]>
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 | 429 | struct cpupri cpupri; |
57d885fe GH |
430 | }; |
431 | ||
dc938520 GH |
432 | /* |
433 | * By default the system creates a single root-domain with all cpus as | |
434 | * members (mimicking the global state we have today). | |
435 | */ | |
57d885fe GH |
436 | static struct root_domain def_root_domain; |
437 | ||
ed2d372c | 438 | #endif /* CONFIG_SMP */ |
57d885fe | 439 | |
1da177e4 LT |
440 | /* |
441 | * This is the main, per-CPU runqueue data structure. | |
442 | * | |
443 | * Locking rule: those places that want to lock multiple runqueues | |
444 | * (such as the load balancing or the thread migration code), lock | |
445 | * acquire operations must be ordered by ascending &runqueue. | |
446 | */ | |
70b97a7f | 447 | struct rq { |
d8016491 | 448 | /* runqueue lock: */ |
05fa785c | 449 | raw_spinlock_t lock; |
1da177e4 LT |
450 | |
451 | /* | |
452 | * nr_running and cpu_load should be in the same cacheline because | |
453 | * remote CPUs use both these fields when doing load calculation. | |
454 | */ | |
455 | unsigned long nr_running; | |
6aa645ea IM |
456 | #define CPU_LOAD_IDX_MAX 5 |
457 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 458 | unsigned long last_load_update_tick; |
46cb4b7c | 459 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 460 | u64 nohz_stamp; |
83cd4fe2 | 461 | unsigned char nohz_balance_kick; |
46cb4b7c | 462 | #endif |
a64692a3 MG |
463 | unsigned int skip_clock_update; |
464 | ||
d8016491 IM |
465 | /* capture load from *all* tasks on this cpu: */ |
466 | struct load_weight load; | |
6aa645ea IM |
467 | unsigned long nr_load_updates; |
468 | u64 nr_switches; | |
469 | ||
470 | struct cfs_rq cfs; | |
6f505b16 | 471 | struct rt_rq rt; |
6f505b16 | 472 | |
6aa645ea | 473 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
474 | /* list of leaf cfs_rq on this cpu: */ |
475 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
476 | #endif |
477 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 478 | struct list_head leaf_rt_rq_list; |
1da177e4 | 479 | #endif |
1da177e4 LT |
480 | |
481 | /* | |
482 | * This is part of a global counter where only the total sum | |
483 | * over all CPUs matters. A task can increase this counter on | |
484 | * one CPU and if it got migrated afterwards it may decrease | |
485 | * it on another CPU. Always updated under the runqueue lock: | |
486 | */ | |
487 | unsigned long nr_uninterruptible; | |
488 | ||
34f971f6 | 489 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 490 | unsigned long next_balance; |
1da177e4 | 491 | struct mm_struct *prev_mm; |
6aa645ea | 492 | |
3e51f33f | 493 | u64 clock; |
305e6835 | 494 | u64 clock_task; |
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 | ||
aa483808 VP |
522 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
523 | u64 prev_irq_time; | |
524 | #endif | |
525 | ||
dce48a84 TG |
526 | /* calc_load related fields */ |
527 | unsigned long calc_load_update; | |
528 | long calc_load_active; | |
529 | ||
8f4d37ec | 530 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
531 | #ifdef CONFIG_SMP |
532 | int hrtick_csd_pending; | |
533 | struct call_single_data hrtick_csd; | |
534 | #endif | |
8f4d37ec PZ |
535 | struct hrtimer hrtick_timer; |
536 | #endif | |
537 | ||
1da177e4 LT |
538 | #ifdef CONFIG_SCHEDSTATS |
539 | /* latency stats */ | |
540 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
541 | unsigned long long rq_cpu_time; |
542 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
543 | |
544 | /* sys_sched_yield() stats */ | |
480b9434 | 545 | unsigned int yld_count; |
1da177e4 LT |
546 | |
547 | /* schedule() stats */ | |
480b9434 KC |
548 | unsigned int sched_switch; |
549 | unsigned int sched_count; | |
550 | unsigned int sched_goidle; | |
1da177e4 LT |
551 | |
552 | /* try_to_wake_up() stats */ | |
480b9434 KC |
553 | unsigned int ttwu_count; |
554 | unsigned int ttwu_local; | |
b8efb561 IM |
555 | |
556 | /* BKL stats */ | |
480b9434 | 557 | unsigned int bkl_count; |
1da177e4 LT |
558 | #endif |
559 | }; | |
560 | ||
f34e3b61 | 561 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 562 | |
a64692a3 | 563 | |
1e5a7405 | 564 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 565 | |
0a2966b4 CL |
566 | static inline int cpu_of(struct rq *rq) |
567 | { | |
568 | #ifdef CONFIG_SMP | |
569 | return rq->cpu; | |
570 | #else | |
571 | return 0; | |
572 | #endif | |
573 | } | |
574 | ||
497f0ab3 | 575 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
576 | rcu_dereference_check((p), \ |
577 | rcu_read_lock_sched_held() || \ | |
578 | lockdep_is_held(&sched_domains_mutex)) | |
579 | ||
674311d5 NP |
580 | /* |
581 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 582 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
583 | * |
584 | * The domain tree of any CPU may only be accessed from within | |
585 | * preempt-disabled sections. | |
586 | */ | |
48f24c4d | 587 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 588 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
589 | |
590 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
591 | #define this_rq() (&__get_cpu_var(runqueues)) | |
592 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
593 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 594 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 595 | |
dc61b1d6 PZ |
596 | #ifdef CONFIG_CGROUP_SCHED |
597 | ||
598 | /* | |
599 | * Return the group to which this tasks belongs. | |
600 | * | |
601 | * We use task_subsys_state_check() and extend the RCU verification | |
602 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
603 | * holds that lock for each task it moves into the cgroup. Therefore | |
604 | * by holding that lock, we pin the task to the current cgroup. | |
605 | */ | |
606 | static inline struct task_group *task_group(struct task_struct *p) | |
607 | { | |
608 | struct cgroup_subsys_state *css; | |
609 | ||
610 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
611 | lockdep_is_held(&task_rq(p)->lock)); | |
612 | return container_of(css, struct task_group, css); | |
613 | } | |
614 | ||
615 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
616 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
617 | { | |
618 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
619 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
620 | p->se.parent = task_group(p)->se[cpu]; | |
621 | #endif | |
622 | ||
623 | #ifdef CONFIG_RT_GROUP_SCHED | |
624 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
625 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
626 | #endif | |
627 | } | |
628 | ||
629 | #else /* CONFIG_CGROUP_SCHED */ | |
630 | ||
631 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
632 | static inline struct task_group *task_group(struct task_struct *p) | |
633 | { | |
634 | return NULL; | |
635 | } | |
636 | ||
637 | #endif /* CONFIG_CGROUP_SCHED */ | |
638 | ||
305e6835 | 639 | static u64 irq_time_cpu(int cpu); |
aa483808 | 640 | static void sched_irq_time_avg_update(struct rq *rq, u64 irq_time); |
305e6835 | 641 | |
aa9c4c0f | 642 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 643 | { |
305e6835 VP |
644 | if (!rq->skip_clock_update) { |
645 | int cpu = cpu_of(rq); | |
646 | u64 irq_time; | |
647 | ||
648 | rq->clock = sched_clock_cpu(cpu); | |
649 | irq_time = irq_time_cpu(cpu); | |
650 | if (rq->clock - irq_time > rq->clock_task) | |
651 | rq->clock_task = rq->clock - irq_time; | |
aa483808 VP |
652 | |
653 | sched_irq_time_avg_update(rq, irq_time); | |
305e6835 | 654 | } |
3e51f33f PZ |
655 | } |
656 | ||
bf5c91ba IM |
657 | /* |
658 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
659 | */ | |
660 | #ifdef CONFIG_SCHED_DEBUG | |
661 | # define const_debug __read_mostly | |
662 | #else | |
663 | # define const_debug static const | |
664 | #endif | |
665 | ||
017730c1 IM |
666 | /** |
667 | * runqueue_is_locked | |
e17b38bf | 668 | * @cpu: the processor in question. |
017730c1 IM |
669 | * |
670 | * Returns true if the current cpu runqueue is locked. | |
671 | * This interface allows printk to be called with the runqueue lock | |
672 | * held and know whether or not it is OK to wake up the klogd. | |
673 | */ | |
89f19f04 | 674 | int runqueue_is_locked(int cpu) |
017730c1 | 675 | { |
05fa785c | 676 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
677 | } |
678 | ||
bf5c91ba IM |
679 | /* |
680 | * Debugging: various feature bits | |
681 | */ | |
f00b45c1 PZ |
682 | |
683 | #define SCHED_FEAT(name, enabled) \ | |
684 | __SCHED_FEAT_##name , | |
685 | ||
bf5c91ba | 686 | enum { |
f00b45c1 | 687 | #include "sched_features.h" |
bf5c91ba IM |
688 | }; |
689 | ||
f00b45c1 PZ |
690 | #undef SCHED_FEAT |
691 | ||
692 | #define SCHED_FEAT(name, enabled) \ | |
693 | (1UL << __SCHED_FEAT_##name) * enabled | | |
694 | ||
bf5c91ba | 695 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
696 | #include "sched_features.h" |
697 | 0; | |
698 | ||
699 | #undef SCHED_FEAT | |
700 | ||
701 | #ifdef CONFIG_SCHED_DEBUG | |
702 | #define SCHED_FEAT(name, enabled) \ | |
703 | #name , | |
704 | ||
983ed7a6 | 705 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
706 | #include "sched_features.h" |
707 | NULL | |
708 | }; | |
709 | ||
710 | #undef SCHED_FEAT | |
711 | ||
34f3a814 | 712 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 713 | { |
f00b45c1 PZ |
714 | int i; |
715 | ||
716 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
717 | if (!(sysctl_sched_features & (1UL << i))) |
718 | seq_puts(m, "NO_"); | |
719 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 720 | } |
34f3a814 | 721 | seq_puts(m, "\n"); |
f00b45c1 | 722 | |
34f3a814 | 723 | return 0; |
f00b45c1 PZ |
724 | } |
725 | ||
726 | static ssize_t | |
727 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
728 | size_t cnt, loff_t *ppos) | |
729 | { | |
730 | char buf[64]; | |
7740191c | 731 | char *cmp; |
f00b45c1 PZ |
732 | int neg = 0; |
733 | int i; | |
734 | ||
735 | if (cnt > 63) | |
736 | cnt = 63; | |
737 | ||
738 | if (copy_from_user(&buf, ubuf, cnt)) | |
739 | return -EFAULT; | |
740 | ||
741 | buf[cnt] = 0; | |
7740191c | 742 | cmp = strstrip(buf); |
f00b45c1 | 743 | |
c24b7c52 | 744 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
745 | neg = 1; |
746 | cmp += 3; | |
747 | } | |
748 | ||
749 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 750 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
751 | if (neg) |
752 | sysctl_sched_features &= ~(1UL << i); | |
753 | else | |
754 | sysctl_sched_features |= (1UL << i); | |
755 | break; | |
756 | } | |
757 | } | |
758 | ||
759 | if (!sched_feat_names[i]) | |
760 | return -EINVAL; | |
761 | ||
42994724 | 762 | *ppos += cnt; |
f00b45c1 PZ |
763 | |
764 | return cnt; | |
765 | } | |
766 | ||
34f3a814 LZ |
767 | static int sched_feat_open(struct inode *inode, struct file *filp) |
768 | { | |
769 | return single_open(filp, sched_feat_show, NULL); | |
770 | } | |
771 | ||
828c0950 | 772 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
773 | .open = sched_feat_open, |
774 | .write = sched_feat_write, | |
775 | .read = seq_read, | |
776 | .llseek = seq_lseek, | |
777 | .release = single_release, | |
f00b45c1 PZ |
778 | }; |
779 | ||
780 | static __init int sched_init_debug(void) | |
781 | { | |
f00b45c1 PZ |
782 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
783 | &sched_feat_fops); | |
784 | ||
785 | return 0; | |
786 | } | |
787 | late_initcall(sched_init_debug); | |
788 | ||
789 | #endif | |
790 | ||
791 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 792 | |
b82d9fdd PZ |
793 | /* |
794 | * Number of tasks to iterate in a single balance run. | |
795 | * Limited because this is done with IRQs disabled. | |
796 | */ | |
797 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
798 | ||
2398f2c6 PZ |
799 | /* |
800 | * ratelimit for updating the group shares. | |
55cd5340 | 801 | * default: 0.25ms |
2398f2c6 | 802 | */ |
55cd5340 | 803 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 804 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 805 | |
ffda12a1 PZ |
806 | /* |
807 | * Inject some fuzzyness into changing the per-cpu group shares | |
808 | * this avoids remote rq-locks at the expense of fairness. | |
809 | * default: 4 | |
810 | */ | |
811 | unsigned int sysctl_sched_shares_thresh = 4; | |
812 | ||
e9e9250b PZ |
813 | /* |
814 | * period over which we average the RT time consumption, measured | |
815 | * in ms. | |
816 | * | |
817 | * default: 1s | |
818 | */ | |
819 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
820 | ||
fa85ae24 | 821 | /* |
9f0c1e56 | 822 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
823 | * default: 1s |
824 | */ | |
9f0c1e56 | 825 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 826 | |
6892b75e IM |
827 | static __read_mostly int scheduler_running; |
828 | ||
9f0c1e56 PZ |
829 | /* |
830 | * part of the period that we allow rt tasks to run in us. | |
831 | * default: 0.95s | |
832 | */ | |
833 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 834 | |
d0b27fa7 PZ |
835 | static inline u64 global_rt_period(void) |
836 | { | |
837 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
838 | } | |
839 | ||
840 | static inline u64 global_rt_runtime(void) | |
841 | { | |
e26873bb | 842 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
843 | return RUNTIME_INF; |
844 | ||
845 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
846 | } | |
fa85ae24 | 847 | |
1da177e4 | 848 | #ifndef prepare_arch_switch |
4866cde0 NP |
849 | # define prepare_arch_switch(next) do { } while (0) |
850 | #endif | |
851 | #ifndef finish_arch_switch | |
852 | # define finish_arch_switch(prev) do { } while (0) | |
853 | #endif | |
854 | ||
051a1d1a DA |
855 | static inline int task_current(struct rq *rq, struct task_struct *p) |
856 | { | |
857 | return rq->curr == p; | |
858 | } | |
859 | ||
4866cde0 | 860 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 861 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 862 | { |
051a1d1a | 863 | return task_current(rq, p); |
4866cde0 NP |
864 | } |
865 | ||
70b97a7f | 866 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
867 | { |
868 | } | |
869 | ||
70b97a7f | 870 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 871 | { |
da04c035 IM |
872 | #ifdef CONFIG_DEBUG_SPINLOCK |
873 | /* this is a valid case when another task releases the spinlock */ | |
874 | rq->lock.owner = current; | |
875 | #endif | |
8a25d5de IM |
876 | /* |
877 | * If we are tracking spinlock dependencies then we have to | |
878 | * fix up the runqueue lock - which gets 'carried over' from | |
879 | * prev into current: | |
880 | */ | |
881 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
882 | ||
05fa785c | 883 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
884 | } |
885 | ||
886 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 887 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
888 | { |
889 | #ifdef CONFIG_SMP | |
890 | return p->oncpu; | |
891 | #else | |
051a1d1a | 892 | return task_current(rq, p); |
4866cde0 NP |
893 | #endif |
894 | } | |
895 | ||
70b97a7f | 896 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
897 | { |
898 | #ifdef CONFIG_SMP | |
899 | /* | |
900 | * We can optimise this out completely for !SMP, because the | |
901 | * SMP rebalancing from interrupt is the only thing that cares | |
902 | * here. | |
903 | */ | |
904 | next->oncpu = 1; | |
905 | #endif | |
906 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 907 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 908 | #else |
05fa785c | 909 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
910 | #endif |
911 | } | |
912 | ||
70b97a7f | 913 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
914 | { |
915 | #ifdef CONFIG_SMP | |
916 | /* | |
917 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
918 | * We must ensure this doesn't happen until the switch is completely | |
919 | * finished. | |
920 | */ | |
921 | smp_wmb(); | |
922 | prev->oncpu = 0; | |
923 | #endif | |
924 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
925 | local_irq_enable(); | |
1da177e4 | 926 | #endif |
4866cde0 NP |
927 | } |
928 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 929 | |
0970d299 | 930 | /* |
65cc8e48 PZ |
931 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
932 | * against ttwu(). | |
0970d299 PZ |
933 | */ |
934 | static inline int task_is_waking(struct task_struct *p) | |
935 | { | |
0017d735 | 936 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
937 | } |
938 | ||
b29739f9 IM |
939 | /* |
940 | * __task_rq_lock - lock the runqueue a given task resides on. | |
941 | * Must be called interrupts disabled. | |
942 | */ | |
70b97a7f | 943 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
944 | __acquires(rq->lock) |
945 | { | |
0970d299 PZ |
946 | struct rq *rq; |
947 | ||
3a5c359a | 948 | for (;;) { |
0970d299 | 949 | rq = task_rq(p); |
05fa785c | 950 | raw_spin_lock(&rq->lock); |
65cc8e48 | 951 | if (likely(rq == task_rq(p))) |
3a5c359a | 952 | return rq; |
05fa785c | 953 | raw_spin_unlock(&rq->lock); |
b29739f9 | 954 | } |
b29739f9 IM |
955 | } |
956 | ||
1da177e4 LT |
957 | /* |
958 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 959 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
960 | * explicitly disabling preemption. |
961 | */ | |
70b97a7f | 962 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
963 | __acquires(rq->lock) |
964 | { | |
70b97a7f | 965 | struct rq *rq; |
1da177e4 | 966 | |
3a5c359a AK |
967 | for (;;) { |
968 | local_irq_save(*flags); | |
969 | rq = task_rq(p); | |
05fa785c | 970 | raw_spin_lock(&rq->lock); |
65cc8e48 | 971 | if (likely(rq == task_rq(p))) |
3a5c359a | 972 | return rq; |
05fa785c | 973 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 974 | } |
1da177e4 LT |
975 | } |
976 | ||
a9957449 | 977 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
978 | __releases(rq->lock) |
979 | { | |
05fa785c | 980 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
981 | } |
982 | ||
70b97a7f | 983 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
984 | __releases(rq->lock) |
985 | { | |
05fa785c | 986 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
987 | } |
988 | ||
1da177e4 | 989 | /* |
cc2a73b5 | 990 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 991 | */ |
a9957449 | 992 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
993 | __acquires(rq->lock) |
994 | { | |
70b97a7f | 995 | struct rq *rq; |
1da177e4 LT |
996 | |
997 | local_irq_disable(); | |
998 | rq = this_rq(); | |
05fa785c | 999 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
1000 | |
1001 | return rq; | |
1002 | } | |
1003 | ||
8f4d37ec PZ |
1004 | #ifdef CONFIG_SCHED_HRTICK |
1005 | /* | |
1006 | * Use HR-timers to deliver accurate preemption points. | |
1007 | * | |
1008 | * Its all a bit involved since we cannot program an hrt while holding the | |
1009 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1010 | * reschedule event. | |
1011 | * | |
1012 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1013 | * rq->lock. | |
1014 | */ | |
8f4d37ec PZ |
1015 | |
1016 | /* | |
1017 | * Use hrtick when: | |
1018 | * - enabled by features | |
1019 | * - hrtimer is actually high res | |
1020 | */ | |
1021 | static inline int hrtick_enabled(struct rq *rq) | |
1022 | { | |
1023 | if (!sched_feat(HRTICK)) | |
1024 | return 0; | |
ba42059f | 1025 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1026 | return 0; |
8f4d37ec PZ |
1027 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1028 | } | |
1029 | ||
8f4d37ec PZ |
1030 | static void hrtick_clear(struct rq *rq) |
1031 | { | |
1032 | if (hrtimer_active(&rq->hrtick_timer)) | |
1033 | hrtimer_cancel(&rq->hrtick_timer); | |
1034 | } | |
1035 | ||
8f4d37ec PZ |
1036 | /* |
1037 | * High-resolution timer tick. | |
1038 | * Runs from hardirq context with interrupts disabled. | |
1039 | */ | |
1040 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1041 | { | |
1042 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1043 | ||
1044 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1045 | ||
05fa785c | 1046 | raw_spin_lock(&rq->lock); |
3e51f33f | 1047 | update_rq_clock(rq); |
8f4d37ec | 1048 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1049 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1050 | |
1051 | return HRTIMER_NORESTART; | |
1052 | } | |
1053 | ||
95e904c7 | 1054 | #ifdef CONFIG_SMP |
31656519 PZ |
1055 | /* |
1056 | * called from hardirq (IPI) context | |
1057 | */ | |
1058 | static void __hrtick_start(void *arg) | |
b328ca18 | 1059 | { |
31656519 | 1060 | struct rq *rq = arg; |
b328ca18 | 1061 | |
05fa785c | 1062 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1063 | hrtimer_restart(&rq->hrtick_timer); |
1064 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1065 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1066 | } |
1067 | ||
31656519 PZ |
1068 | /* |
1069 | * Called to set the hrtick timer state. | |
1070 | * | |
1071 | * called with rq->lock held and irqs disabled | |
1072 | */ | |
1073 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1074 | { |
31656519 PZ |
1075 | struct hrtimer *timer = &rq->hrtick_timer; |
1076 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1077 | |
cc584b21 | 1078 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1079 | |
1080 | if (rq == this_rq()) { | |
1081 | hrtimer_restart(timer); | |
1082 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1083 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1084 | rq->hrtick_csd_pending = 1; |
1085 | } | |
b328ca18 PZ |
1086 | } |
1087 | ||
1088 | static int | |
1089 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1090 | { | |
1091 | int cpu = (int)(long)hcpu; | |
1092 | ||
1093 | switch (action) { | |
1094 | case CPU_UP_CANCELED: | |
1095 | case CPU_UP_CANCELED_FROZEN: | |
1096 | case CPU_DOWN_PREPARE: | |
1097 | case CPU_DOWN_PREPARE_FROZEN: | |
1098 | case CPU_DEAD: | |
1099 | case CPU_DEAD_FROZEN: | |
31656519 | 1100 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1101 | return NOTIFY_OK; |
1102 | } | |
1103 | ||
1104 | return NOTIFY_DONE; | |
1105 | } | |
1106 | ||
fa748203 | 1107 | static __init void init_hrtick(void) |
b328ca18 PZ |
1108 | { |
1109 | hotcpu_notifier(hotplug_hrtick, 0); | |
1110 | } | |
31656519 PZ |
1111 | #else |
1112 | /* | |
1113 | * Called to set the hrtick timer state. | |
1114 | * | |
1115 | * called with rq->lock held and irqs disabled | |
1116 | */ | |
1117 | static void hrtick_start(struct rq *rq, u64 delay) | |
1118 | { | |
7f1e2ca9 | 1119 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1120 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1121 | } |
b328ca18 | 1122 | |
006c75f1 | 1123 | static inline void init_hrtick(void) |
8f4d37ec | 1124 | { |
8f4d37ec | 1125 | } |
31656519 | 1126 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1127 | |
31656519 | 1128 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1129 | { |
31656519 PZ |
1130 | #ifdef CONFIG_SMP |
1131 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1132 | |
31656519 PZ |
1133 | rq->hrtick_csd.flags = 0; |
1134 | rq->hrtick_csd.func = __hrtick_start; | |
1135 | rq->hrtick_csd.info = rq; | |
1136 | #endif | |
8f4d37ec | 1137 | |
31656519 PZ |
1138 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1139 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1140 | } |
006c75f1 | 1141 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1142 | static inline void hrtick_clear(struct rq *rq) |
1143 | { | |
1144 | } | |
1145 | ||
8f4d37ec PZ |
1146 | static inline void init_rq_hrtick(struct rq *rq) |
1147 | { | |
1148 | } | |
1149 | ||
b328ca18 PZ |
1150 | static inline void init_hrtick(void) |
1151 | { | |
1152 | } | |
006c75f1 | 1153 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1154 | |
c24d20db IM |
1155 | /* |
1156 | * resched_task - mark a task 'to be rescheduled now'. | |
1157 | * | |
1158 | * On UP this means the setting of the need_resched flag, on SMP it | |
1159 | * might also involve a cross-CPU call to trigger the scheduler on | |
1160 | * the target CPU. | |
1161 | */ | |
1162 | #ifdef CONFIG_SMP | |
1163 | ||
1164 | #ifndef tsk_is_polling | |
1165 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1166 | #endif | |
1167 | ||
31656519 | 1168 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1169 | { |
1170 | int cpu; | |
1171 | ||
05fa785c | 1172 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1173 | |
5ed0cec0 | 1174 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1175 | return; |
1176 | ||
5ed0cec0 | 1177 | set_tsk_need_resched(p); |
c24d20db IM |
1178 | |
1179 | cpu = task_cpu(p); | |
1180 | if (cpu == smp_processor_id()) | |
1181 | return; | |
1182 | ||
1183 | /* NEED_RESCHED must be visible before we test polling */ | |
1184 | smp_mb(); | |
1185 | if (!tsk_is_polling(p)) | |
1186 | smp_send_reschedule(cpu); | |
1187 | } | |
1188 | ||
1189 | static void resched_cpu(int cpu) | |
1190 | { | |
1191 | struct rq *rq = cpu_rq(cpu); | |
1192 | unsigned long flags; | |
1193 | ||
05fa785c | 1194 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1195 | return; |
1196 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1197 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1198 | } |
06d8308c TG |
1199 | |
1200 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1201 | /* |
1202 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1203 | * from an idle cpu. This is good for power-savings. | |
1204 | * | |
1205 | * We don't do similar optimization for completely idle system, as | |
1206 | * selecting an idle cpu will add more delays to the timers than intended | |
1207 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1208 | */ | |
1209 | int get_nohz_timer_target(void) | |
1210 | { | |
1211 | int cpu = smp_processor_id(); | |
1212 | int i; | |
1213 | struct sched_domain *sd; | |
1214 | ||
1215 | for_each_domain(cpu, sd) { | |
1216 | for_each_cpu(i, sched_domain_span(sd)) | |
1217 | if (!idle_cpu(i)) | |
1218 | return i; | |
1219 | } | |
1220 | return cpu; | |
1221 | } | |
06d8308c TG |
1222 | /* |
1223 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1224 | * idle CPU then this timer might expire before the next timer event | |
1225 | * which is scheduled to wake up that CPU. In case of a completely | |
1226 | * idle system the next event might even be infinite time into the | |
1227 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1228 | * leaves the inner idle loop so the newly added timer is taken into | |
1229 | * account when the CPU goes back to idle and evaluates the timer | |
1230 | * wheel for the next timer event. | |
1231 | */ | |
1232 | void wake_up_idle_cpu(int cpu) | |
1233 | { | |
1234 | struct rq *rq = cpu_rq(cpu); | |
1235 | ||
1236 | if (cpu == smp_processor_id()) | |
1237 | return; | |
1238 | ||
1239 | /* | |
1240 | * This is safe, as this function is called with the timer | |
1241 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1242 | * to idle and has not yet set rq->curr to idle then it will | |
1243 | * be serialized on the timer wheel base lock and take the new | |
1244 | * timer into account automatically. | |
1245 | */ | |
1246 | if (rq->curr != rq->idle) | |
1247 | return; | |
1248 | ||
1249 | /* | |
1250 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1251 | * lockless. The worst case is that the other CPU runs the | |
1252 | * idle task through an additional NOOP schedule() | |
1253 | */ | |
5ed0cec0 | 1254 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1255 | |
1256 | /* NEED_RESCHED must be visible before we test polling */ | |
1257 | smp_mb(); | |
1258 | if (!tsk_is_polling(rq->idle)) | |
1259 | smp_send_reschedule(cpu); | |
1260 | } | |
39c0cbe2 | 1261 | |
6d6bc0ad | 1262 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1263 | |
e9e9250b PZ |
1264 | static u64 sched_avg_period(void) |
1265 | { | |
1266 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1267 | } | |
1268 | ||
1269 | static void sched_avg_update(struct rq *rq) | |
1270 | { | |
1271 | s64 period = sched_avg_period(); | |
1272 | ||
1273 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1274 | /* |
1275 | * Inline assembly required to prevent the compiler | |
1276 | * optimising this loop into a divmod call. | |
1277 | * See __iter_div_u64_rem() for another example of this. | |
1278 | */ | |
1279 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1280 | rq->age_stamp += period; |
1281 | rq->rt_avg /= 2; | |
1282 | } | |
1283 | } | |
1284 | ||
1285 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1286 | { | |
1287 | rq->rt_avg += rt_delta; | |
1288 | sched_avg_update(rq); | |
1289 | } | |
1290 | ||
6d6bc0ad | 1291 | #else /* !CONFIG_SMP */ |
31656519 | 1292 | static void resched_task(struct task_struct *p) |
c24d20db | 1293 | { |
05fa785c | 1294 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1295 | set_tsk_need_resched(p); |
c24d20db | 1296 | } |
e9e9250b PZ |
1297 | |
1298 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1299 | { | |
1300 | } | |
da2b71ed SS |
1301 | |
1302 | static void sched_avg_update(struct rq *rq) | |
1303 | { | |
1304 | } | |
6d6bc0ad | 1305 | #endif /* CONFIG_SMP */ |
c24d20db | 1306 | |
45bf76df IM |
1307 | #if BITS_PER_LONG == 32 |
1308 | # define WMULT_CONST (~0UL) | |
1309 | #else | |
1310 | # define WMULT_CONST (1UL << 32) | |
1311 | #endif | |
1312 | ||
1313 | #define WMULT_SHIFT 32 | |
1314 | ||
194081eb IM |
1315 | /* |
1316 | * Shift right and round: | |
1317 | */ | |
cf2ab469 | 1318 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1319 | |
a7be37ac PZ |
1320 | /* |
1321 | * delta *= weight / lw | |
1322 | */ | |
cb1c4fc9 | 1323 | static unsigned long |
45bf76df IM |
1324 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1325 | struct load_weight *lw) | |
1326 | { | |
1327 | u64 tmp; | |
1328 | ||
7a232e03 LJ |
1329 | if (!lw->inv_weight) { |
1330 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1331 | lw->inv_weight = 1; | |
1332 | else | |
1333 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1334 | / (lw->weight+1); | |
1335 | } | |
45bf76df IM |
1336 | |
1337 | tmp = (u64)delta_exec * weight; | |
1338 | /* | |
1339 | * Check whether we'd overflow the 64-bit multiplication: | |
1340 | */ | |
194081eb | 1341 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1342 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1343 | WMULT_SHIFT/2); |
1344 | else | |
cf2ab469 | 1345 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1346 | |
ecf691da | 1347 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1348 | } |
1349 | ||
1091985b | 1350 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1351 | { |
1352 | lw->weight += inc; | |
e89996ae | 1353 | lw->inv_weight = 0; |
45bf76df IM |
1354 | } |
1355 | ||
1091985b | 1356 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1357 | { |
1358 | lw->weight -= dec; | |
e89996ae | 1359 | lw->inv_weight = 0; |
45bf76df IM |
1360 | } |
1361 | ||
2dd73a4f PW |
1362 | /* |
1363 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1364 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1365 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1366 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1367 | * scaled version of the new time slice allocation that they receive on time |
1368 | * slice expiry etc. | |
1369 | */ | |
1370 | ||
cce7ade8 PZ |
1371 | #define WEIGHT_IDLEPRIO 3 |
1372 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1373 | |
1374 | /* | |
1375 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1376 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1377 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1378 | * that remained on nice 0. | |
1379 | * | |
1380 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1381 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1382 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1383 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1384 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1385 | */ |
1386 | static const int prio_to_weight[40] = { | |
254753dc IM |
1387 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1388 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1389 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1390 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1391 | /* 0 */ 1024, 820, 655, 526, 423, | |
1392 | /* 5 */ 335, 272, 215, 172, 137, | |
1393 | /* 10 */ 110, 87, 70, 56, 45, | |
1394 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1395 | }; |
1396 | ||
5714d2de IM |
1397 | /* |
1398 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1399 | * | |
1400 | * In cases where the weight does not change often, we can use the | |
1401 | * precalculated inverse to speed up arithmetics by turning divisions | |
1402 | * into multiplications: | |
1403 | */ | |
dd41f596 | 1404 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1405 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1406 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1407 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1408 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1409 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1410 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1411 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1412 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1413 | }; |
2dd73a4f | 1414 | |
ef12fefa BR |
1415 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1416 | enum cpuacct_stat_index { | |
1417 | CPUACCT_STAT_USER, /* ... user mode */ | |
1418 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1419 | ||
1420 | CPUACCT_STAT_NSTATS, | |
1421 | }; | |
1422 | ||
d842de87 SV |
1423 | #ifdef CONFIG_CGROUP_CPUACCT |
1424 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1425 | static void cpuacct_update_stats(struct task_struct *tsk, |
1426 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1427 | #else |
1428 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1429 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1430 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1431 | #endif |
1432 | ||
18d95a28 PZ |
1433 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1434 | { | |
1435 | update_load_add(&rq->load, load); | |
1436 | } | |
1437 | ||
1438 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1439 | { | |
1440 | update_load_sub(&rq->load, load); | |
1441 | } | |
1442 | ||
7940ca36 | 1443 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1444 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1445 | |
1446 | /* | |
1447 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1448 | * leaving it for the final time. | |
1449 | */ | |
eb755805 | 1450 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1451 | { |
1452 | struct task_group *parent, *child; | |
eb755805 | 1453 | int ret; |
c09595f6 PZ |
1454 | |
1455 | rcu_read_lock(); | |
1456 | parent = &root_task_group; | |
1457 | down: | |
eb755805 PZ |
1458 | ret = (*down)(parent, data); |
1459 | if (ret) | |
1460 | goto out_unlock; | |
c09595f6 PZ |
1461 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1462 | parent = child; | |
1463 | goto down; | |
1464 | ||
1465 | up: | |
1466 | continue; | |
1467 | } | |
eb755805 PZ |
1468 | ret = (*up)(parent, data); |
1469 | if (ret) | |
1470 | goto out_unlock; | |
c09595f6 PZ |
1471 | |
1472 | child = parent; | |
1473 | parent = parent->parent; | |
1474 | if (parent) | |
1475 | goto up; | |
eb755805 | 1476 | out_unlock: |
c09595f6 | 1477 | rcu_read_unlock(); |
eb755805 PZ |
1478 | |
1479 | return ret; | |
c09595f6 PZ |
1480 | } |
1481 | ||
eb755805 PZ |
1482 | static int tg_nop(struct task_group *tg, void *data) |
1483 | { | |
1484 | return 0; | |
c09595f6 | 1485 | } |
eb755805 PZ |
1486 | #endif |
1487 | ||
1488 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1489 | /* Used instead of source_load when we know the type == 0 */ |
1490 | static unsigned long weighted_cpuload(const int cpu) | |
1491 | { | |
1492 | return cpu_rq(cpu)->load.weight; | |
1493 | } | |
1494 | ||
1495 | /* | |
1496 | * Return a low guess at the load of a migration-source cpu weighted | |
1497 | * according to the scheduling class and "nice" value. | |
1498 | * | |
1499 | * We want to under-estimate the load of migration sources, to | |
1500 | * balance conservatively. | |
1501 | */ | |
1502 | static unsigned long source_load(int cpu, int type) | |
1503 | { | |
1504 | struct rq *rq = cpu_rq(cpu); | |
1505 | unsigned long total = weighted_cpuload(cpu); | |
1506 | ||
1507 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1508 | return total; | |
1509 | ||
1510 | return min(rq->cpu_load[type-1], total); | |
1511 | } | |
1512 | ||
1513 | /* | |
1514 | * Return a high guess at the load of a migration-target cpu weighted | |
1515 | * according to the scheduling class and "nice" value. | |
1516 | */ | |
1517 | static unsigned long target_load(int cpu, int type) | |
1518 | { | |
1519 | struct rq *rq = cpu_rq(cpu); | |
1520 | unsigned long total = weighted_cpuload(cpu); | |
1521 | ||
1522 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1523 | return total; | |
1524 | ||
1525 | return max(rq->cpu_load[type-1], total); | |
1526 | } | |
1527 | ||
ae154be1 PZ |
1528 | static unsigned long power_of(int cpu) |
1529 | { | |
e51fd5e2 | 1530 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1531 | } |
1532 | ||
eb755805 PZ |
1533 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1534 | ||
1535 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1536 | { | |
1537 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1538 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1539 | |
4cd42620 SR |
1540 | if (nr_running) |
1541 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1542 | else |
1543 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1544 | |
1545 | return rq->avg_load_per_task; | |
1546 | } | |
1547 | ||
1548 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1549 | |
43cf38eb | 1550 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1551 | |
c09595f6 PZ |
1552 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1553 | ||
1554 | /* | |
1555 | * Calculate and set the cpu's group shares. | |
1556 | */ | |
34d76c41 PZ |
1557 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1558 | unsigned long sd_shares, | |
1559 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1560 | unsigned long *usd_rq_weight) |
18d95a28 | 1561 | { |
34d76c41 | 1562 | unsigned long shares, rq_weight; |
a5004278 | 1563 | int boost = 0; |
c09595f6 | 1564 | |
4a6cc4bd | 1565 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1566 | if (!rq_weight) { |
1567 | boost = 1; | |
1568 | rq_weight = NICE_0_LOAD; | |
1569 | } | |
c8cba857 | 1570 | |
c09595f6 | 1571 | /* |
a8af7246 PZ |
1572 | * \Sum_j shares_j * rq_weight_i |
1573 | * shares_i = ----------------------------- | |
1574 | * \Sum_j rq_weight_j | |
c09595f6 | 1575 | */ |
ec4e0e2f | 1576 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1577 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1578 | |
ffda12a1 PZ |
1579 | if (abs(shares - tg->se[cpu]->load.weight) > |
1580 | sysctl_sched_shares_thresh) { | |
1581 | struct rq *rq = cpu_rq(cpu); | |
1582 | unsigned long flags; | |
c09595f6 | 1583 | |
05fa785c | 1584 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1585 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1586 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1587 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1588 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1589 | } |
18d95a28 | 1590 | } |
c09595f6 PZ |
1591 | |
1592 | /* | |
c8cba857 PZ |
1593 | * Re-compute the task group their per cpu shares over the given domain. |
1594 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1595 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1596 | */ |
eb755805 | 1597 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1598 | { |
cd8ad40d | 1599 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1600 | unsigned long *usd_rq_weight; |
eb755805 | 1601 | struct sched_domain *sd = data; |
34d76c41 | 1602 | unsigned long flags; |
c8cba857 | 1603 | int i; |
c09595f6 | 1604 | |
34d76c41 PZ |
1605 | if (!tg->se[0]) |
1606 | return 0; | |
1607 | ||
1608 | local_irq_save(flags); | |
4a6cc4bd | 1609 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1610 | |
758b2cdc | 1611 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1612 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1613 | usd_rq_weight[i] = weight; |
34d76c41 | 1614 | |
cd8ad40d | 1615 | rq_weight += weight; |
ec4e0e2f KC |
1616 | /* |
1617 | * If there are currently no tasks on the cpu pretend there | |
1618 | * is one of average load so that when a new task gets to | |
1619 | * run here it will not get delayed by group starvation. | |
1620 | */ | |
ec4e0e2f KC |
1621 | if (!weight) |
1622 | weight = NICE_0_LOAD; | |
1623 | ||
cd8ad40d | 1624 | sum_weight += weight; |
c8cba857 | 1625 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1626 | } |
c09595f6 | 1627 | |
cd8ad40d PZ |
1628 | if (!rq_weight) |
1629 | rq_weight = sum_weight; | |
1630 | ||
c8cba857 PZ |
1631 | if ((!shares && rq_weight) || shares > tg->shares) |
1632 | shares = tg->shares; | |
1633 | ||
1634 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1635 | shares = tg->shares; | |
c09595f6 | 1636 | |
758b2cdc | 1637 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1638 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1639 | |
1640 | local_irq_restore(flags); | |
eb755805 PZ |
1641 | |
1642 | return 0; | |
c09595f6 PZ |
1643 | } |
1644 | ||
1645 | /* | |
c8cba857 PZ |
1646 | * Compute the cpu's hierarchical load factor for each task group. |
1647 | * This needs to be done in a top-down fashion because the load of a child | |
1648 | * group is a fraction of its parents load. | |
c09595f6 | 1649 | */ |
eb755805 | 1650 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1651 | { |
c8cba857 | 1652 | unsigned long load; |
eb755805 | 1653 | long cpu = (long)data; |
c09595f6 | 1654 | |
c8cba857 PZ |
1655 | if (!tg->parent) { |
1656 | load = cpu_rq(cpu)->load.weight; | |
1657 | } else { | |
1658 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1659 | load *= tg->cfs_rq[cpu]->shares; | |
1660 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1661 | } | |
c09595f6 | 1662 | |
c8cba857 | 1663 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1664 | |
eb755805 | 1665 | return 0; |
c09595f6 PZ |
1666 | } |
1667 | ||
c8cba857 | 1668 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1669 | { |
e7097159 PZ |
1670 | s64 elapsed; |
1671 | u64 now; | |
1672 | ||
1673 | if (root_task_group_empty()) | |
1674 | return; | |
1675 | ||
c676329a | 1676 | now = local_clock(); |
e7097159 | 1677 | elapsed = now - sd->last_update; |
2398f2c6 PZ |
1678 | |
1679 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1680 | sd->last_update = now; | |
eb755805 | 1681 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1682 | } |
4d8d595d PZ |
1683 | } |
1684 | ||
eb755805 | 1685 | static void update_h_load(long cpu) |
c09595f6 | 1686 | { |
eb755805 | 1687 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1688 | } |
1689 | ||
c09595f6 PZ |
1690 | #else |
1691 | ||
c8cba857 | 1692 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1693 | { |
1694 | } | |
1695 | ||
18d95a28 PZ |
1696 | #endif |
1697 | ||
8f45e2b5 GH |
1698 | #ifdef CONFIG_PREEMPT |
1699 | ||
b78bb868 PZ |
1700 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1701 | ||
70574a99 | 1702 | /* |
8f45e2b5 GH |
1703 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1704 | * way at the expense of forcing extra atomic operations in all | |
1705 | * invocations. This assures that the double_lock is acquired using the | |
1706 | * same underlying policy as the spinlock_t on this architecture, which | |
1707 | * reduces latency compared to the unfair variant below. However, it | |
1708 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1709 | */ |
8f45e2b5 GH |
1710 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1711 | __releases(this_rq->lock) | |
1712 | __acquires(busiest->lock) | |
1713 | __acquires(this_rq->lock) | |
1714 | { | |
05fa785c | 1715 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1716 | double_rq_lock(this_rq, busiest); |
1717 | ||
1718 | return 1; | |
1719 | } | |
1720 | ||
1721 | #else | |
1722 | /* | |
1723 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1724 | * latency by eliminating extra atomic operations when the locks are | |
1725 | * already in proper order on entry. This favors lower cpu-ids and will | |
1726 | * grant the double lock to lower cpus over higher ids under contention, | |
1727 | * regardless of entry order into the function. | |
1728 | */ | |
1729 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1730 | __releases(this_rq->lock) |
1731 | __acquires(busiest->lock) | |
1732 | __acquires(this_rq->lock) | |
1733 | { | |
1734 | int ret = 0; | |
1735 | ||
05fa785c | 1736 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1737 | if (busiest < this_rq) { |
05fa785c TG |
1738 | raw_spin_unlock(&this_rq->lock); |
1739 | raw_spin_lock(&busiest->lock); | |
1740 | raw_spin_lock_nested(&this_rq->lock, | |
1741 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1742 | ret = 1; |
1743 | } else | |
05fa785c TG |
1744 | raw_spin_lock_nested(&busiest->lock, |
1745 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1746 | } |
1747 | return ret; | |
1748 | } | |
1749 | ||
8f45e2b5 GH |
1750 | #endif /* CONFIG_PREEMPT */ |
1751 | ||
1752 | /* | |
1753 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1754 | */ | |
1755 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1756 | { | |
1757 | if (unlikely(!irqs_disabled())) { | |
1758 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1759 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1760 | BUG_ON(1); |
1761 | } | |
1762 | ||
1763 | return _double_lock_balance(this_rq, busiest); | |
1764 | } | |
1765 | ||
70574a99 AD |
1766 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1767 | __releases(busiest->lock) | |
1768 | { | |
05fa785c | 1769 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1770 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1771 | } | |
1e3c88bd PZ |
1772 | |
1773 | /* | |
1774 | * double_rq_lock - safely lock two runqueues | |
1775 | * | |
1776 | * Note this does not disable interrupts like task_rq_lock, | |
1777 | * you need to do so manually before calling. | |
1778 | */ | |
1779 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1780 | __acquires(rq1->lock) | |
1781 | __acquires(rq2->lock) | |
1782 | { | |
1783 | BUG_ON(!irqs_disabled()); | |
1784 | if (rq1 == rq2) { | |
1785 | raw_spin_lock(&rq1->lock); | |
1786 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1787 | } else { | |
1788 | if (rq1 < rq2) { | |
1789 | raw_spin_lock(&rq1->lock); | |
1790 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1791 | } else { | |
1792 | raw_spin_lock(&rq2->lock); | |
1793 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1794 | } | |
1795 | } | |
1e3c88bd PZ |
1796 | } |
1797 | ||
1798 | /* | |
1799 | * double_rq_unlock - safely unlock two runqueues | |
1800 | * | |
1801 | * Note this does not restore interrupts like task_rq_unlock, | |
1802 | * you need to do so manually after calling. | |
1803 | */ | |
1804 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1805 | __releases(rq1->lock) | |
1806 | __releases(rq2->lock) | |
1807 | { | |
1808 | raw_spin_unlock(&rq1->lock); | |
1809 | if (rq1 != rq2) | |
1810 | raw_spin_unlock(&rq2->lock); | |
1811 | else | |
1812 | __release(rq2->lock); | |
1813 | } | |
1814 | ||
18d95a28 PZ |
1815 | #endif |
1816 | ||
30432094 | 1817 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1818 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1819 | { | |
30432094 | 1820 | #ifdef CONFIG_SMP |
34e83e85 IM |
1821 | cfs_rq->shares = shares; |
1822 | #endif | |
1823 | } | |
30432094 | 1824 | #endif |
e7693a36 | 1825 | |
74f5187a | 1826 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1827 | static void update_sysctl(void); |
acb4a848 | 1828 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1829 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1830 | |
cd29fe6f PZ |
1831 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1832 | { | |
1833 | set_task_rq(p, cpu); | |
1834 | #ifdef CONFIG_SMP | |
1835 | /* | |
1836 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1837 | * successfuly executed on another CPU. We must ensure that updates of | |
1838 | * per-task data have been completed by this moment. | |
1839 | */ | |
1840 | smp_wmb(); | |
1841 | task_thread_info(p)->cpu = cpu; | |
1842 | #endif | |
1843 | } | |
dce48a84 | 1844 | |
1e3c88bd | 1845 | static const struct sched_class rt_sched_class; |
dd41f596 | 1846 | |
34f971f6 | 1847 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1848 | #define for_each_class(class) \ |
1849 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1850 | |
1e3c88bd PZ |
1851 | #include "sched_stats.h" |
1852 | ||
c09595f6 | 1853 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1854 | { |
1855 | rq->nr_running++; | |
9c217245 IM |
1856 | } |
1857 | ||
c09595f6 | 1858 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1859 | { |
1860 | rq->nr_running--; | |
9c217245 IM |
1861 | } |
1862 | ||
45bf76df IM |
1863 | static void set_load_weight(struct task_struct *p) |
1864 | { | |
dd41f596 IM |
1865 | /* |
1866 | * SCHED_IDLE tasks get minimal weight: | |
1867 | */ | |
1868 | if (p->policy == SCHED_IDLE) { | |
1869 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1870 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1871 | return; | |
1872 | } | |
71f8bd46 | 1873 | |
dd41f596 IM |
1874 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1875 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1876 | } |
1877 | ||
371fd7e7 | 1878 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1879 | { |
a64692a3 | 1880 | update_rq_clock(rq); |
dd41f596 | 1881 | sched_info_queued(p); |
371fd7e7 | 1882 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1883 | p->se.on_rq = 1; |
71f8bd46 IM |
1884 | } |
1885 | ||
371fd7e7 | 1886 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1887 | { |
a64692a3 | 1888 | update_rq_clock(rq); |
46ac22ba | 1889 | sched_info_dequeued(p); |
371fd7e7 | 1890 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1891 | p->se.on_rq = 0; |
71f8bd46 IM |
1892 | } |
1893 | ||
1e3c88bd PZ |
1894 | /* |
1895 | * activate_task - move a task to the runqueue. | |
1896 | */ | |
371fd7e7 | 1897 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1898 | { |
1899 | if (task_contributes_to_load(p)) | |
1900 | rq->nr_uninterruptible--; | |
1901 | ||
371fd7e7 | 1902 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1903 | inc_nr_running(rq); |
1904 | } | |
1905 | ||
1906 | /* | |
1907 | * deactivate_task - remove a task from the runqueue. | |
1908 | */ | |
371fd7e7 | 1909 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1910 | { |
1911 | if (task_contributes_to_load(p)) | |
1912 | rq->nr_uninterruptible++; | |
1913 | ||
371fd7e7 | 1914 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1915 | dec_nr_running(rq); |
1916 | } | |
1917 | ||
b52bfee4 VP |
1918 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1919 | ||
305e6835 VP |
1920 | /* |
1921 | * There are no locks covering percpu hardirq/softirq time. | |
1922 | * They are only modified in account_system_vtime, on corresponding CPU | |
1923 | * with interrupts disabled. So, writes are safe. | |
1924 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1925 | * This may result in other CPU reading this CPU's irq time and can | |
1926 | * race with irq/account_system_vtime on this CPU. We would either get old | |
1927 | * or new value (or semi updated value on 32 bit) with a side effect of | |
1928 | * accounting a slice of irq time to wrong task when irq is in progress | |
1929 | * while we read rq->clock. That is a worthy compromise in place of having | |
1930 | * locks on each irq in account_system_time. | |
1931 | */ | |
b52bfee4 VP |
1932 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1933 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1934 | ||
1935 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1936 | static int sched_clock_irqtime; | |
1937 | ||
1938 | void enable_sched_clock_irqtime(void) | |
1939 | { | |
1940 | sched_clock_irqtime = 1; | |
1941 | } | |
1942 | ||
1943 | void disable_sched_clock_irqtime(void) | |
1944 | { | |
1945 | sched_clock_irqtime = 0; | |
1946 | } | |
1947 | ||
305e6835 VP |
1948 | static u64 irq_time_cpu(int cpu) |
1949 | { | |
1950 | if (!sched_clock_irqtime) | |
1951 | return 0; | |
1952 | ||
1953 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | |
1954 | } | |
1955 | ||
b52bfee4 VP |
1956 | void account_system_vtime(struct task_struct *curr) |
1957 | { | |
1958 | unsigned long flags; | |
1959 | int cpu; | |
1960 | u64 now, delta; | |
1961 | ||
1962 | if (!sched_clock_irqtime) | |
1963 | return; | |
1964 | ||
1965 | local_irq_save(flags); | |
1966 | ||
b52bfee4 | 1967 | cpu = smp_processor_id(); |
d267f87f | 1968 | now = sched_clock_cpu(cpu); |
b52bfee4 VP |
1969 | delta = now - per_cpu(irq_start_time, cpu); |
1970 | per_cpu(irq_start_time, cpu) = now; | |
1971 | /* | |
1972 | * We do not account for softirq time from ksoftirqd here. | |
1973 | * We want to continue accounting softirq time to ksoftirqd thread | |
1974 | * in that case, so as not to confuse scheduler with a special task | |
1975 | * that do not consume any time, but still wants to run. | |
1976 | */ | |
1977 | if (hardirq_count()) | |
1978 | per_cpu(cpu_hardirq_time, cpu) += delta; | |
1979 | else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD)) | |
1980 | per_cpu(cpu_softirq_time, cpu) += delta; | |
1981 | ||
1982 | local_irq_restore(flags); | |
1983 | } | |
b7dadc38 | 1984 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1985 | |
aa483808 VP |
1986 | static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) |
1987 | { | |
1988 | if (sched_clock_irqtime && sched_feat(NONIRQ_POWER)) { | |
1989 | u64 delta_irq = curr_irq_time - rq->prev_irq_time; | |
1990 | rq->prev_irq_time = curr_irq_time; | |
1991 | sched_rt_avg_update(rq, delta_irq); | |
1992 | } | |
1993 | } | |
1994 | ||
305e6835 VP |
1995 | #else |
1996 | ||
1997 | static u64 irq_time_cpu(int cpu) | |
1998 | { | |
1999 | return 0; | |
2000 | } | |
2001 | ||
aa483808 VP |
2002 | static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) { } |
2003 | ||
b52bfee4 VP |
2004 | #endif |
2005 | ||
1e3c88bd PZ |
2006 | #include "sched_idletask.c" |
2007 | #include "sched_fair.c" | |
2008 | #include "sched_rt.c" | |
34f971f6 | 2009 | #include "sched_stoptask.c" |
1e3c88bd PZ |
2010 | #ifdef CONFIG_SCHED_DEBUG |
2011 | # include "sched_debug.c" | |
2012 | #endif | |
2013 | ||
34f971f6 PZ |
2014 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2015 | { | |
2016 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2017 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2018 | ||
2019 | if (stop) { | |
2020 | /* | |
2021 | * Make it appear like a SCHED_FIFO task, its something | |
2022 | * userspace knows about and won't get confused about. | |
2023 | * | |
2024 | * Also, it will make PI more or less work without too | |
2025 | * much confusion -- but then, stop work should not | |
2026 | * rely on PI working anyway. | |
2027 | */ | |
2028 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2029 | ||
2030 | stop->sched_class = &stop_sched_class; | |
2031 | } | |
2032 | ||
2033 | cpu_rq(cpu)->stop = stop; | |
2034 | ||
2035 | if (old_stop) { | |
2036 | /* | |
2037 | * Reset it back to a normal scheduling class so that | |
2038 | * it can die in pieces. | |
2039 | */ | |
2040 | old_stop->sched_class = &rt_sched_class; | |
2041 | } | |
2042 | } | |
2043 | ||
14531189 | 2044 | /* |
dd41f596 | 2045 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2046 | */ |
14531189 IM |
2047 | static inline int __normal_prio(struct task_struct *p) |
2048 | { | |
dd41f596 | 2049 | return p->static_prio; |
14531189 IM |
2050 | } |
2051 | ||
b29739f9 IM |
2052 | /* |
2053 | * Calculate the expected normal priority: i.e. priority | |
2054 | * without taking RT-inheritance into account. Might be | |
2055 | * boosted by interactivity modifiers. Changes upon fork, | |
2056 | * setprio syscalls, and whenever the interactivity | |
2057 | * estimator recalculates. | |
2058 | */ | |
36c8b586 | 2059 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2060 | { |
2061 | int prio; | |
2062 | ||
e05606d3 | 2063 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2064 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2065 | else | |
2066 | prio = __normal_prio(p); | |
2067 | return prio; | |
2068 | } | |
2069 | ||
2070 | /* | |
2071 | * Calculate the current priority, i.e. the priority | |
2072 | * taken into account by the scheduler. This value might | |
2073 | * be boosted by RT tasks, or might be boosted by | |
2074 | * interactivity modifiers. Will be RT if the task got | |
2075 | * RT-boosted. If not then it returns p->normal_prio. | |
2076 | */ | |
36c8b586 | 2077 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2078 | { |
2079 | p->normal_prio = normal_prio(p); | |
2080 | /* | |
2081 | * If we are RT tasks or we were boosted to RT priority, | |
2082 | * keep the priority unchanged. Otherwise, update priority | |
2083 | * to the normal priority: | |
2084 | */ | |
2085 | if (!rt_prio(p->prio)) | |
2086 | return p->normal_prio; | |
2087 | return p->prio; | |
2088 | } | |
2089 | ||
1da177e4 LT |
2090 | /** |
2091 | * task_curr - is this task currently executing on a CPU? | |
2092 | * @p: the task in question. | |
2093 | */ | |
36c8b586 | 2094 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2095 | { |
2096 | return cpu_curr(task_cpu(p)) == p; | |
2097 | } | |
2098 | ||
cb469845 SR |
2099 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2100 | const struct sched_class *prev_class, | |
2101 | int oldprio, int running) | |
2102 | { | |
2103 | if (prev_class != p->sched_class) { | |
2104 | if (prev_class->switched_from) | |
2105 | prev_class->switched_from(rq, p, running); | |
2106 | p->sched_class->switched_to(rq, p, running); | |
2107 | } else | |
2108 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
2109 | } | |
2110 | ||
1e5a7405 PZ |
2111 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2112 | { | |
2113 | const struct sched_class *class; | |
2114 | ||
2115 | if (p->sched_class == rq->curr->sched_class) { | |
2116 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2117 | } else { | |
2118 | for_each_class(class) { | |
2119 | if (class == rq->curr->sched_class) | |
2120 | break; | |
2121 | if (class == p->sched_class) { | |
2122 | resched_task(rq->curr); | |
2123 | break; | |
2124 | } | |
2125 | } | |
2126 | } | |
2127 | ||
2128 | /* | |
2129 | * A queue event has occurred, and we're going to schedule. In | |
2130 | * this case, we can save a useless back to back clock update. | |
2131 | */ | |
2132 | if (test_tsk_need_resched(rq->curr)) | |
2133 | rq->skip_clock_update = 1; | |
2134 | } | |
2135 | ||
1da177e4 | 2136 | #ifdef CONFIG_SMP |
cc367732 IM |
2137 | /* |
2138 | * Is this task likely cache-hot: | |
2139 | */ | |
e7693a36 | 2140 | static int |
cc367732 IM |
2141 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2142 | { | |
2143 | s64 delta; | |
2144 | ||
e6c8fba7 PZ |
2145 | if (p->sched_class != &fair_sched_class) |
2146 | return 0; | |
2147 | ||
ef8002f6 NR |
2148 | if (unlikely(p->policy == SCHED_IDLE)) |
2149 | return 0; | |
2150 | ||
f540a608 IM |
2151 | /* |
2152 | * Buddy candidates are cache hot: | |
2153 | */ | |
f685ceac | 2154 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2155 | (&p->se == cfs_rq_of(&p->se)->next || |
2156 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2157 | return 1; |
2158 | ||
6bc1665b IM |
2159 | if (sysctl_sched_migration_cost == -1) |
2160 | return 1; | |
2161 | if (sysctl_sched_migration_cost == 0) | |
2162 | return 0; | |
2163 | ||
cc367732 IM |
2164 | delta = now - p->se.exec_start; |
2165 | ||
2166 | return delta < (s64)sysctl_sched_migration_cost; | |
2167 | } | |
2168 | ||
dd41f596 | 2169 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2170 | { |
e2912009 PZ |
2171 | #ifdef CONFIG_SCHED_DEBUG |
2172 | /* | |
2173 | * We should never call set_task_cpu() on a blocked task, | |
2174 | * ttwu() will sort out the placement. | |
2175 | */ | |
077614ee PZ |
2176 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2177 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2178 | #endif |
2179 | ||
de1d7286 | 2180 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2181 | |
0c69774e PZ |
2182 | if (task_cpu(p) != new_cpu) { |
2183 | p->se.nr_migrations++; | |
2184 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2185 | } | |
dd41f596 IM |
2186 | |
2187 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2188 | } |
2189 | ||
969c7921 | 2190 | struct migration_arg { |
36c8b586 | 2191 | struct task_struct *task; |
1da177e4 | 2192 | int dest_cpu; |
70b97a7f | 2193 | }; |
1da177e4 | 2194 | |
969c7921 TH |
2195 | static int migration_cpu_stop(void *data); |
2196 | ||
1da177e4 LT |
2197 | /* |
2198 | * The task's runqueue lock must be held. | |
2199 | * Returns true if you have to wait for migration thread. | |
2200 | */ | |
969c7921 | 2201 | static bool migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2202 | { |
70b97a7f | 2203 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2204 | |
2205 | /* | |
2206 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2207 | * the next wake-up will properly place the task. |
1da177e4 | 2208 | */ |
969c7921 | 2209 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2210 | } |
2211 | ||
2212 | /* | |
2213 | * wait_task_inactive - wait for a thread to unschedule. | |
2214 | * | |
85ba2d86 RM |
2215 | * If @match_state is nonzero, it's the @p->state value just checked and |
2216 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2217 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2218 | * we return a positive number (its total switch count). If a second call | |
2219 | * a short while later returns the same number, the caller can be sure that | |
2220 | * @p has remained unscheduled the whole time. | |
2221 | * | |
1da177e4 LT |
2222 | * The caller must ensure that the task *will* unschedule sometime soon, |
2223 | * else this function might spin for a *long* time. This function can't | |
2224 | * be called with interrupts off, or it may introduce deadlock with | |
2225 | * smp_call_function() if an IPI is sent by the same process we are | |
2226 | * waiting to become inactive. | |
2227 | */ | |
85ba2d86 | 2228 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2229 | { |
2230 | unsigned long flags; | |
dd41f596 | 2231 | int running, on_rq; |
85ba2d86 | 2232 | unsigned long ncsw; |
70b97a7f | 2233 | struct rq *rq; |
1da177e4 | 2234 | |
3a5c359a AK |
2235 | for (;;) { |
2236 | /* | |
2237 | * We do the initial early heuristics without holding | |
2238 | * any task-queue locks at all. We'll only try to get | |
2239 | * the runqueue lock when things look like they will | |
2240 | * work out! | |
2241 | */ | |
2242 | rq = task_rq(p); | |
fa490cfd | 2243 | |
3a5c359a AK |
2244 | /* |
2245 | * If the task is actively running on another CPU | |
2246 | * still, just relax and busy-wait without holding | |
2247 | * any locks. | |
2248 | * | |
2249 | * NOTE! Since we don't hold any locks, it's not | |
2250 | * even sure that "rq" stays as the right runqueue! | |
2251 | * But we don't care, since "task_running()" will | |
2252 | * return false if the runqueue has changed and p | |
2253 | * is actually now running somewhere else! | |
2254 | */ | |
85ba2d86 RM |
2255 | while (task_running(rq, p)) { |
2256 | if (match_state && unlikely(p->state != match_state)) | |
2257 | return 0; | |
3a5c359a | 2258 | cpu_relax(); |
85ba2d86 | 2259 | } |
fa490cfd | 2260 | |
3a5c359a AK |
2261 | /* |
2262 | * Ok, time to look more closely! We need the rq | |
2263 | * lock now, to be *sure*. If we're wrong, we'll | |
2264 | * just go back and repeat. | |
2265 | */ | |
2266 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2267 | trace_sched_wait_task(p); |
3a5c359a AK |
2268 | running = task_running(rq, p); |
2269 | on_rq = p->se.on_rq; | |
85ba2d86 | 2270 | ncsw = 0; |
f31e11d8 | 2271 | if (!match_state || p->state == match_state) |
93dcf55f | 2272 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2273 | task_rq_unlock(rq, &flags); |
fa490cfd | 2274 | |
85ba2d86 RM |
2275 | /* |
2276 | * If it changed from the expected state, bail out now. | |
2277 | */ | |
2278 | if (unlikely(!ncsw)) | |
2279 | break; | |
2280 | ||
3a5c359a AK |
2281 | /* |
2282 | * Was it really running after all now that we | |
2283 | * checked with the proper locks actually held? | |
2284 | * | |
2285 | * Oops. Go back and try again.. | |
2286 | */ | |
2287 | if (unlikely(running)) { | |
2288 | cpu_relax(); | |
2289 | continue; | |
2290 | } | |
fa490cfd | 2291 | |
3a5c359a AK |
2292 | /* |
2293 | * It's not enough that it's not actively running, | |
2294 | * it must be off the runqueue _entirely_, and not | |
2295 | * preempted! | |
2296 | * | |
80dd99b3 | 2297 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2298 | * running right now), it's preempted, and we should |
2299 | * yield - it could be a while. | |
2300 | */ | |
2301 | if (unlikely(on_rq)) { | |
2302 | schedule_timeout_uninterruptible(1); | |
2303 | continue; | |
2304 | } | |
fa490cfd | 2305 | |
3a5c359a AK |
2306 | /* |
2307 | * Ahh, all good. It wasn't running, and it wasn't | |
2308 | * runnable, which means that it will never become | |
2309 | * running in the future either. We're all done! | |
2310 | */ | |
2311 | break; | |
2312 | } | |
85ba2d86 RM |
2313 | |
2314 | return ncsw; | |
1da177e4 LT |
2315 | } |
2316 | ||
2317 | /*** | |
2318 | * kick_process - kick a running thread to enter/exit the kernel | |
2319 | * @p: the to-be-kicked thread | |
2320 | * | |
2321 | * Cause a process which is running on another CPU to enter | |
2322 | * kernel-mode, without any delay. (to get signals handled.) | |
2323 | * | |
2324 | * NOTE: this function doesnt have to take the runqueue lock, | |
2325 | * because all it wants to ensure is that the remote task enters | |
2326 | * the kernel. If the IPI races and the task has been migrated | |
2327 | * to another CPU then no harm is done and the purpose has been | |
2328 | * achieved as well. | |
2329 | */ | |
36c8b586 | 2330 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2331 | { |
2332 | int cpu; | |
2333 | ||
2334 | preempt_disable(); | |
2335 | cpu = task_cpu(p); | |
2336 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2337 | smp_send_reschedule(cpu); | |
2338 | preempt_enable(); | |
2339 | } | |
b43e3521 | 2340 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2341 | #endif /* CONFIG_SMP */ |
1da177e4 | 2342 | |
0793a61d TG |
2343 | /** |
2344 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2345 | * @p: the task to evaluate | |
2346 | * @func: the function to be called | |
2347 | * @info: the function call argument | |
2348 | * | |
2349 | * Calls the function @func when the task is currently running. This might | |
2350 | * be on the current CPU, which just calls the function directly | |
2351 | */ | |
2352 | void task_oncpu_function_call(struct task_struct *p, | |
2353 | void (*func) (void *info), void *info) | |
2354 | { | |
2355 | int cpu; | |
2356 | ||
2357 | preempt_disable(); | |
2358 | cpu = task_cpu(p); | |
2359 | if (task_curr(p)) | |
2360 | smp_call_function_single(cpu, func, info, 1); | |
2361 | preempt_enable(); | |
2362 | } | |
2363 | ||
970b13ba | 2364 | #ifdef CONFIG_SMP |
30da688e ON |
2365 | /* |
2366 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2367 | */ | |
5da9a0fb PZ |
2368 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2369 | { | |
2370 | int dest_cpu; | |
2371 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2372 | ||
2373 | /* Look for allowed, online CPU in same node. */ | |
2374 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2375 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2376 | return dest_cpu; | |
2377 | ||
2378 | /* Any allowed, online CPU? */ | |
2379 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2380 | if (dest_cpu < nr_cpu_ids) | |
2381 | return dest_cpu; | |
2382 | ||
2383 | /* No more Mr. Nice Guy. */ | |
897f0b3c | 2384 | if (unlikely(dest_cpu >= nr_cpu_ids)) { |
9084bb82 | 2385 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
5da9a0fb PZ |
2386 | /* |
2387 | * Don't tell them about moving exiting tasks or | |
2388 | * kernel threads (both mm NULL), since they never | |
2389 | * leave kernel. | |
2390 | */ | |
2391 | if (p->mm && printk_ratelimit()) { | |
2392 | printk(KERN_INFO "process %d (%s) no " | |
2393 | "longer affine to cpu%d\n", | |
2394 | task_pid_nr(p), p->comm, cpu); | |
2395 | } | |
2396 | } | |
2397 | ||
2398 | return dest_cpu; | |
2399 | } | |
2400 | ||
e2912009 | 2401 | /* |
30da688e | 2402 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2403 | */ |
970b13ba | 2404 | static inline |
0017d735 | 2405 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2406 | { |
0017d735 | 2407 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2408 | |
2409 | /* | |
2410 | * In order not to call set_task_cpu() on a blocking task we need | |
2411 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2412 | * cpu. | |
2413 | * | |
2414 | * Since this is common to all placement strategies, this lives here. | |
2415 | * | |
2416 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2417 | * not worry about this generic constraint ] | |
2418 | */ | |
2419 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2420 | !cpu_online(cpu))) |
5da9a0fb | 2421 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2422 | |
2423 | return cpu; | |
970b13ba | 2424 | } |
09a40af5 MG |
2425 | |
2426 | static void update_avg(u64 *avg, u64 sample) | |
2427 | { | |
2428 | s64 diff = sample - *avg; | |
2429 | *avg += diff >> 3; | |
2430 | } | |
970b13ba PZ |
2431 | #endif |
2432 | ||
9ed3811a TH |
2433 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2434 | bool is_sync, bool is_migrate, bool is_local, | |
2435 | unsigned long en_flags) | |
2436 | { | |
2437 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2438 | if (is_sync) | |
2439 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2440 | if (is_migrate) | |
2441 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2442 | if (is_local) | |
2443 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2444 | else | |
2445 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2446 | ||
2447 | activate_task(rq, p, en_flags); | |
2448 | } | |
2449 | ||
2450 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2451 | int wake_flags, bool success) | |
2452 | { | |
2453 | trace_sched_wakeup(p, success); | |
2454 | check_preempt_curr(rq, p, wake_flags); | |
2455 | ||
2456 | p->state = TASK_RUNNING; | |
2457 | #ifdef CONFIG_SMP | |
2458 | if (p->sched_class->task_woken) | |
2459 | p->sched_class->task_woken(rq, p); | |
2460 | ||
2461 | if (unlikely(rq->idle_stamp)) { | |
2462 | u64 delta = rq->clock - rq->idle_stamp; | |
2463 | u64 max = 2*sysctl_sched_migration_cost; | |
2464 | ||
2465 | if (delta > max) | |
2466 | rq->avg_idle = max; | |
2467 | else | |
2468 | update_avg(&rq->avg_idle, delta); | |
2469 | rq->idle_stamp = 0; | |
2470 | } | |
2471 | #endif | |
21aa9af0 TH |
2472 | /* if a worker is waking up, notify workqueue */ |
2473 | if ((p->flags & PF_WQ_WORKER) && success) | |
2474 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2475 | } |
2476 | ||
2477 | /** | |
1da177e4 | 2478 | * try_to_wake_up - wake up a thread |
9ed3811a | 2479 | * @p: the thread to be awakened |
1da177e4 | 2480 | * @state: the mask of task states that can be woken |
9ed3811a | 2481 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2482 | * |
2483 | * Put it on the run-queue if it's not already there. The "current" | |
2484 | * thread is always on the run-queue (except when the actual | |
2485 | * re-schedule is in progress), and as such you're allowed to do | |
2486 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2487 | * runnable without the overhead of this. | |
2488 | * | |
9ed3811a TH |
2489 | * Returns %true if @p was woken up, %false if it was already running |
2490 | * or @state didn't match @p's state. | |
1da177e4 | 2491 | */ |
7d478721 PZ |
2492 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2493 | int wake_flags) | |
1da177e4 | 2494 | { |
cc367732 | 2495 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2496 | unsigned long flags; |
371fd7e7 | 2497 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2498 | struct rq *rq; |
1da177e4 | 2499 | |
e9c84311 | 2500 | this_cpu = get_cpu(); |
2398f2c6 | 2501 | |
04e2f174 | 2502 | smp_wmb(); |
ab3b3aa5 | 2503 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2504 | if (!(p->state & state)) |
1da177e4 LT |
2505 | goto out; |
2506 | ||
dd41f596 | 2507 | if (p->se.on_rq) |
1da177e4 LT |
2508 | goto out_running; |
2509 | ||
2510 | cpu = task_cpu(p); | |
cc367732 | 2511 | orig_cpu = cpu; |
1da177e4 LT |
2512 | |
2513 | #ifdef CONFIG_SMP | |
2514 | if (unlikely(task_running(rq, p))) | |
2515 | goto out_activate; | |
2516 | ||
e9c84311 PZ |
2517 | /* |
2518 | * In order to handle concurrent wakeups and release the rq->lock | |
2519 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2520 | * |
2521 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2522 | */ |
cc87f76a PZ |
2523 | if (task_contributes_to_load(p)) { |
2524 | if (likely(cpu_online(orig_cpu))) | |
2525 | rq->nr_uninterruptible--; | |
2526 | else | |
2527 | this_rq()->nr_uninterruptible--; | |
2528 | } | |
e9c84311 | 2529 | p->state = TASK_WAKING; |
efbbd05a | 2530 | |
371fd7e7 | 2531 | if (p->sched_class->task_waking) { |
efbbd05a | 2532 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2533 | en_flags |= ENQUEUE_WAKING; |
2534 | } | |
efbbd05a | 2535 | |
0017d735 PZ |
2536 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2537 | if (cpu != orig_cpu) | |
5d2f5a61 | 2538 | set_task_cpu(p, cpu); |
0017d735 | 2539 | __task_rq_unlock(rq); |
ab19cb23 | 2540 | |
0970d299 PZ |
2541 | rq = cpu_rq(cpu); |
2542 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2543 | |
0970d299 PZ |
2544 | /* |
2545 | * We migrated the task without holding either rq->lock, however | |
2546 | * since the task is not on the task list itself, nobody else | |
2547 | * will try and migrate the task, hence the rq should match the | |
2548 | * cpu we just moved it to. | |
2549 | */ | |
2550 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2551 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2552 | |
e7693a36 GH |
2553 | #ifdef CONFIG_SCHEDSTATS |
2554 | schedstat_inc(rq, ttwu_count); | |
2555 | if (cpu == this_cpu) | |
2556 | schedstat_inc(rq, ttwu_local); | |
2557 | else { | |
2558 | struct sched_domain *sd; | |
2559 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2560 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2561 | schedstat_inc(sd, ttwu_wake_remote); |
2562 | break; | |
2563 | } | |
2564 | } | |
2565 | } | |
6d6bc0ad | 2566 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2567 | |
1da177e4 LT |
2568 | out_activate: |
2569 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2570 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2571 | cpu == this_cpu, en_flags); | |
1da177e4 | 2572 | success = 1; |
1da177e4 | 2573 | out_running: |
9ed3811a | 2574 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2575 | out: |
2576 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2577 | put_cpu(); |
1da177e4 LT |
2578 | |
2579 | return success; | |
2580 | } | |
2581 | ||
21aa9af0 TH |
2582 | /** |
2583 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2584 | * @p: the thread to be awakened | |
2585 | * | |
2586 | * Put @p on the run-queue if it's not alredy there. The caller must | |
2587 | * ensure that this_rq() is locked, @p is bound to this_rq() and not | |
2588 | * the current task. this_rq() stays locked over invocation. | |
2589 | */ | |
2590 | static void try_to_wake_up_local(struct task_struct *p) | |
2591 | { | |
2592 | struct rq *rq = task_rq(p); | |
2593 | bool success = false; | |
2594 | ||
2595 | BUG_ON(rq != this_rq()); | |
2596 | BUG_ON(p == current); | |
2597 | lockdep_assert_held(&rq->lock); | |
2598 | ||
2599 | if (!(p->state & TASK_NORMAL)) | |
2600 | return; | |
2601 | ||
2602 | if (!p->se.on_rq) { | |
2603 | if (likely(!task_running(rq, p))) { | |
2604 | schedstat_inc(rq, ttwu_count); | |
2605 | schedstat_inc(rq, ttwu_local); | |
2606 | } | |
2607 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2608 | success = true; | |
2609 | } | |
2610 | ttwu_post_activation(p, rq, 0, success); | |
2611 | } | |
2612 | ||
50fa610a DH |
2613 | /** |
2614 | * wake_up_process - Wake up a specific process | |
2615 | * @p: The process to be woken up. | |
2616 | * | |
2617 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2618 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2619 | * running. | |
2620 | * | |
2621 | * It may be assumed that this function implies a write memory barrier before | |
2622 | * changing the task state if and only if any tasks are woken up. | |
2623 | */ | |
7ad5b3a5 | 2624 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2625 | { |
d9514f6c | 2626 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2627 | } |
1da177e4 LT |
2628 | EXPORT_SYMBOL(wake_up_process); |
2629 | ||
7ad5b3a5 | 2630 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2631 | { |
2632 | return try_to_wake_up(p, state, 0); | |
2633 | } | |
2634 | ||
1da177e4 LT |
2635 | /* |
2636 | * Perform scheduler related setup for a newly forked process p. | |
2637 | * p is forked by current. | |
dd41f596 IM |
2638 | * |
2639 | * __sched_fork() is basic setup used by init_idle() too: | |
2640 | */ | |
2641 | static void __sched_fork(struct task_struct *p) | |
2642 | { | |
dd41f596 IM |
2643 | p->se.exec_start = 0; |
2644 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2645 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2646 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2647 | |
2648 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2649 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2650 | #endif |
476d139c | 2651 | |
fa717060 | 2652 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2653 | p->se.on_rq = 0; |
4a55bd5e | 2654 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2655 | |
e107be36 AK |
2656 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2657 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2658 | #endif | |
dd41f596 IM |
2659 | } |
2660 | ||
2661 | /* | |
2662 | * fork()/clone()-time setup: | |
2663 | */ | |
2664 | void sched_fork(struct task_struct *p, int clone_flags) | |
2665 | { | |
2666 | int cpu = get_cpu(); | |
2667 | ||
2668 | __sched_fork(p); | |
06b83b5f | 2669 | /* |
0017d735 | 2670 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2671 | * nobody will actually run it, and a signal or other external |
2672 | * event cannot wake it up and insert it on the runqueue either. | |
2673 | */ | |
0017d735 | 2674 | p->state = TASK_RUNNING; |
dd41f596 | 2675 | |
b9dc29e7 MG |
2676 | /* |
2677 | * Revert to default priority/policy on fork if requested. | |
2678 | */ | |
2679 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2680 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2681 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2682 | p->normal_prio = p->static_prio; |
2683 | } | |
b9dc29e7 | 2684 | |
6c697bdf MG |
2685 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2686 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2687 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2688 | set_load_weight(p); |
2689 | } | |
2690 | ||
b9dc29e7 MG |
2691 | /* |
2692 | * We don't need the reset flag anymore after the fork. It has | |
2693 | * fulfilled its duty: | |
2694 | */ | |
2695 | p->sched_reset_on_fork = 0; | |
2696 | } | |
ca94c442 | 2697 | |
f83f9ac2 PW |
2698 | /* |
2699 | * Make sure we do not leak PI boosting priority to the child. | |
2700 | */ | |
2701 | p->prio = current->normal_prio; | |
2702 | ||
2ddbf952 HS |
2703 | if (!rt_prio(p->prio)) |
2704 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2705 | |
cd29fe6f PZ |
2706 | if (p->sched_class->task_fork) |
2707 | p->sched_class->task_fork(p); | |
2708 | ||
86951599 PZ |
2709 | /* |
2710 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2711 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2712 | * is ran before sched_fork(). | |
2713 | * | |
2714 | * Silence PROVE_RCU. | |
2715 | */ | |
2716 | rcu_read_lock(); | |
5f3edc1b | 2717 | set_task_cpu(p, cpu); |
86951599 | 2718 | rcu_read_unlock(); |
5f3edc1b | 2719 | |
52f17b6c | 2720 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2721 | if (likely(sched_info_on())) |
52f17b6c | 2722 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2723 | #endif |
d6077cb8 | 2724 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2725 | p->oncpu = 0; |
2726 | #endif | |
1da177e4 | 2727 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2728 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2729 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2730 | #endif |
917b627d GH |
2731 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2732 | ||
476d139c | 2733 | put_cpu(); |
1da177e4 LT |
2734 | } |
2735 | ||
2736 | /* | |
2737 | * wake_up_new_task - wake up a newly created task for the first time. | |
2738 | * | |
2739 | * This function will do some initial scheduler statistics housekeeping | |
2740 | * that must be done for every newly created context, then puts the task | |
2741 | * on the runqueue and wakes it. | |
2742 | */ | |
7ad5b3a5 | 2743 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2744 | { |
2745 | unsigned long flags; | |
dd41f596 | 2746 | struct rq *rq; |
c890692b | 2747 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2748 | |
2749 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2750 | rq = task_rq_lock(p, &flags); |
2751 | p->state = TASK_WAKING; | |
2752 | ||
fabf318e PZ |
2753 | /* |
2754 | * Fork balancing, do it here and not earlier because: | |
2755 | * - cpus_allowed can change in the fork path | |
2756 | * - any previously selected cpu might disappear through hotplug | |
2757 | * | |
0017d735 PZ |
2758 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2759 | * without people poking at ->cpus_allowed. | |
fabf318e | 2760 | */ |
0017d735 | 2761 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2762 | set_task_cpu(p, cpu); |
1da177e4 | 2763 | |
06b83b5f | 2764 | p->state = TASK_RUNNING; |
0017d735 PZ |
2765 | task_rq_unlock(rq, &flags); |
2766 | #endif | |
2767 | ||
2768 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2769 | activate_task(rq, p, 0); |
27a9da65 | 2770 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2771 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2772 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2773 | if (p->sched_class->task_woken) |
2774 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2775 | #endif |
dd41f596 | 2776 | task_rq_unlock(rq, &flags); |
fabf318e | 2777 | put_cpu(); |
1da177e4 LT |
2778 | } |
2779 | ||
e107be36 AK |
2780 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2781 | ||
2782 | /** | |
80dd99b3 | 2783 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2784 | * @notifier: notifier struct to register |
e107be36 AK |
2785 | */ |
2786 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2787 | { | |
2788 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2789 | } | |
2790 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2791 | ||
2792 | /** | |
2793 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2794 | * @notifier: notifier struct to unregister |
e107be36 AK |
2795 | * |
2796 | * This is safe to call from within a preemption notifier. | |
2797 | */ | |
2798 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2799 | { | |
2800 | hlist_del(¬ifier->link); | |
2801 | } | |
2802 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2803 | ||
2804 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2805 | { | |
2806 | struct preempt_notifier *notifier; | |
2807 | struct hlist_node *node; | |
2808 | ||
2809 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2810 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2811 | } | |
2812 | ||
2813 | static void | |
2814 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2815 | struct task_struct *next) | |
2816 | { | |
2817 | struct preempt_notifier *notifier; | |
2818 | struct hlist_node *node; | |
2819 | ||
2820 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2821 | notifier->ops->sched_out(notifier, next); | |
2822 | } | |
2823 | ||
6d6bc0ad | 2824 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2825 | |
2826 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2827 | { | |
2828 | } | |
2829 | ||
2830 | static void | |
2831 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2832 | struct task_struct *next) | |
2833 | { | |
2834 | } | |
2835 | ||
6d6bc0ad | 2836 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2837 | |
4866cde0 NP |
2838 | /** |
2839 | * prepare_task_switch - prepare to switch tasks | |
2840 | * @rq: the runqueue preparing to switch | |
421cee29 | 2841 | * @prev: the current task that is being switched out |
4866cde0 NP |
2842 | * @next: the task we are going to switch to. |
2843 | * | |
2844 | * This is called with the rq lock held and interrupts off. It must | |
2845 | * be paired with a subsequent finish_task_switch after the context | |
2846 | * switch. | |
2847 | * | |
2848 | * prepare_task_switch sets up locking and calls architecture specific | |
2849 | * hooks. | |
2850 | */ | |
e107be36 AK |
2851 | static inline void |
2852 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2853 | struct task_struct *next) | |
4866cde0 | 2854 | { |
e107be36 | 2855 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2856 | prepare_lock_switch(rq, next); |
2857 | prepare_arch_switch(next); | |
2858 | } | |
2859 | ||
1da177e4 LT |
2860 | /** |
2861 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2862 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2863 | * @prev: the thread we just switched away from. |
2864 | * | |
4866cde0 NP |
2865 | * finish_task_switch must be called after the context switch, paired |
2866 | * with a prepare_task_switch call before the context switch. | |
2867 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2868 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2869 | * |
2870 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2871 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2872 | * with the lock held can cause deadlocks; see schedule() for |
2873 | * details.) | |
2874 | */ | |
a9957449 | 2875 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2876 | __releases(rq->lock) |
2877 | { | |
1da177e4 | 2878 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2879 | long prev_state; |
1da177e4 LT |
2880 | |
2881 | rq->prev_mm = NULL; | |
2882 | ||
2883 | /* | |
2884 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2885 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2886 | * schedule one last time. The schedule call will never return, and |
2887 | * the scheduled task must drop that reference. | |
c394cc9f | 2888 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2889 | * still held, otherwise prev could be scheduled on another cpu, die |
2890 | * there before we look at prev->state, and then the reference would | |
2891 | * be dropped twice. | |
2892 | * Manfred Spraul <manfred@colorfullife.com> | |
2893 | */ | |
55a101f8 | 2894 | prev_state = prev->state; |
4866cde0 | 2895 | finish_arch_switch(prev); |
8381f65d JI |
2896 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2897 | local_irq_disable(); | |
2898 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2899 | perf_event_task_sched_in(current); |
8381f65d JI |
2900 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2901 | local_irq_enable(); | |
2902 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2903 | finish_lock_switch(rq, prev); |
e8fa1362 | 2904 | |
e107be36 | 2905 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2906 | if (mm) |
2907 | mmdrop(mm); | |
c394cc9f | 2908 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2909 | /* |
2910 | * Remove function-return probe instances associated with this | |
2911 | * task and put them back on the free list. | |
9761eea8 | 2912 | */ |
c6fd91f0 | 2913 | kprobe_flush_task(prev); |
1da177e4 | 2914 | put_task_struct(prev); |
c6fd91f0 | 2915 | } |
1da177e4 LT |
2916 | } |
2917 | ||
3f029d3c GH |
2918 | #ifdef CONFIG_SMP |
2919 | ||
2920 | /* assumes rq->lock is held */ | |
2921 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2922 | { | |
2923 | if (prev->sched_class->pre_schedule) | |
2924 | prev->sched_class->pre_schedule(rq, prev); | |
2925 | } | |
2926 | ||
2927 | /* rq->lock is NOT held, but preemption is disabled */ | |
2928 | static inline void post_schedule(struct rq *rq) | |
2929 | { | |
2930 | if (rq->post_schedule) { | |
2931 | unsigned long flags; | |
2932 | ||
05fa785c | 2933 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2934 | if (rq->curr->sched_class->post_schedule) |
2935 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2936 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2937 | |
2938 | rq->post_schedule = 0; | |
2939 | } | |
2940 | } | |
2941 | ||
2942 | #else | |
da19ab51 | 2943 | |
3f029d3c GH |
2944 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2945 | { | |
2946 | } | |
2947 | ||
2948 | static inline void post_schedule(struct rq *rq) | |
2949 | { | |
1da177e4 LT |
2950 | } |
2951 | ||
3f029d3c GH |
2952 | #endif |
2953 | ||
1da177e4 LT |
2954 | /** |
2955 | * schedule_tail - first thing a freshly forked thread must call. | |
2956 | * @prev: the thread we just switched away from. | |
2957 | */ | |
36c8b586 | 2958 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2959 | __releases(rq->lock) |
2960 | { | |
70b97a7f IM |
2961 | struct rq *rq = this_rq(); |
2962 | ||
4866cde0 | 2963 | finish_task_switch(rq, prev); |
da19ab51 | 2964 | |
3f029d3c GH |
2965 | /* |
2966 | * FIXME: do we need to worry about rq being invalidated by the | |
2967 | * task_switch? | |
2968 | */ | |
2969 | post_schedule(rq); | |
70b97a7f | 2970 | |
4866cde0 NP |
2971 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2972 | /* In this case, finish_task_switch does not reenable preemption */ | |
2973 | preempt_enable(); | |
2974 | #endif | |
1da177e4 | 2975 | if (current->set_child_tid) |
b488893a | 2976 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2977 | } |
2978 | ||
2979 | /* | |
2980 | * context_switch - switch to the new MM and the new | |
2981 | * thread's register state. | |
2982 | */ | |
dd41f596 | 2983 | static inline void |
70b97a7f | 2984 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2985 | struct task_struct *next) |
1da177e4 | 2986 | { |
dd41f596 | 2987 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2988 | |
e107be36 | 2989 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2990 | trace_sched_switch(prev, next); |
dd41f596 IM |
2991 | mm = next->mm; |
2992 | oldmm = prev->active_mm; | |
9226d125 ZA |
2993 | /* |
2994 | * For paravirt, this is coupled with an exit in switch_to to | |
2995 | * combine the page table reload and the switch backend into | |
2996 | * one hypercall. | |
2997 | */ | |
224101ed | 2998 | arch_start_context_switch(prev); |
9226d125 | 2999 | |
31915ab4 | 3000 | if (!mm) { |
1da177e4 LT |
3001 | next->active_mm = oldmm; |
3002 | atomic_inc(&oldmm->mm_count); | |
3003 | enter_lazy_tlb(oldmm, next); | |
3004 | } else | |
3005 | switch_mm(oldmm, mm, next); | |
3006 | ||
31915ab4 | 3007 | if (!prev->mm) { |
1da177e4 | 3008 | prev->active_mm = NULL; |
1da177e4 LT |
3009 | rq->prev_mm = oldmm; |
3010 | } | |
3a5f5e48 IM |
3011 | /* |
3012 | * Since the runqueue lock will be released by the next | |
3013 | * task (which is an invalid locking op but in the case | |
3014 | * of the scheduler it's an obvious special-case), so we | |
3015 | * do an early lockdep release here: | |
3016 | */ | |
3017 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3018 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3019 | #endif |
1da177e4 LT |
3020 | |
3021 | /* Here we just switch the register state and the stack. */ | |
3022 | switch_to(prev, next, prev); | |
3023 | ||
dd41f596 IM |
3024 | barrier(); |
3025 | /* | |
3026 | * this_rq must be evaluated again because prev may have moved | |
3027 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3028 | * frame will be invalid. | |
3029 | */ | |
3030 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3031 | } |
3032 | ||
3033 | /* | |
3034 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3035 | * | |
3036 | * externally visible scheduler statistics: current number of runnable | |
3037 | * threads, current number of uninterruptible-sleeping threads, total | |
3038 | * number of context switches performed since bootup. | |
3039 | */ | |
3040 | unsigned long nr_running(void) | |
3041 | { | |
3042 | unsigned long i, sum = 0; | |
3043 | ||
3044 | for_each_online_cpu(i) | |
3045 | sum += cpu_rq(i)->nr_running; | |
3046 | ||
3047 | return sum; | |
f711f609 | 3048 | } |
1da177e4 LT |
3049 | |
3050 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3051 | { |
1da177e4 | 3052 | unsigned long i, sum = 0; |
f711f609 | 3053 | |
0a945022 | 3054 | for_each_possible_cpu(i) |
1da177e4 | 3055 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3056 | |
3057 | /* | |
1da177e4 LT |
3058 | * Since we read the counters lockless, it might be slightly |
3059 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3060 | */ |
1da177e4 LT |
3061 | if (unlikely((long)sum < 0)) |
3062 | sum = 0; | |
f711f609 | 3063 | |
1da177e4 | 3064 | return sum; |
f711f609 | 3065 | } |
f711f609 | 3066 | |
1da177e4 | 3067 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3068 | { |
cc94abfc SR |
3069 | int i; |
3070 | unsigned long long sum = 0; | |
46cb4b7c | 3071 | |
0a945022 | 3072 | for_each_possible_cpu(i) |
1da177e4 | 3073 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3074 | |
1da177e4 LT |
3075 | return sum; |
3076 | } | |
483b4ee6 | 3077 | |
1da177e4 LT |
3078 | unsigned long nr_iowait(void) |
3079 | { | |
3080 | unsigned long i, sum = 0; | |
483b4ee6 | 3081 | |
0a945022 | 3082 | for_each_possible_cpu(i) |
1da177e4 | 3083 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3084 | |
1da177e4 LT |
3085 | return sum; |
3086 | } | |
483b4ee6 | 3087 | |
8c215bd3 | 3088 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3089 | { |
8c215bd3 | 3090 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3091 | return atomic_read(&this->nr_iowait); |
3092 | } | |
46cb4b7c | 3093 | |
69d25870 AV |
3094 | unsigned long this_cpu_load(void) |
3095 | { | |
3096 | struct rq *this = this_rq(); | |
3097 | return this->cpu_load[0]; | |
3098 | } | |
e790fb0b | 3099 | |
46cb4b7c | 3100 | |
dce48a84 TG |
3101 | /* Variables and functions for calc_load */ |
3102 | static atomic_long_t calc_load_tasks; | |
3103 | static unsigned long calc_load_update; | |
3104 | unsigned long avenrun[3]; | |
3105 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3106 | |
74f5187a PZ |
3107 | static long calc_load_fold_active(struct rq *this_rq) |
3108 | { | |
3109 | long nr_active, delta = 0; | |
3110 | ||
3111 | nr_active = this_rq->nr_running; | |
3112 | nr_active += (long) this_rq->nr_uninterruptible; | |
3113 | ||
3114 | if (nr_active != this_rq->calc_load_active) { | |
3115 | delta = nr_active - this_rq->calc_load_active; | |
3116 | this_rq->calc_load_active = nr_active; | |
3117 | } | |
3118 | ||
3119 | return delta; | |
3120 | } | |
3121 | ||
0f004f5a PZ |
3122 | static unsigned long |
3123 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3124 | { | |
3125 | load *= exp; | |
3126 | load += active * (FIXED_1 - exp); | |
3127 | load += 1UL << (FSHIFT - 1); | |
3128 | return load >> FSHIFT; | |
3129 | } | |
3130 | ||
74f5187a PZ |
3131 | #ifdef CONFIG_NO_HZ |
3132 | /* | |
3133 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3134 | * | |
3135 | * When making the ILB scale, we should try to pull this in as well. | |
3136 | */ | |
3137 | static atomic_long_t calc_load_tasks_idle; | |
3138 | ||
3139 | static void calc_load_account_idle(struct rq *this_rq) | |
3140 | { | |
3141 | long delta; | |
3142 | ||
3143 | delta = calc_load_fold_active(this_rq); | |
3144 | if (delta) | |
3145 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3146 | } | |
3147 | ||
3148 | static long calc_load_fold_idle(void) | |
3149 | { | |
3150 | long delta = 0; | |
3151 | ||
3152 | /* | |
3153 | * Its got a race, we don't care... | |
3154 | */ | |
3155 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3156 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3157 | ||
3158 | return delta; | |
3159 | } | |
0f004f5a PZ |
3160 | |
3161 | /** | |
3162 | * fixed_power_int - compute: x^n, in O(log n) time | |
3163 | * | |
3164 | * @x: base of the power | |
3165 | * @frac_bits: fractional bits of @x | |
3166 | * @n: power to raise @x to. | |
3167 | * | |
3168 | * By exploiting the relation between the definition of the natural power | |
3169 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3170 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3171 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3172 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3173 | * of course trivially computable in O(log_2 n), the length of our binary | |
3174 | * vector. | |
3175 | */ | |
3176 | static unsigned long | |
3177 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3178 | { | |
3179 | unsigned long result = 1UL << frac_bits; | |
3180 | ||
3181 | if (n) for (;;) { | |
3182 | if (n & 1) { | |
3183 | result *= x; | |
3184 | result += 1UL << (frac_bits - 1); | |
3185 | result >>= frac_bits; | |
3186 | } | |
3187 | n >>= 1; | |
3188 | if (!n) | |
3189 | break; | |
3190 | x *= x; | |
3191 | x += 1UL << (frac_bits - 1); | |
3192 | x >>= frac_bits; | |
3193 | } | |
3194 | ||
3195 | return result; | |
3196 | } | |
3197 | ||
3198 | /* | |
3199 | * a1 = a0 * e + a * (1 - e) | |
3200 | * | |
3201 | * a2 = a1 * e + a * (1 - e) | |
3202 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3203 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3204 | * | |
3205 | * a3 = a2 * e + a * (1 - e) | |
3206 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3207 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3208 | * | |
3209 | * ... | |
3210 | * | |
3211 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3212 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3213 | * = a0 * e^n + a * (1 - e^n) | |
3214 | * | |
3215 | * [1] application of the geometric series: | |
3216 | * | |
3217 | * n 1 - x^(n+1) | |
3218 | * S_n := \Sum x^i = ------------- | |
3219 | * i=0 1 - x | |
3220 | */ | |
3221 | static unsigned long | |
3222 | calc_load_n(unsigned long load, unsigned long exp, | |
3223 | unsigned long active, unsigned int n) | |
3224 | { | |
3225 | ||
3226 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3227 | } | |
3228 | ||
3229 | /* | |
3230 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3231 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3232 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3233 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3234 | * | |
3235 | * Once we've updated the global active value, we need to apply the exponential | |
3236 | * weights adjusted to the number of cycles missed. | |
3237 | */ | |
3238 | static void calc_global_nohz(unsigned long ticks) | |
3239 | { | |
3240 | long delta, active, n; | |
3241 | ||
3242 | if (time_before(jiffies, calc_load_update)) | |
3243 | return; | |
3244 | ||
3245 | /* | |
3246 | * If we crossed a calc_load_update boundary, make sure to fold | |
3247 | * any pending idle changes, the respective CPUs might have | |
3248 | * missed the tick driven calc_load_account_active() update | |
3249 | * due to NO_HZ. | |
3250 | */ | |
3251 | delta = calc_load_fold_idle(); | |
3252 | if (delta) | |
3253 | atomic_long_add(delta, &calc_load_tasks); | |
3254 | ||
3255 | /* | |
3256 | * If we were idle for multiple load cycles, apply them. | |
3257 | */ | |
3258 | if (ticks >= LOAD_FREQ) { | |
3259 | n = ticks / LOAD_FREQ; | |
3260 | ||
3261 | active = atomic_long_read(&calc_load_tasks); | |
3262 | active = active > 0 ? active * FIXED_1 : 0; | |
3263 | ||
3264 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3265 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3266 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3267 | ||
3268 | calc_load_update += n * LOAD_FREQ; | |
3269 | } | |
3270 | ||
3271 | /* | |
3272 | * Its possible the remainder of the above division also crosses | |
3273 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3274 | * which comes after this will take care of that. | |
3275 | * | |
3276 | * Consider us being 11 ticks before a cycle completion, and us | |
3277 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3278 | * age us 4 cycles, and the test in calc_global_load() will | |
3279 | * pick up the final one. | |
3280 | */ | |
3281 | } | |
74f5187a PZ |
3282 | #else |
3283 | static void calc_load_account_idle(struct rq *this_rq) | |
3284 | { | |
3285 | } | |
3286 | ||
3287 | static inline long calc_load_fold_idle(void) | |
3288 | { | |
3289 | return 0; | |
3290 | } | |
0f004f5a PZ |
3291 | |
3292 | static void calc_global_nohz(unsigned long ticks) | |
3293 | { | |
3294 | } | |
74f5187a PZ |
3295 | #endif |
3296 | ||
2d02494f TG |
3297 | /** |
3298 | * get_avenrun - get the load average array | |
3299 | * @loads: pointer to dest load array | |
3300 | * @offset: offset to add | |
3301 | * @shift: shift count to shift the result left | |
3302 | * | |
3303 | * These values are estimates at best, so no need for locking. | |
3304 | */ | |
3305 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3306 | { | |
3307 | loads[0] = (avenrun[0] + offset) << shift; | |
3308 | loads[1] = (avenrun[1] + offset) << shift; | |
3309 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3310 | } |
46cb4b7c | 3311 | |
46cb4b7c | 3312 | /* |
dce48a84 TG |
3313 | * calc_load - update the avenrun load estimates 10 ticks after the |
3314 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3315 | */ |
0f004f5a | 3316 | void calc_global_load(unsigned long ticks) |
7835b98b | 3317 | { |
dce48a84 | 3318 | long active; |
1da177e4 | 3319 | |
0f004f5a PZ |
3320 | calc_global_nohz(ticks); |
3321 | ||
3322 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3323 | return; |
1da177e4 | 3324 | |
dce48a84 TG |
3325 | active = atomic_long_read(&calc_load_tasks); |
3326 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3327 | |
dce48a84 TG |
3328 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3329 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3330 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3331 | |
dce48a84 TG |
3332 | calc_load_update += LOAD_FREQ; |
3333 | } | |
1da177e4 | 3334 | |
dce48a84 | 3335 | /* |
74f5187a PZ |
3336 | * Called from update_cpu_load() to periodically update this CPU's |
3337 | * active count. | |
dce48a84 TG |
3338 | */ |
3339 | static void calc_load_account_active(struct rq *this_rq) | |
3340 | { | |
74f5187a | 3341 | long delta; |
08c183f3 | 3342 | |
74f5187a PZ |
3343 | if (time_before(jiffies, this_rq->calc_load_update)) |
3344 | return; | |
783609c6 | 3345 | |
74f5187a PZ |
3346 | delta = calc_load_fold_active(this_rq); |
3347 | delta += calc_load_fold_idle(); | |
3348 | if (delta) | |
dce48a84 | 3349 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3350 | |
3351 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3352 | } |
3353 | ||
fdf3e95d VP |
3354 | /* |
3355 | * The exact cpuload at various idx values, calculated at every tick would be | |
3356 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3357 | * | |
3358 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3359 | * on nth tick when cpu may be busy, then we have: | |
3360 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3361 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3362 | * | |
3363 | * decay_load_missed() below does efficient calculation of | |
3364 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3365 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3366 | * | |
3367 | * The calculation is approximated on a 128 point scale. | |
3368 | * degrade_zero_ticks is the number of ticks after which load at any | |
3369 | * particular idx is approximated to be zero. | |
3370 | * degrade_factor is a precomputed table, a row for each load idx. | |
3371 | * Each column corresponds to degradation factor for a power of two ticks, | |
3372 | * based on 128 point scale. | |
3373 | * Example: | |
3374 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3375 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3376 | * | |
3377 | * With this power of 2 load factors, we can degrade the load n times | |
3378 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3379 | * n mult/shifts needed by the exact degradation. | |
3380 | */ | |
3381 | #define DEGRADE_SHIFT 7 | |
3382 | static const unsigned char | |
3383 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3384 | static const unsigned char | |
3385 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3386 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3387 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3388 | {96, 72, 40, 12, 1, 0, 0}, | |
3389 | {112, 98, 75, 43, 15, 1, 0}, | |
3390 | {120, 112, 98, 76, 45, 16, 2} }; | |
3391 | ||
3392 | /* | |
3393 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3394 | * would be when CPU is idle and so we just decay the old load without | |
3395 | * adding any new load. | |
3396 | */ | |
3397 | static unsigned long | |
3398 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3399 | { | |
3400 | int j = 0; | |
3401 | ||
3402 | if (!missed_updates) | |
3403 | return load; | |
3404 | ||
3405 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3406 | return 0; | |
3407 | ||
3408 | if (idx == 1) | |
3409 | return load >> missed_updates; | |
3410 | ||
3411 | while (missed_updates) { | |
3412 | if (missed_updates % 2) | |
3413 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3414 | ||
3415 | missed_updates >>= 1; | |
3416 | j++; | |
3417 | } | |
3418 | return load; | |
3419 | } | |
3420 | ||
46cb4b7c | 3421 | /* |
dd41f596 | 3422 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3423 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3424 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3425 | */ |
dd41f596 | 3426 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3427 | { |
495eca49 | 3428 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3429 | unsigned long curr_jiffies = jiffies; |
3430 | unsigned long pending_updates; | |
dd41f596 | 3431 | int i, scale; |
46cb4b7c | 3432 | |
dd41f596 | 3433 | this_rq->nr_load_updates++; |
46cb4b7c | 3434 | |
fdf3e95d VP |
3435 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3436 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3437 | return; | |
3438 | ||
3439 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3440 | this_rq->last_load_update_tick = curr_jiffies; | |
3441 | ||
dd41f596 | 3442 | /* Update our load: */ |
fdf3e95d VP |
3443 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3444 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3445 | unsigned long old_load, new_load; |
7d1e6a9b | 3446 | |
dd41f596 | 3447 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3448 | |
dd41f596 | 3449 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3450 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3451 | new_load = this_load; |
a25707f3 IM |
3452 | /* |
3453 | * Round up the averaging division if load is increasing. This | |
3454 | * prevents us from getting stuck on 9 if the load is 10, for | |
3455 | * example. | |
3456 | */ | |
3457 | if (new_load > old_load) | |
fdf3e95d VP |
3458 | new_load += scale - 1; |
3459 | ||
3460 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3461 | } |
da2b71ed SS |
3462 | |
3463 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3464 | } |
3465 | ||
3466 | static void update_cpu_load_active(struct rq *this_rq) | |
3467 | { | |
3468 | update_cpu_load(this_rq); | |
46cb4b7c | 3469 | |
74f5187a | 3470 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3471 | } |
3472 | ||
dd41f596 | 3473 | #ifdef CONFIG_SMP |
8a0be9ef | 3474 | |
46cb4b7c | 3475 | /* |
38022906 PZ |
3476 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3477 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3478 | */ |
38022906 | 3479 | void sched_exec(void) |
46cb4b7c | 3480 | { |
38022906 | 3481 | struct task_struct *p = current; |
1da177e4 | 3482 | unsigned long flags; |
70b97a7f | 3483 | struct rq *rq; |
0017d735 | 3484 | int dest_cpu; |
46cb4b7c | 3485 | |
1da177e4 | 3486 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3487 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3488 | if (dest_cpu == smp_processor_id()) | |
3489 | goto unlock; | |
38022906 | 3490 | |
46cb4b7c | 3491 | /* |
38022906 | 3492 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3493 | */ |
30da688e | 3494 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
969c7921 TH |
3495 | likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) { |
3496 | struct migration_arg arg = { p, dest_cpu }; | |
46cb4b7c | 3497 | |
1da177e4 | 3498 | task_rq_unlock(rq, &flags); |
969c7921 | 3499 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3500 | return; |
3501 | } | |
0017d735 | 3502 | unlock: |
1da177e4 | 3503 | task_rq_unlock(rq, &flags); |
1da177e4 | 3504 | } |
dd41f596 | 3505 | |
1da177e4 LT |
3506 | #endif |
3507 | ||
1da177e4 LT |
3508 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3509 | ||
3510 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3511 | ||
3512 | /* | |
c5f8d995 | 3513 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3514 | * @p in case that task is currently running. |
c5f8d995 HS |
3515 | * |
3516 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3517 | */ |
c5f8d995 HS |
3518 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3519 | { | |
3520 | u64 ns = 0; | |
3521 | ||
3522 | if (task_current(rq, p)) { | |
3523 | update_rq_clock(rq); | |
305e6835 | 3524 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3525 | if ((s64)ns < 0) |
3526 | ns = 0; | |
3527 | } | |
3528 | ||
3529 | return ns; | |
3530 | } | |
3531 | ||
bb34d92f | 3532 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3533 | { |
1da177e4 | 3534 | unsigned long flags; |
41b86e9c | 3535 | struct rq *rq; |
bb34d92f | 3536 | u64 ns = 0; |
48f24c4d | 3537 | |
41b86e9c | 3538 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3539 | ns = do_task_delta_exec(p, rq); |
3540 | task_rq_unlock(rq, &flags); | |
1508487e | 3541 | |
c5f8d995 HS |
3542 | return ns; |
3543 | } | |
f06febc9 | 3544 | |
c5f8d995 HS |
3545 | /* |
3546 | * Return accounted runtime for the task. | |
3547 | * In case the task is currently running, return the runtime plus current's | |
3548 | * pending runtime that have not been accounted yet. | |
3549 | */ | |
3550 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3551 | { | |
3552 | unsigned long flags; | |
3553 | struct rq *rq; | |
3554 | u64 ns = 0; | |
3555 | ||
3556 | rq = task_rq_lock(p, &flags); | |
3557 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3558 | task_rq_unlock(rq, &flags); | |
3559 | ||
3560 | return ns; | |
3561 | } | |
48f24c4d | 3562 | |
c5f8d995 HS |
3563 | /* |
3564 | * Return sum_exec_runtime for the thread group. | |
3565 | * In case the task is currently running, return the sum plus current's | |
3566 | * pending runtime that have not been accounted yet. | |
3567 | * | |
3568 | * Note that the thread group might have other running tasks as well, | |
3569 | * so the return value not includes other pending runtime that other | |
3570 | * running tasks might have. | |
3571 | */ | |
3572 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3573 | { | |
3574 | struct task_cputime totals; | |
3575 | unsigned long flags; | |
3576 | struct rq *rq; | |
3577 | u64 ns; | |
3578 | ||
3579 | rq = task_rq_lock(p, &flags); | |
3580 | thread_group_cputime(p, &totals); | |
3581 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3582 | task_rq_unlock(rq, &flags); |
48f24c4d | 3583 | |
1da177e4 LT |
3584 | return ns; |
3585 | } | |
3586 | ||
1da177e4 LT |
3587 | /* |
3588 | * Account user cpu time to a process. | |
3589 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3590 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3591 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3592 | */ |
457533a7 MS |
3593 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3594 | cputime_t cputime_scaled) | |
1da177e4 LT |
3595 | { |
3596 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3597 | cputime64_t tmp; | |
3598 | ||
457533a7 | 3599 | /* Add user time to process. */ |
1da177e4 | 3600 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3601 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3602 | account_group_user_time(p, cputime); |
1da177e4 LT |
3603 | |
3604 | /* Add user time to cpustat. */ | |
3605 | tmp = cputime_to_cputime64(cputime); | |
3606 | if (TASK_NICE(p) > 0) | |
3607 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3608 | else | |
3609 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3610 | |
3611 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3612 | /* Account for user time used */ |
3613 | acct_update_integrals(p); | |
1da177e4 LT |
3614 | } |
3615 | ||
94886b84 LV |
3616 | /* |
3617 | * Account guest cpu time to a process. | |
3618 | * @p: the process that the cpu time gets accounted to | |
3619 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3620 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3621 | */ |
457533a7 MS |
3622 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3623 | cputime_t cputime_scaled) | |
94886b84 LV |
3624 | { |
3625 | cputime64_t tmp; | |
3626 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3627 | ||
3628 | tmp = cputime_to_cputime64(cputime); | |
3629 | ||
457533a7 | 3630 | /* Add guest time to process. */ |
94886b84 | 3631 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3632 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3633 | account_group_user_time(p, cputime); |
94886b84 LV |
3634 | p->gtime = cputime_add(p->gtime, cputime); |
3635 | ||
457533a7 | 3636 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3637 | if (TASK_NICE(p) > 0) { |
3638 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3639 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3640 | } else { | |
3641 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3642 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3643 | } | |
94886b84 LV |
3644 | } |
3645 | ||
1da177e4 LT |
3646 | /* |
3647 | * Account system cpu time to a process. | |
3648 | * @p: the process that the cpu time gets accounted to | |
3649 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3650 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3651 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3652 | */ |
3653 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3654 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3655 | { |
3656 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3657 | cputime64_t tmp; |
3658 | ||
983ed7a6 | 3659 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3660 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3661 | return; |
3662 | } | |
94886b84 | 3663 | |
457533a7 | 3664 | /* Add system time to process. */ |
1da177e4 | 3665 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3666 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3667 | account_group_system_time(p, cputime); |
1da177e4 LT |
3668 | |
3669 | /* Add system time to cpustat. */ | |
3670 | tmp = cputime_to_cputime64(cputime); | |
3671 | if (hardirq_count() - hardirq_offset) | |
3672 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
75e1056f | 3673 | else if (in_serving_softirq()) |
1da177e4 | 3674 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); |
1da177e4 | 3675 | else |
79741dd3 MS |
3676 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3677 | ||
ef12fefa BR |
3678 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3679 | ||
1da177e4 LT |
3680 | /* Account for system time used */ |
3681 | acct_update_integrals(p); | |
1da177e4 LT |
3682 | } |
3683 | ||
c66f08be | 3684 | /* |
1da177e4 | 3685 | * Account for involuntary wait time. |
1da177e4 | 3686 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3687 | */ |
79741dd3 | 3688 | void account_steal_time(cputime_t cputime) |
c66f08be | 3689 | { |
79741dd3 MS |
3690 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3691 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3692 | ||
3693 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3694 | } |
3695 | ||
1da177e4 | 3696 | /* |
79741dd3 MS |
3697 | * Account for idle time. |
3698 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3699 | */ |
79741dd3 | 3700 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3701 | { |
3702 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3703 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3704 | struct rq *rq = this_rq(); |
1da177e4 | 3705 | |
79741dd3 MS |
3706 | if (atomic_read(&rq->nr_iowait) > 0) |
3707 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3708 | else | |
3709 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3710 | } |
3711 | ||
79741dd3 MS |
3712 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3713 | ||
3714 | /* | |
3715 | * Account a single tick of cpu time. | |
3716 | * @p: the process that the cpu time gets accounted to | |
3717 | * @user_tick: indicates if the tick is a user or a system tick | |
3718 | */ | |
3719 | void account_process_tick(struct task_struct *p, int user_tick) | |
3720 | { | |
a42548a1 | 3721 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3722 | struct rq *rq = this_rq(); |
3723 | ||
3724 | if (user_tick) | |
a42548a1 | 3725 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3726 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3727 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3728 | one_jiffy_scaled); |
3729 | else | |
a42548a1 | 3730 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3731 | } |
3732 | ||
3733 | /* | |
3734 | * Account multiple ticks of steal time. | |
3735 | * @p: the process from which the cpu time has been stolen | |
3736 | * @ticks: number of stolen ticks | |
3737 | */ | |
3738 | void account_steal_ticks(unsigned long ticks) | |
3739 | { | |
3740 | account_steal_time(jiffies_to_cputime(ticks)); | |
3741 | } | |
3742 | ||
3743 | /* | |
3744 | * Account multiple ticks of idle time. | |
3745 | * @ticks: number of stolen ticks | |
3746 | */ | |
3747 | void account_idle_ticks(unsigned long ticks) | |
3748 | { | |
3749 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3750 | } |
3751 | ||
79741dd3 MS |
3752 | #endif |
3753 | ||
49048622 BS |
3754 | /* |
3755 | * Use precise platform statistics if available: | |
3756 | */ | |
3757 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3758 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3759 | { |
d99ca3b9 HS |
3760 | *ut = p->utime; |
3761 | *st = p->stime; | |
49048622 BS |
3762 | } |
3763 | ||
0cf55e1e | 3764 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3765 | { |
0cf55e1e HS |
3766 | struct task_cputime cputime; |
3767 | ||
3768 | thread_group_cputime(p, &cputime); | |
3769 | ||
3770 | *ut = cputime.utime; | |
3771 | *st = cputime.stime; | |
49048622 BS |
3772 | } |
3773 | #else | |
761b1d26 HS |
3774 | |
3775 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3776 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3777 | #endif |
3778 | ||
d180c5bc | 3779 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3780 | { |
d99ca3b9 | 3781 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3782 | |
3783 | /* | |
3784 | * Use CFS's precise accounting: | |
3785 | */ | |
d180c5bc | 3786 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3787 | |
3788 | if (total) { | |
e75e863d | 3789 | u64 temp = rtime; |
d180c5bc | 3790 | |
e75e863d | 3791 | temp *= utime; |
49048622 | 3792 | do_div(temp, total); |
d180c5bc HS |
3793 | utime = (cputime_t)temp; |
3794 | } else | |
3795 | utime = rtime; | |
49048622 | 3796 | |
d180c5bc HS |
3797 | /* |
3798 | * Compare with previous values, to keep monotonicity: | |
3799 | */ | |
761b1d26 | 3800 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3801 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3802 | |
d99ca3b9 HS |
3803 | *ut = p->prev_utime; |
3804 | *st = p->prev_stime; | |
49048622 BS |
3805 | } |
3806 | ||
0cf55e1e HS |
3807 | /* |
3808 | * Must be called with siglock held. | |
3809 | */ | |
3810 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3811 | { |
0cf55e1e HS |
3812 | struct signal_struct *sig = p->signal; |
3813 | struct task_cputime cputime; | |
3814 | cputime_t rtime, utime, total; | |
49048622 | 3815 | |
0cf55e1e | 3816 | thread_group_cputime(p, &cputime); |
49048622 | 3817 | |
0cf55e1e HS |
3818 | total = cputime_add(cputime.utime, cputime.stime); |
3819 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3820 | |
0cf55e1e | 3821 | if (total) { |
e75e863d | 3822 | u64 temp = rtime; |
49048622 | 3823 | |
e75e863d | 3824 | temp *= cputime.utime; |
0cf55e1e HS |
3825 | do_div(temp, total); |
3826 | utime = (cputime_t)temp; | |
3827 | } else | |
3828 | utime = rtime; | |
3829 | ||
3830 | sig->prev_utime = max(sig->prev_utime, utime); | |
3831 | sig->prev_stime = max(sig->prev_stime, | |
3832 | cputime_sub(rtime, sig->prev_utime)); | |
3833 | ||
3834 | *ut = sig->prev_utime; | |
3835 | *st = sig->prev_stime; | |
49048622 | 3836 | } |
49048622 | 3837 | #endif |
49048622 | 3838 | |
7835b98b CL |
3839 | /* |
3840 | * This function gets called by the timer code, with HZ frequency. | |
3841 | * We call it with interrupts disabled. | |
3842 | * | |
3843 | * It also gets called by the fork code, when changing the parent's | |
3844 | * timeslices. | |
3845 | */ | |
3846 | void scheduler_tick(void) | |
3847 | { | |
7835b98b CL |
3848 | int cpu = smp_processor_id(); |
3849 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3850 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3851 | |
3852 | sched_clock_tick(); | |
dd41f596 | 3853 | |
05fa785c | 3854 | raw_spin_lock(&rq->lock); |
3e51f33f | 3855 | update_rq_clock(rq); |
fdf3e95d | 3856 | update_cpu_load_active(rq); |
fa85ae24 | 3857 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3858 | raw_spin_unlock(&rq->lock); |
7835b98b | 3859 | |
e9d2b064 | 3860 | perf_event_task_tick(); |
e220d2dc | 3861 | |
e418e1c2 | 3862 | #ifdef CONFIG_SMP |
dd41f596 IM |
3863 | rq->idle_at_tick = idle_cpu(cpu); |
3864 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3865 | #endif |
1da177e4 LT |
3866 | } |
3867 | ||
132380a0 | 3868 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3869 | { |
3870 | if (in_lock_functions(addr)) { | |
3871 | addr = CALLER_ADDR2; | |
3872 | if (in_lock_functions(addr)) | |
3873 | addr = CALLER_ADDR3; | |
3874 | } | |
3875 | return addr; | |
3876 | } | |
1da177e4 | 3877 | |
7e49fcce SR |
3878 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3879 | defined(CONFIG_PREEMPT_TRACER)) | |
3880 | ||
43627582 | 3881 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3882 | { |
6cd8a4bb | 3883 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3884 | /* |
3885 | * Underflow? | |
3886 | */ | |
9a11b49a IM |
3887 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3888 | return; | |
6cd8a4bb | 3889 | #endif |
1da177e4 | 3890 | preempt_count() += val; |
6cd8a4bb | 3891 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3892 | /* |
3893 | * Spinlock count overflowing soon? | |
3894 | */ | |
33859f7f MOS |
3895 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3896 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3897 | #endif |
3898 | if (preempt_count() == val) | |
3899 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3900 | } |
3901 | EXPORT_SYMBOL(add_preempt_count); | |
3902 | ||
43627582 | 3903 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3904 | { |
6cd8a4bb | 3905 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3906 | /* |
3907 | * Underflow? | |
3908 | */ | |
01e3eb82 | 3909 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3910 | return; |
1da177e4 LT |
3911 | /* |
3912 | * Is the spinlock portion underflowing? | |
3913 | */ | |
9a11b49a IM |
3914 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3915 | !(preempt_count() & PREEMPT_MASK))) | |
3916 | return; | |
6cd8a4bb | 3917 | #endif |
9a11b49a | 3918 | |
6cd8a4bb SR |
3919 | if (preempt_count() == val) |
3920 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3921 | preempt_count() -= val; |
3922 | } | |
3923 | EXPORT_SYMBOL(sub_preempt_count); | |
3924 | ||
3925 | #endif | |
3926 | ||
3927 | /* | |
dd41f596 | 3928 | * Print scheduling while atomic bug: |
1da177e4 | 3929 | */ |
dd41f596 | 3930 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3931 | { |
838225b4 SS |
3932 | struct pt_regs *regs = get_irq_regs(); |
3933 | ||
3df0fc5b PZ |
3934 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3935 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3936 | |
dd41f596 | 3937 | debug_show_held_locks(prev); |
e21f5b15 | 3938 | print_modules(); |
dd41f596 IM |
3939 | if (irqs_disabled()) |
3940 | print_irqtrace_events(prev); | |
838225b4 SS |
3941 | |
3942 | if (regs) | |
3943 | show_regs(regs); | |
3944 | else | |
3945 | dump_stack(); | |
dd41f596 | 3946 | } |
1da177e4 | 3947 | |
dd41f596 IM |
3948 | /* |
3949 | * Various schedule()-time debugging checks and statistics: | |
3950 | */ | |
3951 | static inline void schedule_debug(struct task_struct *prev) | |
3952 | { | |
1da177e4 | 3953 | /* |
41a2d6cf | 3954 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3955 | * schedule() atomically, we ignore that path for now. |
3956 | * Otherwise, whine if we are scheduling when we should not be. | |
3957 | */ | |
3f33a7ce | 3958 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3959 | __schedule_bug(prev); |
3960 | ||
1da177e4 LT |
3961 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3962 | ||
2d72376b | 3963 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3964 | #ifdef CONFIG_SCHEDSTATS |
3965 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3966 | schedstat_inc(this_rq(), bkl_count); |
3967 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3968 | } |
3969 | #endif | |
dd41f596 IM |
3970 | } |
3971 | ||
6cecd084 | 3972 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3973 | { |
a64692a3 MG |
3974 | if (prev->se.on_rq) |
3975 | update_rq_clock(rq); | |
3976 | rq->skip_clock_update = 0; | |
6cecd084 | 3977 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3978 | } |
3979 | ||
dd41f596 IM |
3980 | /* |
3981 | * Pick up the highest-prio task: | |
3982 | */ | |
3983 | static inline struct task_struct * | |
b67802ea | 3984 | pick_next_task(struct rq *rq) |
dd41f596 | 3985 | { |
5522d5d5 | 3986 | const struct sched_class *class; |
dd41f596 | 3987 | struct task_struct *p; |
1da177e4 LT |
3988 | |
3989 | /* | |
dd41f596 IM |
3990 | * Optimization: we know that if all tasks are in |
3991 | * the fair class we can call that function directly: | |
1da177e4 | 3992 | */ |
dd41f596 | 3993 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3994 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3995 | if (likely(p)) |
3996 | return p; | |
1da177e4 LT |
3997 | } |
3998 | ||
34f971f6 | 3999 | for_each_class(class) { |
fb8d4724 | 4000 | p = class->pick_next_task(rq); |
dd41f596 IM |
4001 | if (p) |
4002 | return p; | |
dd41f596 | 4003 | } |
34f971f6 PZ |
4004 | |
4005 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4006 | } |
1da177e4 | 4007 | |
dd41f596 IM |
4008 | /* |
4009 | * schedule() is the main scheduler function. | |
4010 | */ | |
ff743345 | 4011 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4012 | { |
4013 | struct task_struct *prev, *next; | |
67ca7bde | 4014 | unsigned long *switch_count; |
dd41f596 | 4015 | struct rq *rq; |
31656519 | 4016 | int cpu; |
dd41f596 | 4017 | |
ff743345 PZ |
4018 | need_resched: |
4019 | preempt_disable(); | |
dd41f596 IM |
4020 | cpu = smp_processor_id(); |
4021 | rq = cpu_rq(cpu); | |
25502a6c | 4022 | rcu_note_context_switch(cpu); |
dd41f596 | 4023 | prev = rq->curr; |
dd41f596 IM |
4024 | |
4025 | release_kernel_lock(prev); | |
4026 | need_resched_nonpreemptible: | |
4027 | ||
4028 | schedule_debug(prev); | |
1da177e4 | 4029 | |
31656519 | 4030 | if (sched_feat(HRTICK)) |
f333fdc9 | 4031 | hrtick_clear(rq); |
8f4d37ec | 4032 | |
05fa785c | 4033 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 4034 | clear_tsk_need_resched(prev); |
1da177e4 | 4035 | |
246d86b5 | 4036 | switch_count = &prev->nivcsw; |
1da177e4 | 4037 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4038 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4039 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
4040 | } else { |
4041 | /* | |
4042 | * If a worker is going to sleep, notify and | |
4043 | * ask workqueue whether it wants to wake up a | |
4044 | * task to maintain concurrency. If so, wake | |
4045 | * up the task. | |
4046 | */ | |
4047 | if (prev->flags & PF_WQ_WORKER) { | |
4048 | struct task_struct *to_wakeup; | |
4049 | ||
4050 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4051 | if (to_wakeup) | |
4052 | try_to_wake_up_local(to_wakeup); | |
4053 | } | |
371fd7e7 | 4054 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 4055 | } |
dd41f596 | 4056 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4057 | } |
4058 | ||
3f029d3c | 4059 | pre_schedule(rq, prev); |
f65eda4f | 4060 | |
dd41f596 | 4061 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4062 | idle_balance(cpu, rq); |
1da177e4 | 4063 | |
df1c99d4 | 4064 | put_prev_task(rq, prev); |
b67802ea | 4065 | next = pick_next_task(rq); |
1da177e4 | 4066 | |
1da177e4 | 4067 | if (likely(prev != next)) { |
673a90a1 | 4068 | sched_info_switch(prev, next); |
49f47433 | 4069 | perf_event_task_sched_out(prev, next); |
673a90a1 | 4070 | |
1da177e4 LT |
4071 | rq->nr_switches++; |
4072 | rq->curr = next; | |
4073 | ++*switch_count; | |
4074 | ||
dd41f596 | 4075 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4076 | /* |
246d86b5 ON |
4077 | * The context switch have flipped the stack from under us |
4078 | * and restored the local variables which were saved when | |
4079 | * this task called schedule() in the past. prev == current | |
4080 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4081 | */ |
4082 | cpu = smp_processor_id(); | |
4083 | rq = cpu_rq(cpu); | |
1da177e4 | 4084 | } else |
05fa785c | 4085 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4086 | |
3f029d3c | 4087 | post_schedule(rq); |
1da177e4 | 4088 | |
246d86b5 | 4089 | if (unlikely(reacquire_kernel_lock(prev))) |
1da177e4 | 4090 | goto need_resched_nonpreemptible; |
8f4d37ec | 4091 | |
1da177e4 | 4092 | preempt_enable_no_resched(); |
ff743345 | 4093 | if (need_resched()) |
1da177e4 LT |
4094 | goto need_resched; |
4095 | } | |
1da177e4 LT |
4096 | EXPORT_SYMBOL(schedule); |
4097 | ||
c08f7829 | 4098 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
4099 | /* |
4100 | * Look out! "owner" is an entirely speculative pointer | |
4101 | * access and not reliable. | |
4102 | */ | |
4103 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
4104 | { | |
4105 | unsigned int cpu; | |
4106 | struct rq *rq; | |
4107 | ||
4108 | if (!sched_feat(OWNER_SPIN)) | |
4109 | return 0; | |
4110 | ||
4111 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
4112 | /* | |
4113 | * Need to access the cpu field knowing that | |
4114 | * DEBUG_PAGEALLOC could have unmapped it if | |
4115 | * the mutex owner just released it and exited. | |
4116 | */ | |
4117 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 4118 | return 0; |
0d66bf6d PZ |
4119 | #else |
4120 | cpu = owner->cpu; | |
4121 | #endif | |
4122 | ||
4123 | /* | |
4124 | * Even if the access succeeded (likely case), | |
4125 | * the cpu field may no longer be valid. | |
4126 | */ | |
4127 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 4128 | return 0; |
0d66bf6d PZ |
4129 | |
4130 | /* | |
4131 | * We need to validate that we can do a | |
4132 | * get_cpu() and that we have the percpu area. | |
4133 | */ | |
4134 | if (!cpu_online(cpu)) | |
4b402210 | 4135 | return 0; |
0d66bf6d PZ |
4136 | |
4137 | rq = cpu_rq(cpu); | |
4138 | ||
4139 | for (;;) { | |
4140 | /* | |
4141 | * Owner changed, break to re-assess state. | |
4142 | */ | |
9d0f4dcc TC |
4143 | if (lock->owner != owner) { |
4144 | /* | |
4145 | * If the lock has switched to a different owner, | |
4146 | * we likely have heavy contention. Return 0 to quit | |
4147 | * optimistic spinning and not contend further: | |
4148 | */ | |
4149 | if (lock->owner) | |
4150 | return 0; | |
0d66bf6d | 4151 | break; |
9d0f4dcc | 4152 | } |
0d66bf6d PZ |
4153 | |
4154 | /* | |
4155 | * Is that owner really running on that cpu? | |
4156 | */ | |
4157 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
4158 | return 0; | |
4159 | ||
4160 | cpu_relax(); | |
4161 | } | |
4b402210 | 4162 | |
0d66bf6d PZ |
4163 | return 1; |
4164 | } | |
4165 | #endif | |
4166 | ||
1da177e4 LT |
4167 | #ifdef CONFIG_PREEMPT |
4168 | /* | |
2ed6e34f | 4169 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4170 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4171 | * occur there and call schedule directly. |
4172 | */ | |
d1f74e20 | 4173 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4174 | { |
4175 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4176 | |
1da177e4 LT |
4177 | /* |
4178 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4179 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4180 | */ |
beed33a8 | 4181 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4182 | return; |
4183 | ||
3a5c359a | 4184 | do { |
d1f74e20 | 4185 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4186 | schedule(); |
d1f74e20 | 4187 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4188 | |
3a5c359a AK |
4189 | /* |
4190 | * Check again in case we missed a preemption opportunity | |
4191 | * between schedule and now. | |
4192 | */ | |
4193 | barrier(); | |
5ed0cec0 | 4194 | } while (need_resched()); |
1da177e4 | 4195 | } |
1da177e4 LT |
4196 | EXPORT_SYMBOL(preempt_schedule); |
4197 | ||
4198 | /* | |
2ed6e34f | 4199 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4200 | * off of irq context. |
4201 | * Note, that this is called and return with irqs disabled. This will | |
4202 | * protect us against recursive calling from irq. | |
4203 | */ | |
4204 | asmlinkage void __sched preempt_schedule_irq(void) | |
4205 | { | |
4206 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4207 | |
2ed6e34f | 4208 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4209 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4210 | ||
3a5c359a AK |
4211 | do { |
4212 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4213 | local_irq_enable(); |
4214 | schedule(); | |
4215 | local_irq_disable(); | |
3a5c359a | 4216 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4217 | |
3a5c359a AK |
4218 | /* |
4219 | * Check again in case we missed a preemption opportunity | |
4220 | * between schedule and now. | |
4221 | */ | |
4222 | barrier(); | |
5ed0cec0 | 4223 | } while (need_resched()); |
1da177e4 LT |
4224 | } |
4225 | ||
4226 | #endif /* CONFIG_PREEMPT */ | |
4227 | ||
63859d4f | 4228 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4229 | void *key) |
1da177e4 | 4230 | { |
63859d4f | 4231 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4232 | } |
1da177e4 LT |
4233 | EXPORT_SYMBOL(default_wake_function); |
4234 | ||
4235 | /* | |
41a2d6cf IM |
4236 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4237 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4238 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4239 | * | |
4240 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4241 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4242 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4243 | */ | |
78ddb08f | 4244 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4245 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4246 | { |
2e45874c | 4247 | wait_queue_t *curr, *next; |
1da177e4 | 4248 | |
2e45874c | 4249 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4250 | unsigned flags = curr->flags; |
4251 | ||
63859d4f | 4252 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4253 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4254 | break; |
4255 | } | |
4256 | } | |
4257 | ||
4258 | /** | |
4259 | * __wake_up - wake up threads blocked on a waitqueue. | |
4260 | * @q: the waitqueue | |
4261 | * @mode: which threads | |
4262 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4263 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4264 | * |
4265 | * It may be assumed that this function implies a write memory barrier before | |
4266 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4267 | */ |
7ad5b3a5 | 4268 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4269 | int nr_exclusive, void *key) |
1da177e4 LT |
4270 | { |
4271 | unsigned long flags; | |
4272 | ||
4273 | spin_lock_irqsave(&q->lock, flags); | |
4274 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4275 | spin_unlock_irqrestore(&q->lock, flags); | |
4276 | } | |
1da177e4 LT |
4277 | EXPORT_SYMBOL(__wake_up); |
4278 | ||
4279 | /* | |
4280 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4281 | */ | |
7ad5b3a5 | 4282 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4283 | { |
4284 | __wake_up_common(q, mode, 1, 0, NULL); | |
4285 | } | |
22c43c81 | 4286 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4287 | |
4ede816a DL |
4288 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4289 | { | |
4290 | __wake_up_common(q, mode, 1, 0, key); | |
4291 | } | |
4292 | ||
1da177e4 | 4293 | /** |
4ede816a | 4294 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4295 | * @q: the waitqueue |
4296 | * @mode: which threads | |
4297 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4298 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4299 | * |
4300 | * The sync wakeup differs that the waker knows that it will schedule | |
4301 | * away soon, so while the target thread will be woken up, it will not | |
4302 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4303 | * with each other. This can prevent needless bouncing between CPUs. | |
4304 | * | |
4305 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4306 | * |
4307 | * It may be assumed that this function implies a write memory barrier before | |
4308 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4309 | */ |
4ede816a DL |
4310 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4311 | int nr_exclusive, void *key) | |
1da177e4 LT |
4312 | { |
4313 | unsigned long flags; | |
7d478721 | 4314 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4315 | |
4316 | if (unlikely(!q)) | |
4317 | return; | |
4318 | ||
4319 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4320 | wake_flags = 0; |
1da177e4 LT |
4321 | |
4322 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4323 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4324 | spin_unlock_irqrestore(&q->lock, flags); |
4325 | } | |
4ede816a DL |
4326 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4327 | ||
4328 | /* | |
4329 | * __wake_up_sync - see __wake_up_sync_key() | |
4330 | */ | |
4331 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4332 | { | |
4333 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4334 | } | |
1da177e4 LT |
4335 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4336 | ||
65eb3dc6 KD |
4337 | /** |
4338 | * complete: - signals a single thread waiting on this completion | |
4339 | * @x: holds the state of this particular completion | |
4340 | * | |
4341 | * This will wake up a single thread waiting on this completion. Threads will be | |
4342 | * awakened in the same order in which they were queued. | |
4343 | * | |
4344 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4345 | * |
4346 | * It may be assumed that this function implies a write memory barrier before | |
4347 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4348 | */ |
b15136e9 | 4349 | void complete(struct completion *x) |
1da177e4 LT |
4350 | { |
4351 | unsigned long flags; | |
4352 | ||
4353 | spin_lock_irqsave(&x->wait.lock, flags); | |
4354 | x->done++; | |
d9514f6c | 4355 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4356 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4357 | } | |
4358 | EXPORT_SYMBOL(complete); | |
4359 | ||
65eb3dc6 KD |
4360 | /** |
4361 | * complete_all: - signals all threads waiting on this completion | |
4362 | * @x: holds the state of this particular completion | |
4363 | * | |
4364 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4365 | * |
4366 | * It may be assumed that this function implies a write memory barrier before | |
4367 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4368 | */ |
b15136e9 | 4369 | void complete_all(struct completion *x) |
1da177e4 LT |
4370 | { |
4371 | unsigned long flags; | |
4372 | ||
4373 | spin_lock_irqsave(&x->wait.lock, flags); | |
4374 | x->done += UINT_MAX/2; | |
d9514f6c | 4375 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4376 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4377 | } | |
4378 | EXPORT_SYMBOL(complete_all); | |
4379 | ||
8cbbe86d AK |
4380 | static inline long __sched |
4381 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4382 | { |
1da177e4 LT |
4383 | if (!x->done) { |
4384 | DECLARE_WAITQUEUE(wait, current); | |
4385 | ||
a93d2f17 | 4386 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4387 | do { |
94d3d824 | 4388 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4389 | timeout = -ERESTARTSYS; |
4390 | break; | |
8cbbe86d AK |
4391 | } |
4392 | __set_current_state(state); | |
1da177e4 LT |
4393 | spin_unlock_irq(&x->wait.lock); |
4394 | timeout = schedule_timeout(timeout); | |
4395 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4396 | } while (!x->done && timeout); |
1da177e4 | 4397 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4398 | if (!x->done) |
4399 | return timeout; | |
1da177e4 LT |
4400 | } |
4401 | x->done--; | |
ea71a546 | 4402 | return timeout ?: 1; |
1da177e4 | 4403 | } |
1da177e4 | 4404 | |
8cbbe86d AK |
4405 | static long __sched |
4406 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4407 | { |
1da177e4 LT |
4408 | might_sleep(); |
4409 | ||
4410 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4411 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4412 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4413 | return timeout; |
4414 | } | |
1da177e4 | 4415 | |
65eb3dc6 KD |
4416 | /** |
4417 | * wait_for_completion: - waits for completion of a task | |
4418 | * @x: holds the state of this particular completion | |
4419 | * | |
4420 | * This waits to be signaled for completion of a specific task. It is NOT | |
4421 | * interruptible and there is no timeout. | |
4422 | * | |
4423 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4424 | * and interrupt capability. Also see complete(). | |
4425 | */ | |
b15136e9 | 4426 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4427 | { |
4428 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4429 | } |
8cbbe86d | 4430 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4431 | |
65eb3dc6 KD |
4432 | /** |
4433 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4434 | * @x: holds the state of this particular completion | |
4435 | * @timeout: timeout value in jiffies | |
4436 | * | |
4437 | * This waits for either a completion of a specific task to be signaled or for a | |
4438 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4439 | * interruptible. | |
4440 | */ | |
b15136e9 | 4441 | unsigned long __sched |
8cbbe86d | 4442 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4443 | { |
8cbbe86d | 4444 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4445 | } |
8cbbe86d | 4446 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4447 | |
65eb3dc6 KD |
4448 | /** |
4449 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4450 | * @x: holds the state of this particular completion | |
4451 | * | |
4452 | * This waits for completion of a specific task to be signaled. It is | |
4453 | * interruptible. | |
4454 | */ | |
8cbbe86d | 4455 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4456 | { |
51e97990 AK |
4457 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4458 | if (t == -ERESTARTSYS) | |
4459 | return t; | |
4460 | return 0; | |
0fec171c | 4461 | } |
8cbbe86d | 4462 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4463 | |
65eb3dc6 KD |
4464 | /** |
4465 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4466 | * @x: holds the state of this particular completion | |
4467 | * @timeout: timeout value in jiffies | |
4468 | * | |
4469 | * This waits for either a completion of a specific task to be signaled or for a | |
4470 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4471 | */ | |
b15136e9 | 4472 | unsigned long __sched |
8cbbe86d AK |
4473 | wait_for_completion_interruptible_timeout(struct completion *x, |
4474 | unsigned long timeout) | |
0fec171c | 4475 | { |
8cbbe86d | 4476 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4477 | } |
8cbbe86d | 4478 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4479 | |
65eb3dc6 KD |
4480 | /** |
4481 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4482 | * @x: holds the state of this particular completion | |
4483 | * | |
4484 | * This waits to be signaled for completion of a specific task. It can be | |
4485 | * interrupted by a kill signal. | |
4486 | */ | |
009e577e MW |
4487 | int __sched wait_for_completion_killable(struct completion *x) |
4488 | { | |
4489 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4490 | if (t == -ERESTARTSYS) | |
4491 | return t; | |
4492 | return 0; | |
4493 | } | |
4494 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4495 | ||
0aa12fb4 SW |
4496 | /** |
4497 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4498 | * @x: holds the state of this particular completion | |
4499 | * @timeout: timeout value in jiffies | |
4500 | * | |
4501 | * This waits for either a completion of a specific task to be | |
4502 | * signaled or for a specified timeout to expire. It can be | |
4503 | * interrupted by a kill signal. The timeout is in jiffies. | |
4504 | */ | |
4505 | unsigned long __sched | |
4506 | wait_for_completion_killable_timeout(struct completion *x, | |
4507 | unsigned long timeout) | |
4508 | { | |
4509 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4510 | } | |
4511 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4512 | ||
be4de352 DC |
4513 | /** |
4514 | * try_wait_for_completion - try to decrement a completion without blocking | |
4515 | * @x: completion structure | |
4516 | * | |
4517 | * Returns: 0 if a decrement cannot be done without blocking | |
4518 | * 1 if a decrement succeeded. | |
4519 | * | |
4520 | * If a completion is being used as a counting completion, | |
4521 | * attempt to decrement the counter without blocking. This | |
4522 | * enables us to avoid waiting if the resource the completion | |
4523 | * is protecting is not available. | |
4524 | */ | |
4525 | bool try_wait_for_completion(struct completion *x) | |
4526 | { | |
7539a3b3 | 4527 | unsigned long flags; |
be4de352 DC |
4528 | int ret = 1; |
4529 | ||
7539a3b3 | 4530 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4531 | if (!x->done) |
4532 | ret = 0; | |
4533 | else | |
4534 | x->done--; | |
7539a3b3 | 4535 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4536 | return ret; |
4537 | } | |
4538 | EXPORT_SYMBOL(try_wait_for_completion); | |
4539 | ||
4540 | /** | |
4541 | * completion_done - Test to see if a completion has any waiters | |
4542 | * @x: completion structure | |
4543 | * | |
4544 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4545 | * 1 if there are no waiters. | |
4546 | * | |
4547 | */ | |
4548 | bool completion_done(struct completion *x) | |
4549 | { | |
7539a3b3 | 4550 | unsigned long flags; |
be4de352 DC |
4551 | int ret = 1; |
4552 | ||
7539a3b3 | 4553 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4554 | if (!x->done) |
4555 | ret = 0; | |
7539a3b3 | 4556 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4557 | return ret; |
4558 | } | |
4559 | EXPORT_SYMBOL(completion_done); | |
4560 | ||
8cbbe86d AK |
4561 | static long __sched |
4562 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4563 | { |
0fec171c IM |
4564 | unsigned long flags; |
4565 | wait_queue_t wait; | |
4566 | ||
4567 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4568 | |
8cbbe86d | 4569 | __set_current_state(state); |
1da177e4 | 4570 | |
8cbbe86d AK |
4571 | spin_lock_irqsave(&q->lock, flags); |
4572 | __add_wait_queue(q, &wait); | |
4573 | spin_unlock(&q->lock); | |
4574 | timeout = schedule_timeout(timeout); | |
4575 | spin_lock_irq(&q->lock); | |
4576 | __remove_wait_queue(q, &wait); | |
4577 | spin_unlock_irqrestore(&q->lock, flags); | |
4578 | ||
4579 | return timeout; | |
4580 | } | |
4581 | ||
4582 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4583 | { | |
4584 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4585 | } |
1da177e4 LT |
4586 | EXPORT_SYMBOL(interruptible_sleep_on); |
4587 | ||
0fec171c | 4588 | long __sched |
95cdf3b7 | 4589 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4590 | { |
8cbbe86d | 4591 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4592 | } |
1da177e4 LT |
4593 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4594 | ||
0fec171c | 4595 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4596 | { |
8cbbe86d | 4597 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4598 | } |
1da177e4 LT |
4599 | EXPORT_SYMBOL(sleep_on); |
4600 | ||
0fec171c | 4601 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4602 | { |
8cbbe86d | 4603 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4604 | } |
1da177e4 LT |
4605 | EXPORT_SYMBOL(sleep_on_timeout); |
4606 | ||
b29739f9 IM |
4607 | #ifdef CONFIG_RT_MUTEXES |
4608 | ||
4609 | /* | |
4610 | * rt_mutex_setprio - set the current priority of a task | |
4611 | * @p: task | |
4612 | * @prio: prio value (kernel-internal form) | |
4613 | * | |
4614 | * This function changes the 'effective' priority of a task. It does | |
4615 | * not touch ->normal_prio like __setscheduler(). | |
4616 | * | |
4617 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4618 | */ | |
36c8b586 | 4619 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4620 | { |
4621 | unsigned long flags; | |
83b699ed | 4622 | int oldprio, on_rq, running; |
70b97a7f | 4623 | struct rq *rq; |
83ab0aa0 | 4624 | const struct sched_class *prev_class; |
b29739f9 IM |
4625 | |
4626 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4627 | ||
4628 | rq = task_rq_lock(p, &flags); | |
4629 | ||
a8027073 | 4630 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4631 | oldprio = p->prio; |
83ab0aa0 | 4632 | prev_class = p->sched_class; |
dd41f596 | 4633 | on_rq = p->se.on_rq; |
051a1d1a | 4634 | running = task_current(rq, p); |
0e1f3483 | 4635 | if (on_rq) |
69be72c1 | 4636 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4637 | if (running) |
4638 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4639 | |
4640 | if (rt_prio(prio)) | |
4641 | p->sched_class = &rt_sched_class; | |
4642 | else | |
4643 | p->sched_class = &fair_sched_class; | |
4644 | ||
b29739f9 IM |
4645 | p->prio = prio; |
4646 | ||
0e1f3483 HS |
4647 | if (running) |
4648 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4649 | if (on_rq) { |
371fd7e7 | 4650 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4651 | |
4652 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4653 | } |
4654 | task_rq_unlock(rq, &flags); | |
4655 | } | |
4656 | ||
4657 | #endif | |
4658 | ||
36c8b586 | 4659 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4660 | { |
dd41f596 | 4661 | int old_prio, delta, on_rq; |
1da177e4 | 4662 | unsigned long flags; |
70b97a7f | 4663 | struct rq *rq; |
1da177e4 LT |
4664 | |
4665 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4666 | return; | |
4667 | /* | |
4668 | * We have to be careful, if called from sys_setpriority(), | |
4669 | * the task might be in the middle of scheduling on another CPU. | |
4670 | */ | |
4671 | rq = task_rq_lock(p, &flags); | |
4672 | /* | |
4673 | * The RT priorities are set via sched_setscheduler(), but we still | |
4674 | * allow the 'normal' nice value to be set - but as expected | |
4675 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4676 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4677 | */ |
e05606d3 | 4678 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4679 | p->static_prio = NICE_TO_PRIO(nice); |
4680 | goto out_unlock; | |
4681 | } | |
dd41f596 | 4682 | on_rq = p->se.on_rq; |
c09595f6 | 4683 | if (on_rq) |
69be72c1 | 4684 | dequeue_task(rq, p, 0); |
1da177e4 | 4685 | |
1da177e4 | 4686 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4687 | set_load_weight(p); |
b29739f9 IM |
4688 | old_prio = p->prio; |
4689 | p->prio = effective_prio(p); | |
4690 | delta = p->prio - old_prio; | |
1da177e4 | 4691 | |
dd41f596 | 4692 | if (on_rq) { |
371fd7e7 | 4693 | enqueue_task(rq, p, 0); |
1da177e4 | 4694 | /* |
d5f9f942 AM |
4695 | * If the task increased its priority or is running and |
4696 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4697 | */ |
d5f9f942 | 4698 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4699 | resched_task(rq->curr); |
4700 | } | |
4701 | out_unlock: | |
4702 | task_rq_unlock(rq, &flags); | |
4703 | } | |
1da177e4 LT |
4704 | EXPORT_SYMBOL(set_user_nice); |
4705 | ||
e43379f1 MM |
4706 | /* |
4707 | * can_nice - check if a task can reduce its nice value | |
4708 | * @p: task | |
4709 | * @nice: nice value | |
4710 | */ | |
36c8b586 | 4711 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4712 | { |
024f4747 MM |
4713 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4714 | int nice_rlim = 20 - nice; | |
48f24c4d | 4715 | |
78d7d407 | 4716 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4717 | capable(CAP_SYS_NICE)); |
4718 | } | |
4719 | ||
1da177e4 LT |
4720 | #ifdef __ARCH_WANT_SYS_NICE |
4721 | ||
4722 | /* | |
4723 | * sys_nice - change the priority of the current process. | |
4724 | * @increment: priority increment | |
4725 | * | |
4726 | * sys_setpriority is a more generic, but much slower function that | |
4727 | * does similar things. | |
4728 | */ | |
5add95d4 | 4729 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4730 | { |
48f24c4d | 4731 | long nice, retval; |
1da177e4 LT |
4732 | |
4733 | /* | |
4734 | * Setpriority might change our priority at the same moment. | |
4735 | * We don't have to worry. Conceptually one call occurs first | |
4736 | * and we have a single winner. | |
4737 | */ | |
e43379f1 MM |
4738 | if (increment < -40) |
4739 | increment = -40; | |
1da177e4 LT |
4740 | if (increment > 40) |
4741 | increment = 40; | |
4742 | ||
2b8f836f | 4743 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4744 | if (nice < -20) |
4745 | nice = -20; | |
4746 | if (nice > 19) | |
4747 | nice = 19; | |
4748 | ||
e43379f1 MM |
4749 | if (increment < 0 && !can_nice(current, nice)) |
4750 | return -EPERM; | |
4751 | ||
1da177e4 LT |
4752 | retval = security_task_setnice(current, nice); |
4753 | if (retval) | |
4754 | return retval; | |
4755 | ||
4756 | set_user_nice(current, nice); | |
4757 | return 0; | |
4758 | } | |
4759 | ||
4760 | #endif | |
4761 | ||
4762 | /** | |
4763 | * task_prio - return the priority value of a given task. | |
4764 | * @p: the task in question. | |
4765 | * | |
4766 | * This is the priority value as seen by users in /proc. | |
4767 | * RT tasks are offset by -200. Normal tasks are centered | |
4768 | * around 0, value goes from -16 to +15. | |
4769 | */ | |
36c8b586 | 4770 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4771 | { |
4772 | return p->prio - MAX_RT_PRIO; | |
4773 | } | |
4774 | ||
4775 | /** | |
4776 | * task_nice - return the nice value of a given task. | |
4777 | * @p: the task in question. | |
4778 | */ | |
36c8b586 | 4779 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4780 | { |
4781 | return TASK_NICE(p); | |
4782 | } | |
150d8bed | 4783 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4784 | |
4785 | /** | |
4786 | * idle_cpu - is a given cpu idle currently? | |
4787 | * @cpu: the processor in question. | |
4788 | */ | |
4789 | int idle_cpu(int cpu) | |
4790 | { | |
4791 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4792 | } | |
4793 | ||
1da177e4 LT |
4794 | /** |
4795 | * idle_task - return the idle task for a given cpu. | |
4796 | * @cpu: the processor in question. | |
4797 | */ | |
36c8b586 | 4798 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4799 | { |
4800 | return cpu_rq(cpu)->idle; | |
4801 | } | |
4802 | ||
4803 | /** | |
4804 | * find_process_by_pid - find a process with a matching PID value. | |
4805 | * @pid: the pid in question. | |
4806 | */ | |
a9957449 | 4807 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4808 | { |
228ebcbe | 4809 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4810 | } |
4811 | ||
4812 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4813 | static void |
4814 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4815 | { |
dd41f596 | 4816 | BUG_ON(p->se.on_rq); |
48f24c4d | 4817 | |
1da177e4 LT |
4818 | p->policy = policy; |
4819 | p->rt_priority = prio; | |
b29739f9 IM |
4820 | p->normal_prio = normal_prio(p); |
4821 | /* we are holding p->pi_lock already */ | |
4822 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4823 | if (rt_prio(p->prio)) |
4824 | p->sched_class = &rt_sched_class; | |
4825 | else | |
4826 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4827 | set_load_weight(p); |
1da177e4 LT |
4828 | } |
4829 | ||
c69e8d9c DH |
4830 | /* |
4831 | * check the target process has a UID that matches the current process's | |
4832 | */ | |
4833 | static bool check_same_owner(struct task_struct *p) | |
4834 | { | |
4835 | const struct cred *cred = current_cred(), *pcred; | |
4836 | bool match; | |
4837 | ||
4838 | rcu_read_lock(); | |
4839 | pcred = __task_cred(p); | |
4840 | match = (cred->euid == pcred->euid || | |
4841 | cred->euid == pcred->uid); | |
4842 | rcu_read_unlock(); | |
4843 | return match; | |
4844 | } | |
4845 | ||
961ccddd RR |
4846 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4847 | struct sched_param *param, bool user) | |
1da177e4 | 4848 | { |
83b699ed | 4849 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4850 | unsigned long flags; |
83ab0aa0 | 4851 | const struct sched_class *prev_class; |
70b97a7f | 4852 | struct rq *rq; |
ca94c442 | 4853 | int reset_on_fork; |
1da177e4 | 4854 | |
66e5393a SR |
4855 | /* may grab non-irq protected spin_locks */ |
4856 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4857 | recheck: |
4858 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4859 | if (policy < 0) { |
4860 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4861 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4862 | } else { |
4863 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4864 | policy &= ~SCHED_RESET_ON_FORK; | |
4865 | ||
4866 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4867 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4868 | policy != SCHED_IDLE) | |
4869 | return -EINVAL; | |
4870 | } | |
4871 | ||
1da177e4 LT |
4872 | /* |
4873 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4874 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4875 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4876 | */ |
4877 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4878 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4879 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4880 | return -EINVAL; |
e05606d3 | 4881 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4882 | return -EINVAL; |
4883 | ||
37e4ab3f OC |
4884 | /* |
4885 | * Allow unprivileged RT tasks to decrease priority: | |
4886 | */ | |
961ccddd | 4887 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4888 | if (rt_policy(policy)) { |
a44702e8 ON |
4889 | unsigned long rlim_rtprio = |
4890 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4891 | |
4892 | /* can't set/change the rt policy */ | |
4893 | if (policy != p->policy && !rlim_rtprio) | |
4894 | return -EPERM; | |
4895 | ||
4896 | /* can't increase priority */ | |
4897 | if (param->sched_priority > p->rt_priority && | |
4898 | param->sched_priority > rlim_rtprio) | |
4899 | return -EPERM; | |
4900 | } | |
dd41f596 IM |
4901 | /* |
4902 | * Like positive nice levels, dont allow tasks to | |
4903 | * move out of SCHED_IDLE either: | |
4904 | */ | |
4905 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4906 | return -EPERM; | |
5fe1d75f | 4907 | |
37e4ab3f | 4908 | /* can't change other user's priorities */ |
c69e8d9c | 4909 | if (!check_same_owner(p)) |
37e4ab3f | 4910 | return -EPERM; |
ca94c442 LP |
4911 | |
4912 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4913 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4914 | return -EPERM; | |
37e4ab3f | 4915 | } |
1da177e4 | 4916 | |
725aad24 | 4917 | if (user) { |
b0ae1981 | 4918 | retval = security_task_setscheduler(p); |
725aad24 JF |
4919 | if (retval) |
4920 | return retval; | |
4921 | } | |
4922 | ||
b29739f9 IM |
4923 | /* |
4924 | * make sure no PI-waiters arrive (or leave) while we are | |
4925 | * changing the priority of the task: | |
4926 | */ | |
1d615482 | 4927 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4928 | /* |
4929 | * To be able to change p->policy safely, the apropriate | |
4930 | * runqueue lock must be held. | |
4931 | */ | |
b29739f9 | 4932 | rq = __task_rq_lock(p); |
dc61b1d6 | 4933 | |
34f971f6 PZ |
4934 | /* |
4935 | * Changing the policy of the stop threads its a very bad idea | |
4936 | */ | |
4937 | if (p == rq->stop) { | |
4938 | __task_rq_unlock(rq); | |
4939 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4940 | return -EINVAL; | |
4941 | } | |
4942 | ||
dc61b1d6 PZ |
4943 | #ifdef CONFIG_RT_GROUP_SCHED |
4944 | if (user) { | |
4945 | /* | |
4946 | * Do not allow realtime tasks into groups that have no runtime | |
4947 | * assigned. | |
4948 | */ | |
4949 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
4950 | task_group(p)->rt_bandwidth.rt_runtime == 0) { | |
4951 | __task_rq_unlock(rq); | |
4952 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4953 | return -EPERM; | |
4954 | } | |
4955 | } | |
4956 | #endif | |
4957 | ||
1da177e4 LT |
4958 | /* recheck policy now with rq lock held */ |
4959 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4960 | policy = oldpolicy = -1; | |
b29739f9 | 4961 | __task_rq_unlock(rq); |
1d615482 | 4962 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4963 | goto recheck; |
4964 | } | |
dd41f596 | 4965 | on_rq = p->se.on_rq; |
051a1d1a | 4966 | running = task_current(rq, p); |
0e1f3483 | 4967 | if (on_rq) |
2e1cb74a | 4968 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4969 | if (running) |
4970 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4971 | |
ca94c442 LP |
4972 | p->sched_reset_on_fork = reset_on_fork; |
4973 | ||
1da177e4 | 4974 | oldprio = p->prio; |
83ab0aa0 | 4975 | prev_class = p->sched_class; |
dd41f596 | 4976 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4977 | |
0e1f3483 HS |
4978 | if (running) |
4979 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4980 | if (on_rq) { |
4981 | activate_task(rq, p, 0); | |
cb469845 SR |
4982 | |
4983 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4984 | } |
b29739f9 | 4985 | __task_rq_unlock(rq); |
1d615482 | 4986 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4987 | |
95e02ca9 TG |
4988 | rt_mutex_adjust_pi(p); |
4989 | ||
1da177e4 LT |
4990 | return 0; |
4991 | } | |
961ccddd RR |
4992 | |
4993 | /** | |
4994 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4995 | * @p: the task in question. | |
4996 | * @policy: new policy. | |
4997 | * @param: structure containing the new RT priority. | |
4998 | * | |
4999 | * NOTE that the task may be already dead. | |
5000 | */ | |
5001 | int sched_setscheduler(struct task_struct *p, int policy, | |
5002 | struct sched_param *param) | |
5003 | { | |
5004 | return __sched_setscheduler(p, policy, param, true); | |
5005 | } | |
1da177e4 LT |
5006 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5007 | ||
961ccddd RR |
5008 | /** |
5009 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5010 | * @p: the task in question. | |
5011 | * @policy: new policy. | |
5012 | * @param: structure containing the new RT priority. | |
5013 | * | |
5014 | * Just like sched_setscheduler, only don't bother checking if the | |
5015 | * current context has permission. For example, this is needed in | |
5016 | * stop_machine(): we create temporary high priority worker threads, | |
5017 | * but our caller might not have that capability. | |
5018 | */ | |
5019 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5020 | struct sched_param *param) | |
5021 | { | |
5022 | return __sched_setscheduler(p, policy, param, false); | |
5023 | } | |
5024 | ||
95cdf3b7 IM |
5025 | static int |
5026 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5027 | { |
1da177e4 LT |
5028 | struct sched_param lparam; |
5029 | struct task_struct *p; | |
36c8b586 | 5030 | int retval; |
1da177e4 LT |
5031 | |
5032 | if (!param || pid < 0) | |
5033 | return -EINVAL; | |
5034 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5035 | return -EFAULT; | |
5fe1d75f ON |
5036 | |
5037 | rcu_read_lock(); | |
5038 | retval = -ESRCH; | |
1da177e4 | 5039 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5040 | if (p != NULL) |
5041 | retval = sched_setscheduler(p, policy, &lparam); | |
5042 | rcu_read_unlock(); | |
36c8b586 | 5043 | |
1da177e4 LT |
5044 | return retval; |
5045 | } | |
5046 | ||
5047 | /** | |
5048 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5049 | * @pid: the pid in question. | |
5050 | * @policy: new policy. | |
5051 | * @param: structure containing the new RT priority. | |
5052 | */ | |
5add95d4 HC |
5053 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5054 | struct sched_param __user *, param) | |
1da177e4 | 5055 | { |
c21761f1 JB |
5056 | /* negative values for policy are not valid */ |
5057 | if (policy < 0) | |
5058 | return -EINVAL; | |
5059 | ||
1da177e4 LT |
5060 | return do_sched_setscheduler(pid, policy, param); |
5061 | } | |
5062 | ||
5063 | /** | |
5064 | * sys_sched_setparam - set/change the RT priority of a thread | |
5065 | * @pid: the pid in question. | |
5066 | * @param: structure containing the new RT priority. | |
5067 | */ | |
5add95d4 | 5068 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5069 | { |
5070 | return do_sched_setscheduler(pid, -1, param); | |
5071 | } | |
5072 | ||
5073 | /** | |
5074 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5075 | * @pid: the pid in question. | |
5076 | */ | |
5add95d4 | 5077 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5078 | { |
36c8b586 | 5079 | struct task_struct *p; |
3a5c359a | 5080 | int retval; |
1da177e4 LT |
5081 | |
5082 | if (pid < 0) | |
3a5c359a | 5083 | return -EINVAL; |
1da177e4 LT |
5084 | |
5085 | retval = -ESRCH; | |
5fe85be0 | 5086 | rcu_read_lock(); |
1da177e4 LT |
5087 | p = find_process_by_pid(pid); |
5088 | if (p) { | |
5089 | retval = security_task_getscheduler(p); | |
5090 | if (!retval) | |
ca94c442 LP |
5091 | retval = p->policy |
5092 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5093 | } |
5fe85be0 | 5094 | rcu_read_unlock(); |
1da177e4 LT |
5095 | return retval; |
5096 | } | |
5097 | ||
5098 | /** | |
ca94c442 | 5099 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5100 | * @pid: the pid in question. |
5101 | * @param: structure containing the RT priority. | |
5102 | */ | |
5add95d4 | 5103 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5104 | { |
5105 | struct sched_param lp; | |
36c8b586 | 5106 | struct task_struct *p; |
3a5c359a | 5107 | int retval; |
1da177e4 LT |
5108 | |
5109 | if (!param || pid < 0) | |
3a5c359a | 5110 | return -EINVAL; |
1da177e4 | 5111 | |
5fe85be0 | 5112 | rcu_read_lock(); |
1da177e4 LT |
5113 | p = find_process_by_pid(pid); |
5114 | retval = -ESRCH; | |
5115 | if (!p) | |
5116 | goto out_unlock; | |
5117 | ||
5118 | retval = security_task_getscheduler(p); | |
5119 | if (retval) | |
5120 | goto out_unlock; | |
5121 | ||
5122 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5123 | rcu_read_unlock(); |
1da177e4 LT |
5124 | |
5125 | /* | |
5126 | * This one might sleep, we cannot do it with a spinlock held ... | |
5127 | */ | |
5128 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5129 | ||
1da177e4 LT |
5130 | return retval; |
5131 | ||
5132 | out_unlock: | |
5fe85be0 | 5133 | rcu_read_unlock(); |
1da177e4 LT |
5134 | return retval; |
5135 | } | |
5136 | ||
96f874e2 | 5137 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5138 | { |
5a16f3d3 | 5139 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5140 | struct task_struct *p; |
5141 | int retval; | |
1da177e4 | 5142 | |
95402b38 | 5143 | get_online_cpus(); |
23f5d142 | 5144 | rcu_read_lock(); |
1da177e4 LT |
5145 | |
5146 | p = find_process_by_pid(pid); | |
5147 | if (!p) { | |
23f5d142 | 5148 | rcu_read_unlock(); |
95402b38 | 5149 | put_online_cpus(); |
1da177e4 LT |
5150 | return -ESRCH; |
5151 | } | |
5152 | ||
23f5d142 | 5153 | /* Prevent p going away */ |
1da177e4 | 5154 | get_task_struct(p); |
23f5d142 | 5155 | rcu_read_unlock(); |
1da177e4 | 5156 | |
5a16f3d3 RR |
5157 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5158 | retval = -ENOMEM; | |
5159 | goto out_put_task; | |
5160 | } | |
5161 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5162 | retval = -ENOMEM; | |
5163 | goto out_free_cpus_allowed; | |
5164 | } | |
1da177e4 | 5165 | retval = -EPERM; |
c69e8d9c | 5166 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
5167 | goto out_unlock; |
5168 | ||
b0ae1981 | 5169 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5170 | if (retval) |
5171 | goto out_unlock; | |
5172 | ||
5a16f3d3 RR |
5173 | cpuset_cpus_allowed(p, cpus_allowed); |
5174 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5175 | again: |
5a16f3d3 | 5176 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5177 | |
8707d8b8 | 5178 | if (!retval) { |
5a16f3d3 RR |
5179 | cpuset_cpus_allowed(p, cpus_allowed); |
5180 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5181 | /* |
5182 | * We must have raced with a concurrent cpuset | |
5183 | * update. Just reset the cpus_allowed to the | |
5184 | * cpuset's cpus_allowed | |
5185 | */ | |
5a16f3d3 | 5186 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5187 | goto again; |
5188 | } | |
5189 | } | |
1da177e4 | 5190 | out_unlock: |
5a16f3d3 RR |
5191 | free_cpumask_var(new_mask); |
5192 | out_free_cpus_allowed: | |
5193 | free_cpumask_var(cpus_allowed); | |
5194 | out_put_task: | |
1da177e4 | 5195 | put_task_struct(p); |
95402b38 | 5196 | put_online_cpus(); |
1da177e4 LT |
5197 | return retval; |
5198 | } | |
5199 | ||
5200 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5201 | struct cpumask *new_mask) |
1da177e4 | 5202 | { |
96f874e2 RR |
5203 | if (len < cpumask_size()) |
5204 | cpumask_clear(new_mask); | |
5205 | else if (len > cpumask_size()) | |
5206 | len = cpumask_size(); | |
5207 | ||
1da177e4 LT |
5208 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5209 | } | |
5210 | ||
5211 | /** | |
5212 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5213 | * @pid: pid of the process | |
5214 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5215 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5216 | */ | |
5add95d4 HC |
5217 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5218 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5219 | { |
5a16f3d3 | 5220 | cpumask_var_t new_mask; |
1da177e4 LT |
5221 | int retval; |
5222 | ||
5a16f3d3 RR |
5223 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5224 | return -ENOMEM; | |
1da177e4 | 5225 | |
5a16f3d3 RR |
5226 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5227 | if (retval == 0) | |
5228 | retval = sched_setaffinity(pid, new_mask); | |
5229 | free_cpumask_var(new_mask); | |
5230 | return retval; | |
1da177e4 LT |
5231 | } |
5232 | ||
96f874e2 | 5233 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5234 | { |
36c8b586 | 5235 | struct task_struct *p; |
31605683 TG |
5236 | unsigned long flags; |
5237 | struct rq *rq; | |
1da177e4 | 5238 | int retval; |
1da177e4 | 5239 | |
95402b38 | 5240 | get_online_cpus(); |
23f5d142 | 5241 | rcu_read_lock(); |
1da177e4 LT |
5242 | |
5243 | retval = -ESRCH; | |
5244 | p = find_process_by_pid(pid); | |
5245 | if (!p) | |
5246 | goto out_unlock; | |
5247 | ||
e7834f8f DQ |
5248 | retval = security_task_getscheduler(p); |
5249 | if (retval) | |
5250 | goto out_unlock; | |
5251 | ||
31605683 | 5252 | rq = task_rq_lock(p, &flags); |
96f874e2 | 5253 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 5254 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5255 | |
5256 | out_unlock: | |
23f5d142 | 5257 | rcu_read_unlock(); |
95402b38 | 5258 | put_online_cpus(); |
1da177e4 | 5259 | |
9531b62f | 5260 | return retval; |
1da177e4 LT |
5261 | } |
5262 | ||
5263 | /** | |
5264 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5265 | * @pid: pid of the process | |
5266 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5267 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5268 | */ | |
5add95d4 HC |
5269 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5270 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5271 | { |
5272 | int ret; | |
f17c8607 | 5273 | cpumask_var_t mask; |
1da177e4 | 5274 | |
84fba5ec | 5275 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5276 | return -EINVAL; |
5277 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5278 | return -EINVAL; |
5279 | ||
f17c8607 RR |
5280 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5281 | return -ENOMEM; | |
1da177e4 | 5282 | |
f17c8607 RR |
5283 | ret = sched_getaffinity(pid, mask); |
5284 | if (ret == 0) { | |
8bc037fb | 5285 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5286 | |
5287 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5288 | ret = -EFAULT; |
5289 | else | |
cd3d8031 | 5290 | ret = retlen; |
f17c8607 RR |
5291 | } |
5292 | free_cpumask_var(mask); | |
1da177e4 | 5293 | |
f17c8607 | 5294 | return ret; |
1da177e4 LT |
5295 | } |
5296 | ||
5297 | /** | |
5298 | * sys_sched_yield - yield the current processor to other threads. | |
5299 | * | |
dd41f596 IM |
5300 | * This function yields the current CPU to other tasks. If there are no |
5301 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5302 | */ |
5add95d4 | 5303 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5304 | { |
70b97a7f | 5305 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5306 | |
2d72376b | 5307 | schedstat_inc(rq, yld_count); |
4530d7ab | 5308 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5309 | |
5310 | /* | |
5311 | * Since we are going to call schedule() anyway, there's | |
5312 | * no need to preempt or enable interrupts: | |
5313 | */ | |
5314 | __release(rq->lock); | |
8a25d5de | 5315 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5316 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5317 | preempt_enable_no_resched(); |
5318 | ||
5319 | schedule(); | |
5320 | ||
5321 | return 0; | |
5322 | } | |
5323 | ||
d86ee480 PZ |
5324 | static inline int should_resched(void) |
5325 | { | |
5326 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5327 | } | |
5328 | ||
e7b38404 | 5329 | static void __cond_resched(void) |
1da177e4 | 5330 | { |
e7aaaa69 FW |
5331 | add_preempt_count(PREEMPT_ACTIVE); |
5332 | schedule(); | |
5333 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5334 | } |
5335 | ||
02b67cc3 | 5336 | int __sched _cond_resched(void) |
1da177e4 | 5337 | { |
d86ee480 | 5338 | if (should_resched()) { |
1da177e4 LT |
5339 | __cond_resched(); |
5340 | return 1; | |
5341 | } | |
5342 | return 0; | |
5343 | } | |
02b67cc3 | 5344 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5345 | |
5346 | /* | |
613afbf8 | 5347 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5348 | * call schedule, and on return reacquire the lock. |
5349 | * | |
41a2d6cf | 5350 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5351 | * operations here to prevent schedule() from being called twice (once via |
5352 | * spin_unlock(), once by hand). | |
5353 | */ | |
613afbf8 | 5354 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5355 | { |
d86ee480 | 5356 | int resched = should_resched(); |
6df3cecb JK |
5357 | int ret = 0; |
5358 | ||
f607c668 PZ |
5359 | lockdep_assert_held(lock); |
5360 | ||
95c354fe | 5361 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5362 | spin_unlock(lock); |
d86ee480 | 5363 | if (resched) |
95c354fe NP |
5364 | __cond_resched(); |
5365 | else | |
5366 | cpu_relax(); | |
6df3cecb | 5367 | ret = 1; |
1da177e4 | 5368 | spin_lock(lock); |
1da177e4 | 5369 | } |
6df3cecb | 5370 | return ret; |
1da177e4 | 5371 | } |
613afbf8 | 5372 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5373 | |
613afbf8 | 5374 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5375 | { |
5376 | BUG_ON(!in_softirq()); | |
5377 | ||
d86ee480 | 5378 | if (should_resched()) { |
98d82567 | 5379 | local_bh_enable(); |
1da177e4 LT |
5380 | __cond_resched(); |
5381 | local_bh_disable(); | |
5382 | return 1; | |
5383 | } | |
5384 | return 0; | |
5385 | } | |
613afbf8 | 5386 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5387 | |
1da177e4 LT |
5388 | /** |
5389 | * yield - yield the current processor to other threads. | |
5390 | * | |
72fd4a35 | 5391 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5392 | * thread runnable and calls sys_sched_yield(). |
5393 | */ | |
5394 | void __sched yield(void) | |
5395 | { | |
5396 | set_current_state(TASK_RUNNING); | |
5397 | sys_sched_yield(); | |
5398 | } | |
1da177e4 LT |
5399 | EXPORT_SYMBOL(yield); |
5400 | ||
5401 | /* | |
41a2d6cf | 5402 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5403 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5404 | */ |
5405 | void __sched io_schedule(void) | |
5406 | { | |
54d35f29 | 5407 | struct rq *rq = raw_rq(); |
1da177e4 | 5408 | |
0ff92245 | 5409 | delayacct_blkio_start(); |
1da177e4 | 5410 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5411 | current->in_iowait = 1; |
1da177e4 | 5412 | schedule(); |
8f0dfc34 | 5413 | current->in_iowait = 0; |
1da177e4 | 5414 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5415 | delayacct_blkio_end(); |
1da177e4 | 5416 | } |
1da177e4 LT |
5417 | EXPORT_SYMBOL(io_schedule); |
5418 | ||
5419 | long __sched io_schedule_timeout(long timeout) | |
5420 | { | |
54d35f29 | 5421 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5422 | long ret; |
5423 | ||
0ff92245 | 5424 | delayacct_blkio_start(); |
1da177e4 | 5425 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5426 | current->in_iowait = 1; |
1da177e4 | 5427 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5428 | current->in_iowait = 0; |
1da177e4 | 5429 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5430 | delayacct_blkio_end(); |
1da177e4 LT |
5431 | return ret; |
5432 | } | |
5433 | ||
5434 | /** | |
5435 | * sys_sched_get_priority_max - return maximum RT priority. | |
5436 | * @policy: scheduling class. | |
5437 | * | |
5438 | * this syscall returns the maximum rt_priority that can be used | |
5439 | * by a given scheduling class. | |
5440 | */ | |
5add95d4 | 5441 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5442 | { |
5443 | int ret = -EINVAL; | |
5444 | ||
5445 | switch (policy) { | |
5446 | case SCHED_FIFO: | |
5447 | case SCHED_RR: | |
5448 | ret = MAX_USER_RT_PRIO-1; | |
5449 | break; | |
5450 | case SCHED_NORMAL: | |
b0a9499c | 5451 | case SCHED_BATCH: |
dd41f596 | 5452 | case SCHED_IDLE: |
1da177e4 LT |
5453 | ret = 0; |
5454 | break; | |
5455 | } | |
5456 | return ret; | |
5457 | } | |
5458 | ||
5459 | /** | |
5460 | * sys_sched_get_priority_min - return minimum RT priority. | |
5461 | * @policy: scheduling class. | |
5462 | * | |
5463 | * this syscall returns the minimum rt_priority that can be used | |
5464 | * by a given scheduling class. | |
5465 | */ | |
5add95d4 | 5466 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5467 | { |
5468 | int ret = -EINVAL; | |
5469 | ||
5470 | switch (policy) { | |
5471 | case SCHED_FIFO: | |
5472 | case SCHED_RR: | |
5473 | ret = 1; | |
5474 | break; | |
5475 | case SCHED_NORMAL: | |
b0a9499c | 5476 | case SCHED_BATCH: |
dd41f596 | 5477 | case SCHED_IDLE: |
1da177e4 LT |
5478 | ret = 0; |
5479 | } | |
5480 | return ret; | |
5481 | } | |
5482 | ||
5483 | /** | |
5484 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5485 | * @pid: pid of the process. | |
5486 | * @interval: userspace pointer to the timeslice value. | |
5487 | * | |
5488 | * this syscall writes the default timeslice value of a given process | |
5489 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5490 | */ | |
17da2bd9 | 5491 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5492 | struct timespec __user *, interval) |
1da177e4 | 5493 | { |
36c8b586 | 5494 | struct task_struct *p; |
a4ec24b4 | 5495 | unsigned int time_slice; |
dba091b9 TG |
5496 | unsigned long flags; |
5497 | struct rq *rq; | |
3a5c359a | 5498 | int retval; |
1da177e4 | 5499 | struct timespec t; |
1da177e4 LT |
5500 | |
5501 | if (pid < 0) | |
3a5c359a | 5502 | return -EINVAL; |
1da177e4 LT |
5503 | |
5504 | retval = -ESRCH; | |
1a551ae7 | 5505 | rcu_read_lock(); |
1da177e4 LT |
5506 | p = find_process_by_pid(pid); |
5507 | if (!p) | |
5508 | goto out_unlock; | |
5509 | ||
5510 | retval = security_task_getscheduler(p); | |
5511 | if (retval) | |
5512 | goto out_unlock; | |
5513 | ||
dba091b9 TG |
5514 | rq = task_rq_lock(p, &flags); |
5515 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5516 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5517 | |
1a551ae7 | 5518 | rcu_read_unlock(); |
a4ec24b4 | 5519 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5520 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5521 | return retval; |
3a5c359a | 5522 | |
1da177e4 | 5523 | out_unlock: |
1a551ae7 | 5524 | rcu_read_unlock(); |
1da177e4 LT |
5525 | return retval; |
5526 | } | |
5527 | ||
7c731e0a | 5528 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5529 | |
82a1fcb9 | 5530 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5531 | { |
1da177e4 | 5532 | unsigned long free = 0; |
36c8b586 | 5533 | unsigned state; |
1da177e4 | 5534 | |
1da177e4 | 5535 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5536 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5537 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5538 | #if BITS_PER_LONG == 32 |
1da177e4 | 5539 | if (state == TASK_RUNNING) |
3df0fc5b | 5540 | printk(KERN_CONT " running "); |
1da177e4 | 5541 | else |
3df0fc5b | 5542 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5543 | #else |
5544 | if (state == TASK_RUNNING) | |
3df0fc5b | 5545 | printk(KERN_CONT " running task "); |
1da177e4 | 5546 | else |
3df0fc5b | 5547 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5548 | #endif |
5549 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5550 | free = stack_not_used(p); |
1da177e4 | 5551 | #endif |
3df0fc5b | 5552 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5553 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5554 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5555 | |
5fb5e6de | 5556 | show_stack(p, NULL); |
1da177e4 LT |
5557 | } |
5558 | ||
e59e2ae2 | 5559 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5560 | { |
36c8b586 | 5561 | struct task_struct *g, *p; |
1da177e4 | 5562 | |
4bd77321 | 5563 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5564 | printk(KERN_INFO |
5565 | " task PC stack pid father\n"); | |
1da177e4 | 5566 | #else |
3df0fc5b PZ |
5567 | printk(KERN_INFO |
5568 | " task PC stack pid father\n"); | |
1da177e4 LT |
5569 | #endif |
5570 | read_lock(&tasklist_lock); | |
5571 | do_each_thread(g, p) { | |
5572 | /* | |
5573 | * reset the NMI-timeout, listing all files on a slow | |
5574 | * console might take alot of time: | |
5575 | */ | |
5576 | touch_nmi_watchdog(); | |
39bc89fd | 5577 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5578 | sched_show_task(p); |
1da177e4 LT |
5579 | } while_each_thread(g, p); |
5580 | ||
04c9167f JF |
5581 | touch_all_softlockup_watchdogs(); |
5582 | ||
dd41f596 IM |
5583 | #ifdef CONFIG_SCHED_DEBUG |
5584 | sysrq_sched_debug_show(); | |
5585 | #endif | |
1da177e4 | 5586 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5587 | /* |
5588 | * Only show locks if all tasks are dumped: | |
5589 | */ | |
93335a21 | 5590 | if (!state_filter) |
e59e2ae2 | 5591 | debug_show_all_locks(); |
1da177e4 LT |
5592 | } |
5593 | ||
1df21055 IM |
5594 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5595 | { | |
dd41f596 | 5596 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5597 | } |
5598 | ||
f340c0d1 IM |
5599 | /** |
5600 | * init_idle - set up an idle thread for a given CPU | |
5601 | * @idle: task in question | |
5602 | * @cpu: cpu the idle task belongs to | |
5603 | * | |
5604 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5605 | * flag, to make booting more robust. | |
5606 | */ | |
5c1e1767 | 5607 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5608 | { |
70b97a7f | 5609 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5610 | unsigned long flags; |
5611 | ||
05fa785c | 5612 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5613 | |
dd41f596 | 5614 | __sched_fork(idle); |
06b83b5f | 5615 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5616 | idle->se.exec_start = sched_clock(); |
5617 | ||
96f874e2 | 5618 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5619 | /* |
5620 | * We're having a chicken and egg problem, even though we are | |
5621 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5622 | * lockdep check in task_group() will fail. | |
5623 | * | |
5624 | * Similar case to sched_fork(). / Alternatively we could | |
5625 | * use task_rq_lock() here and obtain the other rq->lock. | |
5626 | * | |
5627 | * Silence PROVE_RCU | |
5628 | */ | |
5629 | rcu_read_lock(); | |
dd41f596 | 5630 | __set_task_cpu(idle, cpu); |
6506cf6c | 5631 | rcu_read_unlock(); |
1da177e4 | 5632 | |
1da177e4 | 5633 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5634 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5635 | idle->oncpu = 1; | |
5636 | #endif | |
05fa785c | 5637 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5638 | |
5639 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5640 | #if defined(CONFIG_PREEMPT) |
5641 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5642 | #else | |
a1261f54 | 5643 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5644 | #endif |
dd41f596 IM |
5645 | /* |
5646 | * The idle tasks have their own, simple scheduling class: | |
5647 | */ | |
5648 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5649 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5650 | } |
5651 | ||
5652 | /* | |
5653 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5654 | * indicates which cpus entered this state. This is used | |
5655 | * in the rcu update to wait only for active cpus. For system | |
5656 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5657 | * always be CPU_BITS_NONE. |
1da177e4 | 5658 | */ |
6a7b3dc3 | 5659 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5660 | |
19978ca6 IM |
5661 | /* |
5662 | * Increase the granularity value when there are more CPUs, | |
5663 | * because with more CPUs the 'effective latency' as visible | |
5664 | * to users decreases. But the relationship is not linear, | |
5665 | * so pick a second-best guess by going with the log2 of the | |
5666 | * number of CPUs. | |
5667 | * | |
5668 | * This idea comes from the SD scheduler of Con Kolivas: | |
5669 | */ | |
acb4a848 | 5670 | static int get_update_sysctl_factor(void) |
19978ca6 | 5671 | { |
4ca3ef71 | 5672 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5673 | unsigned int factor; |
5674 | ||
5675 | switch (sysctl_sched_tunable_scaling) { | |
5676 | case SCHED_TUNABLESCALING_NONE: | |
5677 | factor = 1; | |
5678 | break; | |
5679 | case SCHED_TUNABLESCALING_LINEAR: | |
5680 | factor = cpus; | |
5681 | break; | |
5682 | case SCHED_TUNABLESCALING_LOG: | |
5683 | default: | |
5684 | factor = 1 + ilog2(cpus); | |
5685 | break; | |
5686 | } | |
19978ca6 | 5687 | |
acb4a848 CE |
5688 | return factor; |
5689 | } | |
19978ca6 | 5690 | |
acb4a848 CE |
5691 | static void update_sysctl(void) |
5692 | { | |
5693 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5694 | |
0bcdcf28 CE |
5695 | #define SET_SYSCTL(name) \ |
5696 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5697 | SET_SYSCTL(sched_min_granularity); | |
5698 | SET_SYSCTL(sched_latency); | |
5699 | SET_SYSCTL(sched_wakeup_granularity); | |
5700 | SET_SYSCTL(sched_shares_ratelimit); | |
5701 | #undef SET_SYSCTL | |
5702 | } | |
55cd5340 | 5703 | |
0bcdcf28 CE |
5704 | static inline void sched_init_granularity(void) |
5705 | { | |
5706 | update_sysctl(); | |
19978ca6 IM |
5707 | } |
5708 | ||
1da177e4 LT |
5709 | #ifdef CONFIG_SMP |
5710 | /* | |
5711 | * This is how migration works: | |
5712 | * | |
969c7921 TH |
5713 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5714 | * stop_one_cpu(). | |
5715 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5716 | * off the CPU) | |
5717 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5718 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5719 | * it and puts it into the right queue. |
969c7921 TH |
5720 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5721 | * is done. | |
1da177e4 LT |
5722 | */ |
5723 | ||
5724 | /* | |
5725 | * Change a given task's CPU affinity. Migrate the thread to a | |
5726 | * proper CPU and schedule it away if the CPU it's executing on | |
5727 | * is removed from the allowed bitmask. | |
5728 | * | |
5729 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5730 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5731 | * call is not atomic; no spinlocks may be held. |
5732 | */ | |
96f874e2 | 5733 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5734 | { |
5735 | unsigned long flags; | |
70b97a7f | 5736 | struct rq *rq; |
969c7921 | 5737 | unsigned int dest_cpu; |
48f24c4d | 5738 | int ret = 0; |
1da177e4 | 5739 | |
65cc8e48 PZ |
5740 | /* |
5741 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5742 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5743 | */ | |
5744 | again: | |
5745 | while (task_is_waking(p)) | |
5746 | cpu_relax(); | |
1da177e4 | 5747 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5748 | if (task_is_waking(p)) { |
5749 | task_rq_unlock(rq, &flags); | |
5750 | goto again; | |
5751 | } | |
e2912009 | 5752 | |
6ad4c188 | 5753 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5754 | ret = -EINVAL; |
5755 | goto out; | |
5756 | } | |
5757 | ||
9985b0ba | 5758 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5759 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5760 | ret = -EINVAL; |
5761 | goto out; | |
5762 | } | |
5763 | ||
73fe6aae | 5764 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5765 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5766 | else { |
96f874e2 RR |
5767 | cpumask_copy(&p->cpus_allowed, new_mask); |
5768 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5769 | } |
5770 | ||
1da177e4 | 5771 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5772 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5773 | goto out; |
5774 | ||
969c7921 TH |
5775 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
5776 | if (migrate_task(p, dest_cpu)) { | |
5777 | struct migration_arg arg = { p, dest_cpu }; | |
1da177e4 LT |
5778 | /* Need help from migration thread: drop lock and wait. */ |
5779 | task_rq_unlock(rq, &flags); | |
969c7921 | 5780 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5781 | tlb_migrate_finish(p->mm); |
5782 | return 0; | |
5783 | } | |
5784 | out: | |
5785 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5786 | |
1da177e4 LT |
5787 | return ret; |
5788 | } | |
cd8ba7cd | 5789 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5790 | |
5791 | /* | |
41a2d6cf | 5792 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5793 | * this because either it can't run here any more (set_cpus_allowed() |
5794 | * away from this CPU, or CPU going down), or because we're | |
5795 | * attempting to rebalance this task on exec (sched_exec). | |
5796 | * | |
5797 | * So we race with normal scheduler movements, but that's OK, as long | |
5798 | * as the task is no longer on this CPU. | |
efc30814 KK |
5799 | * |
5800 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5801 | */ |
efc30814 | 5802 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5803 | { |
70b97a7f | 5804 | struct rq *rq_dest, *rq_src; |
e2912009 | 5805 | int ret = 0; |
1da177e4 | 5806 | |
e761b772 | 5807 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5808 | return ret; |
1da177e4 LT |
5809 | |
5810 | rq_src = cpu_rq(src_cpu); | |
5811 | rq_dest = cpu_rq(dest_cpu); | |
5812 | ||
5813 | double_rq_lock(rq_src, rq_dest); | |
5814 | /* Already moved. */ | |
5815 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5816 | goto done; |
1da177e4 | 5817 | /* Affinity changed (again). */ |
96f874e2 | 5818 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5819 | goto fail; |
1da177e4 | 5820 | |
e2912009 PZ |
5821 | /* |
5822 | * If we're not on a rq, the next wake-up will ensure we're | |
5823 | * placed properly. | |
5824 | */ | |
5825 | if (p->se.on_rq) { | |
2e1cb74a | 5826 | deactivate_task(rq_src, p, 0); |
e2912009 | 5827 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5828 | activate_task(rq_dest, p, 0); |
15afe09b | 5829 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5830 | } |
b1e38734 | 5831 | done: |
efc30814 | 5832 | ret = 1; |
b1e38734 | 5833 | fail: |
1da177e4 | 5834 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5835 | return ret; |
1da177e4 LT |
5836 | } |
5837 | ||
5838 | /* | |
969c7921 TH |
5839 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5840 | * and performs thread migration by bumping thread off CPU then | |
5841 | * 'pushing' onto another runqueue. | |
1da177e4 | 5842 | */ |
969c7921 | 5843 | static int migration_cpu_stop(void *data) |
1da177e4 | 5844 | { |
969c7921 | 5845 | struct migration_arg *arg = data; |
f7b4cddc | 5846 | |
969c7921 TH |
5847 | /* |
5848 | * The original target cpu might have gone down and we might | |
5849 | * be on another cpu but it doesn't matter. | |
5850 | */ | |
f7b4cddc | 5851 | local_irq_disable(); |
969c7921 | 5852 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5853 | local_irq_enable(); |
1da177e4 | 5854 | return 0; |
f7b4cddc ON |
5855 | } |
5856 | ||
1da177e4 | 5857 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5858 | /* |
3a4fa0a2 | 5859 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5860 | */ |
6a1bdc1b | 5861 | void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5862 | { |
1445c08d ON |
5863 | struct rq *rq = cpu_rq(dead_cpu); |
5864 | int needs_cpu, uninitialized_var(dest_cpu); | |
5865 | unsigned long flags; | |
e76bd8d9 | 5866 | |
1445c08d | 5867 | local_irq_save(flags); |
e76bd8d9 | 5868 | |
1445c08d ON |
5869 | raw_spin_lock(&rq->lock); |
5870 | needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING); | |
5871 | if (needs_cpu) | |
5872 | dest_cpu = select_fallback_rq(dead_cpu, p); | |
5873 | raw_spin_unlock(&rq->lock); | |
c1804d54 ON |
5874 | /* |
5875 | * It can only fail if we race with set_cpus_allowed(), | |
5876 | * in the racer should migrate the task anyway. | |
5877 | */ | |
1445c08d | 5878 | if (needs_cpu) |
c1804d54 | 5879 | __migrate_task(p, dead_cpu, dest_cpu); |
1445c08d | 5880 | local_irq_restore(flags); |
1da177e4 LT |
5881 | } |
5882 | ||
5883 | /* | |
5884 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5885 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5886 | * for performance reasons the counter is not stricly tracking tasks to | |
5887 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5888 | * to keep the global sum constant after CPU-down: | |
5889 | */ | |
70b97a7f | 5890 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5891 | { |
6ad4c188 | 5892 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5893 | unsigned long flags; |
5894 | ||
5895 | local_irq_save(flags); | |
5896 | double_rq_lock(rq_src, rq_dest); | |
5897 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5898 | rq_src->nr_uninterruptible = 0; | |
5899 | double_rq_unlock(rq_src, rq_dest); | |
5900 | local_irq_restore(flags); | |
5901 | } | |
5902 | ||
5903 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5904 | static void migrate_live_tasks(int src_cpu) | |
5905 | { | |
48f24c4d | 5906 | struct task_struct *p, *t; |
1da177e4 | 5907 | |
f7b4cddc | 5908 | read_lock(&tasklist_lock); |
1da177e4 | 5909 | |
48f24c4d IM |
5910 | do_each_thread(t, p) { |
5911 | if (p == current) | |
1da177e4 LT |
5912 | continue; |
5913 | ||
48f24c4d IM |
5914 | if (task_cpu(p) == src_cpu) |
5915 | move_task_off_dead_cpu(src_cpu, p); | |
5916 | } while_each_thread(t, p); | |
1da177e4 | 5917 | |
f7b4cddc | 5918 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5919 | } |
5920 | ||
dd41f596 IM |
5921 | /* |
5922 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5923 | * It does so by boosting its priority to highest possible. |
5924 | * Used by CPU offline code. | |
1da177e4 LT |
5925 | */ |
5926 | void sched_idle_next(void) | |
5927 | { | |
48f24c4d | 5928 | int this_cpu = smp_processor_id(); |
70b97a7f | 5929 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5930 | struct task_struct *p = rq->idle; |
5931 | unsigned long flags; | |
5932 | ||
5933 | /* cpu has to be offline */ | |
48f24c4d | 5934 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5935 | |
48f24c4d IM |
5936 | /* |
5937 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5938 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5939 | */ |
05fa785c | 5940 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5941 | |
dd41f596 | 5942 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5943 | |
94bc9a7b | 5944 | activate_task(rq, p, 0); |
1da177e4 | 5945 | |
05fa785c | 5946 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5947 | } |
5948 | ||
48f24c4d IM |
5949 | /* |
5950 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5951 | * offline. |
5952 | */ | |
5953 | void idle_task_exit(void) | |
5954 | { | |
5955 | struct mm_struct *mm = current->active_mm; | |
5956 | ||
5957 | BUG_ON(cpu_online(smp_processor_id())); | |
5958 | ||
5959 | if (mm != &init_mm) | |
5960 | switch_mm(mm, &init_mm, current); | |
5961 | mmdrop(mm); | |
5962 | } | |
5963 | ||
054b9108 | 5964 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5965 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5966 | { |
70b97a7f | 5967 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5968 | |
5969 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5970 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5971 | |
5972 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5973 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5974 | |
48f24c4d | 5975 | get_task_struct(p); |
1da177e4 LT |
5976 | |
5977 | /* | |
5978 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5979 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5980 | * fine. |
5981 | */ | |
05fa785c | 5982 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5983 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5984 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5985 | |
48f24c4d | 5986 | put_task_struct(p); |
1da177e4 LT |
5987 | } |
5988 | ||
5989 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5990 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5991 | { | |
70b97a7f | 5992 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5993 | struct task_struct *next; |
48f24c4d | 5994 | |
dd41f596 IM |
5995 | for ( ; ; ) { |
5996 | if (!rq->nr_running) | |
5997 | break; | |
b67802ea | 5998 | next = pick_next_task(rq); |
dd41f596 IM |
5999 | if (!next) |
6000 | break; | |
79c53799 | 6001 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6002 | migrate_dead(dead_cpu, next); |
e692ab53 | 6003 | |
1da177e4 LT |
6004 | } |
6005 | } | |
dce48a84 TG |
6006 | |
6007 | /* | |
6008 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
6009 | */ | |
6010 | static void calc_global_load_remove(struct rq *rq) | |
6011 | { | |
6012 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 6013 | rq->calc_load_active = 0; |
dce48a84 | 6014 | } |
1da177e4 LT |
6015 | #endif /* CONFIG_HOTPLUG_CPU */ |
6016 | ||
e692ab53 NP |
6017 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6018 | ||
6019 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6020 | { |
6021 | .procname = "sched_domain", | |
c57baf1e | 6022 | .mode = 0555, |
e0361851 | 6023 | }, |
56992309 | 6024 | {} |
e692ab53 NP |
6025 | }; |
6026 | ||
6027 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6028 | { |
6029 | .procname = "kernel", | |
c57baf1e | 6030 | .mode = 0555, |
e0361851 AD |
6031 | .child = sd_ctl_dir, |
6032 | }, | |
56992309 | 6033 | {} |
e692ab53 NP |
6034 | }; |
6035 | ||
6036 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6037 | { | |
6038 | struct ctl_table *entry = | |
5cf9f062 | 6039 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6040 | |
e692ab53 NP |
6041 | return entry; |
6042 | } | |
6043 | ||
6382bc90 MM |
6044 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6045 | { | |
cd790076 | 6046 | struct ctl_table *entry; |
6382bc90 | 6047 | |
cd790076 MM |
6048 | /* |
6049 | * In the intermediate directories, both the child directory and | |
6050 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6051 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6052 | * static strings and all have proc handlers. |
6053 | */ | |
6054 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6055 | if (entry->child) |
6056 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6057 | if (entry->proc_handler == NULL) |
6058 | kfree(entry->procname); | |
6059 | } | |
6382bc90 MM |
6060 | |
6061 | kfree(*tablep); | |
6062 | *tablep = NULL; | |
6063 | } | |
6064 | ||
e692ab53 | 6065 | static void |
e0361851 | 6066 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6067 | const char *procname, void *data, int maxlen, |
6068 | mode_t mode, proc_handler *proc_handler) | |
6069 | { | |
e692ab53 NP |
6070 | entry->procname = procname; |
6071 | entry->data = data; | |
6072 | entry->maxlen = maxlen; | |
6073 | entry->mode = mode; | |
6074 | entry->proc_handler = proc_handler; | |
6075 | } | |
6076 | ||
6077 | static struct ctl_table * | |
6078 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6079 | { | |
a5d8c348 | 6080 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6081 | |
ad1cdc1d MM |
6082 | if (table == NULL) |
6083 | return NULL; | |
6084 | ||
e0361851 | 6085 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6086 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6087 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6088 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6089 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6090 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6091 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6092 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6093 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6094 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6095 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6096 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6097 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6098 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6099 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6100 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6101 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6102 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6103 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6104 | &sd->cache_nice_tries, |
6105 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6106 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6107 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6108 | set_table_entry(&table[11], "name", sd->name, |
6109 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6110 | /* &table[12] is terminator */ | |
e692ab53 NP |
6111 | |
6112 | return table; | |
6113 | } | |
6114 | ||
9a4e7159 | 6115 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6116 | { |
6117 | struct ctl_table *entry, *table; | |
6118 | struct sched_domain *sd; | |
6119 | int domain_num = 0, i; | |
6120 | char buf[32]; | |
6121 | ||
6122 | for_each_domain(cpu, sd) | |
6123 | domain_num++; | |
6124 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6125 | if (table == NULL) |
6126 | return NULL; | |
e692ab53 NP |
6127 | |
6128 | i = 0; | |
6129 | for_each_domain(cpu, sd) { | |
6130 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6131 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6132 | entry->mode = 0555; |
e692ab53 NP |
6133 | entry->child = sd_alloc_ctl_domain_table(sd); |
6134 | entry++; | |
6135 | i++; | |
6136 | } | |
6137 | return table; | |
6138 | } | |
6139 | ||
6140 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6141 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6142 | { |
6ad4c188 | 6143 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6144 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6145 | char buf[32]; | |
6146 | ||
7378547f MM |
6147 | WARN_ON(sd_ctl_dir[0].child); |
6148 | sd_ctl_dir[0].child = entry; | |
6149 | ||
ad1cdc1d MM |
6150 | if (entry == NULL) |
6151 | return; | |
6152 | ||
6ad4c188 | 6153 | for_each_possible_cpu(i) { |
e692ab53 | 6154 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6155 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6156 | entry->mode = 0555; |
e692ab53 | 6157 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6158 | entry++; |
e692ab53 | 6159 | } |
7378547f MM |
6160 | |
6161 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6162 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6163 | } | |
6382bc90 | 6164 | |
7378547f | 6165 | /* may be called multiple times per register */ |
6382bc90 MM |
6166 | static void unregister_sched_domain_sysctl(void) |
6167 | { | |
7378547f MM |
6168 | if (sd_sysctl_header) |
6169 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6170 | sd_sysctl_header = NULL; |
7378547f MM |
6171 | if (sd_ctl_dir[0].child) |
6172 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6173 | } |
e692ab53 | 6174 | #else |
6382bc90 MM |
6175 | static void register_sched_domain_sysctl(void) |
6176 | { | |
6177 | } | |
6178 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6179 | { |
6180 | } | |
6181 | #endif | |
6182 | ||
1f11eb6a GH |
6183 | static void set_rq_online(struct rq *rq) |
6184 | { | |
6185 | if (!rq->online) { | |
6186 | const struct sched_class *class; | |
6187 | ||
c6c4927b | 6188 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6189 | rq->online = 1; |
6190 | ||
6191 | for_each_class(class) { | |
6192 | if (class->rq_online) | |
6193 | class->rq_online(rq); | |
6194 | } | |
6195 | } | |
6196 | } | |
6197 | ||
6198 | static void set_rq_offline(struct rq *rq) | |
6199 | { | |
6200 | if (rq->online) { | |
6201 | const struct sched_class *class; | |
6202 | ||
6203 | for_each_class(class) { | |
6204 | if (class->rq_offline) | |
6205 | class->rq_offline(rq); | |
6206 | } | |
6207 | ||
c6c4927b | 6208 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6209 | rq->online = 0; |
6210 | } | |
6211 | } | |
6212 | ||
1da177e4 LT |
6213 | /* |
6214 | * migration_call - callback that gets triggered when a CPU is added. | |
6215 | * Here we can start up the necessary migration thread for the new CPU. | |
6216 | */ | |
48f24c4d IM |
6217 | static int __cpuinit |
6218 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6219 | { |
48f24c4d | 6220 | int cpu = (long)hcpu; |
1da177e4 | 6221 | unsigned long flags; |
969c7921 | 6222 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6223 | |
6224 | switch (action) { | |
5be9361c | 6225 | |
1da177e4 | 6226 | case CPU_UP_PREPARE: |
8bb78442 | 6227 | case CPU_UP_PREPARE_FROZEN: |
a468d389 | 6228 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6229 | break; |
48f24c4d | 6230 | |
1da177e4 | 6231 | case CPU_ONLINE: |
8bb78442 | 6232 | case CPU_ONLINE_FROZEN: |
1f94ef59 | 6233 | /* Update our root-domain */ |
05fa785c | 6234 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6235 | if (rq->rd) { |
c6c4927b | 6236 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6237 | |
6238 | set_rq_online(rq); | |
1f94ef59 | 6239 | } |
05fa785c | 6240 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6241 | break; |
48f24c4d | 6242 | |
1da177e4 | 6243 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 6244 | case CPU_DEAD: |
8bb78442 | 6245 | case CPU_DEAD_FROZEN: |
1da177e4 | 6246 | migrate_live_tasks(cpu); |
1da177e4 | 6247 | /* Idle task back to normal (off runqueue, low prio) */ |
05fa785c | 6248 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 6249 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
6250 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6251 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6252 | migrate_dead_tasks(cpu); |
05fa785c | 6253 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
6254 | migrate_nr_uninterruptible(rq); |
6255 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 6256 | calc_global_load_remove(rq); |
1da177e4 | 6257 | break; |
57d885fe | 6258 | |
08f503b0 GH |
6259 | case CPU_DYING: |
6260 | case CPU_DYING_FROZEN: | |
57d885fe | 6261 | /* Update our root-domain */ |
05fa785c | 6262 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6263 | if (rq->rd) { |
c6c4927b | 6264 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6265 | set_rq_offline(rq); |
57d885fe | 6266 | } |
05fa785c | 6267 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 6268 | break; |
1da177e4 LT |
6269 | #endif |
6270 | } | |
6271 | return NOTIFY_OK; | |
6272 | } | |
6273 | ||
f38b0820 PM |
6274 | /* |
6275 | * Register at high priority so that task migration (migrate_all_tasks) | |
6276 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6277 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6278 | */ |
26c2143b | 6279 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6280 | .notifier_call = migration_call, |
50a323b7 | 6281 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6282 | }; |
6283 | ||
3a101d05 TH |
6284 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6285 | unsigned long action, void *hcpu) | |
6286 | { | |
6287 | switch (action & ~CPU_TASKS_FROZEN) { | |
6288 | case CPU_ONLINE: | |
6289 | case CPU_DOWN_FAILED: | |
6290 | set_cpu_active((long)hcpu, true); | |
6291 | return NOTIFY_OK; | |
6292 | default: | |
6293 | return NOTIFY_DONE; | |
6294 | } | |
6295 | } | |
6296 | ||
6297 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6298 | unsigned long action, void *hcpu) | |
6299 | { | |
6300 | switch (action & ~CPU_TASKS_FROZEN) { | |
6301 | case CPU_DOWN_PREPARE: | |
6302 | set_cpu_active((long)hcpu, false); | |
6303 | return NOTIFY_OK; | |
6304 | default: | |
6305 | return NOTIFY_DONE; | |
6306 | } | |
6307 | } | |
6308 | ||
7babe8db | 6309 | static int __init migration_init(void) |
1da177e4 LT |
6310 | { |
6311 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6312 | int err; |
48f24c4d | 6313 | |
3a101d05 | 6314 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6315 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6316 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6317 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6318 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6319 | |
3a101d05 TH |
6320 | /* Register cpu active notifiers */ |
6321 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6322 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6323 | ||
a004cd42 | 6324 | return 0; |
1da177e4 | 6325 | } |
7babe8db | 6326 | early_initcall(migration_init); |
1da177e4 LT |
6327 | #endif |
6328 | ||
6329 | #ifdef CONFIG_SMP | |
476f3534 | 6330 | |
3e9830dc | 6331 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6332 | |
f6630114 MT |
6333 | static __read_mostly int sched_domain_debug_enabled; |
6334 | ||
6335 | static int __init sched_domain_debug_setup(char *str) | |
6336 | { | |
6337 | sched_domain_debug_enabled = 1; | |
6338 | ||
6339 | return 0; | |
6340 | } | |
6341 | early_param("sched_debug", sched_domain_debug_setup); | |
6342 | ||
7c16ec58 | 6343 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6344 | struct cpumask *groupmask) |
1da177e4 | 6345 | { |
4dcf6aff | 6346 | struct sched_group *group = sd->groups; |
434d53b0 | 6347 | char str[256]; |
1da177e4 | 6348 | |
968ea6d8 | 6349 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6350 | cpumask_clear(groupmask); |
4dcf6aff IM |
6351 | |
6352 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6353 | ||
6354 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6355 | printk("does not load-balance\n"); |
4dcf6aff | 6356 | if (sd->parent) |
3df0fc5b PZ |
6357 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6358 | " has parent"); | |
4dcf6aff | 6359 | return -1; |
41c7ce9a NP |
6360 | } |
6361 | ||
3df0fc5b | 6362 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6363 | |
758b2cdc | 6364 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6365 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6366 | "CPU%d\n", cpu); | |
4dcf6aff | 6367 | } |
758b2cdc | 6368 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6369 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6370 | " CPU%d\n", cpu); | |
4dcf6aff | 6371 | } |
1da177e4 | 6372 | |
4dcf6aff | 6373 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6374 | do { |
4dcf6aff | 6375 | if (!group) { |
3df0fc5b PZ |
6376 | printk("\n"); |
6377 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6378 | break; |
6379 | } | |
6380 | ||
18a3885f | 6381 | if (!group->cpu_power) { |
3df0fc5b PZ |
6382 | printk(KERN_CONT "\n"); |
6383 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6384 | "set\n"); | |
4dcf6aff IM |
6385 | break; |
6386 | } | |
1da177e4 | 6387 | |
758b2cdc | 6388 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6389 | printk(KERN_CONT "\n"); |
6390 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6391 | break; |
6392 | } | |
1da177e4 | 6393 | |
758b2cdc | 6394 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6395 | printk(KERN_CONT "\n"); |
6396 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6397 | break; |
6398 | } | |
1da177e4 | 6399 | |
758b2cdc | 6400 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6401 | |
968ea6d8 | 6402 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6403 | |
3df0fc5b | 6404 | printk(KERN_CONT " %s", str); |
18a3885f | 6405 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6406 | printk(KERN_CONT " (cpu_power = %d)", |
6407 | group->cpu_power); | |
381512cf | 6408 | } |
1da177e4 | 6409 | |
4dcf6aff IM |
6410 | group = group->next; |
6411 | } while (group != sd->groups); | |
3df0fc5b | 6412 | printk(KERN_CONT "\n"); |
1da177e4 | 6413 | |
758b2cdc | 6414 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6415 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6416 | |
758b2cdc RR |
6417 | if (sd->parent && |
6418 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6419 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6420 | "of domain->span\n"); | |
4dcf6aff IM |
6421 | return 0; |
6422 | } | |
1da177e4 | 6423 | |
4dcf6aff IM |
6424 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6425 | { | |
d5dd3db1 | 6426 | cpumask_var_t groupmask; |
4dcf6aff | 6427 | int level = 0; |
1da177e4 | 6428 | |
f6630114 MT |
6429 | if (!sched_domain_debug_enabled) |
6430 | return; | |
6431 | ||
4dcf6aff IM |
6432 | if (!sd) { |
6433 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6434 | return; | |
6435 | } | |
1da177e4 | 6436 | |
4dcf6aff IM |
6437 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6438 | ||
d5dd3db1 | 6439 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6440 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6441 | return; | |
6442 | } | |
6443 | ||
4dcf6aff | 6444 | for (;;) { |
7c16ec58 | 6445 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6446 | break; |
1da177e4 LT |
6447 | level++; |
6448 | sd = sd->parent; | |
33859f7f | 6449 | if (!sd) |
4dcf6aff IM |
6450 | break; |
6451 | } | |
d5dd3db1 | 6452 | free_cpumask_var(groupmask); |
1da177e4 | 6453 | } |
6d6bc0ad | 6454 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6455 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6456 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6457 | |
1a20ff27 | 6458 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6459 | { |
758b2cdc | 6460 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6461 | return 1; |
6462 | ||
6463 | /* Following flags need at least 2 groups */ | |
6464 | if (sd->flags & (SD_LOAD_BALANCE | | |
6465 | SD_BALANCE_NEWIDLE | | |
6466 | SD_BALANCE_FORK | | |
89c4710e SS |
6467 | SD_BALANCE_EXEC | |
6468 | SD_SHARE_CPUPOWER | | |
6469 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6470 | if (sd->groups != sd->groups->next) |
6471 | return 0; | |
6472 | } | |
6473 | ||
6474 | /* Following flags don't use groups */ | |
c88d5910 | 6475 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6476 | return 0; |
6477 | ||
6478 | return 1; | |
6479 | } | |
6480 | ||
48f24c4d IM |
6481 | static int |
6482 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6483 | { |
6484 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6485 | ||
6486 | if (sd_degenerate(parent)) | |
6487 | return 1; | |
6488 | ||
758b2cdc | 6489 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6490 | return 0; |
6491 | ||
245af2c7 SS |
6492 | /* Flags needing groups don't count if only 1 group in parent */ |
6493 | if (parent->groups == parent->groups->next) { | |
6494 | pflags &= ~(SD_LOAD_BALANCE | | |
6495 | SD_BALANCE_NEWIDLE | | |
6496 | SD_BALANCE_FORK | | |
89c4710e SS |
6497 | SD_BALANCE_EXEC | |
6498 | SD_SHARE_CPUPOWER | | |
6499 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6500 | if (nr_node_ids == 1) |
6501 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6502 | } |
6503 | if (~cflags & pflags) | |
6504 | return 0; | |
6505 | ||
6506 | return 1; | |
6507 | } | |
6508 | ||
c6c4927b RR |
6509 | static void free_rootdomain(struct root_domain *rd) |
6510 | { | |
047106ad PZ |
6511 | synchronize_sched(); |
6512 | ||
68e74568 RR |
6513 | cpupri_cleanup(&rd->cpupri); |
6514 | ||
c6c4927b RR |
6515 | free_cpumask_var(rd->rto_mask); |
6516 | free_cpumask_var(rd->online); | |
6517 | free_cpumask_var(rd->span); | |
6518 | kfree(rd); | |
6519 | } | |
6520 | ||
57d885fe GH |
6521 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6522 | { | |
a0490fa3 | 6523 | struct root_domain *old_rd = NULL; |
57d885fe | 6524 | unsigned long flags; |
57d885fe | 6525 | |
05fa785c | 6526 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6527 | |
6528 | if (rq->rd) { | |
a0490fa3 | 6529 | old_rd = rq->rd; |
57d885fe | 6530 | |
c6c4927b | 6531 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6532 | set_rq_offline(rq); |
57d885fe | 6533 | |
c6c4927b | 6534 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6535 | |
a0490fa3 IM |
6536 | /* |
6537 | * If we dont want to free the old_rt yet then | |
6538 | * set old_rd to NULL to skip the freeing later | |
6539 | * in this function: | |
6540 | */ | |
6541 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6542 | old_rd = NULL; | |
57d885fe GH |
6543 | } |
6544 | ||
6545 | atomic_inc(&rd->refcount); | |
6546 | rq->rd = rd; | |
6547 | ||
c6c4927b | 6548 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6549 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6550 | set_rq_online(rq); |
57d885fe | 6551 | |
05fa785c | 6552 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6553 | |
6554 | if (old_rd) | |
6555 | free_rootdomain(old_rd); | |
57d885fe GH |
6556 | } |
6557 | ||
68c38fc3 | 6558 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6559 | { |
6560 | memset(rd, 0, sizeof(*rd)); | |
6561 | ||
68c38fc3 | 6562 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6563 | goto out; |
68c38fc3 | 6564 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6565 | goto free_span; |
68c38fc3 | 6566 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6567 | goto free_online; |
6e0534f2 | 6568 | |
68c38fc3 | 6569 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6570 | goto free_rto_mask; |
c6c4927b | 6571 | return 0; |
6e0534f2 | 6572 | |
68e74568 RR |
6573 | free_rto_mask: |
6574 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6575 | free_online: |
6576 | free_cpumask_var(rd->online); | |
6577 | free_span: | |
6578 | free_cpumask_var(rd->span); | |
0c910d28 | 6579 | out: |
c6c4927b | 6580 | return -ENOMEM; |
57d885fe GH |
6581 | } |
6582 | ||
6583 | static void init_defrootdomain(void) | |
6584 | { | |
68c38fc3 | 6585 | init_rootdomain(&def_root_domain); |
c6c4927b | 6586 | |
57d885fe GH |
6587 | atomic_set(&def_root_domain.refcount, 1); |
6588 | } | |
6589 | ||
dc938520 | 6590 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6591 | { |
6592 | struct root_domain *rd; | |
6593 | ||
6594 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6595 | if (!rd) | |
6596 | return NULL; | |
6597 | ||
68c38fc3 | 6598 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6599 | kfree(rd); |
6600 | return NULL; | |
6601 | } | |
57d885fe GH |
6602 | |
6603 | return rd; | |
6604 | } | |
6605 | ||
1da177e4 | 6606 | /* |
0eab9146 | 6607 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6608 | * hold the hotplug lock. |
6609 | */ | |
0eab9146 IM |
6610 | static void |
6611 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6612 | { |
70b97a7f | 6613 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6614 | struct sched_domain *tmp; |
6615 | ||
669c55e9 PZ |
6616 | for (tmp = sd; tmp; tmp = tmp->parent) |
6617 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6618 | ||
245af2c7 | 6619 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6620 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6621 | struct sched_domain *parent = tmp->parent; |
6622 | if (!parent) | |
6623 | break; | |
f29c9b1c | 6624 | |
1a848870 | 6625 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6626 | tmp->parent = parent->parent; |
1a848870 SS |
6627 | if (parent->parent) |
6628 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6629 | } else |
6630 | tmp = tmp->parent; | |
245af2c7 SS |
6631 | } |
6632 | ||
1a848870 | 6633 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6634 | sd = sd->parent; |
1a848870 SS |
6635 | if (sd) |
6636 | sd->child = NULL; | |
6637 | } | |
1da177e4 LT |
6638 | |
6639 | sched_domain_debug(sd, cpu); | |
6640 | ||
57d885fe | 6641 | rq_attach_root(rq, rd); |
674311d5 | 6642 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6643 | } |
6644 | ||
6645 | /* cpus with isolated domains */ | |
dcc30a35 | 6646 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6647 | |
6648 | /* Setup the mask of cpus configured for isolated domains */ | |
6649 | static int __init isolated_cpu_setup(char *str) | |
6650 | { | |
bdddd296 | 6651 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6652 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6653 | return 1; |
6654 | } | |
6655 | ||
8927f494 | 6656 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6657 | |
6658 | /* | |
6711cab4 SS |
6659 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6660 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6661 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6662 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6663 | * |
6664 | * init_sched_build_groups will build a circular linked list of the groups | |
6665 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6666 | * and ->cpu_power to 0. | |
6667 | */ | |
a616058b | 6668 | static void |
96f874e2 RR |
6669 | init_sched_build_groups(const struct cpumask *span, |
6670 | const struct cpumask *cpu_map, | |
6671 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6672 | struct sched_group **sg, |
96f874e2 RR |
6673 | struct cpumask *tmpmask), |
6674 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6675 | { |
6676 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6677 | int i; |
6678 | ||
96f874e2 | 6679 | cpumask_clear(covered); |
7c16ec58 | 6680 | |
abcd083a | 6681 | for_each_cpu(i, span) { |
6711cab4 | 6682 | struct sched_group *sg; |
7c16ec58 | 6683 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6684 | int j; |
6685 | ||
758b2cdc | 6686 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6687 | continue; |
6688 | ||
758b2cdc | 6689 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6690 | sg->cpu_power = 0; |
1da177e4 | 6691 | |
abcd083a | 6692 | for_each_cpu(j, span) { |
7c16ec58 | 6693 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6694 | continue; |
6695 | ||
96f874e2 | 6696 | cpumask_set_cpu(j, covered); |
758b2cdc | 6697 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6698 | } |
6699 | if (!first) | |
6700 | first = sg; | |
6701 | if (last) | |
6702 | last->next = sg; | |
6703 | last = sg; | |
6704 | } | |
6705 | last->next = first; | |
6706 | } | |
6707 | ||
9c1cfda2 | 6708 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6709 | |
9c1cfda2 | 6710 | #ifdef CONFIG_NUMA |
198e2f18 | 6711 | |
9c1cfda2 JH |
6712 | /** |
6713 | * find_next_best_node - find the next node to include in a sched_domain | |
6714 | * @node: node whose sched_domain we're building | |
6715 | * @used_nodes: nodes already in the sched_domain | |
6716 | * | |
41a2d6cf | 6717 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6718 | * finds the closest node not already in the @used_nodes map. |
6719 | * | |
6720 | * Should use nodemask_t. | |
6721 | */ | |
c5f59f08 | 6722 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6723 | { |
6724 | int i, n, val, min_val, best_node = 0; | |
6725 | ||
6726 | min_val = INT_MAX; | |
6727 | ||
076ac2af | 6728 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6729 | /* Start at @node */ |
076ac2af | 6730 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6731 | |
6732 | if (!nr_cpus_node(n)) | |
6733 | continue; | |
6734 | ||
6735 | /* Skip already used nodes */ | |
c5f59f08 | 6736 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6737 | continue; |
6738 | ||
6739 | /* Simple min distance search */ | |
6740 | val = node_distance(node, n); | |
6741 | ||
6742 | if (val < min_val) { | |
6743 | min_val = val; | |
6744 | best_node = n; | |
6745 | } | |
6746 | } | |
6747 | ||
c5f59f08 | 6748 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6749 | return best_node; |
6750 | } | |
6751 | ||
6752 | /** | |
6753 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6754 | * @node: node whose cpumask we're constructing | |
73486722 | 6755 | * @span: resulting cpumask |
9c1cfda2 | 6756 | * |
41a2d6cf | 6757 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6758 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6759 | * out optimally. | |
6760 | */ | |
96f874e2 | 6761 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6762 | { |
c5f59f08 | 6763 | nodemask_t used_nodes; |
48f24c4d | 6764 | int i; |
9c1cfda2 | 6765 | |
6ca09dfc | 6766 | cpumask_clear(span); |
c5f59f08 | 6767 | nodes_clear(used_nodes); |
9c1cfda2 | 6768 | |
6ca09dfc | 6769 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6770 | node_set(node, used_nodes); |
9c1cfda2 JH |
6771 | |
6772 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6773 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6774 | |
6ca09dfc | 6775 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6776 | } |
9c1cfda2 | 6777 | } |
6d6bc0ad | 6778 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6779 | |
5c45bf27 | 6780 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6781 | |
6c99e9ad RR |
6782 | /* |
6783 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6784 | * |
6785 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6786 | * and struct sched_domain. ) | |
6c99e9ad RR |
6787 | */ |
6788 | struct static_sched_group { | |
6789 | struct sched_group sg; | |
6790 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6791 | }; | |
6792 | ||
6793 | struct static_sched_domain { | |
6794 | struct sched_domain sd; | |
6795 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6796 | }; | |
6797 | ||
49a02c51 AH |
6798 | struct s_data { |
6799 | #ifdef CONFIG_NUMA | |
6800 | int sd_allnodes; | |
6801 | cpumask_var_t domainspan; | |
6802 | cpumask_var_t covered; | |
6803 | cpumask_var_t notcovered; | |
6804 | #endif | |
6805 | cpumask_var_t nodemask; | |
6806 | cpumask_var_t this_sibling_map; | |
6807 | cpumask_var_t this_core_map; | |
01a08546 | 6808 | cpumask_var_t this_book_map; |
49a02c51 AH |
6809 | cpumask_var_t send_covered; |
6810 | cpumask_var_t tmpmask; | |
6811 | struct sched_group **sched_group_nodes; | |
6812 | struct root_domain *rd; | |
6813 | }; | |
6814 | ||
2109b99e AH |
6815 | enum s_alloc { |
6816 | sa_sched_groups = 0, | |
6817 | sa_rootdomain, | |
6818 | sa_tmpmask, | |
6819 | sa_send_covered, | |
01a08546 | 6820 | sa_this_book_map, |
2109b99e AH |
6821 | sa_this_core_map, |
6822 | sa_this_sibling_map, | |
6823 | sa_nodemask, | |
6824 | sa_sched_group_nodes, | |
6825 | #ifdef CONFIG_NUMA | |
6826 | sa_notcovered, | |
6827 | sa_covered, | |
6828 | sa_domainspan, | |
6829 | #endif | |
6830 | sa_none, | |
6831 | }; | |
6832 | ||
9c1cfda2 | 6833 | /* |
48f24c4d | 6834 | * SMT sched-domains: |
9c1cfda2 | 6835 | */ |
1da177e4 | 6836 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6837 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6838 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6839 | |
41a2d6cf | 6840 | static int |
96f874e2 RR |
6841 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6842 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6843 | { |
6711cab4 | 6844 | if (sg) |
1871e52c | 6845 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6846 | return cpu; |
6847 | } | |
6d6bc0ad | 6848 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6849 | |
48f24c4d IM |
6850 | /* |
6851 | * multi-core sched-domains: | |
6852 | */ | |
1e9f28fa | 6853 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6854 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6855 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6856 | |
41a2d6cf | 6857 | static int |
96f874e2 RR |
6858 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6859 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6860 | { |
6711cab4 | 6861 | int group; |
f269893c | 6862 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 6863 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6864 | group = cpumask_first(mask); |
f269893c HC |
6865 | #else |
6866 | group = cpu; | |
6867 | #endif | |
6711cab4 | 6868 | if (sg) |
6c99e9ad | 6869 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6870 | return group; |
1e9f28fa | 6871 | } |
f269893c | 6872 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 6873 | |
01a08546 HC |
6874 | /* |
6875 | * book sched-domains: | |
6876 | */ | |
6877 | #ifdef CONFIG_SCHED_BOOK | |
6878 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
6879 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
6880 | ||
41a2d6cf | 6881 | static int |
01a08546 HC |
6882 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
6883 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6884 | { |
01a08546 HC |
6885 | int group = cpu; |
6886 | #ifdef CONFIG_SCHED_MC | |
6887 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
6888 | group = cpumask_first(mask); | |
6889 | #elif defined(CONFIG_SCHED_SMT) | |
6890 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
6891 | group = cpumask_first(mask); | |
6892 | #endif | |
6711cab4 | 6893 | if (sg) |
01a08546 HC |
6894 | *sg = &per_cpu(sched_group_book, group).sg; |
6895 | return group; | |
1e9f28fa | 6896 | } |
01a08546 | 6897 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 6898 | |
6c99e9ad RR |
6899 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6900 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6901 | |
41a2d6cf | 6902 | static int |
96f874e2 RR |
6903 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6904 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6905 | { |
6711cab4 | 6906 | int group; |
01a08546 HC |
6907 | #ifdef CONFIG_SCHED_BOOK |
6908 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
6909 | group = cpumask_first(mask); | |
6910 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 6911 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6912 | group = cpumask_first(mask); |
1e9f28fa | 6913 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6914 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6915 | group = cpumask_first(mask); |
1da177e4 | 6916 | #else |
6711cab4 | 6917 | group = cpu; |
1da177e4 | 6918 | #endif |
6711cab4 | 6919 | if (sg) |
6c99e9ad | 6920 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6921 | return group; |
1da177e4 LT |
6922 | } |
6923 | ||
6924 | #ifdef CONFIG_NUMA | |
1da177e4 | 6925 | /* |
9c1cfda2 JH |
6926 | * The init_sched_build_groups can't handle what we want to do with node |
6927 | * groups, so roll our own. Now each node has its own list of groups which | |
6928 | * gets dynamically allocated. | |
1da177e4 | 6929 | */ |
62ea9ceb | 6930 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6931 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6932 | |
62ea9ceb | 6933 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6934 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6935 | |
96f874e2 RR |
6936 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6937 | struct sched_group **sg, | |
6938 | struct cpumask *nodemask) | |
9c1cfda2 | 6939 | { |
6711cab4 SS |
6940 | int group; |
6941 | ||
6ca09dfc | 6942 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6943 | group = cpumask_first(nodemask); |
6711cab4 SS |
6944 | |
6945 | if (sg) | |
6c99e9ad | 6946 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6947 | return group; |
1da177e4 | 6948 | } |
6711cab4 | 6949 | |
08069033 SS |
6950 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6951 | { | |
6952 | struct sched_group *sg = group_head; | |
6953 | int j; | |
6954 | ||
6955 | if (!sg) | |
6956 | return; | |
3a5c359a | 6957 | do { |
758b2cdc | 6958 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6959 | struct sched_domain *sd; |
08069033 | 6960 | |
6c99e9ad | 6961 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6962 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6963 | /* |
6964 | * Only add "power" once for each | |
6965 | * physical package. | |
6966 | */ | |
6967 | continue; | |
6968 | } | |
08069033 | 6969 | |
18a3885f | 6970 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6971 | } |
6972 | sg = sg->next; | |
6973 | } while (sg != group_head); | |
08069033 | 6974 | } |
0601a88d AH |
6975 | |
6976 | static int build_numa_sched_groups(struct s_data *d, | |
6977 | const struct cpumask *cpu_map, int num) | |
6978 | { | |
6979 | struct sched_domain *sd; | |
6980 | struct sched_group *sg, *prev; | |
6981 | int n, j; | |
6982 | ||
6983 | cpumask_clear(d->covered); | |
6984 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6985 | if (cpumask_empty(d->nodemask)) { | |
6986 | d->sched_group_nodes[num] = NULL; | |
6987 | goto out; | |
6988 | } | |
6989 | ||
6990 | sched_domain_node_span(num, d->domainspan); | |
6991 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6992 | ||
6993 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6994 | GFP_KERNEL, num); | |
6995 | if (!sg) { | |
3df0fc5b PZ |
6996 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6997 | num); | |
0601a88d AH |
6998 | return -ENOMEM; |
6999 | } | |
7000 | d->sched_group_nodes[num] = sg; | |
7001 | ||
7002 | for_each_cpu(j, d->nodemask) { | |
7003 | sd = &per_cpu(node_domains, j).sd; | |
7004 | sd->groups = sg; | |
7005 | } | |
7006 | ||
18a3885f | 7007 | sg->cpu_power = 0; |
0601a88d AH |
7008 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
7009 | sg->next = sg; | |
7010 | cpumask_or(d->covered, d->covered, d->nodemask); | |
7011 | ||
7012 | prev = sg; | |
7013 | for (j = 0; j < nr_node_ids; j++) { | |
7014 | n = (num + j) % nr_node_ids; | |
7015 | cpumask_complement(d->notcovered, d->covered); | |
7016 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
7017 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
7018 | if (cpumask_empty(d->tmpmask)) | |
7019 | break; | |
7020 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
7021 | if (cpumask_empty(d->tmpmask)) | |
7022 | continue; | |
7023 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7024 | GFP_KERNEL, num); | |
7025 | if (!sg) { | |
3df0fc5b PZ |
7026 | printk(KERN_WARNING |
7027 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
7028 | return -ENOMEM; |
7029 | } | |
18a3885f | 7030 | sg->cpu_power = 0; |
0601a88d AH |
7031 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
7032 | sg->next = prev->next; | |
7033 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
7034 | prev->next = sg; | |
7035 | prev = sg; | |
7036 | } | |
7037 | out: | |
7038 | return 0; | |
7039 | } | |
6d6bc0ad | 7040 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7041 | |
a616058b | 7042 | #ifdef CONFIG_NUMA |
51888ca2 | 7043 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7044 | static void free_sched_groups(const struct cpumask *cpu_map, |
7045 | struct cpumask *nodemask) | |
51888ca2 | 7046 | { |
a616058b | 7047 | int cpu, i; |
51888ca2 | 7048 | |
abcd083a | 7049 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7050 | struct sched_group **sched_group_nodes |
7051 | = sched_group_nodes_bycpu[cpu]; | |
7052 | ||
51888ca2 SV |
7053 | if (!sched_group_nodes) |
7054 | continue; | |
7055 | ||
076ac2af | 7056 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7057 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7058 | ||
6ca09dfc | 7059 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7060 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7061 | continue; |
7062 | ||
7063 | if (sg == NULL) | |
7064 | continue; | |
7065 | sg = sg->next; | |
7066 | next_sg: | |
7067 | oldsg = sg; | |
7068 | sg = sg->next; | |
7069 | kfree(oldsg); | |
7070 | if (oldsg != sched_group_nodes[i]) | |
7071 | goto next_sg; | |
7072 | } | |
7073 | kfree(sched_group_nodes); | |
7074 | sched_group_nodes_bycpu[cpu] = NULL; | |
7075 | } | |
51888ca2 | 7076 | } |
6d6bc0ad | 7077 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7078 | static void free_sched_groups(const struct cpumask *cpu_map, |
7079 | struct cpumask *nodemask) | |
a616058b SS |
7080 | { |
7081 | } | |
6d6bc0ad | 7082 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7083 | |
89c4710e SS |
7084 | /* |
7085 | * Initialize sched groups cpu_power. | |
7086 | * | |
7087 | * cpu_power indicates the capacity of sched group, which is used while | |
7088 | * distributing the load between different sched groups in a sched domain. | |
7089 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7090 | * there are asymmetries in the topology. If there are asymmetries, group | |
7091 | * having more cpu_power will pickup more load compared to the group having | |
7092 | * less cpu_power. | |
89c4710e SS |
7093 | */ |
7094 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7095 | { | |
7096 | struct sched_domain *child; | |
7097 | struct sched_group *group; | |
f93e65c1 PZ |
7098 | long power; |
7099 | int weight; | |
89c4710e SS |
7100 | |
7101 | WARN_ON(!sd || !sd->groups); | |
7102 | ||
13318a71 | 7103 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
7104 | return; |
7105 | ||
aae6d3dd SS |
7106 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
7107 | ||
89c4710e SS |
7108 | child = sd->child; |
7109 | ||
18a3885f | 7110 | sd->groups->cpu_power = 0; |
5517d86b | 7111 | |
f93e65c1 PZ |
7112 | if (!child) { |
7113 | power = SCHED_LOAD_SCALE; | |
7114 | weight = cpumask_weight(sched_domain_span(sd)); | |
7115 | /* | |
7116 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
7117 | * Usually multiple threads get a better yield out of |
7118 | * that one core than a single thread would have, | |
7119 | * reflect that in sd->smt_gain. | |
f93e65c1 | 7120 | */ |
a52bfd73 PZ |
7121 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
7122 | power *= sd->smt_gain; | |
f93e65c1 | 7123 | power /= weight; |
a52bfd73 PZ |
7124 | power >>= SCHED_LOAD_SHIFT; |
7125 | } | |
18a3885f | 7126 | sd->groups->cpu_power += power; |
89c4710e SS |
7127 | return; |
7128 | } | |
7129 | ||
89c4710e | 7130 | /* |
f93e65c1 | 7131 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
7132 | */ |
7133 | group = child->groups; | |
7134 | do { | |
18a3885f | 7135 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
7136 | group = group->next; |
7137 | } while (group != child->groups); | |
7138 | } | |
7139 | ||
7c16ec58 MT |
7140 | /* |
7141 | * Initializers for schedule domains | |
7142 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7143 | */ | |
7144 | ||
a5d8c348 IM |
7145 | #ifdef CONFIG_SCHED_DEBUG |
7146 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7147 | #else | |
7148 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7149 | #endif | |
7150 | ||
7c16ec58 | 7151 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7152 | |
7c16ec58 MT |
7153 | #define SD_INIT_FUNC(type) \ |
7154 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7155 | { \ | |
7156 | memset(sd, 0, sizeof(*sd)); \ | |
7157 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7158 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7159 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7160 | } |
7161 | ||
7162 | SD_INIT_FUNC(CPU) | |
7163 | #ifdef CONFIG_NUMA | |
7164 | SD_INIT_FUNC(ALLNODES) | |
7165 | SD_INIT_FUNC(NODE) | |
7166 | #endif | |
7167 | #ifdef CONFIG_SCHED_SMT | |
7168 | SD_INIT_FUNC(SIBLING) | |
7169 | #endif | |
7170 | #ifdef CONFIG_SCHED_MC | |
7171 | SD_INIT_FUNC(MC) | |
7172 | #endif | |
01a08546 HC |
7173 | #ifdef CONFIG_SCHED_BOOK |
7174 | SD_INIT_FUNC(BOOK) | |
7175 | #endif | |
7c16ec58 | 7176 | |
1d3504fc HS |
7177 | static int default_relax_domain_level = -1; |
7178 | ||
7179 | static int __init setup_relax_domain_level(char *str) | |
7180 | { | |
30e0e178 LZ |
7181 | unsigned long val; |
7182 | ||
7183 | val = simple_strtoul(str, NULL, 0); | |
7184 | if (val < SD_LV_MAX) | |
7185 | default_relax_domain_level = val; | |
7186 | ||
1d3504fc HS |
7187 | return 1; |
7188 | } | |
7189 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7190 | ||
7191 | static void set_domain_attribute(struct sched_domain *sd, | |
7192 | struct sched_domain_attr *attr) | |
7193 | { | |
7194 | int request; | |
7195 | ||
7196 | if (!attr || attr->relax_domain_level < 0) { | |
7197 | if (default_relax_domain_level < 0) | |
7198 | return; | |
7199 | else | |
7200 | request = default_relax_domain_level; | |
7201 | } else | |
7202 | request = attr->relax_domain_level; | |
7203 | if (request < sd->level) { | |
7204 | /* turn off idle balance on this domain */ | |
c88d5910 | 7205 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7206 | } else { |
7207 | /* turn on idle balance on this domain */ | |
c88d5910 | 7208 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7209 | } |
7210 | } | |
7211 | ||
2109b99e AH |
7212 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7213 | const struct cpumask *cpu_map) | |
7214 | { | |
7215 | switch (what) { | |
7216 | case sa_sched_groups: | |
7217 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
7218 | d->sched_group_nodes = NULL; | |
7219 | case sa_rootdomain: | |
7220 | free_rootdomain(d->rd); /* fall through */ | |
7221 | case sa_tmpmask: | |
7222 | free_cpumask_var(d->tmpmask); /* fall through */ | |
7223 | case sa_send_covered: | |
7224 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
7225 | case sa_this_book_map: |
7226 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
7227 | case sa_this_core_map: |
7228 | free_cpumask_var(d->this_core_map); /* fall through */ | |
7229 | case sa_this_sibling_map: | |
7230 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
7231 | case sa_nodemask: | |
7232 | free_cpumask_var(d->nodemask); /* fall through */ | |
7233 | case sa_sched_group_nodes: | |
d1b55138 | 7234 | #ifdef CONFIG_NUMA |
2109b99e AH |
7235 | kfree(d->sched_group_nodes); /* fall through */ |
7236 | case sa_notcovered: | |
7237 | free_cpumask_var(d->notcovered); /* fall through */ | |
7238 | case sa_covered: | |
7239 | free_cpumask_var(d->covered); /* fall through */ | |
7240 | case sa_domainspan: | |
7241 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 7242 | #endif |
2109b99e AH |
7243 | case sa_none: |
7244 | break; | |
7245 | } | |
7246 | } | |
3404c8d9 | 7247 | |
2109b99e AH |
7248 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7249 | const struct cpumask *cpu_map) | |
7250 | { | |
3404c8d9 | 7251 | #ifdef CONFIG_NUMA |
2109b99e AH |
7252 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
7253 | return sa_none; | |
7254 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
7255 | return sa_domainspan; | |
7256 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
7257 | return sa_covered; | |
7258 | /* Allocate the per-node list of sched groups */ | |
7259 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
7260 | sizeof(struct sched_group *), GFP_KERNEL); | |
7261 | if (!d->sched_group_nodes) { | |
3df0fc5b | 7262 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 7263 | return sa_notcovered; |
d1b55138 | 7264 | } |
2109b99e | 7265 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 7266 | #endif |
2109b99e AH |
7267 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
7268 | return sa_sched_group_nodes; | |
7269 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
7270 | return sa_nodemask; | |
7271 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
7272 | return sa_this_sibling_map; | |
01a08546 | 7273 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 7274 | return sa_this_core_map; |
01a08546 HC |
7275 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
7276 | return sa_this_book_map; | |
2109b99e AH |
7277 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
7278 | return sa_send_covered; | |
7279 | d->rd = alloc_rootdomain(); | |
7280 | if (!d->rd) { | |
3df0fc5b | 7281 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 7282 | return sa_tmpmask; |
57d885fe | 7283 | } |
2109b99e AH |
7284 | return sa_rootdomain; |
7285 | } | |
57d885fe | 7286 | |
7f4588f3 AH |
7287 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
7288 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
7289 | { | |
7290 | struct sched_domain *sd = NULL; | |
7c16ec58 | 7291 | #ifdef CONFIG_NUMA |
7f4588f3 | 7292 | struct sched_domain *parent; |
1da177e4 | 7293 | |
7f4588f3 AH |
7294 | d->sd_allnodes = 0; |
7295 | if (cpumask_weight(cpu_map) > | |
7296 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
7297 | sd = &per_cpu(allnodes_domains, i).sd; | |
7298 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 7299 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
7300 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7301 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7302 | d->sd_allnodes = 1; | |
7303 | } | |
7304 | parent = sd; | |
7305 | ||
7306 | sd = &per_cpu(node_domains, i).sd; | |
7307 | SD_INIT(sd, NODE); | |
7308 | set_domain_attribute(sd, attr); | |
7309 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
7310 | sd->parent = parent; | |
7311 | if (parent) | |
7312 | parent->child = sd; | |
7313 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 7314 | #endif |
7f4588f3 AH |
7315 | return sd; |
7316 | } | |
1da177e4 | 7317 | |
87cce662 AH |
7318 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7319 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7320 | struct sched_domain *parent, int i) | |
7321 | { | |
7322 | struct sched_domain *sd; | |
7323 | sd = &per_cpu(phys_domains, i).sd; | |
7324 | SD_INIT(sd, CPU); | |
7325 | set_domain_attribute(sd, attr); | |
7326 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
7327 | sd->parent = parent; | |
7328 | if (parent) | |
7329 | parent->child = sd; | |
7330 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7331 | return sd; | |
7332 | } | |
1da177e4 | 7333 | |
01a08546 HC |
7334 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
7335 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7336 | struct sched_domain *parent, int i) | |
7337 | { | |
7338 | struct sched_domain *sd = parent; | |
7339 | #ifdef CONFIG_SCHED_BOOK | |
7340 | sd = &per_cpu(book_domains, i).sd; | |
7341 | SD_INIT(sd, BOOK); | |
7342 | set_domain_attribute(sd, attr); | |
7343 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
7344 | sd->parent = parent; | |
7345 | parent->child = sd; | |
7346 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7347 | #endif | |
7348 | return sd; | |
7349 | } | |
7350 | ||
410c4081 AH |
7351 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7352 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7353 | struct sched_domain *parent, int i) | |
7354 | { | |
7355 | struct sched_domain *sd = parent; | |
1e9f28fa | 7356 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7357 | sd = &per_cpu(core_domains, i).sd; |
7358 | SD_INIT(sd, MC); | |
7359 | set_domain_attribute(sd, attr); | |
7360 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7361 | sd->parent = parent; | |
7362 | parent->child = sd; | |
7363 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7364 | #endif |
410c4081 AH |
7365 | return sd; |
7366 | } | |
1e9f28fa | 7367 | |
d8173535 AH |
7368 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7369 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7370 | struct sched_domain *parent, int i) | |
7371 | { | |
7372 | struct sched_domain *sd = parent; | |
1da177e4 | 7373 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7374 | sd = &per_cpu(cpu_domains, i).sd; |
7375 | SD_INIT(sd, SIBLING); | |
7376 | set_domain_attribute(sd, attr); | |
7377 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7378 | sd->parent = parent; | |
7379 | parent->child = sd; | |
7380 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7381 | #endif |
d8173535 AH |
7382 | return sd; |
7383 | } | |
1da177e4 | 7384 | |
0e8e85c9 AH |
7385 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7386 | const struct cpumask *cpu_map, int cpu) | |
7387 | { | |
7388 | switch (l) { | |
1da177e4 | 7389 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7390 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7391 | cpumask_and(d->this_sibling_map, cpu_map, | |
7392 | topology_thread_cpumask(cpu)); | |
7393 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7394 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7395 | &cpu_to_cpu_group, | |
7396 | d->send_covered, d->tmpmask); | |
7397 | break; | |
1da177e4 | 7398 | #endif |
1e9f28fa | 7399 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7400 | case SD_LV_MC: /* set up multi-core groups */ |
7401 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7402 | if (cpu == cpumask_first(d->this_core_map)) | |
7403 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7404 | &cpu_to_core_group, | |
7405 | d->send_covered, d->tmpmask); | |
7406 | break; | |
01a08546 HC |
7407 | #endif |
7408 | #ifdef CONFIG_SCHED_BOOK | |
7409 | case SD_LV_BOOK: /* set up book groups */ | |
7410 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7411 | if (cpu == cpumask_first(d->this_book_map)) | |
7412 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7413 | &cpu_to_book_group, | |
7414 | d->send_covered, d->tmpmask); | |
7415 | break; | |
1e9f28fa | 7416 | #endif |
86548096 AH |
7417 | case SD_LV_CPU: /* set up physical groups */ |
7418 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7419 | if (!cpumask_empty(d->nodemask)) | |
7420 | init_sched_build_groups(d->nodemask, cpu_map, | |
7421 | &cpu_to_phys_group, | |
7422 | d->send_covered, d->tmpmask); | |
7423 | break; | |
1da177e4 | 7424 | #ifdef CONFIG_NUMA |
de616e36 AH |
7425 | case SD_LV_ALLNODES: |
7426 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7427 | d->send_covered, d->tmpmask); | |
7428 | break; | |
7429 | #endif | |
0e8e85c9 AH |
7430 | default: |
7431 | break; | |
7c16ec58 | 7432 | } |
0e8e85c9 | 7433 | } |
9c1cfda2 | 7434 | |
2109b99e AH |
7435 | /* |
7436 | * Build sched domains for a given set of cpus and attach the sched domains | |
7437 | * to the individual cpus | |
7438 | */ | |
7439 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7440 | struct sched_domain_attr *attr) | |
7441 | { | |
7442 | enum s_alloc alloc_state = sa_none; | |
7443 | struct s_data d; | |
294b0c96 | 7444 | struct sched_domain *sd; |
2109b99e | 7445 | int i; |
7c16ec58 | 7446 | #ifdef CONFIG_NUMA |
2109b99e | 7447 | d.sd_allnodes = 0; |
7c16ec58 | 7448 | #endif |
9c1cfda2 | 7449 | |
2109b99e AH |
7450 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7451 | if (alloc_state != sa_rootdomain) | |
7452 | goto error; | |
7453 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7454 | |
1da177e4 | 7455 | /* |
1a20ff27 | 7456 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7457 | */ |
abcd083a | 7458 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7459 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7460 | cpu_map); | |
9761eea8 | 7461 | |
7f4588f3 | 7462 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7463 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7464 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7465 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7466 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7467 | } |
9c1cfda2 | 7468 | |
abcd083a | 7469 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7470 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7471 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7472 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7473 | } |
9c1cfda2 | 7474 | |
1da177e4 | 7475 | /* Set up physical groups */ |
86548096 AH |
7476 | for (i = 0; i < nr_node_ids; i++) |
7477 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7478 | |
1da177e4 LT |
7479 | #ifdef CONFIG_NUMA |
7480 | /* Set up node groups */ | |
de616e36 AH |
7481 | if (d.sd_allnodes) |
7482 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7483 | |
0601a88d AH |
7484 | for (i = 0; i < nr_node_ids; i++) |
7485 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7486 | goto error; |
1da177e4 LT |
7487 | #endif |
7488 | ||
7489 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7490 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7491 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7492 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7493 | init_sched_groups_power(i, sd); |
5c45bf27 | 7494 | } |
1da177e4 | 7495 | #endif |
1e9f28fa | 7496 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7497 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7498 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7499 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7500 | } |
7501 | #endif | |
01a08546 HC |
7502 | #ifdef CONFIG_SCHED_BOOK |
7503 | for_each_cpu(i, cpu_map) { | |
7504 | sd = &per_cpu(book_domains, i).sd; | |
7505 | init_sched_groups_power(i, sd); | |
7506 | } | |
7507 | #endif | |
1e9f28fa | 7508 | |
abcd083a | 7509 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7510 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7511 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7512 | } |
7513 | ||
9c1cfda2 | 7514 | #ifdef CONFIG_NUMA |
076ac2af | 7515 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7516 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7517 | |
49a02c51 | 7518 | if (d.sd_allnodes) { |
6711cab4 | 7519 | struct sched_group *sg; |
f712c0c7 | 7520 | |
96f874e2 | 7521 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7522 | d.tmpmask); |
f712c0c7 SS |
7523 | init_numa_sched_groups_power(sg); |
7524 | } | |
9c1cfda2 JH |
7525 | #endif |
7526 | ||
1da177e4 | 7527 | /* Attach the domains */ |
abcd083a | 7528 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7529 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7530 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7531 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7532 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7533 | #elif defined(CONFIG_SCHED_BOOK) |
7534 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7535 | #else |
6c99e9ad | 7536 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7537 | #endif |
49a02c51 | 7538 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7539 | } |
51888ca2 | 7540 | |
2109b99e AH |
7541 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7542 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7543 | return 0; | |
51888ca2 | 7544 | |
51888ca2 | 7545 | error: |
2109b99e AH |
7546 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7547 | return -ENOMEM; | |
1da177e4 | 7548 | } |
029190c5 | 7549 | |
96f874e2 | 7550 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7551 | { |
7552 | return __build_sched_domains(cpu_map, NULL); | |
7553 | } | |
7554 | ||
acc3f5d7 | 7555 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7556 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7557 | static struct sched_domain_attr *dattr_cur; |
7558 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7559 | |
7560 | /* | |
7561 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7562 | * cpumask) fails, then fallback to a single sched domain, |
7563 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7564 | */ |
4212823f | 7565 | static cpumask_var_t fallback_doms; |
029190c5 | 7566 | |
ee79d1bd HC |
7567 | /* |
7568 | * arch_update_cpu_topology lets virtualized architectures update the | |
7569 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7570 | * or 0 if it stayed the same. | |
7571 | */ | |
7572 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7573 | { |
ee79d1bd | 7574 | return 0; |
22e52b07 HC |
7575 | } |
7576 | ||
acc3f5d7 RR |
7577 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7578 | { | |
7579 | int i; | |
7580 | cpumask_var_t *doms; | |
7581 | ||
7582 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7583 | if (!doms) | |
7584 | return NULL; | |
7585 | for (i = 0; i < ndoms; i++) { | |
7586 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7587 | free_sched_domains(doms, i); | |
7588 | return NULL; | |
7589 | } | |
7590 | } | |
7591 | return doms; | |
7592 | } | |
7593 | ||
7594 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7595 | { | |
7596 | unsigned int i; | |
7597 | for (i = 0; i < ndoms; i++) | |
7598 | free_cpumask_var(doms[i]); | |
7599 | kfree(doms); | |
7600 | } | |
7601 | ||
1a20ff27 | 7602 | /* |
41a2d6cf | 7603 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7604 | * For now this just excludes isolated cpus, but could be used to |
7605 | * exclude other special cases in the future. | |
1a20ff27 | 7606 | */ |
96f874e2 | 7607 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7608 | { |
7378547f MM |
7609 | int err; |
7610 | ||
22e52b07 | 7611 | arch_update_cpu_topology(); |
029190c5 | 7612 | ndoms_cur = 1; |
acc3f5d7 | 7613 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7614 | if (!doms_cur) |
acc3f5d7 RR |
7615 | doms_cur = &fallback_doms; |
7616 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7617 | dattr_cur = NULL; |
acc3f5d7 | 7618 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7619 | register_sched_domain_sysctl(); |
7378547f MM |
7620 | |
7621 | return err; | |
1a20ff27 DG |
7622 | } |
7623 | ||
96f874e2 RR |
7624 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7625 | struct cpumask *tmpmask) | |
1da177e4 | 7626 | { |
7c16ec58 | 7627 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7628 | } |
1da177e4 | 7629 | |
1a20ff27 DG |
7630 | /* |
7631 | * Detach sched domains from a group of cpus specified in cpu_map | |
7632 | * These cpus will now be attached to the NULL domain | |
7633 | */ | |
96f874e2 | 7634 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7635 | { |
96f874e2 RR |
7636 | /* Save because hotplug lock held. */ |
7637 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7638 | int i; |
7639 | ||
abcd083a | 7640 | for_each_cpu(i, cpu_map) |
57d885fe | 7641 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7642 | synchronize_sched(); |
96f874e2 | 7643 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7644 | } |
7645 | ||
1d3504fc HS |
7646 | /* handle null as "default" */ |
7647 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7648 | struct sched_domain_attr *new, int idx_new) | |
7649 | { | |
7650 | struct sched_domain_attr tmp; | |
7651 | ||
7652 | /* fast path */ | |
7653 | if (!new && !cur) | |
7654 | return 1; | |
7655 | ||
7656 | tmp = SD_ATTR_INIT; | |
7657 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7658 | new ? (new + idx_new) : &tmp, | |
7659 | sizeof(struct sched_domain_attr)); | |
7660 | } | |
7661 | ||
029190c5 PJ |
7662 | /* |
7663 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7664 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7665 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7666 | * It destroys each deleted domain and builds each new domain. | |
7667 | * | |
acc3f5d7 | 7668 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7669 | * The masks don't intersect (don't overlap.) We should setup one |
7670 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7671 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7672 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7673 | * it as it is. | |
7674 | * | |
acc3f5d7 RR |
7675 | * The passed in 'doms_new' should be allocated using |
7676 | * alloc_sched_domains. This routine takes ownership of it and will | |
7677 | * free_sched_domains it when done with it. If the caller failed the | |
7678 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7679 | * and partition_sched_domains() will fallback to the single partition | |
7680 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7681 | * |
96f874e2 | 7682 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7683 | * ndoms_new == 0 is a special case for destroying existing domains, |
7684 | * and it will not create the default domain. | |
dfb512ec | 7685 | * |
029190c5 PJ |
7686 | * Call with hotplug lock held |
7687 | */ | |
acc3f5d7 | 7688 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7689 | struct sched_domain_attr *dattr_new) |
029190c5 | 7690 | { |
dfb512ec | 7691 | int i, j, n; |
d65bd5ec | 7692 | int new_topology; |
029190c5 | 7693 | |
712555ee | 7694 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7695 | |
7378547f MM |
7696 | /* always unregister in case we don't destroy any domains */ |
7697 | unregister_sched_domain_sysctl(); | |
7698 | ||
d65bd5ec HC |
7699 | /* Let architecture update cpu core mappings. */ |
7700 | new_topology = arch_update_cpu_topology(); | |
7701 | ||
dfb512ec | 7702 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7703 | |
7704 | /* Destroy deleted domains */ | |
7705 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7706 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7707 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7708 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7709 | goto match1; |
7710 | } | |
7711 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7712 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7713 | match1: |
7714 | ; | |
7715 | } | |
7716 | ||
e761b772 MK |
7717 | if (doms_new == NULL) { |
7718 | ndoms_cur = 0; | |
acc3f5d7 | 7719 | doms_new = &fallback_doms; |
6ad4c188 | 7720 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7721 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7722 | } |
7723 | ||
029190c5 PJ |
7724 | /* Build new domains */ |
7725 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7726 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7727 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7728 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7729 | goto match2; |
7730 | } | |
7731 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7732 | __build_sched_domains(doms_new[i], |
1d3504fc | 7733 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7734 | match2: |
7735 | ; | |
7736 | } | |
7737 | ||
7738 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7739 | if (doms_cur != &fallback_doms) |
7740 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7741 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7742 | doms_cur = doms_new; |
1d3504fc | 7743 | dattr_cur = dattr_new; |
029190c5 | 7744 | ndoms_cur = ndoms_new; |
7378547f MM |
7745 | |
7746 | register_sched_domain_sysctl(); | |
a1835615 | 7747 | |
712555ee | 7748 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7749 | } |
7750 | ||
5c45bf27 | 7751 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7752 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7753 | { |
95402b38 | 7754 | get_online_cpus(); |
dfb512ec MK |
7755 | |
7756 | /* Destroy domains first to force the rebuild */ | |
7757 | partition_sched_domains(0, NULL, NULL); | |
7758 | ||
e761b772 | 7759 | rebuild_sched_domains(); |
95402b38 | 7760 | put_online_cpus(); |
5c45bf27 SS |
7761 | } |
7762 | ||
7763 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7764 | { | |
afb8a9b7 | 7765 | unsigned int level = 0; |
5c45bf27 | 7766 | |
afb8a9b7 GS |
7767 | if (sscanf(buf, "%u", &level) != 1) |
7768 | return -EINVAL; | |
7769 | ||
7770 | /* | |
7771 | * level is always be positive so don't check for | |
7772 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7773 | * What happens on 0 or 1 byte write, | |
7774 | * need to check for count as well? | |
7775 | */ | |
7776 | ||
7777 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7778 | return -EINVAL; |
7779 | ||
7780 | if (smt) | |
afb8a9b7 | 7781 | sched_smt_power_savings = level; |
5c45bf27 | 7782 | else |
afb8a9b7 | 7783 | sched_mc_power_savings = level; |
5c45bf27 | 7784 | |
c70f22d2 | 7785 | arch_reinit_sched_domains(); |
5c45bf27 | 7786 | |
c70f22d2 | 7787 | return count; |
5c45bf27 SS |
7788 | } |
7789 | ||
5c45bf27 | 7790 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7791 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7792 | struct sysdev_class_attribute *attr, |
f718cd4a | 7793 | char *page) |
5c45bf27 SS |
7794 | { |
7795 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7796 | } | |
f718cd4a | 7797 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7798 | struct sysdev_class_attribute *attr, |
48f24c4d | 7799 | const char *buf, size_t count) |
5c45bf27 SS |
7800 | { |
7801 | return sched_power_savings_store(buf, count, 0); | |
7802 | } | |
f718cd4a AK |
7803 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7804 | sched_mc_power_savings_show, | |
7805 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7806 | #endif |
7807 | ||
7808 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7809 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7810 | struct sysdev_class_attribute *attr, |
f718cd4a | 7811 | char *page) |
5c45bf27 SS |
7812 | { |
7813 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7814 | } | |
f718cd4a | 7815 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7816 | struct sysdev_class_attribute *attr, |
48f24c4d | 7817 | const char *buf, size_t count) |
5c45bf27 SS |
7818 | { |
7819 | return sched_power_savings_store(buf, count, 1); | |
7820 | } | |
f718cd4a AK |
7821 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7822 | sched_smt_power_savings_show, | |
6707de00 AB |
7823 | sched_smt_power_savings_store); |
7824 | #endif | |
7825 | ||
39aac648 | 7826 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7827 | { |
7828 | int err = 0; | |
7829 | ||
7830 | #ifdef CONFIG_SCHED_SMT | |
7831 | if (smt_capable()) | |
7832 | err = sysfs_create_file(&cls->kset.kobj, | |
7833 | &attr_sched_smt_power_savings.attr); | |
7834 | #endif | |
7835 | #ifdef CONFIG_SCHED_MC | |
7836 | if (!err && mc_capable()) | |
7837 | err = sysfs_create_file(&cls->kset.kobj, | |
7838 | &attr_sched_mc_power_savings.attr); | |
7839 | #endif | |
7840 | return err; | |
7841 | } | |
6d6bc0ad | 7842 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7843 | |
1da177e4 | 7844 | /* |
3a101d05 TH |
7845 | * Update cpusets according to cpu_active mask. If cpusets are |
7846 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7847 | * around partition_sched_domains(). | |
1da177e4 | 7848 | */ |
0b2e918a TH |
7849 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7850 | void *hcpu) | |
e761b772 | 7851 | { |
3a101d05 | 7852 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7853 | case CPU_ONLINE: |
6ad4c188 | 7854 | case CPU_DOWN_FAILED: |
3a101d05 | 7855 | cpuset_update_active_cpus(); |
e761b772 | 7856 | return NOTIFY_OK; |
3a101d05 TH |
7857 | default: |
7858 | return NOTIFY_DONE; | |
7859 | } | |
7860 | } | |
e761b772 | 7861 | |
0b2e918a TH |
7862 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7863 | void *hcpu) | |
3a101d05 TH |
7864 | { |
7865 | switch (action & ~CPU_TASKS_FROZEN) { | |
7866 | case CPU_DOWN_PREPARE: | |
7867 | cpuset_update_active_cpus(); | |
7868 | return NOTIFY_OK; | |
e761b772 MK |
7869 | default: |
7870 | return NOTIFY_DONE; | |
7871 | } | |
7872 | } | |
e761b772 MK |
7873 | |
7874 | static int update_runtime(struct notifier_block *nfb, | |
7875 | unsigned long action, void *hcpu) | |
1da177e4 | 7876 | { |
7def2be1 PZ |
7877 | int cpu = (int)(long)hcpu; |
7878 | ||
1da177e4 | 7879 | switch (action) { |
1da177e4 | 7880 | case CPU_DOWN_PREPARE: |
8bb78442 | 7881 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7882 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7883 | return NOTIFY_OK; |
7884 | ||
1da177e4 | 7885 | case CPU_DOWN_FAILED: |
8bb78442 | 7886 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7887 | case CPU_ONLINE: |
8bb78442 | 7888 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7889 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7890 | return NOTIFY_OK; |
7891 | ||
1da177e4 LT |
7892 | default: |
7893 | return NOTIFY_DONE; | |
7894 | } | |
1da177e4 | 7895 | } |
1da177e4 LT |
7896 | |
7897 | void __init sched_init_smp(void) | |
7898 | { | |
dcc30a35 RR |
7899 | cpumask_var_t non_isolated_cpus; |
7900 | ||
7901 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7902 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7903 | |
434d53b0 MT |
7904 | #if defined(CONFIG_NUMA) |
7905 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7906 | GFP_KERNEL); | |
7907 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7908 | #endif | |
95402b38 | 7909 | get_online_cpus(); |
712555ee | 7910 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7911 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7912 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7913 | if (cpumask_empty(non_isolated_cpus)) | |
7914 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7915 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7916 | put_online_cpus(); |
e761b772 | 7917 | |
3a101d05 TH |
7918 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7919 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7920 | |
7921 | /* RT runtime code needs to handle some hotplug events */ | |
7922 | hotcpu_notifier(update_runtime, 0); | |
7923 | ||
b328ca18 | 7924 | init_hrtick(); |
5c1e1767 NP |
7925 | |
7926 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7927 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7928 | BUG(); |
19978ca6 | 7929 | sched_init_granularity(); |
dcc30a35 | 7930 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7931 | |
0e3900e6 | 7932 | init_sched_rt_class(); |
1da177e4 LT |
7933 | } |
7934 | #else | |
7935 | void __init sched_init_smp(void) | |
7936 | { | |
19978ca6 | 7937 | sched_init_granularity(); |
1da177e4 LT |
7938 | } |
7939 | #endif /* CONFIG_SMP */ | |
7940 | ||
cd1bb94b AB |
7941 | const_debug unsigned int sysctl_timer_migration = 1; |
7942 | ||
1da177e4 LT |
7943 | int in_sched_functions(unsigned long addr) |
7944 | { | |
1da177e4 LT |
7945 | return in_lock_functions(addr) || |
7946 | (addr >= (unsigned long)__sched_text_start | |
7947 | && addr < (unsigned long)__sched_text_end); | |
7948 | } | |
7949 | ||
a9957449 | 7950 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7951 | { |
7952 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7953 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7954 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7955 | cfs_rq->rq = rq; | |
7956 | #endif | |
67e9fb2a | 7957 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7958 | } |
7959 | ||
fa85ae24 PZ |
7960 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7961 | { | |
7962 | struct rt_prio_array *array; | |
7963 | int i; | |
7964 | ||
7965 | array = &rt_rq->active; | |
7966 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7967 | INIT_LIST_HEAD(array->queue + i); | |
7968 | __clear_bit(i, array->bitmap); | |
7969 | } | |
7970 | /* delimiter for bitsearch: */ | |
7971 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7972 | ||
052f1dc7 | 7973 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7974 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7975 | #ifdef CONFIG_SMP |
e864c499 | 7976 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7977 | #endif |
48d5e258 | 7978 | #endif |
fa85ae24 PZ |
7979 | #ifdef CONFIG_SMP |
7980 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7981 | rt_rq->overloaded = 0; |
05fa785c | 7982 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7983 | #endif |
7984 | ||
7985 | rt_rq->rt_time = 0; | |
7986 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7987 | rt_rq->rt_runtime = 0; |
0986b11b | 7988 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7989 | |
052f1dc7 | 7990 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7991 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7992 | rt_rq->rq = rq; |
7993 | #endif | |
fa85ae24 PZ |
7994 | } |
7995 | ||
6f505b16 | 7996 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7997 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7998 | struct sched_entity *se, int cpu, int add, | |
7999 | struct sched_entity *parent) | |
6f505b16 | 8000 | { |
ec7dc8ac | 8001 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8002 | tg->cfs_rq[cpu] = cfs_rq; |
8003 | init_cfs_rq(cfs_rq, rq); | |
8004 | cfs_rq->tg = tg; | |
8005 | if (add) | |
8006 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8007 | ||
8008 | tg->se[cpu] = se; | |
354d60c2 DG |
8009 | /* se could be NULL for init_task_group */ |
8010 | if (!se) | |
8011 | return; | |
8012 | ||
ec7dc8ac DG |
8013 | if (!parent) |
8014 | se->cfs_rq = &rq->cfs; | |
8015 | else | |
8016 | se->cfs_rq = parent->my_q; | |
8017 | ||
6f505b16 PZ |
8018 | se->my_q = cfs_rq; |
8019 | se->load.weight = tg->shares; | |
e05510d0 | 8020 | se->load.inv_weight = 0; |
ec7dc8ac | 8021 | se->parent = parent; |
6f505b16 | 8022 | } |
052f1dc7 | 8023 | #endif |
6f505b16 | 8024 | |
052f1dc7 | 8025 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8026 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8027 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8028 | struct sched_rt_entity *parent) | |
6f505b16 | 8029 | { |
ec7dc8ac DG |
8030 | struct rq *rq = cpu_rq(cpu); |
8031 | ||
6f505b16 PZ |
8032 | tg->rt_rq[cpu] = rt_rq; |
8033 | init_rt_rq(rt_rq, rq); | |
8034 | rt_rq->tg = tg; | |
ac086bc2 | 8035 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8036 | if (add) |
8037 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8038 | ||
8039 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8040 | if (!rt_se) |
8041 | return; | |
8042 | ||
ec7dc8ac DG |
8043 | if (!parent) |
8044 | rt_se->rt_rq = &rq->rt; | |
8045 | else | |
8046 | rt_se->rt_rq = parent->my_q; | |
8047 | ||
6f505b16 | 8048 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8049 | rt_se->parent = parent; |
6f505b16 PZ |
8050 | INIT_LIST_HEAD(&rt_se->run_list); |
8051 | } | |
8052 | #endif | |
8053 | ||
1da177e4 LT |
8054 | void __init sched_init(void) |
8055 | { | |
dd41f596 | 8056 | int i, j; |
434d53b0 MT |
8057 | unsigned long alloc_size = 0, ptr; |
8058 | ||
8059 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8060 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8061 | #endif | |
8062 | #ifdef CONFIG_RT_GROUP_SCHED | |
8063 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8064 | #endif |
df7c8e84 | 8065 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 8066 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 8067 | #endif |
434d53b0 | 8068 | if (alloc_size) { |
36b7b6d4 | 8069 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
8070 | |
8071 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8072 | init_task_group.se = (struct sched_entity **)ptr; | |
8073 | ptr += nr_cpu_ids * sizeof(void **); | |
8074 | ||
8075 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8076 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8077 | |
6d6bc0ad | 8078 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
8079 | #ifdef CONFIG_RT_GROUP_SCHED |
8080 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8081 | ptr += nr_cpu_ids * sizeof(void **); | |
8082 | ||
8083 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8084 | ptr += nr_cpu_ids * sizeof(void **); |
8085 | ||
6d6bc0ad | 8086 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
8087 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8088 | for_each_possible_cpu(i) { | |
8089 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
8090 | ptr += cpumask_size(); | |
8091 | } | |
8092 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 8093 | } |
dd41f596 | 8094 | |
57d885fe GH |
8095 | #ifdef CONFIG_SMP |
8096 | init_defrootdomain(); | |
8097 | #endif | |
8098 | ||
d0b27fa7 PZ |
8099 | init_rt_bandwidth(&def_rt_bandwidth, |
8100 | global_rt_period(), global_rt_runtime()); | |
8101 | ||
8102 | #ifdef CONFIG_RT_GROUP_SCHED | |
8103 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8104 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 8105 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8106 | |
7c941438 | 8107 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 8108 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8109 | INIT_LIST_HEAD(&init_task_group.children); |
8110 | ||
7c941438 | 8111 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 8112 | |
4a6cc4bd JK |
8113 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
8114 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
8115 | __alignof__(unsigned long)); | |
8116 | #endif | |
0a945022 | 8117 | for_each_possible_cpu(i) { |
70b97a7f | 8118 | struct rq *rq; |
1da177e4 LT |
8119 | |
8120 | rq = cpu_rq(i); | |
05fa785c | 8121 | raw_spin_lock_init(&rq->lock); |
7897986b | 8122 | rq->nr_running = 0; |
dce48a84 TG |
8123 | rq->calc_load_active = 0; |
8124 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 8125 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8126 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8127 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8128 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8129 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8130 | #ifdef CONFIG_CGROUP_SCHED |
8131 | /* | |
8132 | * How much cpu bandwidth does init_task_group get? | |
8133 | * | |
8134 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8135 | * gets 100% of the cpu resources in the system. This overall | |
8136 | * system cpu resource is divided among the tasks of | |
8137 | * init_task_group and its child task-groups in a fair manner, | |
8138 | * based on each entity's (task or task-group's) weight | |
8139 | * (se->load.weight). | |
8140 | * | |
8141 | * In other words, if init_task_group has 10 tasks of weight | |
8142 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8143 | * then A0's share of the cpu resource is: | |
8144 | * | |
0d905bca | 8145 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
8146 | * |
8147 | * We achieve this by letting init_task_group's tasks sit | |
8148 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8149 | */ | |
ec7dc8ac | 8150 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 8151 | #endif |
354d60c2 DG |
8152 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8153 | ||
8154 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8155 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8156 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8157 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8158 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8159 | #endif |
dd41f596 | 8160 | #endif |
1da177e4 | 8161 | |
dd41f596 IM |
8162 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8163 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
8164 | |
8165 | rq->last_load_update_tick = jiffies; | |
8166 | ||
1da177e4 | 8167 | #ifdef CONFIG_SMP |
41c7ce9a | 8168 | rq->sd = NULL; |
57d885fe | 8169 | rq->rd = NULL; |
e51fd5e2 | 8170 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 8171 | rq->post_schedule = 0; |
1da177e4 | 8172 | rq->active_balance = 0; |
dd41f596 | 8173 | rq->next_balance = jiffies; |
1da177e4 | 8174 | rq->push_cpu = 0; |
0a2966b4 | 8175 | rq->cpu = i; |
1f11eb6a | 8176 | rq->online = 0; |
eae0c9df MG |
8177 | rq->idle_stamp = 0; |
8178 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 8179 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
8180 | #ifdef CONFIG_NO_HZ |
8181 | rq->nohz_balance_kick = 0; | |
8182 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
8183 | #endif | |
1da177e4 | 8184 | #endif |
8f4d37ec | 8185 | init_rq_hrtick(rq); |
1da177e4 | 8186 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8187 | } |
8188 | ||
2dd73a4f | 8189 | set_load_weight(&init_task); |
b50f60ce | 8190 | |
e107be36 AK |
8191 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8192 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8193 | #endif | |
8194 | ||
c9819f45 | 8195 | #ifdef CONFIG_SMP |
962cf36c | 8196 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8197 | #endif |
8198 | ||
b50f60ce | 8199 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 8200 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
8201 | #endif |
8202 | ||
1da177e4 LT |
8203 | /* |
8204 | * The boot idle thread does lazy MMU switching as well: | |
8205 | */ | |
8206 | atomic_inc(&init_mm.mm_count); | |
8207 | enter_lazy_tlb(&init_mm, current); | |
8208 | ||
8209 | /* | |
8210 | * Make us the idle thread. Technically, schedule() should not be | |
8211 | * called from this thread, however somewhere below it might be, | |
8212 | * but because we are the idle thread, we just pick up running again | |
8213 | * when this runqueue becomes "idle". | |
8214 | */ | |
8215 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8216 | |
8217 | calc_load_update = jiffies + LOAD_FREQ; | |
8218 | ||
dd41f596 IM |
8219 | /* |
8220 | * During early bootup we pretend to be a normal task: | |
8221 | */ | |
8222 | current->sched_class = &fair_sched_class; | |
6892b75e | 8223 | |
6a7b3dc3 | 8224 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8225 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8226 | #ifdef CONFIG_SMP |
7d1e6a9b | 8227 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8228 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8229 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8230 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8231 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8232 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8233 | #endif |
bdddd296 RR |
8234 | /* May be allocated at isolcpus cmdline parse time */ |
8235 | if (cpu_isolated_map == NULL) | |
8236 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8237 | #endif /* SMP */ |
6a7b3dc3 | 8238 | |
cdd6c482 | 8239 | perf_event_init(); |
0d905bca | 8240 | |
6892b75e | 8241 | scheduler_running = 1; |
1da177e4 LT |
8242 | } |
8243 | ||
8244 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
8245 | static inline int preempt_count_equals(int preempt_offset) |
8246 | { | |
234da7bc | 8247 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
8248 | |
8249 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
8250 | } | |
8251 | ||
d894837f | 8252 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8253 | { |
48f24c4d | 8254 | #ifdef in_atomic |
1da177e4 LT |
8255 | static unsigned long prev_jiffy; /* ratelimiting */ |
8256 | ||
e4aafea2 FW |
8257 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8258 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8259 | return; |
8260 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8261 | return; | |
8262 | prev_jiffy = jiffies; | |
8263 | ||
3df0fc5b PZ |
8264 | printk(KERN_ERR |
8265 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8266 | file, line); | |
8267 | printk(KERN_ERR | |
8268 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8269 | in_atomic(), irqs_disabled(), | |
8270 | current->pid, current->comm); | |
aef745fc IM |
8271 | |
8272 | debug_show_held_locks(current); | |
8273 | if (irqs_disabled()) | |
8274 | print_irqtrace_events(current); | |
8275 | dump_stack(); | |
1da177e4 LT |
8276 | #endif |
8277 | } | |
8278 | EXPORT_SYMBOL(__might_sleep); | |
8279 | #endif | |
8280 | ||
8281 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8282 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8283 | { | |
8284 | int on_rq; | |
3e51f33f | 8285 | |
3a5e4dc1 AK |
8286 | on_rq = p->se.on_rq; |
8287 | if (on_rq) | |
8288 | deactivate_task(rq, p, 0); | |
8289 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8290 | if (on_rq) { | |
8291 | activate_task(rq, p, 0); | |
8292 | resched_task(rq->curr); | |
8293 | } | |
8294 | } | |
8295 | ||
1da177e4 LT |
8296 | void normalize_rt_tasks(void) |
8297 | { | |
a0f98a1c | 8298 | struct task_struct *g, *p; |
1da177e4 | 8299 | unsigned long flags; |
70b97a7f | 8300 | struct rq *rq; |
1da177e4 | 8301 | |
4cf5d77a | 8302 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8303 | do_each_thread(g, p) { |
178be793 IM |
8304 | /* |
8305 | * Only normalize user tasks: | |
8306 | */ | |
8307 | if (!p->mm) | |
8308 | continue; | |
8309 | ||
6cfb0d5d | 8310 | p->se.exec_start = 0; |
6cfb0d5d | 8311 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8312 | p->se.statistics.wait_start = 0; |
8313 | p->se.statistics.sleep_start = 0; | |
8314 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8315 | #endif |
dd41f596 IM |
8316 | |
8317 | if (!rt_task(p)) { | |
8318 | /* | |
8319 | * Renice negative nice level userspace | |
8320 | * tasks back to 0: | |
8321 | */ | |
8322 | if (TASK_NICE(p) < 0 && p->mm) | |
8323 | set_user_nice(p, 0); | |
1da177e4 | 8324 | continue; |
dd41f596 | 8325 | } |
1da177e4 | 8326 | |
1d615482 | 8327 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8328 | rq = __task_rq_lock(p); |
1da177e4 | 8329 | |
178be793 | 8330 | normalize_task(rq, p); |
3a5e4dc1 | 8331 | |
b29739f9 | 8332 | __task_rq_unlock(rq); |
1d615482 | 8333 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8334 | } while_each_thread(g, p); |
8335 | ||
4cf5d77a | 8336 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8337 | } |
8338 | ||
8339 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8340 | |
67fc4e0c | 8341 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8342 | /* |
67fc4e0c | 8343 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8344 | * |
8345 | * They can only be called when the whole system has been | |
8346 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8347 | * activity can take place. Using them for anything else would | |
8348 | * be a serious bug, and as a result, they aren't even visible | |
8349 | * under any other configuration. | |
8350 | */ | |
8351 | ||
8352 | /** | |
8353 | * curr_task - return the current task for a given cpu. | |
8354 | * @cpu: the processor in question. | |
8355 | * | |
8356 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8357 | */ | |
36c8b586 | 8358 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8359 | { |
8360 | return cpu_curr(cpu); | |
8361 | } | |
8362 | ||
67fc4e0c JW |
8363 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8364 | ||
8365 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8366 | /** |
8367 | * set_curr_task - set the current task for a given cpu. | |
8368 | * @cpu: the processor in question. | |
8369 | * @p: the task pointer to set. | |
8370 | * | |
8371 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8372 | * are serviced on a separate stack. It allows the architecture to switch the |
8373 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8374 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8375 | * and caller must save the original value of the current task (see | |
8376 | * curr_task() above) and restore that value before reenabling interrupts and | |
8377 | * re-starting the system. | |
8378 | * | |
8379 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8380 | */ | |
36c8b586 | 8381 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8382 | { |
8383 | cpu_curr(cpu) = p; | |
8384 | } | |
8385 | ||
8386 | #endif | |
29f59db3 | 8387 | |
bccbe08a PZ |
8388 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8389 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8390 | { |
8391 | int i; | |
8392 | ||
8393 | for_each_possible_cpu(i) { | |
8394 | if (tg->cfs_rq) | |
8395 | kfree(tg->cfs_rq[i]); | |
8396 | if (tg->se) | |
8397 | kfree(tg->se[i]); | |
6f505b16 PZ |
8398 | } |
8399 | ||
8400 | kfree(tg->cfs_rq); | |
8401 | kfree(tg->se); | |
6f505b16 PZ |
8402 | } |
8403 | ||
ec7dc8ac DG |
8404 | static |
8405 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8406 | { |
29f59db3 | 8407 | struct cfs_rq *cfs_rq; |
eab17229 | 8408 | struct sched_entity *se; |
9b5b7751 | 8409 | struct rq *rq; |
29f59db3 SV |
8410 | int i; |
8411 | ||
434d53b0 | 8412 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8413 | if (!tg->cfs_rq) |
8414 | goto err; | |
434d53b0 | 8415 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8416 | if (!tg->se) |
8417 | goto err; | |
052f1dc7 PZ |
8418 | |
8419 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8420 | |
8421 | for_each_possible_cpu(i) { | |
9b5b7751 | 8422 | rq = cpu_rq(i); |
29f59db3 | 8423 | |
eab17229 LZ |
8424 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8425 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8426 | if (!cfs_rq) |
8427 | goto err; | |
8428 | ||
eab17229 LZ |
8429 | se = kzalloc_node(sizeof(struct sched_entity), |
8430 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8431 | if (!se) |
dfc12eb2 | 8432 | goto err_free_rq; |
29f59db3 | 8433 | |
eab17229 | 8434 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
8435 | } |
8436 | ||
8437 | return 1; | |
8438 | ||
49246274 | 8439 | err_free_rq: |
dfc12eb2 | 8440 | kfree(cfs_rq); |
49246274 | 8441 | err: |
bccbe08a PZ |
8442 | return 0; |
8443 | } | |
8444 | ||
8445 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8446 | { | |
8447 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8448 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8449 | } | |
8450 | ||
8451 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8452 | { | |
8453 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8454 | } | |
6d6bc0ad | 8455 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8456 | static inline void free_fair_sched_group(struct task_group *tg) |
8457 | { | |
8458 | } | |
8459 | ||
ec7dc8ac DG |
8460 | static inline |
8461 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8462 | { |
8463 | return 1; | |
8464 | } | |
8465 | ||
8466 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8467 | { | |
8468 | } | |
8469 | ||
8470 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8471 | { | |
8472 | } | |
6d6bc0ad | 8473 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8474 | |
8475 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8476 | static void free_rt_sched_group(struct task_group *tg) |
8477 | { | |
8478 | int i; | |
8479 | ||
d0b27fa7 PZ |
8480 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8481 | ||
bccbe08a PZ |
8482 | for_each_possible_cpu(i) { |
8483 | if (tg->rt_rq) | |
8484 | kfree(tg->rt_rq[i]); | |
8485 | if (tg->rt_se) | |
8486 | kfree(tg->rt_se[i]); | |
8487 | } | |
8488 | ||
8489 | kfree(tg->rt_rq); | |
8490 | kfree(tg->rt_se); | |
8491 | } | |
8492 | ||
ec7dc8ac DG |
8493 | static |
8494 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8495 | { |
8496 | struct rt_rq *rt_rq; | |
eab17229 | 8497 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8498 | struct rq *rq; |
8499 | int i; | |
8500 | ||
434d53b0 | 8501 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8502 | if (!tg->rt_rq) |
8503 | goto err; | |
434d53b0 | 8504 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8505 | if (!tg->rt_se) |
8506 | goto err; | |
8507 | ||
d0b27fa7 PZ |
8508 | init_rt_bandwidth(&tg->rt_bandwidth, |
8509 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8510 | |
8511 | for_each_possible_cpu(i) { | |
8512 | rq = cpu_rq(i); | |
8513 | ||
eab17229 LZ |
8514 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8515 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8516 | if (!rt_rq) |
8517 | goto err; | |
29f59db3 | 8518 | |
eab17229 LZ |
8519 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8520 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8521 | if (!rt_se) |
dfc12eb2 | 8522 | goto err_free_rq; |
29f59db3 | 8523 | |
eab17229 | 8524 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8525 | } |
8526 | ||
bccbe08a PZ |
8527 | return 1; |
8528 | ||
49246274 | 8529 | err_free_rq: |
dfc12eb2 | 8530 | kfree(rt_rq); |
49246274 | 8531 | err: |
bccbe08a PZ |
8532 | return 0; |
8533 | } | |
8534 | ||
8535 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8536 | { | |
8537 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8538 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8539 | } | |
8540 | ||
8541 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8542 | { | |
8543 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8544 | } | |
6d6bc0ad | 8545 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8546 | static inline void free_rt_sched_group(struct task_group *tg) |
8547 | { | |
8548 | } | |
8549 | ||
ec7dc8ac DG |
8550 | static inline |
8551 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8552 | { |
8553 | return 1; | |
8554 | } | |
8555 | ||
8556 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8557 | { | |
8558 | } | |
8559 | ||
8560 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8561 | { | |
8562 | } | |
6d6bc0ad | 8563 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8564 | |
7c941438 | 8565 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8566 | static void free_sched_group(struct task_group *tg) |
8567 | { | |
8568 | free_fair_sched_group(tg); | |
8569 | free_rt_sched_group(tg); | |
8570 | kfree(tg); | |
8571 | } | |
8572 | ||
8573 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8574 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8575 | { |
8576 | struct task_group *tg; | |
8577 | unsigned long flags; | |
8578 | int i; | |
8579 | ||
8580 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8581 | if (!tg) | |
8582 | return ERR_PTR(-ENOMEM); | |
8583 | ||
ec7dc8ac | 8584 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8585 | goto err; |
8586 | ||
ec7dc8ac | 8587 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8588 | goto err; |
8589 | ||
8ed36996 | 8590 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8591 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8592 | register_fair_sched_group(tg, i); |
8593 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8594 | } |
6f505b16 | 8595 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8596 | |
8597 | WARN_ON(!parent); /* root should already exist */ | |
8598 | ||
8599 | tg->parent = parent; | |
f473aa5e | 8600 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8601 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8602 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8603 | |
9b5b7751 | 8604 | return tg; |
29f59db3 SV |
8605 | |
8606 | err: | |
6f505b16 | 8607 | free_sched_group(tg); |
29f59db3 SV |
8608 | return ERR_PTR(-ENOMEM); |
8609 | } | |
8610 | ||
9b5b7751 | 8611 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8612 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8613 | { |
29f59db3 | 8614 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8615 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8616 | } |
8617 | ||
9b5b7751 | 8618 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8619 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8620 | { |
8ed36996 | 8621 | unsigned long flags; |
9b5b7751 | 8622 | int i; |
29f59db3 | 8623 | |
8ed36996 | 8624 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8625 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8626 | unregister_fair_sched_group(tg, i); |
8627 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8628 | } |
6f505b16 | 8629 | list_del_rcu(&tg->list); |
f473aa5e | 8630 | list_del_rcu(&tg->siblings); |
8ed36996 | 8631 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8632 | |
9b5b7751 | 8633 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8634 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8635 | } |
8636 | ||
9b5b7751 | 8637 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8638 | * The caller of this function should have put the task in its new group |
8639 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8640 | * reflect its new group. | |
9b5b7751 SV |
8641 | */ |
8642 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8643 | { |
8644 | int on_rq, running; | |
8645 | unsigned long flags; | |
8646 | struct rq *rq; | |
8647 | ||
8648 | rq = task_rq_lock(tsk, &flags); | |
8649 | ||
051a1d1a | 8650 | running = task_current(rq, tsk); |
29f59db3 SV |
8651 | on_rq = tsk->se.on_rq; |
8652 | ||
0e1f3483 | 8653 | if (on_rq) |
29f59db3 | 8654 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8655 | if (unlikely(running)) |
8656 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8657 | |
810b3817 | 8658 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8659 | if (tsk->sched_class->task_move_group) |
8660 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8661 | else | |
810b3817 | 8662 | #endif |
b2b5ce02 | 8663 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8664 | |
0e1f3483 HS |
8665 | if (unlikely(running)) |
8666 | tsk->sched_class->set_curr_task(rq); | |
8667 | if (on_rq) | |
371fd7e7 | 8668 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8669 | |
29f59db3 SV |
8670 | task_rq_unlock(rq, &flags); |
8671 | } | |
7c941438 | 8672 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8673 | |
052f1dc7 | 8674 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8675 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8676 | { |
8677 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8678 | int on_rq; |
8679 | ||
29f59db3 | 8680 | on_rq = se->on_rq; |
62fb1851 | 8681 | if (on_rq) |
29f59db3 SV |
8682 | dequeue_entity(cfs_rq, se, 0); |
8683 | ||
8684 | se->load.weight = shares; | |
e05510d0 | 8685 | se->load.inv_weight = 0; |
29f59db3 | 8686 | |
62fb1851 | 8687 | if (on_rq) |
29f59db3 | 8688 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8689 | } |
62fb1851 | 8690 | |
c09595f6 PZ |
8691 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8692 | { | |
8693 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8694 | struct rq *rq = cfs_rq->rq; | |
8695 | unsigned long flags; | |
8696 | ||
05fa785c | 8697 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8698 | __set_se_shares(se, shares); |
05fa785c | 8699 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8700 | } |
8701 | ||
8ed36996 PZ |
8702 | static DEFINE_MUTEX(shares_mutex); |
8703 | ||
4cf86d77 | 8704 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8705 | { |
8706 | int i; | |
8ed36996 | 8707 | unsigned long flags; |
c61935fd | 8708 | |
ec7dc8ac DG |
8709 | /* |
8710 | * We can't change the weight of the root cgroup. | |
8711 | */ | |
8712 | if (!tg->se[0]) | |
8713 | return -EINVAL; | |
8714 | ||
18d95a28 PZ |
8715 | if (shares < MIN_SHARES) |
8716 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8717 | else if (shares > MAX_SHARES) |
8718 | shares = MAX_SHARES; | |
62fb1851 | 8719 | |
8ed36996 | 8720 | mutex_lock(&shares_mutex); |
9b5b7751 | 8721 | if (tg->shares == shares) |
5cb350ba | 8722 | goto done; |
29f59db3 | 8723 | |
8ed36996 | 8724 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8725 | for_each_possible_cpu(i) |
8726 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8727 | list_del_rcu(&tg->siblings); |
8ed36996 | 8728 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8729 | |
8730 | /* wait for any ongoing reference to this group to finish */ | |
8731 | synchronize_sched(); | |
8732 | ||
8733 | /* | |
8734 | * Now we are free to modify the group's share on each cpu | |
8735 | * w/o tripping rebalance_share or load_balance_fair. | |
8736 | */ | |
9b5b7751 | 8737 | tg->shares = shares; |
c09595f6 PZ |
8738 | for_each_possible_cpu(i) { |
8739 | /* | |
8740 | * force a rebalance | |
8741 | */ | |
8742 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8743 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8744 | } |
29f59db3 | 8745 | |
6b2d7700 SV |
8746 | /* |
8747 | * Enable load balance activity on this group, by inserting it back on | |
8748 | * each cpu's rq->leaf_cfs_rq_list. | |
8749 | */ | |
8ed36996 | 8750 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8751 | for_each_possible_cpu(i) |
8752 | register_fair_sched_group(tg, i); | |
f473aa5e | 8753 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8754 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8755 | done: |
8ed36996 | 8756 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8757 | return 0; |
29f59db3 SV |
8758 | } |
8759 | ||
5cb350ba DG |
8760 | unsigned long sched_group_shares(struct task_group *tg) |
8761 | { | |
8762 | return tg->shares; | |
8763 | } | |
052f1dc7 | 8764 | #endif |
5cb350ba | 8765 | |
052f1dc7 | 8766 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8767 | /* |
9f0c1e56 | 8768 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8769 | */ |
9f0c1e56 PZ |
8770 | static DEFINE_MUTEX(rt_constraints_mutex); |
8771 | ||
8772 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8773 | { | |
8774 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8775 | return 1ULL << 20; |
9f0c1e56 | 8776 | |
9a7e0b18 | 8777 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8778 | } |
8779 | ||
9a7e0b18 PZ |
8780 | /* Must be called with tasklist_lock held */ |
8781 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8782 | { |
9a7e0b18 | 8783 | struct task_struct *g, *p; |
b40b2e8e | 8784 | |
9a7e0b18 PZ |
8785 | do_each_thread(g, p) { |
8786 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8787 | return 1; | |
8788 | } while_each_thread(g, p); | |
b40b2e8e | 8789 | |
9a7e0b18 PZ |
8790 | return 0; |
8791 | } | |
b40b2e8e | 8792 | |
9a7e0b18 PZ |
8793 | struct rt_schedulable_data { |
8794 | struct task_group *tg; | |
8795 | u64 rt_period; | |
8796 | u64 rt_runtime; | |
8797 | }; | |
b40b2e8e | 8798 | |
9a7e0b18 PZ |
8799 | static int tg_schedulable(struct task_group *tg, void *data) |
8800 | { | |
8801 | struct rt_schedulable_data *d = data; | |
8802 | struct task_group *child; | |
8803 | unsigned long total, sum = 0; | |
8804 | u64 period, runtime; | |
b40b2e8e | 8805 | |
9a7e0b18 PZ |
8806 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8807 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8808 | |
9a7e0b18 PZ |
8809 | if (tg == d->tg) { |
8810 | period = d->rt_period; | |
8811 | runtime = d->rt_runtime; | |
b40b2e8e | 8812 | } |
b40b2e8e | 8813 | |
4653f803 PZ |
8814 | /* |
8815 | * Cannot have more runtime than the period. | |
8816 | */ | |
8817 | if (runtime > period && runtime != RUNTIME_INF) | |
8818 | return -EINVAL; | |
6f505b16 | 8819 | |
4653f803 PZ |
8820 | /* |
8821 | * Ensure we don't starve existing RT tasks. | |
8822 | */ | |
9a7e0b18 PZ |
8823 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8824 | return -EBUSY; | |
6f505b16 | 8825 | |
9a7e0b18 | 8826 | total = to_ratio(period, runtime); |
6f505b16 | 8827 | |
4653f803 PZ |
8828 | /* |
8829 | * Nobody can have more than the global setting allows. | |
8830 | */ | |
8831 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8832 | return -EINVAL; | |
6f505b16 | 8833 | |
4653f803 PZ |
8834 | /* |
8835 | * The sum of our children's runtime should not exceed our own. | |
8836 | */ | |
9a7e0b18 PZ |
8837 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8838 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8839 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8840 | |
9a7e0b18 PZ |
8841 | if (child == d->tg) { |
8842 | period = d->rt_period; | |
8843 | runtime = d->rt_runtime; | |
8844 | } | |
6f505b16 | 8845 | |
9a7e0b18 | 8846 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8847 | } |
6f505b16 | 8848 | |
9a7e0b18 PZ |
8849 | if (sum > total) |
8850 | return -EINVAL; | |
8851 | ||
8852 | return 0; | |
6f505b16 PZ |
8853 | } |
8854 | ||
9a7e0b18 | 8855 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8856 | { |
9a7e0b18 PZ |
8857 | struct rt_schedulable_data data = { |
8858 | .tg = tg, | |
8859 | .rt_period = period, | |
8860 | .rt_runtime = runtime, | |
8861 | }; | |
8862 | ||
8863 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8864 | } |
8865 | ||
d0b27fa7 PZ |
8866 | static int tg_set_bandwidth(struct task_group *tg, |
8867 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8868 | { |
ac086bc2 | 8869 | int i, err = 0; |
9f0c1e56 | 8870 | |
9f0c1e56 | 8871 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8872 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8873 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8874 | if (err) | |
9f0c1e56 | 8875 | goto unlock; |
ac086bc2 | 8876 | |
0986b11b | 8877 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8878 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8879 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8880 | |
8881 | for_each_possible_cpu(i) { | |
8882 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8883 | ||
0986b11b | 8884 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8885 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8886 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8887 | } |
0986b11b | 8888 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8889 | unlock: |
521f1a24 | 8890 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8891 | mutex_unlock(&rt_constraints_mutex); |
8892 | ||
8893 | return err; | |
6f505b16 PZ |
8894 | } |
8895 | ||
d0b27fa7 PZ |
8896 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8897 | { | |
8898 | u64 rt_runtime, rt_period; | |
8899 | ||
8900 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8901 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8902 | if (rt_runtime_us < 0) | |
8903 | rt_runtime = RUNTIME_INF; | |
8904 | ||
8905 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8906 | } | |
8907 | ||
9f0c1e56 PZ |
8908 | long sched_group_rt_runtime(struct task_group *tg) |
8909 | { | |
8910 | u64 rt_runtime_us; | |
8911 | ||
d0b27fa7 | 8912 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8913 | return -1; |
8914 | ||
d0b27fa7 | 8915 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8916 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8917 | return rt_runtime_us; | |
8918 | } | |
d0b27fa7 PZ |
8919 | |
8920 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8921 | { | |
8922 | u64 rt_runtime, rt_period; | |
8923 | ||
8924 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8925 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8926 | ||
619b0488 R |
8927 | if (rt_period == 0) |
8928 | return -EINVAL; | |
8929 | ||
d0b27fa7 PZ |
8930 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8931 | } | |
8932 | ||
8933 | long sched_group_rt_period(struct task_group *tg) | |
8934 | { | |
8935 | u64 rt_period_us; | |
8936 | ||
8937 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8938 | do_div(rt_period_us, NSEC_PER_USEC); | |
8939 | return rt_period_us; | |
8940 | } | |
8941 | ||
8942 | static int sched_rt_global_constraints(void) | |
8943 | { | |
4653f803 | 8944 | u64 runtime, period; |
d0b27fa7 PZ |
8945 | int ret = 0; |
8946 | ||
ec5d4989 HS |
8947 | if (sysctl_sched_rt_period <= 0) |
8948 | return -EINVAL; | |
8949 | ||
4653f803 PZ |
8950 | runtime = global_rt_runtime(); |
8951 | period = global_rt_period(); | |
8952 | ||
8953 | /* | |
8954 | * Sanity check on the sysctl variables. | |
8955 | */ | |
8956 | if (runtime > period && runtime != RUNTIME_INF) | |
8957 | return -EINVAL; | |
10b612f4 | 8958 | |
d0b27fa7 | 8959 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8960 | read_lock(&tasklist_lock); |
4653f803 | 8961 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8962 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8963 | mutex_unlock(&rt_constraints_mutex); |
8964 | ||
8965 | return ret; | |
8966 | } | |
54e99124 DG |
8967 | |
8968 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8969 | { | |
8970 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8971 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8972 | return 0; | |
8973 | ||
8974 | return 1; | |
8975 | } | |
8976 | ||
6d6bc0ad | 8977 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8978 | static int sched_rt_global_constraints(void) |
8979 | { | |
ac086bc2 PZ |
8980 | unsigned long flags; |
8981 | int i; | |
8982 | ||
ec5d4989 HS |
8983 | if (sysctl_sched_rt_period <= 0) |
8984 | return -EINVAL; | |
8985 | ||
60aa605d PZ |
8986 | /* |
8987 | * There's always some RT tasks in the root group | |
8988 | * -- migration, kstopmachine etc.. | |
8989 | */ | |
8990 | if (sysctl_sched_rt_runtime == 0) | |
8991 | return -EBUSY; | |
8992 | ||
0986b11b | 8993 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8994 | for_each_possible_cpu(i) { |
8995 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8996 | ||
0986b11b | 8997 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8998 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8999 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 9000 | } |
0986b11b | 9001 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 9002 | |
d0b27fa7 PZ |
9003 | return 0; |
9004 | } | |
6d6bc0ad | 9005 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9006 | |
9007 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 9008 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
9009 | loff_t *ppos) |
9010 | { | |
9011 | int ret; | |
9012 | int old_period, old_runtime; | |
9013 | static DEFINE_MUTEX(mutex); | |
9014 | ||
9015 | mutex_lock(&mutex); | |
9016 | old_period = sysctl_sched_rt_period; | |
9017 | old_runtime = sysctl_sched_rt_runtime; | |
9018 | ||
8d65af78 | 9019 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
9020 | |
9021 | if (!ret && write) { | |
9022 | ret = sched_rt_global_constraints(); | |
9023 | if (ret) { | |
9024 | sysctl_sched_rt_period = old_period; | |
9025 | sysctl_sched_rt_runtime = old_runtime; | |
9026 | } else { | |
9027 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9028 | def_rt_bandwidth.rt_period = | |
9029 | ns_to_ktime(global_rt_period()); | |
9030 | } | |
9031 | } | |
9032 | mutex_unlock(&mutex); | |
9033 | ||
9034 | return ret; | |
9035 | } | |
68318b8e | 9036 | |
052f1dc7 | 9037 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9038 | |
9039 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9040 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9041 | { |
2b01dfe3 PM |
9042 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9043 | struct task_group, css); | |
68318b8e SV |
9044 | } |
9045 | ||
9046 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9047 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9048 | { |
ec7dc8ac | 9049 | struct task_group *tg, *parent; |
68318b8e | 9050 | |
2b01dfe3 | 9051 | if (!cgrp->parent) { |
68318b8e | 9052 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9053 | return &init_task_group.css; |
9054 | } | |
9055 | ||
ec7dc8ac DG |
9056 | parent = cgroup_tg(cgrp->parent); |
9057 | tg = sched_create_group(parent); | |
68318b8e SV |
9058 | if (IS_ERR(tg)) |
9059 | return ERR_PTR(-ENOMEM); | |
9060 | ||
68318b8e SV |
9061 | return &tg->css; |
9062 | } | |
9063 | ||
41a2d6cf IM |
9064 | static void |
9065 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9066 | { |
2b01dfe3 | 9067 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9068 | |
9069 | sched_destroy_group(tg); | |
9070 | } | |
9071 | ||
41a2d6cf | 9072 | static int |
be367d09 | 9073 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 9074 | { |
b68aa230 | 9075 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9076 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9077 | return -EINVAL; |
9078 | #else | |
68318b8e SV |
9079 | /* We don't support RT-tasks being in separate groups */ |
9080 | if (tsk->sched_class != &fair_sched_class) | |
9081 | return -EINVAL; | |
b68aa230 | 9082 | #endif |
be367d09 BB |
9083 | return 0; |
9084 | } | |
68318b8e | 9085 | |
be367d09 BB |
9086 | static int |
9087 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9088 | struct task_struct *tsk, bool threadgroup) | |
9089 | { | |
9090 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
9091 | if (retval) | |
9092 | return retval; | |
9093 | if (threadgroup) { | |
9094 | struct task_struct *c; | |
9095 | rcu_read_lock(); | |
9096 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9097 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
9098 | if (retval) { | |
9099 | rcu_read_unlock(); | |
9100 | return retval; | |
9101 | } | |
9102 | } | |
9103 | rcu_read_unlock(); | |
9104 | } | |
68318b8e SV |
9105 | return 0; |
9106 | } | |
9107 | ||
9108 | static void | |
2b01dfe3 | 9109 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
9110 | struct cgroup *old_cont, struct task_struct *tsk, |
9111 | bool threadgroup) | |
68318b8e SV |
9112 | { |
9113 | sched_move_task(tsk); | |
be367d09 BB |
9114 | if (threadgroup) { |
9115 | struct task_struct *c; | |
9116 | rcu_read_lock(); | |
9117 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9118 | sched_move_task(c); | |
9119 | } | |
9120 | rcu_read_unlock(); | |
9121 | } | |
68318b8e SV |
9122 | } |
9123 | ||
052f1dc7 | 9124 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9125 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9126 | u64 shareval) |
68318b8e | 9127 | { |
2b01dfe3 | 9128 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9129 | } |
9130 | ||
f4c753b7 | 9131 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9132 | { |
2b01dfe3 | 9133 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9134 | |
9135 | return (u64) tg->shares; | |
9136 | } | |
6d6bc0ad | 9137 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9138 | |
052f1dc7 | 9139 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9140 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9141 | s64 val) |
6f505b16 | 9142 | { |
06ecb27c | 9143 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9144 | } |
9145 | ||
06ecb27c | 9146 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9147 | { |
06ecb27c | 9148 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9149 | } |
d0b27fa7 PZ |
9150 | |
9151 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9152 | u64 rt_period_us) | |
9153 | { | |
9154 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9155 | } | |
9156 | ||
9157 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9158 | { | |
9159 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9160 | } | |
6d6bc0ad | 9161 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9162 | |
fe5c7cc2 | 9163 | static struct cftype cpu_files[] = { |
052f1dc7 | 9164 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9165 | { |
9166 | .name = "shares", | |
f4c753b7 PM |
9167 | .read_u64 = cpu_shares_read_u64, |
9168 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9169 | }, |
052f1dc7 PZ |
9170 | #endif |
9171 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9172 | { |
9f0c1e56 | 9173 | .name = "rt_runtime_us", |
06ecb27c PM |
9174 | .read_s64 = cpu_rt_runtime_read, |
9175 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9176 | }, |
d0b27fa7 PZ |
9177 | { |
9178 | .name = "rt_period_us", | |
f4c753b7 PM |
9179 | .read_u64 = cpu_rt_period_read_uint, |
9180 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9181 | }, |
052f1dc7 | 9182 | #endif |
68318b8e SV |
9183 | }; |
9184 | ||
9185 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9186 | { | |
fe5c7cc2 | 9187 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9188 | } |
9189 | ||
9190 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9191 | .name = "cpu", |
9192 | .create = cpu_cgroup_create, | |
9193 | .destroy = cpu_cgroup_destroy, | |
9194 | .can_attach = cpu_cgroup_can_attach, | |
9195 | .attach = cpu_cgroup_attach, | |
9196 | .populate = cpu_cgroup_populate, | |
9197 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9198 | .early_init = 1, |
9199 | }; | |
9200 | ||
052f1dc7 | 9201 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9202 | |
9203 | #ifdef CONFIG_CGROUP_CPUACCT | |
9204 | ||
9205 | /* | |
9206 | * CPU accounting code for task groups. | |
9207 | * | |
9208 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9209 | * (balbir@in.ibm.com). | |
9210 | */ | |
9211 | ||
934352f2 | 9212 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9213 | struct cpuacct { |
9214 | struct cgroup_subsys_state css; | |
9215 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 9216 | u64 __percpu *cpuusage; |
ef12fefa | 9217 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 9218 | struct cpuacct *parent; |
d842de87 SV |
9219 | }; |
9220 | ||
9221 | struct cgroup_subsys cpuacct_subsys; | |
9222 | ||
9223 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9224 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9225 | { |
32cd756a | 9226 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9227 | struct cpuacct, css); |
9228 | } | |
9229 | ||
9230 | /* return cpu accounting group to which this task belongs */ | |
9231 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9232 | { | |
9233 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9234 | struct cpuacct, css); | |
9235 | } | |
9236 | ||
9237 | /* create a new cpu accounting group */ | |
9238 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9239 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9240 | { |
9241 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 9242 | int i; |
d842de87 SV |
9243 | |
9244 | if (!ca) | |
ef12fefa | 9245 | goto out; |
d842de87 SV |
9246 | |
9247 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9248 | if (!ca->cpuusage) |
9249 | goto out_free_ca; | |
9250 | ||
9251 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9252 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9253 | goto out_free_counters; | |
d842de87 | 9254 | |
934352f2 BR |
9255 | if (cgrp->parent) |
9256 | ca->parent = cgroup_ca(cgrp->parent); | |
9257 | ||
d842de87 | 9258 | return &ca->css; |
ef12fefa BR |
9259 | |
9260 | out_free_counters: | |
9261 | while (--i >= 0) | |
9262 | percpu_counter_destroy(&ca->cpustat[i]); | |
9263 | free_percpu(ca->cpuusage); | |
9264 | out_free_ca: | |
9265 | kfree(ca); | |
9266 | out: | |
9267 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9268 | } |
9269 | ||
9270 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9271 | static void |
32cd756a | 9272 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9273 | { |
32cd756a | 9274 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9275 | int i; |
d842de87 | 9276 | |
ef12fefa BR |
9277 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9278 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9279 | free_percpu(ca->cpuusage); |
9280 | kfree(ca); | |
9281 | } | |
9282 | ||
720f5498 KC |
9283 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9284 | { | |
b36128c8 | 9285 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9286 | u64 data; |
9287 | ||
9288 | #ifndef CONFIG_64BIT | |
9289 | /* | |
9290 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9291 | */ | |
05fa785c | 9292 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9293 | data = *cpuusage; |
05fa785c | 9294 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9295 | #else |
9296 | data = *cpuusage; | |
9297 | #endif | |
9298 | ||
9299 | return data; | |
9300 | } | |
9301 | ||
9302 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9303 | { | |
b36128c8 | 9304 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9305 | |
9306 | #ifndef CONFIG_64BIT | |
9307 | /* | |
9308 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9309 | */ | |
05fa785c | 9310 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9311 | *cpuusage = val; |
05fa785c | 9312 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9313 | #else |
9314 | *cpuusage = val; | |
9315 | #endif | |
9316 | } | |
9317 | ||
d842de87 | 9318 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9319 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9320 | { |
32cd756a | 9321 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9322 | u64 totalcpuusage = 0; |
9323 | int i; | |
9324 | ||
720f5498 KC |
9325 | for_each_present_cpu(i) |
9326 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9327 | |
9328 | return totalcpuusage; | |
9329 | } | |
9330 | ||
0297b803 DG |
9331 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9332 | u64 reset) | |
9333 | { | |
9334 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9335 | int err = 0; | |
9336 | int i; | |
9337 | ||
9338 | if (reset) { | |
9339 | err = -EINVAL; | |
9340 | goto out; | |
9341 | } | |
9342 | ||
720f5498 KC |
9343 | for_each_present_cpu(i) |
9344 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9345 | |
0297b803 DG |
9346 | out: |
9347 | return err; | |
9348 | } | |
9349 | ||
e9515c3c KC |
9350 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9351 | struct seq_file *m) | |
9352 | { | |
9353 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9354 | u64 percpu; | |
9355 | int i; | |
9356 | ||
9357 | for_each_present_cpu(i) { | |
9358 | percpu = cpuacct_cpuusage_read(ca, i); | |
9359 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9360 | } | |
9361 | seq_printf(m, "\n"); | |
9362 | return 0; | |
9363 | } | |
9364 | ||
ef12fefa BR |
9365 | static const char *cpuacct_stat_desc[] = { |
9366 | [CPUACCT_STAT_USER] = "user", | |
9367 | [CPUACCT_STAT_SYSTEM] = "system", | |
9368 | }; | |
9369 | ||
9370 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9371 | struct cgroup_map_cb *cb) | |
9372 | { | |
9373 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9374 | int i; | |
9375 | ||
9376 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9377 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9378 | val = cputime64_to_clock_t(val); | |
9379 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9380 | } | |
9381 | return 0; | |
9382 | } | |
9383 | ||
d842de87 SV |
9384 | static struct cftype files[] = { |
9385 | { | |
9386 | .name = "usage", | |
f4c753b7 PM |
9387 | .read_u64 = cpuusage_read, |
9388 | .write_u64 = cpuusage_write, | |
d842de87 | 9389 | }, |
e9515c3c KC |
9390 | { |
9391 | .name = "usage_percpu", | |
9392 | .read_seq_string = cpuacct_percpu_seq_read, | |
9393 | }, | |
ef12fefa BR |
9394 | { |
9395 | .name = "stat", | |
9396 | .read_map = cpuacct_stats_show, | |
9397 | }, | |
d842de87 SV |
9398 | }; |
9399 | ||
32cd756a | 9400 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9401 | { |
32cd756a | 9402 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9403 | } |
9404 | ||
9405 | /* | |
9406 | * charge this task's execution time to its accounting group. | |
9407 | * | |
9408 | * called with rq->lock held. | |
9409 | */ | |
9410 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9411 | { | |
9412 | struct cpuacct *ca; | |
934352f2 | 9413 | int cpu; |
d842de87 | 9414 | |
c40c6f85 | 9415 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9416 | return; |
9417 | ||
934352f2 | 9418 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9419 | |
9420 | rcu_read_lock(); | |
9421 | ||
d842de87 | 9422 | ca = task_ca(tsk); |
d842de87 | 9423 | |
934352f2 | 9424 | for (; ca; ca = ca->parent) { |
b36128c8 | 9425 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9426 | *cpuusage += cputime; |
9427 | } | |
a18b83b7 BR |
9428 | |
9429 | rcu_read_unlock(); | |
d842de87 SV |
9430 | } |
9431 | ||
fa535a77 AB |
9432 | /* |
9433 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9434 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9435 | * percpu_counter_add with values large enough to always overflow the | |
9436 | * per cpu batch limit causing bad SMP scalability. | |
9437 | * | |
9438 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9439 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9440 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9441 | */ | |
9442 | #ifdef CONFIG_SMP | |
9443 | #define CPUACCT_BATCH \ | |
9444 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9445 | #else | |
9446 | #define CPUACCT_BATCH 0 | |
9447 | #endif | |
9448 | ||
ef12fefa BR |
9449 | /* |
9450 | * Charge the system/user time to the task's accounting group. | |
9451 | */ | |
9452 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9453 | enum cpuacct_stat_index idx, cputime_t val) | |
9454 | { | |
9455 | struct cpuacct *ca; | |
fa535a77 | 9456 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9457 | |
9458 | if (unlikely(!cpuacct_subsys.active)) | |
9459 | return; | |
9460 | ||
9461 | rcu_read_lock(); | |
9462 | ca = task_ca(tsk); | |
9463 | ||
9464 | do { | |
fa535a77 | 9465 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9466 | ca = ca->parent; |
9467 | } while (ca); | |
9468 | rcu_read_unlock(); | |
9469 | } | |
9470 | ||
d842de87 SV |
9471 | struct cgroup_subsys cpuacct_subsys = { |
9472 | .name = "cpuacct", | |
9473 | .create = cpuacct_create, | |
9474 | .destroy = cpuacct_destroy, | |
9475 | .populate = cpuacct_populate, | |
9476 | .subsys_id = cpuacct_subsys_id, | |
9477 | }; | |
9478 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
9479 | |
9480 | #ifndef CONFIG_SMP | |
9481 | ||
03b042bf PM |
9482 | void synchronize_sched_expedited(void) |
9483 | { | |
fc390cde | 9484 | barrier(); |
03b042bf PM |
9485 | } |
9486 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9487 | ||
9488 | #else /* #ifndef CONFIG_SMP */ | |
9489 | ||
cc631fb7 | 9490 | static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0); |
03b042bf | 9491 | |
cc631fb7 | 9492 | static int synchronize_sched_expedited_cpu_stop(void *data) |
03b042bf | 9493 | { |
969c7921 TH |
9494 | /* |
9495 | * There must be a full memory barrier on each affected CPU | |
9496 | * between the time that try_stop_cpus() is called and the | |
9497 | * time that it returns. | |
9498 | * | |
9499 | * In the current initial implementation of cpu_stop, the | |
9500 | * above condition is already met when the control reaches | |
9501 | * this point and the following smp_mb() is not strictly | |
9502 | * necessary. Do smp_mb() anyway for documentation and | |
9503 | * robustness against future implementation changes. | |
9504 | */ | |
cc631fb7 | 9505 | smp_mb(); /* See above comment block. */ |
969c7921 | 9506 | return 0; |
03b042bf | 9507 | } |
03b042bf PM |
9508 | |
9509 | /* | |
9510 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9511 | * approach to force grace period to end quickly. This consumes | |
9512 | * significant time on all CPUs, and is thus not recommended for | |
9513 | * any sort of common-case code. | |
9514 | * | |
9515 | * Note that it is illegal to call this function while holding any | |
9516 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9517 | * observe this restriction will result in deadlock. | |
9518 | */ | |
9519 | void synchronize_sched_expedited(void) | |
9520 | { | |
969c7921 | 9521 | int snap, trycount = 0; |
03b042bf PM |
9522 | |
9523 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
969c7921 | 9524 | snap = atomic_read(&synchronize_sched_expedited_count) + 1; |
03b042bf | 9525 | get_online_cpus(); |
969c7921 TH |
9526 | while (try_stop_cpus(cpu_online_mask, |
9527 | synchronize_sched_expedited_cpu_stop, | |
94458d5e | 9528 | NULL) == -EAGAIN) { |
03b042bf PM |
9529 | put_online_cpus(); |
9530 | if (trycount++ < 10) | |
9531 | udelay(trycount * num_online_cpus()); | |
9532 | else { | |
9533 | synchronize_sched(); | |
9534 | return; | |
9535 | } | |
969c7921 | 9536 | if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) { |
03b042bf PM |
9537 | smp_mb(); /* ensure test happens before caller kfree */ |
9538 | return; | |
9539 | } | |
9540 | get_online_cpus(); | |
9541 | } | |
969c7921 | 9542 | atomic_inc(&synchronize_sched_expedited_count); |
cc631fb7 | 9543 | smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */ |
03b042bf | 9544 | put_online_cpus(); |
03b042bf PM |
9545 | } |
9546 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9547 | ||
9548 | #endif /* #else #ifndef CONFIG_SMP */ |