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 | 79 | #include "sched_cpupri.h" |
21aa9af0 | 80 | #include "workqueue_sched.h" |
6e0534f2 | 81 | |
a8d154b0 | 82 | #define CREATE_TRACE_POINTS |
ad8d75ff | 83 | #include <trace/events/sched.h> |
a8d154b0 | 84 | |
1da177e4 LT |
85 | /* |
86 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
87 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
88 | * and back. | |
89 | */ | |
90 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
91 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
92 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
93 | ||
94 | /* | |
95 | * 'User priority' is the nice value converted to something we | |
96 | * can work with better when scaling various scheduler parameters, | |
97 | * it's a [ 0 ... 39 ] range. | |
98 | */ | |
99 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
100 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
101 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
102 | ||
103 | /* | |
d7876a08 | 104 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 105 | */ |
d6322faf | 106 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 107 | |
6aa645ea IM |
108 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
109 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
110 | ||
1da177e4 LT |
111 | /* |
112 | * These are the 'tuning knobs' of the scheduler: | |
113 | * | |
a4ec24b4 | 114 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
115 | * Timeslices get refilled after they expire. |
116 | */ | |
1da177e4 | 117 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 118 | |
d0b27fa7 PZ |
119 | /* |
120 | * single value that denotes runtime == period, ie unlimited time. | |
121 | */ | |
122 | #define RUNTIME_INF ((u64)~0ULL) | |
123 | ||
e05606d3 IM |
124 | static inline int rt_policy(int policy) |
125 | { | |
3f33a7ce | 126 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
127 | return 1; |
128 | return 0; | |
129 | } | |
130 | ||
131 | static inline int task_has_rt_policy(struct task_struct *p) | |
132 | { | |
133 | return rt_policy(p->policy); | |
134 | } | |
135 | ||
1da177e4 | 136 | /* |
6aa645ea | 137 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 138 | */ |
6aa645ea IM |
139 | struct rt_prio_array { |
140 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
141 | struct list_head queue[MAX_RT_PRIO]; | |
142 | }; | |
143 | ||
d0b27fa7 | 144 | struct rt_bandwidth { |
ea736ed5 | 145 | /* nests inside the rq lock: */ |
0986b11b | 146 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
147 | ktime_t rt_period; |
148 | u64 rt_runtime; | |
149 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
150 | }; |
151 | ||
152 | static struct rt_bandwidth def_rt_bandwidth; | |
153 | ||
154 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
155 | ||
156 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
157 | { | |
158 | struct rt_bandwidth *rt_b = | |
159 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
160 | ktime_t now; | |
161 | int overrun; | |
162 | int idle = 0; | |
163 | ||
164 | for (;;) { | |
165 | now = hrtimer_cb_get_time(timer); | |
166 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
167 | ||
168 | if (!overrun) | |
169 | break; | |
170 | ||
171 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
172 | } | |
173 | ||
174 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
175 | } | |
176 | ||
177 | static | |
178 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
179 | { | |
180 | rt_b->rt_period = ns_to_ktime(period); | |
181 | rt_b->rt_runtime = runtime; | |
182 | ||
0986b11b | 183 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 184 | |
d0b27fa7 PZ |
185 | hrtimer_init(&rt_b->rt_period_timer, |
186 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
187 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
188 | } |
189 | ||
c8bfff6d KH |
190 | static inline int rt_bandwidth_enabled(void) |
191 | { | |
192 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
193 | } |
194 | ||
195 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
196 | { | |
197 | ktime_t now; | |
198 | ||
cac64d00 | 199 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
200 | return; |
201 | ||
202 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
203 | return; | |
204 | ||
0986b11b | 205 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 206 | for (;;) { |
7f1e2ca9 PZ |
207 | unsigned long delta; |
208 | ktime_t soft, hard; | |
209 | ||
d0b27fa7 PZ |
210 | if (hrtimer_active(&rt_b->rt_period_timer)) |
211 | break; | |
212 | ||
213 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
214 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
215 | |
216 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
217 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
218 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
219 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 220 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 221 | } |
0986b11b | 222 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
223 | } |
224 | ||
225 | #ifdef CONFIG_RT_GROUP_SCHED | |
226 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
227 | { | |
228 | hrtimer_cancel(&rt_b->rt_period_timer); | |
229 | } | |
230 | #endif | |
231 | ||
712555ee HC |
232 | /* |
233 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
234 | * detach_destroy_domains and partition_sched_domains. | |
235 | */ | |
236 | static DEFINE_MUTEX(sched_domains_mutex); | |
237 | ||
7c941438 | 238 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 239 | |
68318b8e SV |
240 | #include <linux/cgroup.h> |
241 | ||
29f59db3 SV |
242 | struct cfs_rq; |
243 | ||
6f505b16 PZ |
244 | static LIST_HEAD(task_groups); |
245 | ||
29f59db3 | 246 | /* task group related information */ |
4cf86d77 | 247 | struct task_group { |
68318b8e | 248 | struct cgroup_subsys_state css; |
6c415b92 | 249 | |
052f1dc7 | 250 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
251 | /* schedulable entities of this group on each cpu */ |
252 | struct sched_entity **se; | |
253 | /* runqueue "owned" by this group on each cpu */ | |
254 | struct cfs_rq **cfs_rq; | |
255 | unsigned long shares; | |
052f1dc7 PZ |
256 | #endif |
257 | ||
258 | #ifdef CONFIG_RT_GROUP_SCHED | |
259 | struct sched_rt_entity **rt_se; | |
260 | struct rt_rq **rt_rq; | |
261 | ||
d0b27fa7 | 262 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 263 | #endif |
6b2d7700 | 264 | |
ae8393e5 | 265 | struct rcu_head rcu; |
6f505b16 | 266 | struct list_head list; |
f473aa5e PZ |
267 | |
268 | struct task_group *parent; | |
269 | struct list_head siblings; | |
270 | struct list_head children; | |
29f59db3 SV |
271 | }; |
272 | ||
eff766a6 | 273 | #define root_task_group init_task_group |
6f505b16 | 274 | |
8ed36996 | 275 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
276 | * a task group's cpu shares. |
277 | */ | |
8ed36996 | 278 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 279 | |
e9036b36 CG |
280 | #ifdef CONFIG_FAIR_GROUP_SCHED |
281 | ||
57310a98 PZ |
282 | #ifdef CONFIG_SMP |
283 | static int root_task_group_empty(void) | |
284 | { | |
285 | return list_empty(&root_task_group.children); | |
286 | } | |
287 | #endif | |
288 | ||
052f1dc7 | 289 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 290 | |
cb4ad1ff | 291 | /* |
2e084786 LJ |
292 | * A weight of 0 or 1 can cause arithmetics problems. |
293 | * A weight of a cfs_rq is the sum of weights of which entities | |
294 | * are queued on this cfs_rq, so a weight of a entity should not be | |
295 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
296 | * (The default weight is 1024 - so there's no practical |
297 | * limitation from this.) | |
298 | */ | |
18d95a28 | 299 | #define MIN_SHARES 2 |
2e084786 | 300 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 301 | |
052f1dc7 PZ |
302 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
303 | #endif | |
304 | ||
29f59db3 | 305 | /* Default task group. |
3a252015 | 306 | * Every task in system belong to this group at bootup. |
29f59db3 | 307 | */ |
434d53b0 | 308 | struct task_group init_task_group; |
29f59db3 | 309 | |
7c941438 | 310 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 311 | |
6aa645ea IM |
312 | /* CFS-related fields in a runqueue */ |
313 | struct cfs_rq { | |
314 | struct load_weight load; | |
315 | unsigned long nr_running; | |
316 | ||
6aa645ea | 317 | u64 exec_clock; |
e9acbff6 | 318 | u64 min_vruntime; |
6aa645ea IM |
319 | |
320 | struct rb_root tasks_timeline; | |
321 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
322 | |
323 | struct list_head tasks; | |
324 | struct list_head *balance_iterator; | |
325 | ||
326 | /* | |
327 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
328 | * It is set to NULL otherwise (i.e when none are currently running). |
329 | */ | |
4793241b | 330 | struct sched_entity *curr, *next, *last; |
ddc97297 | 331 | |
5ac5c4d6 | 332 | unsigned int nr_spread_over; |
ddc97297 | 333 | |
62160e3f | 334 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
335 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
336 | ||
41a2d6cf IM |
337 | /* |
338 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
339 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
340 | * (like users, containers etc.) | |
341 | * | |
342 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
343 | * list is used during load balance. | |
344 | */ | |
41a2d6cf IM |
345 | struct list_head leaf_cfs_rq_list; |
346 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
347 | |
348 | #ifdef CONFIG_SMP | |
c09595f6 | 349 | /* |
c8cba857 | 350 | * the part of load.weight contributed by tasks |
c09595f6 | 351 | */ |
c8cba857 | 352 | unsigned long task_weight; |
c09595f6 | 353 | |
c8cba857 PZ |
354 | /* |
355 | * h_load = weight * f(tg) | |
356 | * | |
357 | * Where f(tg) is the recursive weight fraction assigned to | |
358 | * this group. | |
359 | */ | |
360 | unsigned long h_load; | |
c09595f6 | 361 | |
c8cba857 PZ |
362 | /* |
363 | * this cpu's part of tg->shares | |
364 | */ | |
365 | unsigned long shares; | |
f1d239f7 PZ |
366 | |
367 | /* | |
368 | * load.weight at the time we set shares | |
369 | */ | |
370 | unsigned long rq_weight; | |
c09595f6 | 371 | #endif |
6aa645ea IM |
372 | #endif |
373 | }; | |
1da177e4 | 374 | |
6aa645ea IM |
375 | /* Real-Time classes' related field in a runqueue: */ |
376 | struct rt_rq { | |
377 | struct rt_prio_array active; | |
63489e45 | 378 | unsigned long rt_nr_running; |
052f1dc7 | 379 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
380 | struct { |
381 | int curr; /* highest queued rt task prio */ | |
398a153b | 382 | #ifdef CONFIG_SMP |
e864c499 | 383 | int next; /* next highest */ |
398a153b | 384 | #endif |
e864c499 | 385 | } highest_prio; |
6f505b16 | 386 | #endif |
fa85ae24 | 387 | #ifdef CONFIG_SMP |
73fe6aae | 388 | unsigned long rt_nr_migratory; |
a1ba4d8b | 389 | unsigned long rt_nr_total; |
a22d7fc1 | 390 | int overloaded; |
917b627d | 391 | struct plist_head pushable_tasks; |
fa85ae24 | 392 | #endif |
6f505b16 | 393 | int rt_throttled; |
fa85ae24 | 394 | u64 rt_time; |
ac086bc2 | 395 | u64 rt_runtime; |
ea736ed5 | 396 | /* Nests inside the rq lock: */ |
0986b11b | 397 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 398 | |
052f1dc7 | 399 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
400 | unsigned long rt_nr_boosted; |
401 | ||
6f505b16 PZ |
402 | struct rq *rq; |
403 | struct list_head leaf_rt_rq_list; | |
404 | struct task_group *tg; | |
6f505b16 | 405 | #endif |
6aa645ea IM |
406 | }; |
407 | ||
57d885fe GH |
408 | #ifdef CONFIG_SMP |
409 | ||
410 | /* | |
411 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
412 | * variables. Each exclusive cpuset essentially defines an island domain by |
413 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
414 | * exclusive cpuset is created, we also create and attach a new root-domain |
415 | * object. | |
416 | * | |
57d885fe GH |
417 | */ |
418 | struct root_domain { | |
419 | atomic_t refcount; | |
c6c4927b RR |
420 | cpumask_var_t span; |
421 | cpumask_var_t online; | |
637f5085 | 422 | |
0eab9146 | 423 | /* |
637f5085 GH |
424 | * The "RT overload" flag: it gets set if a CPU has more than |
425 | * one runnable RT task. | |
426 | */ | |
c6c4927b | 427 | cpumask_var_t rto_mask; |
0eab9146 | 428 | atomic_t rto_count; |
6e0534f2 GH |
429 | #ifdef CONFIG_SMP |
430 | struct cpupri cpupri; | |
431 | #endif | |
57d885fe GH |
432 | }; |
433 | ||
dc938520 GH |
434 | /* |
435 | * By default the system creates a single root-domain with all cpus as | |
436 | * members (mimicking the global state we have today). | |
437 | */ | |
57d885fe GH |
438 | static struct root_domain def_root_domain; |
439 | ||
440 | #endif | |
441 | ||
1da177e4 LT |
442 | /* |
443 | * This is the main, per-CPU runqueue data structure. | |
444 | * | |
445 | * Locking rule: those places that want to lock multiple runqueues | |
446 | * (such as the load balancing or the thread migration code), lock | |
447 | * acquire operations must be ordered by ascending &runqueue. | |
448 | */ | |
70b97a7f | 449 | struct rq { |
d8016491 | 450 | /* runqueue lock: */ |
05fa785c | 451 | raw_spinlock_t lock; |
1da177e4 LT |
452 | |
453 | /* | |
454 | * nr_running and cpu_load should be in the same cacheline because | |
455 | * remote CPUs use both these fields when doing load calculation. | |
456 | */ | |
457 | unsigned long nr_running; | |
6aa645ea IM |
458 | #define CPU_LOAD_IDX_MAX 5 |
459 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
46cb4b7c | 460 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 461 | u64 nohz_stamp; |
46cb4b7c SS |
462 | unsigned char in_nohz_recently; |
463 | #endif | |
a64692a3 MG |
464 | unsigned int skip_clock_update; |
465 | ||
d8016491 IM |
466 | /* capture load from *all* tasks on this cpu: */ |
467 | struct load_weight load; | |
6aa645ea IM |
468 | unsigned long nr_load_updates; |
469 | u64 nr_switches; | |
470 | ||
471 | struct cfs_rq cfs; | |
6f505b16 | 472 | struct rt_rq rt; |
6f505b16 | 473 | |
6aa645ea | 474 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
475 | /* list of leaf cfs_rq on this cpu: */ |
476 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
477 | #endif |
478 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 479 | struct list_head leaf_rt_rq_list; |
1da177e4 | 480 | #endif |
1da177e4 LT |
481 | |
482 | /* | |
483 | * This is part of a global counter where only the total sum | |
484 | * over all CPUs matters. A task can increase this counter on | |
485 | * one CPU and if it got migrated afterwards it may decrease | |
486 | * it on another CPU. Always updated under the runqueue lock: | |
487 | */ | |
488 | unsigned long nr_uninterruptible; | |
489 | ||
36c8b586 | 490 | struct task_struct *curr, *idle; |
c9819f45 | 491 | unsigned long next_balance; |
1da177e4 | 492 | struct mm_struct *prev_mm; |
6aa645ea | 493 | |
3e51f33f | 494 | u64 clock; |
6aa645ea | 495 | |
1da177e4 LT |
496 | atomic_t nr_iowait; |
497 | ||
498 | #ifdef CONFIG_SMP | |
0eab9146 | 499 | struct root_domain *rd; |
1da177e4 LT |
500 | struct sched_domain *sd; |
501 | ||
e51fd5e2 PZ |
502 | unsigned long cpu_power; |
503 | ||
a0a522ce | 504 | unsigned char idle_at_tick; |
1da177e4 | 505 | /* For active balancing */ |
3f029d3c | 506 | int post_schedule; |
1da177e4 LT |
507 | int active_balance; |
508 | int push_cpu; | |
969c7921 | 509 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
510 | /* cpu of this runqueue: */ |
511 | int cpu; | |
1f11eb6a | 512 | int online; |
1da177e4 | 513 | |
a8a51d5e | 514 | unsigned long avg_load_per_task; |
1da177e4 | 515 | |
e9e9250b PZ |
516 | u64 rt_avg; |
517 | u64 age_stamp; | |
1b9508f6 MG |
518 | u64 idle_stamp; |
519 | u64 avg_idle; | |
1da177e4 LT |
520 | #endif |
521 | ||
dce48a84 TG |
522 | /* calc_load related fields */ |
523 | unsigned long calc_load_update; | |
524 | long calc_load_active; | |
525 | ||
8f4d37ec | 526 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
527 | #ifdef CONFIG_SMP |
528 | int hrtick_csd_pending; | |
529 | struct call_single_data hrtick_csd; | |
530 | #endif | |
8f4d37ec PZ |
531 | struct hrtimer hrtick_timer; |
532 | #endif | |
533 | ||
1da177e4 LT |
534 | #ifdef CONFIG_SCHEDSTATS |
535 | /* latency stats */ | |
536 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
537 | unsigned long long rq_cpu_time; |
538 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
539 | |
540 | /* sys_sched_yield() stats */ | |
480b9434 | 541 | unsigned int yld_count; |
1da177e4 LT |
542 | |
543 | /* schedule() stats */ | |
480b9434 KC |
544 | unsigned int sched_switch; |
545 | unsigned int sched_count; | |
546 | unsigned int sched_goidle; | |
1da177e4 LT |
547 | |
548 | /* try_to_wake_up() stats */ | |
480b9434 KC |
549 | unsigned int ttwu_count; |
550 | unsigned int ttwu_local; | |
b8efb561 IM |
551 | |
552 | /* BKL stats */ | |
480b9434 | 553 | unsigned int bkl_count; |
1da177e4 LT |
554 | #endif |
555 | }; | |
556 | ||
f34e3b61 | 557 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 558 | |
7d478721 PZ |
559 | static inline |
560 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 561 | { |
7d478721 | 562 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
a64692a3 MG |
563 | |
564 | /* | |
565 | * A queue event has occurred, and we're going to schedule. In | |
566 | * this case, we can save a useless back to back clock update. | |
567 | */ | |
568 | if (test_tsk_need_resched(p)) | |
569 | rq->skip_clock_update = 1; | |
dd41f596 IM |
570 | } |
571 | ||
0a2966b4 CL |
572 | static inline int cpu_of(struct rq *rq) |
573 | { | |
574 | #ifdef CONFIG_SMP | |
575 | return rq->cpu; | |
576 | #else | |
577 | return 0; | |
578 | #endif | |
579 | } | |
580 | ||
497f0ab3 | 581 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
582 | rcu_dereference_check((p), \ |
583 | rcu_read_lock_sched_held() || \ | |
584 | lockdep_is_held(&sched_domains_mutex)) | |
585 | ||
674311d5 NP |
586 | /* |
587 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 588 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
589 | * |
590 | * The domain tree of any CPU may only be accessed from within | |
591 | * preempt-disabled sections. | |
592 | */ | |
48f24c4d | 593 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 594 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
595 | |
596 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
597 | #define this_rq() (&__get_cpu_var(runqueues)) | |
598 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
599 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 600 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 601 | |
dc61b1d6 PZ |
602 | #ifdef CONFIG_CGROUP_SCHED |
603 | ||
604 | /* | |
605 | * Return the group to which this tasks belongs. | |
606 | * | |
607 | * We use task_subsys_state_check() and extend the RCU verification | |
608 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
609 | * holds that lock for each task it moves into the cgroup. Therefore | |
610 | * by holding that lock, we pin the task to the current cgroup. | |
611 | */ | |
612 | static inline struct task_group *task_group(struct task_struct *p) | |
613 | { | |
614 | struct cgroup_subsys_state *css; | |
615 | ||
616 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
617 | lockdep_is_held(&task_rq(p)->lock)); | |
618 | return container_of(css, struct task_group, css); | |
619 | } | |
620 | ||
621 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
622 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
623 | { | |
624 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
625 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
626 | p->se.parent = task_group(p)->se[cpu]; | |
627 | #endif | |
628 | ||
629 | #ifdef CONFIG_RT_GROUP_SCHED | |
630 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
631 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
632 | #endif | |
633 | } | |
634 | ||
635 | #else /* CONFIG_CGROUP_SCHED */ | |
636 | ||
637 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
638 | static inline struct task_group *task_group(struct task_struct *p) | |
639 | { | |
640 | return NULL; | |
641 | } | |
642 | ||
643 | #endif /* CONFIG_CGROUP_SCHED */ | |
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 | ||
a9957449 | 972 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
973 | __releases(rq->lock) |
974 | { | |
05fa785c | 975 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
976 | } |
977 | ||
70b97a7f | 978 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
979 | __releases(rq->lock) |
980 | { | |
05fa785c | 981 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
982 | } |
983 | ||
1da177e4 | 984 | /* |
cc2a73b5 | 985 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 986 | */ |
a9957449 | 987 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
988 | __acquires(rq->lock) |
989 | { | |
70b97a7f | 990 | struct rq *rq; |
1da177e4 LT |
991 | |
992 | local_irq_disable(); | |
993 | rq = this_rq(); | |
05fa785c | 994 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
995 | |
996 | return rq; | |
997 | } | |
998 | ||
8f4d37ec PZ |
999 | #ifdef CONFIG_SCHED_HRTICK |
1000 | /* | |
1001 | * Use HR-timers to deliver accurate preemption points. | |
1002 | * | |
1003 | * Its all a bit involved since we cannot program an hrt while holding the | |
1004 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1005 | * reschedule event. | |
1006 | * | |
1007 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1008 | * rq->lock. | |
1009 | */ | |
8f4d37ec PZ |
1010 | |
1011 | /* | |
1012 | * Use hrtick when: | |
1013 | * - enabled by features | |
1014 | * - hrtimer is actually high res | |
1015 | */ | |
1016 | static inline int hrtick_enabled(struct rq *rq) | |
1017 | { | |
1018 | if (!sched_feat(HRTICK)) | |
1019 | return 0; | |
ba42059f | 1020 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1021 | return 0; |
8f4d37ec PZ |
1022 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1023 | } | |
1024 | ||
8f4d37ec PZ |
1025 | static void hrtick_clear(struct rq *rq) |
1026 | { | |
1027 | if (hrtimer_active(&rq->hrtick_timer)) | |
1028 | hrtimer_cancel(&rq->hrtick_timer); | |
1029 | } | |
1030 | ||
8f4d37ec PZ |
1031 | /* |
1032 | * High-resolution timer tick. | |
1033 | * Runs from hardirq context with interrupts disabled. | |
1034 | */ | |
1035 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1036 | { | |
1037 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1038 | ||
1039 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1040 | ||
05fa785c | 1041 | raw_spin_lock(&rq->lock); |
3e51f33f | 1042 | update_rq_clock(rq); |
8f4d37ec | 1043 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1044 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1045 | |
1046 | return HRTIMER_NORESTART; | |
1047 | } | |
1048 | ||
95e904c7 | 1049 | #ifdef CONFIG_SMP |
31656519 PZ |
1050 | /* |
1051 | * called from hardirq (IPI) context | |
1052 | */ | |
1053 | static void __hrtick_start(void *arg) | |
b328ca18 | 1054 | { |
31656519 | 1055 | struct rq *rq = arg; |
b328ca18 | 1056 | |
05fa785c | 1057 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1058 | hrtimer_restart(&rq->hrtick_timer); |
1059 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1060 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1061 | } |
1062 | ||
31656519 PZ |
1063 | /* |
1064 | * Called to set the hrtick timer state. | |
1065 | * | |
1066 | * called with rq->lock held and irqs disabled | |
1067 | */ | |
1068 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1069 | { |
31656519 PZ |
1070 | struct hrtimer *timer = &rq->hrtick_timer; |
1071 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1072 | |
cc584b21 | 1073 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1074 | |
1075 | if (rq == this_rq()) { | |
1076 | hrtimer_restart(timer); | |
1077 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1078 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1079 | rq->hrtick_csd_pending = 1; |
1080 | } | |
b328ca18 PZ |
1081 | } |
1082 | ||
1083 | static int | |
1084 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1085 | { | |
1086 | int cpu = (int)(long)hcpu; | |
1087 | ||
1088 | switch (action) { | |
1089 | case CPU_UP_CANCELED: | |
1090 | case CPU_UP_CANCELED_FROZEN: | |
1091 | case CPU_DOWN_PREPARE: | |
1092 | case CPU_DOWN_PREPARE_FROZEN: | |
1093 | case CPU_DEAD: | |
1094 | case CPU_DEAD_FROZEN: | |
31656519 | 1095 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1096 | return NOTIFY_OK; |
1097 | } | |
1098 | ||
1099 | return NOTIFY_DONE; | |
1100 | } | |
1101 | ||
fa748203 | 1102 | static __init void init_hrtick(void) |
b328ca18 PZ |
1103 | { |
1104 | hotcpu_notifier(hotplug_hrtick, 0); | |
1105 | } | |
31656519 PZ |
1106 | #else |
1107 | /* | |
1108 | * Called to set the hrtick timer state. | |
1109 | * | |
1110 | * called with rq->lock held and irqs disabled | |
1111 | */ | |
1112 | static void hrtick_start(struct rq *rq, u64 delay) | |
1113 | { | |
7f1e2ca9 | 1114 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1115 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1116 | } |
b328ca18 | 1117 | |
006c75f1 | 1118 | static inline void init_hrtick(void) |
8f4d37ec | 1119 | { |
8f4d37ec | 1120 | } |
31656519 | 1121 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1122 | |
31656519 | 1123 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1124 | { |
31656519 PZ |
1125 | #ifdef CONFIG_SMP |
1126 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1127 | |
31656519 PZ |
1128 | rq->hrtick_csd.flags = 0; |
1129 | rq->hrtick_csd.func = __hrtick_start; | |
1130 | rq->hrtick_csd.info = rq; | |
1131 | #endif | |
8f4d37ec | 1132 | |
31656519 PZ |
1133 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1134 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1135 | } |
006c75f1 | 1136 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1137 | static inline void hrtick_clear(struct rq *rq) |
1138 | { | |
1139 | } | |
1140 | ||
8f4d37ec PZ |
1141 | static inline void init_rq_hrtick(struct rq *rq) |
1142 | { | |
1143 | } | |
1144 | ||
b328ca18 PZ |
1145 | static inline void init_hrtick(void) |
1146 | { | |
1147 | } | |
006c75f1 | 1148 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1149 | |
c24d20db IM |
1150 | /* |
1151 | * resched_task - mark a task 'to be rescheduled now'. | |
1152 | * | |
1153 | * On UP this means the setting of the need_resched flag, on SMP it | |
1154 | * might also involve a cross-CPU call to trigger the scheduler on | |
1155 | * the target CPU. | |
1156 | */ | |
1157 | #ifdef CONFIG_SMP | |
1158 | ||
1159 | #ifndef tsk_is_polling | |
1160 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1161 | #endif | |
1162 | ||
31656519 | 1163 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1164 | { |
1165 | int cpu; | |
1166 | ||
05fa785c | 1167 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1168 | |
5ed0cec0 | 1169 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1170 | return; |
1171 | ||
5ed0cec0 | 1172 | set_tsk_need_resched(p); |
c24d20db IM |
1173 | |
1174 | cpu = task_cpu(p); | |
1175 | if (cpu == smp_processor_id()) | |
1176 | return; | |
1177 | ||
1178 | /* NEED_RESCHED must be visible before we test polling */ | |
1179 | smp_mb(); | |
1180 | if (!tsk_is_polling(p)) | |
1181 | smp_send_reschedule(cpu); | |
1182 | } | |
1183 | ||
1184 | static void resched_cpu(int cpu) | |
1185 | { | |
1186 | struct rq *rq = cpu_rq(cpu); | |
1187 | unsigned long flags; | |
1188 | ||
05fa785c | 1189 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1190 | return; |
1191 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1192 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1193 | } |
06d8308c TG |
1194 | |
1195 | #ifdef CONFIG_NO_HZ | |
1196 | /* | |
1197 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1198 | * idle CPU then this timer might expire before the next timer event | |
1199 | * which is scheduled to wake up that CPU. In case of a completely | |
1200 | * idle system the next event might even be infinite time into the | |
1201 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1202 | * leaves the inner idle loop so the newly added timer is taken into | |
1203 | * account when the CPU goes back to idle and evaluates the timer | |
1204 | * wheel for the next timer event. | |
1205 | */ | |
1206 | void wake_up_idle_cpu(int cpu) | |
1207 | { | |
1208 | struct rq *rq = cpu_rq(cpu); | |
1209 | ||
1210 | if (cpu == smp_processor_id()) | |
1211 | return; | |
1212 | ||
1213 | /* | |
1214 | * This is safe, as this function is called with the timer | |
1215 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1216 | * to idle and has not yet set rq->curr to idle then it will | |
1217 | * be serialized on the timer wheel base lock and take the new | |
1218 | * timer into account automatically. | |
1219 | */ | |
1220 | if (rq->curr != rq->idle) | |
1221 | return; | |
1222 | ||
1223 | /* | |
1224 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1225 | * lockless. The worst case is that the other CPU runs the | |
1226 | * idle task through an additional NOOP schedule() | |
1227 | */ | |
5ed0cec0 | 1228 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1229 | |
1230 | /* NEED_RESCHED must be visible before we test polling */ | |
1231 | smp_mb(); | |
1232 | if (!tsk_is_polling(rq->idle)) | |
1233 | smp_send_reschedule(cpu); | |
1234 | } | |
39c0cbe2 MG |
1235 | |
1236 | int nohz_ratelimit(int cpu) | |
1237 | { | |
1238 | struct rq *rq = cpu_rq(cpu); | |
1239 | u64 diff = rq->clock - rq->nohz_stamp; | |
1240 | ||
1241 | rq->nohz_stamp = rq->clock; | |
1242 | ||
1243 | return diff < (NSEC_PER_SEC / HZ) >> 1; | |
1244 | } | |
1245 | ||
6d6bc0ad | 1246 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1247 | |
e9e9250b PZ |
1248 | static u64 sched_avg_period(void) |
1249 | { | |
1250 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1251 | } | |
1252 | ||
1253 | static void sched_avg_update(struct rq *rq) | |
1254 | { | |
1255 | s64 period = sched_avg_period(); | |
1256 | ||
1257 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
1258 | rq->age_stamp += period; | |
1259 | rq->rt_avg /= 2; | |
1260 | } | |
1261 | } | |
1262 | ||
1263 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1264 | { | |
1265 | rq->rt_avg += rt_delta; | |
1266 | sched_avg_update(rq); | |
1267 | } | |
1268 | ||
6d6bc0ad | 1269 | #else /* !CONFIG_SMP */ |
31656519 | 1270 | static void resched_task(struct task_struct *p) |
c24d20db | 1271 | { |
05fa785c | 1272 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1273 | set_tsk_need_resched(p); |
c24d20db | 1274 | } |
e9e9250b PZ |
1275 | |
1276 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1277 | { | |
1278 | } | |
6d6bc0ad | 1279 | #endif /* CONFIG_SMP */ |
c24d20db | 1280 | |
45bf76df IM |
1281 | #if BITS_PER_LONG == 32 |
1282 | # define WMULT_CONST (~0UL) | |
1283 | #else | |
1284 | # define WMULT_CONST (1UL << 32) | |
1285 | #endif | |
1286 | ||
1287 | #define WMULT_SHIFT 32 | |
1288 | ||
194081eb IM |
1289 | /* |
1290 | * Shift right and round: | |
1291 | */ | |
cf2ab469 | 1292 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1293 | |
a7be37ac PZ |
1294 | /* |
1295 | * delta *= weight / lw | |
1296 | */ | |
cb1c4fc9 | 1297 | static unsigned long |
45bf76df IM |
1298 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1299 | struct load_weight *lw) | |
1300 | { | |
1301 | u64 tmp; | |
1302 | ||
7a232e03 LJ |
1303 | if (!lw->inv_weight) { |
1304 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1305 | lw->inv_weight = 1; | |
1306 | else | |
1307 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1308 | / (lw->weight+1); | |
1309 | } | |
45bf76df IM |
1310 | |
1311 | tmp = (u64)delta_exec * weight; | |
1312 | /* | |
1313 | * Check whether we'd overflow the 64-bit multiplication: | |
1314 | */ | |
194081eb | 1315 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1316 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1317 | WMULT_SHIFT/2); |
1318 | else | |
cf2ab469 | 1319 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1320 | |
ecf691da | 1321 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1322 | } |
1323 | ||
1091985b | 1324 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1325 | { |
1326 | lw->weight += inc; | |
e89996ae | 1327 | lw->inv_weight = 0; |
45bf76df IM |
1328 | } |
1329 | ||
1091985b | 1330 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1331 | { |
1332 | lw->weight -= dec; | |
e89996ae | 1333 | lw->inv_weight = 0; |
45bf76df IM |
1334 | } |
1335 | ||
2dd73a4f PW |
1336 | /* |
1337 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1338 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1339 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1340 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1341 | * scaled version of the new time slice allocation that they receive on time |
1342 | * slice expiry etc. | |
1343 | */ | |
1344 | ||
cce7ade8 PZ |
1345 | #define WEIGHT_IDLEPRIO 3 |
1346 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1347 | |
1348 | /* | |
1349 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1350 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1351 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1352 | * that remained on nice 0. | |
1353 | * | |
1354 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1355 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1356 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1357 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1358 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1359 | */ |
1360 | static const int prio_to_weight[40] = { | |
254753dc IM |
1361 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1362 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1363 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1364 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1365 | /* 0 */ 1024, 820, 655, 526, 423, | |
1366 | /* 5 */ 335, 272, 215, 172, 137, | |
1367 | /* 10 */ 110, 87, 70, 56, 45, | |
1368 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1369 | }; |
1370 | ||
5714d2de IM |
1371 | /* |
1372 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1373 | * | |
1374 | * In cases where the weight does not change often, we can use the | |
1375 | * precalculated inverse to speed up arithmetics by turning divisions | |
1376 | * into multiplications: | |
1377 | */ | |
dd41f596 | 1378 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1379 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1380 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1381 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1382 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1383 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1384 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1385 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1386 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1387 | }; |
2dd73a4f | 1388 | |
ef12fefa BR |
1389 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1390 | enum cpuacct_stat_index { | |
1391 | CPUACCT_STAT_USER, /* ... user mode */ | |
1392 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1393 | ||
1394 | CPUACCT_STAT_NSTATS, | |
1395 | }; | |
1396 | ||
d842de87 SV |
1397 | #ifdef CONFIG_CGROUP_CPUACCT |
1398 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1399 | static void cpuacct_update_stats(struct task_struct *tsk, |
1400 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1401 | #else |
1402 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1403 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1404 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1405 | #endif |
1406 | ||
18d95a28 PZ |
1407 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1408 | { | |
1409 | update_load_add(&rq->load, load); | |
1410 | } | |
1411 | ||
1412 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1413 | { | |
1414 | update_load_sub(&rq->load, load); | |
1415 | } | |
1416 | ||
7940ca36 | 1417 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1418 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1419 | |
1420 | /* | |
1421 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1422 | * leaving it for the final time. | |
1423 | */ | |
eb755805 | 1424 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1425 | { |
1426 | struct task_group *parent, *child; | |
eb755805 | 1427 | int ret; |
c09595f6 PZ |
1428 | |
1429 | rcu_read_lock(); | |
1430 | parent = &root_task_group; | |
1431 | down: | |
eb755805 PZ |
1432 | ret = (*down)(parent, data); |
1433 | if (ret) | |
1434 | goto out_unlock; | |
c09595f6 PZ |
1435 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1436 | parent = child; | |
1437 | goto down; | |
1438 | ||
1439 | up: | |
1440 | continue; | |
1441 | } | |
eb755805 PZ |
1442 | ret = (*up)(parent, data); |
1443 | if (ret) | |
1444 | goto out_unlock; | |
c09595f6 PZ |
1445 | |
1446 | child = parent; | |
1447 | parent = parent->parent; | |
1448 | if (parent) | |
1449 | goto up; | |
eb755805 | 1450 | out_unlock: |
c09595f6 | 1451 | rcu_read_unlock(); |
eb755805 PZ |
1452 | |
1453 | return ret; | |
c09595f6 PZ |
1454 | } |
1455 | ||
eb755805 PZ |
1456 | static int tg_nop(struct task_group *tg, void *data) |
1457 | { | |
1458 | return 0; | |
c09595f6 | 1459 | } |
eb755805 PZ |
1460 | #endif |
1461 | ||
1462 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1463 | /* Used instead of source_load when we know the type == 0 */ |
1464 | static unsigned long weighted_cpuload(const int cpu) | |
1465 | { | |
1466 | return cpu_rq(cpu)->load.weight; | |
1467 | } | |
1468 | ||
1469 | /* | |
1470 | * Return a low guess at the load of a migration-source cpu weighted | |
1471 | * according to the scheduling class and "nice" value. | |
1472 | * | |
1473 | * We want to under-estimate the load of migration sources, to | |
1474 | * balance conservatively. | |
1475 | */ | |
1476 | static unsigned long source_load(int cpu, int type) | |
1477 | { | |
1478 | struct rq *rq = cpu_rq(cpu); | |
1479 | unsigned long total = weighted_cpuload(cpu); | |
1480 | ||
1481 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1482 | return total; | |
1483 | ||
1484 | return min(rq->cpu_load[type-1], total); | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Return a high guess at the load of a migration-target cpu weighted | |
1489 | * according to the scheduling class and "nice" value. | |
1490 | */ | |
1491 | static unsigned long target_load(int cpu, int type) | |
1492 | { | |
1493 | struct rq *rq = cpu_rq(cpu); | |
1494 | unsigned long total = weighted_cpuload(cpu); | |
1495 | ||
1496 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1497 | return total; | |
1498 | ||
1499 | return max(rq->cpu_load[type-1], total); | |
1500 | } | |
1501 | ||
ae154be1 PZ |
1502 | static unsigned long power_of(int cpu) |
1503 | { | |
e51fd5e2 | 1504 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1505 | } |
1506 | ||
eb755805 PZ |
1507 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1508 | ||
1509 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1510 | { | |
1511 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1512 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1513 | |
4cd42620 SR |
1514 | if (nr_running) |
1515 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1516 | else |
1517 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1518 | |
1519 | return rq->avg_load_per_task; | |
1520 | } | |
1521 | ||
1522 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1523 | |
43cf38eb | 1524 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1525 | |
c09595f6 PZ |
1526 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1527 | ||
1528 | /* | |
1529 | * Calculate and set the cpu's group shares. | |
1530 | */ | |
34d76c41 PZ |
1531 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1532 | unsigned long sd_shares, | |
1533 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1534 | unsigned long *usd_rq_weight) |
18d95a28 | 1535 | { |
34d76c41 | 1536 | unsigned long shares, rq_weight; |
a5004278 | 1537 | int boost = 0; |
c09595f6 | 1538 | |
4a6cc4bd | 1539 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1540 | if (!rq_weight) { |
1541 | boost = 1; | |
1542 | rq_weight = NICE_0_LOAD; | |
1543 | } | |
c8cba857 | 1544 | |
c09595f6 | 1545 | /* |
a8af7246 PZ |
1546 | * \Sum_j shares_j * rq_weight_i |
1547 | * shares_i = ----------------------------- | |
1548 | * \Sum_j rq_weight_j | |
c09595f6 | 1549 | */ |
ec4e0e2f | 1550 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1551 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1552 | |
ffda12a1 PZ |
1553 | if (abs(shares - tg->se[cpu]->load.weight) > |
1554 | sysctl_sched_shares_thresh) { | |
1555 | struct rq *rq = cpu_rq(cpu); | |
1556 | unsigned long flags; | |
c09595f6 | 1557 | |
05fa785c | 1558 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1559 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1560 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1561 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1562 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1563 | } |
18d95a28 | 1564 | } |
c09595f6 PZ |
1565 | |
1566 | /* | |
c8cba857 PZ |
1567 | * Re-compute the task group their per cpu shares over the given domain. |
1568 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1569 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1570 | */ |
eb755805 | 1571 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1572 | { |
cd8ad40d | 1573 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1574 | unsigned long *usd_rq_weight; |
eb755805 | 1575 | struct sched_domain *sd = data; |
34d76c41 | 1576 | unsigned long flags; |
c8cba857 | 1577 | int i; |
c09595f6 | 1578 | |
34d76c41 PZ |
1579 | if (!tg->se[0]) |
1580 | return 0; | |
1581 | ||
1582 | local_irq_save(flags); | |
4a6cc4bd | 1583 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1584 | |
758b2cdc | 1585 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1586 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1587 | usd_rq_weight[i] = weight; |
34d76c41 | 1588 | |
cd8ad40d | 1589 | rq_weight += weight; |
ec4e0e2f KC |
1590 | /* |
1591 | * If there are currently no tasks on the cpu pretend there | |
1592 | * is one of average load so that when a new task gets to | |
1593 | * run here it will not get delayed by group starvation. | |
1594 | */ | |
ec4e0e2f KC |
1595 | if (!weight) |
1596 | weight = NICE_0_LOAD; | |
1597 | ||
cd8ad40d | 1598 | sum_weight += weight; |
c8cba857 | 1599 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1600 | } |
c09595f6 | 1601 | |
cd8ad40d PZ |
1602 | if (!rq_weight) |
1603 | rq_weight = sum_weight; | |
1604 | ||
c8cba857 PZ |
1605 | if ((!shares && rq_weight) || shares > tg->shares) |
1606 | shares = tg->shares; | |
1607 | ||
1608 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1609 | shares = tg->shares; | |
c09595f6 | 1610 | |
758b2cdc | 1611 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1612 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1613 | |
1614 | local_irq_restore(flags); | |
eb755805 PZ |
1615 | |
1616 | return 0; | |
c09595f6 PZ |
1617 | } |
1618 | ||
1619 | /* | |
c8cba857 PZ |
1620 | * Compute the cpu's hierarchical load factor for each task group. |
1621 | * This needs to be done in a top-down fashion because the load of a child | |
1622 | * group is a fraction of its parents load. | |
c09595f6 | 1623 | */ |
eb755805 | 1624 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1625 | { |
c8cba857 | 1626 | unsigned long load; |
eb755805 | 1627 | long cpu = (long)data; |
c09595f6 | 1628 | |
c8cba857 PZ |
1629 | if (!tg->parent) { |
1630 | load = cpu_rq(cpu)->load.weight; | |
1631 | } else { | |
1632 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1633 | load *= tg->cfs_rq[cpu]->shares; | |
1634 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1635 | } | |
c09595f6 | 1636 | |
c8cba857 | 1637 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1638 | |
eb755805 | 1639 | return 0; |
c09595f6 PZ |
1640 | } |
1641 | ||
c8cba857 | 1642 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1643 | { |
e7097159 PZ |
1644 | s64 elapsed; |
1645 | u64 now; | |
1646 | ||
1647 | if (root_task_group_empty()) | |
1648 | return; | |
1649 | ||
c676329a | 1650 | now = local_clock(); |
e7097159 | 1651 | elapsed = now - sd->last_update; |
2398f2c6 PZ |
1652 | |
1653 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1654 | sd->last_update = now; | |
eb755805 | 1655 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1656 | } |
4d8d595d PZ |
1657 | } |
1658 | ||
eb755805 | 1659 | static void update_h_load(long cpu) |
c09595f6 | 1660 | { |
e7097159 PZ |
1661 | if (root_task_group_empty()) |
1662 | return; | |
1663 | ||
eb755805 | 1664 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1665 | } |
1666 | ||
c09595f6 PZ |
1667 | #else |
1668 | ||
c8cba857 | 1669 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1670 | { |
1671 | } | |
1672 | ||
18d95a28 PZ |
1673 | #endif |
1674 | ||
8f45e2b5 GH |
1675 | #ifdef CONFIG_PREEMPT |
1676 | ||
b78bb868 PZ |
1677 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1678 | ||
70574a99 | 1679 | /* |
8f45e2b5 GH |
1680 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1681 | * way at the expense of forcing extra atomic operations in all | |
1682 | * invocations. This assures that the double_lock is acquired using the | |
1683 | * same underlying policy as the spinlock_t on this architecture, which | |
1684 | * reduces latency compared to the unfair variant below. However, it | |
1685 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1686 | */ |
8f45e2b5 GH |
1687 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1688 | __releases(this_rq->lock) | |
1689 | __acquires(busiest->lock) | |
1690 | __acquires(this_rq->lock) | |
1691 | { | |
05fa785c | 1692 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1693 | double_rq_lock(this_rq, busiest); |
1694 | ||
1695 | return 1; | |
1696 | } | |
1697 | ||
1698 | #else | |
1699 | /* | |
1700 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1701 | * latency by eliminating extra atomic operations when the locks are | |
1702 | * already in proper order on entry. This favors lower cpu-ids and will | |
1703 | * grant the double lock to lower cpus over higher ids under contention, | |
1704 | * regardless of entry order into the function. | |
1705 | */ | |
1706 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1707 | __releases(this_rq->lock) |
1708 | __acquires(busiest->lock) | |
1709 | __acquires(this_rq->lock) | |
1710 | { | |
1711 | int ret = 0; | |
1712 | ||
05fa785c | 1713 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1714 | if (busiest < this_rq) { |
05fa785c TG |
1715 | raw_spin_unlock(&this_rq->lock); |
1716 | raw_spin_lock(&busiest->lock); | |
1717 | raw_spin_lock_nested(&this_rq->lock, | |
1718 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1719 | ret = 1; |
1720 | } else | |
05fa785c TG |
1721 | raw_spin_lock_nested(&busiest->lock, |
1722 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1723 | } |
1724 | return ret; | |
1725 | } | |
1726 | ||
8f45e2b5 GH |
1727 | #endif /* CONFIG_PREEMPT */ |
1728 | ||
1729 | /* | |
1730 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1731 | */ | |
1732 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1733 | { | |
1734 | if (unlikely(!irqs_disabled())) { | |
1735 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1736 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1737 | BUG_ON(1); |
1738 | } | |
1739 | ||
1740 | return _double_lock_balance(this_rq, busiest); | |
1741 | } | |
1742 | ||
70574a99 AD |
1743 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1744 | __releases(busiest->lock) | |
1745 | { | |
05fa785c | 1746 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1747 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1748 | } | |
1e3c88bd PZ |
1749 | |
1750 | /* | |
1751 | * double_rq_lock - safely lock two runqueues | |
1752 | * | |
1753 | * Note this does not disable interrupts like task_rq_lock, | |
1754 | * you need to do so manually before calling. | |
1755 | */ | |
1756 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1757 | __acquires(rq1->lock) | |
1758 | __acquires(rq2->lock) | |
1759 | { | |
1760 | BUG_ON(!irqs_disabled()); | |
1761 | if (rq1 == rq2) { | |
1762 | raw_spin_lock(&rq1->lock); | |
1763 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1764 | } else { | |
1765 | if (rq1 < rq2) { | |
1766 | raw_spin_lock(&rq1->lock); | |
1767 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1768 | } else { | |
1769 | raw_spin_lock(&rq2->lock); | |
1770 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1771 | } | |
1772 | } | |
1e3c88bd PZ |
1773 | } |
1774 | ||
1775 | /* | |
1776 | * double_rq_unlock - safely unlock two runqueues | |
1777 | * | |
1778 | * Note this does not restore interrupts like task_rq_unlock, | |
1779 | * you need to do so manually after calling. | |
1780 | */ | |
1781 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1782 | __releases(rq1->lock) | |
1783 | __releases(rq2->lock) | |
1784 | { | |
1785 | raw_spin_unlock(&rq1->lock); | |
1786 | if (rq1 != rq2) | |
1787 | raw_spin_unlock(&rq2->lock); | |
1788 | else | |
1789 | __release(rq2->lock); | |
1790 | } | |
1791 | ||
18d95a28 PZ |
1792 | #endif |
1793 | ||
30432094 | 1794 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1795 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1796 | { | |
30432094 | 1797 | #ifdef CONFIG_SMP |
34e83e85 IM |
1798 | cfs_rq->shares = shares; |
1799 | #endif | |
1800 | } | |
30432094 | 1801 | #endif |
e7693a36 | 1802 | |
74f5187a | 1803 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1804 | static void update_sysctl(void); |
acb4a848 | 1805 | static int get_update_sysctl_factor(void); |
dce48a84 | 1806 | |
cd29fe6f PZ |
1807 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1808 | { | |
1809 | set_task_rq(p, cpu); | |
1810 | #ifdef CONFIG_SMP | |
1811 | /* | |
1812 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1813 | * successfuly executed on another CPU. We must ensure that updates of | |
1814 | * per-task data have been completed by this moment. | |
1815 | */ | |
1816 | smp_wmb(); | |
1817 | task_thread_info(p)->cpu = cpu; | |
1818 | #endif | |
1819 | } | |
dce48a84 | 1820 | |
1e3c88bd | 1821 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1822 | |
1823 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1824 | #define for_each_class(class) \ |
1825 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1826 | |
1e3c88bd PZ |
1827 | #include "sched_stats.h" |
1828 | ||
c09595f6 | 1829 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1830 | { |
1831 | rq->nr_running++; | |
9c217245 IM |
1832 | } |
1833 | ||
c09595f6 | 1834 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1835 | { |
1836 | rq->nr_running--; | |
9c217245 IM |
1837 | } |
1838 | ||
45bf76df IM |
1839 | static void set_load_weight(struct task_struct *p) |
1840 | { | |
1841 | if (task_has_rt_policy(p)) { | |
e51fd5e2 PZ |
1842 | p->se.load.weight = 0; |
1843 | p->se.load.inv_weight = WMULT_CONST; | |
dd41f596 IM |
1844 | return; |
1845 | } | |
45bf76df | 1846 | |
dd41f596 IM |
1847 | /* |
1848 | * SCHED_IDLE tasks get minimal weight: | |
1849 | */ | |
1850 | if (p->policy == SCHED_IDLE) { | |
1851 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1852 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1853 | return; | |
1854 | } | |
71f8bd46 | 1855 | |
dd41f596 IM |
1856 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1857 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1858 | } |
1859 | ||
371fd7e7 | 1860 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1861 | { |
a64692a3 | 1862 | update_rq_clock(rq); |
dd41f596 | 1863 | sched_info_queued(p); |
371fd7e7 | 1864 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1865 | p->se.on_rq = 1; |
71f8bd46 IM |
1866 | } |
1867 | ||
371fd7e7 | 1868 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1869 | { |
a64692a3 | 1870 | update_rq_clock(rq); |
46ac22ba | 1871 | sched_info_dequeued(p); |
371fd7e7 | 1872 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1873 | p->se.on_rq = 0; |
71f8bd46 IM |
1874 | } |
1875 | ||
1e3c88bd PZ |
1876 | /* |
1877 | * activate_task - move a task to the runqueue. | |
1878 | */ | |
371fd7e7 | 1879 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1880 | { |
1881 | if (task_contributes_to_load(p)) | |
1882 | rq->nr_uninterruptible--; | |
1883 | ||
371fd7e7 | 1884 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1885 | inc_nr_running(rq); |
1886 | } | |
1887 | ||
1888 | /* | |
1889 | * deactivate_task - remove a task from the runqueue. | |
1890 | */ | |
371fd7e7 | 1891 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1892 | { |
1893 | if (task_contributes_to_load(p)) | |
1894 | rq->nr_uninterruptible++; | |
1895 | ||
371fd7e7 | 1896 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1897 | dec_nr_running(rq); |
1898 | } | |
1899 | ||
1900 | #include "sched_idletask.c" | |
1901 | #include "sched_fair.c" | |
1902 | #include "sched_rt.c" | |
1903 | #ifdef CONFIG_SCHED_DEBUG | |
1904 | # include "sched_debug.c" | |
1905 | #endif | |
1906 | ||
14531189 | 1907 | /* |
dd41f596 | 1908 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1909 | */ |
14531189 IM |
1910 | static inline int __normal_prio(struct task_struct *p) |
1911 | { | |
dd41f596 | 1912 | return p->static_prio; |
14531189 IM |
1913 | } |
1914 | ||
b29739f9 IM |
1915 | /* |
1916 | * Calculate the expected normal priority: i.e. priority | |
1917 | * without taking RT-inheritance into account. Might be | |
1918 | * boosted by interactivity modifiers. Changes upon fork, | |
1919 | * setprio syscalls, and whenever the interactivity | |
1920 | * estimator recalculates. | |
1921 | */ | |
36c8b586 | 1922 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1923 | { |
1924 | int prio; | |
1925 | ||
e05606d3 | 1926 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1927 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1928 | else | |
1929 | prio = __normal_prio(p); | |
1930 | return prio; | |
1931 | } | |
1932 | ||
1933 | /* | |
1934 | * Calculate the current priority, i.e. the priority | |
1935 | * taken into account by the scheduler. This value might | |
1936 | * be boosted by RT tasks, or might be boosted by | |
1937 | * interactivity modifiers. Will be RT if the task got | |
1938 | * RT-boosted. If not then it returns p->normal_prio. | |
1939 | */ | |
36c8b586 | 1940 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1941 | { |
1942 | p->normal_prio = normal_prio(p); | |
1943 | /* | |
1944 | * If we are RT tasks or we were boosted to RT priority, | |
1945 | * keep the priority unchanged. Otherwise, update priority | |
1946 | * to the normal priority: | |
1947 | */ | |
1948 | if (!rt_prio(p->prio)) | |
1949 | return p->normal_prio; | |
1950 | return p->prio; | |
1951 | } | |
1952 | ||
1da177e4 LT |
1953 | /** |
1954 | * task_curr - is this task currently executing on a CPU? | |
1955 | * @p: the task in question. | |
1956 | */ | |
36c8b586 | 1957 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1958 | { |
1959 | return cpu_curr(task_cpu(p)) == p; | |
1960 | } | |
1961 | ||
cb469845 SR |
1962 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1963 | const struct sched_class *prev_class, | |
1964 | int oldprio, int running) | |
1965 | { | |
1966 | if (prev_class != p->sched_class) { | |
1967 | if (prev_class->switched_from) | |
1968 | prev_class->switched_from(rq, p, running); | |
1969 | p->sched_class->switched_to(rq, p, running); | |
1970 | } else | |
1971 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1972 | } | |
1973 | ||
1da177e4 | 1974 | #ifdef CONFIG_SMP |
cc367732 IM |
1975 | /* |
1976 | * Is this task likely cache-hot: | |
1977 | */ | |
e7693a36 | 1978 | static int |
cc367732 IM |
1979 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
1980 | { | |
1981 | s64 delta; | |
1982 | ||
e6c8fba7 PZ |
1983 | if (p->sched_class != &fair_sched_class) |
1984 | return 0; | |
1985 | ||
f540a608 IM |
1986 | /* |
1987 | * Buddy candidates are cache hot: | |
1988 | */ | |
f685ceac | 1989 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
1990 | (&p->se == cfs_rq_of(&p->se)->next || |
1991 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
1992 | return 1; |
1993 | ||
6bc1665b IM |
1994 | if (sysctl_sched_migration_cost == -1) |
1995 | return 1; | |
1996 | if (sysctl_sched_migration_cost == 0) | |
1997 | return 0; | |
1998 | ||
cc367732 IM |
1999 | delta = now - p->se.exec_start; |
2000 | ||
2001 | return delta < (s64)sysctl_sched_migration_cost; | |
2002 | } | |
2003 | ||
dd41f596 | 2004 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2005 | { |
e2912009 PZ |
2006 | #ifdef CONFIG_SCHED_DEBUG |
2007 | /* | |
2008 | * We should never call set_task_cpu() on a blocked task, | |
2009 | * ttwu() will sort out the placement. | |
2010 | */ | |
077614ee PZ |
2011 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2012 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2013 | #endif |
2014 | ||
de1d7286 | 2015 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2016 | |
0c69774e PZ |
2017 | if (task_cpu(p) != new_cpu) { |
2018 | p->se.nr_migrations++; | |
2019 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2020 | } | |
dd41f596 IM |
2021 | |
2022 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2023 | } |
2024 | ||
969c7921 | 2025 | struct migration_arg { |
36c8b586 | 2026 | struct task_struct *task; |
1da177e4 | 2027 | int dest_cpu; |
70b97a7f | 2028 | }; |
1da177e4 | 2029 | |
969c7921 TH |
2030 | static int migration_cpu_stop(void *data); |
2031 | ||
1da177e4 LT |
2032 | /* |
2033 | * The task's runqueue lock must be held. | |
2034 | * Returns true if you have to wait for migration thread. | |
2035 | */ | |
969c7921 | 2036 | static bool migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2037 | { |
70b97a7f | 2038 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2039 | |
2040 | /* | |
2041 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2042 | * the next wake-up will properly place the task. |
1da177e4 | 2043 | */ |
969c7921 | 2044 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2045 | } |
2046 | ||
2047 | /* | |
2048 | * wait_task_inactive - wait for a thread to unschedule. | |
2049 | * | |
85ba2d86 RM |
2050 | * If @match_state is nonzero, it's the @p->state value just checked and |
2051 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2052 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2053 | * we return a positive number (its total switch count). If a second call | |
2054 | * a short while later returns the same number, the caller can be sure that | |
2055 | * @p has remained unscheduled the whole time. | |
2056 | * | |
1da177e4 LT |
2057 | * The caller must ensure that the task *will* unschedule sometime soon, |
2058 | * else this function might spin for a *long* time. This function can't | |
2059 | * be called with interrupts off, or it may introduce deadlock with | |
2060 | * smp_call_function() if an IPI is sent by the same process we are | |
2061 | * waiting to become inactive. | |
2062 | */ | |
85ba2d86 | 2063 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2064 | { |
2065 | unsigned long flags; | |
dd41f596 | 2066 | int running, on_rq; |
85ba2d86 | 2067 | unsigned long ncsw; |
70b97a7f | 2068 | struct rq *rq; |
1da177e4 | 2069 | |
3a5c359a AK |
2070 | for (;;) { |
2071 | /* | |
2072 | * We do the initial early heuristics without holding | |
2073 | * any task-queue locks at all. We'll only try to get | |
2074 | * the runqueue lock when things look like they will | |
2075 | * work out! | |
2076 | */ | |
2077 | rq = task_rq(p); | |
fa490cfd | 2078 | |
3a5c359a AK |
2079 | /* |
2080 | * If the task is actively running on another CPU | |
2081 | * still, just relax and busy-wait without holding | |
2082 | * any locks. | |
2083 | * | |
2084 | * NOTE! Since we don't hold any locks, it's not | |
2085 | * even sure that "rq" stays as the right runqueue! | |
2086 | * But we don't care, since "task_running()" will | |
2087 | * return false if the runqueue has changed and p | |
2088 | * is actually now running somewhere else! | |
2089 | */ | |
85ba2d86 RM |
2090 | while (task_running(rq, p)) { |
2091 | if (match_state && unlikely(p->state != match_state)) | |
2092 | return 0; | |
3a5c359a | 2093 | cpu_relax(); |
85ba2d86 | 2094 | } |
fa490cfd | 2095 | |
3a5c359a AK |
2096 | /* |
2097 | * Ok, time to look more closely! We need the rq | |
2098 | * lock now, to be *sure*. If we're wrong, we'll | |
2099 | * just go back and repeat. | |
2100 | */ | |
2101 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2102 | trace_sched_wait_task(p); |
3a5c359a AK |
2103 | running = task_running(rq, p); |
2104 | on_rq = p->se.on_rq; | |
85ba2d86 | 2105 | ncsw = 0; |
f31e11d8 | 2106 | if (!match_state || p->state == match_state) |
93dcf55f | 2107 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2108 | task_rq_unlock(rq, &flags); |
fa490cfd | 2109 | |
85ba2d86 RM |
2110 | /* |
2111 | * If it changed from the expected state, bail out now. | |
2112 | */ | |
2113 | if (unlikely(!ncsw)) | |
2114 | break; | |
2115 | ||
3a5c359a AK |
2116 | /* |
2117 | * Was it really running after all now that we | |
2118 | * checked with the proper locks actually held? | |
2119 | * | |
2120 | * Oops. Go back and try again.. | |
2121 | */ | |
2122 | if (unlikely(running)) { | |
2123 | cpu_relax(); | |
2124 | continue; | |
2125 | } | |
fa490cfd | 2126 | |
3a5c359a AK |
2127 | /* |
2128 | * It's not enough that it's not actively running, | |
2129 | * it must be off the runqueue _entirely_, and not | |
2130 | * preempted! | |
2131 | * | |
80dd99b3 | 2132 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2133 | * running right now), it's preempted, and we should |
2134 | * yield - it could be a while. | |
2135 | */ | |
2136 | if (unlikely(on_rq)) { | |
2137 | schedule_timeout_uninterruptible(1); | |
2138 | continue; | |
2139 | } | |
fa490cfd | 2140 | |
3a5c359a AK |
2141 | /* |
2142 | * Ahh, all good. It wasn't running, and it wasn't | |
2143 | * runnable, which means that it will never become | |
2144 | * running in the future either. We're all done! | |
2145 | */ | |
2146 | break; | |
2147 | } | |
85ba2d86 RM |
2148 | |
2149 | return ncsw; | |
1da177e4 LT |
2150 | } |
2151 | ||
2152 | /*** | |
2153 | * kick_process - kick a running thread to enter/exit the kernel | |
2154 | * @p: the to-be-kicked thread | |
2155 | * | |
2156 | * Cause a process which is running on another CPU to enter | |
2157 | * kernel-mode, without any delay. (to get signals handled.) | |
2158 | * | |
2159 | * NOTE: this function doesnt have to take the runqueue lock, | |
2160 | * because all it wants to ensure is that the remote task enters | |
2161 | * the kernel. If the IPI races and the task has been migrated | |
2162 | * to another CPU then no harm is done and the purpose has been | |
2163 | * achieved as well. | |
2164 | */ | |
36c8b586 | 2165 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2166 | { |
2167 | int cpu; | |
2168 | ||
2169 | preempt_disable(); | |
2170 | cpu = task_cpu(p); | |
2171 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2172 | smp_send_reschedule(cpu); | |
2173 | preempt_enable(); | |
2174 | } | |
b43e3521 | 2175 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2176 | #endif /* CONFIG_SMP */ |
1da177e4 | 2177 | |
0793a61d TG |
2178 | /** |
2179 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2180 | * @p: the task to evaluate | |
2181 | * @func: the function to be called | |
2182 | * @info: the function call argument | |
2183 | * | |
2184 | * Calls the function @func when the task is currently running. This might | |
2185 | * be on the current CPU, which just calls the function directly | |
2186 | */ | |
2187 | void task_oncpu_function_call(struct task_struct *p, | |
2188 | void (*func) (void *info), void *info) | |
2189 | { | |
2190 | int cpu; | |
2191 | ||
2192 | preempt_disable(); | |
2193 | cpu = task_cpu(p); | |
2194 | if (task_curr(p)) | |
2195 | smp_call_function_single(cpu, func, info, 1); | |
2196 | preempt_enable(); | |
2197 | } | |
2198 | ||
970b13ba | 2199 | #ifdef CONFIG_SMP |
30da688e ON |
2200 | /* |
2201 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2202 | */ | |
5da9a0fb PZ |
2203 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2204 | { | |
2205 | int dest_cpu; | |
2206 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2207 | ||
2208 | /* Look for allowed, online CPU in same node. */ | |
2209 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2210 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2211 | return dest_cpu; | |
2212 | ||
2213 | /* Any allowed, online CPU? */ | |
2214 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2215 | if (dest_cpu < nr_cpu_ids) | |
2216 | return dest_cpu; | |
2217 | ||
2218 | /* No more Mr. Nice Guy. */ | |
897f0b3c | 2219 | if (unlikely(dest_cpu >= nr_cpu_ids)) { |
9084bb82 | 2220 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
5da9a0fb PZ |
2221 | /* |
2222 | * Don't tell them about moving exiting tasks or | |
2223 | * kernel threads (both mm NULL), since they never | |
2224 | * leave kernel. | |
2225 | */ | |
2226 | if (p->mm && printk_ratelimit()) { | |
2227 | printk(KERN_INFO "process %d (%s) no " | |
2228 | "longer affine to cpu%d\n", | |
2229 | task_pid_nr(p), p->comm, cpu); | |
2230 | } | |
2231 | } | |
2232 | ||
2233 | return dest_cpu; | |
2234 | } | |
2235 | ||
e2912009 | 2236 | /* |
30da688e | 2237 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2238 | */ |
970b13ba | 2239 | static inline |
0017d735 | 2240 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2241 | { |
0017d735 | 2242 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2243 | |
2244 | /* | |
2245 | * In order not to call set_task_cpu() on a blocking task we need | |
2246 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2247 | * cpu. | |
2248 | * | |
2249 | * Since this is common to all placement strategies, this lives here. | |
2250 | * | |
2251 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2252 | * not worry about this generic constraint ] | |
2253 | */ | |
2254 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2255 | !cpu_online(cpu))) |
5da9a0fb | 2256 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2257 | |
2258 | return cpu; | |
970b13ba | 2259 | } |
09a40af5 MG |
2260 | |
2261 | static void update_avg(u64 *avg, u64 sample) | |
2262 | { | |
2263 | s64 diff = sample - *avg; | |
2264 | *avg += diff >> 3; | |
2265 | } | |
970b13ba PZ |
2266 | #endif |
2267 | ||
9ed3811a TH |
2268 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2269 | bool is_sync, bool is_migrate, bool is_local, | |
2270 | unsigned long en_flags) | |
2271 | { | |
2272 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2273 | if (is_sync) | |
2274 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2275 | if (is_migrate) | |
2276 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2277 | if (is_local) | |
2278 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2279 | else | |
2280 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2281 | ||
2282 | activate_task(rq, p, en_flags); | |
2283 | } | |
2284 | ||
2285 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2286 | int wake_flags, bool success) | |
2287 | { | |
2288 | trace_sched_wakeup(p, success); | |
2289 | check_preempt_curr(rq, p, wake_flags); | |
2290 | ||
2291 | p->state = TASK_RUNNING; | |
2292 | #ifdef CONFIG_SMP | |
2293 | if (p->sched_class->task_woken) | |
2294 | p->sched_class->task_woken(rq, p); | |
2295 | ||
2296 | if (unlikely(rq->idle_stamp)) { | |
2297 | u64 delta = rq->clock - rq->idle_stamp; | |
2298 | u64 max = 2*sysctl_sched_migration_cost; | |
2299 | ||
2300 | if (delta > max) | |
2301 | rq->avg_idle = max; | |
2302 | else | |
2303 | update_avg(&rq->avg_idle, delta); | |
2304 | rq->idle_stamp = 0; | |
2305 | } | |
2306 | #endif | |
21aa9af0 TH |
2307 | /* if a worker is waking up, notify workqueue */ |
2308 | if ((p->flags & PF_WQ_WORKER) && success) | |
2309 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2310 | } |
2311 | ||
2312 | /** | |
1da177e4 | 2313 | * try_to_wake_up - wake up a thread |
9ed3811a | 2314 | * @p: the thread to be awakened |
1da177e4 | 2315 | * @state: the mask of task states that can be woken |
9ed3811a | 2316 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2317 | * |
2318 | * Put it on the run-queue if it's not already there. The "current" | |
2319 | * thread is always on the run-queue (except when the actual | |
2320 | * re-schedule is in progress), and as such you're allowed to do | |
2321 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2322 | * runnable without the overhead of this. | |
2323 | * | |
9ed3811a TH |
2324 | * Returns %true if @p was woken up, %false if it was already running |
2325 | * or @state didn't match @p's state. | |
1da177e4 | 2326 | */ |
7d478721 PZ |
2327 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2328 | int wake_flags) | |
1da177e4 | 2329 | { |
cc367732 | 2330 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2331 | unsigned long flags; |
371fd7e7 | 2332 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2333 | struct rq *rq; |
1da177e4 | 2334 | |
e9c84311 | 2335 | this_cpu = get_cpu(); |
2398f2c6 | 2336 | |
04e2f174 | 2337 | smp_wmb(); |
ab3b3aa5 | 2338 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2339 | if (!(p->state & state)) |
1da177e4 LT |
2340 | goto out; |
2341 | ||
dd41f596 | 2342 | if (p->se.on_rq) |
1da177e4 LT |
2343 | goto out_running; |
2344 | ||
2345 | cpu = task_cpu(p); | |
cc367732 | 2346 | orig_cpu = cpu; |
1da177e4 LT |
2347 | |
2348 | #ifdef CONFIG_SMP | |
2349 | if (unlikely(task_running(rq, p))) | |
2350 | goto out_activate; | |
2351 | ||
e9c84311 PZ |
2352 | /* |
2353 | * In order to handle concurrent wakeups and release the rq->lock | |
2354 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2355 | * |
2356 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2357 | */ |
cc87f76a PZ |
2358 | if (task_contributes_to_load(p)) { |
2359 | if (likely(cpu_online(orig_cpu))) | |
2360 | rq->nr_uninterruptible--; | |
2361 | else | |
2362 | this_rq()->nr_uninterruptible--; | |
2363 | } | |
e9c84311 | 2364 | p->state = TASK_WAKING; |
efbbd05a | 2365 | |
371fd7e7 | 2366 | if (p->sched_class->task_waking) { |
efbbd05a | 2367 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2368 | en_flags |= ENQUEUE_WAKING; |
2369 | } | |
efbbd05a | 2370 | |
0017d735 PZ |
2371 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2372 | if (cpu != orig_cpu) | |
5d2f5a61 | 2373 | set_task_cpu(p, cpu); |
0017d735 | 2374 | __task_rq_unlock(rq); |
ab19cb23 | 2375 | |
0970d299 PZ |
2376 | rq = cpu_rq(cpu); |
2377 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2378 | |
0970d299 PZ |
2379 | /* |
2380 | * We migrated the task without holding either rq->lock, however | |
2381 | * since the task is not on the task list itself, nobody else | |
2382 | * will try and migrate the task, hence the rq should match the | |
2383 | * cpu we just moved it to. | |
2384 | */ | |
2385 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2386 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2387 | |
e7693a36 GH |
2388 | #ifdef CONFIG_SCHEDSTATS |
2389 | schedstat_inc(rq, ttwu_count); | |
2390 | if (cpu == this_cpu) | |
2391 | schedstat_inc(rq, ttwu_local); | |
2392 | else { | |
2393 | struct sched_domain *sd; | |
2394 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2395 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2396 | schedstat_inc(sd, ttwu_wake_remote); |
2397 | break; | |
2398 | } | |
2399 | } | |
2400 | } | |
6d6bc0ad | 2401 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2402 | |
1da177e4 LT |
2403 | out_activate: |
2404 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2405 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2406 | cpu == this_cpu, en_flags); | |
1da177e4 | 2407 | success = 1; |
1da177e4 | 2408 | out_running: |
9ed3811a | 2409 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2410 | out: |
2411 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2412 | put_cpu(); |
1da177e4 LT |
2413 | |
2414 | return success; | |
2415 | } | |
2416 | ||
21aa9af0 TH |
2417 | /** |
2418 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2419 | * @p: the thread to be awakened | |
2420 | * | |
2421 | * Put @p on the run-queue if it's not alredy there. The caller must | |
2422 | * ensure that this_rq() is locked, @p is bound to this_rq() and not | |
2423 | * the current task. this_rq() stays locked over invocation. | |
2424 | */ | |
2425 | static void try_to_wake_up_local(struct task_struct *p) | |
2426 | { | |
2427 | struct rq *rq = task_rq(p); | |
2428 | bool success = false; | |
2429 | ||
2430 | BUG_ON(rq != this_rq()); | |
2431 | BUG_ON(p == current); | |
2432 | lockdep_assert_held(&rq->lock); | |
2433 | ||
2434 | if (!(p->state & TASK_NORMAL)) | |
2435 | return; | |
2436 | ||
2437 | if (!p->se.on_rq) { | |
2438 | if (likely(!task_running(rq, p))) { | |
2439 | schedstat_inc(rq, ttwu_count); | |
2440 | schedstat_inc(rq, ttwu_local); | |
2441 | } | |
2442 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2443 | success = true; | |
2444 | } | |
2445 | ttwu_post_activation(p, rq, 0, success); | |
2446 | } | |
2447 | ||
50fa610a DH |
2448 | /** |
2449 | * wake_up_process - Wake up a specific process | |
2450 | * @p: The process to be woken up. | |
2451 | * | |
2452 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2453 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2454 | * running. | |
2455 | * | |
2456 | * It may be assumed that this function implies a write memory barrier before | |
2457 | * changing the task state if and only if any tasks are woken up. | |
2458 | */ | |
7ad5b3a5 | 2459 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2460 | { |
d9514f6c | 2461 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2462 | } |
1da177e4 LT |
2463 | EXPORT_SYMBOL(wake_up_process); |
2464 | ||
7ad5b3a5 | 2465 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2466 | { |
2467 | return try_to_wake_up(p, state, 0); | |
2468 | } | |
2469 | ||
1da177e4 LT |
2470 | /* |
2471 | * Perform scheduler related setup for a newly forked process p. | |
2472 | * p is forked by current. | |
dd41f596 IM |
2473 | * |
2474 | * __sched_fork() is basic setup used by init_idle() too: | |
2475 | */ | |
2476 | static void __sched_fork(struct task_struct *p) | |
2477 | { | |
dd41f596 IM |
2478 | p->se.exec_start = 0; |
2479 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2480 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2481 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2482 | |
2483 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2484 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2485 | #endif |
476d139c | 2486 | |
fa717060 | 2487 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2488 | p->se.on_rq = 0; |
4a55bd5e | 2489 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2490 | |
e107be36 AK |
2491 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2492 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2493 | #endif | |
dd41f596 IM |
2494 | } |
2495 | ||
2496 | /* | |
2497 | * fork()/clone()-time setup: | |
2498 | */ | |
2499 | void sched_fork(struct task_struct *p, int clone_flags) | |
2500 | { | |
2501 | int cpu = get_cpu(); | |
2502 | ||
2503 | __sched_fork(p); | |
06b83b5f | 2504 | /* |
0017d735 | 2505 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2506 | * nobody will actually run it, and a signal or other external |
2507 | * event cannot wake it up and insert it on the runqueue either. | |
2508 | */ | |
0017d735 | 2509 | p->state = TASK_RUNNING; |
dd41f596 | 2510 | |
b9dc29e7 MG |
2511 | /* |
2512 | * Revert to default priority/policy on fork if requested. | |
2513 | */ | |
2514 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2515 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2516 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2517 | p->normal_prio = p->static_prio; |
2518 | } | |
b9dc29e7 | 2519 | |
6c697bdf MG |
2520 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2521 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2522 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2523 | set_load_weight(p); |
2524 | } | |
2525 | ||
b9dc29e7 MG |
2526 | /* |
2527 | * We don't need the reset flag anymore after the fork. It has | |
2528 | * fulfilled its duty: | |
2529 | */ | |
2530 | p->sched_reset_on_fork = 0; | |
2531 | } | |
ca94c442 | 2532 | |
f83f9ac2 PW |
2533 | /* |
2534 | * Make sure we do not leak PI boosting priority to the child. | |
2535 | */ | |
2536 | p->prio = current->normal_prio; | |
2537 | ||
2ddbf952 HS |
2538 | if (!rt_prio(p->prio)) |
2539 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2540 | |
cd29fe6f PZ |
2541 | if (p->sched_class->task_fork) |
2542 | p->sched_class->task_fork(p); | |
2543 | ||
5f3edc1b PZ |
2544 | set_task_cpu(p, cpu); |
2545 | ||
52f17b6c | 2546 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2547 | if (likely(sched_info_on())) |
52f17b6c | 2548 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2549 | #endif |
d6077cb8 | 2550 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2551 | p->oncpu = 0; |
2552 | #endif | |
1da177e4 | 2553 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2554 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2555 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2556 | #endif |
917b627d GH |
2557 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2558 | ||
476d139c | 2559 | put_cpu(); |
1da177e4 LT |
2560 | } |
2561 | ||
2562 | /* | |
2563 | * wake_up_new_task - wake up a newly created task for the first time. | |
2564 | * | |
2565 | * This function will do some initial scheduler statistics housekeeping | |
2566 | * that must be done for every newly created context, then puts the task | |
2567 | * on the runqueue and wakes it. | |
2568 | */ | |
7ad5b3a5 | 2569 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2570 | { |
2571 | unsigned long flags; | |
dd41f596 | 2572 | struct rq *rq; |
c890692b | 2573 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2574 | |
2575 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2576 | rq = task_rq_lock(p, &flags); |
2577 | p->state = TASK_WAKING; | |
2578 | ||
fabf318e PZ |
2579 | /* |
2580 | * Fork balancing, do it here and not earlier because: | |
2581 | * - cpus_allowed can change in the fork path | |
2582 | * - any previously selected cpu might disappear through hotplug | |
2583 | * | |
0017d735 PZ |
2584 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2585 | * without people poking at ->cpus_allowed. | |
fabf318e | 2586 | */ |
0017d735 | 2587 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2588 | set_task_cpu(p, cpu); |
1da177e4 | 2589 | |
06b83b5f | 2590 | p->state = TASK_RUNNING; |
0017d735 PZ |
2591 | task_rq_unlock(rq, &flags); |
2592 | #endif | |
2593 | ||
2594 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2595 | activate_task(rq, p, 0); |
27a9da65 | 2596 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2597 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2598 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2599 | if (p->sched_class->task_woken) |
2600 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2601 | #endif |
dd41f596 | 2602 | task_rq_unlock(rq, &flags); |
fabf318e | 2603 | put_cpu(); |
1da177e4 LT |
2604 | } |
2605 | ||
e107be36 AK |
2606 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2607 | ||
2608 | /** | |
80dd99b3 | 2609 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2610 | * @notifier: notifier struct to register |
e107be36 AK |
2611 | */ |
2612 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2613 | { | |
2614 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2615 | } | |
2616 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2617 | ||
2618 | /** | |
2619 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2620 | * @notifier: notifier struct to unregister |
e107be36 AK |
2621 | * |
2622 | * This is safe to call from within a preemption notifier. | |
2623 | */ | |
2624 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2625 | { | |
2626 | hlist_del(¬ifier->link); | |
2627 | } | |
2628 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2629 | ||
2630 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2631 | { | |
2632 | struct preempt_notifier *notifier; | |
2633 | struct hlist_node *node; | |
2634 | ||
2635 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2636 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2637 | } | |
2638 | ||
2639 | static void | |
2640 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2641 | struct task_struct *next) | |
2642 | { | |
2643 | struct preempt_notifier *notifier; | |
2644 | struct hlist_node *node; | |
2645 | ||
2646 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2647 | notifier->ops->sched_out(notifier, next); | |
2648 | } | |
2649 | ||
6d6bc0ad | 2650 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2651 | |
2652 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2653 | { | |
2654 | } | |
2655 | ||
2656 | static void | |
2657 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2658 | struct task_struct *next) | |
2659 | { | |
2660 | } | |
2661 | ||
6d6bc0ad | 2662 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2663 | |
4866cde0 NP |
2664 | /** |
2665 | * prepare_task_switch - prepare to switch tasks | |
2666 | * @rq: the runqueue preparing to switch | |
421cee29 | 2667 | * @prev: the current task that is being switched out |
4866cde0 NP |
2668 | * @next: the task we are going to switch to. |
2669 | * | |
2670 | * This is called with the rq lock held and interrupts off. It must | |
2671 | * be paired with a subsequent finish_task_switch after the context | |
2672 | * switch. | |
2673 | * | |
2674 | * prepare_task_switch sets up locking and calls architecture specific | |
2675 | * hooks. | |
2676 | */ | |
e107be36 AK |
2677 | static inline void |
2678 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2679 | struct task_struct *next) | |
4866cde0 | 2680 | { |
e107be36 | 2681 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2682 | prepare_lock_switch(rq, next); |
2683 | prepare_arch_switch(next); | |
2684 | } | |
2685 | ||
1da177e4 LT |
2686 | /** |
2687 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2688 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2689 | * @prev: the thread we just switched away from. |
2690 | * | |
4866cde0 NP |
2691 | * finish_task_switch must be called after the context switch, paired |
2692 | * with a prepare_task_switch call before the context switch. | |
2693 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2694 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2695 | * |
2696 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2697 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2698 | * with the lock held can cause deadlocks; see schedule() for |
2699 | * details.) | |
2700 | */ | |
a9957449 | 2701 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2702 | __releases(rq->lock) |
2703 | { | |
1da177e4 | 2704 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2705 | long prev_state; |
1da177e4 LT |
2706 | |
2707 | rq->prev_mm = NULL; | |
2708 | ||
2709 | /* | |
2710 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2711 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2712 | * schedule one last time. The schedule call will never return, and |
2713 | * the scheduled task must drop that reference. | |
c394cc9f | 2714 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2715 | * still held, otherwise prev could be scheduled on another cpu, die |
2716 | * there before we look at prev->state, and then the reference would | |
2717 | * be dropped twice. | |
2718 | * Manfred Spraul <manfred@colorfullife.com> | |
2719 | */ | |
55a101f8 | 2720 | prev_state = prev->state; |
4866cde0 | 2721 | finish_arch_switch(prev); |
8381f65d JI |
2722 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2723 | local_irq_disable(); | |
2724 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2725 | perf_event_task_sched_in(current); |
8381f65d JI |
2726 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2727 | local_irq_enable(); | |
2728 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2729 | finish_lock_switch(rq, prev); |
e8fa1362 | 2730 | |
e107be36 | 2731 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2732 | if (mm) |
2733 | mmdrop(mm); | |
c394cc9f | 2734 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2735 | /* |
2736 | * Remove function-return probe instances associated with this | |
2737 | * task and put them back on the free list. | |
9761eea8 | 2738 | */ |
c6fd91f0 | 2739 | kprobe_flush_task(prev); |
1da177e4 | 2740 | put_task_struct(prev); |
c6fd91f0 | 2741 | } |
1da177e4 LT |
2742 | } |
2743 | ||
3f029d3c GH |
2744 | #ifdef CONFIG_SMP |
2745 | ||
2746 | /* assumes rq->lock is held */ | |
2747 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2748 | { | |
2749 | if (prev->sched_class->pre_schedule) | |
2750 | prev->sched_class->pre_schedule(rq, prev); | |
2751 | } | |
2752 | ||
2753 | /* rq->lock is NOT held, but preemption is disabled */ | |
2754 | static inline void post_schedule(struct rq *rq) | |
2755 | { | |
2756 | if (rq->post_schedule) { | |
2757 | unsigned long flags; | |
2758 | ||
05fa785c | 2759 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2760 | if (rq->curr->sched_class->post_schedule) |
2761 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2762 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2763 | |
2764 | rq->post_schedule = 0; | |
2765 | } | |
2766 | } | |
2767 | ||
2768 | #else | |
da19ab51 | 2769 | |
3f029d3c GH |
2770 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2771 | { | |
2772 | } | |
2773 | ||
2774 | static inline void post_schedule(struct rq *rq) | |
2775 | { | |
1da177e4 LT |
2776 | } |
2777 | ||
3f029d3c GH |
2778 | #endif |
2779 | ||
1da177e4 LT |
2780 | /** |
2781 | * schedule_tail - first thing a freshly forked thread must call. | |
2782 | * @prev: the thread we just switched away from. | |
2783 | */ | |
36c8b586 | 2784 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2785 | __releases(rq->lock) |
2786 | { | |
70b97a7f IM |
2787 | struct rq *rq = this_rq(); |
2788 | ||
4866cde0 | 2789 | finish_task_switch(rq, prev); |
da19ab51 | 2790 | |
3f029d3c GH |
2791 | /* |
2792 | * FIXME: do we need to worry about rq being invalidated by the | |
2793 | * task_switch? | |
2794 | */ | |
2795 | post_schedule(rq); | |
70b97a7f | 2796 | |
4866cde0 NP |
2797 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2798 | /* In this case, finish_task_switch does not reenable preemption */ | |
2799 | preempt_enable(); | |
2800 | #endif | |
1da177e4 | 2801 | if (current->set_child_tid) |
b488893a | 2802 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2803 | } |
2804 | ||
2805 | /* | |
2806 | * context_switch - switch to the new MM and the new | |
2807 | * thread's register state. | |
2808 | */ | |
dd41f596 | 2809 | static inline void |
70b97a7f | 2810 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2811 | struct task_struct *next) |
1da177e4 | 2812 | { |
dd41f596 | 2813 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2814 | |
e107be36 | 2815 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2816 | trace_sched_switch(prev, next); |
dd41f596 IM |
2817 | mm = next->mm; |
2818 | oldmm = prev->active_mm; | |
9226d125 ZA |
2819 | /* |
2820 | * For paravirt, this is coupled with an exit in switch_to to | |
2821 | * combine the page table reload and the switch backend into | |
2822 | * one hypercall. | |
2823 | */ | |
224101ed | 2824 | arch_start_context_switch(prev); |
9226d125 | 2825 | |
710390d9 | 2826 | if (likely(!mm)) { |
1da177e4 LT |
2827 | next->active_mm = oldmm; |
2828 | atomic_inc(&oldmm->mm_count); | |
2829 | enter_lazy_tlb(oldmm, next); | |
2830 | } else | |
2831 | switch_mm(oldmm, mm, next); | |
2832 | ||
710390d9 | 2833 | if (likely(!prev->mm)) { |
1da177e4 | 2834 | prev->active_mm = NULL; |
1da177e4 LT |
2835 | rq->prev_mm = oldmm; |
2836 | } | |
3a5f5e48 IM |
2837 | /* |
2838 | * Since the runqueue lock will be released by the next | |
2839 | * task (which is an invalid locking op but in the case | |
2840 | * of the scheduler it's an obvious special-case), so we | |
2841 | * do an early lockdep release here: | |
2842 | */ | |
2843 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2844 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2845 | #endif |
1da177e4 LT |
2846 | |
2847 | /* Here we just switch the register state and the stack. */ | |
2848 | switch_to(prev, next, prev); | |
2849 | ||
dd41f596 IM |
2850 | barrier(); |
2851 | /* | |
2852 | * this_rq must be evaluated again because prev may have moved | |
2853 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2854 | * frame will be invalid. | |
2855 | */ | |
2856 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2857 | } |
2858 | ||
2859 | /* | |
2860 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2861 | * | |
2862 | * externally visible scheduler statistics: current number of runnable | |
2863 | * threads, current number of uninterruptible-sleeping threads, total | |
2864 | * number of context switches performed since bootup. | |
2865 | */ | |
2866 | unsigned long nr_running(void) | |
2867 | { | |
2868 | unsigned long i, sum = 0; | |
2869 | ||
2870 | for_each_online_cpu(i) | |
2871 | sum += cpu_rq(i)->nr_running; | |
2872 | ||
2873 | return sum; | |
f711f609 | 2874 | } |
1da177e4 LT |
2875 | |
2876 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2877 | { |
1da177e4 | 2878 | unsigned long i, sum = 0; |
f711f609 | 2879 | |
0a945022 | 2880 | for_each_possible_cpu(i) |
1da177e4 | 2881 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2882 | |
2883 | /* | |
1da177e4 LT |
2884 | * Since we read the counters lockless, it might be slightly |
2885 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2886 | */ |
1da177e4 LT |
2887 | if (unlikely((long)sum < 0)) |
2888 | sum = 0; | |
f711f609 | 2889 | |
1da177e4 | 2890 | return sum; |
f711f609 | 2891 | } |
f711f609 | 2892 | |
1da177e4 | 2893 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2894 | { |
cc94abfc SR |
2895 | int i; |
2896 | unsigned long long sum = 0; | |
46cb4b7c | 2897 | |
0a945022 | 2898 | for_each_possible_cpu(i) |
1da177e4 | 2899 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2900 | |
1da177e4 LT |
2901 | return sum; |
2902 | } | |
483b4ee6 | 2903 | |
1da177e4 LT |
2904 | unsigned long nr_iowait(void) |
2905 | { | |
2906 | unsigned long i, sum = 0; | |
483b4ee6 | 2907 | |
0a945022 | 2908 | for_each_possible_cpu(i) |
1da177e4 | 2909 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2910 | |
1da177e4 LT |
2911 | return sum; |
2912 | } | |
483b4ee6 | 2913 | |
69d25870 AV |
2914 | unsigned long nr_iowait_cpu(void) |
2915 | { | |
2916 | struct rq *this = this_rq(); | |
2917 | return atomic_read(&this->nr_iowait); | |
2918 | } | |
46cb4b7c | 2919 | |
69d25870 AV |
2920 | unsigned long this_cpu_load(void) |
2921 | { | |
2922 | struct rq *this = this_rq(); | |
2923 | return this->cpu_load[0]; | |
2924 | } | |
e790fb0b | 2925 | |
46cb4b7c | 2926 | |
dce48a84 TG |
2927 | /* Variables and functions for calc_load */ |
2928 | static atomic_long_t calc_load_tasks; | |
2929 | static unsigned long calc_load_update; | |
2930 | unsigned long avenrun[3]; | |
2931 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2932 | |
74f5187a PZ |
2933 | static long calc_load_fold_active(struct rq *this_rq) |
2934 | { | |
2935 | long nr_active, delta = 0; | |
2936 | ||
2937 | nr_active = this_rq->nr_running; | |
2938 | nr_active += (long) this_rq->nr_uninterruptible; | |
2939 | ||
2940 | if (nr_active != this_rq->calc_load_active) { | |
2941 | delta = nr_active - this_rq->calc_load_active; | |
2942 | this_rq->calc_load_active = nr_active; | |
2943 | } | |
2944 | ||
2945 | return delta; | |
2946 | } | |
2947 | ||
2948 | #ifdef CONFIG_NO_HZ | |
2949 | /* | |
2950 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2951 | * | |
2952 | * When making the ILB scale, we should try to pull this in as well. | |
2953 | */ | |
2954 | static atomic_long_t calc_load_tasks_idle; | |
2955 | ||
2956 | static void calc_load_account_idle(struct rq *this_rq) | |
2957 | { | |
2958 | long delta; | |
2959 | ||
2960 | delta = calc_load_fold_active(this_rq); | |
2961 | if (delta) | |
2962 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2963 | } | |
2964 | ||
2965 | static long calc_load_fold_idle(void) | |
2966 | { | |
2967 | long delta = 0; | |
2968 | ||
2969 | /* | |
2970 | * Its got a race, we don't care... | |
2971 | */ | |
2972 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2973 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2974 | ||
2975 | return delta; | |
2976 | } | |
2977 | #else | |
2978 | static void calc_load_account_idle(struct rq *this_rq) | |
2979 | { | |
2980 | } | |
2981 | ||
2982 | static inline long calc_load_fold_idle(void) | |
2983 | { | |
2984 | return 0; | |
2985 | } | |
2986 | #endif | |
2987 | ||
2d02494f TG |
2988 | /** |
2989 | * get_avenrun - get the load average array | |
2990 | * @loads: pointer to dest load array | |
2991 | * @offset: offset to add | |
2992 | * @shift: shift count to shift the result left | |
2993 | * | |
2994 | * These values are estimates at best, so no need for locking. | |
2995 | */ | |
2996 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
2997 | { | |
2998 | loads[0] = (avenrun[0] + offset) << shift; | |
2999 | loads[1] = (avenrun[1] + offset) << shift; | |
3000 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3001 | } |
46cb4b7c | 3002 | |
dce48a84 TG |
3003 | static unsigned long |
3004 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3005 | { |
dce48a84 TG |
3006 | load *= exp; |
3007 | load += active * (FIXED_1 - exp); | |
3008 | return load >> FSHIFT; | |
3009 | } | |
46cb4b7c SS |
3010 | |
3011 | /* | |
dce48a84 TG |
3012 | * calc_load - update the avenrun load estimates 10 ticks after the |
3013 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3014 | */ |
dce48a84 | 3015 | void calc_global_load(void) |
7835b98b | 3016 | { |
dce48a84 TG |
3017 | unsigned long upd = calc_load_update + 10; |
3018 | long active; | |
1da177e4 | 3019 | |
dce48a84 TG |
3020 | if (time_before(jiffies, upd)) |
3021 | return; | |
1da177e4 | 3022 | |
dce48a84 TG |
3023 | active = atomic_long_read(&calc_load_tasks); |
3024 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3025 | |
dce48a84 TG |
3026 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3027 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3028 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3029 | |
dce48a84 TG |
3030 | calc_load_update += LOAD_FREQ; |
3031 | } | |
1da177e4 | 3032 | |
dce48a84 | 3033 | /* |
74f5187a PZ |
3034 | * Called from update_cpu_load() to periodically update this CPU's |
3035 | * active count. | |
dce48a84 TG |
3036 | */ |
3037 | static void calc_load_account_active(struct rq *this_rq) | |
3038 | { | |
74f5187a | 3039 | long delta; |
08c183f3 | 3040 | |
74f5187a PZ |
3041 | if (time_before(jiffies, this_rq->calc_load_update)) |
3042 | return; | |
783609c6 | 3043 | |
74f5187a PZ |
3044 | delta = calc_load_fold_active(this_rq); |
3045 | delta += calc_load_fold_idle(); | |
3046 | if (delta) | |
dce48a84 | 3047 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3048 | |
3049 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3050 | } |
3051 | ||
3052 | /* | |
dd41f596 IM |
3053 | * Update rq->cpu_load[] statistics. This function is usually called every |
3054 | * scheduler tick (TICK_NSEC). | |
46cb4b7c | 3055 | */ |
dd41f596 | 3056 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3057 | { |
495eca49 | 3058 | unsigned long this_load = this_rq->load.weight; |
dd41f596 | 3059 | int i, scale; |
46cb4b7c | 3060 | |
dd41f596 | 3061 | this_rq->nr_load_updates++; |
46cb4b7c | 3062 | |
dd41f596 IM |
3063 | /* Update our load: */ |
3064 | for (i = 0, scale = 1; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
3065 | unsigned long old_load, new_load; | |
7d1e6a9b | 3066 | |
dd41f596 | 3067 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3068 | |
dd41f596 IM |
3069 | old_load = this_rq->cpu_load[i]; |
3070 | new_load = this_load; | |
a25707f3 IM |
3071 | /* |
3072 | * Round up the averaging division if load is increasing. This | |
3073 | * prevents us from getting stuck on 9 if the load is 10, for | |
3074 | * example. | |
3075 | */ | |
3076 | if (new_load > old_load) | |
3077 | new_load += scale-1; | |
dd41f596 IM |
3078 | this_rq->cpu_load[i] = (old_load*(scale-1) + new_load) >> i; |
3079 | } | |
46cb4b7c | 3080 | |
74f5187a | 3081 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3082 | } |
3083 | ||
dd41f596 | 3084 | #ifdef CONFIG_SMP |
8a0be9ef | 3085 | |
46cb4b7c | 3086 | /* |
38022906 PZ |
3087 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3088 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3089 | */ |
38022906 | 3090 | void sched_exec(void) |
46cb4b7c | 3091 | { |
38022906 | 3092 | struct task_struct *p = current; |
1da177e4 | 3093 | unsigned long flags; |
70b97a7f | 3094 | struct rq *rq; |
0017d735 | 3095 | int dest_cpu; |
46cb4b7c | 3096 | |
1da177e4 | 3097 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3098 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3099 | if (dest_cpu == smp_processor_id()) | |
3100 | goto unlock; | |
38022906 | 3101 | |
46cb4b7c | 3102 | /* |
38022906 | 3103 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3104 | */ |
30da688e | 3105 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
969c7921 TH |
3106 | likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) { |
3107 | struct migration_arg arg = { p, dest_cpu }; | |
46cb4b7c | 3108 | |
1da177e4 | 3109 | task_rq_unlock(rq, &flags); |
969c7921 | 3110 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3111 | return; |
3112 | } | |
0017d735 | 3113 | unlock: |
1da177e4 | 3114 | task_rq_unlock(rq, &flags); |
1da177e4 | 3115 | } |
dd41f596 | 3116 | |
1da177e4 LT |
3117 | #endif |
3118 | ||
1da177e4 LT |
3119 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3120 | ||
3121 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3122 | ||
3123 | /* | |
c5f8d995 | 3124 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3125 | * @p in case that task is currently running. |
c5f8d995 HS |
3126 | * |
3127 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3128 | */ |
c5f8d995 HS |
3129 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3130 | { | |
3131 | u64 ns = 0; | |
3132 | ||
3133 | if (task_current(rq, p)) { | |
3134 | update_rq_clock(rq); | |
3135 | ns = rq->clock - p->se.exec_start; | |
3136 | if ((s64)ns < 0) | |
3137 | ns = 0; | |
3138 | } | |
3139 | ||
3140 | return ns; | |
3141 | } | |
3142 | ||
bb34d92f | 3143 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3144 | { |
1da177e4 | 3145 | unsigned long flags; |
41b86e9c | 3146 | struct rq *rq; |
bb34d92f | 3147 | u64 ns = 0; |
48f24c4d | 3148 | |
41b86e9c | 3149 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3150 | ns = do_task_delta_exec(p, rq); |
3151 | task_rq_unlock(rq, &flags); | |
1508487e | 3152 | |
c5f8d995 HS |
3153 | return ns; |
3154 | } | |
f06febc9 | 3155 | |
c5f8d995 HS |
3156 | /* |
3157 | * Return accounted runtime for the task. | |
3158 | * In case the task is currently running, return the runtime plus current's | |
3159 | * pending runtime that have not been accounted yet. | |
3160 | */ | |
3161 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3162 | { | |
3163 | unsigned long flags; | |
3164 | struct rq *rq; | |
3165 | u64 ns = 0; | |
3166 | ||
3167 | rq = task_rq_lock(p, &flags); | |
3168 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3169 | task_rq_unlock(rq, &flags); | |
3170 | ||
3171 | return ns; | |
3172 | } | |
48f24c4d | 3173 | |
c5f8d995 HS |
3174 | /* |
3175 | * Return sum_exec_runtime for the thread group. | |
3176 | * In case the task is currently running, return the sum plus current's | |
3177 | * pending runtime that have not been accounted yet. | |
3178 | * | |
3179 | * Note that the thread group might have other running tasks as well, | |
3180 | * so the return value not includes other pending runtime that other | |
3181 | * running tasks might have. | |
3182 | */ | |
3183 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3184 | { | |
3185 | struct task_cputime totals; | |
3186 | unsigned long flags; | |
3187 | struct rq *rq; | |
3188 | u64 ns; | |
3189 | ||
3190 | rq = task_rq_lock(p, &flags); | |
3191 | thread_group_cputime(p, &totals); | |
3192 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3193 | task_rq_unlock(rq, &flags); |
48f24c4d | 3194 | |
1da177e4 LT |
3195 | return ns; |
3196 | } | |
3197 | ||
1da177e4 LT |
3198 | /* |
3199 | * Account user cpu time to a process. | |
3200 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3201 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3202 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3203 | */ |
457533a7 MS |
3204 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3205 | cputime_t cputime_scaled) | |
1da177e4 LT |
3206 | { |
3207 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3208 | cputime64_t tmp; | |
3209 | ||
457533a7 | 3210 | /* Add user time to process. */ |
1da177e4 | 3211 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3212 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3213 | account_group_user_time(p, cputime); |
1da177e4 LT |
3214 | |
3215 | /* Add user time to cpustat. */ | |
3216 | tmp = cputime_to_cputime64(cputime); | |
3217 | if (TASK_NICE(p) > 0) | |
3218 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3219 | else | |
3220 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3221 | |
3222 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3223 | /* Account for user time used */ |
3224 | acct_update_integrals(p); | |
1da177e4 LT |
3225 | } |
3226 | ||
94886b84 LV |
3227 | /* |
3228 | * Account guest cpu time to a process. | |
3229 | * @p: the process that the cpu time gets accounted to | |
3230 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3231 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3232 | */ |
457533a7 MS |
3233 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3234 | cputime_t cputime_scaled) | |
94886b84 LV |
3235 | { |
3236 | cputime64_t tmp; | |
3237 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3238 | ||
3239 | tmp = cputime_to_cputime64(cputime); | |
3240 | ||
457533a7 | 3241 | /* Add guest time to process. */ |
94886b84 | 3242 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3243 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3244 | account_group_user_time(p, cputime); |
94886b84 LV |
3245 | p->gtime = cputime_add(p->gtime, cputime); |
3246 | ||
457533a7 | 3247 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3248 | if (TASK_NICE(p) > 0) { |
3249 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3250 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3251 | } else { | |
3252 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3253 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3254 | } | |
94886b84 LV |
3255 | } |
3256 | ||
1da177e4 LT |
3257 | /* |
3258 | * Account system cpu time to a process. | |
3259 | * @p: the process that the cpu time gets accounted to | |
3260 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3261 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3262 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3263 | */ |
3264 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3265 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3266 | { |
3267 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3268 | cputime64_t tmp; |
3269 | ||
983ed7a6 | 3270 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3271 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3272 | return; |
3273 | } | |
94886b84 | 3274 | |
457533a7 | 3275 | /* Add system time to process. */ |
1da177e4 | 3276 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3277 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3278 | account_group_system_time(p, cputime); |
1da177e4 LT |
3279 | |
3280 | /* Add system time to cpustat. */ | |
3281 | tmp = cputime_to_cputime64(cputime); | |
3282 | if (hardirq_count() - hardirq_offset) | |
3283 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3284 | else if (softirq_count()) | |
3285 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3286 | else |
79741dd3 MS |
3287 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3288 | ||
ef12fefa BR |
3289 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3290 | ||
1da177e4 LT |
3291 | /* Account for system time used */ |
3292 | acct_update_integrals(p); | |
1da177e4 LT |
3293 | } |
3294 | ||
c66f08be | 3295 | /* |
1da177e4 | 3296 | * Account for involuntary wait time. |
1da177e4 | 3297 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3298 | */ |
79741dd3 | 3299 | void account_steal_time(cputime_t cputime) |
c66f08be | 3300 | { |
79741dd3 MS |
3301 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3302 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3303 | ||
3304 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3305 | } |
3306 | ||
1da177e4 | 3307 | /* |
79741dd3 MS |
3308 | * Account for idle time. |
3309 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3310 | */ |
79741dd3 | 3311 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3312 | { |
3313 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3314 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3315 | struct rq *rq = this_rq(); |
1da177e4 | 3316 | |
79741dd3 MS |
3317 | if (atomic_read(&rq->nr_iowait) > 0) |
3318 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3319 | else | |
3320 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3321 | } |
3322 | ||
79741dd3 MS |
3323 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3324 | ||
3325 | /* | |
3326 | * Account a single tick of cpu time. | |
3327 | * @p: the process that the cpu time gets accounted to | |
3328 | * @user_tick: indicates if the tick is a user or a system tick | |
3329 | */ | |
3330 | void account_process_tick(struct task_struct *p, int user_tick) | |
3331 | { | |
a42548a1 | 3332 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3333 | struct rq *rq = this_rq(); |
3334 | ||
3335 | if (user_tick) | |
a42548a1 | 3336 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3337 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3338 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3339 | one_jiffy_scaled); |
3340 | else | |
a42548a1 | 3341 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3342 | } |
3343 | ||
3344 | /* | |
3345 | * Account multiple ticks of steal time. | |
3346 | * @p: the process from which the cpu time has been stolen | |
3347 | * @ticks: number of stolen ticks | |
3348 | */ | |
3349 | void account_steal_ticks(unsigned long ticks) | |
3350 | { | |
3351 | account_steal_time(jiffies_to_cputime(ticks)); | |
3352 | } | |
3353 | ||
3354 | /* | |
3355 | * Account multiple ticks of idle time. | |
3356 | * @ticks: number of stolen ticks | |
3357 | */ | |
3358 | void account_idle_ticks(unsigned long ticks) | |
3359 | { | |
3360 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3361 | } |
3362 | ||
79741dd3 MS |
3363 | #endif |
3364 | ||
49048622 BS |
3365 | /* |
3366 | * Use precise platform statistics if available: | |
3367 | */ | |
3368 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3369 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3370 | { |
d99ca3b9 HS |
3371 | *ut = p->utime; |
3372 | *st = p->stime; | |
49048622 BS |
3373 | } |
3374 | ||
0cf55e1e | 3375 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3376 | { |
0cf55e1e HS |
3377 | struct task_cputime cputime; |
3378 | ||
3379 | thread_group_cputime(p, &cputime); | |
3380 | ||
3381 | *ut = cputime.utime; | |
3382 | *st = cputime.stime; | |
49048622 BS |
3383 | } |
3384 | #else | |
761b1d26 HS |
3385 | |
3386 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3387 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3388 | #endif |
3389 | ||
d180c5bc | 3390 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3391 | { |
d99ca3b9 | 3392 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3393 | |
3394 | /* | |
3395 | * Use CFS's precise accounting: | |
3396 | */ | |
d180c5bc | 3397 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3398 | |
3399 | if (total) { | |
d180c5bc HS |
3400 | u64 temp; |
3401 | ||
3402 | temp = (u64)(rtime * utime); | |
49048622 | 3403 | do_div(temp, total); |
d180c5bc HS |
3404 | utime = (cputime_t)temp; |
3405 | } else | |
3406 | utime = rtime; | |
49048622 | 3407 | |
d180c5bc HS |
3408 | /* |
3409 | * Compare with previous values, to keep monotonicity: | |
3410 | */ | |
761b1d26 | 3411 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3412 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3413 | |
d99ca3b9 HS |
3414 | *ut = p->prev_utime; |
3415 | *st = p->prev_stime; | |
49048622 BS |
3416 | } |
3417 | ||
0cf55e1e HS |
3418 | /* |
3419 | * Must be called with siglock held. | |
3420 | */ | |
3421 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3422 | { |
0cf55e1e HS |
3423 | struct signal_struct *sig = p->signal; |
3424 | struct task_cputime cputime; | |
3425 | cputime_t rtime, utime, total; | |
49048622 | 3426 | |
0cf55e1e | 3427 | thread_group_cputime(p, &cputime); |
49048622 | 3428 | |
0cf55e1e HS |
3429 | total = cputime_add(cputime.utime, cputime.stime); |
3430 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3431 | |
0cf55e1e HS |
3432 | if (total) { |
3433 | u64 temp; | |
49048622 | 3434 | |
0cf55e1e HS |
3435 | temp = (u64)(rtime * cputime.utime); |
3436 | do_div(temp, total); | |
3437 | utime = (cputime_t)temp; | |
3438 | } else | |
3439 | utime = rtime; | |
3440 | ||
3441 | sig->prev_utime = max(sig->prev_utime, utime); | |
3442 | sig->prev_stime = max(sig->prev_stime, | |
3443 | cputime_sub(rtime, sig->prev_utime)); | |
3444 | ||
3445 | *ut = sig->prev_utime; | |
3446 | *st = sig->prev_stime; | |
49048622 | 3447 | } |
49048622 | 3448 | #endif |
49048622 | 3449 | |
7835b98b CL |
3450 | /* |
3451 | * This function gets called by the timer code, with HZ frequency. | |
3452 | * We call it with interrupts disabled. | |
3453 | * | |
3454 | * It also gets called by the fork code, when changing the parent's | |
3455 | * timeslices. | |
3456 | */ | |
3457 | void scheduler_tick(void) | |
3458 | { | |
7835b98b CL |
3459 | int cpu = smp_processor_id(); |
3460 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3461 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3462 | |
3463 | sched_clock_tick(); | |
dd41f596 | 3464 | |
05fa785c | 3465 | raw_spin_lock(&rq->lock); |
3e51f33f | 3466 | update_rq_clock(rq); |
f1a438d8 | 3467 | update_cpu_load(rq); |
fa85ae24 | 3468 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3469 | raw_spin_unlock(&rq->lock); |
7835b98b | 3470 | |
49f47433 | 3471 | perf_event_task_tick(curr); |
e220d2dc | 3472 | |
e418e1c2 | 3473 | #ifdef CONFIG_SMP |
dd41f596 IM |
3474 | rq->idle_at_tick = idle_cpu(cpu); |
3475 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3476 | #endif |
1da177e4 LT |
3477 | } |
3478 | ||
132380a0 | 3479 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3480 | { |
3481 | if (in_lock_functions(addr)) { | |
3482 | addr = CALLER_ADDR2; | |
3483 | if (in_lock_functions(addr)) | |
3484 | addr = CALLER_ADDR3; | |
3485 | } | |
3486 | return addr; | |
3487 | } | |
1da177e4 | 3488 | |
7e49fcce SR |
3489 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3490 | defined(CONFIG_PREEMPT_TRACER)) | |
3491 | ||
43627582 | 3492 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3493 | { |
6cd8a4bb | 3494 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3495 | /* |
3496 | * Underflow? | |
3497 | */ | |
9a11b49a IM |
3498 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3499 | return; | |
6cd8a4bb | 3500 | #endif |
1da177e4 | 3501 | preempt_count() += val; |
6cd8a4bb | 3502 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3503 | /* |
3504 | * Spinlock count overflowing soon? | |
3505 | */ | |
33859f7f MOS |
3506 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3507 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3508 | #endif |
3509 | if (preempt_count() == val) | |
3510 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3511 | } |
3512 | EXPORT_SYMBOL(add_preempt_count); | |
3513 | ||
43627582 | 3514 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3515 | { |
6cd8a4bb | 3516 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3517 | /* |
3518 | * Underflow? | |
3519 | */ | |
01e3eb82 | 3520 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3521 | return; |
1da177e4 LT |
3522 | /* |
3523 | * Is the spinlock portion underflowing? | |
3524 | */ | |
9a11b49a IM |
3525 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3526 | !(preempt_count() & PREEMPT_MASK))) | |
3527 | return; | |
6cd8a4bb | 3528 | #endif |
9a11b49a | 3529 | |
6cd8a4bb SR |
3530 | if (preempt_count() == val) |
3531 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3532 | preempt_count() -= val; |
3533 | } | |
3534 | EXPORT_SYMBOL(sub_preempt_count); | |
3535 | ||
3536 | #endif | |
3537 | ||
3538 | /* | |
dd41f596 | 3539 | * Print scheduling while atomic bug: |
1da177e4 | 3540 | */ |
dd41f596 | 3541 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3542 | { |
838225b4 SS |
3543 | struct pt_regs *regs = get_irq_regs(); |
3544 | ||
3df0fc5b PZ |
3545 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3546 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3547 | |
dd41f596 | 3548 | debug_show_held_locks(prev); |
e21f5b15 | 3549 | print_modules(); |
dd41f596 IM |
3550 | if (irqs_disabled()) |
3551 | print_irqtrace_events(prev); | |
838225b4 SS |
3552 | |
3553 | if (regs) | |
3554 | show_regs(regs); | |
3555 | else | |
3556 | dump_stack(); | |
dd41f596 | 3557 | } |
1da177e4 | 3558 | |
dd41f596 IM |
3559 | /* |
3560 | * Various schedule()-time debugging checks and statistics: | |
3561 | */ | |
3562 | static inline void schedule_debug(struct task_struct *prev) | |
3563 | { | |
1da177e4 | 3564 | /* |
41a2d6cf | 3565 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3566 | * schedule() atomically, we ignore that path for now. |
3567 | * Otherwise, whine if we are scheduling when we should not be. | |
3568 | */ | |
3f33a7ce | 3569 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3570 | __schedule_bug(prev); |
3571 | ||
1da177e4 LT |
3572 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3573 | ||
2d72376b | 3574 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3575 | #ifdef CONFIG_SCHEDSTATS |
3576 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3577 | schedstat_inc(this_rq(), bkl_count); |
3578 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3579 | } |
3580 | #endif | |
dd41f596 IM |
3581 | } |
3582 | ||
6cecd084 | 3583 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3584 | { |
a64692a3 MG |
3585 | if (prev->se.on_rq) |
3586 | update_rq_clock(rq); | |
3587 | rq->skip_clock_update = 0; | |
6cecd084 | 3588 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3589 | } |
3590 | ||
dd41f596 IM |
3591 | /* |
3592 | * Pick up the highest-prio task: | |
3593 | */ | |
3594 | static inline struct task_struct * | |
b67802ea | 3595 | pick_next_task(struct rq *rq) |
dd41f596 | 3596 | { |
5522d5d5 | 3597 | const struct sched_class *class; |
dd41f596 | 3598 | struct task_struct *p; |
1da177e4 LT |
3599 | |
3600 | /* | |
dd41f596 IM |
3601 | * Optimization: we know that if all tasks are in |
3602 | * the fair class we can call that function directly: | |
1da177e4 | 3603 | */ |
dd41f596 | 3604 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3605 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3606 | if (likely(p)) |
3607 | return p; | |
1da177e4 LT |
3608 | } |
3609 | ||
dd41f596 IM |
3610 | class = sched_class_highest; |
3611 | for ( ; ; ) { | |
fb8d4724 | 3612 | p = class->pick_next_task(rq); |
dd41f596 IM |
3613 | if (p) |
3614 | return p; | |
3615 | /* | |
3616 | * Will never be NULL as the idle class always | |
3617 | * returns a non-NULL p: | |
3618 | */ | |
3619 | class = class->next; | |
3620 | } | |
3621 | } | |
1da177e4 | 3622 | |
dd41f596 IM |
3623 | /* |
3624 | * schedule() is the main scheduler function. | |
3625 | */ | |
ff743345 | 3626 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3627 | { |
3628 | struct task_struct *prev, *next; | |
67ca7bde | 3629 | unsigned long *switch_count; |
dd41f596 | 3630 | struct rq *rq; |
31656519 | 3631 | int cpu; |
dd41f596 | 3632 | |
ff743345 PZ |
3633 | need_resched: |
3634 | preempt_disable(); | |
dd41f596 IM |
3635 | cpu = smp_processor_id(); |
3636 | rq = cpu_rq(cpu); | |
25502a6c | 3637 | rcu_note_context_switch(cpu); |
dd41f596 IM |
3638 | prev = rq->curr; |
3639 | switch_count = &prev->nivcsw; | |
3640 | ||
3641 | release_kernel_lock(prev); | |
3642 | need_resched_nonpreemptible: | |
3643 | ||
3644 | schedule_debug(prev); | |
1da177e4 | 3645 | |
31656519 | 3646 | if (sched_feat(HRTICK)) |
f333fdc9 | 3647 | hrtick_clear(rq); |
8f4d37ec | 3648 | |
05fa785c | 3649 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3650 | clear_tsk_need_resched(prev); |
1da177e4 | 3651 | |
1da177e4 | 3652 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3653 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3654 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
3655 | } else { |
3656 | /* | |
3657 | * If a worker is going to sleep, notify and | |
3658 | * ask workqueue whether it wants to wake up a | |
3659 | * task to maintain concurrency. If so, wake | |
3660 | * up the task. | |
3661 | */ | |
3662 | if (prev->flags & PF_WQ_WORKER) { | |
3663 | struct task_struct *to_wakeup; | |
3664 | ||
3665 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3666 | if (to_wakeup) | |
3667 | try_to_wake_up_local(to_wakeup); | |
3668 | } | |
371fd7e7 | 3669 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 3670 | } |
dd41f596 | 3671 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3672 | } |
3673 | ||
3f029d3c | 3674 | pre_schedule(rq, prev); |
f65eda4f | 3675 | |
dd41f596 | 3676 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3677 | idle_balance(cpu, rq); |
1da177e4 | 3678 | |
df1c99d4 | 3679 | put_prev_task(rq, prev); |
b67802ea | 3680 | next = pick_next_task(rq); |
1da177e4 | 3681 | |
1da177e4 | 3682 | if (likely(prev != next)) { |
673a90a1 | 3683 | sched_info_switch(prev, next); |
49f47433 | 3684 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3685 | |
1da177e4 LT |
3686 | rq->nr_switches++; |
3687 | rq->curr = next; | |
3688 | ++*switch_count; | |
3689 | ||
dd41f596 | 3690 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec PZ |
3691 | /* |
3692 | * the context switch might have flipped the stack from under | |
3693 | * us, hence refresh the local variables. | |
3694 | */ | |
3695 | cpu = smp_processor_id(); | |
3696 | rq = cpu_rq(cpu); | |
1da177e4 | 3697 | } else |
05fa785c | 3698 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3699 | |
3f029d3c | 3700 | post_schedule(rq); |
1da177e4 | 3701 | |
6d558c3a YZ |
3702 | if (unlikely(reacquire_kernel_lock(current) < 0)) { |
3703 | prev = rq->curr; | |
3704 | switch_count = &prev->nivcsw; | |
1da177e4 | 3705 | goto need_resched_nonpreemptible; |
6d558c3a | 3706 | } |
8f4d37ec | 3707 | |
1da177e4 | 3708 | preempt_enable_no_resched(); |
ff743345 | 3709 | if (need_resched()) |
1da177e4 LT |
3710 | goto need_resched; |
3711 | } | |
1da177e4 LT |
3712 | EXPORT_SYMBOL(schedule); |
3713 | ||
c08f7829 | 3714 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3715 | /* |
3716 | * Look out! "owner" is an entirely speculative pointer | |
3717 | * access and not reliable. | |
3718 | */ | |
3719 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3720 | { | |
3721 | unsigned int cpu; | |
3722 | struct rq *rq; | |
3723 | ||
3724 | if (!sched_feat(OWNER_SPIN)) | |
3725 | return 0; | |
3726 | ||
3727 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3728 | /* | |
3729 | * Need to access the cpu field knowing that | |
3730 | * DEBUG_PAGEALLOC could have unmapped it if | |
3731 | * the mutex owner just released it and exited. | |
3732 | */ | |
3733 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 3734 | return 0; |
0d66bf6d PZ |
3735 | #else |
3736 | cpu = owner->cpu; | |
3737 | #endif | |
3738 | ||
3739 | /* | |
3740 | * Even if the access succeeded (likely case), | |
3741 | * the cpu field may no longer be valid. | |
3742 | */ | |
3743 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 3744 | return 0; |
0d66bf6d PZ |
3745 | |
3746 | /* | |
3747 | * We need to validate that we can do a | |
3748 | * get_cpu() and that we have the percpu area. | |
3749 | */ | |
3750 | if (!cpu_online(cpu)) | |
4b402210 | 3751 | return 0; |
0d66bf6d PZ |
3752 | |
3753 | rq = cpu_rq(cpu); | |
3754 | ||
3755 | for (;;) { | |
3756 | /* | |
3757 | * Owner changed, break to re-assess state. | |
3758 | */ | |
3759 | if (lock->owner != owner) | |
3760 | break; | |
3761 | ||
3762 | /* | |
3763 | * Is that owner really running on that cpu? | |
3764 | */ | |
3765 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3766 | return 0; | |
3767 | ||
3768 | cpu_relax(); | |
3769 | } | |
4b402210 | 3770 | |
0d66bf6d PZ |
3771 | return 1; |
3772 | } | |
3773 | #endif | |
3774 | ||
1da177e4 LT |
3775 | #ifdef CONFIG_PREEMPT |
3776 | /* | |
2ed6e34f | 3777 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3778 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3779 | * occur there and call schedule directly. |
3780 | */ | |
3781 | asmlinkage void __sched preempt_schedule(void) | |
3782 | { | |
3783 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3784 | |
1da177e4 LT |
3785 | /* |
3786 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3787 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3788 | */ |
beed33a8 | 3789 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3790 | return; |
3791 | ||
3a5c359a AK |
3792 | do { |
3793 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3794 | schedule(); |
3a5c359a | 3795 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3796 | |
3a5c359a AK |
3797 | /* |
3798 | * Check again in case we missed a preemption opportunity | |
3799 | * between schedule and now. | |
3800 | */ | |
3801 | barrier(); | |
5ed0cec0 | 3802 | } while (need_resched()); |
1da177e4 | 3803 | } |
1da177e4 LT |
3804 | EXPORT_SYMBOL(preempt_schedule); |
3805 | ||
3806 | /* | |
2ed6e34f | 3807 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3808 | * off of irq context. |
3809 | * Note, that this is called and return with irqs disabled. This will | |
3810 | * protect us against recursive calling from irq. | |
3811 | */ | |
3812 | asmlinkage void __sched preempt_schedule_irq(void) | |
3813 | { | |
3814 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3815 | |
2ed6e34f | 3816 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3817 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3818 | ||
3a5c359a AK |
3819 | do { |
3820 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3821 | local_irq_enable(); |
3822 | schedule(); | |
3823 | local_irq_disable(); | |
3a5c359a | 3824 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3825 | |
3a5c359a AK |
3826 | /* |
3827 | * Check again in case we missed a preemption opportunity | |
3828 | * between schedule and now. | |
3829 | */ | |
3830 | barrier(); | |
5ed0cec0 | 3831 | } while (need_resched()); |
1da177e4 LT |
3832 | } |
3833 | ||
3834 | #endif /* CONFIG_PREEMPT */ | |
3835 | ||
63859d4f | 3836 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3837 | void *key) |
1da177e4 | 3838 | { |
63859d4f | 3839 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3840 | } |
1da177e4 LT |
3841 | EXPORT_SYMBOL(default_wake_function); |
3842 | ||
3843 | /* | |
41a2d6cf IM |
3844 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3845 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3846 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3847 | * | |
3848 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3849 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3850 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3851 | */ | |
78ddb08f | 3852 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3853 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3854 | { |
2e45874c | 3855 | wait_queue_t *curr, *next; |
1da177e4 | 3856 | |
2e45874c | 3857 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3858 | unsigned flags = curr->flags; |
3859 | ||
63859d4f | 3860 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3861 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3862 | break; |
3863 | } | |
3864 | } | |
3865 | ||
3866 | /** | |
3867 | * __wake_up - wake up threads blocked on a waitqueue. | |
3868 | * @q: the waitqueue | |
3869 | * @mode: which threads | |
3870 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3871 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3872 | * |
3873 | * It may be assumed that this function implies a write memory barrier before | |
3874 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3875 | */ |
7ad5b3a5 | 3876 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3877 | int nr_exclusive, void *key) |
1da177e4 LT |
3878 | { |
3879 | unsigned long flags; | |
3880 | ||
3881 | spin_lock_irqsave(&q->lock, flags); | |
3882 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3883 | spin_unlock_irqrestore(&q->lock, flags); | |
3884 | } | |
1da177e4 LT |
3885 | EXPORT_SYMBOL(__wake_up); |
3886 | ||
3887 | /* | |
3888 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3889 | */ | |
7ad5b3a5 | 3890 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3891 | { |
3892 | __wake_up_common(q, mode, 1, 0, NULL); | |
3893 | } | |
22c43c81 | 3894 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 3895 | |
4ede816a DL |
3896 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
3897 | { | |
3898 | __wake_up_common(q, mode, 1, 0, key); | |
3899 | } | |
3900 | ||
1da177e4 | 3901 | /** |
4ede816a | 3902 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
3903 | * @q: the waitqueue |
3904 | * @mode: which threads | |
3905 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 3906 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
3907 | * |
3908 | * The sync wakeup differs that the waker knows that it will schedule | |
3909 | * away soon, so while the target thread will be woken up, it will not | |
3910 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
3911 | * with each other. This can prevent needless bouncing between CPUs. | |
3912 | * | |
3913 | * On UP it can prevent extra preemption. | |
50fa610a DH |
3914 | * |
3915 | * It may be assumed that this function implies a write memory barrier before | |
3916 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3917 | */ |
4ede816a DL |
3918 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
3919 | int nr_exclusive, void *key) | |
1da177e4 LT |
3920 | { |
3921 | unsigned long flags; | |
7d478721 | 3922 | int wake_flags = WF_SYNC; |
1da177e4 LT |
3923 | |
3924 | if (unlikely(!q)) | |
3925 | return; | |
3926 | ||
3927 | if (unlikely(!nr_exclusive)) | |
7d478721 | 3928 | wake_flags = 0; |
1da177e4 LT |
3929 | |
3930 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 3931 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
3932 | spin_unlock_irqrestore(&q->lock, flags); |
3933 | } | |
4ede816a DL |
3934 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
3935 | ||
3936 | /* | |
3937 | * __wake_up_sync - see __wake_up_sync_key() | |
3938 | */ | |
3939 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
3940 | { | |
3941 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
3942 | } | |
1da177e4 LT |
3943 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
3944 | ||
65eb3dc6 KD |
3945 | /** |
3946 | * complete: - signals a single thread waiting on this completion | |
3947 | * @x: holds the state of this particular completion | |
3948 | * | |
3949 | * This will wake up a single thread waiting on this completion. Threads will be | |
3950 | * awakened in the same order in which they were queued. | |
3951 | * | |
3952 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
3953 | * |
3954 | * It may be assumed that this function implies a write memory barrier before | |
3955 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3956 | */ |
b15136e9 | 3957 | void complete(struct completion *x) |
1da177e4 LT |
3958 | { |
3959 | unsigned long flags; | |
3960 | ||
3961 | spin_lock_irqsave(&x->wait.lock, flags); | |
3962 | x->done++; | |
d9514f6c | 3963 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
3964 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3965 | } | |
3966 | EXPORT_SYMBOL(complete); | |
3967 | ||
65eb3dc6 KD |
3968 | /** |
3969 | * complete_all: - signals all threads waiting on this completion | |
3970 | * @x: holds the state of this particular completion | |
3971 | * | |
3972 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
3973 | * |
3974 | * It may be assumed that this function implies a write memory barrier before | |
3975 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 3976 | */ |
b15136e9 | 3977 | void complete_all(struct completion *x) |
1da177e4 LT |
3978 | { |
3979 | unsigned long flags; | |
3980 | ||
3981 | spin_lock_irqsave(&x->wait.lock, flags); | |
3982 | x->done += UINT_MAX/2; | |
d9514f6c | 3983 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
3984 | spin_unlock_irqrestore(&x->wait.lock, flags); |
3985 | } | |
3986 | EXPORT_SYMBOL(complete_all); | |
3987 | ||
8cbbe86d AK |
3988 | static inline long __sched |
3989 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 3990 | { |
1da177e4 LT |
3991 | if (!x->done) { |
3992 | DECLARE_WAITQUEUE(wait, current); | |
3993 | ||
a93d2f17 | 3994 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 3995 | do { |
94d3d824 | 3996 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
3997 | timeout = -ERESTARTSYS; |
3998 | break; | |
8cbbe86d AK |
3999 | } |
4000 | __set_current_state(state); | |
1da177e4 LT |
4001 | spin_unlock_irq(&x->wait.lock); |
4002 | timeout = schedule_timeout(timeout); | |
4003 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4004 | } while (!x->done && timeout); |
1da177e4 | 4005 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4006 | if (!x->done) |
4007 | return timeout; | |
1da177e4 LT |
4008 | } |
4009 | x->done--; | |
ea71a546 | 4010 | return timeout ?: 1; |
1da177e4 | 4011 | } |
1da177e4 | 4012 | |
8cbbe86d AK |
4013 | static long __sched |
4014 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4015 | { |
1da177e4 LT |
4016 | might_sleep(); |
4017 | ||
4018 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4019 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4020 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4021 | return timeout; |
4022 | } | |
1da177e4 | 4023 | |
65eb3dc6 KD |
4024 | /** |
4025 | * wait_for_completion: - waits for completion of a task | |
4026 | * @x: holds the state of this particular completion | |
4027 | * | |
4028 | * This waits to be signaled for completion of a specific task. It is NOT | |
4029 | * interruptible and there is no timeout. | |
4030 | * | |
4031 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4032 | * and interrupt capability. Also see complete(). | |
4033 | */ | |
b15136e9 | 4034 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4035 | { |
4036 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4037 | } |
8cbbe86d | 4038 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4039 | |
65eb3dc6 KD |
4040 | /** |
4041 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4042 | * @x: holds the state of this particular completion | |
4043 | * @timeout: timeout value in jiffies | |
4044 | * | |
4045 | * This waits for either a completion of a specific task to be signaled or for a | |
4046 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4047 | * interruptible. | |
4048 | */ | |
b15136e9 | 4049 | unsigned long __sched |
8cbbe86d | 4050 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4051 | { |
8cbbe86d | 4052 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4053 | } |
8cbbe86d | 4054 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4055 | |
65eb3dc6 KD |
4056 | /** |
4057 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4058 | * @x: holds the state of this particular completion | |
4059 | * | |
4060 | * This waits for completion of a specific task to be signaled. It is | |
4061 | * interruptible. | |
4062 | */ | |
8cbbe86d | 4063 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4064 | { |
51e97990 AK |
4065 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4066 | if (t == -ERESTARTSYS) | |
4067 | return t; | |
4068 | return 0; | |
0fec171c | 4069 | } |
8cbbe86d | 4070 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4071 | |
65eb3dc6 KD |
4072 | /** |
4073 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4074 | * @x: holds the state of this particular completion | |
4075 | * @timeout: timeout value in jiffies | |
4076 | * | |
4077 | * This waits for either a completion of a specific task to be signaled or for a | |
4078 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4079 | */ | |
b15136e9 | 4080 | unsigned long __sched |
8cbbe86d AK |
4081 | wait_for_completion_interruptible_timeout(struct completion *x, |
4082 | unsigned long timeout) | |
0fec171c | 4083 | { |
8cbbe86d | 4084 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4085 | } |
8cbbe86d | 4086 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4087 | |
65eb3dc6 KD |
4088 | /** |
4089 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4090 | * @x: holds the state of this particular completion | |
4091 | * | |
4092 | * This waits to be signaled for completion of a specific task. It can be | |
4093 | * interrupted by a kill signal. | |
4094 | */ | |
009e577e MW |
4095 | int __sched wait_for_completion_killable(struct completion *x) |
4096 | { | |
4097 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4098 | if (t == -ERESTARTSYS) | |
4099 | return t; | |
4100 | return 0; | |
4101 | } | |
4102 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4103 | ||
0aa12fb4 SW |
4104 | /** |
4105 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4106 | * @x: holds the state of this particular completion | |
4107 | * @timeout: timeout value in jiffies | |
4108 | * | |
4109 | * This waits for either a completion of a specific task to be | |
4110 | * signaled or for a specified timeout to expire. It can be | |
4111 | * interrupted by a kill signal. The timeout is in jiffies. | |
4112 | */ | |
4113 | unsigned long __sched | |
4114 | wait_for_completion_killable_timeout(struct completion *x, | |
4115 | unsigned long timeout) | |
4116 | { | |
4117 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4118 | } | |
4119 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4120 | ||
be4de352 DC |
4121 | /** |
4122 | * try_wait_for_completion - try to decrement a completion without blocking | |
4123 | * @x: completion structure | |
4124 | * | |
4125 | * Returns: 0 if a decrement cannot be done without blocking | |
4126 | * 1 if a decrement succeeded. | |
4127 | * | |
4128 | * If a completion is being used as a counting completion, | |
4129 | * attempt to decrement the counter without blocking. This | |
4130 | * enables us to avoid waiting if the resource the completion | |
4131 | * is protecting is not available. | |
4132 | */ | |
4133 | bool try_wait_for_completion(struct completion *x) | |
4134 | { | |
7539a3b3 | 4135 | unsigned long flags; |
be4de352 DC |
4136 | int ret = 1; |
4137 | ||
7539a3b3 | 4138 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4139 | if (!x->done) |
4140 | ret = 0; | |
4141 | else | |
4142 | x->done--; | |
7539a3b3 | 4143 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4144 | return ret; |
4145 | } | |
4146 | EXPORT_SYMBOL(try_wait_for_completion); | |
4147 | ||
4148 | /** | |
4149 | * completion_done - Test to see if a completion has any waiters | |
4150 | * @x: completion structure | |
4151 | * | |
4152 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4153 | * 1 if there are no waiters. | |
4154 | * | |
4155 | */ | |
4156 | bool completion_done(struct completion *x) | |
4157 | { | |
7539a3b3 | 4158 | unsigned long flags; |
be4de352 DC |
4159 | int ret = 1; |
4160 | ||
7539a3b3 | 4161 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4162 | if (!x->done) |
4163 | ret = 0; | |
7539a3b3 | 4164 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4165 | return ret; |
4166 | } | |
4167 | EXPORT_SYMBOL(completion_done); | |
4168 | ||
8cbbe86d AK |
4169 | static long __sched |
4170 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4171 | { |
0fec171c IM |
4172 | unsigned long flags; |
4173 | wait_queue_t wait; | |
4174 | ||
4175 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4176 | |
8cbbe86d | 4177 | __set_current_state(state); |
1da177e4 | 4178 | |
8cbbe86d AK |
4179 | spin_lock_irqsave(&q->lock, flags); |
4180 | __add_wait_queue(q, &wait); | |
4181 | spin_unlock(&q->lock); | |
4182 | timeout = schedule_timeout(timeout); | |
4183 | spin_lock_irq(&q->lock); | |
4184 | __remove_wait_queue(q, &wait); | |
4185 | spin_unlock_irqrestore(&q->lock, flags); | |
4186 | ||
4187 | return timeout; | |
4188 | } | |
4189 | ||
4190 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4191 | { | |
4192 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4193 | } |
1da177e4 LT |
4194 | EXPORT_SYMBOL(interruptible_sleep_on); |
4195 | ||
0fec171c | 4196 | long __sched |
95cdf3b7 | 4197 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4198 | { |
8cbbe86d | 4199 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4200 | } |
1da177e4 LT |
4201 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4202 | ||
0fec171c | 4203 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4204 | { |
8cbbe86d | 4205 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4206 | } |
1da177e4 LT |
4207 | EXPORT_SYMBOL(sleep_on); |
4208 | ||
0fec171c | 4209 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4210 | { |
8cbbe86d | 4211 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4212 | } |
1da177e4 LT |
4213 | EXPORT_SYMBOL(sleep_on_timeout); |
4214 | ||
b29739f9 IM |
4215 | #ifdef CONFIG_RT_MUTEXES |
4216 | ||
4217 | /* | |
4218 | * rt_mutex_setprio - set the current priority of a task | |
4219 | * @p: task | |
4220 | * @prio: prio value (kernel-internal form) | |
4221 | * | |
4222 | * This function changes the 'effective' priority of a task. It does | |
4223 | * not touch ->normal_prio like __setscheduler(). | |
4224 | * | |
4225 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4226 | */ | |
36c8b586 | 4227 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4228 | { |
4229 | unsigned long flags; | |
83b699ed | 4230 | int oldprio, on_rq, running; |
70b97a7f | 4231 | struct rq *rq; |
83ab0aa0 | 4232 | const struct sched_class *prev_class; |
b29739f9 IM |
4233 | |
4234 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4235 | ||
4236 | rq = task_rq_lock(p, &flags); | |
4237 | ||
d5f9f942 | 4238 | oldprio = p->prio; |
83ab0aa0 | 4239 | prev_class = p->sched_class; |
dd41f596 | 4240 | on_rq = p->se.on_rq; |
051a1d1a | 4241 | running = task_current(rq, p); |
0e1f3483 | 4242 | if (on_rq) |
69be72c1 | 4243 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4244 | if (running) |
4245 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4246 | |
4247 | if (rt_prio(prio)) | |
4248 | p->sched_class = &rt_sched_class; | |
4249 | else | |
4250 | p->sched_class = &fair_sched_class; | |
4251 | ||
b29739f9 IM |
4252 | p->prio = prio; |
4253 | ||
0e1f3483 HS |
4254 | if (running) |
4255 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4256 | if (on_rq) { |
371fd7e7 | 4257 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4258 | |
4259 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4260 | } |
4261 | task_rq_unlock(rq, &flags); | |
4262 | } | |
4263 | ||
4264 | #endif | |
4265 | ||
36c8b586 | 4266 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4267 | { |
dd41f596 | 4268 | int old_prio, delta, on_rq; |
1da177e4 | 4269 | unsigned long flags; |
70b97a7f | 4270 | struct rq *rq; |
1da177e4 LT |
4271 | |
4272 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4273 | return; | |
4274 | /* | |
4275 | * We have to be careful, if called from sys_setpriority(), | |
4276 | * the task might be in the middle of scheduling on another CPU. | |
4277 | */ | |
4278 | rq = task_rq_lock(p, &flags); | |
4279 | /* | |
4280 | * The RT priorities are set via sched_setscheduler(), but we still | |
4281 | * allow the 'normal' nice value to be set - but as expected | |
4282 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4283 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4284 | */ |
e05606d3 | 4285 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4286 | p->static_prio = NICE_TO_PRIO(nice); |
4287 | goto out_unlock; | |
4288 | } | |
dd41f596 | 4289 | on_rq = p->se.on_rq; |
c09595f6 | 4290 | if (on_rq) |
69be72c1 | 4291 | dequeue_task(rq, p, 0); |
1da177e4 | 4292 | |
1da177e4 | 4293 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4294 | set_load_weight(p); |
b29739f9 IM |
4295 | old_prio = p->prio; |
4296 | p->prio = effective_prio(p); | |
4297 | delta = p->prio - old_prio; | |
1da177e4 | 4298 | |
dd41f596 | 4299 | if (on_rq) { |
371fd7e7 | 4300 | enqueue_task(rq, p, 0); |
1da177e4 | 4301 | /* |
d5f9f942 AM |
4302 | * If the task increased its priority or is running and |
4303 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4304 | */ |
d5f9f942 | 4305 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4306 | resched_task(rq->curr); |
4307 | } | |
4308 | out_unlock: | |
4309 | task_rq_unlock(rq, &flags); | |
4310 | } | |
1da177e4 LT |
4311 | EXPORT_SYMBOL(set_user_nice); |
4312 | ||
e43379f1 MM |
4313 | /* |
4314 | * can_nice - check if a task can reduce its nice value | |
4315 | * @p: task | |
4316 | * @nice: nice value | |
4317 | */ | |
36c8b586 | 4318 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4319 | { |
024f4747 MM |
4320 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4321 | int nice_rlim = 20 - nice; | |
48f24c4d | 4322 | |
78d7d407 | 4323 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4324 | capable(CAP_SYS_NICE)); |
4325 | } | |
4326 | ||
1da177e4 LT |
4327 | #ifdef __ARCH_WANT_SYS_NICE |
4328 | ||
4329 | /* | |
4330 | * sys_nice - change the priority of the current process. | |
4331 | * @increment: priority increment | |
4332 | * | |
4333 | * sys_setpriority is a more generic, but much slower function that | |
4334 | * does similar things. | |
4335 | */ | |
5add95d4 | 4336 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4337 | { |
48f24c4d | 4338 | long nice, retval; |
1da177e4 LT |
4339 | |
4340 | /* | |
4341 | * Setpriority might change our priority at the same moment. | |
4342 | * We don't have to worry. Conceptually one call occurs first | |
4343 | * and we have a single winner. | |
4344 | */ | |
e43379f1 MM |
4345 | if (increment < -40) |
4346 | increment = -40; | |
1da177e4 LT |
4347 | if (increment > 40) |
4348 | increment = 40; | |
4349 | ||
2b8f836f | 4350 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4351 | if (nice < -20) |
4352 | nice = -20; | |
4353 | if (nice > 19) | |
4354 | nice = 19; | |
4355 | ||
e43379f1 MM |
4356 | if (increment < 0 && !can_nice(current, nice)) |
4357 | return -EPERM; | |
4358 | ||
1da177e4 LT |
4359 | retval = security_task_setnice(current, nice); |
4360 | if (retval) | |
4361 | return retval; | |
4362 | ||
4363 | set_user_nice(current, nice); | |
4364 | return 0; | |
4365 | } | |
4366 | ||
4367 | #endif | |
4368 | ||
4369 | /** | |
4370 | * task_prio - return the priority value of a given task. | |
4371 | * @p: the task in question. | |
4372 | * | |
4373 | * This is the priority value as seen by users in /proc. | |
4374 | * RT tasks are offset by -200. Normal tasks are centered | |
4375 | * around 0, value goes from -16 to +15. | |
4376 | */ | |
36c8b586 | 4377 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4378 | { |
4379 | return p->prio - MAX_RT_PRIO; | |
4380 | } | |
4381 | ||
4382 | /** | |
4383 | * task_nice - return the nice value of a given task. | |
4384 | * @p: the task in question. | |
4385 | */ | |
36c8b586 | 4386 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4387 | { |
4388 | return TASK_NICE(p); | |
4389 | } | |
150d8bed | 4390 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4391 | |
4392 | /** | |
4393 | * idle_cpu - is a given cpu idle currently? | |
4394 | * @cpu: the processor in question. | |
4395 | */ | |
4396 | int idle_cpu(int cpu) | |
4397 | { | |
4398 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4399 | } | |
4400 | ||
1da177e4 LT |
4401 | /** |
4402 | * idle_task - return the idle task for a given cpu. | |
4403 | * @cpu: the processor in question. | |
4404 | */ | |
36c8b586 | 4405 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4406 | { |
4407 | return cpu_rq(cpu)->idle; | |
4408 | } | |
4409 | ||
4410 | /** | |
4411 | * find_process_by_pid - find a process with a matching PID value. | |
4412 | * @pid: the pid in question. | |
4413 | */ | |
a9957449 | 4414 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4415 | { |
228ebcbe | 4416 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4417 | } |
4418 | ||
4419 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4420 | static void |
4421 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4422 | { |
dd41f596 | 4423 | BUG_ON(p->se.on_rq); |
48f24c4d | 4424 | |
1da177e4 LT |
4425 | p->policy = policy; |
4426 | p->rt_priority = prio; | |
b29739f9 IM |
4427 | p->normal_prio = normal_prio(p); |
4428 | /* we are holding p->pi_lock already */ | |
4429 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4430 | if (rt_prio(p->prio)) |
4431 | p->sched_class = &rt_sched_class; | |
4432 | else | |
4433 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4434 | set_load_weight(p); |
1da177e4 LT |
4435 | } |
4436 | ||
c69e8d9c DH |
4437 | /* |
4438 | * check the target process has a UID that matches the current process's | |
4439 | */ | |
4440 | static bool check_same_owner(struct task_struct *p) | |
4441 | { | |
4442 | const struct cred *cred = current_cred(), *pcred; | |
4443 | bool match; | |
4444 | ||
4445 | rcu_read_lock(); | |
4446 | pcred = __task_cred(p); | |
4447 | match = (cred->euid == pcred->euid || | |
4448 | cred->euid == pcred->uid); | |
4449 | rcu_read_unlock(); | |
4450 | return match; | |
4451 | } | |
4452 | ||
961ccddd RR |
4453 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4454 | struct sched_param *param, bool user) | |
1da177e4 | 4455 | { |
83b699ed | 4456 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4457 | unsigned long flags; |
83ab0aa0 | 4458 | const struct sched_class *prev_class; |
70b97a7f | 4459 | struct rq *rq; |
ca94c442 | 4460 | int reset_on_fork; |
1da177e4 | 4461 | |
66e5393a SR |
4462 | /* may grab non-irq protected spin_locks */ |
4463 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4464 | recheck: |
4465 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4466 | if (policy < 0) { |
4467 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4468 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4469 | } else { |
4470 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4471 | policy &= ~SCHED_RESET_ON_FORK; | |
4472 | ||
4473 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4474 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4475 | policy != SCHED_IDLE) | |
4476 | return -EINVAL; | |
4477 | } | |
4478 | ||
1da177e4 LT |
4479 | /* |
4480 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4481 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4482 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4483 | */ |
4484 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4485 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4486 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4487 | return -EINVAL; |
e05606d3 | 4488 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4489 | return -EINVAL; |
4490 | ||
37e4ab3f OC |
4491 | /* |
4492 | * Allow unprivileged RT tasks to decrease priority: | |
4493 | */ | |
961ccddd | 4494 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4495 | if (rt_policy(policy)) { |
8dc3e909 | 4496 | unsigned long rlim_rtprio; |
8dc3e909 ON |
4497 | |
4498 | if (!lock_task_sighand(p, &flags)) | |
4499 | return -ESRCH; | |
78d7d407 | 4500 | rlim_rtprio = task_rlimit(p, RLIMIT_RTPRIO); |
8dc3e909 ON |
4501 | unlock_task_sighand(p, &flags); |
4502 | ||
4503 | /* can't set/change the rt policy */ | |
4504 | if (policy != p->policy && !rlim_rtprio) | |
4505 | return -EPERM; | |
4506 | ||
4507 | /* can't increase priority */ | |
4508 | if (param->sched_priority > p->rt_priority && | |
4509 | param->sched_priority > rlim_rtprio) | |
4510 | return -EPERM; | |
4511 | } | |
dd41f596 IM |
4512 | /* |
4513 | * Like positive nice levels, dont allow tasks to | |
4514 | * move out of SCHED_IDLE either: | |
4515 | */ | |
4516 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4517 | return -EPERM; | |
5fe1d75f | 4518 | |
37e4ab3f | 4519 | /* can't change other user's priorities */ |
c69e8d9c | 4520 | if (!check_same_owner(p)) |
37e4ab3f | 4521 | return -EPERM; |
ca94c442 LP |
4522 | |
4523 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4524 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4525 | return -EPERM; | |
37e4ab3f | 4526 | } |
1da177e4 | 4527 | |
725aad24 | 4528 | if (user) { |
725aad24 JF |
4529 | retval = security_task_setscheduler(p, policy, param); |
4530 | if (retval) | |
4531 | return retval; | |
4532 | } | |
4533 | ||
b29739f9 IM |
4534 | /* |
4535 | * make sure no PI-waiters arrive (or leave) while we are | |
4536 | * changing the priority of the task: | |
4537 | */ | |
1d615482 | 4538 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4539 | /* |
4540 | * To be able to change p->policy safely, the apropriate | |
4541 | * runqueue lock must be held. | |
4542 | */ | |
b29739f9 | 4543 | rq = __task_rq_lock(p); |
dc61b1d6 PZ |
4544 | |
4545 | #ifdef CONFIG_RT_GROUP_SCHED | |
4546 | if (user) { | |
4547 | /* | |
4548 | * Do not allow realtime tasks into groups that have no runtime | |
4549 | * assigned. | |
4550 | */ | |
4551 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
4552 | task_group(p)->rt_bandwidth.rt_runtime == 0) { | |
4553 | __task_rq_unlock(rq); | |
4554 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4555 | return -EPERM; | |
4556 | } | |
4557 | } | |
4558 | #endif | |
4559 | ||
1da177e4 LT |
4560 | /* recheck policy now with rq lock held */ |
4561 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4562 | policy = oldpolicy = -1; | |
b29739f9 | 4563 | __task_rq_unlock(rq); |
1d615482 | 4564 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4565 | goto recheck; |
4566 | } | |
dd41f596 | 4567 | on_rq = p->se.on_rq; |
051a1d1a | 4568 | running = task_current(rq, p); |
0e1f3483 | 4569 | if (on_rq) |
2e1cb74a | 4570 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4571 | if (running) |
4572 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4573 | |
ca94c442 LP |
4574 | p->sched_reset_on_fork = reset_on_fork; |
4575 | ||
1da177e4 | 4576 | oldprio = p->prio; |
83ab0aa0 | 4577 | prev_class = p->sched_class; |
dd41f596 | 4578 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4579 | |
0e1f3483 HS |
4580 | if (running) |
4581 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4582 | if (on_rq) { |
4583 | activate_task(rq, p, 0); | |
cb469845 SR |
4584 | |
4585 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4586 | } |
b29739f9 | 4587 | __task_rq_unlock(rq); |
1d615482 | 4588 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4589 | |
95e02ca9 TG |
4590 | rt_mutex_adjust_pi(p); |
4591 | ||
1da177e4 LT |
4592 | return 0; |
4593 | } | |
961ccddd RR |
4594 | |
4595 | /** | |
4596 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4597 | * @p: the task in question. | |
4598 | * @policy: new policy. | |
4599 | * @param: structure containing the new RT priority. | |
4600 | * | |
4601 | * NOTE that the task may be already dead. | |
4602 | */ | |
4603 | int sched_setscheduler(struct task_struct *p, int policy, | |
4604 | struct sched_param *param) | |
4605 | { | |
4606 | return __sched_setscheduler(p, policy, param, true); | |
4607 | } | |
1da177e4 LT |
4608 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4609 | ||
961ccddd RR |
4610 | /** |
4611 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4612 | * @p: the task in question. | |
4613 | * @policy: new policy. | |
4614 | * @param: structure containing the new RT priority. | |
4615 | * | |
4616 | * Just like sched_setscheduler, only don't bother checking if the | |
4617 | * current context has permission. For example, this is needed in | |
4618 | * stop_machine(): we create temporary high priority worker threads, | |
4619 | * but our caller might not have that capability. | |
4620 | */ | |
4621 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4622 | struct sched_param *param) | |
4623 | { | |
4624 | return __sched_setscheduler(p, policy, param, false); | |
4625 | } | |
4626 | ||
95cdf3b7 IM |
4627 | static int |
4628 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4629 | { |
1da177e4 LT |
4630 | struct sched_param lparam; |
4631 | struct task_struct *p; | |
36c8b586 | 4632 | int retval; |
1da177e4 LT |
4633 | |
4634 | if (!param || pid < 0) | |
4635 | return -EINVAL; | |
4636 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4637 | return -EFAULT; | |
5fe1d75f ON |
4638 | |
4639 | rcu_read_lock(); | |
4640 | retval = -ESRCH; | |
1da177e4 | 4641 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4642 | if (p != NULL) |
4643 | retval = sched_setscheduler(p, policy, &lparam); | |
4644 | rcu_read_unlock(); | |
36c8b586 | 4645 | |
1da177e4 LT |
4646 | return retval; |
4647 | } | |
4648 | ||
4649 | /** | |
4650 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4651 | * @pid: the pid in question. | |
4652 | * @policy: new policy. | |
4653 | * @param: structure containing the new RT priority. | |
4654 | */ | |
5add95d4 HC |
4655 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4656 | struct sched_param __user *, param) | |
1da177e4 | 4657 | { |
c21761f1 JB |
4658 | /* negative values for policy are not valid */ |
4659 | if (policy < 0) | |
4660 | return -EINVAL; | |
4661 | ||
1da177e4 LT |
4662 | return do_sched_setscheduler(pid, policy, param); |
4663 | } | |
4664 | ||
4665 | /** | |
4666 | * sys_sched_setparam - set/change the RT priority of a thread | |
4667 | * @pid: the pid in question. | |
4668 | * @param: structure containing the new RT priority. | |
4669 | */ | |
5add95d4 | 4670 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4671 | { |
4672 | return do_sched_setscheduler(pid, -1, param); | |
4673 | } | |
4674 | ||
4675 | /** | |
4676 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4677 | * @pid: the pid in question. | |
4678 | */ | |
5add95d4 | 4679 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4680 | { |
36c8b586 | 4681 | struct task_struct *p; |
3a5c359a | 4682 | int retval; |
1da177e4 LT |
4683 | |
4684 | if (pid < 0) | |
3a5c359a | 4685 | return -EINVAL; |
1da177e4 LT |
4686 | |
4687 | retval = -ESRCH; | |
5fe85be0 | 4688 | rcu_read_lock(); |
1da177e4 LT |
4689 | p = find_process_by_pid(pid); |
4690 | if (p) { | |
4691 | retval = security_task_getscheduler(p); | |
4692 | if (!retval) | |
ca94c442 LP |
4693 | retval = p->policy |
4694 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4695 | } |
5fe85be0 | 4696 | rcu_read_unlock(); |
1da177e4 LT |
4697 | return retval; |
4698 | } | |
4699 | ||
4700 | /** | |
ca94c442 | 4701 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4702 | * @pid: the pid in question. |
4703 | * @param: structure containing the RT priority. | |
4704 | */ | |
5add95d4 | 4705 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4706 | { |
4707 | struct sched_param lp; | |
36c8b586 | 4708 | struct task_struct *p; |
3a5c359a | 4709 | int retval; |
1da177e4 LT |
4710 | |
4711 | if (!param || pid < 0) | |
3a5c359a | 4712 | return -EINVAL; |
1da177e4 | 4713 | |
5fe85be0 | 4714 | rcu_read_lock(); |
1da177e4 LT |
4715 | p = find_process_by_pid(pid); |
4716 | retval = -ESRCH; | |
4717 | if (!p) | |
4718 | goto out_unlock; | |
4719 | ||
4720 | retval = security_task_getscheduler(p); | |
4721 | if (retval) | |
4722 | goto out_unlock; | |
4723 | ||
4724 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4725 | rcu_read_unlock(); |
1da177e4 LT |
4726 | |
4727 | /* | |
4728 | * This one might sleep, we cannot do it with a spinlock held ... | |
4729 | */ | |
4730 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4731 | ||
1da177e4 LT |
4732 | return retval; |
4733 | ||
4734 | out_unlock: | |
5fe85be0 | 4735 | rcu_read_unlock(); |
1da177e4 LT |
4736 | return retval; |
4737 | } | |
4738 | ||
96f874e2 | 4739 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4740 | { |
5a16f3d3 | 4741 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4742 | struct task_struct *p; |
4743 | int retval; | |
1da177e4 | 4744 | |
95402b38 | 4745 | get_online_cpus(); |
23f5d142 | 4746 | rcu_read_lock(); |
1da177e4 LT |
4747 | |
4748 | p = find_process_by_pid(pid); | |
4749 | if (!p) { | |
23f5d142 | 4750 | rcu_read_unlock(); |
95402b38 | 4751 | put_online_cpus(); |
1da177e4 LT |
4752 | return -ESRCH; |
4753 | } | |
4754 | ||
23f5d142 | 4755 | /* Prevent p going away */ |
1da177e4 | 4756 | get_task_struct(p); |
23f5d142 | 4757 | rcu_read_unlock(); |
1da177e4 | 4758 | |
5a16f3d3 RR |
4759 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4760 | retval = -ENOMEM; | |
4761 | goto out_put_task; | |
4762 | } | |
4763 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4764 | retval = -ENOMEM; | |
4765 | goto out_free_cpus_allowed; | |
4766 | } | |
1da177e4 | 4767 | retval = -EPERM; |
c69e8d9c | 4768 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4769 | goto out_unlock; |
4770 | ||
e7834f8f DQ |
4771 | retval = security_task_setscheduler(p, 0, NULL); |
4772 | if (retval) | |
4773 | goto out_unlock; | |
4774 | ||
5a16f3d3 RR |
4775 | cpuset_cpus_allowed(p, cpus_allowed); |
4776 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4777 | again: |
5a16f3d3 | 4778 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4779 | |
8707d8b8 | 4780 | if (!retval) { |
5a16f3d3 RR |
4781 | cpuset_cpus_allowed(p, cpus_allowed); |
4782 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4783 | /* |
4784 | * We must have raced with a concurrent cpuset | |
4785 | * update. Just reset the cpus_allowed to the | |
4786 | * cpuset's cpus_allowed | |
4787 | */ | |
5a16f3d3 | 4788 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4789 | goto again; |
4790 | } | |
4791 | } | |
1da177e4 | 4792 | out_unlock: |
5a16f3d3 RR |
4793 | free_cpumask_var(new_mask); |
4794 | out_free_cpus_allowed: | |
4795 | free_cpumask_var(cpus_allowed); | |
4796 | out_put_task: | |
1da177e4 | 4797 | put_task_struct(p); |
95402b38 | 4798 | put_online_cpus(); |
1da177e4 LT |
4799 | return retval; |
4800 | } | |
4801 | ||
4802 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4803 | struct cpumask *new_mask) |
1da177e4 | 4804 | { |
96f874e2 RR |
4805 | if (len < cpumask_size()) |
4806 | cpumask_clear(new_mask); | |
4807 | else if (len > cpumask_size()) | |
4808 | len = cpumask_size(); | |
4809 | ||
1da177e4 LT |
4810 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4811 | } | |
4812 | ||
4813 | /** | |
4814 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4815 | * @pid: pid of the process | |
4816 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4817 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4818 | */ | |
5add95d4 HC |
4819 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4820 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4821 | { |
5a16f3d3 | 4822 | cpumask_var_t new_mask; |
1da177e4 LT |
4823 | int retval; |
4824 | ||
5a16f3d3 RR |
4825 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4826 | return -ENOMEM; | |
1da177e4 | 4827 | |
5a16f3d3 RR |
4828 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4829 | if (retval == 0) | |
4830 | retval = sched_setaffinity(pid, new_mask); | |
4831 | free_cpumask_var(new_mask); | |
4832 | return retval; | |
1da177e4 LT |
4833 | } |
4834 | ||
96f874e2 | 4835 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4836 | { |
36c8b586 | 4837 | struct task_struct *p; |
31605683 TG |
4838 | unsigned long flags; |
4839 | struct rq *rq; | |
1da177e4 | 4840 | int retval; |
1da177e4 | 4841 | |
95402b38 | 4842 | get_online_cpus(); |
23f5d142 | 4843 | rcu_read_lock(); |
1da177e4 LT |
4844 | |
4845 | retval = -ESRCH; | |
4846 | p = find_process_by_pid(pid); | |
4847 | if (!p) | |
4848 | goto out_unlock; | |
4849 | ||
e7834f8f DQ |
4850 | retval = security_task_getscheduler(p); |
4851 | if (retval) | |
4852 | goto out_unlock; | |
4853 | ||
31605683 | 4854 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4855 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4856 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4857 | |
4858 | out_unlock: | |
23f5d142 | 4859 | rcu_read_unlock(); |
95402b38 | 4860 | put_online_cpus(); |
1da177e4 | 4861 | |
9531b62f | 4862 | return retval; |
1da177e4 LT |
4863 | } |
4864 | ||
4865 | /** | |
4866 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4867 | * @pid: pid of the process | |
4868 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4869 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4870 | */ | |
5add95d4 HC |
4871 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4872 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4873 | { |
4874 | int ret; | |
f17c8607 | 4875 | cpumask_var_t mask; |
1da177e4 | 4876 | |
84fba5ec | 4877 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4878 | return -EINVAL; |
4879 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4880 | return -EINVAL; |
4881 | ||
f17c8607 RR |
4882 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4883 | return -ENOMEM; | |
1da177e4 | 4884 | |
f17c8607 RR |
4885 | ret = sched_getaffinity(pid, mask); |
4886 | if (ret == 0) { | |
8bc037fb | 4887 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4888 | |
4889 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4890 | ret = -EFAULT; |
4891 | else | |
cd3d8031 | 4892 | ret = retlen; |
f17c8607 RR |
4893 | } |
4894 | free_cpumask_var(mask); | |
1da177e4 | 4895 | |
f17c8607 | 4896 | return ret; |
1da177e4 LT |
4897 | } |
4898 | ||
4899 | /** | |
4900 | * sys_sched_yield - yield the current processor to other threads. | |
4901 | * | |
dd41f596 IM |
4902 | * This function yields the current CPU to other tasks. If there are no |
4903 | * other threads running on this CPU then this function will return. | |
1da177e4 | 4904 | */ |
5add95d4 | 4905 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 4906 | { |
70b97a7f | 4907 | struct rq *rq = this_rq_lock(); |
1da177e4 | 4908 | |
2d72376b | 4909 | schedstat_inc(rq, yld_count); |
4530d7ab | 4910 | current->sched_class->yield_task(rq); |
1da177e4 LT |
4911 | |
4912 | /* | |
4913 | * Since we are going to call schedule() anyway, there's | |
4914 | * no need to preempt or enable interrupts: | |
4915 | */ | |
4916 | __release(rq->lock); | |
8a25d5de | 4917 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 4918 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
4919 | preempt_enable_no_resched(); |
4920 | ||
4921 | schedule(); | |
4922 | ||
4923 | return 0; | |
4924 | } | |
4925 | ||
d86ee480 PZ |
4926 | static inline int should_resched(void) |
4927 | { | |
4928 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
4929 | } | |
4930 | ||
e7b38404 | 4931 | static void __cond_resched(void) |
1da177e4 | 4932 | { |
e7aaaa69 FW |
4933 | add_preempt_count(PREEMPT_ACTIVE); |
4934 | schedule(); | |
4935 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
4936 | } |
4937 | ||
02b67cc3 | 4938 | int __sched _cond_resched(void) |
1da177e4 | 4939 | { |
d86ee480 | 4940 | if (should_resched()) { |
1da177e4 LT |
4941 | __cond_resched(); |
4942 | return 1; | |
4943 | } | |
4944 | return 0; | |
4945 | } | |
02b67cc3 | 4946 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
4947 | |
4948 | /* | |
613afbf8 | 4949 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
4950 | * call schedule, and on return reacquire the lock. |
4951 | * | |
41a2d6cf | 4952 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
4953 | * operations here to prevent schedule() from being called twice (once via |
4954 | * spin_unlock(), once by hand). | |
4955 | */ | |
613afbf8 | 4956 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 4957 | { |
d86ee480 | 4958 | int resched = should_resched(); |
6df3cecb JK |
4959 | int ret = 0; |
4960 | ||
f607c668 PZ |
4961 | lockdep_assert_held(lock); |
4962 | ||
95c354fe | 4963 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 4964 | spin_unlock(lock); |
d86ee480 | 4965 | if (resched) |
95c354fe NP |
4966 | __cond_resched(); |
4967 | else | |
4968 | cpu_relax(); | |
6df3cecb | 4969 | ret = 1; |
1da177e4 | 4970 | spin_lock(lock); |
1da177e4 | 4971 | } |
6df3cecb | 4972 | return ret; |
1da177e4 | 4973 | } |
613afbf8 | 4974 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 4975 | |
613afbf8 | 4976 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
4977 | { |
4978 | BUG_ON(!in_softirq()); | |
4979 | ||
d86ee480 | 4980 | if (should_resched()) { |
98d82567 | 4981 | local_bh_enable(); |
1da177e4 LT |
4982 | __cond_resched(); |
4983 | local_bh_disable(); | |
4984 | return 1; | |
4985 | } | |
4986 | return 0; | |
4987 | } | |
613afbf8 | 4988 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 4989 | |
1da177e4 LT |
4990 | /** |
4991 | * yield - yield the current processor to other threads. | |
4992 | * | |
72fd4a35 | 4993 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
4994 | * thread runnable and calls sys_sched_yield(). |
4995 | */ | |
4996 | void __sched yield(void) | |
4997 | { | |
4998 | set_current_state(TASK_RUNNING); | |
4999 | sys_sched_yield(); | |
5000 | } | |
1da177e4 LT |
5001 | EXPORT_SYMBOL(yield); |
5002 | ||
5003 | /* | |
41a2d6cf | 5004 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5005 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5006 | */ |
5007 | void __sched io_schedule(void) | |
5008 | { | |
54d35f29 | 5009 | struct rq *rq = raw_rq(); |
1da177e4 | 5010 | |
0ff92245 | 5011 | delayacct_blkio_start(); |
1da177e4 | 5012 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5013 | current->in_iowait = 1; |
1da177e4 | 5014 | schedule(); |
8f0dfc34 | 5015 | current->in_iowait = 0; |
1da177e4 | 5016 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5017 | delayacct_blkio_end(); |
1da177e4 | 5018 | } |
1da177e4 LT |
5019 | EXPORT_SYMBOL(io_schedule); |
5020 | ||
5021 | long __sched io_schedule_timeout(long timeout) | |
5022 | { | |
54d35f29 | 5023 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5024 | long ret; |
5025 | ||
0ff92245 | 5026 | delayacct_blkio_start(); |
1da177e4 | 5027 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5028 | current->in_iowait = 1; |
1da177e4 | 5029 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5030 | current->in_iowait = 0; |
1da177e4 | 5031 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5032 | delayacct_blkio_end(); |
1da177e4 LT |
5033 | return ret; |
5034 | } | |
5035 | ||
5036 | /** | |
5037 | * sys_sched_get_priority_max - return maximum RT priority. | |
5038 | * @policy: scheduling class. | |
5039 | * | |
5040 | * this syscall returns the maximum rt_priority that can be used | |
5041 | * by a given scheduling class. | |
5042 | */ | |
5add95d4 | 5043 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5044 | { |
5045 | int ret = -EINVAL; | |
5046 | ||
5047 | switch (policy) { | |
5048 | case SCHED_FIFO: | |
5049 | case SCHED_RR: | |
5050 | ret = MAX_USER_RT_PRIO-1; | |
5051 | break; | |
5052 | case SCHED_NORMAL: | |
b0a9499c | 5053 | case SCHED_BATCH: |
dd41f596 | 5054 | case SCHED_IDLE: |
1da177e4 LT |
5055 | ret = 0; |
5056 | break; | |
5057 | } | |
5058 | return ret; | |
5059 | } | |
5060 | ||
5061 | /** | |
5062 | * sys_sched_get_priority_min - return minimum RT priority. | |
5063 | * @policy: scheduling class. | |
5064 | * | |
5065 | * this syscall returns the minimum rt_priority that can be used | |
5066 | * by a given scheduling class. | |
5067 | */ | |
5add95d4 | 5068 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5069 | { |
5070 | int ret = -EINVAL; | |
5071 | ||
5072 | switch (policy) { | |
5073 | case SCHED_FIFO: | |
5074 | case SCHED_RR: | |
5075 | ret = 1; | |
5076 | break; | |
5077 | case SCHED_NORMAL: | |
b0a9499c | 5078 | case SCHED_BATCH: |
dd41f596 | 5079 | case SCHED_IDLE: |
1da177e4 LT |
5080 | ret = 0; |
5081 | } | |
5082 | return ret; | |
5083 | } | |
5084 | ||
5085 | /** | |
5086 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5087 | * @pid: pid of the process. | |
5088 | * @interval: userspace pointer to the timeslice value. | |
5089 | * | |
5090 | * this syscall writes the default timeslice value of a given process | |
5091 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5092 | */ | |
17da2bd9 | 5093 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5094 | struct timespec __user *, interval) |
1da177e4 | 5095 | { |
36c8b586 | 5096 | struct task_struct *p; |
a4ec24b4 | 5097 | unsigned int time_slice; |
dba091b9 TG |
5098 | unsigned long flags; |
5099 | struct rq *rq; | |
3a5c359a | 5100 | int retval; |
1da177e4 | 5101 | struct timespec t; |
1da177e4 LT |
5102 | |
5103 | if (pid < 0) | |
3a5c359a | 5104 | return -EINVAL; |
1da177e4 LT |
5105 | |
5106 | retval = -ESRCH; | |
1a551ae7 | 5107 | rcu_read_lock(); |
1da177e4 LT |
5108 | p = find_process_by_pid(pid); |
5109 | if (!p) | |
5110 | goto out_unlock; | |
5111 | ||
5112 | retval = security_task_getscheduler(p); | |
5113 | if (retval) | |
5114 | goto out_unlock; | |
5115 | ||
dba091b9 TG |
5116 | rq = task_rq_lock(p, &flags); |
5117 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5118 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5119 | |
1a551ae7 | 5120 | rcu_read_unlock(); |
a4ec24b4 | 5121 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5122 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5123 | return retval; |
3a5c359a | 5124 | |
1da177e4 | 5125 | out_unlock: |
1a551ae7 | 5126 | rcu_read_unlock(); |
1da177e4 LT |
5127 | return retval; |
5128 | } | |
5129 | ||
7c731e0a | 5130 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5131 | |
82a1fcb9 | 5132 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5133 | { |
1da177e4 | 5134 | unsigned long free = 0; |
36c8b586 | 5135 | unsigned state; |
1da177e4 | 5136 | |
1da177e4 | 5137 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5138 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5139 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5140 | #if BITS_PER_LONG == 32 |
1da177e4 | 5141 | if (state == TASK_RUNNING) |
3df0fc5b | 5142 | printk(KERN_CONT " running "); |
1da177e4 | 5143 | else |
3df0fc5b | 5144 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5145 | #else |
5146 | if (state == TASK_RUNNING) | |
3df0fc5b | 5147 | printk(KERN_CONT " running task "); |
1da177e4 | 5148 | else |
3df0fc5b | 5149 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5150 | #endif |
5151 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5152 | free = stack_not_used(p); |
1da177e4 | 5153 | #endif |
3df0fc5b | 5154 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5155 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5156 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5157 | |
5fb5e6de | 5158 | show_stack(p, NULL); |
1da177e4 LT |
5159 | } |
5160 | ||
e59e2ae2 | 5161 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5162 | { |
36c8b586 | 5163 | struct task_struct *g, *p; |
1da177e4 | 5164 | |
4bd77321 | 5165 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5166 | printk(KERN_INFO |
5167 | " task PC stack pid father\n"); | |
1da177e4 | 5168 | #else |
3df0fc5b PZ |
5169 | printk(KERN_INFO |
5170 | " task PC stack pid father\n"); | |
1da177e4 LT |
5171 | #endif |
5172 | read_lock(&tasklist_lock); | |
5173 | do_each_thread(g, p) { | |
5174 | /* | |
5175 | * reset the NMI-timeout, listing all files on a slow | |
5176 | * console might take alot of time: | |
5177 | */ | |
5178 | touch_nmi_watchdog(); | |
39bc89fd | 5179 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5180 | sched_show_task(p); |
1da177e4 LT |
5181 | } while_each_thread(g, p); |
5182 | ||
04c9167f JF |
5183 | touch_all_softlockup_watchdogs(); |
5184 | ||
dd41f596 IM |
5185 | #ifdef CONFIG_SCHED_DEBUG |
5186 | sysrq_sched_debug_show(); | |
5187 | #endif | |
1da177e4 | 5188 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5189 | /* |
5190 | * Only show locks if all tasks are dumped: | |
5191 | */ | |
93335a21 | 5192 | if (!state_filter) |
e59e2ae2 | 5193 | debug_show_all_locks(); |
1da177e4 LT |
5194 | } |
5195 | ||
1df21055 IM |
5196 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5197 | { | |
dd41f596 | 5198 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5199 | } |
5200 | ||
f340c0d1 IM |
5201 | /** |
5202 | * init_idle - set up an idle thread for a given CPU | |
5203 | * @idle: task in question | |
5204 | * @cpu: cpu the idle task belongs to | |
5205 | * | |
5206 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5207 | * flag, to make booting more robust. | |
5208 | */ | |
5c1e1767 | 5209 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5210 | { |
70b97a7f | 5211 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5212 | unsigned long flags; |
5213 | ||
05fa785c | 5214 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5215 | |
dd41f596 | 5216 | __sched_fork(idle); |
06b83b5f | 5217 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5218 | idle->se.exec_start = sched_clock(); |
5219 | ||
96f874e2 | 5220 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5221 | __set_task_cpu(idle, cpu); |
1da177e4 | 5222 | |
1da177e4 | 5223 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5224 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5225 | idle->oncpu = 1; | |
5226 | #endif | |
05fa785c | 5227 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5228 | |
5229 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5230 | #if defined(CONFIG_PREEMPT) |
5231 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5232 | #else | |
a1261f54 | 5233 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5234 | #endif |
dd41f596 IM |
5235 | /* |
5236 | * The idle tasks have their own, simple scheduling class: | |
5237 | */ | |
5238 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5239 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5240 | } |
5241 | ||
5242 | /* | |
5243 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5244 | * indicates which cpus entered this state. This is used | |
5245 | * in the rcu update to wait only for active cpus. For system | |
5246 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5247 | * always be CPU_BITS_NONE. |
1da177e4 | 5248 | */ |
6a7b3dc3 | 5249 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5250 | |
19978ca6 IM |
5251 | /* |
5252 | * Increase the granularity value when there are more CPUs, | |
5253 | * because with more CPUs the 'effective latency' as visible | |
5254 | * to users decreases. But the relationship is not linear, | |
5255 | * so pick a second-best guess by going with the log2 of the | |
5256 | * number of CPUs. | |
5257 | * | |
5258 | * This idea comes from the SD scheduler of Con Kolivas: | |
5259 | */ | |
acb4a848 | 5260 | static int get_update_sysctl_factor(void) |
19978ca6 | 5261 | { |
4ca3ef71 | 5262 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5263 | unsigned int factor; |
5264 | ||
5265 | switch (sysctl_sched_tunable_scaling) { | |
5266 | case SCHED_TUNABLESCALING_NONE: | |
5267 | factor = 1; | |
5268 | break; | |
5269 | case SCHED_TUNABLESCALING_LINEAR: | |
5270 | factor = cpus; | |
5271 | break; | |
5272 | case SCHED_TUNABLESCALING_LOG: | |
5273 | default: | |
5274 | factor = 1 + ilog2(cpus); | |
5275 | break; | |
5276 | } | |
19978ca6 | 5277 | |
acb4a848 CE |
5278 | return factor; |
5279 | } | |
19978ca6 | 5280 | |
acb4a848 CE |
5281 | static void update_sysctl(void) |
5282 | { | |
5283 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5284 | |
0bcdcf28 CE |
5285 | #define SET_SYSCTL(name) \ |
5286 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5287 | SET_SYSCTL(sched_min_granularity); | |
5288 | SET_SYSCTL(sched_latency); | |
5289 | SET_SYSCTL(sched_wakeup_granularity); | |
5290 | SET_SYSCTL(sched_shares_ratelimit); | |
5291 | #undef SET_SYSCTL | |
5292 | } | |
55cd5340 | 5293 | |
0bcdcf28 CE |
5294 | static inline void sched_init_granularity(void) |
5295 | { | |
5296 | update_sysctl(); | |
19978ca6 IM |
5297 | } |
5298 | ||
1da177e4 LT |
5299 | #ifdef CONFIG_SMP |
5300 | /* | |
5301 | * This is how migration works: | |
5302 | * | |
969c7921 TH |
5303 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5304 | * stop_one_cpu(). | |
5305 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5306 | * off the CPU) | |
5307 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5308 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5309 | * it and puts it into the right queue. |
969c7921 TH |
5310 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5311 | * is done. | |
1da177e4 LT |
5312 | */ |
5313 | ||
5314 | /* | |
5315 | * Change a given task's CPU affinity. Migrate the thread to a | |
5316 | * proper CPU and schedule it away if the CPU it's executing on | |
5317 | * is removed from the allowed bitmask. | |
5318 | * | |
5319 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5320 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5321 | * call is not atomic; no spinlocks may be held. |
5322 | */ | |
96f874e2 | 5323 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5324 | { |
5325 | unsigned long flags; | |
70b97a7f | 5326 | struct rq *rq; |
969c7921 | 5327 | unsigned int dest_cpu; |
48f24c4d | 5328 | int ret = 0; |
1da177e4 | 5329 | |
65cc8e48 PZ |
5330 | /* |
5331 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5332 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5333 | */ | |
5334 | again: | |
5335 | while (task_is_waking(p)) | |
5336 | cpu_relax(); | |
1da177e4 | 5337 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5338 | if (task_is_waking(p)) { |
5339 | task_rq_unlock(rq, &flags); | |
5340 | goto again; | |
5341 | } | |
e2912009 | 5342 | |
6ad4c188 | 5343 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5344 | ret = -EINVAL; |
5345 | goto out; | |
5346 | } | |
5347 | ||
9985b0ba | 5348 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5349 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5350 | ret = -EINVAL; |
5351 | goto out; | |
5352 | } | |
5353 | ||
73fe6aae | 5354 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5355 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5356 | else { |
96f874e2 RR |
5357 | cpumask_copy(&p->cpus_allowed, new_mask); |
5358 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5359 | } |
5360 | ||
1da177e4 | 5361 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5362 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5363 | goto out; |
5364 | ||
969c7921 TH |
5365 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
5366 | if (migrate_task(p, dest_cpu)) { | |
5367 | struct migration_arg arg = { p, dest_cpu }; | |
1da177e4 LT |
5368 | /* Need help from migration thread: drop lock and wait. */ |
5369 | task_rq_unlock(rq, &flags); | |
969c7921 | 5370 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5371 | tlb_migrate_finish(p->mm); |
5372 | return 0; | |
5373 | } | |
5374 | out: | |
5375 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5376 | |
1da177e4 LT |
5377 | return ret; |
5378 | } | |
cd8ba7cd | 5379 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5380 | |
5381 | /* | |
41a2d6cf | 5382 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5383 | * this because either it can't run here any more (set_cpus_allowed() |
5384 | * away from this CPU, or CPU going down), or because we're | |
5385 | * attempting to rebalance this task on exec (sched_exec). | |
5386 | * | |
5387 | * So we race with normal scheduler movements, but that's OK, as long | |
5388 | * as the task is no longer on this CPU. | |
efc30814 KK |
5389 | * |
5390 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5391 | */ |
efc30814 | 5392 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5393 | { |
70b97a7f | 5394 | struct rq *rq_dest, *rq_src; |
e2912009 | 5395 | int ret = 0; |
1da177e4 | 5396 | |
e761b772 | 5397 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5398 | return ret; |
1da177e4 LT |
5399 | |
5400 | rq_src = cpu_rq(src_cpu); | |
5401 | rq_dest = cpu_rq(dest_cpu); | |
5402 | ||
5403 | double_rq_lock(rq_src, rq_dest); | |
5404 | /* Already moved. */ | |
5405 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5406 | goto done; |
1da177e4 | 5407 | /* Affinity changed (again). */ |
96f874e2 | 5408 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5409 | goto fail; |
1da177e4 | 5410 | |
e2912009 PZ |
5411 | /* |
5412 | * If we're not on a rq, the next wake-up will ensure we're | |
5413 | * placed properly. | |
5414 | */ | |
5415 | if (p->se.on_rq) { | |
2e1cb74a | 5416 | deactivate_task(rq_src, p, 0); |
e2912009 | 5417 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5418 | activate_task(rq_dest, p, 0); |
15afe09b | 5419 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5420 | } |
b1e38734 | 5421 | done: |
efc30814 | 5422 | ret = 1; |
b1e38734 | 5423 | fail: |
1da177e4 | 5424 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5425 | return ret; |
1da177e4 LT |
5426 | } |
5427 | ||
5428 | /* | |
969c7921 TH |
5429 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5430 | * and performs thread migration by bumping thread off CPU then | |
5431 | * 'pushing' onto another runqueue. | |
1da177e4 | 5432 | */ |
969c7921 | 5433 | static int migration_cpu_stop(void *data) |
1da177e4 | 5434 | { |
969c7921 | 5435 | struct migration_arg *arg = data; |
f7b4cddc | 5436 | |
969c7921 TH |
5437 | /* |
5438 | * The original target cpu might have gone down and we might | |
5439 | * be on another cpu but it doesn't matter. | |
5440 | */ | |
f7b4cddc | 5441 | local_irq_disable(); |
969c7921 | 5442 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5443 | local_irq_enable(); |
1da177e4 | 5444 | return 0; |
f7b4cddc ON |
5445 | } |
5446 | ||
1da177e4 | 5447 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5448 | /* |
3a4fa0a2 | 5449 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5450 | */ |
6a1bdc1b | 5451 | void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5452 | { |
1445c08d ON |
5453 | struct rq *rq = cpu_rq(dead_cpu); |
5454 | int needs_cpu, uninitialized_var(dest_cpu); | |
5455 | unsigned long flags; | |
e76bd8d9 | 5456 | |
1445c08d | 5457 | local_irq_save(flags); |
e76bd8d9 | 5458 | |
1445c08d ON |
5459 | raw_spin_lock(&rq->lock); |
5460 | needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING); | |
5461 | if (needs_cpu) | |
5462 | dest_cpu = select_fallback_rq(dead_cpu, p); | |
5463 | raw_spin_unlock(&rq->lock); | |
c1804d54 ON |
5464 | /* |
5465 | * It can only fail if we race with set_cpus_allowed(), | |
5466 | * in the racer should migrate the task anyway. | |
5467 | */ | |
1445c08d | 5468 | if (needs_cpu) |
c1804d54 | 5469 | __migrate_task(p, dead_cpu, dest_cpu); |
1445c08d | 5470 | local_irq_restore(flags); |
1da177e4 LT |
5471 | } |
5472 | ||
5473 | /* | |
5474 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5475 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5476 | * for performance reasons the counter is not stricly tracking tasks to | |
5477 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5478 | * to keep the global sum constant after CPU-down: | |
5479 | */ | |
70b97a7f | 5480 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5481 | { |
6ad4c188 | 5482 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5483 | unsigned long flags; |
5484 | ||
5485 | local_irq_save(flags); | |
5486 | double_rq_lock(rq_src, rq_dest); | |
5487 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5488 | rq_src->nr_uninterruptible = 0; | |
5489 | double_rq_unlock(rq_src, rq_dest); | |
5490 | local_irq_restore(flags); | |
5491 | } | |
5492 | ||
5493 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5494 | static void migrate_live_tasks(int src_cpu) | |
5495 | { | |
48f24c4d | 5496 | struct task_struct *p, *t; |
1da177e4 | 5497 | |
f7b4cddc | 5498 | read_lock(&tasklist_lock); |
1da177e4 | 5499 | |
48f24c4d IM |
5500 | do_each_thread(t, p) { |
5501 | if (p == current) | |
1da177e4 LT |
5502 | continue; |
5503 | ||
48f24c4d IM |
5504 | if (task_cpu(p) == src_cpu) |
5505 | move_task_off_dead_cpu(src_cpu, p); | |
5506 | } while_each_thread(t, p); | |
1da177e4 | 5507 | |
f7b4cddc | 5508 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5509 | } |
5510 | ||
dd41f596 IM |
5511 | /* |
5512 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5513 | * It does so by boosting its priority to highest possible. |
5514 | * Used by CPU offline code. | |
1da177e4 LT |
5515 | */ |
5516 | void sched_idle_next(void) | |
5517 | { | |
48f24c4d | 5518 | int this_cpu = smp_processor_id(); |
70b97a7f | 5519 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5520 | struct task_struct *p = rq->idle; |
5521 | unsigned long flags; | |
5522 | ||
5523 | /* cpu has to be offline */ | |
48f24c4d | 5524 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5525 | |
48f24c4d IM |
5526 | /* |
5527 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5528 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5529 | */ |
05fa785c | 5530 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5531 | |
dd41f596 | 5532 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5533 | |
94bc9a7b | 5534 | activate_task(rq, p, 0); |
1da177e4 | 5535 | |
05fa785c | 5536 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5537 | } |
5538 | ||
48f24c4d IM |
5539 | /* |
5540 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5541 | * offline. |
5542 | */ | |
5543 | void idle_task_exit(void) | |
5544 | { | |
5545 | struct mm_struct *mm = current->active_mm; | |
5546 | ||
5547 | BUG_ON(cpu_online(smp_processor_id())); | |
5548 | ||
5549 | if (mm != &init_mm) | |
5550 | switch_mm(mm, &init_mm, current); | |
5551 | mmdrop(mm); | |
5552 | } | |
5553 | ||
054b9108 | 5554 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5555 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5556 | { |
70b97a7f | 5557 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5558 | |
5559 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5560 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5561 | |
5562 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5563 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5564 | |
48f24c4d | 5565 | get_task_struct(p); |
1da177e4 LT |
5566 | |
5567 | /* | |
5568 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5569 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5570 | * fine. |
5571 | */ | |
05fa785c | 5572 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5573 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5574 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5575 | |
48f24c4d | 5576 | put_task_struct(p); |
1da177e4 LT |
5577 | } |
5578 | ||
5579 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5580 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5581 | { | |
70b97a7f | 5582 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5583 | struct task_struct *next; |
48f24c4d | 5584 | |
dd41f596 IM |
5585 | for ( ; ; ) { |
5586 | if (!rq->nr_running) | |
5587 | break; | |
b67802ea | 5588 | next = pick_next_task(rq); |
dd41f596 IM |
5589 | if (!next) |
5590 | break; | |
79c53799 | 5591 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5592 | migrate_dead(dead_cpu, next); |
e692ab53 | 5593 | |
1da177e4 LT |
5594 | } |
5595 | } | |
dce48a84 TG |
5596 | |
5597 | /* | |
5598 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5599 | */ | |
5600 | static void calc_global_load_remove(struct rq *rq) | |
5601 | { | |
5602 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5603 | rq->calc_load_active = 0; |
dce48a84 | 5604 | } |
1da177e4 LT |
5605 | #endif /* CONFIG_HOTPLUG_CPU */ |
5606 | ||
e692ab53 NP |
5607 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5608 | ||
5609 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5610 | { |
5611 | .procname = "sched_domain", | |
c57baf1e | 5612 | .mode = 0555, |
e0361851 | 5613 | }, |
56992309 | 5614 | {} |
e692ab53 NP |
5615 | }; |
5616 | ||
5617 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5618 | { |
5619 | .procname = "kernel", | |
c57baf1e | 5620 | .mode = 0555, |
e0361851 AD |
5621 | .child = sd_ctl_dir, |
5622 | }, | |
56992309 | 5623 | {} |
e692ab53 NP |
5624 | }; |
5625 | ||
5626 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5627 | { | |
5628 | struct ctl_table *entry = | |
5cf9f062 | 5629 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5630 | |
e692ab53 NP |
5631 | return entry; |
5632 | } | |
5633 | ||
6382bc90 MM |
5634 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5635 | { | |
cd790076 | 5636 | struct ctl_table *entry; |
6382bc90 | 5637 | |
cd790076 MM |
5638 | /* |
5639 | * In the intermediate directories, both the child directory and | |
5640 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5641 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5642 | * static strings and all have proc handlers. |
5643 | */ | |
5644 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5645 | if (entry->child) |
5646 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5647 | if (entry->proc_handler == NULL) |
5648 | kfree(entry->procname); | |
5649 | } | |
6382bc90 MM |
5650 | |
5651 | kfree(*tablep); | |
5652 | *tablep = NULL; | |
5653 | } | |
5654 | ||
e692ab53 | 5655 | static void |
e0361851 | 5656 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5657 | const char *procname, void *data, int maxlen, |
5658 | mode_t mode, proc_handler *proc_handler) | |
5659 | { | |
e692ab53 NP |
5660 | entry->procname = procname; |
5661 | entry->data = data; | |
5662 | entry->maxlen = maxlen; | |
5663 | entry->mode = mode; | |
5664 | entry->proc_handler = proc_handler; | |
5665 | } | |
5666 | ||
5667 | static struct ctl_table * | |
5668 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5669 | { | |
a5d8c348 | 5670 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5671 | |
ad1cdc1d MM |
5672 | if (table == NULL) |
5673 | return NULL; | |
5674 | ||
e0361851 | 5675 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5676 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5677 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5678 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5679 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5680 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5681 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5682 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5683 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5684 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5685 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5686 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5687 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5688 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5689 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5690 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5691 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5692 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5693 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5694 | &sd->cache_nice_tries, |
5695 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5696 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5697 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5698 | set_table_entry(&table[11], "name", sd->name, |
5699 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5700 | /* &table[12] is terminator */ | |
e692ab53 NP |
5701 | |
5702 | return table; | |
5703 | } | |
5704 | ||
9a4e7159 | 5705 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5706 | { |
5707 | struct ctl_table *entry, *table; | |
5708 | struct sched_domain *sd; | |
5709 | int domain_num = 0, i; | |
5710 | char buf[32]; | |
5711 | ||
5712 | for_each_domain(cpu, sd) | |
5713 | domain_num++; | |
5714 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5715 | if (table == NULL) |
5716 | return NULL; | |
e692ab53 NP |
5717 | |
5718 | i = 0; | |
5719 | for_each_domain(cpu, sd) { | |
5720 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5721 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5722 | entry->mode = 0555; |
e692ab53 NP |
5723 | entry->child = sd_alloc_ctl_domain_table(sd); |
5724 | entry++; | |
5725 | i++; | |
5726 | } | |
5727 | return table; | |
5728 | } | |
5729 | ||
5730 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5731 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5732 | { |
6ad4c188 | 5733 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5734 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5735 | char buf[32]; | |
5736 | ||
7378547f MM |
5737 | WARN_ON(sd_ctl_dir[0].child); |
5738 | sd_ctl_dir[0].child = entry; | |
5739 | ||
ad1cdc1d MM |
5740 | if (entry == NULL) |
5741 | return; | |
5742 | ||
6ad4c188 | 5743 | for_each_possible_cpu(i) { |
e692ab53 | 5744 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5745 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5746 | entry->mode = 0555; |
e692ab53 | 5747 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5748 | entry++; |
e692ab53 | 5749 | } |
7378547f MM |
5750 | |
5751 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5752 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5753 | } | |
6382bc90 | 5754 | |
7378547f | 5755 | /* may be called multiple times per register */ |
6382bc90 MM |
5756 | static void unregister_sched_domain_sysctl(void) |
5757 | { | |
7378547f MM |
5758 | if (sd_sysctl_header) |
5759 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5760 | sd_sysctl_header = NULL; |
7378547f MM |
5761 | if (sd_ctl_dir[0].child) |
5762 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5763 | } |
e692ab53 | 5764 | #else |
6382bc90 MM |
5765 | static void register_sched_domain_sysctl(void) |
5766 | { | |
5767 | } | |
5768 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5769 | { |
5770 | } | |
5771 | #endif | |
5772 | ||
1f11eb6a GH |
5773 | static void set_rq_online(struct rq *rq) |
5774 | { | |
5775 | if (!rq->online) { | |
5776 | const struct sched_class *class; | |
5777 | ||
c6c4927b | 5778 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5779 | rq->online = 1; |
5780 | ||
5781 | for_each_class(class) { | |
5782 | if (class->rq_online) | |
5783 | class->rq_online(rq); | |
5784 | } | |
5785 | } | |
5786 | } | |
5787 | ||
5788 | static void set_rq_offline(struct rq *rq) | |
5789 | { | |
5790 | if (rq->online) { | |
5791 | const struct sched_class *class; | |
5792 | ||
5793 | for_each_class(class) { | |
5794 | if (class->rq_offline) | |
5795 | class->rq_offline(rq); | |
5796 | } | |
5797 | ||
c6c4927b | 5798 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5799 | rq->online = 0; |
5800 | } | |
5801 | } | |
5802 | ||
1da177e4 LT |
5803 | /* |
5804 | * migration_call - callback that gets triggered when a CPU is added. | |
5805 | * Here we can start up the necessary migration thread for the new CPU. | |
5806 | */ | |
48f24c4d IM |
5807 | static int __cpuinit |
5808 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5809 | { |
48f24c4d | 5810 | int cpu = (long)hcpu; |
1da177e4 | 5811 | unsigned long flags; |
969c7921 | 5812 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5813 | |
5814 | switch (action) { | |
5be9361c | 5815 | |
1da177e4 | 5816 | case CPU_UP_PREPARE: |
8bb78442 | 5817 | case CPU_UP_PREPARE_FROZEN: |
a468d389 | 5818 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5819 | break; |
48f24c4d | 5820 | |
1da177e4 | 5821 | case CPU_ONLINE: |
8bb78442 | 5822 | case CPU_ONLINE_FROZEN: |
1f94ef59 | 5823 | /* Update our root-domain */ |
05fa785c | 5824 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5825 | if (rq->rd) { |
c6c4927b | 5826 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5827 | |
5828 | set_rq_online(rq); | |
1f94ef59 | 5829 | } |
05fa785c | 5830 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5831 | break; |
48f24c4d | 5832 | |
1da177e4 | 5833 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 5834 | case CPU_DEAD: |
8bb78442 | 5835 | case CPU_DEAD_FROZEN: |
1da177e4 | 5836 | migrate_live_tasks(cpu); |
1da177e4 | 5837 | /* Idle task back to normal (off runqueue, low prio) */ |
05fa785c | 5838 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 5839 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5840 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5841 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5842 | migrate_dead_tasks(cpu); |
05fa785c | 5843 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5844 | migrate_nr_uninterruptible(rq); |
5845 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5846 | calc_global_load_remove(rq); |
1da177e4 | 5847 | break; |
57d885fe | 5848 | |
08f503b0 GH |
5849 | case CPU_DYING: |
5850 | case CPU_DYING_FROZEN: | |
57d885fe | 5851 | /* Update our root-domain */ |
05fa785c | 5852 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5853 | if (rq->rd) { |
c6c4927b | 5854 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5855 | set_rq_offline(rq); |
57d885fe | 5856 | } |
05fa785c | 5857 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 5858 | break; |
1da177e4 LT |
5859 | #endif |
5860 | } | |
5861 | return NOTIFY_OK; | |
5862 | } | |
5863 | ||
f38b0820 PM |
5864 | /* |
5865 | * Register at high priority so that task migration (migrate_all_tasks) | |
5866 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5867 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5868 | */ |
26c2143b | 5869 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5870 | .notifier_call = migration_call, |
50a323b7 | 5871 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5872 | }; |
5873 | ||
3a101d05 TH |
5874 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5875 | unsigned long action, void *hcpu) | |
5876 | { | |
5877 | switch (action & ~CPU_TASKS_FROZEN) { | |
5878 | case CPU_ONLINE: | |
5879 | case CPU_DOWN_FAILED: | |
5880 | set_cpu_active((long)hcpu, true); | |
5881 | return NOTIFY_OK; | |
5882 | default: | |
5883 | return NOTIFY_DONE; | |
5884 | } | |
5885 | } | |
5886 | ||
5887 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5888 | unsigned long action, void *hcpu) | |
5889 | { | |
5890 | switch (action & ~CPU_TASKS_FROZEN) { | |
5891 | case CPU_DOWN_PREPARE: | |
5892 | set_cpu_active((long)hcpu, false); | |
5893 | return NOTIFY_OK; | |
5894 | default: | |
5895 | return NOTIFY_DONE; | |
5896 | } | |
5897 | } | |
5898 | ||
7babe8db | 5899 | static int __init migration_init(void) |
1da177e4 LT |
5900 | { |
5901 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5902 | int err; |
48f24c4d | 5903 | |
3a101d05 | 5904 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5905 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5906 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5907 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5908 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5909 | |
3a101d05 TH |
5910 | /* Register cpu active notifiers */ |
5911 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5912 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5913 | ||
a004cd42 | 5914 | return 0; |
1da177e4 | 5915 | } |
7babe8db | 5916 | early_initcall(migration_init); |
1da177e4 LT |
5917 | #endif |
5918 | ||
5919 | #ifdef CONFIG_SMP | |
476f3534 | 5920 | |
3e9830dc | 5921 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5922 | |
f6630114 MT |
5923 | static __read_mostly int sched_domain_debug_enabled; |
5924 | ||
5925 | static int __init sched_domain_debug_setup(char *str) | |
5926 | { | |
5927 | sched_domain_debug_enabled = 1; | |
5928 | ||
5929 | return 0; | |
5930 | } | |
5931 | early_param("sched_debug", sched_domain_debug_setup); | |
5932 | ||
7c16ec58 | 5933 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5934 | struct cpumask *groupmask) |
1da177e4 | 5935 | { |
4dcf6aff | 5936 | struct sched_group *group = sd->groups; |
434d53b0 | 5937 | char str[256]; |
1da177e4 | 5938 | |
968ea6d8 | 5939 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 5940 | cpumask_clear(groupmask); |
4dcf6aff IM |
5941 | |
5942 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
5943 | ||
5944 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 5945 | printk("does not load-balance\n"); |
4dcf6aff | 5946 | if (sd->parent) |
3df0fc5b PZ |
5947 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
5948 | " has parent"); | |
4dcf6aff | 5949 | return -1; |
41c7ce9a NP |
5950 | } |
5951 | ||
3df0fc5b | 5952 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 5953 | |
758b2cdc | 5954 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
5955 | printk(KERN_ERR "ERROR: domain->span does not contain " |
5956 | "CPU%d\n", cpu); | |
4dcf6aff | 5957 | } |
758b2cdc | 5958 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
5959 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
5960 | " CPU%d\n", cpu); | |
4dcf6aff | 5961 | } |
1da177e4 | 5962 | |
4dcf6aff | 5963 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 5964 | do { |
4dcf6aff | 5965 | if (!group) { |
3df0fc5b PZ |
5966 | printk("\n"); |
5967 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
5968 | break; |
5969 | } | |
5970 | ||
18a3885f | 5971 | if (!group->cpu_power) { |
3df0fc5b PZ |
5972 | printk(KERN_CONT "\n"); |
5973 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
5974 | "set\n"); | |
4dcf6aff IM |
5975 | break; |
5976 | } | |
1da177e4 | 5977 | |
758b2cdc | 5978 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
5979 | printk(KERN_CONT "\n"); |
5980 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
5981 | break; |
5982 | } | |
1da177e4 | 5983 | |
758b2cdc | 5984 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
5985 | printk(KERN_CONT "\n"); |
5986 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
5987 | break; |
5988 | } | |
1da177e4 | 5989 | |
758b2cdc | 5990 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 5991 | |
968ea6d8 | 5992 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 5993 | |
3df0fc5b | 5994 | printk(KERN_CONT " %s", str); |
18a3885f | 5995 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
5996 | printk(KERN_CONT " (cpu_power = %d)", |
5997 | group->cpu_power); | |
381512cf | 5998 | } |
1da177e4 | 5999 | |
4dcf6aff IM |
6000 | group = group->next; |
6001 | } while (group != sd->groups); | |
3df0fc5b | 6002 | printk(KERN_CONT "\n"); |
1da177e4 | 6003 | |
758b2cdc | 6004 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6005 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6006 | |
758b2cdc RR |
6007 | if (sd->parent && |
6008 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6009 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6010 | "of domain->span\n"); | |
4dcf6aff IM |
6011 | return 0; |
6012 | } | |
1da177e4 | 6013 | |
4dcf6aff IM |
6014 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6015 | { | |
d5dd3db1 | 6016 | cpumask_var_t groupmask; |
4dcf6aff | 6017 | int level = 0; |
1da177e4 | 6018 | |
f6630114 MT |
6019 | if (!sched_domain_debug_enabled) |
6020 | return; | |
6021 | ||
4dcf6aff IM |
6022 | if (!sd) { |
6023 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6024 | return; | |
6025 | } | |
1da177e4 | 6026 | |
4dcf6aff IM |
6027 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6028 | ||
d5dd3db1 | 6029 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6030 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6031 | return; | |
6032 | } | |
6033 | ||
4dcf6aff | 6034 | for (;;) { |
7c16ec58 | 6035 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6036 | break; |
1da177e4 LT |
6037 | level++; |
6038 | sd = sd->parent; | |
33859f7f | 6039 | if (!sd) |
4dcf6aff IM |
6040 | break; |
6041 | } | |
d5dd3db1 | 6042 | free_cpumask_var(groupmask); |
1da177e4 | 6043 | } |
6d6bc0ad | 6044 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6045 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6046 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6047 | |
1a20ff27 | 6048 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6049 | { |
758b2cdc | 6050 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6051 | return 1; |
6052 | ||
6053 | /* Following flags need at least 2 groups */ | |
6054 | if (sd->flags & (SD_LOAD_BALANCE | | |
6055 | SD_BALANCE_NEWIDLE | | |
6056 | SD_BALANCE_FORK | | |
89c4710e SS |
6057 | SD_BALANCE_EXEC | |
6058 | SD_SHARE_CPUPOWER | | |
6059 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6060 | if (sd->groups != sd->groups->next) |
6061 | return 0; | |
6062 | } | |
6063 | ||
6064 | /* Following flags don't use groups */ | |
c88d5910 | 6065 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6066 | return 0; |
6067 | ||
6068 | return 1; | |
6069 | } | |
6070 | ||
48f24c4d IM |
6071 | static int |
6072 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6073 | { |
6074 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6075 | ||
6076 | if (sd_degenerate(parent)) | |
6077 | return 1; | |
6078 | ||
758b2cdc | 6079 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6080 | return 0; |
6081 | ||
245af2c7 SS |
6082 | /* Flags needing groups don't count if only 1 group in parent */ |
6083 | if (parent->groups == parent->groups->next) { | |
6084 | pflags &= ~(SD_LOAD_BALANCE | | |
6085 | SD_BALANCE_NEWIDLE | | |
6086 | SD_BALANCE_FORK | | |
89c4710e SS |
6087 | SD_BALANCE_EXEC | |
6088 | SD_SHARE_CPUPOWER | | |
6089 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6090 | if (nr_node_ids == 1) |
6091 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6092 | } |
6093 | if (~cflags & pflags) | |
6094 | return 0; | |
6095 | ||
6096 | return 1; | |
6097 | } | |
6098 | ||
c6c4927b RR |
6099 | static void free_rootdomain(struct root_domain *rd) |
6100 | { | |
047106ad PZ |
6101 | synchronize_sched(); |
6102 | ||
68e74568 RR |
6103 | cpupri_cleanup(&rd->cpupri); |
6104 | ||
c6c4927b RR |
6105 | free_cpumask_var(rd->rto_mask); |
6106 | free_cpumask_var(rd->online); | |
6107 | free_cpumask_var(rd->span); | |
6108 | kfree(rd); | |
6109 | } | |
6110 | ||
57d885fe GH |
6111 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6112 | { | |
a0490fa3 | 6113 | struct root_domain *old_rd = NULL; |
57d885fe | 6114 | unsigned long flags; |
57d885fe | 6115 | |
05fa785c | 6116 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6117 | |
6118 | if (rq->rd) { | |
a0490fa3 | 6119 | old_rd = rq->rd; |
57d885fe | 6120 | |
c6c4927b | 6121 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6122 | set_rq_offline(rq); |
57d885fe | 6123 | |
c6c4927b | 6124 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6125 | |
a0490fa3 IM |
6126 | /* |
6127 | * If we dont want to free the old_rt yet then | |
6128 | * set old_rd to NULL to skip the freeing later | |
6129 | * in this function: | |
6130 | */ | |
6131 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6132 | old_rd = NULL; | |
57d885fe GH |
6133 | } |
6134 | ||
6135 | atomic_inc(&rd->refcount); | |
6136 | rq->rd = rd; | |
6137 | ||
c6c4927b | 6138 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6139 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6140 | set_rq_online(rq); |
57d885fe | 6141 | |
05fa785c | 6142 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6143 | |
6144 | if (old_rd) | |
6145 | free_rootdomain(old_rd); | |
57d885fe GH |
6146 | } |
6147 | ||
fd5e1b5d | 6148 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6149 | { |
36b7b6d4 PE |
6150 | gfp_t gfp = GFP_KERNEL; |
6151 | ||
57d885fe GH |
6152 | memset(rd, 0, sizeof(*rd)); |
6153 | ||
36b7b6d4 PE |
6154 | if (bootmem) |
6155 | gfp = GFP_NOWAIT; | |
c6c4927b | 6156 | |
36b7b6d4 | 6157 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6158 | goto out; |
36b7b6d4 | 6159 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6160 | goto free_span; |
36b7b6d4 | 6161 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6162 | goto free_online; |
6e0534f2 | 6163 | |
0fb53029 | 6164 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6165 | goto free_rto_mask; |
c6c4927b | 6166 | return 0; |
6e0534f2 | 6167 | |
68e74568 RR |
6168 | free_rto_mask: |
6169 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6170 | free_online: |
6171 | free_cpumask_var(rd->online); | |
6172 | free_span: | |
6173 | free_cpumask_var(rd->span); | |
0c910d28 | 6174 | out: |
c6c4927b | 6175 | return -ENOMEM; |
57d885fe GH |
6176 | } |
6177 | ||
6178 | static void init_defrootdomain(void) | |
6179 | { | |
c6c4927b RR |
6180 | init_rootdomain(&def_root_domain, true); |
6181 | ||
57d885fe GH |
6182 | atomic_set(&def_root_domain.refcount, 1); |
6183 | } | |
6184 | ||
dc938520 | 6185 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6186 | { |
6187 | struct root_domain *rd; | |
6188 | ||
6189 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6190 | if (!rd) | |
6191 | return NULL; | |
6192 | ||
c6c4927b RR |
6193 | if (init_rootdomain(rd, false) != 0) { |
6194 | kfree(rd); | |
6195 | return NULL; | |
6196 | } | |
57d885fe GH |
6197 | |
6198 | return rd; | |
6199 | } | |
6200 | ||
1da177e4 | 6201 | /* |
0eab9146 | 6202 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6203 | * hold the hotplug lock. |
6204 | */ | |
0eab9146 IM |
6205 | static void |
6206 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6207 | { |
70b97a7f | 6208 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6209 | struct sched_domain *tmp; |
6210 | ||
669c55e9 PZ |
6211 | for (tmp = sd; tmp; tmp = tmp->parent) |
6212 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6213 | ||
245af2c7 | 6214 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6215 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6216 | struct sched_domain *parent = tmp->parent; |
6217 | if (!parent) | |
6218 | break; | |
f29c9b1c | 6219 | |
1a848870 | 6220 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6221 | tmp->parent = parent->parent; |
1a848870 SS |
6222 | if (parent->parent) |
6223 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6224 | } else |
6225 | tmp = tmp->parent; | |
245af2c7 SS |
6226 | } |
6227 | ||
1a848870 | 6228 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6229 | sd = sd->parent; |
1a848870 SS |
6230 | if (sd) |
6231 | sd->child = NULL; | |
6232 | } | |
1da177e4 LT |
6233 | |
6234 | sched_domain_debug(sd, cpu); | |
6235 | ||
57d885fe | 6236 | rq_attach_root(rq, rd); |
674311d5 | 6237 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6238 | } |
6239 | ||
6240 | /* cpus with isolated domains */ | |
dcc30a35 | 6241 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6242 | |
6243 | /* Setup the mask of cpus configured for isolated domains */ | |
6244 | static int __init isolated_cpu_setup(char *str) | |
6245 | { | |
bdddd296 | 6246 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6247 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6248 | return 1; |
6249 | } | |
6250 | ||
8927f494 | 6251 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6252 | |
6253 | /* | |
6711cab4 SS |
6254 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6255 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6256 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6257 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6258 | * |
6259 | * init_sched_build_groups will build a circular linked list of the groups | |
6260 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6261 | * and ->cpu_power to 0. | |
6262 | */ | |
a616058b | 6263 | static void |
96f874e2 RR |
6264 | init_sched_build_groups(const struct cpumask *span, |
6265 | const struct cpumask *cpu_map, | |
6266 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6267 | struct sched_group **sg, |
96f874e2 RR |
6268 | struct cpumask *tmpmask), |
6269 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6270 | { |
6271 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6272 | int i; |
6273 | ||
96f874e2 | 6274 | cpumask_clear(covered); |
7c16ec58 | 6275 | |
abcd083a | 6276 | for_each_cpu(i, span) { |
6711cab4 | 6277 | struct sched_group *sg; |
7c16ec58 | 6278 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6279 | int j; |
6280 | ||
758b2cdc | 6281 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6282 | continue; |
6283 | ||
758b2cdc | 6284 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6285 | sg->cpu_power = 0; |
1da177e4 | 6286 | |
abcd083a | 6287 | for_each_cpu(j, span) { |
7c16ec58 | 6288 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6289 | continue; |
6290 | ||
96f874e2 | 6291 | cpumask_set_cpu(j, covered); |
758b2cdc | 6292 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6293 | } |
6294 | if (!first) | |
6295 | first = sg; | |
6296 | if (last) | |
6297 | last->next = sg; | |
6298 | last = sg; | |
6299 | } | |
6300 | last->next = first; | |
6301 | } | |
6302 | ||
9c1cfda2 | 6303 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6304 | |
9c1cfda2 | 6305 | #ifdef CONFIG_NUMA |
198e2f18 | 6306 | |
9c1cfda2 JH |
6307 | /** |
6308 | * find_next_best_node - find the next node to include in a sched_domain | |
6309 | * @node: node whose sched_domain we're building | |
6310 | * @used_nodes: nodes already in the sched_domain | |
6311 | * | |
41a2d6cf | 6312 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6313 | * finds the closest node not already in the @used_nodes map. |
6314 | * | |
6315 | * Should use nodemask_t. | |
6316 | */ | |
c5f59f08 | 6317 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6318 | { |
6319 | int i, n, val, min_val, best_node = 0; | |
6320 | ||
6321 | min_val = INT_MAX; | |
6322 | ||
076ac2af | 6323 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6324 | /* Start at @node */ |
076ac2af | 6325 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6326 | |
6327 | if (!nr_cpus_node(n)) | |
6328 | continue; | |
6329 | ||
6330 | /* Skip already used nodes */ | |
c5f59f08 | 6331 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6332 | continue; |
6333 | ||
6334 | /* Simple min distance search */ | |
6335 | val = node_distance(node, n); | |
6336 | ||
6337 | if (val < min_val) { | |
6338 | min_val = val; | |
6339 | best_node = n; | |
6340 | } | |
6341 | } | |
6342 | ||
c5f59f08 | 6343 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6344 | return best_node; |
6345 | } | |
6346 | ||
6347 | /** | |
6348 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6349 | * @node: node whose cpumask we're constructing | |
73486722 | 6350 | * @span: resulting cpumask |
9c1cfda2 | 6351 | * |
41a2d6cf | 6352 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6353 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6354 | * out optimally. | |
6355 | */ | |
96f874e2 | 6356 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6357 | { |
c5f59f08 | 6358 | nodemask_t used_nodes; |
48f24c4d | 6359 | int i; |
9c1cfda2 | 6360 | |
6ca09dfc | 6361 | cpumask_clear(span); |
c5f59f08 | 6362 | nodes_clear(used_nodes); |
9c1cfda2 | 6363 | |
6ca09dfc | 6364 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6365 | node_set(node, used_nodes); |
9c1cfda2 JH |
6366 | |
6367 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6368 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6369 | |
6ca09dfc | 6370 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6371 | } |
9c1cfda2 | 6372 | } |
6d6bc0ad | 6373 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6374 | |
5c45bf27 | 6375 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6376 | |
6c99e9ad RR |
6377 | /* |
6378 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6379 | * |
6380 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6381 | * and struct sched_domain. ) | |
6c99e9ad RR |
6382 | */ |
6383 | struct static_sched_group { | |
6384 | struct sched_group sg; | |
6385 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6386 | }; | |
6387 | ||
6388 | struct static_sched_domain { | |
6389 | struct sched_domain sd; | |
6390 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6391 | }; | |
6392 | ||
49a02c51 AH |
6393 | struct s_data { |
6394 | #ifdef CONFIG_NUMA | |
6395 | int sd_allnodes; | |
6396 | cpumask_var_t domainspan; | |
6397 | cpumask_var_t covered; | |
6398 | cpumask_var_t notcovered; | |
6399 | #endif | |
6400 | cpumask_var_t nodemask; | |
6401 | cpumask_var_t this_sibling_map; | |
6402 | cpumask_var_t this_core_map; | |
6403 | cpumask_var_t send_covered; | |
6404 | cpumask_var_t tmpmask; | |
6405 | struct sched_group **sched_group_nodes; | |
6406 | struct root_domain *rd; | |
6407 | }; | |
6408 | ||
2109b99e AH |
6409 | enum s_alloc { |
6410 | sa_sched_groups = 0, | |
6411 | sa_rootdomain, | |
6412 | sa_tmpmask, | |
6413 | sa_send_covered, | |
6414 | sa_this_core_map, | |
6415 | sa_this_sibling_map, | |
6416 | sa_nodemask, | |
6417 | sa_sched_group_nodes, | |
6418 | #ifdef CONFIG_NUMA | |
6419 | sa_notcovered, | |
6420 | sa_covered, | |
6421 | sa_domainspan, | |
6422 | #endif | |
6423 | sa_none, | |
6424 | }; | |
6425 | ||
9c1cfda2 | 6426 | /* |
48f24c4d | 6427 | * SMT sched-domains: |
9c1cfda2 | 6428 | */ |
1da177e4 | 6429 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6430 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6431 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6432 | |
41a2d6cf | 6433 | static int |
96f874e2 RR |
6434 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6435 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6436 | { |
6711cab4 | 6437 | if (sg) |
1871e52c | 6438 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6439 | return cpu; |
6440 | } | |
6d6bc0ad | 6441 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6442 | |
48f24c4d IM |
6443 | /* |
6444 | * multi-core sched-domains: | |
6445 | */ | |
1e9f28fa | 6446 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6447 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6448 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6449 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6450 | |
6451 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6452 | static int |
96f874e2 RR |
6453 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6454 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6455 | { |
6711cab4 | 6456 | int group; |
7c16ec58 | 6457 | |
c69fc56d | 6458 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6459 | group = cpumask_first(mask); |
6711cab4 | 6460 | if (sg) |
6c99e9ad | 6461 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6462 | return group; |
1e9f28fa SS |
6463 | } |
6464 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6465 | static int |
96f874e2 RR |
6466 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6467 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6468 | { |
6711cab4 | 6469 | if (sg) |
6c99e9ad | 6470 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6471 | return cpu; |
6472 | } | |
6473 | #endif | |
6474 | ||
6c99e9ad RR |
6475 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6476 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6477 | |
41a2d6cf | 6478 | static int |
96f874e2 RR |
6479 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6480 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6481 | { |
6711cab4 | 6482 | int group; |
48f24c4d | 6483 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6484 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6485 | group = cpumask_first(mask); |
1e9f28fa | 6486 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6487 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6488 | group = cpumask_first(mask); |
1da177e4 | 6489 | #else |
6711cab4 | 6490 | group = cpu; |
1da177e4 | 6491 | #endif |
6711cab4 | 6492 | if (sg) |
6c99e9ad | 6493 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6494 | return group; |
1da177e4 LT |
6495 | } |
6496 | ||
6497 | #ifdef CONFIG_NUMA | |
1da177e4 | 6498 | /* |
9c1cfda2 JH |
6499 | * The init_sched_build_groups can't handle what we want to do with node |
6500 | * groups, so roll our own. Now each node has its own list of groups which | |
6501 | * gets dynamically allocated. | |
1da177e4 | 6502 | */ |
62ea9ceb | 6503 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6504 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6505 | |
62ea9ceb | 6506 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6507 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6508 | |
96f874e2 RR |
6509 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6510 | struct sched_group **sg, | |
6511 | struct cpumask *nodemask) | |
9c1cfda2 | 6512 | { |
6711cab4 SS |
6513 | int group; |
6514 | ||
6ca09dfc | 6515 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6516 | group = cpumask_first(nodemask); |
6711cab4 SS |
6517 | |
6518 | if (sg) | |
6c99e9ad | 6519 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6520 | return group; |
1da177e4 | 6521 | } |
6711cab4 | 6522 | |
08069033 SS |
6523 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6524 | { | |
6525 | struct sched_group *sg = group_head; | |
6526 | int j; | |
6527 | ||
6528 | if (!sg) | |
6529 | return; | |
3a5c359a | 6530 | do { |
758b2cdc | 6531 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6532 | struct sched_domain *sd; |
08069033 | 6533 | |
6c99e9ad | 6534 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6535 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6536 | /* |
6537 | * Only add "power" once for each | |
6538 | * physical package. | |
6539 | */ | |
6540 | continue; | |
6541 | } | |
08069033 | 6542 | |
18a3885f | 6543 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6544 | } |
6545 | sg = sg->next; | |
6546 | } while (sg != group_head); | |
08069033 | 6547 | } |
0601a88d AH |
6548 | |
6549 | static int build_numa_sched_groups(struct s_data *d, | |
6550 | const struct cpumask *cpu_map, int num) | |
6551 | { | |
6552 | struct sched_domain *sd; | |
6553 | struct sched_group *sg, *prev; | |
6554 | int n, j; | |
6555 | ||
6556 | cpumask_clear(d->covered); | |
6557 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6558 | if (cpumask_empty(d->nodemask)) { | |
6559 | d->sched_group_nodes[num] = NULL; | |
6560 | goto out; | |
6561 | } | |
6562 | ||
6563 | sched_domain_node_span(num, d->domainspan); | |
6564 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6565 | ||
6566 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6567 | GFP_KERNEL, num); | |
6568 | if (!sg) { | |
3df0fc5b PZ |
6569 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6570 | num); | |
0601a88d AH |
6571 | return -ENOMEM; |
6572 | } | |
6573 | d->sched_group_nodes[num] = sg; | |
6574 | ||
6575 | for_each_cpu(j, d->nodemask) { | |
6576 | sd = &per_cpu(node_domains, j).sd; | |
6577 | sd->groups = sg; | |
6578 | } | |
6579 | ||
18a3885f | 6580 | sg->cpu_power = 0; |
0601a88d AH |
6581 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6582 | sg->next = sg; | |
6583 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6584 | ||
6585 | prev = sg; | |
6586 | for (j = 0; j < nr_node_ids; j++) { | |
6587 | n = (num + j) % nr_node_ids; | |
6588 | cpumask_complement(d->notcovered, d->covered); | |
6589 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6590 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6591 | if (cpumask_empty(d->tmpmask)) | |
6592 | break; | |
6593 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6594 | if (cpumask_empty(d->tmpmask)) | |
6595 | continue; | |
6596 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6597 | GFP_KERNEL, num); | |
6598 | if (!sg) { | |
3df0fc5b PZ |
6599 | printk(KERN_WARNING |
6600 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6601 | return -ENOMEM; |
6602 | } | |
18a3885f | 6603 | sg->cpu_power = 0; |
0601a88d AH |
6604 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6605 | sg->next = prev->next; | |
6606 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6607 | prev->next = sg; | |
6608 | prev = sg; | |
6609 | } | |
6610 | out: | |
6611 | return 0; | |
6612 | } | |
6d6bc0ad | 6613 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6614 | |
a616058b | 6615 | #ifdef CONFIG_NUMA |
51888ca2 | 6616 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6617 | static void free_sched_groups(const struct cpumask *cpu_map, |
6618 | struct cpumask *nodemask) | |
51888ca2 | 6619 | { |
a616058b | 6620 | int cpu, i; |
51888ca2 | 6621 | |
abcd083a | 6622 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6623 | struct sched_group **sched_group_nodes |
6624 | = sched_group_nodes_bycpu[cpu]; | |
6625 | ||
51888ca2 SV |
6626 | if (!sched_group_nodes) |
6627 | continue; | |
6628 | ||
076ac2af | 6629 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6630 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6631 | ||
6ca09dfc | 6632 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6633 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6634 | continue; |
6635 | ||
6636 | if (sg == NULL) | |
6637 | continue; | |
6638 | sg = sg->next; | |
6639 | next_sg: | |
6640 | oldsg = sg; | |
6641 | sg = sg->next; | |
6642 | kfree(oldsg); | |
6643 | if (oldsg != sched_group_nodes[i]) | |
6644 | goto next_sg; | |
6645 | } | |
6646 | kfree(sched_group_nodes); | |
6647 | sched_group_nodes_bycpu[cpu] = NULL; | |
6648 | } | |
51888ca2 | 6649 | } |
6d6bc0ad | 6650 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6651 | static void free_sched_groups(const struct cpumask *cpu_map, |
6652 | struct cpumask *nodemask) | |
a616058b SS |
6653 | { |
6654 | } | |
6d6bc0ad | 6655 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6656 | |
89c4710e SS |
6657 | /* |
6658 | * Initialize sched groups cpu_power. | |
6659 | * | |
6660 | * cpu_power indicates the capacity of sched group, which is used while | |
6661 | * distributing the load between different sched groups in a sched domain. | |
6662 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6663 | * there are asymmetries in the topology. If there are asymmetries, group | |
6664 | * having more cpu_power will pickup more load compared to the group having | |
6665 | * less cpu_power. | |
89c4710e SS |
6666 | */ |
6667 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6668 | { | |
6669 | struct sched_domain *child; | |
6670 | struct sched_group *group; | |
f93e65c1 PZ |
6671 | long power; |
6672 | int weight; | |
89c4710e SS |
6673 | |
6674 | WARN_ON(!sd || !sd->groups); | |
6675 | ||
13318a71 | 6676 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6677 | return; |
6678 | ||
6679 | child = sd->child; | |
6680 | ||
18a3885f | 6681 | sd->groups->cpu_power = 0; |
5517d86b | 6682 | |
f93e65c1 PZ |
6683 | if (!child) { |
6684 | power = SCHED_LOAD_SCALE; | |
6685 | weight = cpumask_weight(sched_domain_span(sd)); | |
6686 | /* | |
6687 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6688 | * Usually multiple threads get a better yield out of |
6689 | * that one core than a single thread would have, | |
6690 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6691 | */ |
a52bfd73 PZ |
6692 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6693 | power *= sd->smt_gain; | |
f93e65c1 | 6694 | power /= weight; |
a52bfd73 PZ |
6695 | power >>= SCHED_LOAD_SHIFT; |
6696 | } | |
18a3885f | 6697 | sd->groups->cpu_power += power; |
89c4710e SS |
6698 | return; |
6699 | } | |
6700 | ||
89c4710e | 6701 | /* |
f93e65c1 | 6702 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6703 | */ |
6704 | group = child->groups; | |
6705 | do { | |
18a3885f | 6706 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6707 | group = group->next; |
6708 | } while (group != child->groups); | |
6709 | } | |
6710 | ||
7c16ec58 MT |
6711 | /* |
6712 | * Initializers for schedule domains | |
6713 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6714 | */ | |
6715 | ||
a5d8c348 IM |
6716 | #ifdef CONFIG_SCHED_DEBUG |
6717 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6718 | #else | |
6719 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6720 | #endif | |
6721 | ||
7c16ec58 | 6722 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6723 | |
7c16ec58 MT |
6724 | #define SD_INIT_FUNC(type) \ |
6725 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6726 | { \ | |
6727 | memset(sd, 0, sizeof(*sd)); \ | |
6728 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6729 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6730 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6731 | } |
6732 | ||
6733 | SD_INIT_FUNC(CPU) | |
6734 | #ifdef CONFIG_NUMA | |
6735 | SD_INIT_FUNC(ALLNODES) | |
6736 | SD_INIT_FUNC(NODE) | |
6737 | #endif | |
6738 | #ifdef CONFIG_SCHED_SMT | |
6739 | SD_INIT_FUNC(SIBLING) | |
6740 | #endif | |
6741 | #ifdef CONFIG_SCHED_MC | |
6742 | SD_INIT_FUNC(MC) | |
6743 | #endif | |
6744 | ||
1d3504fc HS |
6745 | static int default_relax_domain_level = -1; |
6746 | ||
6747 | static int __init setup_relax_domain_level(char *str) | |
6748 | { | |
30e0e178 LZ |
6749 | unsigned long val; |
6750 | ||
6751 | val = simple_strtoul(str, NULL, 0); | |
6752 | if (val < SD_LV_MAX) | |
6753 | default_relax_domain_level = val; | |
6754 | ||
1d3504fc HS |
6755 | return 1; |
6756 | } | |
6757 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6758 | ||
6759 | static void set_domain_attribute(struct sched_domain *sd, | |
6760 | struct sched_domain_attr *attr) | |
6761 | { | |
6762 | int request; | |
6763 | ||
6764 | if (!attr || attr->relax_domain_level < 0) { | |
6765 | if (default_relax_domain_level < 0) | |
6766 | return; | |
6767 | else | |
6768 | request = default_relax_domain_level; | |
6769 | } else | |
6770 | request = attr->relax_domain_level; | |
6771 | if (request < sd->level) { | |
6772 | /* turn off idle balance on this domain */ | |
c88d5910 | 6773 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6774 | } else { |
6775 | /* turn on idle balance on this domain */ | |
c88d5910 | 6776 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6777 | } |
6778 | } | |
6779 | ||
2109b99e AH |
6780 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6781 | const struct cpumask *cpu_map) | |
6782 | { | |
6783 | switch (what) { | |
6784 | case sa_sched_groups: | |
6785 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6786 | d->sched_group_nodes = NULL; | |
6787 | case sa_rootdomain: | |
6788 | free_rootdomain(d->rd); /* fall through */ | |
6789 | case sa_tmpmask: | |
6790 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6791 | case sa_send_covered: | |
6792 | free_cpumask_var(d->send_covered); /* fall through */ | |
6793 | case sa_this_core_map: | |
6794 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6795 | case sa_this_sibling_map: | |
6796 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6797 | case sa_nodemask: | |
6798 | free_cpumask_var(d->nodemask); /* fall through */ | |
6799 | case sa_sched_group_nodes: | |
d1b55138 | 6800 | #ifdef CONFIG_NUMA |
2109b99e AH |
6801 | kfree(d->sched_group_nodes); /* fall through */ |
6802 | case sa_notcovered: | |
6803 | free_cpumask_var(d->notcovered); /* fall through */ | |
6804 | case sa_covered: | |
6805 | free_cpumask_var(d->covered); /* fall through */ | |
6806 | case sa_domainspan: | |
6807 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6808 | #endif |
2109b99e AH |
6809 | case sa_none: |
6810 | break; | |
6811 | } | |
6812 | } | |
3404c8d9 | 6813 | |
2109b99e AH |
6814 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6815 | const struct cpumask *cpu_map) | |
6816 | { | |
3404c8d9 | 6817 | #ifdef CONFIG_NUMA |
2109b99e AH |
6818 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6819 | return sa_none; | |
6820 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6821 | return sa_domainspan; | |
6822 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6823 | return sa_covered; | |
6824 | /* Allocate the per-node list of sched groups */ | |
6825 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6826 | sizeof(struct sched_group *), GFP_KERNEL); | |
6827 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6828 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6829 | return sa_notcovered; |
d1b55138 | 6830 | } |
2109b99e | 6831 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6832 | #endif |
2109b99e AH |
6833 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6834 | return sa_sched_group_nodes; | |
6835 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6836 | return sa_nodemask; | |
6837 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6838 | return sa_this_sibling_map; | |
6839 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6840 | return sa_this_core_map; | |
6841 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6842 | return sa_send_covered; | |
6843 | d->rd = alloc_rootdomain(); | |
6844 | if (!d->rd) { | |
3df0fc5b | 6845 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6846 | return sa_tmpmask; |
57d885fe | 6847 | } |
2109b99e AH |
6848 | return sa_rootdomain; |
6849 | } | |
57d885fe | 6850 | |
7f4588f3 AH |
6851 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6852 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6853 | { | |
6854 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6855 | #ifdef CONFIG_NUMA |
7f4588f3 | 6856 | struct sched_domain *parent; |
1da177e4 | 6857 | |
7f4588f3 AH |
6858 | d->sd_allnodes = 0; |
6859 | if (cpumask_weight(cpu_map) > | |
6860 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6861 | sd = &per_cpu(allnodes_domains, i).sd; | |
6862 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6863 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6864 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6865 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6866 | d->sd_allnodes = 1; | |
6867 | } | |
6868 | parent = sd; | |
6869 | ||
6870 | sd = &per_cpu(node_domains, i).sd; | |
6871 | SD_INIT(sd, NODE); | |
6872 | set_domain_attribute(sd, attr); | |
6873 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6874 | sd->parent = parent; | |
6875 | if (parent) | |
6876 | parent->child = sd; | |
6877 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6878 | #endif |
7f4588f3 AH |
6879 | return sd; |
6880 | } | |
1da177e4 | 6881 | |
87cce662 AH |
6882 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6883 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6884 | struct sched_domain *parent, int i) | |
6885 | { | |
6886 | struct sched_domain *sd; | |
6887 | sd = &per_cpu(phys_domains, i).sd; | |
6888 | SD_INIT(sd, CPU); | |
6889 | set_domain_attribute(sd, attr); | |
6890 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6891 | sd->parent = parent; | |
6892 | if (parent) | |
6893 | parent->child = sd; | |
6894 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6895 | return sd; | |
6896 | } | |
1da177e4 | 6897 | |
410c4081 AH |
6898 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
6899 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6900 | struct sched_domain *parent, int i) | |
6901 | { | |
6902 | struct sched_domain *sd = parent; | |
1e9f28fa | 6903 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
6904 | sd = &per_cpu(core_domains, i).sd; |
6905 | SD_INIT(sd, MC); | |
6906 | set_domain_attribute(sd, attr); | |
6907 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
6908 | sd->parent = parent; | |
6909 | parent->child = sd; | |
6910 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 6911 | #endif |
410c4081 AH |
6912 | return sd; |
6913 | } | |
1e9f28fa | 6914 | |
d8173535 AH |
6915 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
6916 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6917 | struct sched_domain *parent, int i) | |
6918 | { | |
6919 | struct sched_domain *sd = parent; | |
1da177e4 | 6920 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
6921 | sd = &per_cpu(cpu_domains, i).sd; |
6922 | SD_INIT(sd, SIBLING); | |
6923 | set_domain_attribute(sd, attr); | |
6924 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
6925 | sd->parent = parent; | |
6926 | parent->child = sd; | |
6927 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 6928 | #endif |
d8173535 AH |
6929 | return sd; |
6930 | } | |
1da177e4 | 6931 | |
0e8e85c9 AH |
6932 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
6933 | const struct cpumask *cpu_map, int cpu) | |
6934 | { | |
6935 | switch (l) { | |
1da177e4 | 6936 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
6937 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
6938 | cpumask_and(d->this_sibling_map, cpu_map, | |
6939 | topology_thread_cpumask(cpu)); | |
6940 | if (cpu == cpumask_first(d->this_sibling_map)) | |
6941 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
6942 | &cpu_to_cpu_group, | |
6943 | d->send_covered, d->tmpmask); | |
6944 | break; | |
1da177e4 | 6945 | #endif |
1e9f28fa | 6946 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
6947 | case SD_LV_MC: /* set up multi-core groups */ |
6948 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
6949 | if (cpu == cpumask_first(d->this_core_map)) | |
6950 | init_sched_build_groups(d->this_core_map, cpu_map, | |
6951 | &cpu_to_core_group, | |
6952 | d->send_covered, d->tmpmask); | |
6953 | break; | |
1e9f28fa | 6954 | #endif |
86548096 AH |
6955 | case SD_LV_CPU: /* set up physical groups */ |
6956 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
6957 | if (!cpumask_empty(d->nodemask)) | |
6958 | init_sched_build_groups(d->nodemask, cpu_map, | |
6959 | &cpu_to_phys_group, | |
6960 | d->send_covered, d->tmpmask); | |
6961 | break; | |
1da177e4 | 6962 | #ifdef CONFIG_NUMA |
de616e36 AH |
6963 | case SD_LV_ALLNODES: |
6964 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
6965 | d->send_covered, d->tmpmask); | |
6966 | break; | |
6967 | #endif | |
0e8e85c9 AH |
6968 | default: |
6969 | break; | |
7c16ec58 | 6970 | } |
0e8e85c9 | 6971 | } |
9c1cfda2 | 6972 | |
2109b99e AH |
6973 | /* |
6974 | * Build sched domains for a given set of cpus and attach the sched domains | |
6975 | * to the individual cpus | |
6976 | */ | |
6977 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
6978 | struct sched_domain_attr *attr) | |
6979 | { | |
6980 | enum s_alloc alloc_state = sa_none; | |
6981 | struct s_data d; | |
294b0c96 | 6982 | struct sched_domain *sd; |
2109b99e | 6983 | int i; |
7c16ec58 | 6984 | #ifdef CONFIG_NUMA |
2109b99e | 6985 | d.sd_allnodes = 0; |
7c16ec58 | 6986 | #endif |
9c1cfda2 | 6987 | |
2109b99e AH |
6988 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
6989 | if (alloc_state != sa_rootdomain) | |
6990 | goto error; | |
6991 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 6992 | |
1da177e4 | 6993 | /* |
1a20ff27 | 6994 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 6995 | */ |
abcd083a | 6996 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
6997 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
6998 | cpu_map); | |
9761eea8 | 6999 | |
7f4588f3 | 7000 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7001 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7002 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7003 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7004 | } |
9c1cfda2 | 7005 | |
abcd083a | 7006 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7007 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 7008 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7009 | } |
9c1cfda2 | 7010 | |
1da177e4 | 7011 | /* Set up physical groups */ |
86548096 AH |
7012 | for (i = 0; i < nr_node_ids; i++) |
7013 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7014 | |
1da177e4 LT |
7015 | #ifdef CONFIG_NUMA |
7016 | /* Set up node groups */ | |
de616e36 AH |
7017 | if (d.sd_allnodes) |
7018 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7019 | |
0601a88d AH |
7020 | for (i = 0; i < nr_node_ids; i++) |
7021 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7022 | goto error; |
1da177e4 LT |
7023 | #endif |
7024 | ||
7025 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7026 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7027 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7028 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7029 | init_sched_groups_power(i, sd); |
5c45bf27 | 7030 | } |
1da177e4 | 7031 | #endif |
1e9f28fa | 7032 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7033 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7034 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7035 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7036 | } |
7037 | #endif | |
1e9f28fa | 7038 | |
abcd083a | 7039 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7040 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7041 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7042 | } |
7043 | ||
9c1cfda2 | 7044 | #ifdef CONFIG_NUMA |
076ac2af | 7045 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7046 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7047 | |
49a02c51 | 7048 | if (d.sd_allnodes) { |
6711cab4 | 7049 | struct sched_group *sg; |
f712c0c7 | 7050 | |
96f874e2 | 7051 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7052 | d.tmpmask); |
f712c0c7 SS |
7053 | init_numa_sched_groups_power(sg); |
7054 | } | |
9c1cfda2 JH |
7055 | #endif |
7056 | ||
1da177e4 | 7057 | /* Attach the domains */ |
abcd083a | 7058 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7059 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7060 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7061 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7062 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7063 | #else |
6c99e9ad | 7064 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7065 | #endif |
49a02c51 | 7066 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7067 | } |
51888ca2 | 7068 | |
2109b99e AH |
7069 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7070 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7071 | return 0; | |
51888ca2 | 7072 | |
51888ca2 | 7073 | error: |
2109b99e AH |
7074 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7075 | return -ENOMEM; | |
1da177e4 | 7076 | } |
029190c5 | 7077 | |
96f874e2 | 7078 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7079 | { |
7080 | return __build_sched_domains(cpu_map, NULL); | |
7081 | } | |
7082 | ||
acc3f5d7 | 7083 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7084 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7085 | static struct sched_domain_attr *dattr_cur; |
7086 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7087 | |
7088 | /* | |
7089 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7090 | * cpumask) fails, then fallback to a single sched domain, |
7091 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7092 | */ |
4212823f | 7093 | static cpumask_var_t fallback_doms; |
029190c5 | 7094 | |
ee79d1bd HC |
7095 | /* |
7096 | * arch_update_cpu_topology lets virtualized architectures update the | |
7097 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7098 | * or 0 if it stayed the same. | |
7099 | */ | |
7100 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7101 | { |
ee79d1bd | 7102 | return 0; |
22e52b07 HC |
7103 | } |
7104 | ||
acc3f5d7 RR |
7105 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7106 | { | |
7107 | int i; | |
7108 | cpumask_var_t *doms; | |
7109 | ||
7110 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7111 | if (!doms) | |
7112 | return NULL; | |
7113 | for (i = 0; i < ndoms; i++) { | |
7114 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7115 | free_sched_domains(doms, i); | |
7116 | return NULL; | |
7117 | } | |
7118 | } | |
7119 | return doms; | |
7120 | } | |
7121 | ||
7122 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7123 | { | |
7124 | unsigned int i; | |
7125 | for (i = 0; i < ndoms; i++) | |
7126 | free_cpumask_var(doms[i]); | |
7127 | kfree(doms); | |
7128 | } | |
7129 | ||
1a20ff27 | 7130 | /* |
41a2d6cf | 7131 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7132 | * For now this just excludes isolated cpus, but could be used to |
7133 | * exclude other special cases in the future. | |
1a20ff27 | 7134 | */ |
96f874e2 | 7135 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7136 | { |
7378547f MM |
7137 | int err; |
7138 | ||
22e52b07 | 7139 | arch_update_cpu_topology(); |
029190c5 | 7140 | ndoms_cur = 1; |
acc3f5d7 | 7141 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7142 | if (!doms_cur) |
acc3f5d7 RR |
7143 | doms_cur = &fallback_doms; |
7144 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7145 | dattr_cur = NULL; |
acc3f5d7 | 7146 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7147 | register_sched_domain_sysctl(); |
7378547f MM |
7148 | |
7149 | return err; | |
1a20ff27 DG |
7150 | } |
7151 | ||
96f874e2 RR |
7152 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7153 | struct cpumask *tmpmask) | |
1da177e4 | 7154 | { |
7c16ec58 | 7155 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7156 | } |
1da177e4 | 7157 | |
1a20ff27 DG |
7158 | /* |
7159 | * Detach sched domains from a group of cpus specified in cpu_map | |
7160 | * These cpus will now be attached to the NULL domain | |
7161 | */ | |
96f874e2 | 7162 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7163 | { |
96f874e2 RR |
7164 | /* Save because hotplug lock held. */ |
7165 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7166 | int i; |
7167 | ||
abcd083a | 7168 | for_each_cpu(i, cpu_map) |
57d885fe | 7169 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7170 | synchronize_sched(); |
96f874e2 | 7171 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7172 | } |
7173 | ||
1d3504fc HS |
7174 | /* handle null as "default" */ |
7175 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7176 | struct sched_domain_attr *new, int idx_new) | |
7177 | { | |
7178 | struct sched_domain_attr tmp; | |
7179 | ||
7180 | /* fast path */ | |
7181 | if (!new && !cur) | |
7182 | return 1; | |
7183 | ||
7184 | tmp = SD_ATTR_INIT; | |
7185 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7186 | new ? (new + idx_new) : &tmp, | |
7187 | sizeof(struct sched_domain_attr)); | |
7188 | } | |
7189 | ||
029190c5 PJ |
7190 | /* |
7191 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7192 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7193 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7194 | * It destroys each deleted domain and builds each new domain. | |
7195 | * | |
acc3f5d7 | 7196 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7197 | * The masks don't intersect (don't overlap.) We should setup one |
7198 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7199 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7200 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7201 | * it as it is. | |
7202 | * | |
acc3f5d7 RR |
7203 | * The passed in 'doms_new' should be allocated using |
7204 | * alloc_sched_domains. This routine takes ownership of it and will | |
7205 | * free_sched_domains it when done with it. If the caller failed the | |
7206 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7207 | * and partition_sched_domains() will fallback to the single partition | |
7208 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7209 | * |
96f874e2 | 7210 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7211 | * ndoms_new == 0 is a special case for destroying existing domains, |
7212 | * and it will not create the default domain. | |
dfb512ec | 7213 | * |
029190c5 PJ |
7214 | * Call with hotplug lock held |
7215 | */ | |
acc3f5d7 | 7216 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7217 | struct sched_domain_attr *dattr_new) |
029190c5 | 7218 | { |
dfb512ec | 7219 | int i, j, n; |
d65bd5ec | 7220 | int new_topology; |
029190c5 | 7221 | |
712555ee | 7222 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7223 | |
7378547f MM |
7224 | /* always unregister in case we don't destroy any domains */ |
7225 | unregister_sched_domain_sysctl(); | |
7226 | ||
d65bd5ec HC |
7227 | /* Let architecture update cpu core mappings. */ |
7228 | new_topology = arch_update_cpu_topology(); | |
7229 | ||
dfb512ec | 7230 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7231 | |
7232 | /* Destroy deleted domains */ | |
7233 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7234 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7235 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7236 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7237 | goto match1; |
7238 | } | |
7239 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7240 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7241 | match1: |
7242 | ; | |
7243 | } | |
7244 | ||
e761b772 MK |
7245 | if (doms_new == NULL) { |
7246 | ndoms_cur = 0; | |
acc3f5d7 | 7247 | doms_new = &fallback_doms; |
6ad4c188 | 7248 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7249 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7250 | } |
7251 | ||
029190c5 PJ |
7252 | /* Build new domains */ |
7253 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7254 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7255 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7256 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7257 | goto match2; |
7258 | } | |
7259 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7260 | __build_sched_domains(doms_new[i], |
1d3504fc | 7261 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7262 | match2: |
7263 | ; | |
7264 | } | |
7265 | ||
7266 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7267 | if (doms_cur != &fallback_doms) |
7268 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7269 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7270 | doms_cur = doms_new; |
1d3504fc | 7271 | dattr_cur = dattr_new; |
029190c5 | 7272 | ndoms_cur = ndoms_new; |
7378547f MM |
7273 | |
7274 | register_sched_domain_sysctl(); | |
a1835615 | 7275 | |
712555ee | 7276 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7277 | } |
7278 | ||
5c45bf27 | 7279 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7280 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7281 | { |
95402b38 | 7282 | get_online_cpus(); |
dfb512ec MK |
7283 | |
7284 | /* Destroy domains first to force the rebuild */ | |
7285 | partition_sched_domains(0, NULL, NULL); | |
7286 | ||
e761b772 | 7287 | rebuild_sched_domains(); |
95402b38 | 7288 | put_online_cpus(); |
5c45bf27 SS |
7289 | } |
7290 | ||
7291 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7292 | { | |
afb8a9b7 | 7293 | unsigned int level = 0; |
5c45bf27 | 7294 | |
afb8a9b7 GS |
7295 | if (sscanf(buf, "%u", &level) != 1) |
7296 | return -EINVAL; | |
7297 | ||
7298 | /* | |
7299 | * level is always be positive so don't check for | |
7300 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7301 | * What happens on 0 or 1 byte write, | |
7302 | * need to check for count as well? | |
7303 | */ | |
7304 | ||
7305 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7306 | return -EINVAL; |
7307 | ||
7308 | if (smt) | |
afb8a9b7 | 7309 | sched_smt_power_savings = level; |
5c45bf27 | 7310 | else |
afb8a9b7 | 7311 | sched_mc_power_savings = level; |
5c45bf27 | 7312 | |
c70f22d2 | 7313 | arch_reinit_sched_domains(); |
5c45bf27 | 7314 | |
c70f22d2 | 7315 | return count; |
5c45bf27 SS |
7316 | } |
7317 | ||
5c45bf27 | 7318 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7319 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7320 | struct sysdev_class_attribute *attr, |
f718cd4a | 7321 | char *page) |
5c45bf27 SS |
7322 | { |
7323 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7324 | } | |
f718cd4a | 7325 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7326 | struct sysdev_class_attribute *attr, |
48f24c4d | 7327 | const char *buf, size_t count) |
5c45bf27 SS |
7328 | { |
7329 | return sched_power_savings_store(buf, count, 0); | |
7330 | } | |
f718cd4a AK |
7331 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7332 | sched_mc_power_savings_show, | |
7333 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7334 | #endif |
7335 | ||
7336 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7337 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7338 | struct sysdev_class_attribute *attr, |
f718cd4a | 7339 | char *page) |
5c45bf27 SS |
7340 | { |
7341 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7342 | } | |
f718cd4a | 7343 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7344 | struct sysdev_class_attribute *attr, |
48f24c4d | 7345 | const char *buf, size_t count) |
5c45bf27 SS |
7346 | { |
7347 | return sched_power_savings_store(buf, count, 1); | |
7348 | } | |
f718cd4a AK |
7349 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7350 | sched_smt_power_savings_show, | |
6707de00 AB |
7351 | sched_smt_power_savings_store); |
7352 | #endif | |
7353 | ||
39aac648 | 7354 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7355 | { |
7356 | int err = 0; | |
7357 | ||
7358 | #ifdef CONFIG_SCHED_SMT | |
7359 | if (smt_capable()) | |
7360 | err = sysfs_create_file(&cls->kset.kobj, | |
7361 | &attr_sched_smt_power_savings.attr); | |
7362 | #endif | |
7363 | #ifdef CONFIG_SCHED_MC | |
7364 | if (!err && mc_capable()) | |
7365 | err = sysfs_create_file(&cls->kset.kobj, | |
7366 | &attr_sched_mc_power_savings.attr); | |
7367 | #endif | |
7368 | return err; | |
7369 | } | |
6d6bc0ad | 7370 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7371 | |
1da177e4 | 7372 | /* |
3a101d05 TH |
7373 | * Update cpusets according to cpu_active mask. If cpusets are |
7374 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7375 | * around partition_sched_domains(). | |
1da177e4 | 7376 | */ |
3a101d05 TH |
7377 | static int __cpuexit cpuset_cpu_active(struct notifier_block *nfb, |
7378 | unsigned long action, void *hcpu) | |
e761b772 | 7379 | { |
3a101d05 | 7380 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7381 | case CPU_ONLINE: |
6ad4c188 | 7382 | case CPU_DOWN_FAILED: |
3a101d05 | 7383 | cpuset_update_active_cpus(); |
e761b772 | 7384 | return NOTIFY_OK; |
3a101d05 TH |
7385 | default: |
7386 | return NOTIFY_DONE; | |
7387 | } | |
7388 | } | |
e761b772 | 7389 | |
3a101d05 TH |
7390 | static int __cpuexit cpuset_cpu_inactive(struct notifier_block *nfb, |
7391 | unsigned long action, void *hcpu) | |
7392 | { | |
7393 | switch (action & ~CPU_TASKS_FROZEN) { | |
7394 | case CPU_DOWN_PREPARE: | |
7395 | cpuset_update_active_cpus(); | |
7396 | return NOTIFY_OK; | |
e761b772 MK |
7397 | default: |
7398 | return NOTIFY_DONE; | |
7399 | } | |
7400 | } | |
e761b772 MK |
7401 | |
7402 | static int update_runtime(struct notifier_block *nfb, | |
7403 | unsigned long action, void *hcpu) | |
1da177e4 | 7404 | { |
7def2be1 PZ |
7405 | int cpu = (int)(long)hcpu; |
7406 | ||
1da177e4 | 7407 | switch (action) { |
1da177e4 | 7408 | case CPU_DOWN_PREPARE: |
8bb78442 | 7409 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7410 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7411 | return NOTIFY_OK; |
7412 | ||
1da177e4 | 7413 | case CPU_DOWN_FAILED: |
8bb78442 | 7414 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7415 | case CPU_ONLINE: |
8bb78442 | 7416 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7417 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7418 | return NOTIFY_OK; |
7419 | ||
1da177e4 LT |
7420 | default: |
7421 | return NOTIFY_DONE; | |
7422 | } | |
1da177e4 | 7423 | } |
1da177e4 LT |
7424 | |
7425 | void __init sched_init_smp(void) | |
7426 | { | |
dcc30a35 RR |
7427 | cpumask_var_t non_isolated_cpus; |
7428 | ||
7429 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7430 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7431 | |
434d53b0 MT |
7432 | #if defined(CONFIG_NUMA) |
7433 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7434 | GFP_KERNEL); | |
7435 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7436 | #endif | |
95402b38 | 7437 | get_online_cpus(); |
712555ee | 7438 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7439 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7440 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7441 | if (cpumask_empty(non_isolated_cpus)) | |
7442 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7443 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7444 | put_online_cpus(); |
e761b772 | 7445 | |
3a101d05 TH |
7446 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7447 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7448 | |
7449 | /* RT runtime code needs to handle some hotplug events */ | |
7450 | hotcpu_notifier(update_runtime, 0); | |
7451 | ||
b328ca18 | 7452 | init_hrtick(); |
5c1e1767 NP |
7453 | |
7454 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7455 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7456 | BUG(); |
19978ca6 | 7457 | sched_init_granularity(); |
dcc30a35 | 7458 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7459 | |
0e3900e6 | 7460 | init_sched_rt_class(); |
1da177e4 LT |
7461 | } |
7462 | #else | |
7463 | void __init sched_init_smp(void) | |
7464 | { | |
19978ca6 | 7465 | sched_init_granularity(); |
1da177e4 LT |
7466 | } |
7467 | #endif /* CONFIG_SMP */ | |
7468 | ||
cd1bb94b AB |
7469 | const_debug unsigned int sysctl_timer_migration = 1; |
7470 | ||
1da177e4 LT |
7471 | int in_sched_functions(unsigned long addr) |
7472 | { | |
1da177e4 LT |
7473 | return in_lock_functions(addr) || |
7474 | (addr >= (unsigned long)__sched_text_start | |
7475 | && addr < (unsigned long)__sched_text_end); | |
7476 | } | |
7477 | ||
a9957449 | 7478 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7479 | { |
7480 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7481 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7482 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7483 | cfs_rq->rq = rq; | |
7484 | #endif | |
67e9fb2a | 7485 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7486 | } |
7487 | ||
fa85ae24 PZ |
7488 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7489 | { | |
7490 | struct rt_prio_array *array; | |
7491 | int i; | |
7492 | ||
7493 | array = &rt_rq->active; | |
7494 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7495 | INIT_LIST_HEAD(array->queue + i); | |
7496 | __clear_bit(i, array->bitmap); | |
7497 | } | |
7498 | /* delimiter for bitsearch: */ | |
7499 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7500 | ||
052f1dc7 | 7501 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7502 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7503 | #ifdef CONFIG_SMP |
e864c499 | 7504 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7505 | #endif |
48d5e258 | 7506 | #endif |
fa85ae24 PZ |
7507 | #ifdef CONFIG_SMP |
7508 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7509 | rt_rq->overloaded = 0; |
05fa785c | 7510 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7511 | #endif |
7512 | ||
7513 | rt_rq->rt_time = 0; | |
7514 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7515 | rt_rq->rt_runtime = 0; |
0986b11b | 7516 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7517 | |
052f1dc7 | 7518 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7519 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7520 | rt_rq->rq = rq; |
7521 | #endif | |
fa85ae24 PZ |
7522 | } |
7523 | ||
6f505b16 | 7524 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7525 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7526 | struct sched_entity *se, int cpu, int add, | |
7527 | struct sched_entity *parent) | |
6f505b16 | 7528 | { |
ec7dc8ac | 7529 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7530 | tg->cfs_rq[cpu] = cfs_rq; |
7531 | init_cfs_rq(cfs_rq, rq); | |
7532 | cfs_rq->tg = tg; | |
7533 | if (add) | |
7534 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7535 | ||
7536 | tg->se[cpu] = se; | |
354d60c2 DG |
7537 | /* se could be NULL for init_task_group */ |
7538 | if (!se) | |
7539 | return; | |
7540 | ||
ec7dc8ac DG |
7541 | if (!parent) |
7542 | se->cfs_rq = &rq->cfs; | |
7543 | else | |
7544 | se->cfs_rq = parent->my_q; | |
7545 | ||
6f505b16 PZ |
7546 | se->my_q = cfs_rq; |
7547 | se->load.weight = tg->shares; | |
e05510d0 | 7548 | se->load.inv_weight = 0; |
ec7dc8ac | 7549 | se->parent = parent; |
6f505b16 | 7550 | } |
052f1dc7 | 7551 | #endif |
6f505b16 | 7552 | |
052f1dc7 | 7553 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7554 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7555 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7556 | struct sched_rt_entity *parent) | |
6f505b16 | 7557 | { |
ec7dc8ac DG |
7558 | struct rq *rq = cpu_rq(cpu); |
7559 | ||
6f505b16 PZ |
7560 | tg->rt_rq[cpu] = rt_rq; |
7561 | init_rt_rq(rt_rq, rq); | |
7562 | rt_rq->tg = tg; | |
ac086bc2 | 7563 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7564 | if (add) |
7565 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7566 | ||
7567 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7568 | if (!rt_se) |
7569 | return; | |
7570 | ||
ec7dc8ac DG |
7571 | if (!parent) |
7572 | rt_se->rt_rq = &rq->rt; | |
7573 | else | |
7574 | rt_se->rt_rq = parent->my_q; | |
7575 | ||
6f505b16 | 7576 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7577 | rt_se->parent = parent; |
6f505b16 PZ |
7578 | INIT_LIST_HEAD(&rt_se->run_list); |
7579 | } | |
7580 | #endif | |
7581 | ||
1da177e4 LT |
7582 | void __init sched_init(void) |
7583 | { | |
dd41f596 | 7584 | int i, j; |
434d53b0 MT |
7585 | unsigned long alloc_size = 0, ptr; |
7586 | ||
7587 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7588 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7589 | #endif | |
7590 | #ifdef CONFIG_RT_GROUP_SCHED | |
7591 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7592 | #endif |
df7c8e84 | 7593 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7594 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7595 | #endif |
434d53b0 | 7596 | if (alloc_size) { |
36b7b6d4 | 7597 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7598 | |
7599 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7600 | init_task_group.se = (struct sched_entity **)ptr; | |
7601 | ptr += nr_cpu_ids * sizeof(void **); | |
7602 | ||
7603 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7604 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7605 | |
6d6bc0ad | 7606 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7607 | #ifdef CONFIG_RT_GROUP_SCHED |
7608 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7609 | ptr += nr_cpu_ids * sizeof(void **); | |
7610 | ||
7611 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7612 | ptr += nr_cpu_ids * sizeof(void **); |
7613 | ||
6d6bc0ad | 7614 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7615 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7616 | for_each_possible_cpu(i) { | |
7617 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7618 | ptr += cpumask_size(); | |
7619 | } | |
7620 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7621 | } |
dd41f596 | 7622 | |
57d885fe GH |
7623 | #ifdef CONFIG_SMP |
7624 | init_defrootdomain(); | |
7625 | #endif | |
7626 | ||
d0b27fa7 PZ |
7627 | init_rt_bandwidth(&def_rt_bandwidth, |
7628 | global_rt_period(), global_rt_runtime()); | |
7629 | ||
7630 | #ifdef CONFIG_RT_GROUP_SCHED | |
7631 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7632 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7633 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7634 | |
7c941438 | 7635 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7636 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7637 | INIT_LIST_HEAD(&init_task_group.children); |
7638 | ||
7c941438 | 7639 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7640 | |
4a6cc4bd JK |
7641 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7642 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7643 | __alignof__(unsigned long)); | |
7644 | #endif | |
0a945022 | 7645 | for_each_possible_cpu(i) { |
70b97a7f | 7646 | struct rq *rq; |
1da177e4 LT |
7647 | |
7648 | rq = cpu_rq(i); | |
05fa785c | 7649 | raw_spin_lock_init(&rq->lock); |
7897986b | 7650 | rq->nr_running = 0; |
dce48a84 TG |
7651 | rq->calc_load_active = 0; |
7652 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7653 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7654 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7655 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7656 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7657 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7658 | #ifdef CONFIG_CGROUP_SCHED |
7659 | /* | |
7660 | * How much cpu bandwidth does init_task_group get? | |
7661 | * | |
7662 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7663 | * gets 100% of the cpu resources in the system. This overall | |
7664 | * system cpu resource is divided among the tasks of | |
7665 | * init_task_group and its child task-groups in a fair manner, | |
7666 | * based on each entity's (task or task-group's) weight | |
7667 | * (se->load.weight). | |
7668 | * | |
7669 | * In other words, if init_task_group has 10 tasks of weight | |
7670 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7671 | * then A0's share of the cpu resource is: | |
7672 | * | |
0d905bca | 7673 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7674 | * |
7675 | * We achieve this by letting init_task_group's tasks sit | |
7676 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7677 | */ | |
ec7dc8ac | 7678 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7679 | #endif |
354d60c2 DG |
7680 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7681 | ||
7682 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7683 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7684 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7685 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7686 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7687 | #endif |
dd41f596 | 7688 | #endif |
1da177e4 | 7689 | |
dd41f596 IM |
7690 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7691 | rq->cpu_load[j] = 0; | |
1da177e4 | 7692 | #ifdef CONFIG_SMP |
41c7ce9a | 7693 | rq->sd = NULL; |
57d885fe | 7694 | rq->rd = NULL; |
e51fd5e2 | 7695 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 7696 | rq->post_schedule = 0; |
1da177e4 | 7697 | rq->active_balance = 0; |
dd41f596 | 7698 | rq->next_balance = jiffies; |
1da177e4 | 7699 | rq->push_cpu = 0; |
0a2966b4 | 7700 | rq->cpu = i; |
1f11eb6a | 7701 | rq->online = 0; |
eae0c9df MG |
7702 | rq->idle_stamp = 0; |
7703 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7704 | rq_attach_root(rq, &def_root_domain); |
1da177e4 | 7705 | #endif |
8f4d37ec | 7706 | init_rq_hrtick(rq); |
1da177e4 | 7707 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7708 | } |
7709 | ||
2dd73a4f | 7710 | set_load_weight(&init_task); |
b50f60ce | 7711 | |
e107be36 AK |
7712 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7713 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7714 | #endif | |
7715 | ||
c9819f45 | 7716 | #ifdef CONFIG_SMP |
962cf36c | 7717 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7718 | #endif |
7719 | ||
b50f60ce | 7720 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7721 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7722 | #endif |
7723 | ||
1da177e4 LT |
7724 | /* |
7725 | * The boot idle thread does lazy MMU switching as well: | |
7726 | */ | |
7727 | atomic_inc(&init_mm.mm_count); | |
7728 | enter_lazy_tlb(&init_mm, current); | |
7729 | ||
7730 | /* | |
7731 | * Make us the idle thread. Technically, schedule() should not be | |
7732 | * called from this thread, however somewhere below it might be, | |
7733 | * but because we are the idle thread, we just pick up running again | |
7734 | * when this runqueue becomes "idle". | |
7735 | */ | |
7736 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7737 | |
7738 | calc_load_update = jiffies + LOAD_FREQ; | |
7739 | ||
dd41f596 IM |
7740 | /* |
7741 | * During early bootup we pretend to be a normal task: | |
7742 | */ | |
7743 | current->sched_class = &fair_sched_class; | |
6892b75e | 7744 | |
6a7b3dc3 | 7745 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7746 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7747 | #ifdef CONFIG_SMP |
7d1e6a9b | 7748 | #ifdef CONFIG_NO_HZ |
49557e62 | 7749 | zalloc_cpumask_var(&nohz.cpu_mask, GFP_NOWAIT); |
4bdddf8f | 7750 | alloc_cpumask_var(&nohz.ilb_grp_nohz_mask, GFP_NOWAIT); |
7d1e6a9b | 7751 | #endif |
bdddd296 RR |
7752 | /* May be allocated at isolcpus cmdline parse time */ |
7753 | if (cpu_isolated_map == NULL) | |
7754 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7755 | #endif /* SMP */ |
6a7b3dc3 | 7756 | |
cdd6c482 | 7757 | perf_event_init(); |
0d905bca | 7758 | |
6892b75e | 7759 | scheduler_running = 1; |
1da177e4 LT |
7760 | } |
7761 | ||
7762 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7763 | static inline int preempt_count_equals(int preempt_offset) |
7764 | { | |
234da7bc | 7765 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7766 | |
7767 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7768 | } | |
7769 | ||
d894837f | 7770 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7771 | { |
48f24c4d | 7772 | #ifdef in_atomic |
1da177e4 LT |
7773 | static unsigned long prev_jiffy; /* ratelimiting */ |
7774 | ||
e4aafea2 FW |
7775 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7776 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7777 | return; |
7778 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7779 | return; | |
7780 | prev_jiffy = jiffies; | |
7781 | ||
3df0fc5b PZ |
7782 | printk(KERN_ERR |
7783 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7784 | file, line); | |
7785 | printk(KERN_ERR | |
7786 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7787 | in_atomic(), irqs_disabled(), | |
7788 | current->pid, current->comm); | |
aef745fc IM |
7789 | |
7790 | debug_show_held_locks(current); | |
7791 | if (irqs_disabled()) | |
7792 | print_irqtrace_events(current); | |
7793 | dump_stack(); | |
1da177e4 LT |
7794 | #endif |
7795 | } | |
7796 | EXPORT_SYMBOL(__might_sleep); | |
7797 | #endif | |
7798 | ||
7799 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7800 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7801 | { | |
7802 | int on_rq; | |
3e51f33f | 7803 | |
3a5e4dc1 AK |
7804 | on_rq = p->se.on_rq; |
7805 | if (on_rq) | |
7806 | deactivate_task(rq, p, 0); | |
7807 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7808 | if (on_rq) { | |
7809 | activate_task(rq, p, 0); | |
7810 | resched_task(rq->curr); | |
7811 | } | |
7812 | } | |
7813 | ||
1da177e4 LT |
7814 | void normalize_rt_tasks(void) |
7815 | { | |
a0f98a1c | 7816 | struct task_struct *g, *p; |
1da177e4 | 7817 | unsigned long flags; |
70b97a7f | 7818 | struct rq *rq; |
1da177e4 | 7819 | |
4cf5d77a | 7820 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7821 | do_each_thread(g, p) { |
178be793 IM |
7822 | /* |
7823 | * Only normalize user tasks: | |
7824 | */ | |
7825 | if (!p->mm) | |
7826 | continue; | |
7827 | ||
6cfb0d5d | 7828 | p->se.exec_start = 0; |
6cfb0d5d | 7829 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7830 | p->se.statistics.wait_start = 0; |
7831 | p->se.statistics.sleep_start = 0; | |
7832 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7833 | #endif |
dd41f596 IM |
7834 | |
7835 | if (!rt_task(p)) { | |
7836 | /* | |
7837 | * Renice negative nice level userspace | |
7838 | * tasks back to 0: | |
7839 | */ | |
7840 | if (TASK_NICE(p) < 0 && p->mm) | |
7841 | set_user_nice(p, 0); | |
1da177e4 | 7842 | continue; |
dd41f596 | 7843 | } |
1da177e4 | 7844 | |
1d615482 | 7845 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7846 | rq = __task_rq_lock(p); |
1da177e4 | 7847 | |
178be793 | 7848 | normalize_task(rq, p); |
3a5e4dc1 | 7849 | |
b29739f9 | 7850 | __task_rq_unlock(rq); |
1d615482 | 7851 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7852 | } while_each_thread(g, p); |
7853 | ||
4cf5d77a | 7854 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7855 | } |
7856 | ||
7857 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7858 | |
67fc4e0c | 7859 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7860 | /* |
67fc4e0c | 7861 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7862 | * |
7863 | * They can only be called when the whole system has been | |
7864 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7865 | * activity can take place. Using them for anything else would | |
7866 | * be a serious bug, and as a result, they aren't even visible | |
7867 | * under any other configuration. | |
7868 | */ | |
7869 | ||
7870 | /** | |
7871 | * curr_task - return the current task for a given cpu. | |
7872 | * @cpu: the processor in question. | |
7873 | * | |
7874 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7875 | */ | |
36c8b586 | 7876 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7877 | { |
7878 | return cpu_curr(cpu); | |
7879 | } | |
7880 | ||
67fc4e0c JW |
7881 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7882 | ||
7883 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7884 | /** |
7885 | * set_curr_task - set the current task for a given cpu. | |
7886 | * @cpu: the processor in question. | |
7887 | * @p: the task pointer to set. | |
7888 | * | |
7889 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
7890 | * are serviced on a separate stack. It allows the architecture to switch the |
7891 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
7892 | * must be called with all CPU's synchronized, and interrupts disabled, the |
7893 | * and caller must save the original value of the current task (see | |
7894 | * curr_task() above) and restore that value before reenabling interrupts and | |
7895 | * re-starting the system. | |
7896 | * | |
7897 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7898 | */ | |
36c8b586 | 7899 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
7900 | { |
7901 | cpu_curr(cpu) = p; | |
7902 | } | |
7903 | ||
7904 | #endif | |
29f59db3 | 7905 | |
bccbe08a PZ |
7906 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7907 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
7908 | { |
7909 | int i; | |
7910 | ||
7911 | for_each_possible_cpu(i) { | |
7912 | if (tg->cfs_rq) | |
7913 | kfree(tg->cfs_rq[i]); | |
7914 | if (tg->se) | |
7915 | kfree(tg->se[i]); | |
6f505b16 PZ |
7916 | } |
7917 | ||
7918 | kfree(tg->cfs_rq); | |
7919 | kfree(tg->se); | |
6f505b16 PZ |
7920 | } |
7921 | ||
ec7dc8ac DG |
7922 | static |
7923 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 7924 | { |
29f59db3 | 7925 | struct cfs_rq *cfs_rq; |
eab17229 | 7926 | struct sched_entity *se; |
9b5b7751 | 7927 | struct rq *rq; |
29f59db3 SV |
7928 | int i; |
7929 | ||
434d53b0 | 7930 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7931 | if (!tg->cfs_rq) |
7932 | goto err; | |
434d53b0 | 7933 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
7934 | if (!tg->se) |
7935 | goto err; | |
052f1dc7 PZ |
7936 | |
7937 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
7938 | |
7939 | for_each_possible_cpu(i) { | |
9b5b7751 | 7940 | rq = cpu_rq(i); |
29f59db3 | 7941 | |
eab17229 LZ |
7942 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
7943 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
7944 | if (!cfs_rq) |
7945 | goto err; | |
7946 | ||
eab17229 LZ |
7947 | se = kzalloc_node(sizeof(struct sched_entity), |
7948 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 7949 | if (!se) |
dfc12eb2 | 7950 | goto err_free_rq; |
29f59db3 | 7951 | |
eab17229 | 7952 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
7953 | } |
7954 | ||
7955 | return 1; | |
7956 | ||
dfc12eb2 PC |
7957 | err_free_rq: |
7958 | kfree(cfs_rq); | |
bccbe08a PZ |
7959 | err: |
7960 | return 0; | |
7961 | } | |
7962 | ||
7963 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7964 | { | |
7965 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
7966 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
7967 | } | |
7968 | ||
7969 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7970 | { | |
7971 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
7972 | } | |
6d6bc0ad | 7973 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
7974 | static inline void free_fair_sched_group(struct task_group *tg) |
7975 | { | |
7976 | } | |
7977 | ||
ec7dc8ac DG |
7978 | static inline |
7979 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
7980 | { |
7981 | return 1; | |
7982 | } | |
7983 | ||
7984 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
7985 | { | |
7986 | } | |
7987 | ||
7988 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
7989 | { | |
7990 | } | |
6d6bc0ad | 7991 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
7992 | |
7993 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
7994 | static void free_rt_sched_group(struct task_group *tg) |
7995 | { | |
7996 | int i; | |
7997 | ||
d0b27fa7 PZ |
7998 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
7999 | ||
bccbe08a PZ |
8000 | for_each_possible_cpu(i) { |
8001 | if (tg->rt_rq) | |
8002 | kfree(tg->rt_rq[i]); | |
8003 | if (tg->rt_se) | |
8004 | kfree(tg->rt_se[i]); | |
8005 | } | |
8006 | ||
8007 | kfree(tg->rt_rq); | |
8008 | kfree(tg->rt_se); | |
8009 | } | |
8010 | ||
ec7dc8ac DG |
8011 | static |
8012 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8013 | { |
8014 | struct rt_rq *rt_rq; | |
eab17229 | 8015 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8016 | struct rq *rq; |
8017 | int i; | |
8018 | ||
434d53b0 | 8019 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8020 | if (!tg->rt_rq) |
8021 | goto err; | |
434d53b0 | 8022 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8023 | if (!tg->rt_se) |
8024 | goto err; | |
8025 | ||
d0b27fa7 PZ |
8026 | init_rt_bandwidth(&tg->rt_bandwidth, |
8027 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8028 | |
8029 | for_each_possible_cpu(i) { | |
8030 | rq = cpu_rq(i); | |
8031 | ||
eab17229 LZ |
8032 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8033 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8034 | if (!rt_rq) |
8035 | goto err; | |
29f59db3 | 8036 | |
eab17229 LZ |
8037 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8038 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8039 | if (!rt_se) |
dfc12eb2 | 8040 | goto err_free_rq; |
29f59db3 | 8041 | |
eab17229 | 8042 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8043 | } |
8044 | ||
bccbe08a PZ |
8045 | return 1; |
8046 | ||
dfc12eb2 PC |
8047 | err_free_rq: |
8048 | kfree(rt_rq); | |
bccbe08a PZ |
8049 | err: |
8050 | return 0; | |
8051 | } | |
8052 | ||
8053 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8054 | { | |
8055 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8056 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8057 | } | |
8058 | ||
8059 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8060 | { | |
8061 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8062 | } | |
6d6bc0ad | 8063 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8064 | static inline void free_rt_sched_group(struct task_group *tg) |
8065 | { | |
8066 | } | |
8067 | ||
ec7dc8ac DG |
8068 | static inline |
8069 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8070 | { |
8071 | return 1; | |
8072 | } | |
8073 | ||
8074 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8075 | { | |
8076 | } | |
8077 | ||
8078 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8079 | { | |
8080 | } | |
6d6bc0ad | 8081 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8082 | |
7c941438 | 8083 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8084 | static void free_sched_group(struct task_group *tg) |
8085 | { | |
8086 | free_fair_sched_group(tg); | |
8087 | free_rt_sched_group(tg); | |
8088 | kfree(tg); | |
8089 | } | |
8090 | ||
8091 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8092 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8093 | { |
8094 | struct task_group *tg; | |
8095 | unsigned long flags; | |
8096 | int i; | |
8097 | ||
8098 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8099 | if (!tg) | |
8100 | return ERR_PTR(-ENOMEM); | |
8101 | ||
ec7dc8ac | 8102 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8103 | goto err; |
8104 | ||
ec7dc8ac | 8105 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8106 | goto err; |
8107 | ||
8ed36996 | 8108 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8109 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8110 | register_fair_sched_group(tg, i); |
8111 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8112 | } |
6f505b16 | 8113 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8114 | |
8115 | WARN_ON(!parent); /* root should already exist */ | |
8116 | ||
8117 | tg->parent = parent; | |
f473aa5e | 8118 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8119 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8120 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8121 | |
9b5b7751 | 8122 | return tg; |
29f59db3 SV |
8123 | |
8124 | err: | |
6f505b16 | 8125 | free_sched_group(tg); |
29f59db3 SV |
8126 | return ERR_PTR(-ENOMEM); |
8127 | } | |
8128 | ||
9b5b7751 | 8129 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8130 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8131 | { |
29f59db3 | 8132 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8133 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8134 | } |
8135 | ||
9b5b7751 | 8136 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8137 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8138 | { |
8ed36996 | 8139 | unsigned long flags; |
9b5b7751 | 8140 | int i; |
29f59db3 | 8141 | |
8ed36996 | 8142 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8143 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8144 | unregister_fair_sched_group(tg, i); |
8145 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8146 | } |
6f505b16 | 8147 | list_del_rcu(&tg->list); |
f473aa5e | 8148 | list_del_rcu(&tg->siblings); |
8ed36996 | 8149 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8150 | |
9b5b7751 | 8151 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8152 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8153 | } |
8154 | ||
9b5b7751 | 8155 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8156 | * The caller of this function should have put the task in its new group |
8157 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8158 | * reflect its new group. | |
9b5b7751 SV |
8159 | */ |
8160 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8161 | { |
8162 | int on_rq, running; | |
8163 | unsigned long flags; | |
8164 | struct rq *rq; | |
8165 | ||
8166 | rq = task_rq_lock(tsk, &flags); | |
8167 | ||
051a1d1a | 8168 | running = task_current(rq, tsk); |
29f59db3 SV |
8169 | on_rq = tsk->se.on_rq; |
8170 | ||
0e1f3483 | 8171 | if (on_rq) |
29f59db3 | 8172 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8173 | if (unlikely(running)) |
8174 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8175 | |
6f505b16 | 8176 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8177 | |
810b3817 PZ |
8178 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8179 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8180 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8181 | #endif |
8182 | ||
0e1f3483 HS |
8183 | if (unlikely(running)) |
8184 | tsk->sched_class->set_curr_task(rq); | |
8185 | if (on_rq) | |
371fd7e7 | 8186 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8187 | |
29f59db3 SV |
8188 | task_rq_unlock(rq, &flags); |
8189 | } | |
7c941438 | 8190 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8191 | |
052f1dc7 | 8192 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8193 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8194 | { |
8195 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8196 | int on_rq; |
8197 | ||
29f59db3 | 8198 | on_rq = se->on_rq; |
62fb1851 | 8199 | if (on_rq) |
29f59db3 SV |
8200 | dequeue_entity(cfs_rq, se, 0); |
8201 | ||
8202 | se->load.weight = shares; | |
e05510d0 | 8203 | se->load.inv_weight = 0; |
29f59db3 | 8204 | |
62fb1851 | 8205 | if (on_rq) |
29f59db3 | 8206 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8207 | } |
62fb1851 | 8208 | |
c09595f6 PZ |
8209 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8210 | { | |
8211 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8212 | struct rq *rq = cfs_rq->rq; | |
8213 | unsigned long flags; | |
8214 | ||
05fa785c | 8215 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8216 | __set_se_shares(se, shares); |
05fa785c | 8217 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8218 | } |
8219 | ||
8ed36996 PZ |
8220 | static DEFINE_MUTEX(shares_mutex); |
8221 | ||
4cf86d77 | 8222 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8223 | { |
8224 | int i; | |
8ed36996 | 8225 | unsigned long flags; |
c61935fd | 8226 | |
ec7dc8ac DG |
8227 | /* |
8228 | * We can't change the weight of the root cgroup. | |
8229 | */ | |
8230 | if (!tg->se[0]) | |
8231 | return -EINVAL; | |
8232 | ||
18d95a28 PZ |
8233 | if (shares < MIN_SHARES) |
8234 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8235 | else if (shares > MAX_SHARES) |
8236 | shares = MAX_SHARES; | |
62fb1851 | 8237 | |
8ed36996 | 8238 | mutex_lock(&shares_mutex); |
9b5b7751 | 8239 | if (tg->shares == shares) |
5cb350ba | 8240 | goto done; |
29f59db3 | 8241 | |
8ed36996 | 8242 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8243 | for_each_possible_cpu(i) |
8244 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8245 | list_del_rcu(&tg->siblings); |
8ed36996 | 8246 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8247 | |
8248 | /* wait for any ongoing reference to this group to finish */ | |
8249 | synchronize_sched(); | |
8250 | ||
8251 | /* | |
8252 | * Now we are free to modify the group's share on each cpu | |
8253 | * w/o tripping rebalance_share or load_balance_fair. | |
8254 | */ | |
9b5b7751 | 8255 | tg->shares = shares; |
c09595f6 PZ |
8256 | for_each_possible_cpu(i) { |
8257 | /* | |
8258 | * force a rebalance | |
8259 | */ | |
8260 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8261 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8262 | } |
29f59db3 | 8263 | |
6b2d7700 SV |
8264 | /* |
8265 | * Enable load balance activity on this group, by inserting it back on | |
8266 | * each cpu's rq->leaf_cfs_rq_list. | |
8267 | */ | |
8ed36996 | 8268 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8269 | for_each_possible_cpu(i) |
8270 | register_fair_sched_group(tg, i); | |
f473aa5e | 8271 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8272 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8273 | done: |
8ed36996 | 8274 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8275 | return 0; |
29f59db3 SV |
8276 | } |
8277 | ||
5cb350ba DG |
8278 | unsigned long sched_group_shares(struct task_group *tg) |
8279 | { | |
8280 | return tg->shares; | |
8281 | } | |
052f1dc7 | 8282 | #endif |
5cb350ba | 8283 | |
052f1dc7 | 8284 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8285 | /* |
9f0c1e56 | 8286 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8287 | */ |
9f0c1e56 PZ |
8288 | static DEFINE_MUTEX(rt_constraints_mutex); |
8289 | ||
8290 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8291 | { | |
8292 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8293 | return 1ULL << 20; |
9f0c1e56 | 8294 | |
9a7e0b18 | 8295 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8296 | } |
8297 | ||
9a7e0b18 PZ |
8298 | /* Must be called with tasklist_lock held */ |
8299 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8300 | { |
9a7e0b18 | 8301 | struct task_struct *g, *p; |
b40b2e8e | 8302 | |
9a7e0b18 PZ |
8303 | do_each_thread(g, p) { |
8304 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8305 | return 1; | |
8306 | } while_each_thread(g, p); | |
b40b2e8e | 8307 | |
9a7e0b18 PZ |
8308 | return 0; |
8309 | } | |
b40b2e8e | 8310 | |
9a7e0b18 PZ |
8311 | struct rt_schedulable_data { |
8312 | struct task_group *tg; | |
8313 | u64 rt_period; | |
8314 | u64 rt_runtime; | |
8315 | }; | |
b40b2e8e | 8316 | |
9a7e0b18 PZ |
8317 | static int tg_schedulable(struct task_group *tg, void *data) |
8318 | { | |
8319 | struct rt_schedulable_data *d = data; | |
8320 | struct task_group *child; | |
8321 | unsigned long total, sum = 0; | |
8322 | u64 period, runtime; | |
b40b2e8e | 8323 | |
9a7e0b18 PZ |
8324 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8325 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8326 | |
9a7e0b18 PZ |
8327 | if (tg == d->tg) { |
8328 | period = d->rt_period; | |
8329 | runtime = d->rt_runtime; | |
b40b2e8e | 8330 | } |
b40b2e8e | 8331 | |
4653f803 PZ |
8332 | /* |
8333 | * Cannot have more runtime than the period. | |
8334 | */ | |
8335 | if (runtime > period && runtime != RUNTIME_INF) | |
8336 | return -EINVAL; | |
6f505b16 | 8337 | |
4653f803 PZ |
8338 | /* |
8339 | * Ensure we don't starve existing RT tasks. | |
8340 | */ | |
9a7e0b18 PZ |
8341 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8342 | return -EBUSY; | |
6f505b16 | 8343 | |
9a7e0b18 | 8344 | total = to_ratio(period, runtime); |
6f505b16 | 8345 | |
4653f803 PZ |
8346 | /* |
8347 | * Nobody can have more than the global setting allows. | |
8348 | */ | |
8349 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8350 | return -EINVAL; | |
6f505b16 | 8351 | |
4653f803 PZ |
8352 | /* |
8353 | * The sum of our children's runtime should not exceed our own. | |
8354 | */ | |
9a7e0b18 PZ |
8355 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8356 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8357 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8358 | |
9a7e0b18 PZ |
8359 | if (child == d->tg) { |
8360 | period = d->rt_period; | |
8361 | runtime = d->rt_runtime; | |
8362 | } | |
6f505b16 | 8363 | |
9a7e0b18 | 8364 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8365 | } |
6f505b16 | 8366 | |
9a7e0b18 PZ |
8367 | if (sum > total) |
8368 | return -EINVAL; | |
8369 | ||
8370 | return 0; | |
6f505b16 PZ |
8371 | } |
8372 | ||
9a7e0b18 | 8373 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8374 | { |
9a7e0b18 PZ |
8375 | struct rt_schedulable_data data = { |
8376 | .tg = tg, | |
8377 | .rt_period = period, | |
8378 | .rt_runtime = runtime, | |
8379 | }; | |
8380 | ||
8381 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8382 | } |
8383 | ||
d0b27fa7 PZ |
8384 | static int tg_set_bandwidth(struct task_group *tg, |
8385 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8386 | { |
ac086bc2 | 8387 | int i, err = 0; |
9f0c1e56 | 8388 | |
9f0c1e56 | 8389 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8390 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8391 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8392 | if (err) | |
9f0c1e56 | 8393 | goto unlock; |
ac086bc2 | 8394 | |
0986b11b | 8395 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8396 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8397 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8398 | |
8399 | for_each_possible_cpu(i) { | |
8400 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8401 | ||
0986b11b | 8402 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8403 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8404 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8405 | } |
0986b11b | 8406 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8407 | unlock: |
521f1a24 | 8408 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8409 | mutex_unlock(&rt_constraints_mutex); |
8410 | ||
8411 | return err; | |
6f505b16 PZ |
8412 | } |
8413 | ||
d0b27fa7 PZ |
8414 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8415 | { | |
8416 | u64 rt_runtime, rt_period; | |
8417 | ||
8418 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8419 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8420 | if (rt_runtime_us < 0) | |
8421 | rt_runtime = RUNTIME_INF; | |
8422 | ||
8423 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8424 | } | |
8425 | ||
9f0c1e56 PZ |
8426 | long sched_group_rt_runtime(struct task_group *tg) |
8427 | { | |
8428 | u64 rt_runtime_us; | |
8429 | ||
d0b27fa7 | 8430 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8431 | return -1; |
8432 | ||
d0b27fa7 | 8433 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8434 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8435 | return rt_runtime_us; | |
8436 | } | |
d0b27fa7 PZ |
8437 | |
8438 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8439 | { | |
8440 | u64 rt_runtime, rt_period; | |
8441 | ||
8442 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8443 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8444 | ||
619b0488 R |
8445 | if (rt_period == 0) |
8446 | return -EINVAL; | |
8447 | ||
d0b27fa7 PZ |
8448 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8449 | } | |
8450 | ||
8451 | long sched_group_rt_period(struct task_group *tg) | |
8452 | { | |
8453 | u64 rt_period_us; | |
8454 | ||
8455 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8456 | do_div(rt_period_us, NSEC_PER_USEC); | |
8457 | return rt_period_us; | |
8458 | } | |
8459 | ||
8460 | static int sched_rt_global_constraints(void) | |
8461 | { | |
4653f803 | 8462 | u64 runtime, period; |
d0b27fa7 PZ |
8463 | int ret = 0; |
8464 | ||
ec5d4989 HS |
8465 | if (sysctl_sched_rt_period <= 0) |
8466 | return -EINVAL; | |
8467 | ||
4653f803 PZ |
8468 | runtime = global_rt_runtime(); |
8469 | period = global_rt_period(); | |
8470 | ||
8471 | /* | |
8472 | * Sanity check on the sysctl variables. | |
8473 | */ | |
8474 | if (runtime > period && runtime != RUNTIME_INF) | |
8475 | return -EINVAL; | |
10b612f4 | 8476 | |
d0b27fa7 | 8477 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8478 | read_lock(&tasklist_lock); |
4653f803 | 8479 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8480 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8481 | mutex_unlock(&rt_constraints_mutex); |
8482 | ||
8483 | return ret; | |
8484 | } | |
54e99124 DG |
8485 | |
8486 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8487 | { | |
8488 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8489 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8490 | return 0; | |
8491 | ||
8492 | return 1; | |
8493 | } | |
8494 | ||
6d6bc0ad | 8495 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8496 | static int sched_rt_global_constraints(void) |
8497 | { | |
ac086bc2 PZ |
8498 | unsigned long flags; |
8499 | int i; | |
8500 | ||
ec5d4989 HS |
8501 | if (sysctl_sched_rt_period <= 0) |
8502 | return -EINVAL; | |
8503 | ||
60aa605d PZ |
8504 | /* |
8505 | * There's always some RT tasks in the root group | |
8506 | * -- migration, kstopmachine etc.. | |
8507 | */ | |
8508 | if (sysctl_sched_rt_runtime == 0) | |
8509 | return -EBUSY; | |
8510 | ||
0986b11b | 8511 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8512 | for_each_possible_cpu(i) { |
8513 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8514 | ||
0986b11b | 8515 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8516 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8517 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8518 | } |
0986b11b | 8519 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8520 | |
d0b27fa7 PZ |
8521 | return 0; |
8522 | } | |
6d6bc0ad | 8523 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8524 | |
8525 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8526 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8527 | loff_t *ppos) |
8528 | { | |
8529 | int ret; | |
8530 | int old_period, old_runtime; | |
8531 | static DEFINE_MUTEX(mutex); | |
8532 | ||
8533 | mutex_lock(&mutex); | |
8534 | old_period = sysctl_sched_rt_period; | |
8535 | old_runtime = sysctl_sched_rt_runtime; | |
8536 | ||
8d65af78 | 8537 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8538 | |
8539 | if (!ret && write) { | |
8540 | ret = sched_rt_global_constraints(); | |
8541 | if (ret) { | |
8542 | sysctl_sched_rt_period = old_period; | |
8543 | sysctl_sched_rt_runtime = old_runtime; | |
8544 | } else { | |
8545 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8546 | def_rt_bandwidth.rt_period = | |
8547 | ns_to_ktime(global_rt_period()); | |
8548 | } | |
8549 | } | |
8550 | mutex_unlock(&mutex); | |
8551 | ||
8552 | return ret; | |
8553 | } | |
68318b8e | 8554 | |
052f1dc7 | 8555 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8556 | |
8557 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8558 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8559 | { |
2b01dfe3 PM |
8560 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8561 | struct task_group, css); | |
68318b8e SV |
8562 | } |
8563 | ||
8564 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8565 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8566 | { |
ec7dc8ac | 8567 | struct task_group *tg, *parent; |
68318b8e | 8568 | |
2b01dfe3 | 8569 | if (!cgrp->parent) { |
68318b8e | 8570 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8571 | return &init_task_group.css; |
8572 | } | |
8573 | ||
ec7dc8ac DG |
8574 | parent = cgroup_tg(cgrp->parent); |
8575 | tg = sched_create_group(parent); | |
68318b8e SV |
8576 | if (IS_ERR(tg)) |
8577 | return ERR_PTR(-ENOMEM); | |
8578 | ||
68318b8e SV |
8579 | return &tg->css; |
8580 | } | |
8581 | ||
41a2d6cf IM |
8582 | static void |
8583 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8584 | { |
2b01dfe3 | 8585 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8586 | |
8587 | sched_destroy_group(tg); | |
8588 | } | |
8589 | ||
41a2d6cf | 8590 | static int |
be367d09 | 8591 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8592 | { |
b68aa230 | 8593 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8594 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8595 | return -EINVAL; |
8596 | #else | |
68318b8e SV |
8597 | /* We don't support RT-tasks being in separate groups */ |
8598 | if (tsk->sched_class != &fair_sched_class) | |
8599 | return -EINVAL; | |
b68aa230 | 8600 | #endif |
be367d09 BB |
8601 | return 0; |
8602 | } | |
68318b8e | 8603 | |
be367d09 BB |
8604 | static int |
8605 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8606 | struct task_struct *tsk, bool threadgroup) | |
8607 | { | |
8608 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8609 | if (retval) | |
8610 | return retval; | |
8611 | if (threadgroup) { | |
8612 | struct task_struct *c; | |
8613 | rcu_read_lock(); | |
8614 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8615 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8616 | if (retval) { | |
8617 | rcu_read_unlock(); | |
8618 | return retval; | |
8619 | } | |
8620 | } | |
8621 | rcu_read_unlock(); | |
8622 | } | |
68318b8e SV |
8623 | return 0; |
8624 | } | |
8625 | ||
8626 | static void | |
2b01dfe3 | 8627 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8628 | struct cgroup *old_cont, struct task_struct *tsk, |
8629 | bool threadgroup) | |
68318b8e SV |
8630 | { |
8631 | sched_move_task(tsk); | |
be367d09 BB |
8632 | if (threadgroup) { |
8633 | struct task_struct *c; | |
8634 | rcu_read_lock(); | |
8635 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8636 | sched_move_task(c); | |
8637 | } | |
8638 | rcu_read_unlock(); | |
8639 | } | |
68318b8e SV |
8640 | } |
8641 | ||
052f1dc7 | 8642 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8643 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8644 | u64 shareval) |
68318b8e | 8645 | { |
2b01dfe3 | 8646 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8647 | } |
8648 | ||
f4c753b7 | 8649 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8650 | { |
2b01dfe3 | 8651 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8652 | |
8653 | return (u64) tg->shares; | |
8654 | } | |
6d6bc0ad | 8655 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8656 | |
052f1dc7 | 8657 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8658 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8659 | s64 val) |
6f505b16 | 8660 | { |
06ecb27c | 8661 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8662 | } |
8663 | ||
06ecb27c | 8664 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8665 | { |
06ecb27c | 8666 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8667 | } |
d0b27fa7 PZ |
8668 | |
8669 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8670 | u64 rt_period_us) | |
8671 | { | |
8672 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8673 | } | |
8674 | ||
8675 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8676 | { | |
8677 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8678 | } | |
6d6bc0ad | 8679 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8680 | |
fe5c7cc2 | 8681 | static struct cftype cpu_files[] = { |
052f1dc7 | 8682 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8683 | { |
8684 | .name = "shares", | |
f4c753b7 PM |
8685 | .read_u64 = cpu_shares_read_u64, |
8686 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8687 | }, |
052f1dc7 PZ |
8688 | #endif |
8689 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8690 | { |
9f0c1e56 | 8691 | .name = "rt_runtime_us", |
06ecb27c PM |
8692 | .read_s64 = cpu_rt_runtime_read, |
8693 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8694 | }, |
d0b27fa7 PZ |
8695 | { |
8696 | .name = "rt_period_us", | |
f4c753b7 PM |
8697 | .read_u64 = cpu_rt_period_read_uint, |
8698 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8699 | }, |
052f1dc7 | 8700 | #endif |
68318b8e SV |
8701 | }; |
8702 | ||
8703 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8704 | { | |
fe5c7cc2 | 8705 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8706 | } |
8707 | ||
8708 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8709 | .name = "cpu", |
8710 | .create = cpu_cgroup_create, | |
8711 | .destroy = cpu_cgroup_destroy, | |
8712 | .can_attach = cpu_cgroup_can_attach, | |
8713 | .attach = cpu_cgroup_attach, | |
8714 | .populate = cpu_cgroup_populate, | |
8715 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8716 | .early_init = 1, |
8717 | }; | |
8718 | ||
052f1dc7 | 8719 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8720 | |
8721 | #ifdef CONFIG_CGROUP_CPUACCT | |
8722 | ||
8723 | /* | |
8724 | * CPU accounting code for task groups. | |
8725 | * | |
8726 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8727 | * (balbir@in.ibm.com). | |
8728 | */ | |
8729 | ||
934352f2 | 8730 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8731 | struct cpuacct { |
8732 | struct cgroup_subsys_state css; | |
8733 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8734 | u64 __percpu *cpuusage; |
ef12fefa | 8735 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8736 | struct cpuacct *parent; |
d842de87 SV |
8737 | }; |
8738 | ||
8739 | struct cgroup_subsys cpuacct_subsys; | |
8740 | ||
8741 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8742 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8743 | { |
32cd756a | 8744 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8745 | struct cpuacct, css); |
8746 | } | |
8747 | ||
8748 | /* return cpu accounting group to which this task belongs */ | |
8749 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8750 | { | |
8751 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8752 | struct cpuacct, css); | |
8753 | } | |
8754 | ||
8755 | /* create a new cpu accounting group */ | |
8756 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8757 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8758 | { |
8759 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8760 | int i; |
d842de87 SV |
8761 | |
8762 | if (!ca) | |
ef12fefa | 8763 | goto out; |
d842de87 SV |
8764 | |
8765 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8766 | if (!ca->cpuusage) |
8767 | goto out_free_ca; | |
8768 | ||
8769 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8770 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8771 | goto out_free_counters; | |
d842de87 | 8772 | |
934352f2 BR |
8773 | if (cgrp->parent) |
8774 | ca->parent = cgroup_ca(cgrp->parent); | |
8775 | ||
d842de87 | 8776 | return &ca->css; |
ef12fefa BR |
8777 | |
8778 | out_free_counters: | |
8779 | while (--i >= 0) | |
8780 | percpu_counter_destroy(&ca->cpustat[i]); | |
8781 | free_percpu(ca->cpuusage); | |
8782 | out_free_ca: | |
8783 | kfree(ca); | |
8784 | out: | |
8785 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8786 | } |
8787 | ||
8788 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8789 | static void |
32cd756a | 8790 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8791 | { |
32cd756a | 8792 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8793 | int i; |
d842de87 | 8794 | |
ef12fefa BR |
8795 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8796 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8797 | free_percpu(ca->cpuusage); |
8798 | kfree(ca); | |
8799 | } | |
8800 | ||
720f5498 KC |
8801 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8802 | { | |
b36128c8 | 8803 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8804 | u64 data; |
8805 | ||
8806 | #ifndef CONFIG_64BIT | |
8807 | /* | |
8808 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8809 | */ | |
05fa785c | 8810 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8811 | data = *cpuusage; |
05fa785c | 8812 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8813 | #else |
8814 | data = *cpuusage; | |
8815 | #endif | |
8816 | ||
8817 | return data; | |
8818 | } | |
8819 | ||
8820 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8821 | { | |
b36128c8 | 8822 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8823 | |
8824 | #ifndef CONFIG_64BIT | |
8825 | /* | |
8826 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8827 | */ | |
05fa785c | 8828 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8829 | *cpuusage = val; |
05fa785c | 8830 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8831 | #else |
8832 | *cpuusage = val; | |
8833 | #endif | |
8834 | } | |
8835 | ||
d842de87 | 8836 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8837 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8838 | { |
32cd756a | 8839 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8840 | u64 totalcpuusage = 0; |
8841 | int i; | |
8842 | ||
720f5498 KC |
8843 | for_each_present_cpu(i) |
8844 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8845 | |
8846 | return totalcpuusage; | |
8847 | } | |
8848 | ||
0297b803 DG |
8849 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8850 | u64 reset) | |
8851 | { | |
8852 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8853 | int err = 0; | |
8854 | int i; | |
8855 | ||
8856 | if (reset) { | |
8857 | err = -EINVAL; | |
8858 | goto out; | |
8859 | } | |
8860 | ||
720f5498 KC |
8861 | for_each_present_cpu(i) |
8862 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8863 | |
0297b803 DG |
8864 | out: |
8865 | return err; | |
8866 | } | |
8867 | ||
e9515c3c KC |
8868 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8869 | struct seq_file *m) | |
8870 | { | |
8871 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8872 | u64 percpu; | |
8873 | int i; | |
8874 | ||
8875 | for_each_present_cpu(i) { | |
8876 | percpu = cpuacct_cpuusage_read(ca, i); | |
8877 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8878 | } | |
8879 | seq_printf(m, "\n"); | |
8880 | return 0; | |
8881 | } | |
8882 | ||
ef12fefa BR |
8883 | static const char *cpuacct_stat_desc[] = { |
8884 | [CPUACCT_STAT_USER] = "user", | |
8885 | [CPUACCT_STAT_SYSTEM] = "system", | |
8886 | }; | |
8887 | ||
8888 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8889 | struct cgroup_map_cb *cb) | |
8890 | { | |
8891 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8892 | int i; | |
8893 | ||
8894 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8895 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8896 | val = cputime64_to_clock_t(val); | |
8897 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8898 | } | |
8899 | return 0; | |
8900 | } | |
8901 | ||
d842de87 SV |
8902 | static struct cftype files[] = { |
8903 | { | |
8904 | .name = "usage", | |
f4c753b7 PM |
8905 | .read_u64 = cpuusage_read, |
8906 | .write_u64 = cpuusage_write, | |
d842de87 | 8907 | }, |
e9515c3c KC |
8908 | { |
8909 | .name = "usage_percpu", | |
8910 | .read_seq_string = cpuacct_percpu_seq_read, | |
8911 | }, | |
ef12fefa BR |
8912 | { |
8913 | .name = "stat", | |
8914 | .read_map = cpuacct_stats_show, | |
8915 | }, | |
d842de87 SV |
8916 | }; |
8917 | ||
32cd756a | 8918 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8919 | { |
32cd756a | 8920 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8921 | } |
8922 | ||
8923 | /* | |
8924 | * charge this task's execution time to its accounting group. | |
8925 | * | |
8926 | * called with rq->lock held. | |
8927 | */ | |
8928 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8929 | { | |
8930 | struct cpuacct *ca; | |
934352f2 | 8931 | int cpu; |
d842de87 | 8932 | |
c40c6f85 | 8933 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8934 | return; |
8935 | ||
934352f2 | 8936 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8937 | |
8938 | rcu_read_lock(); | |
8939 | ||
d842de87 | 8940 | ca = task_ca(tsk); |
d842de87 | 8941 | |
934352f2 | 8942 | for (; ca; ca = ca->parent) { |
b36128c8 | 8943 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
8944 | *cpuusage += cputime; |
8945 | } | |
a18b83b7 BR |
8946 | |
8947 | rcu_read_unlock(); | |
d842de87 SV |
8948 | } |
8949 | ||
fa535a77 AB |
8950 | /* |
8951 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
8952 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
8953 | * percpu_counter_add with values large enough to always overflow the | |
8954 | * per cpu batch limit causing bad SMP scalability. | |
8955 | * | |
8956 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
8957 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
8958 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
8959 | */ | |
8960 | #ifdef CONFIG_SMP | |
8961 | #define CPUACCT_BATCH \ | |
8962 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
8963 | #else | |
8964 | #define CPUACCT_BATCH 0 | |
8965 | #endif | |
8966 | ||
ef12fefa BR |
8967 | /* |
8968 | * Charge the system/user time to the task's accounting group. | |
8969 | */ | |
8970 | static void cpuacct_update_stats(struct task_struct *tsk, | |
8971 | enum cpuacct_stat_index idx, cputime_t val) | |
8972 | { | |
8973 | struct cpuacct *ca; | |
fa535a77 | 8974 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
8975 | |
8976 | if (unlikely(!cpuacct_subsys.active)) | |
8977 | return; | |
8978 | ||
8979 | rcu_read_lock(); | |
8980 | ca = task_ca(tsk); | |
8981 | ||
8982 | do { | |
fa535a77 | 8983 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
8984 | ca = ca->parent; |
8985 | } while (ca); | |
8986 | rcu_read_unlock(); | |
8987 | } | |
8988 | ||
d842de87 SV |
8989 | struct cgroup_subsys cpuacct_subsys = { |
8990 | .name = "cpuacct", | |
8991 | .create = cpuacct_create, | |
8992 | .destroy = cpuacct_destroy, | |
8993 | .populate = cpuacct_populate, | |
8994 | .subsys_id = cpuacct_subsys_id, | |
8995 | }; | |
8996 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
8997 | |
8998 | #ifndef CONFIG_SMP | |
8999 | ||
03b042bf PM |
9000 | void synchronize_sched_expedited(void) |
9001 | { | |
fc390cde | 9002 | barrier(); |
03b042bf PM |
9003 | } |
9004 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9005 | ||
9006 | #else /* #ifndef CONFIG_SMP */ | |
9007 | ||
cc631fb7 | 9008 | static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0); |
03b042bf | 9009 | |
cc631fb7 | 9010 | static int synchronize_sched_expedited_cpu_stop(void *data) |
03b042bf | 9011 | { |
969c7921 TH |
9012 | /* |
9013 | * There must be a full memory barrier on each affected CPU | |
9014 | * between the time that try_stop_cpus() is called and the | |
9015 | * time that it returns. | |
9016 | * | |
9017 | * In the current initial implementation of cpu_stop, the | |
9018 | * above condition is already met when the control reaches | |
9019 | * this point and the following smp_mb() is not strictly | |
9020 | * necessary. Do smp_mb() anyway for documentation and | |
9021 | * robustness against future implementation changes. | |
9022 | */ | |
cc631fb7 | 9023 | smp_mb(); /* See above comment block. */ |
969c7921 | 9024 | return 0; |
03b042bf | 9025 | } |
03b042bf PM |
9026 | |
9027 | /* | |
9028 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9029 | * approach to force grace period to end quickly. This consumes | |
9030 | * significant time on all CPUs, and is thus not recommended for | |
9031 | * any sort of common-case code. | |
9032 | * | |
9033 | * Note that it is illegal to call this function while holding any | |
9034 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9035 | * observe this restriction will result in deadlock. | |
9036 | */ | |
9037 | void synchronize_sched_expedited(void) | |
9038 | { | |
969c7921 | 9039 | int snap, trycount = 0; |
03b042bf PM |
9040 | |
9041 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
969c7921 | 9042 | snap = atomic_read(&synchronize_sched_expedited_count) + 1; |
03b042bf | 9043 | get_online_cpus(); |
969c7921 TH |
9044 | while (try_stop_cpus(cpu_online_mask, |
9045 | synchronize_sched_expedited_cpu_stop, | |
94458d5e | 9046 | NULL) == -EAGAIN) { |
03b042bf PM |
9047 | put_online_cpus(); |
9048 | if (trycount++ < 10) | |
9049 | udelay(trycount * num_online_cpus()); | |
9050 | else { | |
9051 | synchronize_sched(); | |
9052 | return; | |
9053 | } | |
969c7921 | 9054 | if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) { |
03b042bf PM |
9055 | smp_mb(); /* ensure test happens before caller kfree */ |
9056 | return; | |
9057 | } | |
9058 | get_online_cpus(); | |
9059 | } | |
969c7921 | 9060 | atomic_inc(&synchronize_sched_expedited_count); |
cc631fb7 | 9061 | smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */ |
03b042bf | 9062 | put_online_cpus(); |
03b042bf PM |
9063 | } |
9064 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9065 | ||
9066 | #endif /* #else #ifndef CONFIG_SMP */ |