on_rq) |
69be72c1 | 4186 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4187 | if (running) |
4188 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4189 | |
4190 | if (rt_prio(prio)) | |
4191 | p->sched_class = &rt_sched_class; | |
4192 | else | |
4193 | p->sched_class = &fair_sched_class; | |
4194 | ||
b29739f9 IM |
4195 | p->prio = prio; |
4196 | ||
0e1f3483 HS |
4197 | if (running) |
4198 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4199 | if (on_rq) { |
371fd7e7 | 4200 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4201 | |
4202 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4203 | } |
4204 | task_rq_unlock(rq, &flags); | |
4205 | } | |
4206 | ||
4207 | #endif | |
4208 | ||
36c8b586 | 4209 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4210 | { |
dd41f596 | 4211 | int old_prio, delta, on_rq; |
1da177e4 | 4212 | unsigned long flags; |
70b97a7f | 4213 | struct rq *rq; |
1da177e4 LT |
4214 | |
4215 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4216 | return; | |
4217 | /* | |
4218 | * We have to be careful, if called from sys_setpriority(), | |
4219 | * the task might be in the middle of scheduling on another CPU. | |
4220 | */ | |
4221 | rq = task_rq_lock(p, &flags); | |
4222 | /* | |
4223 | * The RT priorities are set via sched_setscheduler(), but we still | |
4224 | * allow the 'normal' nice value to be set - but as expected | |
4225 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4226 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4227 | */ |
e05606d3 | 4228 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4229 | p->static_prio = NICE_TO_PRIO(nice); |
4230 | goto out_unlock; | |
4231 | } | |
dd41f596 | 4232 | on_rq = p->se.on_rq; |
c09595f6 | 4233 | if (on_rq) |
69be72c1 | 4234 | dequeue_task(rq, p, 0); |
1da177e4 | 4235 | |
1da177e4 | 4236 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4237 | set_load_weight(p); |
b29739f9 IM |
4238 | old_prio = p->prio; |
4239 | p->prio = effective_prio(p); | |
4240 | delta = p->prio - old_prio; | |
1da177e4 | 4241 | |
dd41f596 | 4242 | if (on_rq) { |
371fd7e7 | 4243 | enqueue_task(rq, p, 0); |
1da177e4 | 4244 | /* |
d5f9f942 AM |
4245 | * If the task increased its priority or is running and |
4246 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4247 | */ |
d5f9f942 | 4248 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4249 | resched_task(rq->curr); |
4250 | } | |
4251 | out_unlock: | |
4252 | task_rq_unlock(rq, &flags); | |
4253 | } | |
1da177e4 LT |
4254 | EXPORT_SYMBOL(set_user_nice); |
4255 | ||
e43379f1 MM |
4256 | /* |
4257 | * can_nice - check if a task can reduce its nice value | |
4258 | * @p: task | |
4259 | * @nice: nice value | |
4260 | */ | |
36c8b586 | 4261 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4262 | { |
024f4747 MM |
4263 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4264 | int nice_rlim = 20 - nice; | |
48f24c4d | 4265 | |
78d7d407 | 4266 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4267 | capable(CAP_SYS_NICE)); |
4268 | } | |
4269 | ||
1da177e4 LT |
4270 | #ifdef __ARCH_WANT_SYS_NICE |
4271 | ||
4272 | /* | |
4273 | * sys_nice - change the priority of the current process. | |
4274 | * @increment: priority increment | |
4275 | * | |
4276 | * sys_setpriority is a more generic, but much slower function that | |
4277 | * does similar things. | |
4278 | */ | |
5add95d4 | 4279 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4280 | { |
48f24c4d | 4281 | long nice, retval; |
1da177e4 LT |
4282 | |
4283 | /* | |
4284 | * Setpriority might change our priority at the same moment. | |
4285 | * We don't have to worry. Conceptually one call occurs first | |
4286 | * and we have a single winner. | |
4287 | */ | |
e43379f1 MM |
4288 | if (increment < -40) |
4289 | increment = -40; | |
1da177e4 LT |
4290 | if (increment > 40) |
4291 | increment = 40; | |
4292 | ||
2b8f836f | 4293 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4294 | if (nice < -20) |
4295 | nice = -20; | |
4296 | if (nice > 19) | |
4297 | nice = 19; | |
4298 | ||
e43379f1 MM |
4299 | if (increment < 0 && !can_nice(current, nice)) |
4300 | return -EPERM; | |
4301 | ||
1da177e4 LT |
4302 | retval = security_task_setnice(current, nice); |
4303 | if (retval) | |
4304 | return retval; | |
4305 | ||
4306 | set_user_nice(current, nice); | |
4307 | return 0; | |
4308 | } | |
4309 | ||
4310 | #endif | |
4311 | ||
4312 | /** | |
4313 | * task_prio - return the priority value of a given task. | |
4314 | * @p: the task in question. | |
4315 | * | |
4316 | * This is the priority value as seen by users in /proc. | |
4317 | * RT tasks are offset by -200. Normal tasks are centered | |
4318 | * around 0, value goes from -16 to +15. | |
4319 | */ | |
36c8b586 | 4320 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4321 | { |
4322 | return p->prio - MAX_RT_PRIO; | |
4323 | } | |
4324 | ||
4325 | /** | |
4326 | * task_nice - return the nice value of a given task. | |
4327 | * @p: the task in question. | |
4328 | */ | |
36c8b586 | 4329 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4330 | { |
4331 | return TASK_NICE(p); | |
4332 | } | |
150d8bed | 4333 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4334 | |
4335 | /** | |
4336 | * idle_cpu - is a given cpu idle currently? | |
4337 | * @cpu: the processor in question. | |
4338 | */ | |
4339 | int idle_cpu(int cpu) | |
4340 | { | |
4341 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4342 | } | |
4343 | ||
1da177e4 LT |
4344 | /** |
4345 | * idle_task - return the idle task for a given cpu. | |
4346 | * @cpu: the processor in question. | |
4347 | */ | |
36c8b586 | 4348 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4349 | { |
4350 | return cpu_rq(cpu)->idle; | |
4351 | } | |
4352 | ||
4353 | /** | |
4354 | * find_process_by_pid - find a process with a matching PID value. | |
4355 | * @pid: the pid in question. | |
4356 | */ | |
a9957449 | 4357 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4358 | { |
228ebcbe | 4359 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4360 | } |
4361 | ||
4362 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4363 | static void |
4364 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4365 | { |
dd41f596 | 4366 | BUG_ON(p->se.on_rq); |
48f24c4d | 4367 | |
1da177e4 LT |
4368 | p->policy = policy; |
4369 | p->rt_priority = prio; | |
b29739f9 IM |
4370 | p->normal_prio = normal_prio(p); |
4371 | /* we are holding p->pi_lock already */ | |
4372 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4373 | if (rt_prio(p->prio)) |
4374 | p->sched_class = &rt_sched_class; | |
4375 | else | |
4376 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4377 | set_load_weight(p); |
1da177e4 LT |
4378 | } |
4379 | ||
c69e8d9c DH |
4380 | /* |
4381 | * check the target process has a UID that matches the current process's | |
4382 | */ | |
4383 | static bool check_same_owner(struct task_struct *p) | |
4384 | { | |
4385 | const struct cred *cred = current_cred(), *pcred; | |
4386 | bool match; | |
4387 | ||
4388 | rcu_read_lock(); | |
4389 | pcred = __task_cred(p); | |
4390 | match = (cred->euid == pcred->euid || | |
4391 | cred->euid == pcred->uid); | |
4392 | rcu_read_unlock(); | |
4393 | return match; | |
4394 | } | |
4395 | ||
961ccddd RR |
4396 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4397 | struct sched_param *param, bool user) | |
1da177e4 | 4398 | { |
83b699ed | 4399 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4400 | unsigned long flags; |
83ab0aa0 | 4401 | const struct sched_class *prev_class; |
70b97a7f | 4402 | struct rq *rq; |
ca94c442 | 4403 | int reset_on_fork; |
1da177e4 | 4404 | |
66e5393a SR |
4405 | /* may grab non-irq protected spin_locks */ |
4406 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4407 | recheck: |
4408 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4409 | if (policy < 0) { |
4410 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4411 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4412 | } else { |
4413 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4414 | policy &= ~SCHED_RESET_ON_FORK; | |
4415 | ||
4416 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4417 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4418 | policy != SCHED_IDLE) | |
4419 | return -EINVAL; | |
4420 | } | |
4421 | ||
1da177e4 LT |
4422 | /* |
4423 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4424 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4425 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4426 | */ |
4427 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4428 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4429 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4430 | return -EINVAL; |
e05606d3 | 4431 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4432 | return -EINVAL; |
4433 | ||
37e4ab3f OC |
4434 | /* |
4435 | * Allow unprivileged RT tasks to decrease priority: | |
4436 | */ | |
961ccddd | 4437 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4438 | if (rt_policy(policy)) { |
8dc3e909 | 4439 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4440 | |
4441 | if (!lock_task_sighand(p, &flags)) | |
4442 | return -ESRCH; | |
78d7d407 | 4443 | rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); |
8dc3e909 ON |
4444 | unlock_task_sighand(p, &flags); |
4445 | ||
4446 | /* can't set/change the rt policy */ | |
4447 | if (policy != p->policy && !rlim_rtprio) | |
4448 | return -EPERM; | |
4449 | ||
4450 | /* can't increase priority */ | |
4451 | if (param->sched_priority > p->rt_priority && | |
4452 | param->sched_priority > rlim_rtprio) | |
4453 | return -EPERM; | |
4454 | } | |
dd41f596 IM |
4455 | /* |
4456 | * Like positive nice levels, dont allow tasks to | |
4457 | * move out of SCHED_IDLE either: | |
4458 | */ | |
4459 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4460 | return -EPERM; | |
5fe1d75f | 4461 | |
37e4ab3f | 4462 | /* can't change other user's priorities */ |
c69e8d9c | 4463 | if (!check_same_owner(p)) |
37e4ab3f | 4464 | return -EPERM; |
ca94c442 LP |
4465 | |
4466 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4467 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4468 | return -EPERM; | |
37e4ab3f | 4469 | } |
1da177e4 | 4470 | |
725aad24 | 4471 | if (user) { |
b68aa230 | 4472 | #ifdef CONFIG_RT_GROUP_SCHED |
725aad24 JF |
4473 | /* |
4474 | * Do not allow realtime tasks into groups that have no runtime | |
4475 | * assigned. | |
4476 | */ | |
9a7e0b18 PZ |
4477 | if (rt_bandwidth_enabled() && rt_policy(policy) && |
4478 | task_group(p)->rt_bandwidth.rt_runtime == 0) | |
725aad24 | 4479 | return -EPERM; |
b68aa230 PZ |
4480 | #endif |
4481 | ||
725aad24 JF |
4482 | retval = security_task_setscheduler(p, policy, param); |
4483 | if (retval) | |
4484 | return retval; | |
4485 | } | |
4486 | ||
b29739f9 IM |
4487 | /* |
4488 | * make sure no PI-waiters arrive (or leave) while we are | |
4489 | * changing the priority of the task: | |
4490 | */ | |
1d615482 | 4491 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4492 | /* |
4493 | * To be able to change p->policy safely, the apropriate | |
4494 | * runqueue lock must be held. | |
4495 | */ | |
b29739f9 | 4496 | rq = __task_rq_lock(p); |
1da177e4 LT |
4497 | /* recheck policy now with rq lock held */ |
4498 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4499 | policy = oldpolicy = -1; | |
b29739f9 | 4500 | __task_rq_unlock(rq); |
1d615482 | 4501 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4502 | goto recheck; |
4503 | } | |
dd41f596 | 4504 | on_rq = p->se.on_rq; |
051a1d1a | 4505 | running = task_current(rq, p); |
0e1f3483 | 4506 | if (on_rq) |
2e1cb74a | 4507 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4508 | if (running) |
4509 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4510 | |
ca94c442 LP |
4511 | p->sched_reset_on_fork = reset_on_fork; |
4512 | ||
1da177e4 | 4513 | oldprio = p->prio; |
83ab0aa0 | 4514 | prev_class = p->sched_class; |
dd41f596 | 4515 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4516 | |
0e1f3483 HS |
4517 | if (running) |
4518 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4519 | if (on_rq) { |
4520 | activate_task(rq, p, 0); | |
cb469845 SR |
4521 | |
4522 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4523 | } |
b29739f9 | 4524 | __task_rq_unlock(rq); |
1d615482 | 4525 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4526 | |
95e02ca9 TG |
4527 | rt_mutex_adjust_pi(p); |
4528 | ||
1da177e4 LT |
4529 | return 0; |
4530 | } | |
961ccddd RR |
4531 | |
4532 | /** | |
4533 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4534 | * @p: the task in question. | |
4535 | * @policy: new policy. | |
4536 | * @param: structure containing the new RT priority. | |
4537 | * | |
4538 | * NOTE that the task may be already dead. | |
4539 | */ | |
4540 | int sched_setscheduler(struct task_struct *p, int policy, | |
4541 | struct sched_param *param) | |
4542 | { | |
4543 | return __sched_setscheduler(p, policy, param, true); | |
4544 | } | |
1da177e4 LT |
4545 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4546 | ||
961ccddd RR |
4547 | /** |
4548 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4549 | * @p: the task in question. | |
4550 | * @policy: new policy. | |
4551 | * @param: structure containing the new RT priority. | |
4552 | * | |
4553 | * Just like sched_setscheduler, only don't bother checking if the | |
4554 | * current context has permission. For example, this is needed in | |
4555 | * stop_machine(): we create temporary high priority worker threads, | |
4556 | * but our caller might not have that capability. | |
4557 | */ | |
4558 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4559 | struct sched_param *param) | |
4560 | { | |
4561 | return __sched_setscheduler(p, policy, param, false); | |
4562 | } | |
4563 | ||
95cdf3b7 IM |
4564 | static int |
4565 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4566 | { |
1da177e4 LT |
4567 | struct sched_param lparam; |
4568 | struct task_struct *p; | |
36c8b586 | 4569 | int retval; |
1da177e4 LT |
4570 | |
4571 | if (!param || pid < 0) | |
4572 | return -EINVAL; | |
4573 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4574 | return -EFAULT; | |
5fe1d75f ON |
4575 | |
4576 | rcu_read_lock(); | |
4577 | retval = -ESRCH; | |
1da177e4 | 4578 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4579 | if (p != NULL) |
4580 | retval = sched_setscheduler(p, policy, &lparam); | |
4581 | rcu_read_unlock(); | |
36c8b586 | 4582 | |
1da177e4 LT |
4583 | return retval; |
4584 | } | |
4585 | ||
4586 | /** | |
4587 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4588 | * @pid: the pid in question. | |
4589 | * @policy: new policy. | |
4590 | * @param: structure containing the new RT priority. | |
4591 | */ | |
5add95d4 HC |
4592 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4593 | struct sched_param __user *, param) | |
1da177e4 | 4594 | { |
c21761f1 JB |
4595 | /* negative values for policy are not valid */ |
4596 | if (policy < 0) | |
4597 | return -EINVAL; | |
4598 | ||
1da177e4 LT |
4599 | return do_sched_setscheduler(pid, policy, param); |
4600 | } | |
4601 | ||
4602 | /** | |
4603 | * sys_sched_setparam - set/change the RT priority of a thread | |
4604 | * @pid: the pid in question. | |
4605 | * @param: structure containing the new RT priority. | |
4606 | */ | |
5add95d4 | 4607 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4608 | { |
4609 | return do_sched_setscheduler(pid, -1, param); | |
4610 | } | |
4611 | ||
4612 | /** | |
4613 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4614 | * @pid: the pid in question. | |
4615 | */ | |
5add95d4 | 4616 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4617 | { |
36c8b586 | 4618 | struct task_struct *p; |
3a5c359a | 4619 | int retval; |
1da177e4 LT |
4620 | |
4621 | if (pid < 0) | |
3a5c359a | 4622 | return -EINVAL; |
1da177e4 LT |
4623 | |
4624 | retval = -ESRCH; | |
5fe85be0 | 4625 | rcu_read_lock(); |
1da177e4 LT |
4626 | p = find_process_by_pid(pid); |
4627 | if (p) { | |
4628 | retval = security_task_getscheduler(p); | |
4629 | if (!retval) | |
ca94c442 LP |
4630 | retval = p->policy |
4631 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4632 | } |
5fe85be0 | 4633 | rcu_read_unlock(); |
1da177e4 LT |
4634 | return retval; |
4635 | } | |
4636 | ||
4637 | /** | |
ca94c442 | 4638 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4639 | * @pid: the pid in question. |
4640 | * @param: structure containing the RT priority. | |
4641 | */ | |
5add95d4 | 4642 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4643 | { |
4644 | struct sched_param lp; | |
36c8b586 | 4645 | struct task_struct *p; |
3a5c359a | 4646 | int retval; |
1da177e4 LT |
4647 | |
4648 | if (!param || pid < 0) | |
3a5c359a | 4649 | return -EINVAL; |
1da177e4 | 4650 | |
5fe85be0 | 4651 | rcu_read_lock(); |
1da177e4 LT |
4652 | p = find_process_by_pid(pid); |
4653 | retval = -ESRCH; | |
4654 | if (!p) | |
4655 | goto out_unlock; | |
4656 | ||
4657 | retval = security_task_getscheduler(p); | |
4658 | if (retval) | |
4659 | goto out_unlock; | |
4660 | ||
4661 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4662 | rcu_read_unlock(); |
1da177e4 LT |
4663 | |
4664 | /* | |
4665 | * This one might sleep, we cannot do it with a spinlock held ... | |
4666 | */ | |
4667 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4668 | ||
1da177e4 LT |
4669 | return retval; |
4670 | ||
4671 | out_unlock: | |
5fe85be0 | 4672 | rcu_read_unlock(); |
1da177e4 LT |
4673 | return retval; |
4674 | } | |
4675 | ||
96f874e2 | 4676 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4677 | { |
5a16f3d3 | 4678 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4679 | struct task_struct *p; |
4680 | int retval; | |
1da177e4 | 4681 | |
95402b38 | 4682 | get_online_cpus(); |
23f5d142 | 4683 | rcu_read_lock(); |
1da177e4 LT |
4684 | |
4685 | p = find_process_by_pid(pid); | |
4686 | if (!p) { | |
23f5d142 | 4687 | rcu_read_unlock(); |
95402b38 | 4688 | put_online_cpus(); |
1da177e4 LT |
4689 | return -ESRCH; |
4690 | } | |
4691 | ||
23f5d142 | 4692 | /* Prevent p going away */ |
1da177e4 | 4693 | get_task_struct(p); |
23f5d142 | 4694 | rcu_read_unlock(); |
1da177e4 | 4695 | |
5a16f3d3 RR |
4696 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4697 | retval = -ENOMEM; | |
4698 | goto out_put_task; | |
4699 | } | |
4700 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4701 | retval = -ENOMEM; | |
4702 | goto out_free_cpus_allowed; | |
4703 | } | |
1da177e4 | 4704 | retval = -EPERM; |
c69e8d9c | 4705 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4706 | goto out_unlock; |
4707 | ||
e7834f8f DQ |
4708 | retval = security_task_setscheduler(p, 0, NULL); |
4709 | if (retval) | |
4710 | goto out_unlock; | |
4711 | ||
5a16f3d3 RR |
4712 | cpuset_cpus_allowed(p, cpus_allowed); |
4713 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4714 | again: |
5a16f3d3 | 4715 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4716 | |
8707d8b8 | 4717 | if (!retval) { |
5a16f3d3 RR |
4718 | cpuset_cpus_allowed(p, cpus_allowed); |
4719 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4720 | /* |
4721 | * We must have raced with a concurrent cpuset | |
4722 | * update. Just reset the cpus_allowed to the | |
4723 | * cpuset's cpus_allowed | |
4724 | */ | |
5a16f3d3 | 4725 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4726 | goto again; |
4727 | } | |
4728 | } | |
1da177e4 | 4729 | out_unlock: |
5a16f3d3 RR |
4730 | free_cpumask_var(new_mask); |
4731 | out_free_cpus_allowed: | |
4732 | free_cpumask_var(cpus_allowed); | |
4733 | out_put_task: | |
1da177e4 | 4734 | put_task_struct(p); |
95402b38 | 4735 | put_online_cpus(); |
1da177e4 LT |
4736 | return retval; |
4737 | } | |
4738 | ||
4739 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4740 | struct cpumask *new_mask) |
1da177e4 | 4741 | { |
96f874e2 RR |
4742 | if (len < cpumask_size()) |
4743 | cpumask_clear(new_mask); | |
4744 | else if (len > cpumask_size()) | |
4745 | len = cpumask_size(); | |
4746 | ||
1da177e4 LT |
4747 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4748 | } | |
4749 | ||
4750 | /** | |
4751 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4752 | * @pid: pid of the process | |
4753 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4754 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4755 | */ | |
5add95d4 HC |
4756 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4757 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4758 | { |
5a16f3d3 | 4759 | cpumask_var_t new_mask; |
1da177e4 LT |
4760 | int retval; |
4761 | ||
5a16f3d3 RR |
4762 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4763 | return -ENOMEM; | |
1da177e4 | 4764 | |
5a16f3d3 RR |
4765 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4766 | if (retval == 0) | |
4767 | retval = sched_setaffinity(pid, new_mask); | |
4768 | free_cpumask_var(new_mask); | |
4769 | return retval; | |
1da177e4 LT |
4770 | } |
4771 | ||
96f874e2 | 4772 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4773 | { |
36c8b586 | 4774 | struct task_struct *p; |
31605683 TG |
4775 | unsigned long flags; |
4776 | struct rq *rq; | |
1da177e4 | 4777 | int retval; |
1da177e4 | 4778 | |
95402b38 | 4779 | get_online_cpus(); |
23f5d142 | 4780 | rcu_read_lock(); |
1da177e4 LT |
4781 | |
4782 | retval = -ESRCH; | |
4783 | p = find_process_by_pid(pid); | |
4784 | if (!p) | |
4785 | goto out_unlock; | |
4786 | ||
e7834f8f DQ |
4787 | retval = security_task_getscheduler(p); |
4788 | if (retval) | |
4789 | goto out_unlock; | |
4790 | ||
31605683 | 4791 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4792 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4793 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4794 | |
4795 | out_unlock: | |
23f5d142 | 4796 | rcu_read_unlock(); |
95402b38 | 4797 | put_online_cpus(); |
1da177e4 | 4798 | |
9531b62f | 4799 | return retval; |
1da177e4 LT |
4800 | } |
4801 | ||
4802 | /** | |
4803 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4804 | * @pid: pid of the process | |
4805 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4806 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4807 | */ | |
5add95d4 HC |
4808 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4809 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4810 | { |
4811 | int ret; | |
f17c8607 | 4812 | cpumask_var_t mask; |
1da177e4 | 4813 | |
84fba5ec | 4814 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4815 | return -EINVAL; |
4816 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4817 | return -EINVAL; |
4818 | ||
f17c8607 RR |
4819 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4820 | return -ENOMEM; | |
1da177e4 | 4821 | |
f17c8607 RR |
4822 | ret = sched_getaffinity(pid, mask); |
4823 | if (ret == 0) { | |
8bc037fb | 4824 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4825 | |
4826 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4827 | ret = -EFAULT; |
4828 | else | |
cd3d8031 | 4829 | ret = retlen; |
f17c8607 RR |
4830 | } |
4831 | free_cpumask_var(mask); | |
1da177e4 | 4832 | |
f17c8607 | 4833 | return ret; |
1da177e4 LT |
4834 | } |
4835 | ||
4836 | /** | |
4837 | * sys_sched_yield - yield the current processor to other threads. | |
4838 | * | |
dd41f596 IM |
4839 | * This function yields the current CPU to other tasks. If there are no |
4840 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4841 | */ |
5add95d4 | 4842 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4843 | { |
70b97a7f | 4844 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4845 | |
2d72376b | 4846 | schedstat_inc(rq, yld_count); |
4530d7ab | 4847 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4848 | |
4849 | /* | |
4850 | * Since we are going to call schedule() anyway, there's | |
4851 | * no need to preempt or enable interrupts: | |
4852 | */ | |
4853 | __release(rq->lock); | |
8a25d5de | 4854 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4855 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4856 | preempt_enable_no_resched(); |
4857 | ||
4858 | schedule(); | |
4859 | ||
4860 | return 0; | |
4861 | } | |
4862 | ||
d86ee480 PZ |
4863 | static inline int should_resched(void) |
4864 | { | |
4865 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4866 | } | |
4867 | ||
e7b38404 | 4868 | static void __cond_resched(void) |
1da177e4 | 4869 | { |
e7aaaa69 FW |
4870 | add_preempt_count(PREEMPT_ACTIVE); |
4871 | schedule(); | |
4872 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
4873 | } |
4874 | ||
02b67cc3 | 4875 | int __sched _cond_resched(void) |
1da177e4 | 4876 | { |
d86ee480 | 4877 | if (should_resched()) { |
1da177e4 LT |
4878 | __cond_resched(); |
4879 | return 1; | |
4880 | } | |
4881 | return 0; | |
4882 | } | |
02b67cc3 | 4883 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4884 | |
4885 | /* | |
613afbf8 | 4886 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4887 | * call schedule, and on return reacquire the lock. |
4888 | * | |
41a2d6cf | 4889 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4890 | * operations here to prevent schedule() from being called twice (once via |
4891 | * spin_unlock(), once by hand). | |
4892 | */ | |
613afbf8 | 4893 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4894 | { |
d86ee480 | 4895 | int resched = should_resched(); |
6df3cecb JK |
4896 | int ret = 0; |
4897 | ||
f607c668 PZ |
4898 | lockdep_assert_held(lock); |
4899 | ||
95c354fe | 4900 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4901 | spin_unlock(lock); |
d86ee480 | 4902 | if (resched) |
95c354fe NP |
4903 | __cond_resched(); |
4904 | else | |
4905 | cpu_relax(); | |
6df3cecb | 4906 | ret = 1; |
1da177e4 | 4907 | spin_lock(lock); |
1da177e4 | 4908 | } |
6df3cecb | 4909 | return ret; |
1da177e4 | 4910 | } |
613afbf8 | 4911 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4912 | |
613afbf8 | 4913 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4914 | { |
4915 | BUG_ON(!in_softirq()); | |
4916 | ||
d86ee480 | 4917 | if (should_resched()) { |
98d82567 | 4918 | local_bh_enable(); |
1da177e4 LT |
4919 | __cond_resched(); |
4920 | local_bh_disable(); | |
4921 | return 1; | |
4922 | } | |
4923 | return 0; | |
4924 | } | |
613afbf8 | 4925 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4926 | |
1da177e4 LT |
4927 | /** |
4928 | * yield - yield the current processor to other threads. | |
4929 | * | |
72fd4a35 | 4930 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4931 | * thread runnable and calls sys_sched_yield(). |
4932 | */ | |
4933 | void __sched yield(void) | |
4934 | { | |
4935 | set_current_state(TASK_RUNNING); | |
4936 | sys_sched_yield(); | |
4937 | } | |
1da177e4 LT |
4938 | EXPORT_SYMBOL(yield); |
4939 | ||
4940 | /* | |
41a2d6cf | 4941 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 4942 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
4943 | */ |
4944 | void __sched io_schedule(void) | |
4945 | { | |
54d35f29 | 4946 | struct rq *rq = raw_rq(); |
1da177e4 | 4947 | |
0ff92245 | 4948 | delayacct_blkio_start(); |
1da177e4 | 4949 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4950 | current->in_iowait = 1; |
1da177e4 | 4951 | schedule(); |
8f0dfc34 | 4952 | current->in_iowait = 0; |
1da177e4 | 4953 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4954 | delayacct_blkio_end(); |
1da177e4 | 4955 | } |
1da177e4 LT |
4956 | EXPORT_SYMBOL(io_schedule); |
4957 | ||
4958 | long __sched io_schedule_timeout(long timeout) | |
4959 | { | |
54d35f29 | 4960 | struct rq *rq = raw_rq(); |
1da177e4 LT |
4961 | long ret; |
4962 | ||
0ff92245 | 4963 | delayacct_blkio_start(); |
1da177e4 | 4964 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 4965 | current->in_iowait = 1; |
1da177e4 | 4966 | ret = schedule_timeout(timeout); |
8f0dfc34 | 4967 | current->in_iowait = 0; |
1da177e4 | 4968 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 4969 | delayacct_blkio_end(); |
1da177e4 LT |
4970 | return ret; |
4971 | } | |
4972 | ||
4973 | /** | |
4974 | * sys_sched_get_priority_max - return maximum RT priority. | |
4975 | * @policy: scheduling class. | |
4976 | * | |
4977 | * this syscall returns the maximum rt_priority that can be used | |
4978 | * by a given scheduling class. | |
4979 | */ | |
5add95d4 | 4980 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
4981 | { |
4982 | int ret = -EINVAL; | |
4983 | ||
4984 | switch (policy) { | |
4985 | case SCHED_FIFO: | |
4986 | case SCHED_RR: | |
4987 | ret = MAX_USER_RT_PRIO-1; | |
4988 | break; | |
4989 | case SCHED_NORMAL: | |
b0a9499c | 4990 | case SCHED_BATCH: |
dd41f596 | 4991 | case SCHED_IDLE: |
1da177e4 LT |
4992 | ret = 0; |
4993 | break; | |
4994 | } | |
4995 | return ret; | |
4996 | } | |
4997 | ||
4998 | /** | |
4999 | * sys_sched_get_priority_min - return minimum RT priority. | |
5000 | * @policy: scheduling class. | |
5001 | * | |
5002 | * this syscall returns the minimum rt_priority that can be used | |
5003 | * by a given scheduling class. | |
5004 | */ | |
5add95d4 | 5005 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5006 | { |
5007 | int ret = -EINVAL; | |
5008 | ||
5009 | switch (policy) { | |
5010 | case SCHED_FIFO: | |
5011 | case SCHED_RR: | |
5012 | ret = 1; | |
5013 | break; | |
5014 | case SCHED_NORMAL: | |
b0a9499c | 5015 | case SCHED_BATCH: |
dd41f596 | 5016 | case SCHED_IDLE: |
1da177e4 LT |
5017 | ret = 0; |
5018 | } | |
5019 | return ret; | |
5020 | } | |
5021 | ||
5022 | /** | |
5023 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5024 | * @pid: pid of the process. | |
5025 | * @interval: userspace pointer to the timeslice value. | |
5026 | * | |
5027 | * this syscall writes the default timeslice value of a given process | |
5028 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5029 | */ | |
17da2bd9 | 5030 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5031 | struct timespec __user *, interval) |
1da177e4 | 5032 | { |
36c8b586 | 5033 | struct task_struct *p; |
a4ec24b4 | 5034 | unsigned int time_slice; |
dba091b9 TG |
5035 | unsigned long flags; |
5036 | struct rq *rq; | |
3a5c359a | 5037 | int retval; |
1da177e4 | 5038 | struct timespec t; |
1da177e4 LT |
5039 | |
5040 | if (pid < 0) | |
3a5c359a | 5041 | return -EINVAL; |
1da177e4 LT |
5042 | |
5043 | retval = -ESRCH; | |
1a551ae7 | 5044 | rcu_read_lock(); |
1da177e4 LT |
5045 | p = find_process_by_pid(pid); |
5046 | if (!p) | |
5047 | goto out_unlock; | |
5048 | ||
5049 | retval = security_task_getscheduler(p); | |
5050 | if (retval) | |
5051 | goto out_unlock; | |
5052 | ||
dba091b9 TG |
5053 | rq = task_rq_lock(p, &flags); |
5054 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5055 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5056 | |
1a551ae7 | 5057 | rcu_read_unlock(); |
a4ec24b4 | 5058 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5059 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5060 | return retval; |
3a5c359a | 5061 | |
1da177e4 | 5062 | out_unlock: |
1a551ae7 | 5063 | rcu_read_unlock(); |
1da177e4 LT |
5064 | return retval; |
5065 | } | |
5066 | ||
7c731e0a | 5067 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5068 | |
82a1fcb9 | 5069 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5070 | { |
1da177e4 | 5071 | unsigned long free = 0; |
36c8b586 | 5072 | unsigned state; |
1da177e4 | 5073 | |
1da177e4 | 5074 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5075 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5076 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5077 | #if BITS_PER_LONG == 32 |
1da177e4 | 5078 | if (state == TASK_RUNNING) |
3df0fc5b | 5079 | printk(KERN_CONT " running "); |
1da177e4 | 5080 | else |
3df0fc5b | 5081 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5082 | #else |
5083 | if (state == TASK_RUNNING) | |
3df0fc5b | 5084 | printk(KERN_CONT " running task "); |
1da177e4 | 5085 | else |
3df0fc5b | 5086 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5087 | #endif |
5088 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5089 | free = stack_not_used(p); |
1da177e4 | 5090 | #endif |
3df0fc5b | 5091 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5092 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5093 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5094 | |
5fb5e6de | 5095 | show_stack(p, NULL); |
1da177e4 LT |
5096 | } |
5097 | ||
e59e2ae2 | 5098 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5099 | { |
36c8b586 | 5100 | struct task_struct *g, *p; |
1da177e4 | 5101 | |
4bd77321 | 5102 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5103 | printk(KERN_INFO |
5104 | " task PC stack pid father\n"); | |
1da177e4 | 5105 | #else |
3df0fc5b PZ |
5106 | printk(KERN_INFO |
5107 | " task PC stack pid father\n"); | |
1da177e4 LT |
5108 | #endif |
5109 | read_lock(&tasklist_lock); | |
5110 | do_each_thread(g, p) { | |
5111 | /* | |
5112 | * reset the NMI-timeout, listing all files on a slow | |
5113 | * console might take alot of time: | |
5114 | */ | |
5115 | touch_nmi_watchdog(); | |
39bc89fd | 5116 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5117 | sched_show_task(p); |
1da177e4 LT |
5118 | } while_each_thread(g, p); |
5119 | ||
04c9167f JF |
5120 | touch_all_softlockup_watchdogs(); |
5121 | ||
dd41f596 IM |
5122 | #ifdef CONFIG_SCHED_DEBUG |
5123 | sysrq_sched_debug_show(); | |
5124 | #endif | |
1da177e4 | 5125 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5126 | /* |
5127 | * Only show locks if all tasks are dumped: | |
5128 | */ | |
93335a21 | 5129 | if (!state_filter) |
e59e2ae2 | 5130 | debug_show_all_locks(); |
1da177e4 LT |
5131 | } |
5132 | ||
1df21055 IM |
5133 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5134 | { | |
dd41f596 | 5135 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5136 | } |
5137 | ||
f340c0d1 IM |
5138 | /** |
5139 | * init_idle - set up an idle thread for a given CPU | |
5140 | * @idle: task in question | |
5141 | * @cpu: cpu the idle task belongs to | |
5142 | * | |
5143 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5144 | * flag, to make booting more robust. | |
5145 | */ | |
5c1e1767 | 5146 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5147 | { |
70b97a7f | 5148 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5149 | unsigned long flags; |
5150 | ||
05fa785c | 5151 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5152 | |
dd41f596 | 5153 | __sched_fork(idle); |
06b83b5f | 5154 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5155 | idle->se.exec_start = sched_clock(); |
5156 | ||
96f874e2 | 5157 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5158 | __set_task_cpu(idle, cpu); |
1da177e4 | 5159 | |
1da177e4 | 5160 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5161 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5162 | idle->oncpu = 1; | |
5163 | #endif | |
05fa785c | 5164 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5165 | |
5166 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5167 | #if defined(CONFIG_PREEMPT) |
5168 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5169 | #else | |
a1261f54 | 5170 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5171 | #endif |
dd41f596 IM |
5172 | /* |
5173 | * The idle tasks have their own, simple scheduling class: | |
5174 | */ | |
5175 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5176 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5177 | } |
5178 | ||
5179 | /* | |
5180 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5181 | * indicates which cpus entered this state. This is used | |
5182 | * in the rcu update to wait only for active cpus. For system | |
5183 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5184 | * always be CPU_BITS_NONE. |
1da177e4 | 5185 | */ |
6a7b3dc3 | 5186 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5187 | |
19978ca6 IM |
5188 | /* |
5189 | * Increase the granularity value when there are more CPUs, | |
5190 | * because with more CPUs the 'effective latency' as visible | |
5191 | * to users decreases. But the relationship is not linear, | |
5192 | * so pick a second-best guess by going with the log2 of the | |
5193 | * number of CPUs. | |
5194 | * | |
5195 | * This idea comes from the SD scheduler of Con Kolivas: | |
5196 | */ | |
acb4a848 | 5197 | static int get_update_sysctl_factor(void) |
19978ca6 | 5198 | { |
4ca3ef71 | 5199 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5200 | unsigned int factor; |
5201 | ||
5202 | switch (sysctl_sched_tunable_scaling) { | |
5203 | case SCHED_TUNABLESCALING_NONE: | |
5204 | factor = 1; | |
5205 | break; | |
5206 | case SCHED_TUNABLESCALING_LINEAR: | |
5207 | factor = cpus; | |
5208 | break; | |
5209 | case SCHED_TUNABLESCALING_LOG: | |
5210 | default: | |
5211 | factor = 1 + ilog2(cpus); | |
5212 | break; | |
5213 | } | |
19978ca6 | 5214 | |
acb4a848 CE |
5215 | return factor; |
5216 | } | |
19978ca6 | 5217 | |
acb4a848 CE |
5218 | static void update_sysctl(void) |
5219 | { | |
5220 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5221 | |
0bcdcf28 CE |
5222 | #define SET_SYSCTL(name) \ |
5223 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5224 | SET_SYSCTL(sched_min_granularity); | |
5225 | SET_SYSCTL(sched_latency); | |
5226 | SET_SYSCTL(sched_wakeup_granularity); | |
5227 | SET_SYSCTL(sched_shares_ratelimit); | |
5228 | #undef SET_SYSCTL | |
5229 | } | |
55cd5340 | 5230 | |
0bcdcf28 CE |
5231 | static inline void sched_init_granularity(void) |
5232 | { | |
5233 | update_sysctl(); | |
19978ca6 IM |
5234 | } |
5235 | ||
1da177e4 LT |
5236 | #ifdef CONFIG_SMP |
5237 | /* | |
5238 | * This is how migration works: | |
5239 | * | |
969c7921 TH |
5240 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5241 | * stop_one_cpu(). | |
5242 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5243 | * off the CPU) | |
5244 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5245 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5246 | * it and puts it into the right queue. |
969c7921 TH |
5247 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5248 | * is done. | |
1da177e4 LT |
5249 | */ |
5250 | ||
5251 | /* | |
5252 | * Change a given task's CPU affinity. Migrate the thread to a | |
5253 | * proper CPU and schedule it away if the CPU it's executing on | |
5254 | * is removed from the allowed bitmask. | |
5255 | * | |
5256 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5257 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5258 | * call is not atomic; no spinlocks may be held. |
5259 | */ | |
96f874e2 | 5260 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5261 | { |
5262 | unsigned long flags; | |
70b97a7f | 5263 | struct rq *rq; |
969c7921 | 5264 | unsigned int dest_cpu; |
48f24c4d | 5265 | int ret = 0; |
1da177e4 | 5266 | |
65cc8e48 PZ |
5267 | /* |
5268 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5269 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5270 | */ | |
5271 | again: | |
5272 | while (task_is_waking(p)) | |
5273 | cpu_relax(); | |
1da177e4 | 5274 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5275 | if (task_is_waking(p)) { |
5276 | task_rq_unlock(rq, &flags); | |
5277 | goto again; | |
5278 | } | |
e2912009 | 5279 | |
6ad4c188 | 5280 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5281 | ret = -EINVAL; |
5282 | goto out; | |
5283 | } | |
5284 | ||
9985b0ba | 5285 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5286 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5287 | ret = -EINVAL; |
5288 | goto out; | |
5289 | } | |
5290 | ||
73fe6aae | 5291 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5292 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5293 | else { |
96f874e2 RR |
5294 | cpumask_copy(&p->cpus_allowed, new_mask); |
5295 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5296 | } |
5297 | ||
1da177e4 | 5298 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5299 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5300 | goto out; |
5301 | ||
969c7921 TH |
5302 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
5303 | if (migrate_task(p, dest_cpu)) { | |
5304 | struct migration_arg arg = { p, dest_cpu }; | |
1da177e4 LT |
5305 | /* Need help from migration thread: drop lock and wait. */ |
5306 | task_rq_unlock(rq, &flags); | |
969c7921 | 5307 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5308 | tlb_migrate_finish(p->mm); |
5309 | return 0; | |
5310 | } | |
5311 | out: | |
5312 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5313 | |
1da177e4 LT |
5314 | return ret; |
5315 | } | |
cd8ba7cd | 5316 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5317 | |
5318 | /* | |
41a2d6cf | 5319 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5320 | * this because either it can't run here any more (set_cpus_allowed() |
5321 | * away from this CPU, or CPU going down), or because we're | |
5322 | * attempting to rebalance this task on exec (sched_exec). | |
5323 | * | |
5324 | * So we race with normal scheduler movements, but that's OK, as long | |
5325 | * as the task is no longer on this CPU. | |
efc30814 KK |
5326 | * |
5327 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5328 | */ |
efc30814 | 5329 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5330 | { |
70b97a7f | 5331 | struct rq *rq_dest, *rq_src; |
e2912009 | 5332 | int ret = 0; |
1da177e4 | 5333 | |
e761b772 | 5334 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5335 | return ret; |
1da177e4 LT |
5336 | |
5337 | rq_src = cpu_rq(src_cpu); | |
5338 | rq_dest = cpu_rq(dest_cpu); | |
5339 | ||
5340 | double_rq_lock(rq_src, rq_dest); | |
5341 | /* Already moved. */ | |
5342 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5343 | goto done; |
1da177e4 | 5344 | /* Affinity changed (again). */ |
96f874e2 | 5345 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5346 | goto fail; |
1da177e4 | 5347 | |
e2912009 PZ |
5348 | /* |
5349 | * If we're not on a rq, the next wake-up will ensure we're | |
5350 | * placed properly. | |
5351 | */ | |
5352 | if (p->se.on_rq) { | |
2e1cb74a | 5353 | deactivate_task(rq_src, p, 0); |
e2912009 | 5354 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5355 | activate_task(rq_dest, p, 0); |
15afe09b | 5356 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5357 | } |
b1e38734 | 5358 | done: |
efc30814 | 5359 | ret = 1; |
b1e38734 | 5360 | fail: |
1da177e4 | 5361 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5362 | return ret; |
1da177e4 LT |
5363 | } |
5364 | ||
5365 | /* | |
969c7921 TH |
5366 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5367 | * and performs thread migration by bumping thread off CPU then | |
5368 | * 'pushing' onto another runqueue. | |
1da177e4 | 5369 | */ |
969c7921 | 5370 | static int migration_cpu_stop(void *data) |
1da177e4 | 5371 | { |
969c7921 | 5372 | struct migration_arg *arg = data; |
f7b4cddc | 5373 | |
969c7921 TH |
5374 | /* |
5375 | * The original target cpu might have gone down and we might | |
5376 | * be on another cpu but it doesn't matter. | |
5377 | */ | |
f7b4cddc | 5378 | local_irq_disable(); |
969c7921 | 5379 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5380 | local_irq_enable(); |
1da177e4 | 5381 | return 0; |
f7b4cddc ON |
5382 | } |
5383 | ||
1da177e4 | 5384 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5385 | /* |
3a4fa0a2 | 5386 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5387 | */ |
6a1bdc1b | 5388 | void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5389 | { |
1445c08d ON |
5390 | struct rq *rq = cpu_rq(dead_cpu); |
5391 | int needs_cpu, uninitialized_var(dest_cpu); | |
5392 | unsigned long flags; | |
e76bd8d9 | 5393 | |
1445c08d | 5394 | local_irq_save(flags); |
e76bd8d9 | 5395 | |
1445c08d ON |
5396 | raw_spin_lock(&rq->lock); |
5397 | needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING); | |
5398 | if (needs_cpu) | |
5399 | dest_cpu = select_fallback_rq(dead_cpu, p); | |
5400 | raw_spin_unlock(&rq->lock); | |
c1804d54 ON |
5401 | /* |
5402 | * It can only fail if we race with set_cpus_allowed(), | |
5403 | * in the racer should migrate the task anyway. | |
5404 | */ | |
1445c08d | 5405 | if (needs_cpu) |
c1804d54 | 5406 | __migrate_task(p, dead_cpu, dest_cpu); |
1445c08d | 5407 | local_irq_restore(flags); |
1da177e4 LT |
5408 | } |
5409 | ||
5410 | /* | |
5411 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5412 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5413 | * for performance reasons the counter is not stricly tracking tasks to | |
5414 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5415 | * to keep the global sum constant after CPU-down: | |
5416 | */ | |
70b97a7f | 5417 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5418 | { |
6ad4c188 | 5419 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5420 | unsigned long flags; |
5421 | ||
5422 | local_irq_save(flags); | |
5423 | double_rq_lock(rq_src, rq_dest); | |
5424 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5425 | rq_src->nr_uninterruptible = 0; | |
5426 | double_rq_unlock(rq_src, rq_dest); | |
5427 | local_irq_restore(flags); | |
5428 | } | |
5429 | ||
5430 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5431 | static void migrate_live_tasks(int src_cpu) | |
5432 | { | |
48f24c4d | 5433 | struct task_struct *p, *t; |
1da177e4 | 5434 | |
f7b4cddc | 5435 | read_lock(&tasklist_lock); |
1da177e4 | 5436 | |
48f24c4d IM |
5437 | do_each_thread(t, p) { |
5438 | if (p == current) | |
1da177e4 LT |
5439 | continue; |
5440 | ||
48f24c4d IM |
5441 | if (task_cpu(p) == src_cpu) |
5442 | move_task_off_dead_cpu(src_cpu, p); | |
5443 | } while_each_thread(t, p); | |
1da177e4 | 5444 | |
f7b4cddc | 5445 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5446 | } |
5447 | ||
dd41f596 IM |
5448 | /* |
5449 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5450 | * It does so by boosting its priority to highest possible. |
5451 | * Used by CPU offline code. | |
1da177e4 LT |
5452 | */ |
5453 | void sched_idle_next(void) | |
5454 | { | |
48f24c4d | 5455 | int this_cpu = smp_processor_id(); |
70b97a7f | 5456 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5457 | struct task_struct *p = rq->idle; |
5458 | unsigned long flags; | |
5459 | ||
5460 | /* cpu has to be offline */ | |
48f24c4d | 5461 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5462 | |
48f24c4d IM |
5463 | /* |
5464 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5465 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5466 | */ |
05fa785c | 5467 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5468 | |
dd41f596 | 5469 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5470 | |
94bc9a7b | 5471 | activate_task(rq, p, 0); |
1da177e4 | 5472 | |
05fa785c | 5473 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5474 | } |
5475 | ||
48f24c4d IM |
5476 | /* |
5477 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5478 | * offline. |
5479 | */ | |
5480 | void idle_task_exit(void) | |
5481 | { | |
5482 | struct mm_struct *mm = current->active_mm; | |
5483 | ||
5484 | BUG_ON(cpu_online(smp_processor_id())); | |
5485 | ||
5486 | if (mm != &init_mm) | |
5487 | switch_mm(mm, &init_mm, current); | |
5488 | mmdrop(mm); | |
5489 | } | |
5490 | ||
054b9108 | 5491 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5492 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5493 | { |
70b97a7f | 5494 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5495 | |
5496 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5497 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5498 | |
5499 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5500 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5501 | |
48f24c4d | 5502 | get_task_struct(p); |
1da177e4 LT |
5503 | |
5504 | /* | |
5505 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5506 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5507 | * fine. |
5508 | */ | |
05fa785c | 5509 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5510 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5511 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5512 | |
48f24c4d | 5513 | put_task_struct(p); |
1da177e4 LT |
5514 | } |
5515 | ||
5516 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5517 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5518 | { | |
70b97a7f | 5519 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5520 | struct task_struct *next; |
48f24c4d | 5521 | |
dd41f596 IM |
5522 | for ( ; ; ) { |
5523 | if (!rq->nr_running) | |
5524 | break; | |
b67802ea | 5525 | next = pick_next_task(rq); |
dd41f596 IM |
5526 | if (!next) |
5527 | break; | |
79c53799 | 5528 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5529 | migrate_dead(dead_cpu, next); |
e692ab53 | 5530 | |
1da177e4 LT |
5531 | } |
5532 | } | |
dce48a84 TG |
5533 | |
5534 | /* | |
5535 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5536 | */ | |
5537 | static void calc_global_load_remove(struct rq *rq) | |
5538 | { | |
5539 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5540 | rq->calc_load_active = 0; |
dce48a84 | 5541 | } |
1da177e4 LT |
5542 | #endif /* CONFIG_HOTPLUG_CPU */ |
5543 | ||
e692ab53 NP |
5544 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5545 | ||
5546 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5547 | { |
5548 | .procname = "sched_domain", | |
c57baf1e | 5549 | .mode = 0555, |
e0361851 | 5550 | }, |
56992309 | 5551 | {} |
e692ab53 NP |
5552 | }; |
5553 | ||
5554 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5555 | { |
5556 | .procname = "kernel", | |
c57baf1e | 5557 | .mode = 0555, |
e0361851 AD |
5558 | .child = sd_ctl_dir, |
5559 | }, | |
56992309 | 5560 | {} |
e692ab53 NP |
5561 | }; |
5562 | ||
5563 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5564 | { | |
5565 | struct ctl_table *entry = | |
5cf9f062 | 5566 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5567 | |
e692ab53 NP |
5568 | return entry; |
5569 | } | |
5570 | ||
6382bc90 MM |
5571 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5572 | { | |
cd790076 | 5573 | struct ctl_table *entry; |
6382bc90 | 5574 | |
cd790076 MM |
5575 | /* |
5576 | * In the intermediate directories, both the child directory and | |
5577 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5578 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5579 | * static strings and all have proc handlers. |
5580 | */ | |
5581 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5582 | if (entry->child) |
5583 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5584 | if (entry->proc_handler == NULL) |
5585 | kfree(entry->procname); | |
5586 | } | |
6382bc90 MM |
5587 | |
5588 | kfree(*tablep); | |
5589 | *tablep = NULL; | |
5590 | } | |
5591 | ||
e692ab53 | 5592 | static void |
e0361851 | 5593 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5594 | const char *procname, void *data, int maxlen, |
5595 | mode_t mode, proc_handler *proc_handler) | |
5596 | { | |
e692ab53 NP |
5597 | entry->procname = procname; |
5598 | entry->data = data; | |
5599 | entry->maxlen = maxlen; | |
5600 | entry->mode = mode; | |
5601 | entry->proc_handler = proc_handler; | |
5602 | } | |
5603 | ||
5604 | static struct ctl_table * | |
5605 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5606 | { | |
a5d8c348 | 5607 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5608 | |
ad1cdc1d MM |
5609 | if (table == NULL) |
5610 | return NULL; | |
5611 | ||
e0361851 | 5612 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5613 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5614 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5615 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5616 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5617 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5618 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5619 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5620 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5621 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5622 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5623 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5624 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5625 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5626 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5627 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5628 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5629 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5630 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5631 | &sd->cache_nice_tries, |
5632 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5633 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5634 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5635 | set_table_entry(&table[11], "name", sd->name, |
5636 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5637 | /* &table[12] is terminator */ | |
e692ab53 NP |
5638 | |
5639 | return table; | |
5640 | } | |
5641 | ||
9a4e7159 | 5642 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5643 | { |
5644 | struct ctl_table *entry, *table; | |
5645 | struct sched_domain *sd; | |
5646 | int domain_num = 0, i; | |
5647 | char buf[32]; | |
5648 | ||
5649 | for_each_domain(cpu, sd) | |
5650 | domain_num++; | |
5651 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5652 | if (table == NULL) |
5653 | return NULL; | |
e692ab53 NP |
5654 | |
5655 | i = 0; | |
5656 | for_each_domain(cpu, sd) { | |
5657 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5658 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5659 | entry->mode = 0555; |
e692ab53 NP |
5660 | entry->child = sd_alloc_ctl_domain_table(sd); |
5661 | entry++; | |
5662 | i++; | |
5663 | } | |
5664 | return table; | |
5665 | } | |
5666 | ||
5667 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5668 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5669 | { |
6ad4c188 | 5670 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5671 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5672 | char buf[32]; | |
5673 | ||
7378547f MM |
5674 | WARN_ON(sd_ctl_dir[0].child); |
5675 | sd_ctl_dir[0].child = entry; | |
5676 | ||
ad1cdc1d MM |
5677 | if (entry == NULL) |
5678 | return; | |
5679 | ||
6ad4c188 | 5680 | for_each_possible_cpu(i) { |
e692ab53 | 5681 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5682 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5683 | entry->mode = 0555; |
e692ab53 | 5684 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5685 | entry++; |
e692ab53 | 5686 | } |
7378547f MM |
5687 | |
5688 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5689 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5690 | } | |
6382bc90 | 5691 | |
7378547f | 5692 | /* may be called multiple times per register */ |
6382bc90 MM |
5693 | static void unregister_sched_domain_sysctl(void) |
5694 | { | |
7378547f MM |
5695 | if (sd_sysctl_header) |
5696 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5697 | sd_sysctl_header = NULL; |
7378547f MM |
5698 | if (sd_ctl_dir[0].child) |
5699 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5700 | } |
e692ab53 | 5701 | #else |
6382bc90 MM |
5702 | static void register_sched_domain_sysctl(void) |
5703 | { | |
5704 | } | |
5705 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5706 | { |
5707 | } | |
5708 | #endif | |
5709 | ||
1f11eb6a GH |
5710 | static void set_rq_online(struct rq *rq) |
5711 | { | |
5712 | if (!rq->online) { | |
5713 | const struct sched_class *class; | |
5714 | ||
c6c4927b | 5715 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5716 | rq->online = 1; |
5717 | ||
5718 | for_each_class(class) { | |
5719 | if (class->rq_online) | |
5720 | class->rq_online(rq); | |
5721 | } | |
5722 | } | |
5723 | } | |
5724 | ||
5725 | static void set_rq_offline(struct rq *rq) | |
5726 | { | |
5727 | if (rq->online) { | |
5728 | const struct sched_class *class; | |
5729 | ||
5730 | for_each_class(class) { | |
5731 | if (class->rq_offline) | |
5732 | class->rq_offline(rq); | |
5733 | } | |
5734 | ||
c6c4927b | 5735 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5736 | rq->online = 0; |
5737 | } | |
5738 | } | |
5739 | ||
1da177e4 LT |
5740 | /* |
5741 | * migration_call - callback that gets triggered when a CPU is added. | |
5742 | * Here we can start up the necessary migration thread for the new CPU. | |
5743 | */ | |
48f24c4d IM |
5744 | static int __cpuinit |
5745 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5746 | { |
48f24c4d | 5747 | int cpu = (long)hcpu; |
1da177e4 | 5748 | unsigned long flags; |
969c7921 | 5749 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5750 | |
5751 | switch (action) { | |
5be9361c | 5752 | |
1da177e4 | 5753 | case CPU_UP_PREPARE: |
8bb78442 | 5754 | case CPU_UP_PREPARE_FROZEN: |
a468d389 | 5755 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5756 | break; |
48f24c4d | 5757 | |
1da177e4 | 5758 | case CPU_ONLINE: |
8bb78442 | 5759 | case CPU_ONLINE_FROZEN: |
1f94ef59 | 5760 | /* Update our root-domain */ |
05fa785c | 5761 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5762 | if (rq->rd) { |
c6c4927b | 5763 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5764 | |
5765 | set_rq_online(rq); | |
1f94ef59 | 5766 | } |
05fa785c | 5767 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5768 | break; |
48f24c4d | 5769 | |
1da177e4 | 5770 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 5771 | case CPU_DEAD: |
8bb78442 | 5772 | case CPU_DEAD_FROZEN: |
1da177e4 | 5773 | migrate_live_tasks(cpu); |
1da177e4 | 5774 | /* Idle task back to normal (off runqueue, low prio) */ |
05fa785c | 5775 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 5776 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5777 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5778 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5779 | migrate_dead_tasks(cpu); |
05fa785c | 5780 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5781 | migrate_nr_uninterruptible(rq); |
5782 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5783 | calc_global_load_remove(rq); |
1da177e4 | 5784 | break; |
57d885fe | 5785 | |
08f503b0 GH |
5786 | case CPU_DYING: |
5787 | case CPU_DYING_FROZEN: | |
57d885fe | 5788 | /* Update our root-domain */ |
05fa785c | 5789 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5790 | if (rq->rd) { |
c6c4927b | 5791 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5792 | set_rq_offline(rq); |
57d885fe | 5793 | } |
05fa785c | 5794 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 5795 | break; |
1da177e4 LT |
5796 | #endif |
5797 | } | |
5798 | return NOTIFY_OK; | |
5799 | } | |
5800 | ||
f38b0820 PM |
5801 | /* |
5802 | * Register at high priority so that task migration (migrate_all_tasks) | |
5803 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5804 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5805 | */ |
26c2143b | 5806 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 LT |
5807 | .notifier_call = migration_call, |
5808 | .priority = 10 | |
5809 | }; | |
5810 | ||
7babe8db | 5811 | static int __init migration_init(void) |
1da177e4 LT |
5812 | { |
5813 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5814 | int err; |
48f24c4d IM |
5815 | |
5816 | /* Start one for the boot CPU: */ | |
07dccf33 AM |
5817 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5818 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5819 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5820 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5821 | |
a004cd42 | 5822 | return 0; |
1da177e4 | 5823 | } |
7babe8db | 5824 | early_initcall(migration_init); |
1da177e4 LT |
5825 | #endif |
5826 | ||
5827 | #ifdef CONFIG_SMP | |
476f3534 | 5828 | |
3e9830dc | 5829 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5830 | |
f6630114 MT |
5831 | static __read_mostly int sched_domain_debug_enabled; |
5832 | ||
5833 | static int __init sched_domain_debug_setup(char *str) | |
5834 | { | |
5835 | sched_domain_debug_enabled = 1; | |
5836 | ||
5837 | return 0; | |
5838 | } | |
5839 | early_param("sched_debug", sched_domain_debug_setup); | |
5840 | ||
7c16ec58 | 5841 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5842 | struct cpumask *groupmask) |
1da177e4 | 5843 | { |
4dcf6aff | 5844 | struct sched_group *group = sd->groups; |
434d53b0 | 5845 | char str[256]; |
1da177e4 | 5846 | |
968ea6d8 | 5847 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5848 | cpumask_clear(groupmask); |
4dcf6aff IM |
5849 | |
5850 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5851 | ||
5852 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5853 | printk("does not load-balance\n"); |
4dcf6aff | 5854 | if (sd->parent) |
3df0fc5b PZ |
5855 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5856 | " has parent"); | |
4dcf6aff | 5857 | return -1; |
41c7ce9a NP |
5858 | } |
5859 | ||
3df0fc5b | 5860 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5861 | |
758b2cdc | 5862 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5863 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5864 | "CPU%d\n", cpu); | |
4dcf6aff | 5865 | } |
758b2cdc | 5866 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5867 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5868 | " CPU%d\n", cpu); | |
4dcf6aff | 5869 | } |
1da177e4 | 5870 | |
4dcf6aff | 5871 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5872 | do { |
4dcf6aff | 5873 | if (!group) { |
3df0fc5b PZ |
5874 | printk("\n"); |
5875 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5876 | break; |
5877 | } | |
5878 | ||
18a3885f | 5879 | if (!group->cpu_power) { |
3df0fc5b PZ |
5880 | printk(KERN_CONT "\n"); |
5881 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5882 | "set\n"); | |
4dcf6aff IM |
5883 | break; |
5884 | } | |
1da177e4 | 5885 | |
758b2cdc | 5886 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5887 | printk(KERN_CONT "\n"); |
5888 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5889 | break; |
5890 | } | |
1da177e4 | 5891 | |
758b2cdc | 5892 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5893 | printk(KERN_CONT "\n"); |
5894 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5895 | break; |
5896 | } | |
1da177e4 | 5897 | |
758b2cdc | 5898 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5899 | |
968ea6d8 | 5900 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5901 | |
3df0fc5b | 5902 | printk(KERN_CONT " %s", str); |
18a3885f | 5903 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
5904 | printk(KERN_CONT " (cpu_power = %d)", |
5905 | group->cpu_power); | |
381512cf | 5906 | } |
1da177e4 | 5907 | |
4dcf6aff IM |
5908 | group = group->next; |
5909 | } while (group != sd->groups); | |
3df0fc5b | 5910 | printk(KERN_CONT "\n"); |
1da177e4 | 5911 | |
758b2cdc | 5912 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 5913 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 5914 | |
758b2cdc RR |
5915 | if (sd->parent && |
5916 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
5917 | printk(KERN_ERR "ERROR: parent span is not a superset " |
5918 | "of domain->span\n"); | |
4dcf6aff IM |
5919 | return 0; |
5920 | } | |
1da177e4 | 5921 | |
4dcf6aff IM |
5922 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
5923 | { | |
d5dd3db1 | 5924 | cpumask_var_t groupmask; |
4dcf6aff | 5925 | int level = 0; |
1da177e4 | 5926 | |
f6630114 MT |
5927 | if (!sched_domain_debug_enabled) |
5928 | return; | |
5929 | ||
4dcf6aff IM |
5930 | if (!sd) { |
5931 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
5932 | return; | |
5933 | } | |
1da177e4 | 5934 | |
4dcf6aff IM |
5935 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
5936 | ||
d5dd3db1 | 5937 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
5938 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
5939 | return; | |
5940 | } | |
5941 | ||
4dcf6aff | 5942 | for (;;) { |
7c16ec58 | 5943 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 5944 | break; |
1da177e4 LT |
5945 | level++; |
5946 | sd = sd->parent; | |
33859f7f | 5947 | if (!sd) |
4dcf6aff IM |
5948 | break; |
5949 | } | |
d5dd3db1 | 5950 | free_cpumask_var(groupmask); |
1da177e4 | 5951 | } |
6d6bc0ad | 5952 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 5953 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 5954 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 5955 | |
1a20ff27 | 5956 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 5957 | { |
758b2cdc | 5958 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
5959 | return 1; |
5960 | ||
5961 | /* Following flags need at least 2 groups */ | |
5962 | if (sd->flags & (SD_LOAD_BALANCE | | |
5963 | SD_BALANCE_NEWIDLE | | |
5964 | SD_BALANCE_FORK | | |
89c4710e SS |
5965 | SD_BALANCE_EXEC | |
5966 | SD_SHARE_CPUPOWER | | |
5967 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
5968 | if (sd->groups != sd->groups->next) |
5969 | return 0; | |
5970 | } | |
5971 | ||
5972 | /* Following flags don't use groups */ | |
c88d5910 | 5973 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
5974 | return 0; |
5975 | ||
5976 | return 1; | |
5977 | } | |
5978 | ||
48f24c4d IM |
5979 | static int |
5980 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
5981 | { |
5982 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
5983 | ||
5984 | if (sd_degenerate(parent)) | |
5985 | return 1; | |
5986 | ||
758b2cdc | 5987 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
5988 | return 0; |
5989 | ||
245af2c7 SS |
5990 | /* Flags needing groups don't count if only 1 group in parent */ |
5991 | if (parent->groups == parent->groups->next) { | |
5992 | pflags &= ~(SD_LOAD_BALANCE | | |
5993 | SD_BALANCE_NEWIDLE | | |
5994 | SD_BALANCE_FORK | | |
89c4710e SS |
5995 | SD_BALANCE_EXEC | |
5996 | SD_SHARE_CPUPOWER | | |
5997 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
5998 | if (nr_node_ids == 1) |
5999 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6000 | } |
6001 | if (~cflags & pflags) | |
6002 | return 0; | |
6003 | ||
6004 | return 1; | |
6005 | } | |
6006 | ||
c6c4927b RR |
6007 | static void free_rootdomain(struct root_domain *rd) |
6008 | { | |
047106ad PZ |
6009 | synchronize_sched(); |
6010 | ||
68e74568 RR |
6011 | cpupri_cleanup(&rd->cpupri); |
6012 | ||
c6c4927b RR |
6013 | free_cpumask_var(rd->rto_mask); |
6014 | free_cpumask_var(rd->online); | |
6015 | free_cpumask_var(rd->span); | |
6016 | kfree(rd); | |
6017 | } | |
6018 | ||
57d885fe GH |
6019 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6020 | { | |
a0490fa3 | 6021 | struct root_domain *old_rd = NULL; |
57d885fe | 6022 | unsigned long flags; |
57d885fe | 6023 | |
05fa785c | 6024 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6025 | |
6026 | if (rq->rd) { | |
a0490fa3 | 6027 | old_rd = rq->rd; |
57d885fe | 6028 | |
c6c4927b | 6029 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6030 | set_rq_offline(rq); |
57d885fe | 6031 | |
c6c4927b | 6032 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6033 | |
a0490fa3 IM |
6034 | /* |
6035 | * If we dont want to free the old_rt yet then | |
6036 | * set old_rd to NULL to skip the freeing later | |
6037 | * in this function: | |
6038 | */ | |
6039 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6040 | old_rd = NULL; | |
57d885fe GH |
6041 | } |
6042 | ||
6043 | atomic_inc(&rd->refcount); | |
6044 | rq->rd = rd; | |
6045 | ||
c6c4927b | 6046 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6047 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6048 | set_rq_online(rq); |
57d885fe | 6049 | |
05fa785c | 6050 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6051 | |
6052 | if (old_rd) | |
6053 | free_rootdomain(old_rd); | |
57d885fe GH |
6054 | } |
6055 | ||
fd5e1b5d | 6056 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6057 | { |
36b7b6d4 PE |
6058 | gfp_t gfp = GFP_KERNEL; |
6059 | ||
57d885fe GH |
6060 | memset(rd, 0, sizeof(*rd)); |
6061 | ||
36b7b6d4 PE |
6062 | if (bootmem) |
6063 | gfp = GFP_NOWAIT; | |
c6c4927b | 6064 | |
36b7b6d4 | 6065 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6066 | goto out; |
36b7b6d4 | 6067 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6068 | goto free_span; |
36b7b6d4 | 6069 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6070 | goto free_online; |
6e0534f2 | 6071 | |
0fb53029 | 6072 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6073 | goto free_rto_mask; |
c6c4927b | 6074 | return 0; |
6e0534f2 | 6075 | |
68e74568 RR |
6076 | free_rto_mask: |
6077 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6078 | free_online: |
6079 | free_cpumask_var(rd->online); | |
6080 | free_span: | |
6081 | free_cpumask_var(rd->span); | |
0c910d28 | 6082 | out: |
c6c4927b | 6083 | return -ENOMEM; |
57d885fe GH |
6084 | } |
6085 | ||
6086 | static void init_defrootdomain(void) | |
6087 | { | |
c6c4927b RR |
6088 | init_rootdomain(&def_root_domain, true); |
6089 | ||
57d885fe GH |
6090 | atomic_set(&def_root_domain.refcount, 1); |
6091 | } | |
6092 | ||
dc938520 | 6093 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6094 | { |
6095 | struct root_domain *rd; | |
6096 | ||
6097 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6098 | if (!rd) | |
6099 | return NULL; | |
6100 | ||
c6c4927b RR |
6101 | if (init_rootdomain(rd, false) != 0) { |
6102 | kfree(rd); | |
6103 | return NULL; | |
6104 | } | |
57d885fe GH |
6105 | |
6106 | return rd; | |
6107 | } | |
6108 | ||
1da177e4 | 6109 | /* |
0eab9146 | 6110 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6111 | * hold the hotplug lock. |
6112 | */ | |
0eab9146 IM |
6113 | static void |
6114 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6115 | { |
70b97a7f | 6116 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6117 | struct sched_domain *tmp; |
6118 | ||
669c55e9 PZ |
6119 | for (tmp = sd; tmp; tmp = tmp->parent) |
6120 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6121 | ||
245af2c7 | 6122 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6123 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6124 | struct sched_domain *parent = tmp->parent; |
6125 | if (!parent) | |
6126 | break; | |
f29c9b1c | 6127 | |
1a848870 | 6128 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6129 | tmp->parent = parent->parent; |
1a848870 SS |
6130 | if (parent->parent) |
6131 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6132 | } else |
6133 | tmp = tmp->parent; | |
245af2c7 SS |
6134 | } |
6135 | ||
1a848870 | 6136 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6137 | sd = sd->parent; |
1a848870 SS |
6138 | if (sd) |
6139 | sd->child = NULL; | |
6140 | } | |
1da177e4 LT |
6141 | |
6142 | sched_domain_debug(sd, cpu); | |
6143 | ||
57d885fe | 6144 | rq_attach_root(rq, rd); |
674311d5 | 6145 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6146 | } |
6147 | ||
6148 | /* cpus with isolated domains */ | |
dcc30a35 | 6149 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6150 | |
6151 | /* Setup the mask of cpus configured for isolated domains */ | |
6152 | static int __init isolated_cpu_setup(char *str) | |
6153 | { | |
bdddd296 | 6154 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6155 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6156 | return 1; |
6157 | } | |
6158 | ||
8927f494 | 6159 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6160 | |
6161 | /* | |
6711cab4 SS |
6162 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6163 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6164 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6165 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6166 | * |
6167 | * init_sched_build_groups will build a circular linked list of the groups | |
6168 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6169 | * and ->cpu_power to 0. | |
6170 | */ | |
a616058b | 6171 | static void |
96f874e2 RR |
6172 | init_sched_build_groups(const struct cpumask *span, |
6173 | const struct cpumask *cpu_map, | |
6174 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6175 | struct sched_group **sg, |
96f874e2 RR |
6176 | struct cpumask *tmpmask), |
6177 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6178 | { |
6179 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6180 | int i; |
6181 | ||
96f874e2 | 6182 | cpumask_clear(covered); |
7c16ec58 | 6183 | |
abcd083a | 6184 | for_each_cpu(i, span) { |
6711cab4 | 6185 | struct sched_group *sg; |
7c16ec58 | 6186 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6187 | int j; |
6188 | ||
758b2cdc | 6189 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6190 | continue; |
6191 | ||
758b2cdc | 6192 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6193 | sg->cpu_power = 0; |
1da177e4 | 6194 | |
abcd083a | 6195 | for_each_cpu(j, span) { |
7c16ec58 | 6196 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6197 | continue; |
6198 | ||
96f874e2 | 6199 | cpumask_set_cpu(j, covered); |
758b2cdc | 6200 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6201 | } |
6202 | if (!first) | |
6203 | first = sg; | |
6204 | if (last) | |
6205 | last->next = sg; | |
6206 | last = sg; | |
6207 | } | |
6208 | last->next = first; | |
6209 | } | |
6210 | ||
9c1cfda2 | 6211 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6212 | |
9c1cfda2 | 6213 | #ifdef CONFIG_NUMA |
198e2f18 | 6214 | |
9c1cfda2 JH |
6215 | /** |
6216 | * find_next_best_node - find the next node to include in a sched_domain | |
6217 | * @node: node whose sched_domain we're building | |
6218 | * @used_nodes: nodes already in the sched_domain | |
6219 | * | |
41a2d6cf | 6220 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6221 | * finds the closest node not already in the @used_nodes map. |
6222 | * | |
6223 | * Should use nodemask_t. | |
6224 | */ | |
c5f59f08 | 6225 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6226 | { |
6227 | int i, n, val, min_val, best_node = 0; | |
6228 | ||
6229 | min_val = INT_MAX; | |
6230 | ||
076ac2af | 6231 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6232 | /* Start at @node */ |
076ac2af | 6233 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6234 | |
6235 | if (!nr_cpus_node(n)) | |
6236 | continue; | |
6237 | ||
6238 | /* Skip already used nodes */ | |
c5f59f08 | 6239 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6240 | continue; |
6241 | ||
6242 | /* Simple min distance search */ | |
6243 | val = node_distance(node, n); | |
6244 | ||
6245 | if (val < min_val) { | |
6246 | min_val = val; | |
6247 | best_node = n; | |
6248 | } | |
6249 | } | |
6250 | ||
c5f59f08 | 6251 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6252 | return best_node; |
6253 | } | |
6254 | ||
6255 | /** | |
6256 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6257 | * @node: node whose cpumask we're constructing | |
73486722 | 6258 | * @span: resulting cpumask |
9c1cfda2 | 6259 | * |
41a2d6cf | 6260 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6261 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6262 | * out optimally. | |
6263 | */ | |
96f874e2 | 6264 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6265 | { |
c5f59f08 | 6266 | nodemask_t used_nodes; |
48f24c4d | 6267 | int i; |
9c1cfda2 | 6268 | |
6ca09dfc | 6269 | cpumask_clear(span); |
c5f59f08 | 6270 | nodes_clear(used_nodes); |
9c1cfda2 | 6271 | |
6ca09dfc | 6272 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6273 | node_set(node, used_nodes); |
9c1cfda2 JH |
6274 | |
6275 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6276 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6277 | |
6ca09dfc | 6278 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6279 | } |
9c1cfda2 | 6280 | } |
6d6bc0ad | 6281 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6282 | |
5c45bf27 | 6283 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6284 | |
6c99e9ad RR |
6285 | /* |
6286 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6287 | * |
6288 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6289 | * and struct sched_domain. ) | |
6c99e9ad RR |
6290 | */ |
6291 | struct static_sched_group { | |
6292 | struct sched_group sg; | |
6293 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6294 | }; | |
6295 | ||
6296 | struct static_sched_domain { | |
6297 | struct sched_domain sd; | |
6298 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6299 | }; | |
6300 | ||
49a02c51 AH |
6301 | struct s_data { |
6302 | #ifdef CONFIG_NUMA | |
6303 | int sd_allnodes; | |
6304 | cpumask_var_t domainspan; | |
6305 | cpumask_var_t covered; | |
6306 | cpumask_var_t notcovered; | |
6307 | #endif | |
6308 | cpumask_var_t nodemask; | |
6309 | cpumask_var_t this_sibling_map; | |
6310 | cpumask_var_t this_core_map; | |
6311 | cpumask_var_t send_covered; | |
6312 | cpumask_var_t tmpmask; | |
6313 | struct sched_group **sched_group_nodes; | |
6314 | struct root_domain *rd; | |
6315 | }; | |
6316 | ||
2109b99e AH |
6317 | enum s_alloc { |
6318 | sa_sched_groups = 0, | |
6319 | sa_rootdomain, | |
6320 | sa_tmpmask, | |
6321 | sa_send_covered, | |
6322 | sa_this_core_map, | |
6323 | sa_this_sibling_map, | |
6324 | sa_nodemask, | |
6325 | sa_sched_group_nodes, | |
6326 | #ifdef CONFIG_NUMA | |
6327 | sa_notcovered, | |
6328 | sa_covered, | |
6329 | sa_domainspan, | |
6330 | #endif | |
6331 | sa_none, | |
6332 | }; | |
6333 | ||
9c1cfda2 | 6334 | /* |
48f24c4d | 6335 | * SMT sched-domains: |
9c1cfda2 | 6336 | */ |
1da177e4 | 6337 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6338 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6339 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6340 | |
41a2d6cf | 6341 | static int |
96f874e2 RR |
6342 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6343 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6344 | { |
6711cab4 | 6345 | if (sg) |
1871e52c | 6346 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6347 | return cpu; |
6348 | } | |
6d6bc0ad | 6349 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6350 | |
48f24c4d IM |
6351 | /* |
6352 | * multi-core sched-domains: | |
6353 | */ | |
1e9f28fa | 6354 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6355 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6356 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6357 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6358 | |
6359 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6360 | static int |
96f874e2 RR |
6361 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6362 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6363 | { |
6711cab4 | 6364 | int group; |
7c16ec58 | 6365 | |
c69fc56d | 6366 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6367 | group = cpumask_first(mask); |
6711cab4 | 6368 | if (sg) |
6c99e9ad | 6369 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6370 | return group; |
1e9f28fa SS |
6371 | } |
6372 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6373 | static int |
96f874e2 RR |
6374 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6375 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6376 | { |
6711cab4 | 6377 | if (sg) |
6c99e9ad | 6378 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6379 | return cpu; |
6380 | } | |
6381 | #endif | |
6382 | ||
6c99e9ad RR |
6383 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6384 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6385 | |
41a2d6cf | 6386 | static int |
96f874e2 RR |
6387 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6388 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6389 | { |
6711cab4 | 6390 | int group; |
48f24c4d | 6391 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6392 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6393 | group = cpumask_first(mask); |
1e9f28fa | 6394 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6395 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6396 | group = cpumask_first(mask); |
1da177e4 | 6397 | #else |
6711cab4 | 6398 | group = cpu; |
1da177e4 | 6399 | #endif |
6711cab4 | 6400 | if (sg) |
6c99e9ad | 6401 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6402 | return group; |
1da177e4 LT |
6403 | } |
6404 | ||
6405 | #ifdef CONFIG_NUMA | |
1da177e4 | 6406 | /* |
9c1cfda2 JH |
6407 | * The init_sched_build_groups can't handle what we want to do with node |
6408 | * groups, so roll our own. Now each node has its own list of groups which | |
6409 | * gets dynamically allocated. | |
1da177e4 | 6410 | */ |
62ea9ceb | 6411 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6412 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6413 | |
62ea9ceb | 6414 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6415 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6416 | |
96f874e2 RR |
6417 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6418 | struct sched_group **sg, | |
6419 | struct cpumask *nodemask) | |
9c1cfda2 | 6420 | { |
6711cab4 SS |
6421 | int group; |
6422 | ||
6ca09dfc | 6423 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6424 | group = cpumask_first(nodemask); |
6711cab4 SS |
6425 | |
6426 | if (sg) | |
6c99e9ad | 6427 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6428 | return group; |
1da177e4 | 6429 | } |
6711cab4 | 6430 | |
08069033 SS |
6431 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6432 | { | |
6433 | struct sched_group *sg = group_head; | |
6434 | int j; | |
6435 | ||
6436 | if (!sg) | |
6437 | return; | |
3a5c359a | 6438 | do { |
758b2cdc | 6439 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6440 | struct sched_domain *sd; |
08069033 | 6441 | |
6c99e9ad | 6442 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6443 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6444 | /* |
6445 | * Only add "power" once for each | |
6446 | * physical package. | |
6447 | */ | |
6448 | continue; | |
6449 | } | |
08069033 | 6450 | |
18a3885f | 6451 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6452 | } |
6453 | sg = sg->next; | |
6454 | } while (sg != group_head); | |
08069033 | 6455 | } |
0601a88d AH |
6456 | |
6457 | static int build_numa_sched_groups(struct s_data *d, | |
6458 | const struct cpumask *cpu_map, int num) | |
6459 | { | |
6460 | struct sched_domain *sd; | |
6461 | struct sched_group *sg, *prev; | |
6462 | int n, j; | |
6463 | ||
6464 | cpumask_clear(d->covered); | |
6465 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6466 | if (cpumask_empty(d->nodemask)) { | |
6467 | d->sched_group_nodes[num] = NULL; | |
6468 | goto out; | |
6469 | } | |
6470 | ||
6471 | sched_domain_node_span(num, d->domainspan); | |
6472 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6473 | ||
6474 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6475 | GFP_KERNEL, num); | |
6476 | if (!sg) { | |
3df0fc5b PZ |
6477 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6478 | num); | |
0601a88d AH |
6479 | return -ENOMEM; |
6480 | } | |
6481 | d->sched_group_nodes[num] = sg; | |
6482 | ||
6483 | for_each_cpu(j, d->nodemask) { | |
6484 | sd = &per_cpu(node_domains, j).sd; | |
6485 | sd->groups = sg; | |
6486 | } | |
6487 | ||
18a3885f | 6488 | sg->cpu_power = 0; |
0601a88d AH |
6489 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6490 | sg->next = sg; | |
6491 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6492 | ||
6493 | prev = sg; | |
6494 | for (j = 0; j < nr_node_ids; j++) { | |
6495 | n = (num + j) % nr_node_ids; | |
6496 | cpumask_complement(d->notcovered, d->covered); | |
6497 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6498 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6499 | if (cpumask_empty(d->tmpmask)) | |
6500 | break; | |
6501 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6502 | if (cpumask_empty(d->tmpmask)) | |
6503 | continue; | |
6504 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6505 | GFP_KERNEL, num); | |
6506 | if (!sg) { | |
3df0fc5b PZ |
6507 | printk(KERN_WARNING |
6508 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6509 | return -ENOMEM; |
6510 | } | |
18a3885f | 6511 | sg->cpu_power = 0; |
0601a88d AH |
6512 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6513 | sg->next = prev->next; | |
6514 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6515 | prev->next = sg; | |
6516 | prev = sg; | |
6517 | } | |
6518 | out: | |
6519 | return 0; | |
6520 | } | |
6d6bc0ad | 6521 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6522 | |
a616058b | 6523 | #ifdef CONFIG_NUMA |
51888ca2 | 6524 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6525 | static void free_sched_groups(const struct cpumask *cpu_map, |
6526 | struct cpumask *nodemask) | |
51888ca2 | 6527 | { |
a616058b | 6528 | int cpu, i; |
51888ca2 | 6529 | |
abcd083a | 6530 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6531 | struct sched_group **sched_group_nodes |
6532 | = sched_group_nodes_bycpu[cpu]; | |
6533 | ||
51888ca2 SV |
6534 | if (!sched_group_nodes) |
6535 | continue; | |
6536 | ||
076ac2af | 6537 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6538 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6539 | ||
6ca09dfc | 6540 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6541 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6542 | continue; |
6543 | ||
6544 | if (sg == NULL) | |
6545 | continue; | |
6546 | sg = sg->next; | |
6547 | next_sg: | |
6548 | oldsg = sg; | |
6549 | sg = sg->next; | |
6550 | kfree(oldsg); | |
6551 | if (oldsg != sched_group_nodes[i]) | |
6552 | goto next_sg; | |
6553 | } | |
6554 | kfree(sched_group_nodes); | |
6555 | sched_group_nodes_bycpu[cpu] = NULL; | |
6556 | } | |
51888ca2 | 6557 | } |
6d6bc0ad | 6558 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6559 | static void free_sched_groups(const struct cpumask *cpu_map, |
6560 | struct cpumask *nodemask) | |
a616058b SS |
6561 | { |
6562 | } | |
6d6bc0ad | 6563 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6564 | |
89c4710e SS |
6565 | /* |
6566 | * Initialize sched groups cpu_power. | |
6567 | * | |
6568 | * cpu_power indicates the capacity of sched group, which is used while | |
6569 | * distributing the load between different sched groups in a sched domain. | |
6570 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6571 | * there are asymmetries in the topology. If there are asymmetries, group | |
6572 | * having more cpu_power will pickup more load compared to the group having | |
6573 | * less cpu_power. | |
89c4710e SS |
6574 | */ |
6575 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6576 | { | |
6577 | struct sched_domain *child; | |
6578 | struct sched_group *group; | |
f93e65c1 PZ |
6579 | long power; |
6580 | int weight; | |
89c4710e SS |
6581 | |
6582 | WARN_ON(!sd || !sd->groups); | |
6583 | ||
13318a71 | 6584 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6585 | return; |
6586 | ||
6587 | child = sd->child; | |
6588 | ||
18a3885f | 6589 | sd->groups->cpu_power = 0; |
5517d86b | 6590 | |
f93e65c1 PZ |
6591 | if (!child) { |
6592 | power = SCHED_LOAD_SCALE; | |
6593 | weight = cpumask_weight(sched_domain_span(sd)); | |
6594 | /* | |
6595 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6596 | * Usually multiple threads get a better yield out of |
6597 | * that one core than a single thread would have, | |
6598 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6599 | */ |
a52bfd73 PZ |
6600 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6601 | power *= sd->smt_gain; | |
f93e65c1 | 6602 | power /= weight; |
a52bfd73 PZ |
6603 | power >>= SCHED_LOAD_SHIFT; |
6604 | } | |
18a3885f | 6605 | sd->groups->cpu_power += power; |
89c4710e SS |
6606 | return; |
6607 | } | |
6608 | ||
89c4710e | 6609 | /* |
f93e65c1 | 6610 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6611 | */ |
6612 | group = child->groups; | |
6613 | do { | |
18a3885f | 6614 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6615 | group = group->next; |
6616 | } while (group != child->groups); | |
6617 | } | |
6618 | ||
7c16ec58 MT |
6619 | /* |
6620 | * Initializers for schedule domains | |
6621 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6622 | */ | |
6623 | ||
a5d8c348 IM |
6624 | #ifdef CONFIG_SCHED_DEBUG |
6625 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6626 | #else | |
6627 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6628 | #endif | |
6629 | ||
7c16ec58 | 6630 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6631 | |
7c16ec58 MT |
6632 | #define SD_INIT_FUNC(type) \ |
6633 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6634 | { \ | |
6635 | memset(sd, 0, sizeof(*sd)); \ | |
6636 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6637 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6638 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6639 | } |
6640 | ||
6641 | SD_INIT_FUNC(CPU) | |
6642 | #ifdef CONFIG_NUMA | |
6643 | SD_INIT_FUNC(ALLNODES) | |
6644 | SD_INIT_FUNC(NODE) | |
6645 | #endif | |
6646 | #ifdef CONFIG_SCHED_SMT | |
6647 | SD_INIT_FUNC(SIBLING) | |
6648 | #endif | |
6649 | #ifdef CONFIG_SCHED_MC | |
6650 | SD_INIT_FUNC(MC) | |
6651 | #endif | |
6652 | ||
1d3504fc HS |
6653 | static int default_relax_domain_level = -1; |
6654 | ||
6655 | static int __init setup_relax_domain_level(char *str) | |
6656 | { | |
30e0e178 LZ |
6657 | unsigned long val; |
6658 | ||
6659 | val = simple_strtoul(str, NULL, 0); | |
6660 | if (val < SD_LV_MAX) | |
6661 | default_relax_domain_level = val; | |
6662 | ||
1d3504fc HS |
6663 | return 1; |
6664 | } | |
6665 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6666 | ||
6667 | static void set_domain_attribute(struct sched_domain *sd, | |
6668 | struct sched_domain_attr *attr) | |
6669 | { | |
6670 | int request; | |
6671 | ||
6672 | if (!attr || attr->relax_domain_level < 0) { | |
6673 | if (default_relax_domain_level < 0) | |
6674 | return; | |
6675 | else | |
6676 | request = default_relax_domain_level; | |
6677 | } else | |
6678 | request = attr->relax_domain_level; | |
6679 | if (request < sd->level) { | |
6680 | /* turn off idle balance on this domain */ | |
c88d5910 | 6681 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6682 | } else { |
6683 | /* turn on idle balance on this domain */ | |
c88d5910 | 6684 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6685 | } |
6686 | } | |
6687 | ||
2109b99e AH |
6688 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6689 | const struct cpumask *cpu_map) | |
6690 | { | |
6691 | switch (what) { | |
6692 | case sa_sched_groups: | |
6693 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6694 | d->sched_group_nodes = NULL; | |
6695 | case sa_rootdomain: | |
6696 | free_rootdomain(d->rd); /* fall through */ | |
6697 | case sa_tmpmask: | |
6698 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6699 | case sa_send_covered: | |
6700 | free_cpumask_var(d->send_covered); /* fall through */ | |
6701 | case sa_this_core_map: | |
6702 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6703 | case sa_this_sibling_map: | |
6704 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6705 | case sa_nodemask: | |
6706 | free_cpumask_var(d->nodemask); /* fall through */ | |
6707 | case sa_sched_group_nodes: | |
d1b55138 | 6708 | #ifdef CONFIG_NUMA |
2109b99e AH |
6709 | kfree(d->sched_group_nodes); /* fall through */ |
6710 | case sa_notcovered: | |
6711 | free_cpumask_var(d->notcovered); /* fall through */ | |
6712 | case sa_covered: | |
6713 | free_cpumask_var(d->covered); /* fall through */ | |
6714 | case sa_domainspan: | |
6715 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6716 | #endif |
2109b99e AH |
6717 | case sa_none: |
6718 | break; | |
6719 | } | |
6720 | } | |
3404c8d9 | 6721 | |
2109b99e AH |
6722 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6723 | const struct cpumask *cpu_map) | |
6724 | { | |
3404c8d9 | 6725 | #ifdef CONFIG_NUMA |
2109b99e AH |
6726 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6727 | return sa_none; | |
6728 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6729 | return sa_domainspan; | |
6730 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6731 | return sa_covered; | |
6732 | /* Allocate the per-node list of sched groups */ | |
6733 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6734 | sizeof(struct sched_group *), GFP_KERNEL); | |
6735 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6736 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6737 | return sa_notcovered; |
d1b55138 | 6738 | } |
2109b99e | 6739 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6740 | #endif |
2109b99e AH |
6741 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6742 | return sa_sched_group_nodes; | |
6743 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6744 | return sa_nodemask; | |
6745 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6746 | return sa_this_sibling_map; | |
6747 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6748 | return sa_this_core_map; | |
6749 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6750 | return sa_send_covered; | |
6751 | d->rd = alloc_rootdomain(); | |
6752 | if (!d->rd) { | |
3df0fc5b | 6753 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6754 | return sa_tmpmask; |
57d885fe | 6755 | } |
2109b99e AH |
6756 | return sa_rootdomain; |
6757 | } | |
57d885fe | 6758 | |
7f4588f3 AH |
6759 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6760 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6761 | { | |
6762 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6763 | #ifdef CONFIG_NUMA |
7f4588f3 | 6764 | struct sched_domain *parent; |
1da177e4 | 6765 | |
7f4588f3 AH |
6766 | d->sd_allnodes = 0; |
6767 | if (cpumask_weight(cpu_map) > | |
6768 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6769 | sd = &per_cpu(allnodes_domains, i).sd; | |
6770 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6771 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6772 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6773 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6774 | d->sd_allnodes = 1; | |
6775 | } | |
6776 | parent = sd; | |
6777 | ||
6778 | sd = &per_cpu(node_domains, i).sd; | |
6779 | SD_INIT(sd, NODE); | |
6780 | set_domain_attribute(sd, attr); | |
6781 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6782 | sd->parent = parent; | |
6783 | if (parent) | |
6784 | parent->child = sd; | |
6785 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6786 | #endif |
7f4588f3 AH |
6787 | return sd; |
6788 | } | |
1da177e4 | 6789 | |
87cce662 AH |
6790 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6791 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6792 | struct sched_domain *parent, int i) | |
6793 | { | |
6794 | struct sched_domain *sd; | |
6795 | sd = &per_cpu(phys_domains, i).sd; | |
6796 | SD_INIT(sd, CPU); | |
6797 | set_domain_attribute(sd, attr); | |
6798 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6799 | sd->parent = parent; | |
6800 | if (parent) | |
6801 | parent->child = sd; | |
6802 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6803 | return sd; | |
6804 | } | |
1da177e4 | 6805 | |
410c4081 AH |
6806 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
6807 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6808 | struct sched_domain *parent, int i) | |
6809 | { | |
6810 | struct sched_domain *sd = parent; | |
1e9f28fa | 6811 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
6812 | sd = &per_cpu(core_domains, i).sd; |
6813 | SD_INIT(sd, MC); | |
6814 | set_domain_attribute(sd, attr); | |
6815 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
6816 | sd->parent = parent; | |
6817 | parent->child = sd; | |
6818 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 6819 | #endif |
410c4081 AH |
6820 | return sd; |
6821 | } | |
1e9f28fa | 6822 | |
d8173535 AH |
6823 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
6824 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6825 | struct sched_domain *parent, int i) | |
6826 | { | |
6827 | struct sched_domain *sd = parent; | |
1da177e4 | 6828 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
6829 | sd = &per_cpu(cpu_domains, i).sd; |
6830 | SD_INIT(sd, SIBLING); | |
6831 | set_domain_attribute(sd, attr); | |
6832 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
6833 | sd->parent = parent; | |
6834 | parent->child = sd; | |
6835 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 6836 | #endif |
d8173535 AH |
6837 | return sd; |
6838 | } | |
1da177e4 | 6839 | |
0e8e85c9 AH |
6840 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
6841 | const struct cpumask *cpu_map, int cpu) | |
6842 | { | |
6843 | switch (l) { | |
1da177e4 | 6844 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
6845 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
6846 | cpumask_and(d->this_sibling_map, cpu_map, | |
6847 | topology_thread_cpumask(cpu)); | |
6848 | if (cpu == cpumask_first(d->this_sibling_map)) | |
6849 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
6850 | &cpu_to_cpu_group, | |
6851 | d->send_covered, d->tmpmask); | |
6852 | break; | |
1da177e4 | 6853 | #endif |
1e9f28fa | 6854 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
6855 | case SD_LV_MC: /* set up multi-core groups */ |
6856 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
6857 | if (cpu == cpumask_first(d->this_core_map)) | |
6858 | init_sched_build_groups(d->this_core_map, cpu_map, | |
6859 | &cpu_to_core_group, | |
6860 | d->send_covered, d->tmpmask); | |
6861 | break; | |
1e9f28fa | 6862 | #endif |
86548096 AH |
6863 | case SD_LV_CPU: /* set up physical groups */ |
6864 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
6865 | if (!cpumask_empty(d->nodemask)) | |
6866 | init_sched_build_groups(d->nodemask, cpu_map, | |
6867 | &cpu_to_phys_group, | |
6868 | d->send_covered, d->tmpmask); | |
6869 | break; | |
1da177e4 | 6870 | #ifdef CONFIG_NUMA |
de616e36 AH |
6871 | case SD_LV_ALLNODES: |
6872 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
6873 | d->send_covered, d->tmpmask); | |
6874 | break; | |
6875 | #endif | |
0e8e85c9 AH |
6876 | default: |
6877 | break; | |
7c16ec58 | 6878 | } |
0e8e85c9 | 6879 | } |
9c1cfda2 | 6880 | |
2109b99e AH |
6881 | /* |
6882 | * Build sched domains for a given set of cpus and attach the sched domains | |
6883 | * to the individual cpus | |
6884 | */ | |
6885 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
6886 | struct sched_domain_attr *attr) | |
6887 | { | |
6888 | enum s_alloc alloc_state = sa_none; | |
6889 | struct s_data d; | |
294b0c96 | 6890 | struct sched_domain *sd; |
2109b99e | 6891 | int i; |
7c16ec58 | 6892 | #ifdef CONFIG_NUMA |
2109b99e | 6893 | d.sd_allnodes = 0; |
7c16ec58 | 6894 | #endif |
9c1cfda2 | 6895 | |
2109b99e AH |
6896 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6897 | if (alloc_state != sa_rootdomain) | |
6898 | goto error; | |
6899 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 6900 | |
1da177e4 | 6901 | /* |
1a20ff27 | 6902 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6903 | */ |
abcd083a | 6904 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
6905 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
6906 | cpu_map); | |
9761eea8 | 6907 | |
7f4588f3 | 6908 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 6909 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 6910 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 6911 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 6912 | } |
9c1cfda2 | 6913 | |
abcd083a | 6914 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 6915 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 6916 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 6917 | } |
9c1cfda2 | 6918 | |
1da177e4 | 6919 | /* Set up physical groups */ |
86548096 AH |
6920 | for (i = 0; i < nr_node_ids; i++) |
6921 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 6922 | |
1da177e4 LT |
6923 | #ifdef CONFIG_NUMA |
6924 | /* Set up node groups */ | |
de616e36 AH |
6925 | if (d.sd_allnodes) |
6926 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 6927 | |
0601a88d AH |
6928 | for (i = 0; i < nr_node_ids; i++) |
6929 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 6930 | goto error; |
1da177e4 LT |
6931 | #endif |
6932 | ||
6933 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 6934 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 6935 | for_each_cpu(i, cpu_map) { |
294b0c96 | 6936 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 6937 | init_sched_groups_power(i, sd); |
5c45bf27 | 6938 | } |
1da177e4 | 6939 | #endif |
1e9f28fa | 6940 | #ifdef CONFIG_SCHED_MC |
abcd083a | 6941 | for_each_cpu(i, cpu_map) { |
294b0c96 | 6942 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 6943 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
6944 | } |
6945 | #endif | |
1e9f28fa | 6946 | |
abcd083a | 6947 | for_each_cpu(i, cpu_map) { |
294b0c96 | 6948 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 6949 | init_sched_groups_power(i, sd); |
1da177e4 LT |
6950 | } |
6951 | ||
9c1cfda2 | 6952 | #ifdef CONFIG_NUMA |
076ac2af | 6953 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 6954 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 6955 | |
49a02c51 | 6956 | if (d.sd_allnodes) { |
6711cab4 | 6957 | struct sched_group *sg; |
f712c0c7 | 6958 | |
96f874e2 | 6959 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 6960 | d.tmpmask); |
f712c0c7 SS |
6961 | init_numa_sched_groups_power(sg); |
6962 | } | |
9c1cfda2 JH |
6963 | #endif |
6964 | ||
1da177e4 | 6965 | /* Attach the domains */ |
abcd083a | 6966 | for_each_cpu(i, cpu_map) { |
1da177e4 | 6967 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6968 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 6969 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 6970 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 6971 | #else |
6c99e9ad | 6972 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 6973 | #endif |
49a02c51 | 6974 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 6975 | } |
51888ca2 | 6976 | |
2109b99e AH |
6977 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
6978 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
6979 | return 0; | |
51888ca2 | 6980 | |
51888ca2 | 6981 | error: |
2109b99e AH |
6982 | __free_domain_allocs(&d, alloc_state, cpu_map); |
6983 | return -ENOMEM; | |
1da177e4 | 6984 | } |
029190c5 | 6985 | |
96f874e2 | 6986 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
6987 | { |
6988 | return __build_sched_domains(cpu_map, NULL); | |
6989 | } | |
6990 | ||
acc3f5d7 | 6991 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 6992 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
6993 | static struct sched_domain_attr *dattr_cur; |
6994 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
6995 | |
6996 | /* | |
6997 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
6998 | * cpumask) fails, then fallback to a single sched domain, |
6999 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7000 | */ |
4212823f | 7001 | static cpumask_var_t fallback_doms; |
029190c5 | 7002 | |
ee79d1bd HC |
7003 | /* |
7004 | * arch_update_cpu_topology lets virtualized architectures update the | |
7005 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7006 | * or 0 if it stayed the same. | |
7007 | */ | |
7008 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7009 | { |
ee79d1bd | 7010 | return 0; |
22e52b07 HC |
7011 | } |
7012 | ||
acc3f5d7 RR |
7013 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7014 | { | |
7015 | int i; | |
7016 | cpumask_var_t *doms; | |
7017 | ||
7018 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7019 | if (!doms) | |
7020 | return NULL; | |
7021 | for (i = 0; i < ndoms; i++) { | |
7022 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7023 | free_sched_domains(doms, i); | |
7024 | return NULL; | |
7025 | } | |
7026 | } | |
7027 | return doms; | |
7028 | } | |
7029 | ||
7030 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7031 | { | |
7032 | unsigned int i; | |
7033 | for (i = 0; i < ndoms; i++) | |
7034 | free_cpumask_var(doms[i]); | |
7035 | kfree(doms); | |
7036 | } | |
7037 | ||
1a20ff27 | 7038 | /* |
41a2d6cf | 7039 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7040 | * For now this just excludes isolated cpus, but could be used to |
7041 | * exclude other special cases in the future. | |
1a20ff27 | 7042 | */ |
96f874e2 | 7043 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7044 | { |
7378547f MM |
7045 | int err; |
7046 | ||
22e52b07 | 7047 | arch_update_cpu_topology(); |
029190c5 | 7048 | ndoms_cur = 1; |
acc3f5d7 | 7049 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7050 | if (!doms_cur) |
acc3f5d7 RR |
7051 | doms_cur = &fallback_doms; |
7052 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7053 | dattr_cur = NULL; |
acc3f5d7 | 7054 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7055 | register_sched_domain_sysctl(); |
7378547f MM |
7056 | |
7057 | return err; | |
1a20ff27 DG |
7058 | } |
7059 | ||
96f874e2 RR |
7060 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7061 | struct cpumask *tmpmask) | |
1da177e4 | 7062 | { |
7c16ec58 | 7063 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7064 | } |
1da177e4 | 7065 | |
1a20ff27 DG |
7066 | /* |
7067 | * Detach sched domains from a group of cpus specified in cpu_map | |
7068 | * These cpus will now be attached to the NULL domain | |
7069 | */ | |
96f874e2 | 7070 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7071 | { |
96f874e2 RR |
7072 | /* Save because hotplug lock held. */ |
7073 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7074 | int i; |
7075 | ||
abcd083a | 7076 | for_each_cpu(i, cpu_map) |
57d885fe | 7077 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7078 | synchronize_sched(); |
96f874e2 | 7079 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7080 | } |
7081 | ||
1d3504fc HS |
7082 | /* handle null as "default" */ |
7083 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7084 | struct sched_domain_attr *new, int idx_new) | |
7085 | { | |
7086 | struct sched_domain_attr tmp; | |
7087 | ||
7088 | /* fast path */ | |
7089 | if (!new && !cur) | |
7090 | return 1; | |
7091 | ||
7092 | tmp = SD_ATTR_INIT; | |
7093 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7094 | new ? (new + idx_new) : &tmp, | |
7095 | sizeof(struct sched_domain_attr)); | |
7096 | } | |
7097 | ||
029190c5 PJ |
7098 | /* |
7099 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7100 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7101 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7102 | * It destroys each deleted domain and builds each new domain. | |
7103 | * | |
acc3f5d7 | 7104 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7105 | * The masks don't intersect (don't overlap.) We should setup one |
7106 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7107 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7108 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7109 | * it as it is. | |
7110 | * | |
acc3f5d7 RR |
7111 | * The passed in 'doms_new' should be allocated using |
7112 | * alloc_sched_domains. This routine takes ownership of it and will | |
7113 | * free_sched_domains it when done with it. If the caller failed the | |
7114 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7115 | * and partition_sched_domains() will fallback to the single partition | |
7116 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7117 | * |
96f874e2 | 7118 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7119 | * ndoms_new == 0 is a special case for destroying existing domains, |
7120 | * and it will not create the default domain. | |
dfb512ec | 7121 | * |
029190c5 PJ |
7122 | * Call with hotplug lock held |
7123 | */ | |
acc3f5d7 | 7124 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7125 | struct sched_domain_attr *dattr_new) |
029190c5 | 7126 | { |
dfb512ec | 7127 | int i, j, n; |
d65bd5ec | 7128 | int new_topology; |
029190c5 | 7129 | |
712555ee | 7130 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7131 | |
7378547f MM |
7132 | /* always unregister in case we don't destroy any domains */ |
7133 | unregister_sched_domain_sysctl(); | |
7134 | ||
d65bd5ec HC |
7135 | /* Let architecture update cpu core mappings. */ |
7136 | new_topology = arch_update_cpu_topology(); | |
7137 | ||
dfb512ec | 7138 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7139 | |
7140 | /* Destroy deleted domains */ | |
7141 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7142 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7143 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7144 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7145 | goto match1; |
7146 | } | |
7147 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7148 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7149 | match1: |
7150 | ; | |
7151 | } | |
7152 | ||
e761b772 MK |
7153 | if (doms_new == NULL) { |
7154 | ndoms_cur = 0; | |
acc3f5d7 | 7155 | doms_new = &fallback_doms; |
6ad4c188 | 7156 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7157 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7158 | } |
7159 | ||
029190c5 PJ |
7160 | /* Build new domains */ |
7161 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7162 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7163 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7164 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7165 | goto match2; |
7166 | } | |
7167 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7168 | __build_sched_domains(doms_new[i], |
1d3504fc | 7169 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7170 | match2: |
7171 | ; | |
7172 | } | |
7173 | ||
7174 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7175 | if (doms_cur != &fallback_doms) |
7176 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7177 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7178 | doms_cur = doms_new; |
1d3504fc | 7179 | dattr_cur = dattr_new; |
029190c5 | 7180 | ndoms_cur = ndoms_new; |
7378547f MM |
7181 | |
7182 | register_sched_domain_sysctl(); | |
a1835615 | 7183 | |
712555ee | 7184 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7185 | } |
7186 | ||
5c45bf27 | 7187 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7188 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7189 | { |
95402b38 | 7190 | get_online_cpus(); |
dfb512ec MK |
7191 | |
7192 | /* Destroy domains first to force the rebuild */ | |
7193 | partition_sched_domains(0, NULL, NULL); | |
7194 | ||
e761b772 | 7195 | rebuild_sched_domains(); |
95402b38 | 7196 | put_online_cpus(); |
5c45bf27 SS |
7197 | } |
7198 | ||
7199 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7200 | { | |
afb8a9b7 | 7201 | unsigned int level = 0; |
5c45bf27 | 7202 | |
afb8a9b7 GS |
7203 | if (sscanf(buf, "%u", &level) != 1) |
7204 | return -EINVAL; | |
7205 | ||
7206 | /* | |
7207 | * level is always be positive so don't check for | |
7208 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7209 | * What happens on 0 or 1 byte write, | |
7210 | * need to check for count as well? | |
7211 | */ | |
7212 | ||
7213 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7214 | return -EINVAL; |
7215 | ||
7216 | if (smt) | |
afb8a9b7 | 7217 | sched_smt_power_savings = level; |
5c45bf27 | 7218 | else |
afb8a9b7 | 7219 | sched_mc_power_savings = level; |
5c45bf27 | 7220 | |
c70f22d2 | 7221 | arch_reinit_sched_domains(); |
5c45bf27 | 7222 | |
c70f22d2 | 7223 | return count; |
5c45bf27 SS |
7224 | } |
7225 | ||
5c45bf27 | 7226 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7227 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7228 | struct sysdev_class_attribute *attr, |
f718cd4a | 7229 | char *page) |
5c45bf27 SS |
7230 | { |
7231 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7232 | } | |
f718cd4a | 7233 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7234 | struct sysdev_class_attribute *attr, |
48f24c4d | 7235 | const char *buf, size_t count) |
5c45bf27 SS |
7236 | { |
7237 | return sched_power_savings_store(buf, count, 0); | |
7238 | } | |
f718cd4a AK |
7239 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7240 | sched_mc_power_savings_show, | |
7241 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7242 | #endif |
7243 | ||
7244 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7245 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7246 | struct sysdev_class_attribute *attr, |
f718cd4a | 7247 | char *page) |
5c45bf27 SS |
7248 | { |
7249 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7250 | } | |
f718cd4a | 7251 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7252 | struct sysdev_class_attribute *attr, |
48f24c4d | 7253 | const char *buf, size_t count) |
5c45bf27 SS |
7254 | { |
7255 | return sched_power_savings_store(buf, count, 1); | |
7256 | } | |
f718cd4a AK |
7257 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7258 | sched_smt_power_savings_show, | |
6707de00 AB |
7259 | sched_smt_power_savings_store); |
7260 | #endif | |
7261 | ||
39aac648 | 7262 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7263 | { |
7264 | int err = 0; | |
7265 | ||
7266 | #ifdef CONFIG_SCHED_SMT | |
7267 | if (smt_capable()) | |
7268 | err = sysfs_create_file(&cls->kset.kobj, | |
7269 | &attr_sched_smt_power_savings.attr); | |
7270 | #endif | |
7271 | #ifdef CONFIG_SCHED_MC | |
7272 | if (!err && mc_capable()) | |
7273 | err = sysfs_create_file(&cls->kset.kobj, | |
7274 | &attr_sched_mc_power_savings.attr); | |
7275 | #endif | |
7276 | return err; | |
7277 | } | |
6d6bc0ad | 7278 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7279 | |
e761b772 | 7280 | #ifndef CONFIG_CPUSETS |
1da177e4 | 7281 | /* |
e761b772 MK |
7282 | * Add online and remove offline CPUs from the scheduler domains. |
7283 | * When cpusets are enabled they take over this function. | |
1da177e4 LT |
7284 | */ |
7285 | static int update_sched_domains(struct notifier_block *nfb, | |
7286 | unsigned long action, void *hcpu) | |
e761b772 MK |
7287 | { |
7288 | switch (action) { | |
7289 | case CPU_ONLINE: | |
7290 | case CPU_ONLINE_FROZEN: | |
6ad4c188 PZ |
7291 | case CPU_DOWN_PREPARE: |
7292 | case CPU_DOWN_PREPARE_FROZEN: | |
7293 | case CPU_DOWN_FAILED: | |
7294 | case CPU_DOWN_FAILED_FROZEN: | |
dfb512ec | 7295 | partition_sched_domains(1, NULL, NULL); |
e761b772 MK |
7296 | return NOTIFY_OK; |
7297 | ||
7298 | default: | |
7299 | return NOTIFY_DONE; | |
7300 | } | |
7301 | } | |
7302 | #endif | |
7303 | ||
7304 | static int update_runtime(struct notifier_block *nfb, | |
7305 | unsigned long action, void *hcpu) | |
1da177e4 | 7306 | { |
7def2be1 PZ |
7307 | int cpu = (int)(long)hcpu; |
7308 | ||
1da177e4 | 7309 | switch (action) { |
1da177e4 | 7310 | case CPU_DOWN_PREPARE: |
8bb78442 | 7311 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7312 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7313 | return NOTIFY_OK; |
7314 | ||
1da177e4 | 7315 | case CPU_DOWN_FAILED: |
8bb78442 | 7316 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7317 | case CPU_ONLINE: |
8bb78442 | 7318 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7319 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7320 | return NOTIFY_OK; |
7321 | ||
1da177e4 LT |
7322 | default: |
7323 | return NOTIFY_DONE; | |
7324 | } | |
1da177e4 | 7325 | } |
1da177e4 LT |
7326 | |
7327 | void __init sched_init_smp(void) | |
7328 | { | |
dcc30a35 RR |
7329 | cpumask_var_t non_isolated_cpus; |
7330 | ||
7331 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7332 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7333 | |
434d53b0 MT |
7334 | #if defined(CONFIG_NUMA) |
7335 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7336 | GFP_KERNEL); | |
7337 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7338 | #endif | |
95402b38 | 7339 | get_online_cpus(); |
712555ee | 7340 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7341 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7342 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7343 | if (cpumask_empty(non_isolated_cpus)) | |
7344 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7345 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7346 | put_online_cpus(); |
e761b772 MK |
7347 | |
7348 | #ifndef CONFIG_CPUSETS | |
1da177e4 LT |
7349 | /* XXX: Theoretical race here - CPU may be hotplugged now */ |
7350 | hotcpu_notifier(update_sched_domains, 0); | |
e761b772 MK |
7351 | #endif |
7352 | ||
7353 | /* RT runtime code needs to handle some hotplug events */ | |
7354 | hotcpu_notifier(update_runtime, 0); | |
7355 | ||
b328ca18 | 7356 | init_hrtick(); |
5c1e1767 NP |
7357 | |
7358 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7359 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7360 | BUG(); |
19978ca6 | 7361 | sched_init_granularity(); |
dcc30a35 | 7362 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7363 | |
0e3900e6 | 7364 | init_sched_rt_class(); |
1da177e4 LT |
7365 | } |
7366 | #else | |
7367 | void __init sched_init_smp(void) | |
7368 | { | |
19978ca6 | 7369 | sched_init_granularity(); |
1da177e4 LT |
7370 | } |
7371 | #endif /* CONFIG_SMP */ | |
7372 | ||
cd1bb94b AB |
7373 | const_debug unsigned int sysctl_timer_migration = 1; |
7374 | ||
1da177e4 LT |
7375 | int in_sched_functions(unsigned long addr) |
7376 | { | |
1da177e4 LT |
7377 | return in_lock_functions(addr) || |
7378 | (addr >= (unsigned long)__sched_text_start | |
7379 | && addr < (unsigned long)__sched_text_end); | |
7380 | } | |
7381 | ||
a9957449 | 7382 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7383 | { |
7384 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7385 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7386 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7387 | cfs_rq->rq = rq; | |
7388 | #endif | |
67e9fb2a | 7389 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7390 | } |
7391 | ||
fa85ae24 PZ |
7392 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7393 | { | |
7394 | struct rt_prio_array *array; | |
7395 | int i; | |
7396 | ||
7397 | array = &rt_rq->active; | |
7398 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7399 | INIT_LIST_HEAD(array->queue + i); | |
7400 | __clear_bit(i, array->bitmap); | |
7401 | } | |
7402 | /* delimiter for bitsearch: */ | |
7403 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7404 | ||
052f1dc7 | 7405 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7406 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7407 | #ifdef CONFIG_SMP |
e864c499 | 7408 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7409 | #endif |
48d5e258 | 7410 | #endif |
fa85ae24 PZ |
7411 | #ifdef CONFIG_SMP |
7412 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7413 | rt_rq->overloaded = 0; |
05fa785c | 7414 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7415 | #endif |
7416 | ||
7417 | rt_rq->rt_time = 0; | |
7418 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7419 | rt_rq->rt_runtime = 0; |
0986b11b | 7420 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7421 | |
052f1dc7 | 7422 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7423 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7424 | rt_rq->rq = rq; |
7425 | #endif | |
fa85ae24 PZ |
7426 | } |
7427 | ||
6f505b16 | 7428 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7429 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7430 | struct sched_entity *se, int cpu, int add, | |
7431 | struct sched_entity *parent) | |
6f505b16 | 7432 | { |
ec7dc8ac | 7433 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7434 | tg->cfs_rq[cpu] = cfs_rq; |
7435 | init_cfs_rq(cfs_rq, rq); | |
7436 | cfs_rq->tg = tg; | |
7437 | if (add) | |
7438 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7439 | ||
7440 | tg->se[cpu] = se; | |
354d60c2 DG |
7441 | /* se could be NULL for init_task_group */ |
7442 | if (!se) | |
7443 | return; | |
7444 | ||
ec7dc8ac DG |
7445 | if (!parent) |
7446 | se->cfs_rq = &rq->cfs; | |
7447 | else | |
7448 | se->cfs_rq = parent->my_q; | |
7449 | ||
6f505b16 PZ |
7450 | se->my_q = cfs_rq; |
7451 | se->load.weight = tg->shares; | |
e05510d0 | 7452 | se->load.inv_weight = 0; |
ec7dc8ac | 7453 | se->parent = parent; |
6f505b16 | 7454 | } |
052f1dc7 | 7455 | #endif |
6f505b16 | 7456 | |
052f1dc7 | 7457 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7458 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7459 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7460 | struct sched_rt_entity *parent) | |
6f505b16 | 7461 | { |
ec7dc8ac DG |
7462 | struct rq *rq = cpu_rq(cpu); |
7463 | ||
6f505b16 PZ |
7464 | tg->rt_rq[cpu] = rt_rq; |
7465 | init_rt_rq(rt_rq, rq); | |
7466 | rt_rq->tg = tg; | |
ac086bc2 | 7467 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7468 | if (add) |
7469 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7470 | ||
7471 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7472 | if (!rt_se) |
7473 | return; | |
7474 | ||
ec7dc8ac DG |
7475 | if (!parent) |
7476 | rt_se->rt_rq = &rq->rt; | |
7477 | else | |
7478 | rt_se->rt_rq = parent->my_q; | |
7479 | ||
6f505b16 | 7480 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7481 | rt_se->parent = parent; |
6f505b16 PZ |
7482 | INIT_LIST_HEAD(&rt_se->run_list); |
7483 | } | |
7484 | #endif | |
7485 | ||
1da177e4 LT |
7486 | void __init sched_init(void) |
7487 | { | |
dd41f596 | 7488 | int i, j; |
434d53b0 MT |
7489 | unsigned long alloc_size = 0, ptr; |
7490 | ||
7491 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7492 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7493 | #endif | |
7494 | #ifdef CONFIG_RT_GROUP_SCHED | |
7495 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7496 | #endif |
df7c8e84 | 7497 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7498 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7499 | #endif |
434d53b0 | 7500 | if (alloc_size) { |
36b7b6d4 | 7501 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7502 | |
7503 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7504 | init_task_group.se = (struct sched_entity **)ptr; | |
7505 | ptr += nr_cpu_ids * sizeof(void **); | |
7506 | ||
7507 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7508 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7509 | |
6d6bc0ad | 7510 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7511 | #ifdef CONFIG_RT_GROUP_SCHED |
7512 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7513 | ptr += nr_cpu_ids * sizeof(void **); | |
7514 | ||
7515 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7516 | ptr += nr_cpu_ids * sizeof(void **); |
7517 | ||
6d6bc0ad | 7518 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7519 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7520 | for_each_possible_cpu(i) { | |
7521 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7522 | ptr += cpumask_size(); | |
7523 | } | |
7524 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7525 | } |
dd41f596 | 7526 | |
57d885fe GH |
7527 | #ifdef CONFIG_SMP |
7528 | init_defrootdomain(); | |
7529 | #endif | |
7530 | ||
d0b27fa7 PZ |
7531 | init_rt_bandwidth(&def_rt_bandwidth, |
7532 | global_rt_period(), global_rt_runtime()); | |
7533 | ||
7534 | #ifdef CONFIG_RT_GROUP_SCHED | |
7535 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7536 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7537 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7538 | |
7c941438 | 7539 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7540 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7541 | INIT_LIST_HEAD(&init_task_group.children); |
7542 | ||
7c941438 | 7543 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7544 | |
4a6cc4bd JK |
7545 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7546 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7547 | __alignof__(unsigned long)); | |
7548 | #endif | |
0a945022 | 7549 | for_each_possible_cpu(i) { |
70b97a7f | 7550 | struct rq *rq; |
1da177e4 LT |
7551 | |
7552 | rq = cpu_rq(i); | |
05fa785c | 7553 | raw_spin_lock_init(&rq->lock); |
7897986b | 7554 | rq->nr_running = 0; |
dce48a84 TG |
7555 | rq->calc_load_active = 0; |
7556 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7557 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7558 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7559 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7560 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7561 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7562 | #ifdef CONFIG_CGROUP_SCHED |
7563 | /* | |
7564 | * How much cpu bandwidth does init_task_group get? | |
7565 | * | |
7566 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7567 | * gets 100% of the cpu resources in the system. This overall | |
7568 | * system cpu resource is divided among the tasks of | |
7569 | * init_task_group and its child task-groups in a fair manner, | |
7570 | * based on each entity's (task or task-group's) weight | |
7571 | * (se->load.weight). | |
7572 | * | |
7573 | * In other words, if init_task_group has 10 tasks of weight | |
7574 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7575 | * then A0's share of the cpu resource is: | |
7576 | * | |
0d905bca | 7577 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7578 | * |
7579 | * We achieve this by letting init_task_group's tasks sit | |
7580 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7581 | */ | |
ec7dc8ac | 7582 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7583 | #endif |
354d60c2 DG |
7584 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7585 | ||
7586 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7587 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7588 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7589 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7590 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7591 | #endif |
dd41f596 | 7592 | #endif |
1da177e4 | 7593 | |
dd41f596 IM |
7594 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7595 | rq->cpu_load[j] = 0; | |
1da177e4 | 7596 | #ifdef CONFIG_SMP |
41c7ce9a | 7597 | rq->sd = NULL; |
57d885fe | 7598 | rq->rd = NULL; |
3f029d3c | 7599 | rq->post_schedule = 0; |
1da177e4 | 7600 | rq->active_balance = 0; |
dd41f596 | 7601 | rq->next_balance = jiffies; |
1da177e4 | 7602 | rq->push_cpu = 0; |
0a2966b4 | 7603 | rq->cpu = i; |
1f11eb6a | 7604 | rq->online = 0; |
eae0c9df MG |
7605 | rq->idle_stamp = 0; |
7606 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7607 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7608 | #endif |
8f4d37ec | 7609 | init_rq_hrtick(rq); |
1da177e4 | 7610 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7611 | } |
7612 | ||
2dd73a4f | 7613 | set_load_weight(&init_task); |
b50f60ce | 7614 | |
e107be36 AK |
7615 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7616 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7617 | #endif | |
7618 | ||
c9819f45 | 7619 | #ifdef CONFIG_SMP |
962cf36c | 7620 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7621 | #endif |
7622 | ||
b50f60ce | 7623 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7624 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7625 | #endif |
7626 | ||
1da177e4 LT |
7627 | /* |
7628 | * The boot idle thread does lazy MMU switching as well: | |
7629 | */ | |
7630 | atomic_inc(&init_mm.mm_count); | |
7631 | enter_lazy_tlb(&init_mm, current); | |
7632 | ||
7633 | /* | |
7634 | * Make us the idle thread. Technically, schedule() should not be | |
7635 | * called from this thread, however somewhere below it might be, | |
7636 | * but because we are the idle thread, we just pick up running again | |
7637 | * when this runqueue becomes "idle". | |
7638 | */ | |
7639 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7640 | |
7641 | calc_load_update = jiffies + LOAD_FREQ; | |
7642 | ||
dd41f596 IM |
7643 | /* |
7644 | * During early bootup we pretend to be a normal task: | |
7645 | */ | |
7646 | current->sched_class = &fair_sched_class; | |
6892b75e | 7647 | |
6a7b3dc3 | 7648 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7649 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7650 | #ifdef CONFIG_SMP |
7d1e6a9b | 7651 | #ifdef CONFIG_NO_HZ |
49557e62 | 7652 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 7653 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 7654 | #endif |
bdddd296 RR |
7655 | /* May be allocated at isolcpus cmdline parse time */ |
7656 | if (cpu_isolated_map == NULL) | |
7657 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7658 | #endif /* SMP */ |
6a7b3dc3 | 7659 | |
cdd6c482 | 7660 | perf_event_init(); |
0d905bca | 7661 | |
6892b75e | 7662 | scheduler_running = 1; |
1da177e4 LT |
7663 | } |
7664 | ||
7665 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7666 | static inline int preempt_count_equals(int preempt_offset) |
7667 | { | |
234da7bc | 7668 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7669 | |
7670 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7671 | } | |
7672 | ||
d894837f | 7673 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7674 | { |
48f24c4d | 7675 | #ifdef in_atomic |
1da177e4 LT |
7676 | static unsigned long prev_jiffy; /* ratelimiting */ |
7677 | ||
e4aafea2 FW |
7678 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7679 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7680 | return; |
7681 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7682 | return; | |
7683 | prev_jiffy = jiffies; | |
7684 | ||
3df0fc5b PZ |
7685 | printk(KERN_ERR |
7686 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7687 | file, line); | |
7688 | printk(KERN_ERR | |
7689 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7690 | in_atomic(), irqs_disabled(), | |
7691 | current->pid, current->comm); | |
aef745fc IM |
7692 | |
7693 | debug_show_held_locks(current); | |
7694 | if (irqs_disabled()) | |
7695 | print_irqtrace_events(current); | |
7696 | dump_stack(); | |
1da177e4 LT |
7697 | #endif |
7698 | } | |
7699 | EXPORT_SYMBOL(__might_sleep); | |
7700 | #endif | |
7701 | ||
7702 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7703 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7704 | { | |
7705 | int on_rq; | |
3e51f33f | 7706 | |
3a5e4dc1 AK |
7707 | on_rq = p->se.on_rq; |
7708 | if (on_rq) | |
7709 | deactivate_task(rq, p, 0); | |
7710 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7711 | if (on_rq) { | |
7712 | activate_task(rq, p, 0); | |
7713 | resched_task(rq->curr); | |
7714 | } | |
7715 | } | |
7716 | ||
1da177e4 LT |
7717 | void normalize_rt_tasks(void) |
7718 | { | |
a0f98a1c | 7719 | struct task_struct *g, *p; |
1da177e4 | 7720 | unsigned long flags; |
70b97a7f | 7721 | struct rq *rq; |
1da177e4 | 7722 | |
4cf5d77a | 7723 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7724 | do_each_thread(g, p) { |
178be793 IM |
7725 | /* |
7726 | * Only normalize user tasks: | |
7727 | */ | |
7728 | if (!p->mm) | |
7729 | continue; | |
7730 | ||
6cfb0d5d | 7731 | p->se.exec_start = 0; |
6cfb0d5d | 7732 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7733 | p->se.statistics.wait_start = 0; |
7734 | p->se.statistics.sleep_start = 0; | |
7735 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7736 | #endif |
dd41f596 IM |
7737 | |
7738 | if (!rt_task(p)) { | |
7739 | /* | |
7740 | * Renice negative nice level userspace | |
7741 | * tasks back to 0: | |
7742 | */ | |
7743 | if (TASK_NICE(p) < 0 && p->mm) | |
7744 | set_user_nice(p, 0); | |
1da177e4 | 7745 | continue; |
dd41f596 | 7746 | } |
1da177e4 | 7747 | |
1d615482 | 7748 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7749 | rq = __task_rq_lock(p); |
1da177e4 | 7750 | |
178be793 | 7751 | normalize_task(rq, p); |
3a5e4dc1 | 7752 | |
b29739f9 | 7753 | __task_rq_unlock(rq); |
1d615482 | 7754 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7755 | } while_each_thread(g, p); |
7756 | ||
4cf5d77a | 7757 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7758 | } |
7759 | ||
7760 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a LT |
7761 | |
7762 | #ifdef CONFIG_IA64 | |
7763 | /* | |
7764 | * These functions are only useful for the IA64 MCA handling. | |
7765 | * | |
7766 | * They can only be called when the whole system has been | |
7767 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7768 | * activity can take place. Using them for anything else would | |
7769 | * be a serious bug, and as a result, they aren't even visible | |
7770 | * under any other configuration. | |
7771 | */ | |
7772 | ||
7773 | /** | |
7774 | * curr_task - return the current task for a given cpu. | |
7775 | * @cpu: the processor in question. | |
7776 | * | |
7777 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7778 | */ | |
36c8b586 | 7779 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7780 | { |
7781 | return cpu_curr(cpu); | |
7782 | } | |
7783 | ||
7784 | /** | |
7785 | * set_curr_task - set the current task for a given cpu. | |
7786 | * @cpu: the processor in question. | |
7787 | * @p: the task pointer to set. | |
7788 | * | |
7789 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7790 | * are serviced on a separate stack. It allows the architecture to switch the |
7791 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7792 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7793 | * and caller must save the original value of the current task (see | |
7794 | * curr_task() above) and restore that value before reenabling interrupts and | |
7795 | * re-starting the system. | |
7796 | * | |
7797 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7798 | */ | |
36c8b586 | 7799 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7800 | { |
7801 | cpu_curr(cpu) = p; | |
7802 | } | |
7803 | ||
7804 | #endif | |
29f59db3 | 7805 | |
bccbe08a PZ |
7806 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7807 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7808 | { |
7809 | int i; | |
7810 | ||
7811 | for_each_possible_cpu(i) { | |
7812 | if (tg->cfs_rq) | |
7813 | kfree(tg->cfs_rq[i]); | |
7814 | if (tg->se) | |
7815 | kfree(tg->se[i]); | |
6f505b16 PZ |
7816 | } |
7817 | ||
7818 | kfree(tg->cfs_rq); | |
7819 | kfree(tg->se); | |
6f505b16 PZ |
7820 | } |
7821 | ||
ec7dc8ac DG |
7822 | static |
7823 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 7824 | { |
29f59db3 | 7825 | struct cfs_rq *cfs_rq; |
eab17229 | 7826 | struct sched_entity *se; |
9b5b7751 | 7827 | struct rq *rq; |
29f59db3 SV |
7828 | int i; |
7829 | ||
434d53b0 | 7830 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7831 | if (!tg->cfs_rq) |
7832 | goto err; | |
434d53b0 | 7833 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7834 | if (!tg->se) |
7835 | goto err; | |
052f1dc7 PZ |
7836 | |
7837 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7838 | |
7839 | for_each_possible_cpu(i) { | |
9b5b7751 | 7840 | rq = cpu_rq(i); |
29f59db3 | 7841 | |
eab17229 LZ |
7842 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
7843 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
7844 | if (!cfs_rq) |
7845 | goto err; | |
7846 | ||
eab17229 LZ |
7847 | se = kzalloc_node(sizeof(struct sched_entity), |
7848 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 7849 | if (!se) |
dfc12eb2 | 7850 | goto err_free_rq; |
29f59db3 | 7851 | |
eab17229 | 7852 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
7853 | } |
7854 | ||
7855 | return 1; | |
7856 | ||
dfc12eb2 PC |
7857 | err_free_rq: |
7858 | kfree(cfs_rq); | |
bccbe08a PZ |
7859 | err: |
7860 | return 0; | |
7861 | } | |
7862 | ||
7863 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7864 | { | |
7865 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7866 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7867 | } | |
7868 | ||
7869 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7870 | { | |
7871 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7872 | } | |
6d6bc0ad | 7873 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
7874 | static inline void free_fair_sched_group(struct task_group *tg) |
7875 | { | |
7876 | } | |
7877 | ||
ec7dc8ac DG |
7878 | static inline |
7879 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7880 | { |
7881 | return 1; | |
7882 | } | |
7883 | ||
7884 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7885 | { | |
7886 | } | |
7887 | ||
7888 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7889 | { | |
7890 | } | |
6d6bc0ad | 7891 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
7892 | |
7893 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7894 | static void free_rt_sched_group(struct task_group *tg) |
7895 | { | |
7896 | int i; | |
7897 | ||
d0b27fa7 PZ |
7898 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
7899 | ||
bccbe08a PZ |
7900 | for_each_possible_cpu(i) { |
7901 | if (tg->rt_rq) | |
7902 | kfree(tg->rt_rq[i]); | |
7903 | if (tg->rt_se) | |
7904 | kfree(tg->rt_se[i]); | |
7905 | } | |
7906 | ||
7907 | kfree(tg->rt_rq); | |
7908 | kfree(tg->rt_se); | |
7909 | } | |
7910 | ||
ec7dc8ac DG |
7911 | static |
7912 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7913 | { |
7914 | struct rt_rq *rt_rq; | |
eab17229 | 7915 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
7916 | struct rq *rq; |
7917 | int i; | |
7918 | ||
434d53b0 | 7919 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7920 | if (!tg->rt_rq) |
7921 | goto err; | |
434d53b0 | 7922 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
7923 | if (!tg->rt_se) |
7924 | goto err; | |
7925 | ||
d0b27fa7 PZ |
7926 | init_rt_bandwidth(&tg->rt_bandwidth, |
7927 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
7928 | |
7929 | for_each_possible_cpu(i) { | |
7930 | rq = cpu_rq(i); | |
7931 | ||
eab17229 LZ |
7932 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
7933 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
7934 | if (!rt_rq) |
7935 | goto err; | |
29f59db3 | 7936 | |
eab17229 LZ |
7937 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
7938 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 7939 | if (!rt_se) |
dfc12eb2 | 7940 | goto err_free_rq; |
29f59db3 | 7941 | |
eab17229 | 7942 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
7943 | } |
7944 | ||
bccbe08a PZ |
7945 | return 1; |
7946 | ||
dfc12eb2 PC |
7947 | err_free_rq: |
7948 | kfree(rt_rq); | |
bccbe08a PZ |
7949 | err: |
7950 | return 0; | |
7951 | } | |
7952 | ||
7953 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7954 | { | |
7955 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
7956 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
7957 | } | |
7958 | ||
7959 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7960 | { | |
7961 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
7962 | } | |
6d6bc0ad | 7963 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
7964 | static inline void free_rt_sched_group(struct task_group *tg) |
7965 | { | |
7966 | } | |
7967 | ||
ec7dc8ac DG |
7968 | static inline |
7969 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7970 | { |
7971 | return 1; | |
7972 | } | |
7973 | ||
7974 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
7975 | { | |
7976 | } | |
7977 | ||
7978 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
7979 | { | |
7980 | } | |
6d6bc0ad | 7981 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 7982 | |
7c941438 | 7983 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
7984 | static void free_sched_group(struct task_group *tg) |
7985 | { | |
7986 | free_fair_sched_group(tg); | |
7987 | free_rt_sched_group(tg); | |
7988 | kfree(tg); | |
7989 | } | |
7990 | ||
7991 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 7992 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
7993 | { |
7994 | struct task_group *tg; | |
7995 | unsigned long flags; | |
7996 | int i; | |
7997 | ||
7998 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
7999 | if (!tg) | |
8000 | return ERR_PTR(-ENOMEM); | |
8001 | ||
ec7dc8ac | 8002 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8003 | goto err; |
8004 | ||
ec7dc8ac | 8005 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8006 | goto err; |
8007 | ||
8ed36996 | 8008 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8009 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8010 | register_fair_sched_group(tg, i); |
8011 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8012 | } |
6f505b16 | 8013 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8014 | |
8015 | WARN_ON(!parent); /* root should already exist */ | |
8016 | ||
8017 | tg->parent = parent; | |
f473aa5e | 8018 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8019 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8020 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8021 | |
9b5b7751 | 8022 | return tg; |
29f59db3 SV |
8023 | |
8024 | err: | |
6f505b16 | 8025 | free_sched_group(tg); |
29f59db3 SV |
8026 | return ERR_PTR(-ENOMEM); |
8027 | } | |
8028 | ||
9b5b7751 | 8029 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8030 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8031 | { |
29f59db3 | 8032 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8033 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8034 | } |
8035 | ||
9b5b7751 | 8036 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8037 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8038 | { |
8ed36996 | 8039 | unsigned long flags; |
9b5b7751 | 8040 | int i; |
29f59db3 | 8041 | |
8ed36996 | 8042 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8043 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8044 | unregister_fair_sched_group(tg, i); |
8045 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8046 | } |
6f505b16 | 8047 | list_del_rcu(&tg->list); |
f473aa5e | 8048 | list_del_rcu(&tg->siblings); |
8ed36996 | 8049 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8050 | |
9b5b7751 | 8051 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8052 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8053 | } |
8054 | ||
9b5b7751 | 8055 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8056 | * The caller of this function should have put the task in its new group |
8057 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8058 | * reflect its new group. | |
9b5b7751 SV |
8059 | */ |
8060 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8061 | { |
8062 | int on_rq, running; | |
8063 | unsigned long flags; | |
8064 | struct rq *rq; | |
8065 | ||
8066 | rq = task_rq_lock(tsk, &flags); | |
8067 | ||
051a1d1a | 8068 | running = task_current(rq, tsk); |
29f59db3 SV |
8069 | on_rq = tsk->se.on_rq; |
8070 | ||
0e1f3483 | 8071 | if (on_rq) |
29f59db3 | 8072 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8073 | if (unlikely(running)) |
8074 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8075 | |
6f505b16 | 8076 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8077 | |
810b3817 PZ |
8078 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8079 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8080 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8081 | #endif |
8082 | ||
0e1f3483 HS |
8083 | if (unlikely(running)) |
8084 | tsk->sched_class->set_curr_task(rq); | |
8085 | if (on_rq) | |
371fd7e7 | 8086 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8087 | |
29f59db3 SV |
8088 | task_rq_unlock(rq, &flags); |
8089 | } | |
7c941438 | 8090 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8091 | |
052f1dc7 | 8092 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8093 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8094 | { |
8095 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8096 | int on_rq; |
8097 | ||
29f59db3 | 8098 | on_rq = se->on_rq; |
62fb1851 | 8099 | if (on_rq) |
29f59db3 SV |
8100 | dequeue_entity(cfs_rq, se, 0); |
8101 | ||
8102 | se->load.weight = shares; | |
e05510d0 | 8103 | se->load.inv_weight = 0; |
29f59db3 | 8104 | |
62fb1851 | 8105 | if (on_rq) |
29f59db3 | 8106 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8107 | } |
62fb1851 | 8108 | |
c09595f6 PZ |
8109 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8110 | { | |
8111 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8112 | struct rq *rq = cfs_rq->rq; | |
8113 | unsigned long flags; | |
8114 | ||
05fa785c | 8115 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8116 | __set_se_shares(se, shares); |
05fa785c | 8117 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8118 | } |
8119 | ||
8ed36996 PZ |
8120 | static DEFINE_MUTEX(shares_mutex); |
8121 | ||
4cf86d77 | 8122 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8123 | { |
8124 | int i; | |
8ed36996 | 8125 | unsigned long flags; |
c61935fd | 8126 | |
ec7dc8ac DG |
8127 | /* |
8128 | * We can't change the weight of the root cgroup. | |
8129 | */ | |
8130 | if (!tg->se[0]) | |
8131 | return -EINVAL; | |
8132 | ||
18d95a28 PZ |
8133 | if (shares < MIN_SHARES) |
8134 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8135 | else if (shares > MAX_SHARES) |
8136 | shares = MAX_SHARES; | |
62fb1851 | 8137 | |
8ed36996 | 8138 | mutex_lock(&shares_mutex); |
9b5b7751 | 8139 | if (tg->shares == shares) |
5cb350ba | 8140 | goto done; |
29f59db3 | 8141 | |
8ed36996 | 8142 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8143 | for_each_possible_cpu(i) |
8144 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8145 | list_del_rcu(&tg->siblings); |
8ed36996 | 8146 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8147 | |
8148 | /* wait for any ongoing reference to this group to finish */ | |
8149 | synchronize_sched(); | |
8150 | ||
8151 | /* | |
8152 | * Now we are free to modify the group's share on each cpu | |
8153 | * w/o tripping rebalance_share or load_balance_fair. | |
8154 | */ | |
9b5b7751 | 8155 | tg->shares = shares; |
c09595f6 PZ |
8156 | for_each_possible_cpu(i) { |
8157 | /* | |
8158 | * force a rebalance | |
8159 | */ | |
8160 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8161 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8162 | } |
29f59db3 | 8163 | |
6b2d7700 SV |
8164 | /* |
8165 | * Enable load balance activity on this group, by inserting it back on | |
8166 | * each cpu's rq->leaf_cfs_rq_list. | |
8167 | */ | |
8ed36996 | 8168 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8169 | for_each_possible_cpu(i) |
8170 | register_fair_sched_group(tg, i); | |
f473aa5e | 8171 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8172 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8173 | done: |
8ed36996 | 8174 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8175 | return 0; |
29f59db3 SV |
8176 | } |
8177 | ||
5cb350ba DG |
8178 | unsigned long sched_group_shares(struct task_group *tg) |
8179 | { | |
8180 | return tg->shares; | |
8181 | } | |
052f1dc7 | 8182 | #endif |
5cb350ba | 8183 | |
052f1dc7 | 8184 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8185 | /* |
9f0c1e56 | 8186 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8187 | */ |
9f0c1e56 PZ |
8188 | static DEFINE_MUTEX(rt_constraints_mutex); |
8189 | ||
8190 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8191 | { | |
8192 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8193 | return 1ULL << 20; |
9f0c1e56 | 8194 | |
9a7e0b18 | 8195 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8196 | } |
8197 | ||
9a7e0b18 PZ |
8198 | /* Must be called with tasklist_lock held */ |
8199 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8200 | { |
9a7e0b18 | 8201 | struct task_struct *g, *p; |
b40b2e8e | 8202 | |
9a7e0b18 PZ |
8203 | do_each_thread(g, p) { |
8204 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8205 | return 1; | |
8206 | } while_each_thread(g, p); | |
b40b2e8e | 8207 | |
9a7e0b18 PZ |
8208 | return 0; |
8209 | } | |
b40b2e8e | 8210 | |
9a7e0b18 PZ |
8211 | struct rt_schedulable_data { |
8212 | struct task_group *tg; | |
8213 | u64 rt_period; | |
8214 | u64 rt_runtime; | |
8215 | }; | |
b40b2e8e | 8216 | |
9a7e0b18 PZ |
8217 | static int tg_schedulable(struct task_group *tg, void *data) |
8218 | { | |
8219 | struct rt_schedulable_data *d = data; | |
8220 | struct task_group *child; | |
8221 | unsigned long total, sum = 0; | |
8222 | u64 period, runtime; | |
b40b2e8e | 8223 | |
9a7e0b18 PZ |
8224 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8225 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8226 | |
9a7e0b18 PZ |
8227 | if (tg == d->tg) { |
8228 | period = d->rt_period; | |
8229 | runtime = d->rt_runtime; | |
b40b2e8e | 8230 | } |
b40b2e8e | 8231 | |
4653f803 PZ |
8232 | /* |
8233 | * Cannot have more runtime than the period. | |
8234 | */ | |
8235 | if (runtime > period && runtime != RUNTIME_INF) | |
8236 | return -EINVAL; | |
6f505b16 | 8237 | |
4653f803 PZ |
8238 | /* |
8239 | * Ensure we don't starve existing RT tasks. | |
8240 | */ | |
9a7e0b18 PZ |
8241 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8242 | return -EBUSY; | |
6f505b16 | 8243 | |
9a7e0b18 | 8244 | total = to_ratio(period, runtime); |
6f505b16 | 8245 | |
4653f803 PZ |
8246 | /* |
8247 | * Nobody can have more than the global setting allows. | |
8248 | */ | |
8249 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8250 | return -EINVAL; | |
6f505b16 | 8251 | |
4653f803 PZ |
8252 | /* |
8253 | * The sum of our children's runtime should not exceed our own. | |
8254 | */ | |
9a7e0b18 PZ |
8255 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8256 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8257 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8258 | |
9a7e0b18 PZ |
8259 | if (child == d->tg) { |
8260 | period = d->rt_period; | |
8261 | runtime = d->rt_runtime; | |
8262 | } | |
6f505b16 | 8263 | |
9a7e0b18 | 8264 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8265 | } |
6f505b16 | 8266 | |
9a7e0b18 PZ |
8267 | if (sum > total) |
8268 | return -EINVAL; | |
8269 | ||
8270 | return 0; | |
6f505b16 PZ |
8271 | } |
8272 | ||
9a7e0b18 | 8273 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8274 | { |
9a7e0b18 PZ |
8275 | struct rt_schedulable_data data = { |
8276 | .tg = tg, | |
8277 | .rt_period = period, | |
8278 | .rt_runtime = runtime, | |
8279 | }; | |
8280 | ||
8281 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8282 | } |
8283 | ||
d0b27fa7 PZ |
8284 | static int tg_set_bandwidth(struct task_group *tg, |
8285 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8286 | { |
ac086bc2 | 8287 | int i, err = 0; |
9f0c1e56 | 8288 | |
9f0c1e56 | 8289 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8290 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8291 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8292 | if (err) | |
9f0c1e56 | 8293 | goto unlock; |
ac086bc2 | 8294 | |
0986b11b | 8295 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8296 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8297 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8298 | |
8299 | for_each_possible_cpu(i) { | |
8300 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8301 | ||
0986b11b | 8302 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8303 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8304 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8305 | } |
0986b11b | 8306 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8307 | unlock: |
521f1a24 | 8308 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8309 | mutex_unlock(&rt_constraints_mutex); |
8310 | ||
8311 | return err; | |
6f505b16 PZ |
8312 | } |
8313 | ||
d0b27fa7 PZ |
8314 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8315 | { | |
8316 | u64 rt_runtime, rt_period; | |
8317 | ||
8318 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8319 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8320 | if (rt_runtime_us < 0) | |
8321 | rt_runtime = RUNTIME_INF; | |
8322 | ||
8323 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8324 | } | |
8325 | ||
9f0c1e56 PZ |
8326 | long sched_group_rt_runtime(struct task_group *tg) |
8327 | { | |
8328 | u64 rt_runtime_us; | |
8329 | ||
d0b27fa7 | 8330 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8331 | return -1; |
8332 | ||
d0b27fa7 | 8333 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8334 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8335 | return rt_runtime_us; | |
8336 | } | |
d0b27fa7 PZ |
8337 | |
8338 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8339 | { | |
8340 | u64 rt_runtime, rt_period; | |
8341 | ||
8342 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8343 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8344 | ||
619b0488 R |
8345 | if (rt_period == 0) |
8346 | return -EINVAL; | |
8347 | ||
d0b27fa7 PZ |
8348 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8349 | } | |
8350 | ||
8351 | long sched_group_rt_period(struct task_group *tg) | |
8352 | { | |
8353 | u64 rt_period_us; | |
8354 | ||
8355 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8356 | do_div(rt_period_us, NSEC_PER_USEC); | |
8357 | return rt_period_us; | |
8358 | } | |
8359 | ||
8360 | static int sched_rt_global_constraints(void) | |
8361 | { | |
4653f803 | 8362 | u64 runtime, period; |
d0b27fa7 PZ |
8363 | int ret = 0; |
8364 | ||
ec5d4989 HS |
8365 | if (sysctl_sched_rt_period <= 0) |
8366 | return -EINVAL; | |
8367 | ||
4653f803 PZ |
8368 | runtime = global_rt_runtime(); |
8369 | period = global_rt_period(); | |
8370 | ||
8371 | /* | |
8372 | * Sanity check on the sysctl variables. | |
8373 | */ | |
8374 | if (runtime > period && runtime != RUNTIME_INF) | |
8375 | return -EINVAL; | |
10b612f4 | 8376 | |
d0b27fa7 | 8377 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8378 | read_lock(&tasklist_lock); |
4653f803 | 8379 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8380 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8381 | mutex_unlock(&rt_constraints_mutex); |
8382 | ||
8383 | return ret; | |
8384 | } | |
54e99124 DG |
8385 | |
8386 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8387 | { | |
8388 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8389 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8390 | return 0; | |
8391 | ||
8392 | return 1; | |
8393 | } | |
8394 | ||
6d6bc0ad | 8395 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8396 | static int sched_rt_global_constraints(void) |
8397 | { | |
ac086bc2 PZ |
8398 | unsigned long flags; |
8399 | int i; | |
8400 | ||
ec5d4989 HS |
8401 | if (sysctl_sched_rt_period <= 0) |
8402 | return -EINVAL; | |
8403 | ||
60aa605d PZ |
8404 | /* |
8405 | * There's always some RT tasks in the root group | |
8406 | * -- migration, kstopmachine etc.. | |
8407 | */ | |
8408 | if (sysctl_sched_rt_runtime == 0) | |
8409 | return -EBUSY; | |
8410 | ||
0986b11b | 8411 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8412 | for_each_possible_cpu(i) { |
8413 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8414 | ||
0986b11b | 8415 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8416 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8417 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8418 | } |
0986b11b | 8419 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8420 | |
d0b27fa7 PZ |
8421 | return 0; |
8422 | } | |
6d6bc0ad | 8423 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8424 | |
8425 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8426 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8427 | loff_t *ppos) |
8428 | { | |
8429 | int ret; | |
8430 | int old_period, old_runtime; | |
8431 | static DEFINE_MUTEX(mutex); | |
8432 | ||
8433 | mutex_lock(&mutex); | |
8434 | old_period = sysctl_sched_rt_period; | |
8435 | old_runtime = sysctl_sched_rt_runtime; | |
8436 | ||
8d65af78 | 8437 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8438 | |
8439 | if (!ret && write) { | |
8440 | ret = sched_rt_global_constraints(); | |
8441 | if (ret) { | |
8442 | sysctl_sched_rt_period = old_period; | |
8443 | sysctl_sched_rt_runtime = old_runtime; | |
8444 | } else { | |
8445 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8446 | def_rt_bandwidth.rt_period = | |
8447 | ns_to_ktime(global_rt_period()); | |
8448 | } | |
8449 | } | |
8450 | mutex_unlock(&mutex); | |
8451 | ||
8452 | return ret; | |
8453 | } | |
68318b8e | 8454 | |
052f1dc7 | 8455 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8456 | |
8457 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8458 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8459 | { |
2b01dfe3 PM |
8460 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8461 | struct task_group, css); | |
68318b8e SV |
8462 | } |
8463 | ||
8464 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8465 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8466 | { |
ec7dc8ac | 8467 | struct task_group *tg, *parent; |
68318b8e | 8468 | |
2b01dfe3 | 8469 | if (!cgrp->parent) { |
68318b8e | 8470 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8471 | return &init_task_group.css; |
8472 | } | |
8473 | ||
ec7dc8ac DG |
8474 | parent = cgroup_tg(cgrp->parent); |
8475 | tg = sched_create_group(parent); | |
68318b8e SV |
8476 | if (IS_ERR(tg)) |
8477 | return ERR_PTR(-ENOMEM); | |
8478 | ||
68318b8e SV |
8479 | return &tg->css; |
8480 | } | |
8481 | ||
41a2d6cf IM |
8482 | static void |
8483 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8484 | { |
2b01dfe3 | 8485 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8486 | |
8487 | sched_destroy_group(tg); | |
8488 | } | |
8489 | ||
41a2d6cf | 8490 | static int |
be367d09 | 8491 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8492 | { |
b68aa230 | 8493 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8494 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8495 | return -EINVAL; |
8496 | #else | |
68318b8e SV |
8497 | /* We don't support RT-tasks being in separate groups */ |
8498 | if (tsk->sched_class != &fair_sched_class) | |
8499 | return -EINVAL; | |
b68aa230 | 8500 | #endif |
be367d09 BB |
8501 | return 0; |
8502 | } | |
68318b8e | 8503 | |
be367d09 BB |
8504 | static int |
8505 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8506 | struct task_struct *tsk, bool threadgroup) | |
8507 | { | |
8508 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8509 | if (retval) | |
8510 | return retval; | |
8511 | if (threadgroup) { | |
8512 | struct task_struct *c; | |
8513 | rcu_read_lock(); | |
8514 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8515 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8516 | if (retval) { | |
8517 | rcu_read_unlock(); | |
8518 | return retval; | |
8519 | } | |
8520 | } | |
8521 | rcu_read_unlock(); | |
8522 | } | |
68318b8e SV |
8523 | return 0; |
8524 | } | |
8525 | ||
8526 | static void | |
2b01dfe3 | 8527 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8528 | struct cgroup *old_cont, struct task_struct *tsk, |
8529 | bool threadgroup) | |
68318b8e SV |
8530 | { |
8531 | sched_move_task(tsk); | |
be367d09 BB |
8532 | if (threadgroup) { |
8533 | struct task_struct *c; | |
8534 | rcu_read_lock(); | |
8535 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8536 | sched_move_task(c); | |
8537 | } | |
8538 | rcu_read_unlock(); | |
8539 | } | |
68318b8e SV |
8540 | } |
8541 | ||
052f1dc7 | 8542 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8543 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8544 | u64 shareval) |
68318b8e | 8545 | { |
2b01dfe3 | 8546 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8547 | } |
8548 | ||
f4c753b7 | 8549 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8550 | { |
2b01dfe3 | 8551 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8552 | |
8553 | return (u64) tg->shares; | |
8554 | } | |
6d6bc0ad | 8555 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8556 | |
052f1dc7 | 8557 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8558 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8559 | s64 val) |
6f505b16 | 8560 | { |
06ecb27c | 8561 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8562 | } |
8563 | ||
06ecb27c | 8564 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8565 | { |
06ecb27c | 8566 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8567 | } |
d0b27fa7 PZ |
8568 | |
8569 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8570 | u64 rt_period_us) | |
8571 | { | |
8572 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8573 | } | |
8574 | ||
8575 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8576 | { | |
8577 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8578 | } | |
6d6bc0ad | 8579 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8580 | |
fe5c7cc2 | 8581 | static struct cftype cpu_files[] = { |
052f1dc7 | 8582 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8583 | { |
8584 | .name = "shares", | |
f4c753b7 PM |
8585 | .read_u64 = cpu_shares_read_u64, |
8586 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8587 | }, |
052f1dc7 PZ |
8588 | #endif |
8589 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8590 | { |
9f0c1e56 | 8591 | .name = "rt_runtime_us", |
06ecb27c PM |
8592 | .read_s64 = cpu_rt_runtime_read, |
8593 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8594 | }, |
d0b27fa7 PZ |
8595 | { |
8596 | .name = "rt_period_us", | |
f4c753b7 PM |
8597 | .read_u64 = cpu_rt_period_read_uint, |
8598 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8599 | }, |
052f1dc7 | 8600 | #endif |
68318b8e SV |
8601 | }; |
8602 | ||
8603 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8604 | { | |
fe5c7cc2 | 8605 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8606 | } |
8607 | ||
8608 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8609 | .name = "cpu", |
8610 | .create = cpu_cgroup_create, | |
8611 | .destroy = cpu_cgroup_destroy, | |
8612 | .can_attach = cpu_cgroup_can_attach, | |
8613 | .attach = cpu_cgroup_attach, | |
8614 | .populate = cpu_cgroup_populate, | |
8615 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8616 | .early_init = 1, |
8617 | }; | |
8618 | ||
052f1dc7 | 8619 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8620 | |
8621 | #ifdef CONFIG_CGROUP_CPUACCT | |
8622 | ||
8623 | /* | |
8624 | * CPU accounting code for task groups. | |
8625 | * | |
8626 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8627 | * (balbir@in.ibm.com). | |
8628 | */ | |
8629 | ||
934352f2 | 8630 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8631 | struct cpuacct { |
8632 | struct cgroup_subsys_state css; | |
8633 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8634 | u64 __percpu *cpuusage; |
ef12fefa | 8635 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8636 | struct cpuacct *parent; |
d842de87 SV |
8637 | }; |
8638 | ||
8639 | struct cgroup_subsys cpuacct_subsys; | |
8640 | ||
8641 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8642 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8643 | { |
32cd756a | 8644 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8645 | struct cpuacct, css); |
8646 | } | |
8647 | ||
8648 | /* return cpu accounting group to which this task belongs */ | |
8649 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8650 | { | |
8651 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8652 | struct cpuacct, css); | |
8653 | } | |
8654 | ||
8655 | /* create a new cpu accounting group */ | |
8656 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8657 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8658 | { |
8659 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8660 | int i; |
d842de87 SV |
8661 | |
8662 | if (!ca) | |
ef12fefa | 8663 | goto out; |
d842de87 SV |
8664 | |
8665 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8666 | if (!ca->cpuusage) |
8667 | goto out_free_ca; | |
8668 | ||
8669 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8670 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8671 | goto out_free_counters; | |
d842de87 | 8672 | |
934352f2 BR |
8673 | if (cgrp->parent) |
8674 | ca->parent = cgroup_ca(cgrp->parent); | |
8675 | ||
d842de87 | 8676 | return &ca->css; |
ef12fefa BR |
8677 | |
8678 | out_free_counters: | |
8679 | while (--i >= 0) | |
8680 | percpu_counter_destroy(&ca->cpustat[i]); | |
8681 | free_percpu(ca->cpuusage); | |
8682 | out_free_ca: | |
8683 | kfree(ca); | |
8684 | out: | |
8685 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8686 | } |
8687 | ||
8688 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8689 | static void |
32cd756a | 8690 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8691 | { |
32cd756a | 8692 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8693 | int i; |
d842de87 | 8694 | |
ef12fefa BR |
8695 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8696 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8697 | free_percpu(ca->cpuusage); |
8698 | kfree(ca); | |
8699 | } | |
8700 | ||
720f5498 KC |
8701 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8702 | { | |
b36128c8 | 8703 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8704 | u64 data; |
8705 | ||
8706 | #ifndef CONFIG_64BIT | |
8707 | /* | |
8708 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8709 | */ | |
05fa785c | 8710 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8711 | data = *cpuusage; |
05fa785c | 8712 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8713 | #else |
8714 | data = *cpuusage; | |
8715 | #endif | |
8716 | ||
8717 | return data; | |
8718 | } | |
8719 | ||
8720 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8721 | { | |
b36128c8 | 8722 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8723 | |
8724 | #ifndef CONFIG_64BIT | |
8725 | /* | |
8726 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8727 | */ | |
05fa785c | 8728 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8729 | *cpuusage = val; |
05fa785c | 8730 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8731 | #else |
8732 | *cpuusage = val; | |
8733 | #endif | |
8734 | } | |
8735 | ||
d842de87 | 8736 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8737 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8738 | { |
32cd756a | 8739 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8740 | u64 totalcpuusage = 0; |
8741 | int i; | |
8742 | ||
720f5498 KC |
8743 | for_each_present_cpu(i) |
8744 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8745 | |
8746 | return totalcpuusage; | |
8747 | } | |
8748 | ||
0297b803 DG |
8749 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8750 | u64 reset) | |
8751 | { | |
8752 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8753 | int err = 0; | |
8754 | int i; | |
8755 | ||
8756 | if (reset) { | |
8757 | err = -EINVAL; | |
8758 | goto out; | |
8759 | } | |
8760 | ||
720f5498 KC |
8761 | for_each_present_cpu(i) |
8762 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8763 | |
0297b803 DG |
8764 | out: |
8765 | return err; | |
8766 | } | |
8767 | ||
e9515c3c KC |
8768 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8769 | struct seq_file *m) | |
8770 | { | |
8771 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8772 | u64 percpu; | |
8773 | int i; | |
8774 | ||
8775 | for_each_present_cpu(i) { | |
8776 | percpu = cpuacct_cpuusage_read(ca, i); | |
8777 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8778 | } | |
8779 | seq_printf(m, "\n"); | |
8780 | return 0; | |
8781 | } | |
8782 | ||
ef12fefa BR |
8783 | static const char *cpuacct_stat_desc[] = { |
8784 | [CPUACCT_STAT_USER] = "user", | |
8785 | [CPUACCT_STAT_SYSTEM] = "system", | |
8786 | }; | |
8787 | ||
8788 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8789 | struct cgroup_map_cb *cb) | |
8790 | { | |
8791 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8792 | int i; | |
8793 | ||
8794 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8795 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8796 | val = cputime64_to_clock_t(val); | |
8797 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8798 | } | |
8799 | return 0; | |
8800 | } | |
8801 | ||
d842de87 SV |
8802 | static struct cftype files[] = { |
8803 | { | |
8804 | .name = "usage", | |
f4c753b7 PM |
8805 | .read_u64 = cpuusage_read, |
8806 | .write_u64 = cpuusage_write, | |
d842de87 | 8807 | }, |
e9515c3c KC |
8808 | { |
8809 | .name = "usage_percpu", | |
8810 | .read_seq_string = cpuacct_percpu_seq_read, | |
8811 | }, | |
ef12fefa BR |
8812 | { |
8813 | .name = "stat", | |
8814 | .read_map = cpuacct_stats_show, | |
8815 | }, | |
d842de87 SV |
8816 | }; |
8817 | ||
32cd756a | 8818 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8819 | { |
32cd756a | 8820 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8821 | } |
8822 | ||
8823 | /* | |
8824 | * charge this task's execution time to its accounting group. | |
8825 | * | |
8826 | * called with rq->lock held. | |
8827 | */ | |
8828 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8829 | { | |
8830 | struct cpuacct *ca; | |
934352f2 | 8831 | int cpu; |
d842de87 | 8832 | |
c40c6f85 | 8833 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8834 | return; |
8835 | ||
934352f2 | 8836 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8837 | |
8838 | rcu_read_lock(); | |
8839 | ||
d842de87 | 8840 | ca = task_ca(tsk); |
d842de87 | 8841 | |
934352f2 | 8842 | for (; ca; ca = ca->parent) { |
b36128c8 | 8843 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8844 | *cpuusage += cputime; |
8845 | } | |
a18b83b7 BR |
8846 | |
8847 | rcu_read_unlock(); | |
d842de87 SV |
8848 | } |
8849 | ||
fa535a77 AB |
8850 | /* |
8851 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
8852 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
8853 | * percpu_counter_add with values large enough to always overflow the | |
8854 | * per cpu batch limit causing bad SMP scalability. | |
8855 | * | |
8856 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
8857 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
8858 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
8859 | */ | |
8860 | #ifdef CONFIG_SMP | |
8861 | #define CPUACCT_BATCH \ | |
8862 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
8863 | #else | |
8864 | #define CPUACCT_BATCH 0 | |
8865 | #endif | |
8866 | ||
ef12fefa BR |
8867 | /* |
8868 | * Charge the system/user time to the task's accounting group. | |
8869 | */ | |
8870 | static void cpuacct_update_stats(struct task_struct *tsk, | |
8871 | enum cpuacct_stat_index idx, cputime_t val) | |
8872 | { | |
8873 | struct cpuacct *ca; | |
fa535a77 | 8874 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
8875 | |
8876 | if (unlikely(!cpuacct_subsys.active)) | |
8877 | return; | |
8878 | ||
8879 | rcu_read_lock(); | |
8880 | ca = task_ca(tsk); | |
8881 | ||
8882 | do { | |
fa535a77 | 8883 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
8884 | ca = ca->parent; |
8885 | } while (ca); | |
8886 | rcu_read_unlock(); | |
8887 | } | |
8888 | ||
d842de87 SV |
8889 | struct cgroup_subsys cpuacct_subsys = { |
8890 | .name = "cpuacct", | |
8891 | .create = cpuacct_create, | |
8892 | .destroy = cpuacct_destroy, | |
8893 | .populate = cpuacct_populate, | |
8894 | .subsys_id = cpuacct_subsys_id, | |
8895 | }; | |
8896 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
8897 | |
8898 | #ifndef CONFIG_SMP | |
8899 | ||
03b042bf PM |
8900 | void synchronize_sched_expedited(void) |
8901 | { | |
fc390cde | 8902 | barrier(); |
03b042bf PM |
8903 | } |
8904 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
8905 | ||
8906 | #else /* #ifndef CONFIG_SMP */ | |
8907 | ||
cc631fb7 | 8908 | static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0); |
03b042bf | 8909 | |
cc631fb7 | 8910 | static int synchronize_sched_expedited_cpu_stop(void *data) |
03b042bf | 8911 | { |
969c7921 TH |
8912 | /* |
8913 | * There must be a full memory barrier on each affected CPU | |
8914 | * between the time that try_stop_cpus() is called and the | |
8915 | * time that it returns. | |
8916 | * | |
8917 | * In the current initial implementation of cpu_stop, the | |
8918 | * above condition is already met when the control reaches | |
8919 | * this point and the following smp_mb() is not strictly | |
8920 | * necessary. Do smp_mb() anyway for documentation and | |
8921 | * robustness against future implementation changes. | |
8922 | */ | |
cc631fb7 | 8923 | smp_mb(); /* See above comment block. */ |
969c7921 | 8924 | return 0; |
03b042bf | 8925 | } |
03b042bf PM |
8926 | |
8927 | /* | |
8928 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
8929 | * approach to force grace period to end quickly. This consumes | |
8930 | * significant time on all CPUs, and is thus not recommended for | |
8931 | * any sort of common-case code. | |
8932 | * | |
8933 | * Note that it is illegal to call this function while holding any | |
8934 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
8935 | * observe this restriction will result in deadlock. | |
8936 | */ | |
8937 | void synchronize_sched_expedited(void) | |
8938 | { | |
969c7921 | 8939 | int snap, trycount = 0; |
03b042bf PM |
8940 | |
8941 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
969c7921 | 8942 | snap = atomic_read(&synchronize_sched_expedited_count) + 1; |
03b042bf | 8943 | get_online_cpus(); |
969c7921 TH |
8944 | while (try_stop_cpus(cpu_online_mask, |
8945 | synchronize_sched_expedited_cpu_stop, | |
94458d5e | 8946 | NULL) == -EAGAIN) { |
03b042bf PM |
8947 | put_online_cpus(); |
8948 | if (trycount++ < 10) | |
8949 | udelay(trycount * num_online_cpus()); | |
8950 | else { | |
8951 | synchronize_sched(); | |
8952 | return; | |
8953 | } | |
969c7921 | 8954 | if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) { |
03b042bf PM |
8955 | smp_mb(); /* ensure test happens before caller kfree */ |
8956 | return; | |
8957 | } | |
8958 | get_online_cpus(); | |
8959 | } | |
969c7921 | 8960 | atomic_inc(&synchronize_sched_expedited_count); |
cc631fb7 | 8961 | smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */ |
03b042bf | 8962 | put_online_cpus(); |
03b042bf PM |
8963 | } |
8964 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
8965 | ||
8966 | #endif /* #else #ifndef CONFIG_SMP */ |