Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
969c7921 | 60 | #include <linux/stop_machine.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
5a0e3ad6 | 74 | #include <linux/slab.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 | 79 | #include "sched_cpupri.h" |
21aa9af0 | 80 | #include "workqueue_sched.h" |
6e0534f2 | 81 | |
a8d154b0 | 82 | #define CREATE_TRACE_POINTS |
ad8d75ff | 83 | #include <trace/events/sched.h> |
a8d154b0 | 84 | |
1da177e4 LT |
85 | /* |
86 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
87 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
88 | * and back. | |
89 | */ | |
90 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
91 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
92 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
93 | ||
94 | /* | |
95 | * 'User priority' is the nice value converted to something we | |
96 | * can work with better when scaling various scheduler parameters, | |
97 | * it's a [ 0 ... 39 ] range. | |
98 | */ | |
99 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
100 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
101 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
102 | ||
103 | /* | |
d7876a08 | 104 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 105 | */ |
d6322faf | 106 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 107 | |
6aa645ea IM |
108 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
109 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
110 | ||
1da177e4 LT |
111 | /* |
112 | * These are the 'tuning knobs' of the scheduler: | |
113 | * | |
a4ec24b4 | 114 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
115 | * Timeslices get refilled after they expire. |
116 | */ | |
1da177e4 | 117 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 118 | |
d0b27fa7 PZ |
119 | /* |
120 | * single value that denotes runtime == period, ie unlimited time. | |
121 | */ | |
122 | #define RUNTIME_INF ((u64)~0ULL) | |
123 | ||
e05606d3 IM |
124 | static inline int rt_policy(int policy) |
125 | { | |
3f33a7ce | 126 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
127 | return 1; |
128 | return 0; | |
129 | } | |
130 | ||
131 | static inline int task_has_rt_policy(struct task_struct *p) | |
132 | { | |
133 | return rt_policy(p->policy); | |
134 | } | |
135 | ||
1da177e4 | 136 | /* |
6aa645ea | 137 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 138 | */ |
6aa645ea IM |
139 | struct rt_prio_array { |
140 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
141 | struct list_head queue[MAX_RT_PRIO]; | |
142 | }; | |
143 | ||
d0b27fa7 | 144 | struct rt_bandwidth { |
ea736ed5 | 145 | /* nests inside the rq lock: */ |
0986b11b | 146 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
147 | ktime_t rt_period; |
148 | u64 rt_runtime; | |
149 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
150 | }; |
151 | ||
152 | static struct rt_bandwidth def_rt_bandwidth; | |
153 | ||
154 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
155 | ||
156 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
157 | { | |
158 | struct rt_bandwidth *rt_b = | |
159 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
160 | ktime_t now; | |
161 | int overrun; | |
162 | int idle = 0; | |
163 | ||
164 | for (;;) { | |
165 | now = hrtimer_cb_get_time(timer); | |
166 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
167 | ||
168 | if (!overrun) | |
169 | break; | |
170 | ||
171 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
172 | } | |
173 | ||
174 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
175 | } | |
176 | ||
177 | static | |
178 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
179 | { | |
180 | rt_b->rt_period = ns_to_ktime(period); | |
181 | rt_b->rt_runtime = runtime; | |
182 | ||
0986b11b | 183 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 184 | |
d0b27fa7 PZ |
185 | hrtimer_init(&rt_b->rt_period_timer, |
186 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
187 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
188 | } |
189 | ||
c8bfff6d KH |
190 | static inline int rt_bandwidth_enabled(void) |
191 | { | |
192 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
193 | } |
194 | ||
195 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
196 | { | |
197 | ktime_t now; | |
198 | ||
cac64d00 | 199 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
200 | return; |
201 | ||
202 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
203 | return; | |
204 | ||
0986b11b | 205 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 206 | for (;;) { |
7f1e2ca9 PZ |
207 | unsigned long delta; |
208 | ktime_t soft, hard; | |
209 | ||
d0b27fa7 PZ |
210 | if (hrtimer_active(&rt_b->rt_period_timer)) |
211 | break; | |
212 | ||
213 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
214 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
215 | |
216 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
217 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
218 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
219 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 220 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 221 | } |
0986b11b | 222 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
223 | } |
224 | ||
225 | #ifdef CONFIG_RT_GROUP_SCHED | |
226 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
227 | { | |
228 | hrtimer_cancel(&rt_b->rt_period_timer); | |
229 | } | |
230 | #endif | |
231 | ||
712555ee HC |
232 | /* |
233 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
234 | * detach_destroy_domains and partition_sched_domains. | |
235 | */ | |
236 | static DEFINE_MUTEX(sched_domains_mutex); | |
237 | ||
7c941438 | 238 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 239 | |
68318b8e SV |
240 | #include <linux/cgroup.h> |
241 | ||
29f59db3 SV |
242 | struct cfs_rq; |
243 | ||
6f505b16 PZ |
244 | static LIST_HEAD(task_groups); |
245 | ||
29f59db3 | 246 | /* task group related information */ |
4cf86d77 | 247 | struct task_group { |
68318b8e | 248 | struct cgroup_subsys_state css; |
6c415b92 | 249 | |
052f1dc7 | 250 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
251 | /* schedulable entities of this group on each cpu */ |
252 | struct sched_entity **se; | |
253 | /* runqueue "owned" by this group on each cpu */ | |
254 | struct cfs_rq **cfs_rq; | |
255 | unsigned long shares; | |
052f1dc7 PZ |
256 | #endif |
257 | ||
258 | #ifdef CONFIG_RT_GROUP_SCHED | |
259 | struct sched_rt_entity **rt_se; | |
260 | struct rt_rq **rt_rq; | |
261 | ||
d0b27fa7 | 262 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 263 | #endif |
6b2d7700 | 264 | |
ae8393e5 | 265 | struct rcu_head rcu; |
6f505b16 | 266 | struct list_head list; |
f473aa5e PZ |
267 | |
268 | struct task_group *parent; | |
269 | struct list_head siblings; | |
270 | struct list_head children; | |
29f59db3 SV |
271 | }; |
272 | ||
eff766a6 | 273 | #define root_task_group init_task_group |
6f505b16 | 274 | |
8ed36996 | 275 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
276 | * a task group's cpu shares. |
277 | */ | |
8ed36996 | 278 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 279 | |
e9036b36 CG |
280 | #ifdef CONFIG_FAIR_GROUP_SCHED |
281 | ||
57310a98 PZ |
282 | #ifdef CONFIG_SMP |
283 | static int root_task_group_empty(void) | |
284 | { | |
285 | return list_empty(&root_task_group.children); | |
286 | } | |
287 | #endif | |
288 | ||
052f1dc7 | 289 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 290 | |
cb4ad1ff | 291 | /* |
2e084786 LJ |
292 | * A weight of 0 or 1 can cause arithmetics problems. |
293 | * A weight of a cfs_rq is the sum of weights of which entities | |
294 | * are queued on this cfs_rq, so a weight of a entity should not be | |
295 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
296 | * (The default weight is 1024 - so there's no practical |
297 | * limitation from this.) | |
298 | */ | |
18d95a28 | 299 | #define MIN_SHARES 2 |
2e084786 | 300 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 301 | |
052f1dc7 PZ |
302 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
303 | #endif | |
304 | ||
29f59db3 | 305 | /* Default task group. |
3a252015 | 306 | * Every task in system belong to this group at bootup. |
29f59db3 | 307 | */ |
434d53b0 | 308 | struct task_group init_task_group; |
29f59db3 | 309 | |
7c941438 | 310 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 311 | |
6aa645ea IM |
312 | /* CFS-related fields in a runqueue */ |
313 | struct cfs_rq { | |
314 | struct load_weight load; | |
315 | unsigned long nr_running; | |
316 | ||
6aa645ea | 317 | u64 exec_clock; |
e9acbff6 | 318 | u64 min_vruntime; |
6aa645ea IM |
319 | |
320 | struct rb_root tasks_timeline; | |
321 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
322 | |
323 | struct list_head tasks; | |
324 | struct list_head *balance_iterator; | |
325 | ||
326 | /* | |
327 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
328 | * It is set to NULL otherwise (i.e when none are currently running). |
329 | */ | |
4793241b | 330 | struct sched_entity *curr, *next, *last; |
ddc97297 | 331 | |
5ac5c4d6 | 332 | unsigned int nr_spread_over; |
ddc97297 | 333 | |
62160e3f | 334 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
335 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
336 | ||
41a2d6cf IM |
337 | /* |
338 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
339 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
340 | * (like users, containers etc.) | |
341 | * | |
342 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
343 | * list is used during load balance. | |
344 | */ | |
41a2d6cf IM |
345 | struct list_head leaf_cfs_rq_list; |
346 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
347 | |
348 | #ifdef CONFIG_SMP | |
c09595f6 | 349 | /* |
c8cba857 | 350 | * the part of load.weight contributed by tasks |
c09595f6 | 351 | */ |
c8cba857 | 352 | unsigned long task_weight; |
c09595f6 | 353 | |
c8cba857 PZ |
354 | /* |
355 | * h_load = weight * f(tg) | |
356 | * | |
357 | * Where f(tg) is the recursive weight fraction assigned to | |
358 | * this group. | |
359 | */ | |
360 | unsigned long h_load; | |
c09595f6 | 361 | |
c8cba857 PZ |
362 | /* |
363 | * this cpu's part of tg->shares | |
364 | */ | |
365 | unsigned long shares; | |
f1d239f7 PZ |
366 | |
367 | /* | |
368 | * load.weight at the time we set shares | |
369 | */ | |
370 | unsigned long rq_weight; | |
c09595f6 | 371 | #endif |
6aa645ea IM |
372 | #endif |
373 | }; | |
1da177e4 | 374 | |
6aa645ea IM |
375 | /* Real-Time classes' related field in a runqueue: */ |
376 | struct rt_rq { | |
377 | struct rt_prio_array active; | |
63489e45 | 378 | unsigned long rt_nr_running; |
052f1dc7 | 379 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
380 | struct { |
381 | int curr; /* highest queued rt task prio */ | |
398a153b | 382 | #ifdef CONFIG_SMP |
e864c499 | 383 | int next; /* next highest */ |
398a153b | 384 | #endif |
e864c499 | 385 | } highest_prio; |
6f505b16 | 386 | #endif |
fa85ae24 | 387 | #ifdef CONFIG_SMP |
73fe6aae | 388 | unsigned long rt_nr_migratory; |
a1ba4d8b | 389 | unsigned long rt_nr_total; |
a22d7fc1 | 390 | int overloaded; |
917b627d | 391 | struct plist_head pushable_tasks; |
fa85ae24 | 392 | #endif |
6f505b16 | 393 | int rt_throttled; |
fa85ae24 | 394 | u64 rt_time; |
ac086bc2 | 395 | u64 rt_runtime; |
ea736ed5 | 396 | /* Nests inside the rq lock: */ |
0986b11b | 397 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 398 | |
052f1dc7 | 399 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
400 | unsigned long rt_nr_boosted; |
401 | ||
6f505b16 PZ |
402 | struct rq *rq; |
403 | struct list_head leaf_rt_rq_list; | |
404 | struct task_group *tg; | |
6f505b16 | 405 | #endif |
6aa645ea IM |
406 | }; |
407 | ||
57d885fe GH |
408 | #ifdef CONFIG_SMP |
409 | ||
410 | /* | |
411 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
412 | * variables. Each exclusive cpuset essentially defines an island domain by |
413 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
414 | * exclusive cpuset is created, we also create and attach a new root-domain |
415 | * object. | |
416 | * | |
57d885fe GH |
417 | */ |
418 | struct root_domain { | |
419 | atomic_t refcount; | |
c6c4927b RR |
420 | cpumask_var_t span; |
421 | cpumask_var_t online; | |
637f5085 | 422 | |
0eab9146 | 423 | /* |
637f5085 GH |
424 | * The "RT overload" flag: it gets set if a CPU has more than |
425 | * one runnable RT task. | |
426 | */ | |
c6c4927b | 427 | cpumask_var_t rto_mask; |
0eab9146 | 428 | atomic_t rto_count; |
6e0534f2 GH |
429 | #ifdef CONFIG_SMP |
430 | struct cpupri cpupri; | |
431 | #endif | |
57d885fe GH |
432 | }; |
433 | ||
dc938520 GH |
434 | /* |
435 | * By default the system creates a single root-domain with all cpus as | |
436 | * members (mimicking the global state we have today). | |
437 | */ | |
57d885fe GH |
438 | static struct root_domain def_root_domain; |
439 | ||
440 | #endif | |
441 | ||
1da177e4 LT |
442 | /* |
443 | * This is the main, per-CPU runqueue data structure. | |
444 | * | |
445 | * Locking rule: those places that want to lock multiple runqueues | |
446 | * (such as the load balancing or the thread migration code), lock | |
447 | * acquire operations must be ordered by ascending &runqueue. | |
448 | */ | |
70b97a7f | 449 | struct rq { |
d8016491 | 450 | /* runqueue lock: */ |
05fa785c | 451 | raw_spinlock_t lock; |
1da177e4 LT |
452 | |
453 | /* | |
454 | * nr_running and cpu_load should be in the same cacheline because | |
455 | * remote CPUs use both these fields when doing load calculation. | |
456 | */ | |
457 | unsigned long nr_running; | |
6aa645ea IM |
458 | #define CPU_LOAD_IDX_MAX 5 |
459 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 460 | unsigned long last_load_update_tick; |
46cb4b7c | 461 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 462 | u64 nohz_stamp; |
83cd4fe2 | 463 | unsigned char nohz_balance_kick; |
46cb4b7c | 464 | #endif |
a64692a3 MG |
465 | unsigned int skip_clock_update; |
466 | ||
d8016491 IM |
467 | /* capture load from *all* tasks on this cpu: */ |
468 | struct load_weight load; | |
6aa645ea IM |
469 | unsigned long nr_load_updates; |
470 | u64 nr_switches; | |
471 | ||
472 | struct cfs_rq cfs; | |
6f505b16 | 473 | struct rt_rq rt; |
6f505b16 | 474 | |
6aa645ea | 475 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
476 | /* list of leaf cfs_rq on this cpu: */ |
477 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
478 | #endif |
479 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 480 | struct list_head leaf_rt_rq_list; |
1da177e4 | 481 | #endif |
1da177e4 LT |
482 | |
483 | /* | |
484 | * This is part of a global counter where only the total sum | |
485 | * over all CPUs matters. A task can increase this counter on | |
486 | * one CPU and if it got migrated afterwards it may decrease | |
487 | * it on another CPU. Always updated under the runqueue lock: | |
488 | */ | |
489 | unsigned long nr_uninterruptible; | |
490 | ||
36c8b586 | 491 | struct task_struct *curr, *idle; |
c9819f45 | 492 | unsigned long next_balance; |
1da177e4 | 493 | struct mm_struct *prev_mm; |
6aa645ea | 494 | |
3e51f33f | 495 | u64 clock; |
6aa645ea | 496 | |
1da177e4 LT |
497 | atomic_t nr_iowait; |
498 | ||
499 | #ifdef CONFIG_SMP | |
0eab9146 | 500 | struct root_domain *rd; |
1da177e4 LT |
501 | struct sched_domain *sd; |
502 | ||
e51fd5e2 PZ |
503 | unsigned long cpu_power; |
504 | ||
a0a522ce | 505 | unsigned char idle_at_tick; |
1da177e4 | 506 | /* For active balancing */ |
3f029d3c | 507 | int post_schedule; |
1da177e4 LT |
508 | int active_balance; |
509 | int push_cpu; | |
969c7921 | 510 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
511 | /* cpu of this runqueue: */ |
512 | int cpu; | |
1f11eb6a | 513 | int online; |
1da177e4 | 514 | |
a8a51d5e | 515 | unsigned long avg_load_per_task; |
1da177e4 | 516 | |
e9e9250b PZ |
517 | u64 rt_avg; |
518 | u64 age_stamp; | |
1b9508f6 MG |
519 | u64 idle_stamp; |
520 | u64 avg_idle; | |
1da177e4 LT |
521 | #endif |
522 | ||
dce48a84 TG |
523 | /* calc_load related fields */ |
524 | unsigned long calc_load_update; | |
525 | long calc_load_active; | |
526 | ||
8f4d37ec | 527 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
528 | #ifdef CONFIG_SMP |
529 | int hrtick_csd_pending; | |
530 | struct call_single_data hrtick_csd; | |
531 | #endif | |
8f4d37ec PZ |
532 | struct hrtimer hrtick_timer; |
533 | #endif | |
534 | ||
1da177e4 LT |
535 | #ifdef CONFIG_SCHEDSTATS |
536 | /* latency stats */ | |
537 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
538 | unsigned long long rq_cpu_time; |
539 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
540 | |
541 | /* sys_sched_yield() stats */ | |
480b9434 | 542 | unsigned int yld_count; |
1da177e4 LT |
543 | |
544 | /* schedule() stats */ | |
480b9434 KC |
545 | unsigned int sched_switch; |
546 | unsigned int sched_count; | |
547 | unsigned int sched_goidle; | |
1da177e4 LT |
548 | |
549 | /* try_to_wake_up() stats */ | |
480b9434 KC |
550 | unsigned int ttwu_count; |
551 | unsigned int ttwu_local; | |
b8efb561 IM |
552 | |
553 | /* BKL stats */ | |
480b9434 | 554 | unsigned int bkl_count; |
1da177e4 LT |
555 | #endif |
556 | }; | |
557 | ||
f34e3b61 | 558 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 559 | |
7d478721 PZ |
560 | static inline |
561 | void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) | |
dd41f596 | 562 | { |
7d478721 | 563 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); |
a64692a3 MG |
564 | |
565 | /* | |
566 | * A queue event has occurred, and we're going to schedule. In | |
567 | * this case, we can save a useless back to back clock update. | |
568 | */ | |
569 | if (test_tsk_need_resched(p)) | |
570 | rq->skip_clock_update = 1; | |
dd41f596 IM |
571 | } |
572 | ||
0a2966b4 CL |
573 | static inline int cpu_of(struct rq *rq) |
574 | { | |
575 | #ifdef CONFIG_SMP | |
576 | return rq->cpu; | |
577 | #else | |
578 | return 0; | |
579 | #endif | |
580 | } | |
581 | ||
497f0ab3 | 582 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
583 | rcu_dereference_check((p), \ |
584 | rcu_read_lock_sched_held() || \ | |
585 | lockdep_is_held(&sched_domains_mutex)) | |
586 | ||
674311d5 NP |
587 | /* |
588 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 589 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
590 | * |
591 | * The domain tree of any CPU may only be accessed from within | |
592 | * preempt-disabled sections. | |
593 | */ | |
48f24c4d | 594 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 595 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
596 | |
597 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
598 | #define this_rq() (&__get_cpu_var(runqueues)) | |
599 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
600 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 601 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 602 | |
dc61b1d6 PZ |
603 | #ifdef CONFIG_CGROUP_SCHED |
604 | ||
605 | /* | |
606 | * Return the group to which this tasks belongs. | |
607 | * | |
608 | * We use task_subsys_state_check() and extend the RCU verification | |
609 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
610 | * holds that lock for each task it moves into the cgroup. Therefore | |
611 | * by holding that lock, we pin the task to the current cgroup. | |
612 | */ | |
613 | static inline struct task_group *task_group(struct task_struct *p) | |
614 | { | |
615 | struct cgroup_subsys_state *css; | |
616 | ||
617 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
618 | lockdep_is_held(&task_rq(p)->lock)); | |
619 | return container_of(css, struct task_group, css); | |
620 | } | |
621 | ||
622 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
623 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
624 | { | |
625 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
626 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
627 | p->se.parent = task_group(p)->se[cpu]; | |
628 | #endif | |
629 | ||
630 | #ifdef CONFIG_RT_GROUP_SCHED | |
631 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
632 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
633 | #endif | |
634 | } | |
635 | ||
636 | #else /* CONFIG_CGROUP_SCHED */ | |
637 | ||
638 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
639 | static inline struct task_group *task_group(struct task_struct *p) | |
640 | { | |
641 | return NULL; | |
642 | } | |
643 | ||
644 | #endif /* CONFIG_CGROUP_SCHED */ | |
645 | ||
aa9c4c0f | 646 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 647 | { |
a64692a3 MG |
648 | if (!rq->skip_clock_update) |
649 | rq->clock = sched_clock_cpu(cpu_of(rq)); | |
3e51f33f PZ |
650 | } |
651 | ||
bf5c91ba IM |
652 | /* |
653 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
654 | */ | |
655 | #ifdef CONFIG_SCHED_DEBUG | |
656 | # define const_debug __read_mostly | |
657 | #else | |
658 | # define const_debug static const | |
659 | #endif | |
660 | ||
017730c1 IM |
661 | /** |
662 | * runqueue_is_locked | |
e17b38bf | 663 | * @cpu: the processor in question. |
017730c1 IM |
664 | * |
665 | * Returns true if the current cpu runqueue is locked. | |
666 | * This interface allows printk to be called with the runqueue lock | |
667 | * held and know whether or not it is OK to wake up the klogd. | |
668 | */ | |
89f19f04 | 669 | int runqueue_is_locked(int cpu) |
017730c1 | 670 | { |
05fa785c | 671 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
672 | } |
673 | ||
bf5c91ba IM |
674 | /* |
675 | * Debugging: various feature bits | |
676 | */ | |
f00b45c1 PZ |
677 | |
678 | #define SCHED_FEAT(name, enabled) \ | |
679 | __SCHED_FEAT_##name , | |
680 | ||
bf5c91ba | 681 | enum { |
f00b45c1 | 682 | #include "sched_features.h" |
bf5c91ba IM |
683 | }; |
684 | ||
f00b45c1 PZ |
685 | #undef SCHED_FEAT |
686 | ||
687 | #define SCHED_FEAT(name, enabled) \ | |
688 | (1UL << __SCHED_FEAT_##name) * enabled | | |
689 | ||
bf5c91ba | 690 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
691 | #include "sched_features.h" |
692 | 0; | |
693 | ||
694 | #undef SCHED_FEAT | |
695 | ||
696 | #ifdef CONFIG_SCHED_DEBUG | |
697 | #define SCHED_FEAT(name, enabled) \ | |
698 | #name , | |
699 | ||
983ed7a6 | 700 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
701 | #include "sched_features.h" |
702 | NULL | |
703 | }; | |
704 | ||
705 | #undef SCHED_FEAT | |
706 | ||
34f3a814 | 707 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 708 | { |
f00b45c1 PZ |
709 | int i; |
710 | ||
711 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
712 | if (!(sysctl_sched_features & (1UL << i))) |
713 | seq_puts(m, "NO_"); | |
714 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 715 | } |
34f3a814 | 716 | seq_puts(m, "\n"); |
f00b45c1 | 717 | |
34f3a814 | 718 | return 0; |
f00b45c1 PZ |
719 | } |
720 | ||
721 | static ssize_t | |
722 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
723 | size_t cnt, loff_t *ppos) | |
724 | { | |
725 | char buf[64]; | |
726 | char *cmp = buf; | |
727 | int neg = 0; | |
728 | int i; | |
729 | ||
730 | if (cnt > 63) | |
731 | cnt = 63; | |
732 | ||
733 | if (copy_from_user(&buf, ubuf, cnt)) | |
734 | return -EFAULT; | |
735 | ||
736 | buf[cnt] = 0; | |
737 | ||
c24b7c52 | 738 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
739 | neg = 1; |
740 | cmp += 3; | |
741 | } | |
742 | ||
743 | for (i = 0; sched_feat_names[i]; i++) { | |
744 | int len = strlen(sched_feat_names[i]); | |
745 | ||
746 | if (strncmp(cmp, sched_feat_names[i], len) == 0) { | |
747 | if (neg) | |
748 | sysctl_sched_features &= ~(1UL << i); | |
749 | else | |
750 | sysctl_sched_features |= (1UL << i); | |
751 | break; | |
752 | } | |
753 | } | |
754 | ||
755 | if (!sched_feat_names[i]) | |
756 | return -EINVAL; | |
757 | ||
42994724 | 758 | *ppos += cnt; |
f00b45c1 PZ |
759 | |
760 | return cnt; | |
761 | } | |
762 | ||
34f3a814 LZ |
763 | static int sched_feat_open(struct inode *inode, struct file *filp) |
764 | { | |
765 | return single_open(filp, sched_feat_show, NULL); | |
766 | } | |
767 | ||
828c0950 | 768 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
769 | .open = sched_feat_open, |
770 | .write = sched_feat_write, | |
771 | .read = seq_read, | |
772 | .llseek = seq_lseek, | |
773 | .release = single_release, | |
f00b45c1 PZ |
774 | }; |
775 | ||
776 | static __init int sched_init_debug(void) | |
777 | { | |
f00b45c1 PZ |
778 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
779 | &sched_feat_fops); | |
780 | ||
781 | return 0; | |
782 | } | |
783 | late_initcall(sched_init_debug); | |
784 | ||
785 | #endif | |
786 | ||
787 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 788 | |
b82d9fdd PZ |
789 | /* |
790 | * Number of tasks to iterate in a single balance run. | |
791 | * Limited because this is done with IRQs disabled. | |
792 | */ | |
793 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
794 | ||
2398f2c6 PZ |
795 | /* |
796 | * ratelimit for updating the group shares. | |
55cd5340 | 797 | * default: 0.25ms |
2398f2c6 | 798 | */ |
55cd5340 | 799 | unsigned int sysctl_sched_shares_ratelimit = 250000; |
0bcdcf28 | 800 | unsigned int normalized_sysctl_sched_shares_ratelimit = 250000; |
2398f2c6 | 801 | |
ffda12a1 PZ |
802 | /* |
803 | * Inject some fuzzyness into changing the per-cpu group shares | |
804 | * this avoids remote rq-locks at the expense of fairness. | |
805 | * default: 4 | |
806 | */ | |
807 | unsigned int sysctl_sched_shares_thresh = 4; | |
808 | ||
e9e9250b PZ |
809 | /* |
810 | * period over which we average the RT time consumption, measured | |
811 | * in ms. | |
812 | * | |
813 | * default: 1s | |
814 | */ | |
815 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
816 | ||
fa85ae24 | 817 | /* |
9f0c1e56 | 818 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
819 | * default: 1s |
820 | */ | |
9f0c1e56 | 821 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 822 | |
6892b75e IM |
823 | static __read_mostly int scheduler_running; |
824 | ||
9f0c1e56 PZ |
825 | /* |
826 | * part of the period that we allow rt tasks to run in us. | |
827 | * default: 0.95s | |
828 | */ | |
829 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 830 | |
d0b27fa7 PZ |
831 | static inline u64 global_rt_period(void) |
832 | { | |
833 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
834 | } | |
835 | ||
836 | static inline u64 global_rt_runtime(void) | |
837 | { | |
e26873bb | 838 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
839 | return RUNTIME_INF; |
840 | ||
841 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
842 | } | |
fa85ae24 | 843 | |
1da177e4 | 844 | #ifndef prepare_arch_switch |
4866cde0 NP |
845 | # define prepare_arch_switch(next) do { } while (0) |
846 | #endif | |
847 | #ifndef finish_arch_switch | |
848 | # define finish_arch_switch(prev) do { } while (0) | |
849 | #endif | |
850 | ||
051a1d1a DA |
851 | static inline int task_current(struct rq *rq, struct task_struct *p) |
852 | { | |
853 | return rq->curr == p; | |
854 | } | |
855 | ||
4866cde0 | 856 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 857 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 858 | { |
051a1d1a | 859 | return task_current(rq, p); |
4866cde0 NP |
860 | } |
861 | ||
70b97a7f | 862 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
863 | { |
864 | } | |
865 | ||
70b97a7f | 866 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 867 | { |
da04c035 IM |
868 | #ifdef CONFIG_DEBUG_SPINLOCK |
869 | /* this is a valid case when another task releases the spinlock */ | |
870 | rq->lock.owner = current; | |
871 | #endif | |
8a25d5de IM |
872 | /* |
873 | * If we are tracking spinlock dependencies then we have to | |
874 | * fix up the runqueue lock - which gets 'carried over' from | |
875 | * prev into current: | |
876 | */ | |
877 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
878 | ||
05fa785c | 879 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
880 | } |
881 | ||
882 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 883 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
884 | { |
885 | #ifdef CONFIG_SMP | |
886 | return p->oncpu; | |
887 | #else | |
051a1d1a | 888 | return task_current(rq, p); |
4866cde0 NP |
889 | #endif |
890 | } | |
891 | ||
70b97a7f | 892 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
893 | { |
894 | #ifdef CONFIG_SMP | |
895 | /* | |
896 | * We can optimise this out completely for !SMP, because the | |
897 | * SMP rebalancing from interrupt is the only thing that cares | |
898 | * here. | |
899 | */ | |
900 | next->oncpu = 1; | |
901 | #endif | |
902 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 903 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 904 | #else |
05fa785c | 905 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
906 | #endif |
907 | } | |
908 | ||
70b97a7f | 909 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
910 | { |
911 | #ifdef CONFIG_SMP | |
912 | /* | |
913 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
914 | * We must ensure this doesn't happen until the switch is completely | |
915 | * finished. | |
916 | */ | |
917 | smp_wmb(); | |
918 | prev->oncpu = 0; | |
919 | #endif | |
920 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
921 | local_irq_enable(); | |
1da177e4 | 922 | #endif |
4866cde0 NP |
923 | } |
924 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 925 | |
0970d299 | 926 | /* |
65cc8e48 PZ |
927 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
928 | * against ttwu(). | |
0970d299 PZ |
929 | */ |
930 | static inline int task_is_waking(struct task_struct *p) | |
931 | { | |
0017d735 | 932 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
933 | } |
934 | ||
b29739f9 IM |
935 | /* |
936 | * __task_rq_lock - lock the runqueue a given task resides on. | |
937 | * Must be called interrupts disabled. | |
938 | */ | |
70b97a7f | 939 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
940 | __acquires(rq->lock) |
941 | { | |
0970d299 PZ |
942 | struct rq *rq; |
943 | ||
3a5c359a | 944 | for (;;) { |
0970d299 | 945 | rq = task_rq(p); |
05fa785c | 946 | raw_spin_lock(&rq->lock); |
65cc8e48 | 947 | if (likely(rq == task_rq(p))) |
3a5c359a | 948 | return rq; |
05fa785c | 949 | raw_spin_unlock(&rq->lock); |
b29739f9 | 950 | } |
b29739f9 IM |
951 | } |
952 | ||
1da177e4 LT |
953 | /* |
954 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 955 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
956 | * explicitly disabling preemption. |
957 | */ | |
70b97a7f | 958 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
959 | __acquires(rq->lock) |
960 | { | |
70b97a7f | 961 | struct rq *rq; |
1da177e4 | 962 | |
3a5c359a AK |
963 | for (;;) { |
964 | local_irq_save(*flags); | |
965 | rq = task_rq(p); | |
05fa785c | 966 | raw_spin_lock(&rq->lock); |
65cc8e48 | 967 | if (likely(rq == task_rq(p))) |
3a5c359a | 968 | return rq; |
05fa785c | 969 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 970 | } |
1da177e4 LT |
971 | } |
972 | ||
a9957449 | 973 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
974 | __releases(rq->lock) |
975 | { | |
05fa785c | 976 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
977 | } |
978 | ||
70b97a7f | 979 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
980 | __releases(rq->lock) |
981 | { | |
05fa785c | 982 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
983 | } |
984 | ||
1da177e4 | 985 | /* |
cc2a73b5 | 986 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 987 | */ |
a9957449 | 988 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
989 | __acquires(rq->lock) |
990 | { | |
70b97a7f | 991 | struct rq *rq; |
1da177e4 LT |
992 | |
993 | local_irq_disable(); | |
994 | rq = this_rq(); | |
05fa785c | 995 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
996 | |
997 | return rq; | |
998 | } | |
999 | ||
8f4d37ec PZ |
1000 | #ifdef CONFIG_SCHED_HRTICK |
1001 | /* | |
1002 | * Use HR-timers to deliver accurate preemption points. | |
1003 | * | |
1004 | * Its all a bit involved since we cannot program an hrt while holding the | |
1005 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
1006 | * reschedule event. | |
1007 | * | |
1008 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
1009 | * rq->lock. | |
1010 | */ | |
8f4d37ec PZ |
1011 | |
1012 | /* | |
1013 | * Use hrtick when: | |
1014 | * - enabled by features | |
1015 | * - hrtimer is actually high res | |
1016 | */ | |
1017 | static inline int hrtick_enabled(struct rq *rq) | |
1018 | { | |
1019 | if (!sched_feat(HRTICK)) | |
1020 | return 0; | |
ba42059f | 1021 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1022 | return 0; |
8f4d37ec PZ |
1023 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1024 | } | |
1025 | ||
8f4d37ec PZ |
1026 | static void hrtick_clear(struct rq *rq) |
1027 | { | |
1028 | if (hrtimer_active(&rq->hrtick_timer)) | |
1029 | hrtimer_cancel(&rq->hrtick_timer); | |
1030 | } | |
1031 | ||
8f4d37ec PZ |
1032 | /* |
1033 | * High-resolution timer tick. | |
1034 | * Runs from hardirq context with interrupts disabled. | |
1035 | */ | |
1036 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1037 | { | |
1038 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1039 | ||
1040 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1041 | ||
05fa785c | 1042 | raw_spin_lock(&rq->lock); |
3e51f33f | 1043 | update_rq_clock(rq); |
8f4d37ec | 1044 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1045 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1046 | |
1047 | return HRTIMER_NORESTART; | |
1048 | } | |
1049 | ||
95e904c7 | 1050 | #ifdef CONFIG_SMP |
31656519 PZ |
1051 | /* |
1052 | * called from hardirq (IPI) context | |
1053 | */ | |
1054 | static void __hrtick_start(void *arg) | |
b328ca18 | 1055 | { |
31656519 | 1056 | struct rq *rq = arg; |
b328ca18 | 1057 | |
05fa785c | 1058 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1059 | hrtimer_restart(&rq->hrtick_timer); |
1060 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1061 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1062 | } |
1063 | ||
31656519 PZ |
1064 | /* |
1065 | * Called to set the hrtick timer state. | |
1066 | * | |
1067 | * called with rq->lock held and irqs disabled | |
1068 | */ | |
1069 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1070 | { |
31656519 PZ |
1071 | struct hrtimer *timer = &rq->hrtick_timer; |
1072 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1073 | |
cc584b21 | 1074 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1075 | |
1076 | if (rq == this_rq()) { | |
1077 | hrtimer_restart(timer); | |
1078 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1079 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1080 | rq->hrtick_csd_pending = 1; |
1081 | } | |
b328ca18 PZ |
1082 | } |
1083 | ||
1084 | static int | |
1085 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1086 | { | |
1087 | int cpu = (int)(long)hcpu; | |
1088 | ||
1089 | switch (action) { | |
1090 | case CPU_UP_CANCELED: | |
1091 | case CPU_UP_CANCELED_FROZEN: | |
1092 | case CPU_DOWN_PREPARE: | |
1093 | case CPU_DOWN_PREPARE_FROZEN: | |
1094 | case CPU_DEAD: | |
1095 | case CPU_DEAD_FROZEN: | |
31656519 | 1096 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1097 | return NOTIFY_OK; |
1098 | } | |
1099 | ||
1100 | return NOTIFY_DONE; | |
1101 | } | |
1102 | ||
fa748203 | 1103 | static __init void init_hrtick(void) |
b328ca18 PZ |
1104 | { |
1105 | hotcpu_notifier(hotplug_hrtick, 0); | |
1106 | } | |
31656519 PZ |
1107 | #else |
1108 | /* | |
1109 | * Called to set the hrtick timer state. | |
1110 | * | |
1111 | * called with rq->lock held and irqs disabled | |
1112 | */ | |
1113 | static void hrtick_start(struct rq *rq, u64 delay) | |
1114 | { | |
7f1e2ca9 | 1115 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1116 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1117 | } |
b328ca18 | 1118 | |
006c75f1 | 1119 | static inline void init_hrtick(void) |
8f4d37ec | 1120 | { |
8f4d37ec | 1121 | } |
31656519 | 1122 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1123 | |
31656519 | 1124 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1125 | { |
31656519 PZ |
1126 | #ifdef CONFIG_SMP |
1127 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1128 | |
31656519 PZ |
1129 | rq->hrtick_csd.flags = 0; |
1130 | rq->hrtick_csd.func = __hrtick_start; | |
1131 | rq->hrtick_csd.info = rq; | |
1132 | #endif | |
8f4d37ec | 1133 | |
31656519 PZ |
1134 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1135 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1136 | } |
006c75f1 | 1137 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1138 | static inline void hrtick_clear(struct rq *rq) |
1139 | { | |
1140 | } | |
1141 | ||
8f4d37ec PZ |
1142 | static inline void init_rq_hrtick(struct rq *rq) |
1143 | { | |
1144 | } | |
1145 | ||
b328ca18 PZ |
1146 | static inline void init_hrtick(void) |
1147 | { | |
1148 | } | |
006c75f1 | 1149 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1150 | |
c24d20db IM |
1151 | /* |
1152 | * resched_task - mark a task 'to be rescheduled now'. | |
1153 | * | |
1154 | * On UP this means the setting of the need_resched flag, on SMP it | |
1155 | * might also involve a cross-CPU call to trigger the scheduler on | |
1156 | * the target CPU. | |
1157 | */ | |
1158 | #ifdef CONFIG_SMP | |
1159 | ||
1160 | #ifndef tsk_is_polling | |
1161 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1162 | #endif | |
1163 | ||
31656519 | 1164 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1165 | { |
1166 | int cpu; | |
1167 | ||
05fa785c | 1168 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1169 | |
5ed0cec0 | 1170 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1171 | return; |
1172 | ||
5ed0cec0 | 1173 | set_tsk_need_resched(p); |
c24d20db IM |
1174 | |
1175 | cpu = task_cpu(p); | |
1176 | if (cpu == smp_processor_id()) | |
1177 | return; | |
1178 | ||
1179 | /* NEED_RESCHED must be visible before we test polling */ | |
1180 | smp_mb(); | |
1181 | if (!tsk_is_polling(p)) | |
1182 | smp_send_reschedule(cpu); | |
1183 | } | |
1184 | ||
1185 | static void resched_cpu(int cpu) | |
1186 | { | |
1187 | struct rq *rq = cpu_rq(cpu); | |
1188 | unsigned long flags; | |
1189 | ||
05fa785c | 1190 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1191 | return; |
1192 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1193 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1194 | } |
06d8308c TG |
1195 | |
1196 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1197 | /* |
1198 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1199 | * from an idle cpu. This is good for power-savings. | |
1200 | * | |
1201 | * We don't do similar optimization for completely idle system, as | |
1202 | * selecting an idle cpu will add more delays to the timers than intended | |
1203 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1204 | */ | |
1205 | int get_nohz_timer_target(void) | |
1206 | { | |
1207 | int cpu = smp_processor_id(); | |
1208 | int i; | |
1209 | struct sched_domain *sd; | |
1210 | ||
1211 | for_each_domain(cpu, sd) { | |
1212 | for_each_cpu(i, sched_domain_span(sd)) | |
1213 | if (!idle_cpu(i)) | |
1214 | return i; | |
1215 | } | |
1216 | return cpu; | |
1217 | } | |
06d8308c TG |
1218 | /* |
1219 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1220 | * idle CPU then this timer might expire before the next timer event | |
1221 | * which is scheduled to wake up that CPU. In case of a completely | |
1222 | * idle system the next event might even be infinite time into the | |
1223 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1224 | * leaves the inner idle loop so the newly added timer is taken into | |
1225 | * account when the CPU goes back to idle and evaluates the timer | |
1226 | * wheel for the next timer event. | |
1227 | */ | |
1228 | void wake_up_idle_cpu(int cpu) | |
1229 | { | |
1230 | struct rq *rq = cpu_rq(cpu); | |
1231 | ||
1232 | if (cpu == smp_processor_id()) | |
1233 | return; | |
1234 | ||
1235 | /* | |
1236 | * This is safe, as this function is called with the timer | |
1237 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1238 | * to idle and has not yet set rq->curr to idle then it will | |
1239 | * be serialized on the timer wheel base lock and take the new | |
1240 | * timer into account automatically. | |
1241 | */ | |
1242 | if (rq->curr != rq->idle) | |
1243 | return; | |
1244 | ||
1245 | /* | |
1246 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1247 | * lockless. The worst case is that the other CPU runs the | |
1248 | * idle task through an additional NOOP schedule() | |
1249 | */ | |
5ed0cec0 | 1250 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1251 | |
1252 | /* NEED_RESCHED must be visible before we test polling */ | |
1253 | smp_mb(); | |
1254 | if (!tsk_is_polling(rq->idle)) | |
1255 | smp_send_reschedule(cpu); | |
1256 | } | |
39c0cbe2 MG |
1257 | |
1258 | int nohz_ratelimit(int cpu) | |
1259 | { | |
1260 | struct rq *rq = cpu_rq(cpu); | |
1261 | u64 diff = rq->clock - rq->nohz_stamp; | |
1262 | ||
1263 | rq->nohz_stamp = rq->clock; | |
1264 | ||
1265 | return diff < (NSEC_PER_SEC / HZ) >> 1; | |
1266 | } | |
1267 | ||
6d6bc0ad | 1268 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1269 | |
e9e9250b PZ |
1270 | static u64 sched_avg_period(void) |
1271 | { | |
1272 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1273 | } | |
1274 | ||
1275 | static void sched_avg_update(struct rq *rq) | |
1276 | { | |
1277 | s64 period = sched_avg_period(); | |
1278 | ||
1279 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
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 | } | |
6d6bc0ad | 1301 | #endif /* CONFIG_SMP */ |
c24d20db | 1302 | |
45bf76df IM |
1303 | #if BITS_PER_LONG == 32 |
1304 | # define WMULT_CONST (~0UL) | |
1305 | #else | |
1306 | # define WMULT_CONST (1UL << 32) | |
1307 | #endif | |
1308 | ||
1309 | #define WMULT_SHIFT 32 | |
1310 | ||
194081eb IM |
1311 | /* |
1312 | * Shift right and round: | |
1313 | */ | |
cf2ab469 | 1314 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1315 | |
a7be37ac PZ |
1316 | /* |
1317 | * delta *= weight / lw | |
1318 | */ | |
cb1c4fc9 | 1319 | static unsigned long |
45bf76df IM |
1320 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1321 | struct load_weight *lw) | |
1322 | { | |
1323 | u64 tmp; | |
1324 | ||
7a232e03 LJ |
1325 | if (!lw->inv_weight) { |
1326 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1327 | lw->inv_weight = 1; | |
1328 | else | |
1329 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1330 | / (lw->weight+1); | |
1331 | } | |
45bf76df IM |
1332 | |
1333 | tmp = (u64)delta_exec * weight; | |
1334 | /* | |
1335 | * Check whether we'd overflow the 64-bit multiplication: | |
1336 | */ | |
194081eb | 1337 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1338 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1339 | WMULT_SHIFT/2); |
1340 | else | |
cf2ab469 | 1341 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1342 | |
ecf691da | 1343 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1344 | } |
1345 | ||
1091985b | 1346 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1347 | { |
1348 | lw->weight += inc; | |
e89996ae | 1349 | lw->inv_weight = 0; |
45bf76df IM |
1350 | } |
1351 | ||
1091985b | 1352 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1353 | { |
1354 | lw->weight -= dec; | |
e89996ae | 1355 | lw->inv_weight = 0; |
45bf76df IM |
1356 | } |
1357 | ||
2dd73a4f PW |
1358 | /* |
1359 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1360 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1361 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1362 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1363 | * scaled version of the new time slice allocation that they receive on time |
1364 | * slice expiry etc. | |
1365 | */ | |
1366 | ||
cce7ade8 PZ |
1367 | #define WEIGHT_IDLEPRIO 3 |
1368 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1369 | |
1370 | /* | |
1371 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1372 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1373 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1374 | * that remained on nice 0. | |
1375 | * | |
1376 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1377 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1378 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1379 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1380 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1381 | */ |
1382 | static const int prio_to_weight[40] = { | |
254753dc IM |
1383 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1384 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1385 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1386 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1387 | /* 0 */ 1024, 820, 655, 526, 423, | |
1388 | /* 5 */ 335, 272, 215, 172, 137, | |
1389 | /* 10 */ 110, 87, 70, 56, 45, | |
1390 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1391 | }; |
1392 | ||
5714d2de IM |
1393 | /* |
1394 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1395 | * | |
1396 | * In cases where the weight does not change often, we can use the | |
1397 | * precalculated inverse to speed up arithmetics by turning divisions | |
1398 | * into multiplications: | |
1399 | */ | |
dd41f596 | 1400 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1401 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1402 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1403 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1404 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1405 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1406 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1407 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1408 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1409 | }; |
2dd73a4f | 1410 | |
ef12fefa BR |
1411 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1412 | enum cpuacct_stat_index { | |
1413 | CPUACCT_STAT_USER, /* ... user mode */ | |
1414 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1415 | ||
1416 | CPUACCT_STAT_NSTATS, | |
1417 | }; | |
1418 | ||
d842de87 SV |
1419 | #ifdef CONFIG_CGROUP_CPUACCT |
1420 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1421 | static void cpuacct_update_stats(struct task_struct *tsk, |
1422 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1423 | #else |
1424 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1425 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1426 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1427 | #endif |
1428 | ||
18d95a28 PZ |
1429 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1430 | { | |
1431 | update_load_add(&rq->load, load); | |
1432 | } | |
1433 | ||
1434 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1435 | { | |
1436 | update_load_sub(&rq->load, load); | |
1437 | } | |
1438 | ||
7940ca36 | 1439 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1440 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1441 | |
1442 | /* | |
1443 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1444 | * leaving it for the final time. | |
1445 | */ | |
eb755805 | 1446 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1447 | { |
1448 | struct task_group *parent, *child; | |
eb755805 | 1449 | int ret; |
c09595f6 PZ |
1450 | |
1451 | rcu_read_lock(); | |
1452 | parent = &root_task_group; | |
1453 | down: | |
eb755805 PZ |
1454 | ret = (*down)(parent, data); |
1455 | if (ret) | |
1456 | goto out_unlock; | |
c09595f6 PZ |
1457 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1458 | parent = child; | |
1459 | goto down; | |
1460 | ||
1461 | up: | |
1462 | continue; | |
1463 | } | |
eb755805 PZ |
1464 | ret = (*up)(parent, data); |
1465 | if (ret) | |
1466 | goto out_unlock; | |
c09595f6 PZ |
1467 | |
1468 | child = parent; | |
1469 | parent = parent->parent; | |
1470 | if (parent) | |
1471 | goto up; | |
eb755805 | 1472 | out_unlock: |
c09595f6 | 1473 | rcu_read_unlock(); |
eb755805 PZ |
1474 | |
1475 | return ret; | |
c09595f6 PZ |
1476 | } |
1477 | ||
eb755805 PZ |
1478 | static int tg_nop(struct task_group *tg, void *data) |
1479 | { | |
1480 | return 0; | |
c09595f6 | 1481 | } |
eb755805 PZ |
1482 | #endif |
1483 | ||
1484 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1485 | /* Used instead of source_load when we know the type == 0 */ |
1486 | static unsigned long weighted_cpuload(const int cpu) | |
1487 | { | |
1488 | return cpu_rq(cpu)->load.weight; | |
1489 | } | |
1490 | ||
1491 | /* | |
1492 | * Return a low guess at the load of a migration-source cpu weighted | |
1493 | * according to the scheduling class and "nice" value. | |
1494 | * | |
1495 | * We want to under-estimate the load of migration sources, to | |
1496 | * balance conservatively. | |
1497 | */ | |
1498 | static unsigned long source_load(int cpu, int type) | |
1499 | { | |
1500 | struct rq *rq = cpu_rq(cpu); | |
1501 | unsigned long total = weighted_cpuload(cpu); | |
1502 | ||
1503 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1504 | return total; | |
1505 | ||
1506 | return min(rq->cpu_load[type-1], total); | |
1507 | } | |
1508 | ||
1509 | /* | |
1510 | * Return a high guess at the load of a migration-target cpu weighted | |
1511 | * according to the scheduling class and "nice" value. | |
1512 | */ | |
1513 | static unsigned long target_load(int cpu, int type) | |
1514 | { | |
1515 | struct rq *rq = cpu_rq(cpu); | |
1516 | unsigned long total = weighted_cpuload(cpu); | |
1517 | ||
1518 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1519 | return total; | |
1520 | ||
1521 | return max(rq->cpu_load[type-1], total); | |
1522 | } | |
1523 | ||
ae154be1 PZ |
1524 | static unsigned long power_of(int cpu) |
1525 | { | |
e51fd5e2 | 1526 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1527 | } |
1528 | ||
eb755805 PZ |
1529 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1530 | ||
1531 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1532 | { | |
1533 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1534 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1535 | |
4cd42620 SR |
1536 | if (nr_running) |
1537 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1538 | else |
1539 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1540 | |
1541 | return rq->avg_load_per_task; | |
1542 | } | |
1543 | ||
1544 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1545 | |
43cf38eb | 1546 | static __read_mostly unsigned long __percpu *update_shares_data; |
34d76c41 | 1547 | |
c09595f6 PZ |
1548 | static void __set_se_shares(struct sched_entity *se, unsigned long shares); |
1549 | ||
1550 | /* | |
1551 | * Calculate and set the cpu's group shares. | |
1552 | */ | |
34d76c41 PZ |
1553 | static void update_group_shares_cpu(struct task_group *tg, int cpu, |
1554 | unsigned long sd_shares, | |
1555 | unsigned long sd_rq_weight, | |
4a6cc4bd | 1556 | unsigned long *usd_rq_weight) |
18d95a28 | 1557 | { |
34d76c41 | 1558 | unsigned long shares, rq_weight; |
a5004278 | 1559 | int boost = 0; |
c09595f6 | 1560 | |
4a6cc4bd | 1561 | rq_weight = usd_rq_weight[cpu]; |
a5004278 PZ |
1562 | if (!rq_weight) { |
1563 | boost = 1; | |
1564 | rq_weight = NICE_0_LOAD; | |
1565 | } | |
c8cba857 | 1566 | |
c09595f6 | 1567 | /* |
a8af7246 PZ |
1568 | * \Sum_j shares_j * rq_weight_i |
1569 | * shares_i = ----------------------------- | |
1570 | * \Sum_j rq_weight_j | |
c09595f6 | 1571 | */ |
ec4e0e2f | 1572 | shares = (sd_shares * rq_weight) / sd_rq_weight; |
ffda12a1 | 1573 | shares = clamp_t(unsigned long, shares, MIN_SHARES, MAX_SHARES); |
c09595f6 | 1574 | |
ffda12a1 PZ |
1575 | if (abs(shares - tg->se[cpu]->load.weight) > |
1576 | sysctl_sched_shares_thresh) { | |
1577 | struct rq *rq = cpu_rq(cpu); | |
1578 | unsigned long flags; | |
c09595f6 | 1579 | |
05fa785c | 1580 | raw_spin_lock_irqsave(&rq->lock, flags); |
34d76c41 | 1581 | tg->cfs_rq[cpu]->rq_weight = boost ? 0 : rq_weight; |
a5004278 | 1582 | tg->cfs_rq[cpu]->shares = boost ? 0 : shares; |
ffda12a1 | 1583 | __set_se_shares(tg->se[cpu], shares); |
05fa785c | 1584 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
ffda12a1 | 1585 | } |
18d95a28 | 1586 | } |
c09595f6 PZ |
1587 | |
1588 | /* | |
c8cba857 PZ |
1589 | * Re-compute the task group their per cpu shares over the given domain. |
1590 | * This needs to be done in a bottom-up fashion because the rq weight of a | |
1591 | * parent group depends on the shares of its child groups. | |
c09595f6 | 1592 | */ |
eb755805 | 1593 | static int tg_shares_up(struct task_group *tg, void *data) |
c09595f6 | 1594 | { |
cd8ad40d | 1595 | unsigned long weight, rq_weight = 0, sum_weight = 0, shares = 0; |
4a6cc4bd | 1596 | unsigned long *usd_rq_weight; |
eb755805 | 1597 | struct sched_domain *sd = data; |
34d76c41 | 1598 | unsigned long flags; |
c8cba857 | 1599 | int i; |
c09595f6 | 1600 | |
34d76c41 PZ |
1601 | if (!tg->se[0]) |
1602 | return 0; | |
1603 | ||
1604 | local_irq_save(flags); | |
4a6cc4bd | 1605 | usd_rq_weight = per_cpu_ptr(update_shares_data, smp_processor_id()); |
34d76c41 | 1606 | |
758b2cdc | 1607 | for_each_cpu(i, sched_domain_span(sd)) { |
34d76c41 | 1608 | weight = tg->cfs_rq[i]->load.weight; |
4a6cc4bd | 1609 | usd_rq_weight[i] = weight; |
34d76c41 | 1610 | |
cd8ad40d | 1611 | rq_weight += weight; |
ec4e0e2f KC |
1612 | /* |
1613 | * If there are currently no tasks on the cpu pretend there | |
1614 | * is one of average load so that when a new task gets to | |
1615 | * run here it will not get delayed by group starvation. | |
1616 | */ | |
ec4e0e2f KC |
1617 | if (!weight) |
1618 | weight = NICE_0_LOAD; | |
1619 | ||
cd8ad40d | 1620 | sum_weight += weight; |
c8cba857 | 1621 | shares += tg->cfs_rq[i]->shares; |
c09595f6 | 1622 | } |
c09595f6 | 1623 | |
cd8ad40d PZ |
1624 | if (!rq_weight) |
1625 | rq_weight = sum_weight; | |
1626 | ||
c8cba857 PZ |
1627 | if ((!shares && rq_weight) || shares > tg->shares) |
1628 | shares = tg->shares; | |
1629 | ||
1630 | if (!sd->parent || !(sd->parent->flags & SD_LOAD_BALANCE)) | |
1631 | shares = tg->shares; | |
c09595f6 | 1632 | |
758b2cdc | 1633 | for_each_cpu(i, sched_domain_span(sd)) |
4a6cc4bd | 1634 | update_group_shares_cpu(tg, i, shares, rq_weight, usd_rq_weight); |
34d76c41 PZ |
1635 | |
1636 | local_irq_restore(flags); | |
eb755805 PZ |
1637 | |
1638 | return 0; | |
c09595f6 PZ |
1639 | } |
1640 | ||
1641 | /* | |
c8cba857 PZ |
1642 | * Compute the cpu's hierarchical load factor for each task group. |
1643 | * This needs to be done in a top-down fashion because the load of a child | |
1644 | * group is a fraction of its parents load. | |
c09595f6 | 1645 | */ |
eb755805 | 1646 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1647 | { |
c8cba857 | 1648 | unsigned long load; |
eb755805 | 1649 | long cpu = (long)data; |
c09595f6 | 1650 | |
c8cba857 PZ |
1651 | if (!tg->parent) { |
1652 | load = cpu_rq(cpu)->load.weight; | |
1653 | } else { | |
1654 | load = tg->parent->cfs_rq[cpu]->h_load; | |
1655 | load *= tg->cfs_rq[cpu]->shares; | |
1656 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; | |
1657 | } | |
c09595f6 | 1658 | |
c8cba857 | 1659 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1660 | |
eb755805 | 1661 | return 0; |
c09595f6 PZ |
1662 | } |
1663 | ||
c8cba857 | 1664 | static void update_shares(struct sched_domain *sd) |
4d8d595d | 1665 | { |
e7097159 PZ |
1666 | s64 elapsed; |
1667 | u64 now; | |
1668 | ||
1669 | if (root_task_group_empty()) | |
1670 | return; | |
1671 | ||
c676329a | 1672 | now = local_clock(); |
e7097159 | 1673 | elapsed = now - sd->last_update; |
2398f2c6 PZ |
1674 | |
1675 | if (elapsed >= (s64)(u64)sysctl_sched_shares_ratelimit) { | |
1676 | sd->last_update = now; | |
eb755805 | 1677 | walk_tg_tree(tg_nop, tg_shares_up, sd); |
2398f2c6 | 1678 | } |
4d8d595d PZ |
1679 | } |
1680 | ||
eb755805 | 1681 | static void update_h_load(long cpu) |
c09595f6 | 1682 | { |
e7097159 PZ |
1683 | if (root_task_group_empty()) |
1684 | return; | |
1685 | ||
eb755805 | 1686 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1687 | } |
1688 | ||
c09595f6 PZ |
1689 | #else |
1690 | ||
c8cba857 | 1691 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1692 | { |
1693 | } | |
1694 | ||
18d95a28 PZ |
1695 | #endif |
1696 | ||
8f45e2b5 GH |
1697 | #ifdef CONFIG_PREEMPT |
1698 | ||
b78bb868 PZ |
1699 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1700 | ||
70574a99 | 1701 | /* |
8f45e2b5 GH |
1702 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1703 | * way at the expense of forcing extra atomic operations in all | |
1704 | * invocations. This assures that the double_lock is acquired using the | |
1705 | * same underlying policy as the spinlock_t on this architecture, which | |
1706 | * reduces latency compared to the unfair variant below. However, it | |
1707 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1708 | */ |
8f45e2b5 GH |
1709 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1710 | __releases(this_rq->lock) | |
1711 | __acquires(busiest->lock) | |
1712 | __acquires(this_rq->lock) | |
1713 | { | |
05fa785c | 1714 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1715 | double_rq_lock(this_rq, busiest); |
1716 | ||
1717 | return 1; | |
1718 | } | |
1719 | ||
1720 | #else | |
1721 | /* | |
1722 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1723 | * latency by eliminating extra atomic operations when the locks are | |
1724 | * already in proper order on entry. This favors lower cpu-ids and will | |
1725 | * grant the double lock to lower cpus over higher ids under contention, | |
1726 | * regardless of entry order into the function. | |
1727 | */ | |
1728 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1729 | __releases(this_rq->lock) |
1730 | __acquires(busiest->lock) | |
1731 | __acquires(this_rq->lock) | |
1732 | { | |
1733 | int ret = 0; | |
1734 | ||
05fa785c | 1735 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1736 | if (busiest < this_rq) { |
05fa785c TG |
1737 | raw_spin_unlock(&this_rq->lock); |
1738 | raw_spin_lock(&busiest->lock); | |
1739 | raw_spin_lock_nested(&this_rq->lock, | |
1740 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1741 | ret = 1; |
1742 | } else | |
05fa785c TG |
1743 | raw_spin_lock_nested(&busiest->lock, |
1744 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1745 | } |
1746 | return ret; | |
1747 | } | |
1748 | ||
8f45e2b5 GH |
1749 | #endif /* CONFIG_PREEMPT */ |
1750 | ||
1751 | /* | |
1752 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1753 | */ | |
1754 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1755 | { | |
1756 | if (unlikely(!irqs_disabled())) { | |
1757 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1758 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1759 | BUG_ON(1); |
1760 | } | |
1761 | ||
1762 | return _double_lock_balance(this_rq, busiest); | |
1763 | } | |
1764 | ||
70574a99 AD |
1765 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1766 | __releases(busiest->lock) | |
1767 | { | |
05fa785c | 1768 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1769 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1770 | } | |
1e3c88bd PZ |
1771 | |
1772 | /* | |
1773 | * double_rq_lock - safely lock two runqueues | |
1774 | * | |
1775 | * Note this does not disable interrupts like task_rq_lock, | |
1776 | * you need to do so manually before calling. | |
1777 | */ | |
1778 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1779 | __acquires(rq1->lock) | |
1780 | __acquires(rq2->lock) | |
1781 | { | |
1782 | BUG_ON(!irqs_disabled()); | |
1783 | if (rq1 == rq2) { | |
1784 | raw_spin_lock(&rq1->lock); | |
1785 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1786 | } else { | |
1787 | if (rq1 < rq2) { | |
1788 | raw_spin_lock(&rq1->lock); | |
1789 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1790 | } else { | |
1791 | raw_spin_lock(&rq2->lock); | |
1792 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1793 | } | |
1794 | } | |
1e3c88bd PZ |
1795 | } |
1796 | ||
1797 | /* | |
1798 | * double_rq_unlock - safely unlock two runqueues | |
1799 | * | |
1800 | * Note this does not restore interrupts like task_rq_unlock, | |
1801 | * you need to do so manually after calling. | |
1802 | */ | |
1803 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1804 | __releases(rq1->lock) | |
1805 | __releases(rq2->lock) | |
1806 | { | |
1807 | raw_spin_unlock(&rq1->lock); | |
1808 | if (rq1 != rq2) | |
1809 | raw_spin_unlock(&rq2->lock); | |
1810 | else | |
1811 | __release(rq2->lock); | |
1812 | } | |
1813 | ||
18d95a28 PZ |
1814 | #endif |
1815 | ||
30432094 | 1816 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1817 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1818 | { | |
30432094 | 1819 | #ifdef CONFIG_SMP |
34e83e85 IM |
1820 | cfs_rq->shares = shares; |
1821 | #endif | |
1822 | } | |
30432094 | 1823 | #endif |
e7693a36 | 1824 | |
74f5187a | 1825 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1826 | static void update_sysctl(void); |
acb4a848 | 1827 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1828 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1829 | |
cd29fe6f PZ |
1830 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1831 | { | |
1832 | set_task_rq(p, cpu); | |
1833 | #ifdef CONFIG_SMP | |
1834 | /* | |
1835 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1836 | * successfuly executed on another CPU. We must ensure that updates of | |
1837 | * per-task data have been completed by this moment. | |
1838 | */ | |
1839 | smp_wmb(); | |
1840 | task_thread_info(p)->cpu = cpu; | |
1841 | #endif | |
1842 | } | |
dce48a84 | 1843 | |
1e3c88bd | 1844 | static const struct sched_class rt_sched_class; |
dd41f596 IM |
1845 | |
1846 | #define sched_class_highest (&rt_sched_class) | |
1f11eb6a GH |
1847 | #define for_each_class(class) \ |
1848 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1849 | |
1e3c88bd PZ |
1850 | #include "sched_stats.h" |
1851 | ||
c09595f6 | 1852 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1853 | { |
1854 | rq->nr_running++; | |
9c217245 IM |
1855 | } |
1856 | ||
c09595f6 | 1857 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1858 | { |
1859 | rq->nr_running--; | |
9c217245 IM |
1860 | } |
1861 | ||
45bf76df IM |
1862 | static void set_load_weight(struct task_struct *p) |
1863 | { | |
1864 | if (task_has_rt_policy(p)) { | |
e51fd5e2 PZ |
1865 | p->se.load.weight = 0; |
1866 | p->se.load.inv_weight = WMULT_CONST; | |
dd41f596 IM |
1867 | return; |
1868 | } | |
45bf76df | 1869 | |
dd41f596 IM |
1870 | /* |
1871 | * SCHED_IDLE tasks get minimal weight: | |
1872 | */ | |
1873 | if (p->policy == SCHED_IDLE) { | |
1874 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1875 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1876 | return; | |
1877 | } | |
71f8bd46 | 1878 | |
dd41f596 IM |
1879 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1880 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1881 | } |
1882 | ||
371fd7e7 | 1883 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1884 | { |
a64692a3 | 1885 | update_rq_clock(rq); |
dd41f596 | 1886 | sched_info_queued(p); |
371fd7e7 | 1887 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1888 | p->se.on_rq = 1; |
71f8bd46 IM |
1889 | } |
1890 | ||
371fd7e7 | 1891 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1892 | { |
a64692a3 | 1893 | update_rq_clock(rq); |
46ac22ba | 1894 | sched_info_dequeued(p); |
371fd7e7 | 1895 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1896 | p->se.on_rq = 0; |
71f8bd46 IM |
1897 | } |
1898 | ||
1e3c88bd PZ |
1899 | /* |
1900 | * activate_task - move a task to the runqueue. | |
1901 | */ | |
371fd7e7 | 1902 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1903 | { |
1904 | if (task_contributes_to_load(p)) | |
1905 | rq->nr_uninterruptible--; | |
1906 | ||
371fd7e7 | 1907 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1908 | inc_nr_running(rq); |
1909 | } | |
1910 | ||
1911 | /* | |
1912 | * deactivate_task - remove a task from the runqueue. | |
1913 | */ | |
371fd7e7 | 1914 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1915 | { |
1916 | if (task_contributes_to_load(p)) | |
1917 | rq->nr_uninterruptible++; | |
1918 | ||
371fd7e7 | 1919 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1920 | dec_nr_running(rq); |
1921 | } | |
1922 | ||
1923 | #include "sched_idletask.c" | |
1924 | #include "sched_fair.c" | |
1925 | #include "sched_rt.c" | |
1926 | #ifdef CONFIG_SCHED_DEBUG | |
1927 | # include "sched_debug.c" | |
1928 | #endif | |
1929 | ||
14531189 | 1930 | /* |
dd41f596 | 1931 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1932 | */ |
14531189 IM |
1933 | static inline int __normal_prio(struct task_struct *p) |
1934 | { | |
dd41f596 | 1935 | return p->static_prio; |
14531189 IM |
1936 | } |
1937 | ||
b29739f9 IM |
1938 | /* |
1939 | * Calculate the expected normal priority: i.e. priority | |
1940 | * without taking RT-inheritance into account. Might be | |
1941 | * boosted by interactivity modifiers. Changes upon fork, | |
1942 | * setprio syscalls, and whenever the interactivity | |
1943 | * estimator recalculates. | |
1944 | */ | |
36c8b586 | 1945 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1946 | { |
1947 | int prio; | |
1948 | ||
e05606d3 | 1949 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1950 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1951 | else | |
1952 | prio = __normal_prio(p); | |
1953 | return prio; | |
1954 | } | |
1955 | ||
1956 | /* | |
1957 | * Calculate the current priority, i.e. the priority | |
1958 | * taken into account by the scheduler. This value might | |
1959 | * be boosted by RT tasks, or might be boosted by | |
1960 | * interactivity modifiers. Will be RT if the task got | |
1961 | * RT-boosted. If not then it returns p->normal_prio. | |
1962 | */ | |
36c8b586 | 1963 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1964 | { |
1965 | p->normal_prio = normal_prio(p); | |
1966 | /* | |
1967 | * If we are RT tasks or we were boosted to RT priority, | |
1968 | * keep the priority unchanged. Otherwise, update priority | |
1969 | * to the normal priority: | |
1970 | */ | |
1971 | if (!rt_prio(p->prio)) | |
1972 | return p->normal_prio; | |
1973 | return p->prio; | |
1974 | } | |
1975 | ||
1da177e4 LT |
1976 | /** |
1977 | * task_curr - is this task currently executing on a CPU? | |
1978 | * @p: the task in question. | |
1979 | */ | |
36c8b586 | 1980 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1981 | { |
1982 | return cpu_curr(task_cpu(p)) == p; | |
1983 | } | |
1984 | ||
cb469845 SR |
1985 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1986 | const struct sched_class *prev_class, | |
1987 | int oldprio, int running) | |
1988 | { | |
1989 | if (prev_class != p->sched_class) { | |
1990 | if (prev_class->switched_from) | |
1991 | prev_class->switched_from(rq, p, running); | |
1992 | p->sched_class->switched_to(rq, p, running); | |
1993 | } else | |
1994 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1995 | } | |
1996 | ||
1da177e4 | 1997 | #ifdef CONFIG_SMP |
cc367732 IM |
1998 | /* |
1999 | * Is this task likely cache-hot: | |
2000 | */ | |
e7693a36 | 2001 | static int |
cc367732 IM |
2002 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2003 | { | |
2004 | s64 delta; | |
2005 | ||
e6c8fba7 PZ |
2006 | if (p->sched_class != &fair_sched_class) |
2007 | return 0; | |
2008 | ||
f540a608 IM |
2009 | /* |
2010 | * Buddy candidates are cache hot: | |
2011 | */ | |
f685ceac | 2012 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2013 | (&p->se == cfs_rq_of(&p->se)->next || |
2014 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2015 | return 1; |
2016 | ||
6bc1665b IM |
2017 | if (sysctl_sched_migration_cost == -1) |
2018 | return 1; | |
2019 | if (sysctl_sched_migration_cost == 0) | |
2020 | return 0; | |
2021 | ||
cc367732 IM |
2022 | delta = now - p->se.exec_start; |
2023 | ||
2024 | return delta < (s64)sysctl_sched_migration_cost; | |
2025 | } | |
2026 | ||
dd41f596 | 2027 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2028 | { |
e2912009 PZ |
2029 | #ifdef CONFIG_SCHED_DEBUG |
2030 | /* | |
2031 | * We should never call set_task_cpu() on a blocked task, | |
2032 | * ttwu() will sort out the placement. | |
2033 | */ | |
077614ee PZ |
2034 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2035 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2036 | #endif |
2037 | ||
de1d7286 | 2038 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2039 | |
0c69774e PZ |
2040 | if (task_cpu(p) != new_cpu) { |
2041 | p->se.nr_migrations++; | |
2042 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2043 | } | |
dd41f596 IM |
2044 | |
2045 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2046 | } |
2047 | ||
969c7921 | 2048 | struct migration_arg { |
36c8b586 | 2049 | struct task_struct *task; |
1da177e4 | 2050 | int dest_cpu; |
70b97a7f | 2051 | }; |
1da177e4 | 2052 | |
969c7921 TH |
2053 | static int migration_cpu_stop(void *data); |
2054 | ||
1da177e4 LT |
2055 | /* |
2056 | * The task's runqueue lock must be held. | |
2057 | * Returns true if you have to wait for migration thread. | |
2058 | */ | |
969c7921 | 2059 | static bool migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2060 | { |
70b97a7f | 2061 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2062 | |
2063 | /* | |
2064 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2065 | * the next wake-up will properly place the task. |
1da177e4 | 2066 | */ |
969c7921 | 2067 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2068 | } |
2069 | ||
2070 | /* | |
2071 | * wait_task_inactive - wait for a thread to unschedule. | |
2072 | * | |
85ba2d86 RM |
2073 | * If @match_state is nonzero, it's the @p->state value just checked and |
2074 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2075 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2076 | * we return a positive number (its total switch count). If a second call | |
2077 | * a short while later returns the same number, the caller can be sure that | |
2078 | * @p has remained unscheduled the whole time. | |
2079 | * | |
1da177e4 LT |
2080 | * The caller must ensure that the task *will* unschedule sometime soon, |
2081 | * else this function might spin for a *long* time. This function can't | |
2082 | * be called with interrupts off, or it may introduce deadlock with | |
2083 | * smp_call_function() if an IPI is sent by the same process we are | |
2084 | * waiting to become inactive. | |
2085 | */ | |
85ba2d86 | 2086 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2087 | { |
2088 | unsigned long flags; | |
dd41f596 | 2089 | int running, on_rq; |
85ba2d86 | 2090 | unsigned long ncsw; |
70b97a7f | 2091 | struct rq *rq; |
1da177e4 | 2092 | |
3a5c359a AK |
2093 | for (;;) { |
2094 | /* | |
2095 | * We do the initial early heuristics without holding | |
2096 | * any task-queue locks at all. We'll only try to get | |
2097 | * the runqueue lock when things look like they will | |
2098 | * work out! | |
2099 | */ | |
2100 | rq = task_rq(p); | |
fa490cfd | 2101 | |
3a5c359a AK |
2102 | /* |
2103 | * If the task is actively running on another CPU | |
2104 | * still, just relax and busy-wait without holding | |
2105 | * any locks. | |
2106 | * | |
2107 | * NOTE! Since we don't hold any locks, it's not | |
2108 | * even sure that "rq" stays as the right runqueue! | |
2109 | * But we don't care, since "task_running()" will | |
2110 | * return false if the runqueue has changed and p | |
2111 | * is actually now running somewhere else! | |
2112 | */ | |
85ba2d86 RM |
2113 | while (task_running(rq, p)) { |
2114 | if (match_state && unlikely(p->state != match_state)) | |
2115 | return 0; | |
3a5c359a | 2116 | cpu_relax(); |
85ba2d86 | 2117 | } |
fa490cfd | 2118 | |
3a5c359a AK |
2119 | /* |
2120 | * Ok, time to look more closely! We need the rq | |
2121 | * lock now, to be *sure*. If we're wrong, we'll | |
2122 | * just go back and repeat. | |
2123 | */ | |
2124 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2125 | trace_sched_wait_task(p); |
3a5c359a AK |
2126 | running = task_running(rq, p); |
2127 | on_rq = p->se.on_rq; | |
85ba2d86 | 2128 | ncsw = 0; |
f31e11d8 | 2129 | if (!match_state || p->state == match_state) |
93dcf55f | 2130 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2131 | task_rq_unlock(rq, &flags); |
fa490cfd | 2132 | |
85ba2d86 RM |
2133 | /* |
2134 | * If it changed from the expected state, bail out now. | |
2135 | */ | |
2136 | if (unlikely(!ncsw)) | |
2137 | break; | |
2138 | ||
3a5c359a AK |
2139 | /* |
2140 | * Was it really running after all now that we | |
2141 | * checked with the proper locks actually held? | |
2142 | * | |
2143 | * Oops. Go back and try again.. | |
2144 | */ | |
2145 | if (unlikely(running)) { | |
2146 | cpu_relax(); | |
2147 | continue; | |
2148 | } | |
fa490cfd | 2149 | |
3a5c359a AK |
2150 | /* |
2151 | * It's not enough that it's not actively running, | |
2152 | * it must be off the runqueue _entirely_, and not | |
2153 | * preempted! | |
2154 | * | |
80dd99b3 | 2155 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2156 | * running right now), it's preempted, and we should |
2157 | * yield - it could be a while. | |
2158 | */ | |
2159 | if (unlikely(on_rq)) { | |
2160 | schedule_timeout_uninterruptible(1); | |
2161 | continue; | |
2162 | } | |
fa490cfd | 2163 | |
3a5c359a AK |
2164 | /* |
2165 | * Ahh, all good. It wasn't running, and it wasn't | |
2166 | * runnable, which means that it will never become | |
2167 | * running in the future either. We're all done! | |
2168 | */ | |
2169 | break; | |
2170 | } | |
85ba2d86 RM |
2171 | |
2172 | return ncsw; | |
1da177e4 LT |
2173 | } |
2174 | ||
2175 | /*** | |
2176 | * kick_process - kick a running thread to enter/exit the kernel | |
2177 | * @p: the to-be-kicked thread | |
2178 | * | |
2179 | * Cause a process which is running on another CPU to enter | |
2180 | * kernel-mode, without any delay. (to get signals handled.) | |
2181 | * | |
2182 | * NOTE: this function doesnt have to take the runqueue lock, | |
2183 | * because all it wants to ensure is that the remote task enters | |
2184 | * the kernel. If the IPI races and the task has been migrated | |
2185 | * to another CPU then no harm is done and the purpose has been | |
2186 | * achieved as well. | |
2187 | */ | |
36c8b586 | 2188 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2189 | { |
2190 | int cpu; | |
2191 | ||
2192 | preempt_disable(); | |
2193 | cpu = task_cpu(p); | |
2194 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2195 | smp_send_reschedule(cpu); | |
2196 | preempt_enable(); | |
2197 | } | |
b43e3521 | 2198 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2199 | #endif /* CONFIG_SMP */ |
1da177e4 | 2200 | |
0793a61d TG |
2201 | /** |
2202 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2203 | * @p: the task to evaluate | |
2204 | * @func: the function to be called | |
2205 | * @info: the function call argument | |
2206 | * | |
2207 | * Calls the function @func when the task is currently running. This might | |
2208 | * be on the current CPU, which just calls the function directly | |
2209 | */ | |
2210 | void task_oncpu_function_call(struct task_struct *p, | |
2211 | void (*func) (void *info), void *info) | |
2212 | { | |
2213 | int cpu; | |
2214 | ||
2215 | preempt_disable(); | |
2216 | cpu = task_cpu(p); | |
2217 | if (task_curr(p)) | |
2218 | smp_call_function_single(cpu, func, info, 1); | |
2219 | preempt_enable(); | |
2220 | } | |
2221 | ||
970b13ba | 2222 | #ifdef CONFIG_SMP |
30da688e ON |
2223 | /* |
2224 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2225 | */ | |
5da9a0fb PZ |
2226 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2227 | { | |
2228 | int dest_cpu; | |
2229 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2230 | ||
2231 | /* Look for allowed, online CPU in same node. */ | |
2232 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2233 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2234 | return dest_cpu; | |
2235 | ||
2236 | /* Any allowed, online CPU? */ | |
2237 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2238 | if (dest_cpu < nr_cpu_ids) | |
2239 | return dest_cpu; | |
2240 | ||
2241 | /* No more Mr. Nice Guy. */ | |
897f0b3c | 2242 | if (unlikely(dest_cpu >= nr_cpu_ids)) { |
9084bb82 | 2243 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
5da9a0fb PZ |
2244 | /* |
2245 | * Don't tell them about moving exiting tasks or | |
2246 | * kernel threads (both mm NULL), since they never | |
2247 | * leave kernel. | |
2248 | */ | |
2249 | if (p->mm && printk_ratelimit()) { | |
2250 | printk(KERN_INFO "process %d (%s) no " | |
2251 | "longer affine to cpu%d\n", | |
2252 | task_pid_nr(p), p->comm, cpu); | |
2253 | } | |
2254 | } | |
2255 | ||
2256 | return dest_cpu; | |
2257 | } | |
2258 | ||
e2912009 | 2259 | /* |
30da688e | 2260 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2261 | */ |
970b13ba | 2262 | static inline |
0017d735 | 2263 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2264 | { |
0017d735 | 2265 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2266 | |
2267 | /* | |
2268 | * In order not to call set_task_cpu() on a blocking task we need | |
2269 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2270 | * cpu. | |
2271 | * | |
2272 | * Since this is common to all placement strategies, this lives here. | |
2273 | * | |
2274 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2275 | * not worry about this generic constraint ] | |
2276 | */ | |
2277 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2278 | !cpu_online(cpu))) |
5da9a0fb | 2279 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2280 | |
2281 | return cpu; | |
970b13ba | 2282 | } |
09a40af5 MG |
2283 | |
2284 | static void update_avg(u64 *avg, u64 sample) | |
2285 | { | |
2286 | s64 diff = sample - *avg; | |
2287 | *avg += diff >> 3; | |
2288 | } | |
970b13ba PZ |
2289 | #endif |
2290 | ||
9ed3811a TH |
2291 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2292 | bool is_sync, bool is_migrate, bool is_local, | |
2293 | unsigned long en_flags) | |
2294 | { | |
2295 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2296 | if (is_sync) | |
2297 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2298 | if (is_migrate) | |
2299 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2300 | if (is_local) | |
2301 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2302 | else | |
2303 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2304 | ||
2305 | activate_task(rq, p, en_flags); | |
2306 | } | |
2307 | ||
2308 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2309 | int wake_flags, bool success) | |
2310 | { | |
2311 | trace_sched_wakeup(p, success); | |
2312 | check_preempt_curr(rq, p, wake_flags); | |
2313 | ||
2314 | p->state = TASK_RUNNING; | |
2315 | #ifdef CONFIG_SMP | |
2316 | if (p->sched_class->task_woken) | |
2317 | p->sched_class->task_woken(rq, p); | |
2318 | ||
2319 | if (unlikely(rq->idle_stamp)) { | |
2320 | u64 delta = rq->clock - rq->idle_stamp; | |
2321 | u64 max = 2*sysctl_sched_migration_cost; | |
2322 | ||
2323 | if (delta > max) | |
2324 | rq->avg_idle = max; | |
2325 | else | |
2326 | update_avg(&rq->avg_idle, delta); | |
2327 | rq->idle_stamp = 0; | |
2328 | } | |
2329 | #endif | |
21aa9af0 TH |
2330 | /* if a worker is waking up, notify workqueue */ |
2331 | if ((p->flags & PF_WQ_WORKER) && success) | |
2332 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2333 | } |
2334 | ||
2335 | /** | |
1da177e4 | 2336 | * try_to_wake_up - wake up a thread |
9ed3811a | 2337 | * @p: the thread to be awakened |
1da177e4 | 2338 | * @state: the mask of task states that can be woken |
9ed3811a | 2339 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2340 | * |
2341 | * Put it on the run-queue if it's not already there. The "current" | |
2342 | * thread is always on the run-queue (except when the actual | |
2343 | * re-schedule is in progress), and as such you're allowed to do | |
2344 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2345 | * runnable without the overhead of this. | |
2346 | * | |
9ed3811a TH |
2347 | * Returns %true if @p was woken up, %false if it was already running |
2348 | * or @state didn't match @p's state. | |
1da177e4 | 2349 | */ |
7d478721 PZ |
2350 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2351 | int wake_flags) | |
1da177e4 | 2352 | { |
cc367732 | 2353 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2354 | unsigned long flags; |
371fd7e7 | 2355 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2356 | struct rq *rq; |
1da177e4 | 2357 | |
e9c84311 | 2358 | this_cpu = get_cpu(); |
2398f2c6 | 2359 | |
04e2f174 | 2360 | smp_wmb(); |
ab3b3aa5 | 2361 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2362 | if (!(p->state & state)) |
1da177e4 LT |
2363 | goto out; |
2364 | ||
dd41f596 | 2365 | if (p->se.on_rq) |
1da177e4 LT |
2366 | goto out_running; |
2367 | ||
2368 | cpu = task_cpu(p); | |
cc367732 | 2369 | orig_cpu = cpu; |
1da177e4 LT |
2370 | |
2371 | #ifdef CONFIG_SMP | |
2372 | if (unlikely(task_running(rq, p))) | |
2373 | goto out_activate; | |
2374 | ||
e9c84311 PZ |
2375 | /* |
2376 | * In order to handle concurrent wakeups and release the rq->lock | |
2377 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2378 | * |
2379 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2380 | */ |
cc87f76a PZ |
2381 | if (task_contributes_to_load(p)) { |
2382 | if (likely(cpu_online(orig_cpu))) | |
2383 | rq->nr_uninterruptible--; | |
2384 | else | |
2385 | this_rq()->nr_uninterruptible--; | |
2386 | } | |
e9c84311 | 2387 | p->state = TASK_WAKING; |
efbbd05a | 2388 | |
371fd7e7 | 2389 | if (p->sched_class->task_waking) { |
efbbd05a | 2390 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2391 | en_flags |= ENQUEUE_WAKING; |
2392 | } | |
efbbd05a | 2393 | |
0017d735 PZ |
2394 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2395 | if (cpu != orig_cpu) | |
5d2f5a61 | 2396 | set_task_cpu(p, cpu); |
0017d735 | 2397 | __task_rq_unlock(rq); |
ab19cb23 | 2398 | |
0970d299 PZ |
2399 | rq = cpu_rq(cpu); |
2400 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2401 | |
0970d299 PZ |
2402 | /* |
2403 | * We migrated the task without holding either rq->lock, however | |
2404 | * since the task is not on the task list itself, nobody else | |
2405 | * will try and migrate the task, hence the rq should match the | |
2406 | * cpu we just moved it to. | |
2407 | */ | |
2408 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2409 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2410 | |
e7693a36 GH |
2411 | #ifdef CONFIG_SCHEDSTATS |
2412 | schedstat_inc(rq, ttwu_count); | |
2413 | if (cpu == this_cpu) | |
2414 | schedstat_inc(rq, ttwu_local); | |
2415 | else { | |
2416 | struct sched_domain *sd; | |
2417 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2418 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2419 | schedstat_inc(sd, ttwu_wake_remote); |
2420 | break; | |
2421 | } | |
2422 | } | |
2423 | } | |
6d6bc0ad | 2424 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2425 | |
1da177e4 LT |
2426 | out_activate: |
2427 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2428 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2429 | cpu == this_cpu, en_flags); | |
1da177e4 | 2430 | success = 1; |
1da177e4 | 2431 | out_running: |
9ed3811a | 2432 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2433 | out: |
2434 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2435 | put_cpu(); |
1da177e4 LT |
2436 | |
2437 | return success; | |
2438 | } | |
2439 | ||
21aa9af0 TH |
2440 | /** |
2441 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2442 | * @p: the thread to be awakened | |
2443 | * | |
2444 | * Put @p on the run-queue if it's not alredy there. The caller must | |
2445 | * ensure that this_rq() is locked, @p is bound to this_rq() and not | |
2446 | * the current task. this_rq() stays locked over invocation. | |
2447 | */ | |
2448 | static void try_to_wake_up_local(struct task_struct *p) | |
2449 | { | |
2450 | struct rq *rq = task_rq(p); | |
2451 | bool success = false; | |
2452 | ||
2453 | BUG_ON(rq != this_rq()); | |
2454 | BUG_ON(p == current); | |
2455 | lockdep_assert_held(&rq->lock); | |
2456 | ||
2457 | if (!(p->state & TASK_NORMAL)) | |
2458 | return; | |
2459 | ||
2460 | if (!p->se.on_rq) { | |
2461 | if (likely(!task_running(rq, p))) { | |
2462 | schedstat_inc(rq, ttwu_count); | |
2463 | schedstat_inc(rq, ttwu_local); | |
2464 | } | |
2465 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2466 | success = true; | |
2467 | } | |
2468 | ttwu_post_activation(p, rq, 0, success); | |
2469 | } | |
2470 | ||
50fa610a DH |
2471 | /** |
2472 | * wake_up_process - Wake up a specific process | |
2473 | * @p: The process to be woken up. | |
2474 | * | |
2475 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2476 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2477 | * running. | |
2478 | * | |
2479 | * It may be assumed that this function implies a write memory barrier before | |
2480 | * changing the task state if and only if any tasks are woken up. | |
2481 | */ | |
7ad5b3a5 | 2482 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2483 | { |
d9514f6c | 2484 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2485 | } |
1da177e4 LT |
2486 | EXPORT_SYMBOL(wake_up_process); |
2487 | ||
7ad5b3a5 | 2488 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2489 | { |
2490 | return try_to_wake_up(p, state, 0); | |
2491 | } | |
2492 | ||
1da177e4 LT |
2493 | /* |
2494 | * Perform scheduler related setup for a newly forked process p. | |
2495 | * p is forked by current. | |
dd41f596 IM |
2496 | * |
2497 | * __sched_fork() is basic setup used by init_idle() too: | |
2498 | */ | |
2499 | static void __sched_fork(struct task_struct *p) | |
2500 | { | |
dd41f596 IM |
2501 | p->se.exec_start = 0; |
2502 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2503 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2504 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2505 | |
2506 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2507 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2508 | #endif |
476d139c | 2509 | |
fa717060 | 2510 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2511 | p->se.on_rq = 0; |
4a55bd5e | 2512 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2513 | |
e107be36 AK |
2514 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2515 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2516 | #endif | |
dd41f596 IM |
2517 | } |
2518 | ||
2519 | /* | |
2520 | * fork()/clone()-time setup: | |
2521 | */ | |
2522 | void sched_fork(struct task_struct *p, int clone_flags) | |
2523 | { | |
2524 | int cpu = get_cpu(); | |
2525 | ||
2526 | __sched_fork(p); | |
06b83b5f | 2527 | /* |
0017d735 | 2528 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2529 | * nobody will actually run it, and a signal or other external |
2530 | * event cannot wake it up and insert it on the runqueue either. | |
2531 | */ | |
0017d735 | 2532 | p->state = TASK_RUNNING; |
dd41f596 | 2533 | |
b9dc29e7 MG |
2534 | /* |
2535 | * Revert to default priority/policy on fork if requested. | |
2536 | */ | |
2537 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2538 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2539 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2540 | p->normal_prio = p->static_prio; |
2541 | } | |
b9dc29e7 | 2542 | |
6c697bdf MG |
2543 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2544 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2545 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2546 | set_load_weight(p); |
2547 | } | |
2548 | ||
b9dc29e7 MG |
2549 | /* |
2550 | * We don't need the reset flag anymore after the fork. It has | |
2551 | * fulfilled its duty: | |
2552 | */ | |
2553 | p->sched_reset_on_fork = 0; | |
2554 | } | |
ca94c442 | 2555 | |
f83f9ac2 PW |
2556 | /* |
2557 | * Make sure we do not leak PI boosting priority to the child. | |
2558 | */ | |
2559 | p->prio = current->normal_prio; | |
2560 | ||
2ddbf952 HS |
2561 | if (!rt_prio(p->prio)) |
2562 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2563 | |
cd29fe6f PZ |
2564 | if (p->sched_class->task_fork) |
2565 | p->sched_class->task_fork(p); | |
2566 | ||
5f3edc1b PZ |
2567 | set_task_cpu(p, cpu); |
2568 | ||
52f17b6c | 2569 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2570 | if (likely(sched_info_on())) |
52f17b6c | 2571 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2572 | #endif |
d6077cb8 | 2573 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2574 | p->oncpu = 0; |
2575 | #endif | |
1da177e4 | 2576 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2577 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2578 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2579 | #endif |
917b627d GH |
2580 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2581 | ||
476d139c | 2582 | put_cpu(); |
1da177e4 LT |
2583 | } |
2584 | ||
2585 | /* | |
2586 | * wake_up_new_task - wake up a newly created task for the first time. | |
2587 | * | |
2588 | * This function will do some initial scheduler statistics housekeeping | |
2589 | * that must be done for every newly created context, then puts the task | |
2590 | * on the runqueue and wakes it. | |
2591 | */ | |
7ad5b3a5 | 2592 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2593 | { |
2594 | unsigned long flags; | |
dd41f596 | 2595 | struct rq *rq; |
c890692b | 2596 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2597 | |
2598 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2599 | rq = task_rq_lock(p, &flags); |
2600 | p->state = TASK_WAKING; | |
2601 | ||
fabf318e PZ |
2602 | /* |
2603 | * Fork balancing, do it here and not earlier because: | |
2604 | * - cpus_allowed can change in the fork path | |
2605 | * - any previously selected cpu might disappear through hotplug | |
2606 | * | |
0017d735 PZ |
2607 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2608 | * without people poking at ->cpus_allowed. | |
fabf318e | 2609 | */ |
0017d735 | 2610 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2611 | set_task_cpu(p, cpu); |
1da177e4 | 2612 | |
06b83b5f | 2613 | p->state = TASK_RUNNING; |
0017d735 PZ |
2614 | task_rq_unlock(rq, &flags); |
2615 | #endif | |
2616 | ||
2617 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2618 | activate_task(rq, p, 0); |
27a9da65 | 2619 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2620 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2621 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2622 | if (p->sched_class->task_woken) |
2623 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2624 | #endif |
dd41f596 | 2625 | task_rq_unlock(rq, &flags); |
fabf318e | 2626 | put_cpu(); |
1da177e4 LT |
2627 | } |
2628 | ||
e107be36 AK |
2629 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2630 | ||
2631 | /** | |
80dd99b3 | 2632 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2633 | * @notifier: notifier struct to register |
e107be36 AK |
2634 | */ |
2635 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2636 | { | |
2637 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2638 | } | |
2639 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2640 | ||
2641 | /** | |
2642 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2643 | * @notifier: notifier struct to unregister |
e107be36 AK |
2644 | * |
2645 | * This is safe to call from within a preemption notifier. | |
2646 | */ | |
2647 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2648 | { | |
2649 | hlist_del(¬ifier->link); | |
2650 | } | |
2651 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2652 | ||
2653 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2654 | { | |
2655 | struct preempt_notifier *notifier; | |
2656 | struct hlist_node *node; | |
2657 | ||
2658 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2659 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2660 | } | |
2661 | ||
2662 | static void | |
2663 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2664 | struct task_struct *next) | |
2665 | { | |
2666 | struct preempt_notifier *notifier; | |
2667 | struct hlist_node *node; | |
2668 | ||
2669 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2670 | notifier->ops->sched_out(notifier, next); | |
2671 | } | |
2672 | ||
6d6bc0ad | 2673 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2674 | |
2675 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2676 | { | |
2677 | } | |
2678 | ||
2679 | static void | |
2680 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2681 | struct task_struct *next) | |
2682 | { | |
2683 | } | |
2684 | ||
6d6bc0ad | 2685 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2686 | |
4866cde0 NP |
2687 | /** |
2688 | * prepare_task_switch - prepare to switch tasks | |
2689 | * @rq: the runqueue preparing to switch | |
421cee29 | 2690 | * @prev: the current task that is being switched out |
4866cde0 NP |
2691 | * @next: the task we are going to switch to. |
2692 | * | |
2693 | * This is called with the rq lock held and interrupts off. It must | |
2694 | * be paired with a subsequent finish_task_switch after the context | |
2695 | * switch. | |
2696 | * | |
2697 | * prepare_task_switch sets up locking and calls architecture specific | |
2698 | * hooks. | |
2699 | */ | |
e107be36 AK |
2700 | static inline void |
2701 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2702 | struct task_struct *next) | |
4866cde0 | 2703 | { |
e107be36 | 2704 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2705 | prepare_lock_switch(rq, next); |
2706 | prepare_arch_switch(next); | |
2707 | } | |
2708 | ||
1da177e4 LT |
2709 | /** |
2710 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2711 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2712 | * @prev: the thread we just switched away from. |
2713 | * | |
4866cde0 NP |
2714 | * finish_task_switch must be called after the context switch, paired |
2715 | * with a prepare_task_switch call before the context switch. | |
2716 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2717 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2718 | * |
2719 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2720 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2721 | * with the lock held can cause deadlocks; see schedule() for |
2722 | * details.) | |
2723 | */ | |
a9957449 | 2724 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2725 | __releases(rq->lock) |
2726 | { | |
1da177e4 | 2727 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2728 | long prev_state; |
1da177e4 LT |
2729 | |
2730 | rq->prev_mm = NULL; | |
2731 | ||
2732 | /* | |
2733 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2734 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2735 | * schedule one last time. The schedule call will never return, and |
2736 | * the scheduled task must drop that reference. | |
c394cc9f | 2737 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2738 | * still held, otherwise prev could be scheduled on another cpu, die |
2739 | * there before we look at prev->state, and then the reference would | |
2740 | * be dropped twice. | |
2741 | * Manfred Spraul <manfred@colorfullife.com> | |
2742 | */ | |
55a101f8 | 2743 | prev_state = prev->state; |
4866cde0 | 2744 | finish_arch_switch(prev); |
8381f65d JI |
2745 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2746 | local_irq_disable(); | |
2747 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2748 | perf_event_task_sched_in(current); |
8381f65d JI |
2749 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2750 | local_irq_enable(); | |
2751 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2752 | finish_lock_switch(rq, prev); |
e8fa1362 | 2753 | |
e107be36 | 2754 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2755 | if (mm) |
2756 | mmdrop(mm); | |
c394cc9f | 2757 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2758 | /* |
2759 | * Remove function-return probe instances associated with this | |
2760 | * task and put them back on the free list. | |
9761eea8 | 2761 | */ |
c6fd91f0 | 2762 | kprobe_flush_task(prev); |
1da177e4 | 2763 | put_task_struct(prev); |
c6fd91f0 | 2764 | } |
1da177e4 LT |
2765 | } |
2766 | ||
3f029d3c GH |
2767 | #ifdef CONFIG_SMP |
2768 | ||
2769 | /* assumes rq->lock is held */ | |
2770 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2771 | { | |
2772 | if (prev->sched_class->pre_schedule) | |
2773 | prev->sched_class->pre_schedule(rq, prev); | |
2774 | } | |
2775 | ||
2776 | /* rq->lock is NOT held, but preemption is disabled */ | |
2777 | static inline void post_schedule(struct rq *rq) | |
2778 | { | |
2779 | if (rq->post_schedule) { | |
2780 | unsigned long flags; | |
2781 | ||
05fa785c | 2782 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2783 | if (rq->curr->sched_class->post_schedule) |
2784 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2785 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2786 | |
2787 | rq->post_schedule = 0; | |
2788 | } | |
2789 | } | |
2790 | ||
2791 | #else | |
da19ab51 | 2792 | |
3f029d3c GH |
2793 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2794 | { | |
2795 | } | |
2796 | ||
2797 | static inline void post_schedule(struct rq *rq) | |
2798 | { | |
1da177e4 LT |
2799 | } |
2800 | ||
3f029d3c GH |
2801 | #endif |
2802 | ||
1da177e4 LT |
2803 | /** |
2804 | * schedule_tail - first thing a freshly forked thread must call. | |
2805 | * @prev: the thread we just switched away from. | |
2806 | */ | |
36c8b586 | 2807 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2808 | __releases(rq->lock) |
2809 | { | |
70b97a7f IM |
2810 | struct rq *rq = this_rq(); |
2811 | ||
4866cde0 | 2812 | finish_task_switch(rq, prev); |
da19ab51 | 2813 | |
3f029d3c GH |
2814 | /* |
2815 | * FIXME: do we need to worry about rq being invalidated by the | |
2816 | * task_switch? | |
2817 | */ | |
2818 | post_schedule(rq); | |
70b97a7f | 2819 | |
4866cde0 NP |
2820 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2821 | /* In this case, finish_task_switch does not reenable preemption */ | |
2822 | preempt_enable(); | |
2823 | #endif | |
1da177e4 | 2824 | if (current->set_child_tid) |
b488893a | 2825 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2826 | } |
2827 | ||
2828 | /* | |
2829 | * context_switch - switch to the new MM and the new | |
2830 | * thread's register state. | |
2831 | */ | |
dd41f596 | 2832 | static inline void |
70b97a7f | 2833 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2834 | struct task_struct *next) |
1da177e4 | 2835 | { |
dd41f596 | 2836 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2837 | |
e107be36 | 2838 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2839 | trace_sched_switch(prev, next); |
dd41f596 IM |
2840 | mm = next->mm; |
2841 | oldmm = prev->active_mm; | |
9226d125 ZA |
2842 | /* |
2843 | * For paravirt, this is coupled with an exit in switch_to to | |
2844 | * combine the page table reload and the switch backend into | |
2845 | * one hypercall. | |
2846 | */ | |
224101ed | 2847 | arch_start_context_switch(prev); |
9226d125 | 2848 | |
710390d9 | 2849 | if (likely(!mm)) { |
1da177e4 LT |
2850 | next->active_mm = oldmm; |
2851 | atomic_inc(&oldmm->mm_count); | |
2852 | enter_lazy_tlb(oldmm, next); | |
2853 | } else | |
2854 | switch_mm(oldmm, mm, next); | |
2855 | ||
710390d9 | 2856 | if (likely(!prev->mm)) { |
1da177e4 | 2857 | prev->active_mm = NULL; |
1da177e4 LT |
2858 | rq->prev_mm = oldmm; |
2859 | } | |
3a5f5e48 IM |
2860 | /* |
2861 | * Since the runqueue lock will be released by the next | |
2862 | * task (which is an invalid locking op but in the case | |
2863 | * of the scheduler it's an obvious special-case), so we | |
2864 | * do an early lockdep release here: | |
2865 | */ | |
2866 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2867 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2868 | #endif |
1da177e4 LT |
2869 | |
2870 | /* Here we just switch the register state and the stack. */ | |
2871 | switch_to(prev, next, prev); | |
2872 | ||
dd41f596 IM |
2873 | barrier(); |
2874 | /* | |
2875 | * this_rq must be evaluated again because prev may have moved | |
2876 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2877 | * frame will be invalid. | |
2878 | */ | |
2879 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2880 | } |
2881 | ||
2882 | /* | |
2883 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2884 | * | |
2885 | * externally visible scheduler statistics: current number of runnable | |
2886 | * threads, current number of uninterruptible-sleeping threads, total | |
2887 | * number of context switches performed since bootup. | |
2888 | */ | |
2889 | unsigned long nr_running(void) | |
2890 | { | |
2891 | unsigned long i, sum = 0; | |
2892 | ||
2893 | for_each_online_cpu(i) | |
2894 | sum += cpu_rq(i)->nr_running; | |
2895 | ||
2896 | return sum; | |
f711f609 | 2897 | } |
1da177e4 LT |
2898 | |
2899 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2900 | { |
1da177e4 | 2901 | unsigned long i, sum = 0; |
f711f609 | 2902 | |
0a945022 | 2903 | for_each_possible_cpu(i) |
1da177e4 | 2904 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2905 | |
2906 | /* | |
1da177e4 LT |
2907 | * Since we read the counters lockless, it might be slightly |
2908 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2909 | */ |
1da177e4 LT |
2910 | if (unlikely((long)sum < 0)) |
2911 | sum = 0; | |
f711f609 | 2912 | |
1da177e4 | 2913 | return sum; |
f711f609 | 2914 | } |
f711f609 | 2915 | |
1da177e4 | 2916 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2917 | { |
cc94abfc SR |
2918 | int i; |
2919 | unsigned long long sum = 0; | |
46cb4b7c | 2920 | |
0a945022 | 2921 | for_each_possible_cpu(i) |
1da177e4 | 2922 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2923 | |
1da177e4 LT |
2924 | return sum; |
2925 | } | |
483b4ee6 | 2926 | |
1da177e4 LT |
2927 | unsigned long nr_iowait(void) |
2928 | { | |
2929 | unsigned long i, sum = 0; | |
483b4ee6 | 2930 | |
0a945022 | 2931 | for_each_possible_cpu(i) |
1da177e4 | 2932 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2933 | |
1da177e4 LT |
2934 | return sum; |
2935 | } | |
483b4ee6 | 2936 | |
69d25870 AV |
2937 | unsigned long nr_iowait_cpu(void) |
2938 | { | |
2939 | struct rq *this = this_rq(); | |
2940 | return atomic_read(&this->nr_iowait); | |
2941 | } | |
46cb4b7c | 2942 | |
69d25870 AV |
2943 | unsigned long this_cpu_load(void) |
2944 | { | |
2945 | struct rq *this = this_rq(); | |
2946 | return this->cpu_load[0]; | |
2947 | } | |
e790fb0b | 2948 | |
46cb4b7c | 2949 | |
dce48a84 TG |
2950 | /* Variables and functions for calc_load */ |
2951 | static atomic_long_t calc_load_tasks; | |
2952 | static unsigned long calc_load_update; | |
2953 | unsigned long avenrun[3]; | |
2954 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2955 | |
74f5187a PZ |
2956 | static long calc_load_fold_active(struct rq *this_rq) |
2957 | { | |
2958 | long nr_active, delta = 0; | |
2959 | ||
2960 | nr_active = this_rq->nr_running; | |
2961 | nr_active += (long) this_rq->nr_uninterruptible; | |
2962 | ||
2963 | if (nr_active != this_rq->calc_load_active) { | |
2964 | delta = nr_active - this_rq->calc_load_active; | |
2965 | this_rq->calc_load_active = nr_active; | |
2966 | } | |
2967 | ||
2968 | return delta; | |
2969 | } | |
2970 | ||
2971 | #ifdef CONFIG_NO_HZ | |
2972 | /* | |
2973 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2974 | * | |
2975 | * When making the ILB scale, we should try to pull this in as well. | |
2976 | */ | |
2977 | static atomic_long_t calc_load_tasks_idle; | |
2978 | ||
2979 | static void calc_load_account_idle(struct rq *this_rq) | |
2980 | { | |
2981 | long delta; | |
2982 | ||
2983 | delta = calc_load_fold_active(this_rq); | |
2984 | if (delta) | |
2985 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2986 | } | |
2987 | ||
2988 | static long calc_load_fold_idle(void) | |
2989 | { | |
2990 | long delta = 0; | |
2991 | ||
2992 | /* | |
2993 | * Its got a race, we don't care... | |
2994 | */ | |
2995 | if (atomic_long_read(&calc_load_tasks_idle)) | |
2996 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
2997 | ||
2998 | return delta; | |
2999 | } | |
3000 | #else | |
3001 | static void calc_load_account_idle(struct rq *this_rq) | |
3002 | { | |
3003 | } | |
3004 | ||
3005 | static inline long calc_load_fold_idle(void) | |
3006 | { | |
3007 | return 0; | |
3008 | } | |
3009 | #endif | |
3010 | ||
2d02494f TG |
3011 | /** |
3012 | * get_avenrun - get the load average array | |
3013 | * @loads: pointer to dest load array | |
3014 | * @offset: offset to add | |
3015 | * @shift: shift count to shift the result left | |
3016 | * | |
3017 | * These values are estimates at best, so no need for locking. | |
3018 | */ | |
3019 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3020 | { | |
3021 | loads[0] = (avenrun[0] + offset) << shift; | |
3022 | loads[1] = (avenrun[1] + offset) << shift; | |
3023 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3024 | } |
46cb4b7c | 3025 | |
dce48a84 TG |
3026 | static unsigned long |
3027 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3028 | { |
dce48a84 TG |
3029 | load *= exp; |
3030 | load += active * (FIXED_1 - exp); | |
3031 | return load >> FSHIFT; | |
3032 | } | |
46cb4b7c SS |
3033 | |
3034 | /* | |
dce48a84 TG |
3035 | * calc_load - update the avenrun load estimates 10 ticks after the |
3036 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3037 | */ |
dce48a84 | 3038 | void calc_global_load(void) |
7835b98b | 3039 | { |
dce48a84 TG |
3040 | unsigned long upd = calc_load_update + 10; |
3041 | long active; | |
1da177e4 | 3042 | |
dce48a84 TG |
3043 | if (time_before(jiffies, upd)) |
3044 | return; | |
1da177e4 | 3045 | |
dce48a84 TG |
3046 | active = atomic_long_read(&calc_load_tasks); |
3047 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3048 | |
dce48a84 TG |
3049 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3050 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3051 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3052 | |
dce48a84 TG |
3053 | calc_load_update += LOAD_FREQ; |
3054 | } | |
1da177e4 | 3055 | |
dce48a84 | 3056 | /* |
74f5187a PZ |
3057 | * Called from update_cpu_load() to periodically update this CPU's |
3058 | * active count. | |
dce48a84 TG |
3059 | */ |
3060 | static void calc_load_account_active(struct rq *this_rq) | |
3061 | { | |
74f5187a | 3062 | long delta; |
08c183f3 | 3063 | |
74f5187a PZ |
3064 | if (time_before(jiffies, this_rq->calc_load_update)) |
3065 | return; | |
783609c6 | 3066 | |
74f5187a PZ |
3067 | delta = calc_load_fold_active(this_rq); |
3068 | delta += calc_load_fold_idle(); | |
3069 | if (delta) | |
dce48a84 | 3070 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3071 | |
3072 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3073 | } |
3074 | ||
fdf3e95d VP |
3075 | /* |
3076 | * The exact cpuload at various idx values, calculated at every tick would be | |
3077 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3078 | * | |
3079 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3080 | * on nth tick when cpu may be busy, then we have: | |
3081 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3082 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3083 | * | |
3084 | * decay_load_missed() below does efficient calculation of | |
3085 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3086 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3087 | * | |
3088 | * The calculation is approximated on a 128 point scale. | |
3089 | * degrade_zero_ticks is the number of ticks after which load at any | |
3090 | * particular idx is approximated to be zero. | |
3091 | * degrade_factor is a precomputed table, a row for each load idx. | |
3092 | * Each column corresponds to degradation factor for a power of two ticks, | |
3093 | * based on 128 point scale. | |
3094 | * Example: | |
3095 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3096 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3097 | * | |
3098 | * With this power of 2 load factors, we can degrade the load n times | |
3099 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3100 | * n mult/shifts needed by the exact degradation. | |
3101 | */ | |
3102 | #define DEGRADE_SHIFT 7 | |
3103 | static const unsigned char | |
3104 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3105 | static const unsigned char | |
3106 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3107 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3108 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3109 | {96, 72, 40, 12, 1, 0, 0}, | |
3110 | {112, 98, 75, 43, 15, 1, 0}, | |
3111 | {120, 112, 98, 76, 45, 16, 2} }; | |
3112 | ||
3113 | /* | |
3114 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3115 | * would be when CPU is idle and so we just decay the old load without | |
3116 | * adding any new load. | |
3117 | */ | |
3118 | static unsigned long | |
3119 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3120 | { | |
3121 | int j = 0; | |
3122 | ||
3123 | if (!missed_updates) | |
3124 | return load; | |
3125 | ||
3126 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3127 | return 0; | |
3128 | ||
3129 | if (idx == 1) | |
3130 | return load >> missed_updates; | |
3131 | ||
3132 | while (missed_updates) { | |
3133 | if (missed_updates % 2) | |
3134 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3135 | ||
3136 | missed_updates >>= 1; | |
3137 | j++; | |
3138 | } | |
3139 | return load; | |
3140 | } | |
3141 | ||
46cb4b7c | 3142 | /* |
dd41f596 | 3143 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3144 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3145 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3146 | */ |
dd41f596 | 3147 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3148 | { |
495eca49 | 3149 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3150 | unsigned long curr_jiffies = jiffies; |
3151 | unsigned long pending_updates; | |
dd41f596 | 3152 | int i, scale; |
46cb4b7c | 3153 | |
dd41f596 | 3154 | this_rq->nr_load_updates++; |
46cb4b7c | 3155 | |
fdf3e95d VP |
3156 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3157 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3158 | return; | |
3159 | ||
3160 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3161 | this_rq->last_load_update_tick = curr_jiffies; | |
3162 | ||
dd41f596 | 3163 | /* Update our load: */ |
fdf3e95d VP |
3164 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3165 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3166 | unsigned long old_load, new_load; |
7d1e6a9b | 3167 | |
dd41f596 | 3168 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3169 | |
dd41f596 | 3170 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3171 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3172 | new_load = this_load; |
a25707f3 IM |
3173 | /* |
3174 | * Round up the averaging division if load is increasing. This | |
3175 | * prevents us from getting stuck on 9 if the load is 10, for | |
3176 | * example. | |
3177 | */ | |
3178 | if (new_load > old_load) | |
fdf3e95d VP |
3179 | new_load += scale - 1; |
3180 | ||
3181 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3182 | } |
fdf3e95d VP |
3183 | } |
3184 | ||
3185 | static void update_cpu_load_active(struct rq *this_rq) | |
3186 | { | |
3187 | update_cpu_load(this_rq); | |
46cb4b7c | 3188 | |
74f5187a | 3189 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3190 | } |
3191 | ||
dd41f596 | 3192 | #ifdef CONFIG_SMP |
8a0be9ef | 3193 | |
46cb4b7c | 3194 | /* |
38022906 PZ |
3195 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3196 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3197 | */ |
38022906 | 3198 | void sched_exec(void) |
46cb4b7c | 3199 | { |
38022906 | 3200 | struct task_struct *p = current; |
1da177e4 | 3201 | unsigned long flags; |
70b97a7f | 3202 | struct rq *rq; |
0017d735 | 3203 | int dest_cpu; |
46cb4b7c | 3204 | |
1da177e4 | 3205 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3206 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3207 | if (dest_cpu == smp_processor_id()) | |
3208 | goto unlock; | |
38022906 | 3209 | |
46cb4b7c | 3210 | /* |
38022906 | 3211 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3212 | */ |
30da688e | 3213 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
969c7921 TH |
3214 | likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) { |
3215 | struct migration_arg arg = { p, dest_cpu }; | |
46cb4b7c | 3216 | |
1da177e4 | 3217 | task_rq_unlock(rq, &flags); |
969c7921 | 3218 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3219 | return; |
3220 | } | |
0017d735 | 3221 | unlock: |
1da177e4 | 3222 | task_rq_unlock(rq, &flags); |
1da177e4 | 3223 | } |
dd41f596 | 3224 | |
1da177e4 LT |
3225 | #endif |
3226 | ||
1da177e4 LT |
3227 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3228 | ||
3229 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3230 | ||
3231 | /* | |
c5f8d995 | 3232 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3233 | * @p in case that task is currently running. |
c5f8d995 HS |
3234 | * |
3235 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3236 | */ |
c5f8d995 HS |
3237 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3238 | { | |
3239 | u64 ns = 0; | |
3240 | ||
3241 | if (task_current(rq, p)) { | |
3242 | update_rq_clock(rq); | |
3243 | ns = rq->clock - p->se.exec_start; | |
3244 | if ((s64)ns < 0) | |
3245 | ns = 0; | |
3246 | } | |
3247 | ||
3248 | return ns; | |
3249 | } | |
3250 | ||
bb34d92f | 3251 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3252 | { |
1da177e4 | 3253 | unsigned long flags; |
41b86e9c | 3254 | struct rq *rq; |
bb34d92f | 3255 | u64 ns = 0; |
48f24c4d | 3256 | |
41b86e9c | 3257 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3258 | ns = do_task_delta_exec(p, rq); |
3259 | task_rq_unlock(rq, &flags); | |
1508487e | 3260 | |
c5f8d995 HS |
3261 | return ns; |
3262 | } | |
f06febc9 | 3263 | |
c5f8d995 HS |
3264 | /* |
3265 | * Return accounted runtime for the task. | |
3266 | * In case the task is currently running, return the runtime plus current's | |
3267 | * pending runtime that have not been accounted yet. | |
3268 | */ | |
3269 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3270 | { | |
3271 | unsigned long flags; | |
3272 | struct rq *rq; | |
3273 | u64 ns = 0; | |
3274 | ||
3275 | rq = task_rq_lock(p, &flags); | |
3276 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3277 | task_rq_unlock(rq, &flags); | |
3278 | ||
3279 | return ns; | |
3280 | } | |
48f24c4d | 3281 | |
c5f8d995 HS |
3282 | /* |
3283 | * Return sum_exec_runtime for the thread group. | |
3284 | * In case the task is currently running, return the sum plus current's | |
3285 | * pending runtime that have not been accounted yet. | |
3286 | * | |
3287 | * Note that the thread group might have other running tasks as well, | |
3288 | * so the return value not includes other pending runtime that other | |
3289 | * running tasks might have. | |
3290 | */ | |
3291 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3292 | { | |
3293 | struct task_cputime totals; | |
3294 | unsigned long flags; | |
3295 | struct rq *rq; | |
3296 | u64 ns; | |
3297 | ||
3298 | rq = task_rq_lock(p, &flags); | |
3299 | thread_group_cputime(p, &totals); | |
3300 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3301 | task_rq_unlock(rq, &flags); |
48f24c4d | 3302 | |
1da177e4 LT |
3303 | return ns; |
3304 | } | |
3305 | ||
1da177e4 LT |
3306 | /* |
3307 | * Account user cpu time to a process. | |
3308 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3309 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3310 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3311 | */ |
457533a7 MS |
3312 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3313 | cputime_t cputime_scaled) | |
1da177e4 LT |
3314 | { |
3315 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3316 | cputime64_t tmp; | |
3317 | ||
457533a7 | 3318 | /* Add user time to process. */ |
1da177e4 | 3319 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3320 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3321 | account_group_user_time(p, cputime); |
1da177e4 LT |
3322 | |
3323 | /* Add user time to cpustat. */ | |
3324 | tmp = cputime_to_cputime64(cputime); | |
3325 | if (TASK_NICE(p) > 0) | |
3326 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3327 | else | |
3328 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3329 | |
3330 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3331 | /* Account for user time used */ |
3332 | acct_update_integrals(p); | |
1da177e4 LT |
3333 | } |
3334 | ||
94886b84 LV |
3335 | /* |
3336 | * Account guest cpu time to a process. | |
3337 | * @p: the process that the cpu time gets accounted to | |
3338 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3339 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3340 | */ |
457533a7 MS |
3341 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3342 | cputime_t cputime_scaled) | |
94886b84 LV |
3343 | { |
3344 | cputime64_t tmp; | |
3345 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3346 | ||
3347 | tmp = cputime_to_cputime64(cputime); | |
3348 | ||
457533a7 | 3349 | /* Add guest time to process. */ |
94886b84 | 3350 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3351 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3352 | account_group_user_time(p, cputime); |
94886b84 LV |
3353 | p->gtime = cputime_add(p->gtime, cputime); |
3354 | ||
457533a7 | 3355 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3356 | if (TASK_NICE(p) > 0) { |
3357 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3358 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3359 | } else { | |
3360 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3361 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3362 | } | |
94886b84 LV |
3363 | } |
3364 | ||
1da177e4 LT |
3365 | /* |
3366 | * Account system cpu time to a process. | |
3367 | * @p: the process that the cpu time gets accounted to | |
3368 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3369 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3370 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3371 | */ |
3372 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3373 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3374 | { |
3375 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3376 | cputime64_t tmp; |
3377 | ||
983ed7a6 | 3378 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3379 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3380 | return; |
3381 | } | |
94886b84 | 3382 | |
457533a7 | 3383 | /* Add system time to process. */ |
1da177e4 | 3384 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3385 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3386 | account_group_system_time(p, cputime); |
1da177e4 LT |
3387 | |
3388 | /* Add system time to cpustat. */ | |
3389 | tmp = cputime_to_cputime64(cputime); | |
3390 | if (hardirq_count() - hardirq_offset) | |
3391 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
3392 | else if (softirq_count()) | |
3393 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); | |
1da177e4 | 3394 | else |
79741dd3 MS |
3395 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3396 | ||
ef12fefa BR |
3397 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3398 | ||
1da177e4 LT |
3399 | /* Account for system time used */ |
3400 | acct_update_integrals(p); | |
1da177e4 LT |
3401 | } |
3402 | ||
c66f08be | 3403 | /* |
1da177e4 | 3404 | * Account for involuntary wait time. |
1da177e4 | 3405 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3406 | */ |
79741dd3 | 3407 | void account_steal_time(cputime_t cputime) |
c66f08be | 3408 | { |
79741dd3 MS |
3409 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3410 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3411 | ||
3412 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3413 | } |
3414 | ||
1da177e4 | 3415 | /* |
79741dd3 MS |
3416 | * Account for idle time. |
3417 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3418 | */ |
79741dd3 | 3419 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3420 | { |
3421 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3422 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3423 | struct rq *rq = this_rq(); |
1da177e4 | 3424 | |
79741dd3 MS |
3425 | if (atomic_read(&rq->nr_iowait) > 0) |
3426 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3427 | else | |
3428 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3429 | } |
3430 | ||
79741dd3 MS |
3431 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3432 | ||
3433 | /* | |
3434 | * Account a single tick of cpu time. | |
3435 | * @p: the process that the cpu time gets accounted to | |
3436 | * @user_tick: indicates if the tick is a user or a system tick | |
3437 | */ | |
3438 | void account_process_tick(struct task_struct *p, int user_tick) | |
3439 | { | |
a42548a1 | 3440 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3441 | struct rq *rq = this_rq(); |
3442 | ||
3443 | if (user_tick) | |
a42548a1 | 3444 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3445 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3446 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3447 | one_jiffy_scaled); |
3448 | else | |
a42548a1 | 3449 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3450 | } |
3451 | ||
3452 | /* | |
3453 | * Account multiple ticks of steal time. | |
3454 | * @p: the process from which the cpu time has been stolen | |
3455 | * @ticks: number of stolen ticks | |
3456 | */ | |
3457 | void account_steal_ticks(unsigned long ticks) | |
3458 | { | |
3459 | account_steal_time(jiffies_to_cputime(ticks)); | |
3460 | } | |
3461 | ||
3462 | /* | |
3463 | * Account multiple ticks of idle time. | |
3464 | * @ticks: number of stolen ticks | |
3465 | */ | |
3466 | void account_idle_ticks(unsigned long ticks) | |
3467 | { | |
3468 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3469 | } |
3470 | ||
79741dd3 MS |
3471 | #endif |
3472 | ||
49048622 BS |
3473 | /* |
3474 | * Use precise platform statistics if available: | |
3475 | */ | |
3476 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3477 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3478 | { |
d99ca3b9 HS |
3479 | *ut = p->utime; |
3480 | *st = p->stime; | |
49048622 BS |
3481 | } |
3482 | ||
0cf55e1e | 3483 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3484 | { |
0cf55e1e HS |
3485 | struct task_cputime cputime; |
3486 | ||
3487 | thread_group_cputime(p, &cputime); | |
3488 | ||
3489 | *ut = cputime.utime; | |
3490 | *st = cputime.stime; | |
49048622 BS |
3491 | } |
3492 | #else | |
761b1d26 HS |
3493 | |
3494 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3495 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3496 | #endif |
3497 | ||
d180c5bc | 3498 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3499 | { |
d99ca3b9 | 3500 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3501 | |
3502 | /* | |
3503 | * Use CFS's precise accounting: | |
3504 | */ | |
d180c5bc | 3505 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3506 | |
3507 | if (total) { | |
d180c5bc HS |
3508 | u64 temp; |
3509 | ||
3510 | temp = (u64)(rtime * utime); | |
49048622 | 3511 | do_div(temp, total); |
d180c5bc HS |
3512 | utime = (cputime_t)temp; |
3513 | } else | |
3514 | utime = rtime; | |
49048622 | 3515 | |
d180c5bc HS |
3516 | /* |
3517 | * Compare with previous values, to keep monotonicity: | |
3518 | */ | |
761b1d26 | 3519 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3520 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3521 | |
d99ca3b9 HS |
3522 | *ut = p->prev_utime; |
3523 | *st = p->prev_stime; | |
49048622 BS |
3524 | } |
3525 | ||
0cf55e1e HS |
3526 | /* |
3527 | * Must be called with siglock held. | |
3528 | */ | |
3529 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3530 | { |
0cf55e1e HS |
3531 | struct signal_struct *sig = p->signal; |
3532 | struct task_cputime cputime; | |
3533 | cputime_t rtime, utime, total; | |
49048622 | 3534 | |
0cf55e1e | 3535 | thread_group_cputime(p, &cputime); |
49048622 | 3536 | |
0cf55e1e HS |
3537 | total = cputime_add(cputime.utime, cputime.stime); |
3538 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3539 | |
0cf55e1e HS |
3540 | if (total) { |
3541 | u64 temp; | |
49048622 | 3542 | |
0cf55e1e HS |
3543 | temp = (u64)(rtime * cputime.utime); |
3544 | do_div(temp, total); | |
3545 | utime = (cputime_t)temp; | |
3546 | } else | |
3547 | utime = rtime; | |
3548 | ||
3549 | sig->prev_utime = max(sig->prev_utime, utime); | |
3550 | sig->prev_stime = max(sig->prev_stime, | |
3551 | cputime_sub(rtime, sig->prev_utime)); | |
3552 | ||
3553 | *ut = sig->prev_utime; | |
3554 | *st = sig->prev_stime; | |
49048622 | 3555 | } |
49048622 | 3556 | #endif |
49048622 | 3557 | |
7835b98b CL |
3558 | /* |
3559 | * This function gets called by the timer code, with HZ frequency. | |
3560 | * We call it with interrupts disabled. | |
3561 | * | |
3562 | * It also gets called by the fork code, when changing the parent's | |
3563 | * timeslices. | |
3564 | */ | |
3565 | void scheduler_tick(void) | |
3566 | { | |
7835b98b CL |
3567 | int cpu = smp_processor_id(); |
3568 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3569 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3570 | |
3571 | sched_clock_tick(); | |
dd41f596 | 3572 | |
05fa785c | 3573 | raw_spin_lock(&rq->lock); |
3e51f33f | 3574 | update_rq_clock(rq); |
fdf3e95d | 3575 | update_cpu_load_active(rq); |
fa85ae24 | 3576 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3577 | raw_spin_unlock(&rq->lock); |
7835b98b | 3578 | |
49f47433 | 3579 | perf_event_task_tick(curr); |
e220d2dc | 3580 | |
e418e1c2 | 3581 | #ifdef CONFIG_SMP |
dd41f596 IM |
3582 | rq->idle_at_tick = idle_cpu(cpu); |
3583 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3584 | #endif |
1da177e4 LT |
3585 | } |
3586 | ||
132380a0 | 3587 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3588 | { |
3589 | if (in_lock_functions(addr)) { | |
3590 | addr = CALLER_ADDR2; | |
3591 | if (in_lock_functions(addr)) | |
3592 | addr = CALLER_ADDR3; | |
3593 | } | |
3594 | return addr; | |
3595 | } | |
1da177e4 | 3596 | |
7e49fcce SR |
3597 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3598 | defined(CONFIG_PREEMPT_TRACER)) | |
3599 | ||
43627582 | 3600 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3601 | { |
6cd8a4bb | 3602 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3603 | /* |
3604 | * Underflow? | |
3605 | */ | |
9a11b49a IM |
3606 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3607 | return; | |
6cd8a4bb | 3608 | #endif |
1da177e4 | 3609 | preempt_count() += val; |
6cd8a4bb | 3610 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3611 | /* |
3612 | * Spinlock count overflowing soon? | |
3613 | */ | |
33859f7f MOS |
3614 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3615 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3616 | #endif |
3617 | if (preempt_count() == val) | |
3618 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3619 | } |
3620 | EXPORT_SYMBOL(add_preempt_count); | |
3621 | ||
43627582 | 3622 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3623 | { |
6cd8a4bb | 3624 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3625 | /* |
3626 | * Underflow? | |
3627 | */ | |
01e3eb82 | 3628 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3629 | return; |
1da177e4 LT |
3630 | /* |
3631 | * Is the spinlock portion underflowing? | |
3632 | */ | |
9a11b49a IM |
3633 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3634 | !(preempt_count() & PREEMPT_MASK))) | |
3635 | return; | |
6cd8a4bb | 3636 | #endif |
9a11b49a | 3637 | |
6cd8a4bb SR |
3638 | if (preempt_count() == val) |
3639 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3640 | preempt_count() -= val; |
3641 | } | |
3642 | EXPORT_SYMBOL(sub_preempt_count); | |
3643 | ||
3644 | #endif | |
3645 | ||
3646 | /* | |
dd41f596 | 3647 | * Print scheduling while atomic bug: |
1da177e4 | 3648 | */ |
dd41f596 | 3649 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3650 | { |
838225b4 SS |
3651 | struct pt_regs *regs = get_irq_regs(); |
3652 | ||
3df0fc5b PZ |
3653 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3654 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3655 | |
dd41f596 | 3656 | debug_show_held_locks(prev); |
e21f5b15 | 3657 | print_modules(); |
dd41f596 IM |
3658 | if (irqs_disabled()) |
3659 | print_irqtrace_events(prev); | |
838225b4 SS |
3660 | |
3661 | if (regs) | |
3662 | show_regs(regs); | |
3663 | else | |
3664 | dump_stack(); | |
dd41f596 | 3665 | } |
1da177e4 | 3666 | |
dd41f596 IM |
3667 | /* |
3668 | * Various schedule()-time debugging checks and statistics: | |
3669 | */ | |
3670 | static inline void schedule_debug(struct task_struct *prev) | |
3671 | { | |
1da177e4 | 3672 | /* |
41a2d6cf | 3673 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3674 | * schedule() atomically, we ignore that path for now. |
3675 | * Otherwise, whine if we are scheduling when we should not be. | |
3676 | */ | |
3f33a7ce | 3677 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3678 | __schedule_bug(prev); |
3679 | ||
1da177e4 LT |
3680 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3681 | ||
2d72376b | 3682 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3683 | #ifdef CONFIG_SCHEDSTATS |
3684 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3685 | schedstat_inc(this_rq(), bkl_count); |
3686 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3687 | } |
3688 | #endif | |
dd41f596 IM |
3689 | } |
3690 | ||
6cecd084 | 3691 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3692 | { |
a64692a3 MG |
3693 | if (prev->se.on_rq) |
3694 | update_rq_clock(rq); | |
3695 | rq->skip_clock_update = 0; | |
6cecd084 | 3696 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3697 | } |
3698 | ||
dd41f596 IM |
3699 | /* |
3700 | * Pick up the highest-prio task: | |
3701 | */ | |
3702 | static inline struct task_struct * | |
b67802ea | 3703 | pick_next_task(struct rq *rq) |
dd41f596 | 3704 | { |
5522d5d5 | 3705 | const struct sched_class *class; |
dd41f596 | 3706 | struct task_struct *p; |
1da177e4 LT |
3707 | |
3708 | /* | |
dd41f596 IM |
3709 | * Optimization: we know that if all tasks are in |
3710 | * the fair class we can call that function directly: | |
1da177e4 | 3711 | */ |
dd41f596 | 3712 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3713 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3714 | if (likely(p)) |
3715 | return p; | |
1da177e4 LT |
3716 | } |
3717 | ||
dd41f596 IM |
3718 | class = sched_class_highest; |
3719 | for ( ; ; ) { | |
fb8d4724 | 3720 | p = class->pick_next_task(rq); |
dd41f596 IM |
3721 | if (p) |
3722 | return p; | |
3723 | /* | |
3724 | * Will never be NULL as the idle class always | |
3725 | * returns a non-NULL p: | |
3726 | */ | |
3727 | class = class->next; | |
3728 | } | |
3729 | } | |
1da177e4 | 3730 | |
dd41f596 IM |
3731 | /* |
3732 | * schedule() is the main scheduler function. | |
3733 | */ | |
ff743345 | 3734 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3735 | { |
3736 | struct task_struct *prev, *next; | |
67ca7bde | 3737 | unsigned long *switch_count; |
dd41f596 | 3738 | struct rq *rq; |
31656519 | 3739 | int cpu; |
dd41f596 | 3740 | |
ff743345 PZ |
3741 | need_resched: |
3742 | preempt_disable(); | |
dd41f596 IM |
3743 | cpu = smp_processor_id(); |
3744 | rq = cpu_rq(cpu); | |
25502a6c | 3745 | rcu_note_context_switch(cpu); |
dd41f596 | 3746 | prev = rq->curr; |
dd41f596 IM |
3747 | |
3748 | release_kernel_lock(prev); | |
3749 | need_resched_nonpreemptible: | |
3750 | ||
3751 | schedule_debug(prev); | |
1da177e4 | 3752 | |
31656519 | 3753 | if (sched_feat(HRTICK)) |
f333fdc9 | 3754 | hrtick_clear(rq); |
8f4d37ec | 3755 | |
05fa785c | 3756 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3757 | clear_tsk_need_resched(prev); |
1da177e4 | 3758 | |
246d86b5 | 3759 | switch_count = &prev->nivcsw; |
1da177e4 | 3760 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3761 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3762 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
3763 | } else { |
3764 | /* | |
3765 | * If a worker is going to sleep, notify and | |
3766 | * ask workqueue whether it wants to wake up a | |
3767 | * task to maintain concurrency. If so, wake | |
3768 | * up the task. | |
3769 | */ | |
3770 | if (prev->flags & PF_WQ_WORKER) { | |
3771 | struct task_struct *to_wakeup; | |
3772 | ||
3773 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3774 | if (to_wakeup) | |
3775 | try_to_wake_up_local(to_wakeup); | |
3776 | } | |
371fd7e7 | 3777 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 3778 | } |
dd41f596 | 3779 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3780 | } |
3781 | ||
3f029d3c | 3782 | pre_schedule(rq, prev); |
f65eda4f | 3783 | |
dd41f596 | 3784 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3785 | idle_balance(cpu, rq); |
1da177e4 | 3786 | |
df1c99d4 | 3787 | put_prev_task(rq, prev); |
b67802ea | 3788 | next = pick_next_task(rq); |
1da177e4 | 3789 | |
1da177e4 | 3790 | if (likely(prev != next)) { |
673a90a1 | 3791 | sched_info_switch(prev, next); |
49f47433 | 3792 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3793 | |
1da177e4 LT |
3794 | rq->nr_switches++; |
3795 | rq->curr = next; | |
3796 | ++*switch_count; | |
3797 | ||
dd41f596 | 3798 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3799 | /* |
246d86b5 ON |
3800 | * The context switch have flipped the stack from under us |
3801 | * and restored the local variables which were saved when | |
3802 | * this task called schedule() in the past. prev == current | |
3803 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3804 | */ |
3805 | cpu = smp_processor_id(); | |
3806 | rq = cpu_rq(cpu); | |
1da177e4 | 3807 | } else |
05fa785c | 3808 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3809 | |
3f029d3c | 3810 | post_schedule(rq); |
1da177e4 | 3811 | |
246d86b5 | 3812 | if (unlikely(reacquire_kernel_lock(prev))) |
1da177e4 | 3813 | goto need_resched_nonpreemptible; |
8f4d37ec | 3814 | |
1da177e4 | 3815 | preempt_enable_no_resched(); |
ff743345 | 3816 | if (need_resched()) |
1da177e4 LT |
3817 | goto need_resched; |
3818 | } | |
1da177e4 LT |
3819 | EXPORT_SYMBOL(schedule); |
3820 | ||
c08f7829 | 3821 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3822 | /* |
3823 | * Look out! "owner" is an entirely speculative pointer | |
3824 | * access and not reliable. | |
3825 | */ | |
3826 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3827 | { | |
3828 | unsigned int cpu; | |
3829 | struct rq *rq; | |
3830 | ||
3831 | if (!sched_feat(OWNER_SPIN)) | |
3832 | return 0; | |
3833 | ||
3834 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3835 | /* | |
3836 | * Need to access the cpu field knowing that | |
3837 | * DEBUG_PAGEALLOC could have unmapped it if | |
3838 | * the mutex owner just released it and exited. | |
3839 | */ | |
3840 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 3841 | return 0; |
0d66bf6d PZ |
3842 | #else |
3843 | cpu = owner->cpu; | |
3844 | #endif | |
3845 | ||
3846 | /* | |
3847 | * Even if the access succeeded (likely case), | |
3848 | * the cpu field may no longer be valid. | |
3849 | */ | |
3850 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 3851 | return 0; |
0d66bf6d PZ |
3852 | |
3853 | /* | |
3854 | * We need to validate that we can do a | |
3855 | * get_cpu() and that we have the percpu area. | |
3856 | */ | |
3857 | if (!cpu_online(cpu)) | |
4b402210 | 3858 | return 0; |
0d66bf6d PZ |
3859 | |
3860 | rq = cpu_rq(cpu); | |
3861 | ||
3862 | for (;;) { | |
3863 | /* | |
3864 | * Owner changed, break to re-assess state. | |
3865 | */ | |
3866 | if (lock->owner != owner) | |
3867 | break; | |
3868 | ||
3869 | /* | |
3870 | * Is that owner really running on that cpu? | |
3871 | */ | |
3872 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3873 | return 0; | |
3874 | ||
3875 | cpu_relax(); | |
3876 | } | |
4b402210 | 3877 | |
0d66bf6d PZ |
3878 | return 1; |
3879 | } | |
3880 | #endif | |
3881 | ||
1da177e4 LT |
3882 | #ifdef CONFIG_PREEMPT |
3883 | /* | |
2ed6e34f | 3884 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3885 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3886 | * occur there and call schedule directly. |
3887 | */ | |
3888 | asmlinkage void __sched preempt_schedule(void) | |
3889 | { | |
3890 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3891 | |
1da177e4 LT |
3892 | /* |
3893 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3894 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3895 | */ |
beed33a8 | 3896 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3897 | return; |
3898 | ||
3a5c359a AK |
3899 | do { |
3900 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a | 3901 | schedule(); |
3a5c359a | 3902 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3903 | |
3a5c359a AK |
3904 | /* |
3905 | * Check again in case we missed a preemption opportunity | |
3906 | * between schedule and now. | |
3907 | */ | |
3908 | barrier(); | |
5ed0cec0 | 3909 | } while (need_resched()); |
1da177e4 | 3910 | } |
1da177e4 LT |
3911 | EXPORT_SYMBOL(preempt_schedule); |
3912 | ||
3913 | /* | |
2ed6e34f | 3914 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3915 | * off of irq context. |
3916 | * Note, that this is called and return with irqs disabled. This will | |
3917 | * protect us against recursive calling from irq. | |
3918 | */ | |
3919 | asmlinkage void __sched preempt_schedule_irq(void) | |
3920 | { | |
3921 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3922 | |
2ed6e34f | 3923 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3924 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3925 | ||
3a5c359a AK |
3926 | do { |
3927 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3928 | local_irq_enable(); |
3929 | schedule(); | |
3930 | local_irq_disable(); | |
3a5c359a | 3931 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3932 | |
3a5c359a AK |
3933 | /* |
3934 | * Check again in case we missed a preemption opportunity | |
3935 | * between schedule and now. | |
3936 | */ | |
3937 | barrier(); | |
5ed0cec0 | 3938 | } while (need_resched()); |
1da177e4 LT |
3939 | } |
3940 | ||
3941 | #endif /* CONFIG_PREEMPT */ | |
3942 | ||
63859d4f | 3943 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3944 | void *key) |
1da177e4 | 3945 | { |
63859d4f | 3946 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3947 | } |
1da177e4 LT |
3948 | EXPORT_SYMBOL(default_wake_function); |
3949 | ||
3950 | /* | |
41a2d6cf IM |
3951 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3952 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3953 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3954 | * | |
3955 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3956 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3957 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3958 | */ | |
78ddb08f | 3959 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3960 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3961 | { |
2e45874c | 3962 | wait_queue_t *curr, *next; |
1da177e4 | 3963 | |
2e45874c | 3964 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3965 | unsigned flags = curr->flags; |
3966 | ||
63859d4f | 3967 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3968 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3969 | break; |
3970 | } | |
3971 | } | |
3972 | ||
3973 | /** | |
3974 | * __wake_up - wake up threads blocked on a waitqueue. | |
3975 | * @q: the waitqueue | |
3976 | * @mode: which threads | |
3977 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3978 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3979 | * |
3980 | * It may be assumed that this function implies a write memory barrier before | |
3981 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3982 | */ |
7ad5b3a5 | 3983 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 3984 | int nr_exclusive, void *key) |
1da177e4 LT |
3985 | { |
3986 | unsigned long flags; | |
3987 | ||
3988 | spin_lock_irqsave(&q->lock, flags); | |
3989 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
3990 | spin_unlock_irqrestore(&q->lock, flags); | |
3991 | } | |
1da177e4 LT |
3992 | EXPORT_SYMBOL(__wake_up); |
3993 | ||
3994 | /* | |
3995 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
3996 | */ | |
7ad5b3a5 | 3997 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
3998 | { |
3999 | __wake_up_common(q, mode, 1, 0, NULL); | |
4000 | } | |
22c43c81 | 4001 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4002 | |
4ede816a DL |
4003 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4004 | { | |
4005 | __wake_up_common(q, mode, 1, 0, key); | |
4006 | } | |
4007 | ||
1da177e4 | 4008 | /** |
4ede816a | 4009 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4010 | * @q: the waitqueue |
4011 | * @mode: which threads | |
4012 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4013 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4014 | * |
4015 | * The sync wakeup differs that the waker knows that it will schedule | |
4016 | * away soon, so while the target thread will be woken up, it will not | |
4017 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4018 | * with each other. This can prevent needless bouncing between CPUs. | |
4019 | * | |
4020 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4021 | * |
4022 | * It may be assumed that this function implies a write memory barrier before | |
4023 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4024 | */ |
4ede816a DL |
4025 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4026 | int nr_exclusive, void *key) | |
1da177e4 LT |
4027 | { |
4028 | unsigned long flags; | |
7d478721 | 4029 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4030 | |
4031 | if (unlikely(!q)) | |
4032 | return; | |
4033 | ||
4034 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4035 | wake_flags = 0; |
1da177e4 LT |
4036 | |
4037 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4038 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4039 | spin_unlock_irqrestore(&q->lock, flags); |
4040 | } | |
4ede816a DL |
4041 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4042 | ||
4043 | /* | |
4044 | * __wake_up_sync - see __wake_up_sync_key() | |
4045 | */ | |
4046 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4047 | { | |
4048 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4049 | } | |
1da177e4 LT |
4050 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4051 | ||
65eb3dc6 KD |
4052 | /** |
4053 | * complete: - signals a single thread waiting on this completion | |
4054 | * @x: holds the state of this particular completion | |
4055 | * | |
4056 | * This will wake up a single thread waiting on this completion. Threads will be | |
4057 | * awakened in the same order in which they were queued. | |
4058 | * | |
4059 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4060 | * |
4061 | * It may be assumed that this function implies a write memory barrier before | |
4062 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4063 | */ |
b15136e9 | 4064 | void complete(struct completion *x) |
1da177e4 LT |
4065 | { |
4066 | unsigned long flags; | |
4067 | ||
4068 | spin_lock_irqsave(&x->wait.lock, flags); | |
4069 | x->done++; | |
d9514f6c | 4070 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4071 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4072 | } | |
4073 | EXPORT_SYMBOL(complete); | |
4074 | ||
65eb3dc6 KD |
4075 | /** |
4076 | * complete_all: - signals all threads waiting on this completion | |
4077 | * @x: holds the state of this particular completion | |
4078 | * | |
4079 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4080 | * |
4081 | * It may be assumed that this function implies a write memory barrier before | |
4082 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4083 | */ |
b15136e9 | 4084 | void complete_all(struct completion *x) |
1da177e4 LT |
4085 | { |
4086 | unsigned long flags; | |
4087 | ||
4088 | spin_lock_irqsave(&x->wait.lock, flags); | |
4089 | x->done += UINT_MAX/2; | |
d9514f6c | 4090 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4091 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4092 | } | |
4093 | EXPORT_SYMBOL(complete_all); | |
4094 | ||
8cbbe86d AK |
4095 | static inline long __sched |
4096 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4097 | { |
1da177e4 LT |
4098 | if (!x->done) { |
4099 | DECLARE_WAITQUEUE(wait, current); | |
4100 | ||
a93d2f17 | 4101 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4102 | do { |
94d3d824 | 4103 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4104 | timeout = -ERESTARTSYS; |
4105 | break; | |
8cbbe86d AK |
4106 | } |
4107 | __set_current_state(state); | |
1da177e4 LT |
4108 | spin_unlock_irq(&x->wait.lock); |
4109 | timeout = schedule_timeout(timeout); | |
4110 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4111 | } while (!x->done && timeout); |
1da177e4 | 4112 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4113 | if (!x->done) |
4114 | return timeout; | |
1da177e4 LT |
4115 | } |
4116 | x->done--; | |
ea71a546 | 4117 | return timeout ?: 1; |
1da177e4 | 4118 | } |
1da177e4 | 4119 | |
8cbbe86d AK |
4120 | static long __sched |
4121 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4122 | { |
1da177e4 LT |
4123 | might_sleep(); |
4124 | ||
4125 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4126 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4127 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4128 | return timeout; |
4129 | } | |
1da177e4 | 4130 | |
65eb3dc6 KD |
4131 | /** |
4132 | * wait_for_completion: - waits for completion of a task | |
4133 | * @x: holds the state of this particular completion | |
4134 | * | |
4135 | * This waits to be signaled for completion of a specific task. It is NOT | |
4136 | * interruptible and there is no timeout. | |
4137 | * | |
4138 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4139 | * and interrupt capability. Also see complete(). | |
4140 | */ | |
b15136e9 | 4141 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4142 | { |
4143 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4144 | } |
8cbbe86d | 4145 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4146 | |
65eb3dc6 KD |
4147 | /** |
4148 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4149 | * @x: holds the state of this particular completion | |
4150 | * @timeout: timeout value in jiffies | |
4151 | * | |
4152 | * This waits for either a completion of a specific task to be signaled or for a | |
4153 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4154 | * interruptible. | |
4155 | */ | |
b15136e9 | 4156 | unsigned long __sched |
8cbbe86d | 4157 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4158 | { |
8cbbe86d | 4159 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4160 | } |
8cbbe86d | 4161 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4162 | |
65eb3dc6 KD |
4163 | /** |
4164 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4165 | * @x: holds the state of this particular completion | |
4166 | * | |
4167 | * This waits for completion of a specific task to be signaled. It is | |
4168 | * interruptible. | |
4169 | */ | |
8cbbe86d | 4170 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4171 | { |
51e97990 AK |
4172 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4173 | if (t == -ERESTARTSYS) | |
4174 | return t; | |
4175 | return 0; | |
0fec171c | 4176 | } |
8cbbe86d | 4177 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4178 | |
65eb3dc6 KD |
4179 | /** |
4180 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4181 | * @x: holds the state of this particular completion | |
4182 | * @timeout: timeout value in jiffies | |
4183 | * | |
4184 | * This waits for either a completion of a specific task to be signaled or for a | |
4185 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4186 | */ | |
b15136e9 | 4187 | unsigned long __sched |
8cbbe86d AK |
4188 | wait_for_completion_interruptible_timeout(struct completion *x, |
4189 | unsigned long timeout) | |
0fec171c | 4190 | { |
8cbbe86d | 4191 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4192 | } |
8cbbe86d | 4193 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4194 | |
65eb3dc6 KD |
4195 | /** |
4196 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4197 | * @x: holds the state of this particular completion | |
4198 | * | |
4199 | * This waits to be signaled for completion of a specific task. It can be | |
4200 | * interrupted by a kill signal. | |
4201 | */ | |
009e577e MW |
4202 | int __sched wait_for_completion_killable(struct completion *x) |
4203 | { | |
4204 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4205 | if (t == -ERESTARTSYS) | |
4206 | return t; | |
4207 | return 0; | |
4208 | } | |
4209 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4210 | ||
0aa12fb4 SW |
4211 | /** |
4212 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4213 | * @x: holds the state of this particular completion | |
4214 | * @timeout: timeout value in jiffies | |
4215 | * | |
4216 | * This waits for either a completion of a specific task to be | |
4217 | * signaled or for a specified timeout to expire. It can be | |
4218 | * interrupted by a kill signal. The timeout is in jiffies. | |
4219 | */ | |
4220 | unsigned long __sched | |
4221 | wait_for_completion_killable_timeout(struct completion *x, | |
4222 | unsigned long timeout) | |
4223 | { | |
4224 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4225 | } | |
4226 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4227 | ||
be4de352 DC |
4228 | /** |
4229 | * try_wait_for_completion - try to decrement a completion without blocking | |
4230 | * @x: completion structure | |
4231 | * | |
4232 | * Returns: 0 if a decrement cannot be done without blocking | |
4233 | * 1 if a decrement succeeded. | |
4234 | * | |
4235 | * If a completion is being used as a counting completion, | |
4236 | * attempt to decrement the counter without blocking. This | |
4237 | * enables us to avoid waiting if the resource the completion | |
4238 | * is protecting is not available. | |
4239 | */ | |
4240 | bool try_wait_for_completion(struct completion *x) | |
4241 | { | |
7539a3b3 | 4242 | unsigned long flags; |
be4de352 DC |
4243 | int ret = 1; |
4244 | ||
7539a3b3 | 4245 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4246 | if (!x->done) |
4247 | ret = 0; | |
4248 | else | |
4249 | x->done--; | |
7539a3b3 | 4250 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4251 | return ret; |
4252 | } | |
4253 | EXPORT_SYMBOL(try_wait_for_completion); | |
4254 | ||
4255 | /** | |
4256 | * completion_done - Test to see if a completion has any waiters | |
4257 | * @x: completion structure | |
4258 | * | |
4259 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4260 | * 1 if there are no waiters. | |
4261 | * | |
4262 | */ | |
4263 | bool completion_done(struct completion *x) | |
4264 | { | |
7539a3b3 | 4265 | unsigned long flags; |
be4de352 DC |
4266 | int ret = 1; |
4267 | ||
7539a3b3 | 4268 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4269 | if (!x->done) |
4270 | ret = 0; | |
7539a3b3 | 4271 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4272 | return ret; |
4273 | } | |
4274 | EXPORT_SYMBOL(completion_done); | |
4275 | ||
8cbbe86d AK |
4276 | static long __sched |
4277 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4278 | { |
0fec171c IM |
4279 | unsigned long flags; |
4280 | wait_queue_t wait; | |
4281 | ||
4282 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4283 | |
8cbbe86d | 4284 | __set_current_state(state); |
1da177e4 | 4285 | |
8cbbe86d AK |
4286 | spin_lock_irqsave(&q->lock, flags); |
4287 | __add_wait_queue(q, &wait); | |
4288 | spin_unlock(&q->lock); | |
4289 | timeout = schedule_timeout(timeout); | |
4290 | spin_lock_irq(&q->lock); | |
4291 | __remove_wait_queue(q, &wait); | |
4292 | spin_unlock_irqrestore(&q->lock, flags); | |
4293 | ||
4294 | return timeout; | |
4295 | } | |
4296 | ||
4297 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4298 | { | |
4299 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4300 | } |
1da177e4 LT |
4301 | EXPORT_SYMBOL(interruptible_sleep_on); |
4302 | ||
0fec171c | 4303 | long __sched |
95cdf3b7 | 4304 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4305 | { |
8cbbe86d | 4306 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4307 | } |
1da177e4 LT |
4308 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4309 | ||
0fec171c | 4310 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4311 | { |
8cbbe86d | 4312 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4313 | } |
1da177e4 LT |
4314 | EXPORT_SYMBOL(sleep_on); |
4315 | ||
0fec171c | 4316 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4317 | { |
8cbbe86d | 4318 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4319 | } |
1da177e4 LT |
4320 | EXPORT_SYMBOL(sleep_on_timeout); |
4321 | ||
b29739f9 IM |
4322 | #ifdef CONFIG_RT_MUTEXES |
4323 | ||
4324 | /* | |
4325 | * rt_mutex_setprio - set the current priority of a task | |
4326 | * @p: task | |
4327 | * @prio: prio value (kernel-internal form) | |
4328 | * | |
4329 | * This function changes the 'effective' priority of a task. It does | |
4330 | * not touch ->normal_prio like __setscheduler(). | |
4331 | * | |
4332 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4333 | */ | |
36c8b586 | 4334 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4335 | { |
4336 | unsigned long flags; | |
83b699ed | 4337 | int oldprio, on_rq, running; |
70b97a7f | 4338 | struct rq *rq; |
83ab0aa0 | 4339 | const struct sched_class *prev_class; |
b29739f9 IM |
4340 | |
4341 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4342 | ||
4343 | rq = task_rq_lock(p, &flags); | |
4344 | ||
d5f9f942 | 4345 | oldprio = p->prio; |
83ab0aa0 | 4346 | prev_class = p->sched_class; |
dd41f596 | 4347 | on_rq = p->se.on_rq; |
051a1d1a | 4348 | running = task_current(rq, p); |
0e1f3483 | 4349 | if (on_rq) |
69be72c1 | 4350 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4351 | if (running) |
4352 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4353 | |
4354 | if (rt_prio(prio)) | |
4355 | p->sched_class = &rt_sched_class; | |
4356 | else | |
4357 | p->sched_class = &fair_sched_class; | |
4358 | ||
b29739f9 IM |
4359 | p->prio = prio; |
4360 | ||
0e1f3483 HS |
4361 | if (running) |
4362 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4363 | if (on_rq) { |
371fd7e7 | 4364 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4365 | |
4366 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4367 | } |
4368 | task_rq_unlock(rq, &flags); | |
4369 | } | |
4370 | ||
4371 | #endif | |
4372 | ||
36c8b586 | 4373 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4374 | { |
dd41f596 | 4375 | int old_prio, delta, on_rq; |
1da177e4 | 4376 | unsigned long flags; |
70b97a7f | 4377 | struct rq *rq; |
1da177e4 LT |
4378 | |
4379 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4380 | return; | |
4381 | /* | |
4382 | * We have to be careful, if called from sys_setpriority(), | |
4383 | * the task might be in the middle of scheduling on another CPU. | |
4384 | */ | |
4385 | rq = task_rq_lock(p, &flags); | |
4386 | /* | |
4387 | * The RT priorities are set via sched_setscheduler(), but we still | |
4388 | * allow the 'normal' nice value to be set - but as expected | |
4389 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4390 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4391 | */ |
e05606d3 | 4392 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4393 | p->static_prio = NICE_TO_PRIO(nice); |
4394 | goto out_unlock; | |
4395 | } | |
dd41f596 | 4396 | on_rq = p->se.on_rq; |
c09595f6 | 4397 | if (on_rq) |
69be72c1 | 4398 | dequeue_task(rq, p, 0); |
1da177e4 | 4399 | |
1da177e4 | 4400 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4401 | set_load_weight(p); |
b29739f9 IM |
4402 | old_prio = p->prio; |
4403 | p->prio = effective_prio(p); | |
4404 | delta = p->prio - old_prio; | |
1da177e4 | 4405 | |
dd41f596 | 4406 | if (on_rq) { |
371fd7e7 | 4407 | enqueue_task(rq, p, 0); |
1da177e4 | 4408 | /* |
d5f9f942 AM |
4409 | * If the task increased its priority or is running and |
4410 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4411 | */ |
d5f9f942 | 4412 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4413 | resched_task(rq->curr); |
4414 | } | |
4415 | out_unlock: | |
4416 | task_rq_unlock(rq, &flags); | |
4417 | } | |
1da177e4 LT |
4418 | EXPORT_SYMBOL(set_user_nice); |
4419 | ||
e43379f1 MM |
4420 | /* |
4421 | * can_nice - check if a task can reduce its nice value | |
4422 | * @p: task | |
4423 | * @nice: nice value | |
4424 | */ | |
36c8b586 | 4425 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4426 | { |
024f4747 MM |
4427 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4428 | int nice_rlim = 20 - nice; | |
48f24c4d | 4429 | |
78d7d407 | 4430 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4431 | capable(CAP_SYS_NICE)); |
4432 | } | |
4433 | ||
1da177e4 LT |
4434 | #ifdef __ARCH_WANT_SYS_NICE |
4435 | ||
4436 | /* | |
4437 | * sys_nice - change the priority of the current process. | |
4438 | * @increment: priority increment | |
4439 | * | |
4440 | * sys_setpriority is a more generic, but much slower function that | |
4441 | * does similar things. | |
4442 | */ | |
5add95d4 | 4443 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4444 | { |
48f24c4d | 4445 | long nice, retval; |
1da177e4 LT |
4446 | |
4447 | /* | |
4448 | * Setpriority might change our priority at the same moment. | |
4449 | * We don't have to worry. Conceptually one call occurs first | |
4450 | * and we have a single winner. | |
4451 | */ | |
e43379f1 MM |
4452 | if (increment < -40) |
4453 | increment = -40; | |
1da177e4 LT |
4454 | if (increment > 40) |
4455 | increment = 40; | |
4456 | ||
2b8f836f | 4457 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4458 | if (nice < -20) |
4459 | nice = -20; | |
4460 | if (nice > 19) | |
4461 | nice = 19; | |
4462 | ||
e43379f1 MM |
4463 | if (increment < 0 && !can_nice(current, nice)) |
4464 | return -EPERM; | |
4465 | ||
1da177e4 LT |
4466 | retval = security_task_setnice(current, nice); |
4467 | if (retval) | |
4468 | return retval; | |
4469 | ||
4470 | set_user_nice(current, nice); | |
4471 | return 0; | |
4472 | } | |
4473 | ||
4474 | #endif | |
4475 | ||
4476 | /** | |
4477 | * task_prio - return the priority value of a given task. | |
4478 | * @p: the task in question. | |
4479 | * | |
4480 | * This is the priority value as seen by users in /proc. | |
4481 | * RT tasks are offset by -200. Normal tasks are centered | |
4482 | * around 0, value goes from -16 to +15. | |
4483 | */ | |
36c8b586 | 4484 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4485 | { |
4486 | return p->prio - MAX_RT_PRIO; | |
4487 | } | |
4488 | ||
4489 | /** | |
4490 | * task_nice - return the nice value of a given task. | |
4491 | * @p: the task in question. | |
4492 | */ | |
36c8b586 | 4493 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4494 | { |
4495 | return TASK_NICE(p); | |
4496 | } | |
150d8bed | 4497 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4498 | |
4499 | /** | |
4500 | * idle_cpu - is a given cpu idle currently? | |
4501 | * @cpu: the processor in question. | |
4502 | */ | |
4503 | int idle_cpu(int cpu) | |
4504 | { | |
4505 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4506 | } | |
4507 | ||
1da177e4 LT |
4508 | /** |
4509 | * idle_task - return the idle task for a given cpu. | |
4510 | * @cpu: the processor in question. | |
4511 | */ | |
36c8b586 | 4512 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4513 | { |
4514 | return cpu_rq(cpu)->idle; | |
4515 | } | |
4516 | ||
4517 | /** | |
4518 | * find_process_by_pid - find a process with a matching PID value. | |
4519 | * @pid: the pid in question. | |
4520 | */ | |
a9957449 | 4521 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4522 | { |
228ebcbe | 4523 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4524 | } |
4525 | ||
4526 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4527 | static void |
4528 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4529 | { |
dd41f596 | 4530 | BUG_ON(p->se.on_rq); |
48f24c4d | 4531 | |
1da177e4 LT |
4532 | p->policy = policy; |
4533 | p->rt_priority = prio; | |
b29739f9 IM |
4534 | p->normal_prio = normal_prio(p); |
4535 | /* we are holding p->pi_lock already */ | |
4536 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4537 | if (rt_prio(p->prio)) |
4538 | p->sched_class = &rt_sched_class; | |
4539 | else | |
4540 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4541 | set_load_weight(p); |
1da177e4 LT |
4542 | } |
4543 | ||
c69e8d9c DH |
4544 | /* |
4545 | * check the target process has a UID that matches the current process's | |
4546 | */ | |
4547 | static bool check_same_owner(struct task_struct *p) | |
4548 | { | |
4549 | const struct cred *cred = current_cred(), *pcred; | |
4550 | bool match; | |
4551 | ||
4552 | rcu_read_lock(); | |
4553 | pcred = __task_cred(p); | |
4554 | match = (cred->euid == pcred->euid || | |
4555 | cred->euid == pcred->uid); | |
4556 | rcu_read_unlock(); | |
4557 | return match; | |
4558 | } | |
4559 | ||
961ccddd RR |
4560 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4561 | struct sched_param *param, bool user) | |
1da177e4 | 4562 | { |
83b699ed | 4563 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4564 | unsigned long flags; |
83ab0aa0 | 4565 | const struct sched_class *prev_class; |
70b97a7f | 4566 | struct rq *rq; |
ca94c442 | 4567 | int reset_on_fork; |
1da177e4 | 4568 | |
66e5393a SR |
4569 | /* may grab non-irq protected spin_locks */ |
4570 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4571 | recheck: |
4572 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4573 | if (policy < 0) { |
4574 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4575 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4576 | } else { |
4577 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4578 | policy &= ~SCHED_RESET_ON_FORK; | |
4579 | ||
4580 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4581 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4582 | policy != SCHED_IDLE) | |
4583 | return -EINVAL; | |
4584 | } | |
4585 | ||
1da177e4 LT |
4586 | /* |
4587 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4588 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4589 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4590 | */ |
4591 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4592 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4593 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4594 | return -EINVAL; |
e05606d3 | 4595 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4596 | return -EINVAL; |
4597 | ||
37e4ab3f OC |
4598 | /* |
4599 | * Allow unprivileged RT tasks to decrease priority: | |
4600 | */ | |
961ccddd | 4601 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4602 | if (rt_policy(policy)) { |
a44702e8 ON |
4603 | unsigned long rlim_rtprio = |
4604 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4605 | |
4606 | /* can't set/change the rt policy */ | |
4607 | if (policy != p->policy && !rlim_rtprio) | |
4608 | return -EPERM; | |
4609 | ||
4610 | /* can't increase priority */ | |
4611 | if (param->sched_priority > p->rt_priority && | |
4612 | param->sched_priority > rlim_rtprio) | |
4613 | return -EPERM; | |
4614 | } | |
dd41f596 IM |
4615 | /* |
4616 | * Like positive nice levels, dont allow tasks to | |
4617 | * move out of SCHED_IDLE either: | |
4618 | */ | |
4619 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4620 | return -EPERM; | |
5fe1d75f | 4621 | |
37e4ab3f | 4622 | /* can't change other user's priorities */ |
c69e8d9c | 4623 | if (!check_same_owner(p)) |
37e4ab3f | 4624 | return -EPERM; |
ca94c442 LP |
4625 | |
4626 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4627 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4628 | return -EPERM; | |
37e4ab3f | 4629 | } |
1da177e4 | 4630 | |
725aad24 | 4631 | if (user) { |
725aad24 JF |
4632 | retval = security_task_setscheduler(p, policy, param); |
4633 | if (retval) | |
4634 | return retval; | |
4635 | } | |
4636 | ||
b29739f9 IM |
4637 | /* |
4638 | * make sure no PI-waiters arrive (or leave) while we are | |
4639 | * changing the priority of the task: | |
4640 | */ | |
1d615482 | 4641 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4642 | /* |
4643 | * To be able to change p->policy safely, the apropriate | |
4644 | * runqueue lock must be held. | |
4645 | */ | |
b29739f9 | 4646 | rq = __task_rq_lock(p); |
dc61b1d6 PZ |
4647 | |
4648 | #ifdef CONFIG_RT_GROUP_SCHED | |
4649 | if (user) { | |
4650 | /* | |
4651 | * Do not allow realtime tasks into groups that have no runtime | |
4652 | * assigned. | |
4653 | */ | |
4654 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
4655 | task_group(p)->rt_bandwidth.rt_runtime == 0) { | |
4656 | __task_rq_unlock(rq); | |
4657 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4658 | return -EPERM; | |
4659 | } | |
4660 | } | |
4661 | #endif | |
4662 | ||
1da177e4 LT |
4663 | /* recheck policy now with rq lock held */ |
4664 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4665 | policy = oldpolicy = -1; | |
b29739f9 | 4666 | __task_rq_unlock(rq); |
1d615482 | 4667 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4668 | goto recheck; |
4669 | } | |
dd41f596 | 4670 | on_rq = p->se.on_rq; |
051a1d1a | 4671 | running = task_current(rq, p); |
0e1f3483 | 4672 | if (on_rq) |
2e1cb74a | 4673 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4674 | if (running) |
4675 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4676 | |
ca94c442 LP |
4677 | p->sched_reset_on_fork = reset_on_fork; |
4678 | ||
1da177e4 | 4679 | oldprio = p->prio; |
83ab0aa0 | 4680 | prev_class = p->sched_class; |
dd41f596 | 4681 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4682 | |
0e1f3483 HS |
4683 | if (running) |
4684 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4685 | if (on_rq) { |
4686 | activate_task(rq, p, 0); | |
cb469845 SR |
4687 | |
4688 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4689 | } |
b29739f9 | 4690 | __task_rq_unlock(rq); |
1d615482 | 4691 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4692 | |
95e02ca9 TG |
4693 | rt_mutex_adjust_pi(p); |
4694 | ||
1da177e4 LT |
4695 | return 0; |
4696 | } | |
961ccddd RR |
4697 | |
4698 | /** | |
4699 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4700 | * @p: the task in question. | |
4701 | * @policy: new policy. | |
4702 | * @param: structure containing the new RT priority. | |
4703 | * | |
4704 | * NOTE that the task may be already dead. | |
4705 | */ | |
4706 | int sched_setscheduler(struct task_struct *p, int policy, | |
4707 | struct sched_param *param) | |
4708 | { | |
4709 | return __sched_setscheduler(p, policy, param, true); | |
4710 | } | |
1da177e4 LT |
4711 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4712 | ||
961ccddd RR |
4713 | /** |
4714 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4715 | * @p: the task in question. | |
4716 | * @policy: new policy. | |
4717 | * @param: structure containing the new RT priority. | |
4718 | * | |
4719 | * Just like sched_setscheduler, only don't bother checking if the | |
4720 | * current context has permission. For example, this is needed in | |
4721 | * stop_machine(): we create temporary high priority worker threads, | |
4722 | * but our caller might not have that capability. | |
4723 | */ | |
4724 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
4725 | struct sched_param *param) | |
4726 | { | |
4727 | return __sched_setscheduler(p, policy, param, false); | |
4728 | } | |
4729 | ||
95cdf3b7 IM |
4730 | static int |
4731 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4732 | { |
1da177e4 LT |
4733 | struct sched_param lparam; |
4734 | struct task_struct *p; | |
36c8b586 | 4735 | int retval; |
1da177e4 LT |
4736 | |
4737 | if (!param || pid < 0) | |
4738 | return -EINVAL; | |
4739 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4740 | return -EFAULT; | |
5fe1d75f ON |
4741 | |
4742 | rcu_read_lock(); | |
4743 | retval = -ESRCH; | |
1da177e4 | 4744 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4745 | if (p != NULL) |
4746 | retval = sched_setscheduler(p, policy, &lparam); | |
4747 | rcu_read_unlock(); | |
36c8b586 | 4748 | |
1da177e4 LT |
4749 | return retval; |
4750 | } | |
4751 | ||
4752 | /** | |
4753 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4754 | * @pid: the pid in question. | |
4755 | * @policy: new policy. | |
4756 | * @param: structure containing the new RT priority. | |
4757 | */ | |
5add95d4 HC |
4758 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4759 | struct sched_param __user *, param) | |
1da177e4 | 4760 | { |
c21761f1 JB |
4761 | /* negative values for policy are not valid */ |
4762 | if (policy < 0) | |
4763 | return -EINVAL; | |
4764 | ||
1da177e4 LT |
4765 | return do_sched_setscheduler(pid, policy, param); |
4766 | } | |
4767 | ||
4768 | /** | |
4769 | * sys_sched_setparam - set/change the RT priority of a thread | |
4770 | * @pid: the pid in question. | |
4771 | * @param: structure containing the new RT priority. | |
4772 | */ | |
5add95d4 | 4773 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4774 | { |
4775 | return do_sched_setscheduler(pid, -1, param); | |
4776 | } | |
4777 | ||
4778 | /** | |
4779 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4780 | * @pid: the pid in question. | |
4781 | */ | |
5add95d4 | 4782 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4783 | { |
36c8b586 | 4784 | struct task_struct *p; |
3a5c359a | 4785 | int retval; |
1da177e4 LT |
4786 | |
4787 | if (pid < 0) | |
3a5c359a | 4788 | return -EINVAL; |
1da177e4 LT |
4789 | |
4790 | retval = -ESRCH; | |
5fe85be0 | 4791 | rcu_read_lock(); |
1da177e4 LT |
4792 | p = find_process_by_pid(pid); |
4793 | if (p) { | |
4794 | retval = security_task_getscheduler(p); | |
4795 | if (!retval) | |
ca94c442 LP |
4796 | retval = p->policy |
4797 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4798 | } |
5fe85be0 | 4799 | rcu_read_unlock(); |
1da177e4 LT |
4800 | return retval; |
4801 | } | |
4802 | ||
4803 | /** | |
ca94c442 | 4804 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4805 | * @pid: the pid in question. |
4806 | * @param: structure containing the RT priority. | |
4807 | */ | |
5add95d4 | 4808 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4809 | { |
4810 | struct sched_param lp; | |
36c8b586 | 4811 | struct task_struct *p; |
3a5c359a | 4812 | int retval; |
1da177e4 LT |
4813 | |
4814 | if (!param || pid < 0) | |
3a5c359a | 4815 | return -EINVAL; |
1da177e4 | 4816 | |
5fe85be0 | 4817 | rcu_read_lock(); |
1da177e4 LT |
4818 | p = find_process_by_pid(pid); |
4819 | retval = -ESRCH; | |
4820 | if (!p) | |
4821 | goto out_unlock; | |
4822 | ||
4823 | retval = security_task_getscheduler(p); | |
4824 | if (retval) | |
4825 | goto out_unlock; | |
4826 | ||
4827 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4828 | rcu_read_unlock(); |
1da177e4 LT |
4829 | |
4830 | /* | |
4831 | * This one might sleep, we cannot do it with a spinlock held ... | |
4832 | */ | |
4833 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4834 | ||
1da177e4 LT |
4835 | return retval; |
4836 | ||
4837 | out_unlock: | |
5fe85be0 | 4838 | rcu_read_unlock(); |
1da177e4 LT |
4839 | return retval; |
4840 | } | |
4841 | ||
96f874e2 | 4842 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4843 | { |
5a16f3d3 | 4844 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4845 | struct task_struct *p; |
4846 | int retval; | |
1da177e4 | 4847 | |
95402b38 | 4848 | get_online_cpus(); |
23f5d142 | 4849 | rcu_read_lock(); |
1da177e4 LT |
4850 | |
4851 | p = find_process_by_pid(pid); | |
4852 | if (!p) { | |
23f5d142 | 4853 | rcu_read_unlock(); |
95402b38 | 4854 | put_online_cpus(); |
1da177e4 LT |
4855 | return -ESRCH; |
4856 | } | |
4857 | ||
23f5d142 | 4858 | /* Prevent p going away */ |
1da177e4 | 4859 | get_task_struct(p); |
23f5d142 | 4860 | rcu_read_unlock(); |
1da177e4 | 4861 | |
5a16f3d3 RR |
4862 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4863 | retval = -ENOMEM; | |
4864 | goto out_put_task; | |
4865 | } | |
4866 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4867 | retval = -ENOMEM; | |
4868 | goto out_free_cpus_allowed; | |
4869 | } | |
1da177e4 | 4870 | retval = -EPERM; |
c69e8d9c | 4871 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4872 | goto out_unlock; |
4873 | ||
e7834f8f DQ |
4874 | retval = security_task_setscheduler(p, 0, NULL); |
4875 | if (retval) | |
4876 | goto out_unlock; | |
4877 | ||
5a16f3d3 RR |
4878 | cpuset_cpus_allowed(p, cpus_allowed); |
4879 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
8707d8b8 | 4880 | again: |
5a16f3d3 | 4881 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4882 | |
8707d8b8 | 4883 | if (!retval) { |
5a16f3d3 RR |
4884 | cpuset_cpus_allowed(p, cpus_allowed); |
4885 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4886 | /* |
4887 | * We must have raced with a concurrent cpuset | |
4888 | * update. Just reset the cpus_allowed to the | |
4889 | * cpuset's cpus_allowed | |
4890 | */ | |
5a16f3d3 | 4891 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4892 | goto again; |
4893 | } | |
4894 | } | |
1da177e4 | 4895 | out_unlock: |
5a16f3d3 RR |
4896 | free_cpumask_var(new_mask); |
4897 | out_free_cpus_allowed: | |
4898 | free_cpumask_var(cpus_allowed); | |
4899 | out_put_task: | |
1da177e4 | 4900 | put_task_struct(p); |
95402b38 | 4901 | put_online_cpus(); |
1da177e4 LT |
4902 | return retval; |
4903 | } | |
4904 | ||
4905 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4906 | struct cpumask *new_mask) |
1da177e4 | 4907 | { |
96f874e2 RR |
4908 | if (len < cpumask_size()) |
4909 | cpumask_clear(new_mask); | |
4910 | else if (len > cpumask_size()) | |
4911 | len = cpumask_size(); | |
4912 | ||
1da177e4 LT |
4913 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4914 | } | |
4915 | ||
4916 | /** | |
4917 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4918 | * @pid: pid of the process | |
4919 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4920 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4921 | */ | |
5add95d4 HC |
4922 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4923 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4924 | { |
5a16f3d3 | 4925 | cpumask_var_t new_mask; |
1da177e4 LT |
4926 | int retval; |
4927 | ||
5a16f3d3 RR |
4928 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4929 | return -ENOMEM; | |
1da177e4 | 4930 | |
5a16f3d3 RR |
4931 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4932 | if (retval == 0) | |
4933 | retval = sched_setaffinity(pid, new_mask); | |
4934 | free_cpumask_var(new_mask); | |
4935 | return retval; | |
1da177e4 LT |
4936 | } |
4937 | ||
96f874e2 | 4938 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4939 | { |
36c8b586 | 4940 | struct task_struct *p; |
31605683 TG |
4941 | unsigned long flags; |
4942 | struct rq *rq; | |
1da177e4 | 4943 | int retval; |
1da177e4 | 4944 | |
95402b38 | 4945 | get_online_cpus(); |
23f5d142 | 4946 | rcu_read_lock(); |
1da177e4 LT |
4947 | |
4948 | retval = -ESRCH; | |
4949 | p = find_process_by_pid(pid); | |
4950 | if (!p) | |
4951 | goto out_unlock; | |
4952 | ||
e7834f8f DQ |
4953 | retval = security_task_getscheduler(p); |
4954 | if (retval) | |
4955 | goto out_unlock; | |
4956 | ||
31605683 | 4957 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4958 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4959 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4960 | |
4961 | out_unlock: | |
23f5d142 | 4962 | rcu_read_unlock(); |
95402b38 | 4963 | put_online_cpus(); |
1da177e4 | 4964 | |
9531b62f | 4965 | return retval; |
1da177e4 LT |
4966 | } |
4967 | ||
4968 | /** | |
4969 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4970 | * @pid: pid of the process | |
4971 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4972 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
4973 | */ | |
5add95d4 HC |
4974 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
4975 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
4976 | { |
4977 | int ret; | |
f17c8607 | 4978 | cpumask_var_t mask; |
1da177e4 | 4979 | |
84fba5ec | 4980 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
4981 | return -EINVAL; |
4982 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
4983 | return -EINVAL; |
4984 | ||
f17c8607 RR |
4985 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
4986 | return -ENOMEM; | |
1da177e4 | 4987 | |
f17c8607 RR |
4988 | ret = sched_getaffinity(pid, mask); |
4989 | if (ret == 0) { | |
8bc037fb | 4990 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
4991 | |
4992 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
4993 | ret = -EFAULT; |
4994 | else | |
cd3d8031 | 4995 | ret = retlen; |
f17c8607 RR |
4996 | } |
4997 | free_cpumask_var(mask); | |
1da177e4 | 4998 | |
f17c8607 | 4999 | return ret; |
1da177e4 LT |
5000 | } |
5001 | ||
5002 | /** | |
5003 | * sys_sched_yield - yield the current processor to other threads. | |
5004 | * | |
dd41f596 IM |
5005 | * This function yields the current CPU to other tasks. If there are no |
5006 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5007 | */ |
5add95d4 | 5008 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5009 | { |
70b97a7f | 5010 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5011 | |
2d72376b | 5012 | schedstat_inc(rq, yld_count); |
4530d7ab | 5013 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5014 | |
5015 | /* | |
5016 | * Since we are going to call schedule() anyway, there's | |
5017 | * no need to preempt or enable interrupts: | |
5018 | */ | |
5019 | __release(rq->lock); | |
8a25d5de | 5020 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5021 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5022 | preempt_enable_no_resched(); |
5023 | ||
5024 | schedule(); | |
5025 | ||
5026 | return 0; | |
5027 | } | |
5028 | ||
d86ee480 PZ |
5029 | static inline int should_resched(void) |
5030 | { | |
5031 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5032 | } | |
5033 | ||
e7b38404 | 5034 | static void __cond_resched(void) |
1da177e4 | 5035 | { |
e7aaaa69 FW |
5036 | add_preempt_count(PREEMPT_ACTIVE); |
5037 | schedule(); | |
5038 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5039 | } |
5040 | ||
02b67cc3 | 5041 | int __sched _cond_resched(void) |
1da177e4 | 5042 | { |
d86ee480 | 5043 | if (should_resched()) { |
1da177e4 LT |
5044 | __cond_resched(); |
5045 | return 1; | |
5046 | } | |
5047 | return 0; | |
5048 | } | |
02b67cc3 | 5049 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5050 | |
5051 | /* | |
613afbf8 | 5052 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5053 | * call schedule, and on return reacquire the lock. |
5054 | * | |
41a2d6cf | 5055 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5056 | * operations here to prevent schedule() from being called twice (once via |
5057 | * spin_unlock(), once by hand). | |
5058 | */ | |
613afbf8 | 5059 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5060 | { |
d86ee480 | 5061 | int resched = should_resched(); |
6df3cecb JK |
5062 | int ret = 0; |
5063 | ||
f607c668 PZ |
5064 | lockdep_assert_held(lock); |
5065 | ||
95c354fe | 5066 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5067 | spin_unlock(lock); |
d86ee480 | 5068 | if (resched) |
95c354fe NP |
5069 | __cond_resched(); |
5070 | else | |
5071 | cpu_relax(); | |
6df3cecb | 5072 | ret = 1; |
1da177e4 | 5073 | spin_lock(lock); |
1da177e4 | 5074 | } |
6df3cecb | 5075 | return ret; |
1da177e4 | 5076 | } |
613afbf8 | 5077 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5078 | |
613afbf8 | 5079 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5080 | { |
5081 | BUG_ON(!in_softirq()); | |
5082 | ||
d86ee480 | 5083 | if (should_resched()) { |
98d82567 | 5084 | local_bh_enable(); |
1da177e4 LT |
5085 | __cond_resched(); |
5086 | local_bh_disable(); | |
5087 | return 1; | |
5088 | } | |
5089 | return 0; | |
5090 | } | |
613afbf8 | 5091 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5092 | |
1da177e4 LT |
5093 | /** |
5094 | * yield - yield the current processor to other threads. | |
5095 | * | |
72fd4a35 | 5096 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5097 | * thread runnable and calls sys_sched_yield(). |
5098 | */ | |
5099 | void __sched yield(void) | |
5100 | { | |
5101 | set_current_state(TASK_RUNNING); | |
5102 | sys_sched_yield(); | |
5103 | } | |
1da177e4 LT |
5104 | EXPORT_SYMBOL(yield); |
5105 | ||
5106 | /* | |
41a2d6cf | 5107 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5108 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5109 | */ |
5110 | void __sched io_schedule(void) | |
5111 | { | |
54d35f29 | 5112 | struct rq *rq = raw_rq(); |
1da177e4 | 5113 | |
0ff92245 | 5114 | delayacct_blkio_start(); |
1da177e4 | 5115 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5116 | current->in_iowait = 1; |
1da177e4 | 5117 | schedule(); |
8f0dfc34 | 5118 | current->in_iowait = 0; |
1da177e4 | 5119 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5120 | delayacct_blkio_end(); |
1da177e4 | 5121 | } |
1da177e4 LT |
5122 | EXPORT_SYMBOL(io_schedule); |
5123 | ||
5124 | long __sched io_schedule_timeout(long timeout) | |
5125 | { | |
54d35f29 | 5126 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5127 | long ret; |
5128 | ||
0ff92245 | 5129 | delayacct_blkio_start(); |
1da177e4 | 5130 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5131 | current->in_iowait = 1; |
1da177e4 | 5132 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5133 | current->in_iowait = 0; |
1da177e4 | 5134 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5135 | delayacct_blkio_end(); |
1da177e4 LT |
5136 | return ret; |
5137 | } | |
5138 | ||
5139 | /** | |
5140 | * sys_sched_get_priority_max - return maximum RT priority. | |
5141 | * @policy: scheduling class. | |
5142 | * | |
5143 | * this syscall returns the maximum rt_priority that can be used | |
5144 | * by a given scheduling class. | |
5145 | */ | |
5add95d4 | 5146 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5147 | { |
5148 | int ret = -EINVAL; | |
5149 | ||
5150 | switch (policy) { | |
5151 | case SCHED_FIFO: | |
5152 | case SCHED_RR: | |
5153 | ret = MAX_USER_RT_PRIO-1; | |
5154 | break; | |
5155 | case SCHED_NORMAL: | |
b0a9499c | 5156 | case SCHED_BATCH: |
dd41f596 | 5157 | case SCHED_IDLE: |
1da177e4 LT |
5158 | ret = 0; |
5159 | break; | |
5160 | } | |
5161 | return ret; | |
5162 | } | |
5163 | ||
5164 | /** | |
5165 | * sys_sched_get_priority_min - return minimum RT priority. | |
5166 | * @policy: scheduling class. | |
5167 | * | |
5168 | * this syscall returns the minimum rt_priority that can be used | |
5169 | * by a given scheduling class. | |
5170 | */ | |
5add95d4 | 5171 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5172 | { |
5173 | int ret = -EINVAL; | |
5174 | ||
5175 | switch (policy) { | |
5176 | case SCHED_FIFO: | |
5177 | case SCHED_RR: | |
5178 | ret = 1; | |
5179 | break; | |
5180 | case SCHED_NORMAL: | |
b0a9499c | 5181 | case SCHED_BATCH: |
dd41f596 | 5182 | case SCHED_IDLE: |
1da177e4 LT |
5183 | ret = 0; |
5184 | } | |
5185 | return ret; | |
5186 | } | |
5187 | ||
5188 | /** | |
5189 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5190 | * @pid: pid of the process. | |
5191 | * @interval: userspace pointer to the timeslice value. | |
5192 | * | |
5193 | * this syscall writes the default timeslice value of a given process | |
5194 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5195 | */ | |
17da2bd9 | 5196 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5197 | struct timespec __user *, interval) |
1da177e4 | 5198 | { |
36c8b586 | 5199 | struct task_struct *p; |
a4ec24b4 | 5200 | unsigned int time_slice; |
dba091b9 TG |
5201 | unsigned long flags; |
5202 | struct rq *rq; | |
3a5c359a | 5203 | int retval; |
1da177e4 | 5204 | struct timespec t; |
1da177e4 LT |
5205 | |
5206 | if (pid < 0) | |
3a5c359a | 5207 | return -EINVAL; |
1da177e4 LT |
5208 | |
5209 | retval = -ESRCH; | |
1a551ae7 | 5210 | rcu_read_lock(); |
1da177e4 LT |
5211 | p = find_process_by_pid(pid); |
5212 | if (!p) | |
5213 | goto out_unlock; | |
5214 | ||
5215 | retval = security_task_getscheduler(p); | |
5216 | if (retval) | |
5217 | goto out_unlock; | |
5218 | ||
dba091b9 TG |
5219 | rq = task_rq_lock(p, &flags); |
5220 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5221 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5222 | |
1a551ae7 | 5223 | rcu_read_unlock(); |
a4ec24b4 | 5224 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5225 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5226 | return retval; |
3a5c359a | 5227 | |
1da177e4 | 5228 | out_unlock: |
1a551ae7 | 5229 | rcu_read_unlock(); |
1da177e4 LT |
5230 | return retval; |
5231 | } | |
5232 | ||
7c731e0a | 5233 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5234 | |
82a1fcb9 | 5235 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5236 | { |
1da177e4 | 5237 | unsigned long free = 0; |
36c8b586 | 5238 | unsigned state; |
1da177e4 | 5239 | |
1da177e4 | 5240 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5241 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5242 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5243 | #if BITS_PER_LONG == 32 |
1da177e4 | 5244 | if (state == TASK_RUNNING) |
3df0fc5b | 5245 | printk(KERN_CONT " running "); |
1da177e4 | 5246 | else |
3df0fc5b | 5247 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5248 | #else |
5249 | if (state == TASK_RUNNING) | |
3df0fc5b | 5250 | printk(KERN_CONT " running task "); |
1da177e4 | 5251 | else |
3df0fc5b | 5252 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5253 | #endif |
5254 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5255 | free = stack_not_used(p); |
1da177e4 | 5256 | #endif |
3df0fc5b | 5257 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5258 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5259 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5260 | |
5fb5e6de | 5261 | show_stack(p, NULL); |
1da177e4 LT |
5262 | } |
5263 | ||
e59e2ae2 | 5264 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5265 | { |
36c8b586 | 5266 | struct task_struct *g, *p; |
1da177e4 | 5267 | |
4bd77321 | 5268 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5269 | printk(KERN_INFO |
5270 | " task PC stack pid father\n"); | |
1da177e4 | 5271 | #else |
3df0fc5b PZ |
5272 | printk(KERN_INFO |
5273 | " task PC stack pid father\n"); | |
1da177e4 LT |
5274 | #endif |
5275 | read_lock(&tasklist_lock); | |
5276 | do_each_thread(g, p) { | |
5277 | /* | |
5278 | * reset the NMI-timeout, listing all files on a slow | |
5279 | * console might take alot of time: | |
5280 | */ | |
5281 | touch_nmi_watchdog(); | |
39bc89fd | 5282 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5283 | sched_show_task(p); |
1da177e4 LT |
5284 | } while_each_thread(g, p); |
5285 | ||
04c9167f JF |
5286 | touch_all_softlockup_watchdogs(); |
5287 | ||
dd41f596 IM |
5288 | #ifdef CONFIG_SCHED_DEBUG |
5289 | sysrq_sched_debug_show(); | |
5290 | #endif | |
1da177e4 | 5291 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5292 | /* |
5293 | * Only show locks if all tasks are dumped: | |
5294 | */ | |
93335a21 | 5295 | if (!state_filter) |
e59e2ae2 | 5296 | debug_show_all_locks(); |
1da177e4 LT |
5297 | } |
5298 | ||
1df21055 IM |
5299 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5300 | { | |
dd41f596 | 5301 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5302 | } |
5303 | ||
f340c0d1 IM |
5304 | /** |
5305 | * init_idle - set up an idle thread for a given CPU | |
5306 | * @idle: task in question | |
5307 | * @cpu: cpu the idle task belongs to | |
5308 | * | |
5309 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5310 | * flag, to make booting more robust. | |
5311 | */ | |
5c1e1767 | 5312 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5313 | { |
70b97a7f | 5314 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5315 | unsigned long flags; |
5316 | ||
05fa785c | 5317 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5318 | |
dd41f596 | 5319 | __sched_fork(idle); |
06b83b5f | 5320 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5321 | idle->se.exec_start = sched_clock(); |
5322 | ||
96f874e2 | 5323 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
dd41f596 | 5324 | __set_task_cpu(idle, cpu); |
1da177e4 | 5325 | |
1da177e4 | 5326 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5327 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5328 | idle->oncpu = 1; | |
5329 | #endif | |
05fa785c | 5330 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5331 | |
5332 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5333 | #if defined(CONFIG_PREEMPT) |
5334 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5335 | #else | |
a1261f54 | 5336 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5337 | #endif |
dd41f596 IM |
5338 | /* |
5339 | * The idle tasks have their own, simple scheduling class: | |
5340 | */ | |
5341 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5342 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5343 | } |
5344 | ||
5345 | /* | |
5346 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5347 | * indicates which cpus entered this state. This is used | |
5348 | * in the rcu update to wait only for active cpus. For system | |
5349 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5350 | * always be CPU_BITS_NONE. |
1da177e4 | 5351 | */ |
6a7b3dc3 | 5352 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5353 | |
19978ca6 IM |
5354 | /* |
5355 | * Increase the granularity value when there are more CPUs, | |
5356 | * because with more CPUs the 'effective latency' as visible | |
5357 | * to users decreases. But the relationship is not linear, | |
5358 | * so pick a second-best guess by going with the log2 of the | |
5359 | * number of CPUs. | |
5360 | * | |
5361 | * This idea comes from the SD scheduler of Con Kolivas: | |
5362 | */ | |
acb4a848 | 5363 | static int get_update_sysctl_factor(void) |
19978ca6 | 5364 | { |
4ca3ef71 | 5365 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5366 | unsigned int factor; |
5367 | ||
5368 | switch (sysctl_sched_tunable_scaling) { | |
5369 | case SCHED_TUNABLESCALING_NONE: | |
5370 | factor = 1; | |
5371 | break; | |
5372 | case SCHED_TUNABLESCALING_LINEAR: | |
5373 | factor = cpus; | |
5374 | break; | |
5375 | case SCHED_TUNABLESCALING_LOG: | |
5376 | default: | |
5377 | factor = 1 + ilog2(cpus); | |
5378 | break; | |
5379 | } | |
19978ca6 | 5380 | |
acb4a848 CE |
5381 | return factor; |
5382 | } | |
19978ca6 | 5383 | |
acb4a848 CE |
5384 | static void update_sysctl(void) |
5385 | { | |
5386 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5387 | |
0bcdcf28 CE |
5388 | #define SET_SYSCTL(name) \ |
5389 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5390 | SET_SYSCTL(sched_min_granularity); | |
5391 | SET_SYSCTL(sched_latency); | |
5392 | SET_SYSCTL(sched_wakeup_granularity); | |
5393 | SET_SYSCTL(sched_shares_ratelimit); | |
5394 | #undef SET_SYSCTL | |
5395 | } | |
55cd5340 | 5396 | |
0bcdcf28 CE |
5397 | static inline void sched_init_granularity(void) |
5398 | { | |
5399 | update_sysctl(); | |
19978ca6 IM |
5400 | } |
5401 | ||
1da177e4 LT |
5402 | #ifdef CONFIG_SMP |
5403 | /* | |
5404 | * This is how migration works: | |
5405 | * | |
969c7921 TH |
5406 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5407 | * stop_one_cpu(). | |
5408 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5409 | * off the CPU) | |
5410 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5411 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5412 | * it and puts it into the right queue. |
969c7921 TH |
5413 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5414 | * is done. | |
1da177e4 LT |
5415 | */ |
5416 | ||
5417 | /* | |
5418 | * Change a given task's CPU affinity. Migrate the thread to a | |
5419 | * proper CPU and schedule it away if the CPU it's executing on | |
5420 | * is removed from the allowed bitmask. | |
5421 | * | |
5422 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5423 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5424 | * call is not atomic; no spinlocks may be held. |
5425 | */ | |
96f874e2 | 5426 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5427 | { |
5428 | unsigned long flags; | |
70b97a7f | 5429 | struct rq *rq; |
969c7921 | 5430 | unsigned int dest_cpu; |
48f24c4d | 5431 | int ret = 0; |
1da177e4 | 5432 | |
65cc8e48 PZ |
5433 | /* |
5434 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5435 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5436 | */ | |
5437 | again: | |
5438 | while (task_is_waking(p)) | |
5439 | cpu_relax(); | |
1da177e4 | 5440 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5441 | if (task_is_waking(p)) { |
5442 | task_rq_unlock(rq, &flags); | |
5443 | goto again; | |
5444 | } | |
e2912009 | 5445 | |
6ad4c188 | 5446 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5447 | ret = -EINVAL; |
5448 | goto out; | |
5449 | } | |
5450 | ||
9985b0ba | 5451 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5452 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5453 | ret = -EINVAL; |
5454 | goto out; | |
5455 | } | |
5456 | ||
73fe6aae | 5457 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5458 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5459 | else { |
96f874e2 RR |
5460 | cpumask_copy(&p->cpus_allowed, new_mask); |
5461 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5462 | } |
5463 | ||
1da177e4 | 5464 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5465 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5466 | goto out; |
5467 | ||
969c7921 TH |
5468 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
5469 | if (migrate_task(p, dest_cpu)) { | |
5470 | struct migration_arg arg = { p, dest_cpu }; | |
1da177e4 LT |
5471 | /* Need help from migration thread: drop lock and wait. */ |
5472 | task_rq_unlock(rq, &flags); | |
969c7921 | 5473 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5474 | tlb_migrate_finish(p->mm); |
5475 | return 0; | |
5476 | } | |
5477 | out: | |
5478 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5479 | |
1da177e4 LT |
5480 | return ret; |
5481 | } | |
cd8ba7cd | 5482 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5483 | |
5484 | /* | |
41a2d6cf | 5485 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5486 | * this because either it can't run here any more (set_cpus_allowed() |
5487 | * away from this CPU, or CPU going down), or because we're | |
5488 | * attempting to rebalance this task on exec (sched_exec). | |
5489 | * | |
5490 | * So we race with normal scheduler movements, but that's OK, as long | |
5491 | * as the task is no longer on this CPU. | |
efc30814 KK |
5492 | * |
5493 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5494 | */ |
efc30814 | 5495 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5496 | { |
70b97a7f | 5497 | struct rq *rq_dest, *rq_src; |
e2912009 | 5498 | int ret = 0; |
1da177e4 | 5499 | |
e761b772 | 5500 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5501 | return ret; |
1da177e4 LT |
5502 | |
5503 | rq_src = cpu_rq(src_cpu); | |
5504 | rq_dest = cpu_rq(dest_cpu); | |
5505 | ||
5506 | double_rq_lock(rq_src, rq_dest); | |
5507 | /* Already moved. */ | |
5508 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5509 | goto done; |
1da177e4 | 5510 | /* Affinity changed (again). */ |
96f874e2 | 5511 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5512 | goto fail; |
1da177e4 | 5513 | |
e2912009 PZ |
5514 | /* |
5515 | * If we're not on a rq, the next wake-up will ensure we're | |
5516 | * placed properly. | |
5517 | */ | |
5518 | if (p->se.on_rq) { | |
2e1cb74a | 5519 | deactivate_task(rq_src, p, 0); |
e2912009 | 5520 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5521 | activate_task(rq_dest, p, 0); |
15afe09b | 5522 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5523 | } |
b1e38734 | 5524 | done: |
efc30814 | 5525 | ret = 1; |
b1e38734 | 5526 | fail: |
1da177e4 | 5527 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5528 | return ret; |
1da177e4 LT |
5529 | } |
5530 | ||
5531 | /* | |
969c7921 TH |
5532 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5533 | * and performs thread migration by bumping thread off CPU then | |
5534 | * 'pushing' onto another runqueue. | |
1da177e4 | 5535 | */ |
969c7921 | 5536 | static int migration_cpu_stop(void *data) |
1da177e4 | 5537 | { |
969c7921 | 5538 | struct migration_arg *arg = data; |
f7b4cddc | 5539 | |
969c7921 TH |
5540 | /* |
5541 | * The original target cpu might have gone down and we might | |
5542 | * be on another cpu but it doesn't matter. | |
5543 | */ | |
f7b4cddc | 5544 | local_irq_disable(); |
969c7921 | 5545 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5546 | local_irq_enable(); |
1da177e4 | 5547 | return 0; |
f7b4cddc ON |
5548 | } |
5549 | ||
1da177e4 | 5550 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5551 | /* |
3a4fa0a2 | 5552 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5553 | */ |
6a1bdc1b | 5554 | void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5555 | { |
1445c08d ON |
5556 | struct rq *rq = cpu_rq(dead_cpu); |
5557 | int needs_cpu, uninitialized_var(dest_cpu); | |
5558 | unsigned long flags; | |
e76bd8d9 | 5559 | |
1445c08d | 5560 | local_irq_save(flags); |
e76bd8d9 | 5561 | |
1445c08d ON |
5562 | raw_spin_lock(&rq->lock); |
5563 | needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING); | |
5564 | if (needs_cpu) | |
5565 | dest_cpu = select_fallback_rq(dead_cpu, p); | |
5566 | raw_spin_unlock(&rq->lock); | |
c1804d54 ON |
5567 | /* |
5568 | * It can only fail if we race with set_cpus_allowed(), | |
5569 | * in the racer should migrate the task anyway. | |
5570 | */ | |
1445c08d | 5571 | if (needs_cpu) |
c1804d54 | 5572 | __migrate_task(p, dead_cpu, dest_cpu); |
1445c08d | 5573 | local_irq_restore(flags); |
1da177e4 LT |
5574 | } |
5575 | ||
5576 | /* | |
5577 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5578 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5579 | * for performance reasons the counter is not stricly tracking tasks to | |
5580 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5581 | * to keep the global sum constant after CPU-down: | |
5582 | */ | |
70b97a7f | 5583 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5584 | { |
6ad4c188 | 5585 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5586 | unsigned long flags; |
5587 | ||
5588 | local_irq_save(flags); | |
5589 | double_rq_lock(rq_src, rq_dest); | |
5590 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5591 | rq_src->nr_uninterruptible = 0; | |
5592 | double_rq_unlock(rq_src, rq_dest); | |
5593 | local_irq_restore(flags); | |
5594 | } | |
5595 | ||
5596 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5597 | static void migrate_live_tasks(int src_cpu) | |
5598 | { | |
48f24c4d | 5599 | struct task_struct *p, *t; |
1da177e4 | 5600 | |
f7b4cddc | 5601 | read_lock(&tasklist_lock); |
1da177e4 | 5602 | |
48f24c4d IM |
5603 | do_each_thread(t, p) { |
5604 | if (p == current) | |
1da177e4 LT |
5605 | continue; |
5606 | ||
48f24c4d IM |
5607 | if (task_cpu(p) == src_cpu) |
5608 | move_task_off_dead_cpu(src_cpu, p); | |
5609 | } while_each_thread(t, p); | |
1da177e4 | 5610 | |
f7b4cddc | 5611 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5612 | } |
5613 | ||
dd41f596 IM |
5614 | /* |
5615 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5616 | * It does so by boosting its priority to highest possible. |
5617 | * Used by CPU offline code. | |
1da177e4 LT |
5618 | */ |
5619 | void sched_idle_next(void) | |
5620 | { | |
48f24c4d | 5621 | int this_cpu = smp_processor_id(); |
70b97a7f | 5622 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5623 | struct task_struct *p = rq->idle; |
5624 | unsigned long flags; | |
5625 | ||
5626 | /* cpu has to be offline */ | |
48f24c4d | 5627 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5628 | |
48f24c4d IM |
5629 | /* |
5630 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5631 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5632 | */ |
05fa785c | 5633 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5634 | |
dd41f596 | 5635 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 5636 | |
94bc9a7b | 5637 | activate_task(rq, p, 0); |
1da177e4 | 5638 | |
05fa785c | 5639 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5640 | } |
5641 | ||
48f24c4d IM |
5642 | /* |
5643 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
5644 | * offline. |
5645 | */ | |
5646 | void idle_task_exit(void) | |
5647 | { | |
5648 | struct mm_struct *mm = current->active_mm; | |
5649 | ||
5650 | BUG_ON(cpu_online(smp_processor_id())); | |
5651 | ||
5652 | if (mm != &init_mm) | |
5653 | switch_mm(mm, &init_mm, current); | |
5654 | mmdrop(mm); | |
5655 | } | |
5656 | ||
054b9108 | 5657 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 5658 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 5659 | { |
70b97a7f | 5660 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
5661 | |
5662 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 5663 | BUG_ON(!p->exit_state); |
1da177e4 LT |
5664 | |
5665 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 5666 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 5667 | |
48f24c4d | 5668 | get_task_struct(p); |
1da177e4 LT |
5669 | |
5670 | /* | |
5671 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 5672 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
5673 | * fine. |
5674 | */ | |
05fa785c | 5675 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 5676 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 5677 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 5678 | |
48f24c4d | 5679 | put_task_struct(p); |
1da177e4 LT |
5680 | } |
5681 | ||
5682 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
5683 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
5684 | { | |
70b97a7f | 5685 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 5686 | struct task_struct *next; |
48f24c4d | 5687 | |
dd41f596 IM |
5688 | for ( ; ; ) { |
5689 | if (!rq->nr_running) | |
5690 | break; | |
b67802ea | 5691 | next = pick_next_task(rq); |
dd41f596 IM |
5692 | if (!next) |
5693 | break; | |
79c53799 | 5694 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 5695 | migrate_dead(dead_cpu, next); |
e692ab53 | 5696 | |
1da177e4 LT |
5697 | } |
5698 | } | |
dce48a84 TG |
5699 | |
5700 | /* | |
5701 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
5702 | */ | |
5703 | static void calc_global_load_remove(struct rq *rq) | |
5704 | { | |
5705 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 5706 | rq->calc_load_active = 0; |
dce48a84 | 5707 | } |
1da177e4 LT |
5708 | #endif /* CONFIG_HOTPLUG_CPU */ |
5709 | ||
e692ab53 NP |
5710 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5711 | ||
5712 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5713 | { |
5714 | .procname = "sched_domain", | |
c57baf1e | 5715 | .mode = 0555, |
e0361851 | 5716 | }, |
56992309 | 5717 | {} |
e692ab53 NP |
5718 | }; |
5719 | ||
5720 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5721 | { |
5722 | .procname = "kernel", | |
c57baf1e | 5723 | .mode = 0555, |
e0361851 AD |
5724 | .child = sd_ctl_dir, |
5725 | }, | |
56992309 | 5726 | {} |
e692ab53 NP |
5727 | }; |
5728 | ||
5729 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5730 | { | |
5731 | struct ctl_table *entry = | |
5cf9f062 | 5732 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5733 | |
e692ab53 NP |
5734 | return entry; |
5735 | } | |
5736 | ||
6382bc90 MM |
5737 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5738 | { | |
cd790076 | 5739 | struct ctl_table *entry; |
6382bc90 | 5740 | |
cd790076 MM |
5741 | /* |
5742 | * In the intermediate directories, both the child directory and | |
5743 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5744 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5745 | * static strings and all have proc handlers. |
5746 | */ | |
5747 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5748 | if (entry->child) |
5749 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5750 | if (entry->proc_handler == NULL) |
5751 | kfree(entry->procname); | |
5752 | } | |
6382bc90 MM |
5753 | |
5754 | kfree(*tablep); | |
5755 | *tablep = NULL; | |
5756 | } | |
5757 | ||
e692ab53 | 5758 | static void |
e0361851 | 5759 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5760 | const char *procname, void *data, int maxlen, |
5761 | mode_t mode, proc_handler *proc_handler) | |
5762 | { | |
e692ab53 NP |
5763 | entry->procname = procname; |
5764 | entry->data = data; | |
5765 | entry->maxlen = maxlen; | |
5766 | entry->mode = mode; | |
5767 | entry->proc_handler = proc_handler; | |
5768 | } | |
5769 | ||
5770 | static struct ctl_table * | |
5771 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5772 | { | |
a5d8c348 | 5773 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5774 | |
ad1cdc1d MM |
5775 | if (table == NULL) |
5776 | return NULL; | |
5777 | ||
e0361851 | 5778 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5779 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5780 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5781 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5782 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5783 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5784 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5785 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5786 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5787 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5788 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5789 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5790 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5791 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5792 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5793 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5794 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5795 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5796 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5797 | &sd->cache_nice_tries, |
5798 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5799 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5800 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5801 | set_table_entry(&table[11], "name", sd->name, |
5802 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5803 | /* &table[12] is terminator */ | |
e692ab53 NP |
5804 | |
5805 | return table; | |
5806 | } | |
5807 | ||
9a4e7159 | 5808 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5809 | { |
5810 | struct ctl_table *entry, *table; | |
5811 | struct sched_domain *sd; | |
5812 | int domain_num = 0, i; | |
5813 | char buf[32]; | |
5814 | ||
5815 | for_each_domain(cpu, sd) | |
5816 | domain_num++; | |
5817 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5818 | if (table == NULL) |
5819 | return NULL; | |
e692ab53 NP |
5820 | |
5821 | i = 0; | |
5822 | for_each_domain(cpu, sd) { | |
5823 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5824 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5825 | entry->mode = 0555; |
e692ab53 NP |
5826 | entry->child = sd_alloc_ctl_domain_table(sd); |
5827 | entry++; | |
5828 | i++; | |
5829 | } | |
5830 | return table; | |
5831 | } | |
5832 | ||
5833 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5834 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5835 | { |
6ad4c188 | 5836 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5837 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5838 | char buf[32]; | |
5839 | ||
7378547f MM |
5840 | WARN_ON(sd_ctl_dir[0].child); |
5841 | sd_ctl_dir[0].child = entry; | |
5842 | ||
ad1cdc1d MM |
5843 | if (entry == NULL) |
5844 | return; | |
5845 | ||
6ad4c188 | 5846 | for_each_possible_cpu(i) { |
e692ab53 | 5847 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5848 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5849 | entry->mode = 0555; |
e692ab53 | 5850 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5851 | entry++; |
e692ab53 | 5852 | } |
7378547f MM |
5853 | |
5854 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5855 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5856 | } | |
6382bc90 | 5857 | |
7378547f | 5858 | /* may be called multiple times per register */ |
6382bc90 MM |
5859 | static void unregister_sched_domain_sysctl(void) |
5860 | { | |
7378547f MM |
5861 | if (sd_sysctl_header) |
5862 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5863 | sd_sysctl_header = NULL; |
7378547f MM |
5864 | if (sd_ctl_dir[0].child) |
5865 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5866 | } |
e692ab53 | 5867 | #else |
6382bc90 MM |
5868 | static void register_sched_domain_sysctl(void) |
5869 | { | |
5870 | } | |
5871 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5872 | { |
5873 | } | |
5874 | #endif | |
5875 | ||
1f11eb6a GH |
5876 | static void set_rq_online(struct rq *rq) |
5877 | { | |
5878 | if (!rq->online) { | |
5879 | const struct sched_class *class; | |
5880 | ||
c6c4927b | 5881 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5882 | rq->online = 1; |
5883 | ||
5884 | for_each_class(class) { | |
5885 | if (class->rq_online) | |
5886 | class->rq_online(rq); | |
5887 | } | |
5888 | } | |
5889 | } | |
5890 | ||
5891 | static void set_rq_offline(struct rq *rq) | |
5892 | { | |
5893 | if (rq->online) { | |
5894 | const struct sched_class *class; | |
5895 | ||
5896 | for_each_class(class) { | |
5897 | if (class->rq_offline) | |
5898 | class->rq_offline(rq); | |
5899 | } | |
5900 | ||
c6c4927b | 5901 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5902 | rq->online = 0; |
5903 | } | |
5904 | } | |
5905 | ||
1da177e4 LT |
5906 | /* |
5907 | * migration_call - callback that gets triggered when a CPU is added. | |
5908 | * Here we can start up the necessary migration thread for the new CPU. | |
5909 | */ | |
48f24c4d IM |
5910 | static int __cpuinit |
5911 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5912 | { |
48f24c4d | 5913 | int cpu = (long)hcpu; |
1da177e4 | 5914 | unsigned long flags; |
969c7921 | 5915 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5916 | |
5917 | switch (action) { | |
5be9361c | 5918 | |
1da177e4 | 5919 | case CPU_UP_PREPARE: |
8bb78442 | 5920 | case CPU_UP_PREPARE_FROZEN: |
a468d389 | 5921 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5922 | break; |
48f24c4d | 5923 | |
1da177e4 | 5924 | case CPU_ONLINE: |
8bb78442 | 5925 | case CPU_ONLINE_FROZEN: |
1f94ef59 | 5926 | /* Update our root-domain */ |
05fa785c | 5927 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5928 | if (rq->rd) { |
c6c4927b | 5929 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5930 | |
5931 | set_rq_online(rq); | |
1f94ef59 | 5932 | } |
05fa785c | 5933 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5934 | break; |
48f24c4d | 5935 | |
1da177e4 | 5936 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 5937 | case CPU_DEAD: |
8bb78442 | 5938 | case CPU_DEAD_FROZEN: |
1da177e4 | 5939 | migrate_live_tasks(cpu); |
1da177e4 | 5940 | /* Idle task back to normal (off runqueue, low prio) */ |
05fa785c | 5941 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 5942 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
5943 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
5944 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 5945 | migrate_dead_tasks(cpu); |
05fa785c | 5946 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
5947 | migrate_nr_uninterruptible(rq); |
5948 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 5949 | calc_global_load_remove(rq); |
1da177e4 | 5950 | break; |
57d885fe | 5951 | |
08f503b0 GH |
5952 | case CPU_DYING: |
5953 | case CPU_DYING_FROZEN: | |
57d885fe | 5954 | /* Update our root-domain */ |
05fa785c | 5955 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5956 | if (rq->rd) { |
c6c4927b | 5957 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5958 | set_rq_offline(rq); |
57d885fe | 5959 | } |
05fa785c | 5960 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 5961 | break; |
1da177e4 LT |
5962 | #endif |
5963 | } | |
5964 | return NOTIFY_OK; | |
5965 | } | |
5966 | ||
f38b0820 PM |
5967 | /* |
5968 | * Register at high priority so that task migration (migrate_all_tasks) | |
5969 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5970 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5971 | */ |
26c2143b | 5972 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5973 | .notifier_call = migration_call, |
50a323b7 | 5974 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5975 | }; |
5976 | ||
3a101d05 TH |
5977 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5978 | unsigned long action, void *hcpu) | |
5979 | { | |
5980 | switch (action & ~CPU_TASKS_FROZEN) { | |
5981 | case CPU_ONLINE: | |
5982 | case CPU_DOWN_FAILED: | |
5983 | set_cpu_active((long)hcpu, true); | |
5984 | return NOTIFY_OK; | |
5985 | default: | |
5986 | return NOTIFY_DONE; | |
5987 | } | |
5988 | } | |
5989 | ||
5990 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5991 | unsigned long action, void *hcpu) | |
5992 | { | |
5993 | switch (action & ~CPU_TASKS_FROZEN) { | |
5994 | case CPU_DOWN_PREPARE: | |
5995 | set_cpu_active((long)hcpu, false); | |
5996 | return NOTIFY_OK; | |
5997 | default: | |
5998 | return NOTIFY_DONE; | |
5999 | } | |
6000 | } | |
6001 | ||
7babe8db | 6002 | static int __init migration_init(void) |
1da177e4 LT |
6003 | { |
6004 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6005 | int err; |
48f24c4d | 6006 | |
3a101d05 | 6007 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6008 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6009 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6010 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6011 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6012 | |
3a101d05 TH |
6013 | /* Register cpu active notifiers */ |
6014 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6015 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6016 | ||
a004cd42 | 6017 | return 0; |
1da177e4 | 6018 | } |
7babe8db | 6019 | early_initcall(migration_init); |
1da177e4 LT |
6020 | #endif |
6021 | ||
6022 | #ifdef CONFIG_SMP | |
476f3534 | 6023 | |
3e9830dc | 6024 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6025 | |
f6630114 MT |
6026 | static __read_mostly int sched_domain_debug_enabled; |
6027 | ||
6028 | static int __init sched_domain_debug_setup(char *str) | |
6029 | { | |
6030 | sched_domain_debug_enabled = 1; | |
6031 | ||
6032 | return 0; | |
6033 | } | |
6034 | early_param("sched_debug", sched_domain_debug_setup); | |
6035 | ||
7c16ec58 | 6036 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6037 | struct cpumask *groupmask) |
1da177e4 | 6038 | { |
4dcf6aff | 6039 | struct sched_group *group = sd->groups; |
434d53b0 | 6040 | char str[256]; |
1da177e4 | 6041 | |
968ea6d8 | 6042 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6043 | cpumask_clear(groupmask); |
4dcf6aff IM |
6044 | |
6045 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6046 | ||
6047 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6048 | printk("does not load-balance\n"); |
4dcf6aff | 6049 | if (sd->parent) |
3df0fc5b PZ |
6050 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6051 | " has parent"); | |
4dcf6aff | 6052 | return -1; |
41c7ce9a NP |
6053 | } |
6054 | ||
3df0fc5b | 6055 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6056 | |
758b2cdc | 6057 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6058 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6059 | "CPU%d\n", cpu); | |
4dcf6aff | 6060 | } |
758b2cdc | 6061 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6062 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6063 | " CPU%d\n", cpu); | |
4dcf6aff | 6064 | } |
1da177e4 | 6065 | |
4dcf6aff | 6066 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6067 | do { |
4dcf6aff | 6068 | if (!group) { |
3df0fc5b PZ |
6069 | printk("\n"); |
6070 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6071 | break; |
6072 | } | |
6073 | ||
18a3885f | 6074 | if (!group->cpu_power) { |
3df0fc5b PZ |
6075 | printk(KERN_CONT "\n"); |
6076 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6077 | "set\n"); | |
4dcf6aff IM |
6078 | break; |
6079 | } | |
1da177e4 | 6080 | |
758b2cdc | 6081 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6082 | printk(KERN_CONT "\n"); |
6083 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6084 | break; |
6085 | } | |
1da177e4 | 6086 | |
758b2cdc | 6087 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6088 | printk(KERN_CONT "\n"); |
6089 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6090 | break; |
6091 | } | |
1da177e4 | 6092 | |
758b2cdc | 6093 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6094 | |
968ea6d8 | 6095 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6096 | |
3df0fc5b | 6097 | printk(KERN_CONT " %s", str); |
18a3885f | 6098 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6099 | printk(KERN_CONT " (cpu_power = %d)", |
6100 | group->cpu_power); | |
381512cf | 6101 | } |
1da177e4 | 6102 | |
4dcf6aff IM |
6103 | group = group->next; |
6104 | } while (group != sd->groups); | |
3df0fc5b | 6105 | printk(KERN_CONT "\n"); |
1da177e4 | 6106 | |
758b2cdc | 6107 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6108 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6109 | |
758b2cdc RR |
6110 | if (sd->parent && |
6111 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6112 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6113 | "of domain->span\n"); | |
4dcf6aff IM |
6114 | return 0; |
6115 | } | |
1da177e4 | 6116 | |
4dcf6aff IM |
6117 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6118 | { | |
d5dd3db1 | 6119 | cpumask_var_t groupmask; |
4dcf6aff | 6120 | int level = 0; |
1da177e4 | 6121 | |
f6630114 MT |
6122 | if (!sched_domain_debug_enabled) |
6123 | return; | |
6124 | ||
4dcf6aff IM |
6125 | if (!sd) { |
6126 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6127 | return; | |
6128 | } | |
1da177e4 | 6129 | |
4dcf6aff IM |
6130 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6131 | ||
d5dd3db1 | 6132 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6133 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6134 | return; | |
6135 | } | |
6136 | ||
4dcf6aff | 6137 | for (;;) { |
7c16ec58 | 6138 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6139 | break; |
1da177e4 LT |
6140 | level++; |
6141 | sd = sd->parent; | |
33859f7f | 6142 | if (!sd) |
4dcf6aff IM |
6143 | break; |
6144 | } | |
d5dd3db1 | 6145 | free_cpumask_var(groupmask); |
1da177e4 | 6146 | } |
6d6bc0ad | 6147 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6148 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6149 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6150 | |
1a20ff27 | 6151 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6152 | { |
758b2cdc | 6153 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6154 | return 1; |
6155 | ||
6156 | /* Following flags need at least 2 groups */ | |
6157 | if (sd->flags & (SD_LOAD_BALANCE | | |
6158 | SD_BALANCE_NEWIDLE | | |
6159 | SD_BALANCE_FORK | | |
89c4710e SS |
6160 | SD_BALANCE_EXEC | |
6161 | SD_SHARE_CPUPOWER | | |
6162 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6163 | if (sd->groups != sd->groups->next) |
6164 | return 0; | |
6165 | } | |
6166 | ||
6167 | /* Following flags don't use groups */ | |
c88d5910 | 6168 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6169 | return 0; |
6170 | ||
6171 | return 1; | |
6172 | } | |
6173 | ||
48f24c4d IM |
6174 | static int |
6175 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6176 | { |
6177 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6178 | ||
6179 | if (sd_degenerate(parent)) | |
6180 | return 1; | |
6181 | ||
758b2cdc | 6182 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6183 | return 0; |
6184 | ||
245af2c7 SS |
6185 | /* Flags needing groups don't count if only 1 group in parent */ |
6186 | if (parent->groups == parent->groups->next) { | |
6187 | pflags &= ~(SD_LOAD_BALANCE | | |
6188 | SD_BALANCE_NEWIDLE | | |
6189 | SD_BALANCE_FORK | | |
89c4710e SS |
6190 | SD_BALANCE_EXEC | |
6191 | SD_SHARE_CPUPOWER | | |
6192 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6193 | if (nr_node_ids == 1) |
6194 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6195 | } |
6196 | if (~cflags & pflags) | |
6197 | return 0; | |
6198 | ||
6199 | return 1; | |
6200 | } | |
6201 | ||
c6c4927b RR |
6202 | static void free_rootdomain(struct root_domain *rd) |
6203 | { | |
047106ad PZ |
6204 | synchronize_sched(); |
6205 | ||
68e74568 RR |
6206 | cpupri_cleanup(&rd->cpupri); |
6207 | ||
c6c4927b RR |
6208 | free_cpumask_var(rd->rto_mask); |
6209 | free_cpumask_var(rd->online); | |
6210 | free_cpumask_var(rd->span); | |
6211 | kfree(rd); | |
6212 | } | |
6213 | ||
57d885fe GH |
6214 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6215 | { | |
a0490fa3 | 6216 | struct root_domain *old_rd = NULL; |
57d885fe | 6217 | unsigned long flags; |
57d885fe | 6218 | |
05fa785c | 6219 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6220 | |
6221 | if (rq->rd) { | |
a0490fa3 | 6222 | old_rd = rq->rd; |
57d885fe | 6223 | |
c6c4927b | 6224 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6225 | set_rq_offline(rq); |
57d885fe | 6226 | |
c6c4927b | 6227 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6228 | |
a0490fa3 IM |
6229 | /* |
6230 | * If we dont want to free the old_rt yet then | |
6231 | * set old_rd to NULL to skip the freeing later | |
6232 | * in this function: | |
6233 | */ | |
6234 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6235 | old_rd = NULL; | |
57d885fe GH |
6236 | } |
6237 | ||
6238 | atomic_inc(&rd->refcount); | |
6239 | rq->rd = rd; | |
6240 | ||
c6c4927b | 6241 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6242 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6243 | set_rq_online(rq); |
57d885fe | 6244 | |
05fa785c | 6245 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6246 | |
6247 | if (old_rd) | |
6248 | free_rootdomain(old_rd); | |
57d885fe GH |
6249 | } |
6250 | ||
fd5e1b5d | 6251 | static int init_rootdomain(struct root_domain *rd, bool bootmem) |
57d885fe | 6252 | { |
36b7b6d4 PE |
6253 | gfp_t gfp = GFP_KERNEL; |
6254 | ||
57d885fe GH |
6255 | memset(rd, 0, sizeof(*rd)); |
6256 | ||
36b7b6d4 PE |
6257 | if (bootmem) |
6258 | gfp = GFP_NOWAIT; | |
c6c4927b | 6259 | |
36b7b6d4 | 6260 | if (!alloc_cpumask_var(&rd->span, gfp)) |
0c910d28 | 6261 | goto out; |
36b7b6d4 | 6262 | if (!alloc_cpumask_var(&rd->online, gfp)) |
c6c4927b | 6263 | goto free_span; |
36b7b6d4 | 6264 | if (!alloc_cpumask_var(&rd->rto_mask, gfp)) |
c6c4927b | 6265 | goto free_online; |
6e0534f2 | 6266 | |
0fb53029 | 6267 | if (cpupri_init(&rd->cpupri, bootmem) != 0) |
68e74568 | 6268 | goto free_rto_mask; |
c6c4927b | 6269 | return 0; |
6e0534f2 | 6270 | |
68e74568 RR |
6271 | free_rto_mask: |
6272 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6273 | free_online: |
6274 | free_cpumask_var(rd->online); | |
6275 | free_span: | |
6276 | free_cpumask_var(rd->span); | |
0c910d28 | 6277 | out: |
c6c4927b | 6278 | return -ENOMEM; |
57d885fe GH |
6279 | } |
6280 | ||
6281 | static void init_defrootdomain(void) | |
6282 | { | |
c6c4927b RR |
6283 | init_rootdomain(&def_root_domain, true); |
6284 | ||
57d885fe GH |
6285 | atomic_set(&def_root_domain.refcount, 1); |
6286 | } | |
6287 | ||
dc938520 | 6288 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6289 | { |
6290 | struct root_domain *rd; | |
6291 | ||
6292 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6293 | if (!rd) | |
6294 | return NULL; | |
6295 | ||
c6c4927b RR |
6296 | if (init_rootdomain(rd, false) != 0) { |
6297 | kfree(rd); | |
6298 | return NULL; | |
6299 | } | |
57d885fe GH |
6300 | |
6301 | return rd; | |
6302 | } | |
6303 | ||
1da177e4 | 6304 | /* |
0eab9146 | 6305 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6306 | * hold the hotplug lock. |
6307 | */ | |
0eab9146 IM |
6308 | static void |
6309 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6310 | { |
70b97a7f | 6311 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6312 | struct sched_domain *tmp; |
6313 | ||
669c55e9 PZ |
6314 | for (tmp = sd; tmp; tmp = tmp->parent) |
6315 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6316 | ||
245af2c7 | 6317 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6318 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6319 | struct sched_domain *parent = tmp->parent; |
6320 | if (!parent) | |
6321 | break; | |
f29c9b1c | 6322 | |
1a848870 | 6323 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6324 | tmp->parent = parent->parent; |
1a848870 SS |
6325 | if (parent->parent) |
6326 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6327 | } else |
6328 | tmp = tmp->parent; | |
245af2c7 SS |
6329 | } |
6330 | ||
1a848870 | 6331 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6332 | sd = sd->parent; |
1a848870 SS |
6333 | if (sd) |
6334 | sd->child = NULL; | |
6335 | } | |
1da177e4 LT |
6336 | |
6337 | sched_domain_debug(sd, cpu); | |
6338 | ||
57d885fe | 6339 | rq_attach_root(rq, rd); |
674311d5 | 6340 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6341 | } |
6342 | ||
6343 | /* cpus with isolated domains */ | |
dcc30a35 | 6344 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6345 | |
6346 | /* Setup the mask of cpus configured for isolated domains */ | |
6347 | static int __init isolated_cpu_setup(char *str) | |
6348 | { | |
bdddd296 | 6349 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6350 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6351 | return 1; |
6352 | } | |
6353 | ||
8927f494 | 6354 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6355 | |
6356 | /* | |
6711cab4 SS |
6357 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6358 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6359 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6360 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6361 | * |
6362 | * init_sched_build_groups will build a circular linked list of the groups | |
6363 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6364 | * and ->cpu_power to 0. | |
6365 | */ | |
a616058b | 6366 | static void |
96f874e2 RR |
6367 | init_sched_build_groups(const struct cpumask *span, |
6368 | const struct cpumask *cpu_map, | |
6369 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6370 | struct sched_group **sg, |
96f874e2 RR |
6371 | struct cpumask *tmpmask), |
6372 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6373 | { |
6374 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6375 | int i; |
6376 | ||
96f874e2 | 6377 | cpumask_clear(covered); |
7c16ec58 | 6378 | |
abcd083a | 6379 | for_each_cpu(i, span) { |
6711cab4 | 6380 | struct sched_group *sg; |
7c16ec58 | 6381 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6382 | int j; |
6383 | ||
758b2cdc | 6384 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6385 | continue; |
6386 | ||
758b2cdc | 6387 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6388 | sg->cpu_power = 0; |
1da177e4 | 6389 | |
abcd083a | 6390 | for_each_cpu(j, span) { |
7c16ec58 | 6391 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6392 | continue; |
6393 | ||
96f874e2 | 6394 | cpumask_set_cpu(j, covered); |
758b2cdc | 6395 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6396 | } |
6397 | if (!first) | |
6398 | first = sg; | |
6399 | if (last) | |
6400 | last->next = sg; | |
6401 | last = sg; | |
6402 | } | |
6403 | last->next = first; | |
6404 | } | |
6405 | ||
9c1cfda2 | 6406 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6407 | |
9c1cfda2 | 6408 | #ifdef CONFIG_NUMA |
198e2f18 | 6409 | |
9c1cfda2 JH |
6410 | /** |
6411 | * find_next_best_node - find the next node to include in a sched_domain | |
6412 | * @node: node whose sched_domain we're building | |
6413 | * @used_nodes: nodes already in the sched_domain | |
6414 | * | |
41a2d6cf | 6415 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6416 | * finds the closest node not already in the @used_nodes map. |
6417 | * | |
6418 | * Should use nodemask_t. | |
6419 | */ | |
c5f59f08 | 6420 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6421 | { |
6422 | int i, n, val, min_val, best_node = 0; | |
6423 | ||
6424 | min_val = INT_MAX; | |
6425 | ||
076ac2af | 6426 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6427 | /* Start at @node */ |
076ac2af | 6428 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6429 | |
6430 | if (!nr_cpus_node(n)) | |
6431 | continue; | |
6432 | ||
6433 | /* Skip already used nodes */ | |
c5f59f08 | 6434 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6435 | continue; |
6436 | ||
6437 | /* Simple min distance search */ | |
6438 | val = node_distance(node, n); | |
6439 | ||
6440 | if (val < min_val) { | |
6441 | min_val = val; | |
6442 | best_node = n; | |
6443 | } | |
6444 | } | |
6445 | ||
c5f59f08 | 6446 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6447 | return best_node; |
6448 | } | |
6449 | ||
6450 | /** | |
6451 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6452 | * @node: node whose cpumask we're constructing | |
73486722 | 6453 | * @span: resulting cpumask |
9c1cfda2 | 6454 | * |
41a2d6cf | 6455 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6456 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6457 | * out optimally. | |
6458 | */ | |
96f874e2 | 6459 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6460 | { |
c5f59f08 | 6461 | nodemask_t used_nodes; |
48f24c4d | 6462 | int i; |
9c1cfda2 | 6463 | |
6ca09dfc | 6464 | cpumask_clear(span); |
c5f59f08 | 6465 | nodes_clear(used_nodes); |
9c1cfda2 | 6466 | |
6ca09dfc | 6467 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6468 | node_set(node, used_nodes); |
9c1cfda2 JH |
6469 | |
6470 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6471 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6472 | |
6ca09dfc | 6473 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6474 | } |
9c1cfda2 | 6475 | } |
6d6bc0ad | 6476 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6477 | |
5c45bf27 | 6478 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6479 | |
6c99e9ad RR |
6480 | /* |
6481 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6482 | * |
6483 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6484 | * and struct sched_domain. ) | |
6c99e9ad RR |
6485 | */ |
6486 | struct static_sched_group { | |
6487 | struct sched_group sg; | |
6488 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6489 | }; | |
6490 | ||
6491 | struct static_sched_domain { | |
6492 | struct sched_domain sd; | |
6493 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6494 | }; | |
6495 | ||
49a02c51 AH |
6496 | struct s_data { |
6497 | #ifdef CONFIG_NUMA | |
6498 | int sd_allnodes; | |
6499 | cpumask_var_t domainspan; | |
6500 | cpumask_var_t covered; | |
6501 | cpumask_var_t notcovered; | |
6502 | #endif | |
6503 | cpumask_var_t nodemask; | |
6504 | cpumask_var_t this_sibling_map; | |
6505 | cpumask_var_t this_core_map; | |
6506 | cpumask_var_t send_covered; | |
6507 | cpumask_var_t tmpmask; | |
6508 | struct sched_group **sched_group_nodes; | |
6509 | struct root_domain *rd; | |
6510 | }; | |
6511 | ||
2109b99e AH |
6512 | enum s_alloc { |
6513 | sa_sched_groups = 0, | |
6514 | sa_rootdomain, | |
6515 | sa_tmpmask, | |
6516 | sa_send_covered, | |
6517 | sa_this_core_map, | |
6518 | sa_this_sibling_map, | |
6519 | sa_nodemask, | |
6520 | sa_sched_group_nodes, | |
6521 | #ifdef CONFIG_NUMA | |
6522 | sa_notcovered, | |
6523 | sa_covered, | |
6524 | sa_domainspan, | |
6525 | #endif | |
6526 | sa_none, | |
6527 | }; | |
6528 | ||
9c1cfda2 | 6529 | /* |
48f24c4d | 6530 | * SMT sched-domains: |
9c1cfda2 | 6531 | */ |
1da177e4 | 6532 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6533 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6534 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6535 | |
41a2d6cf | 6536 | static int |
96f874e2 RR |
6537 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6538 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6539 | { |
6711cab4 | 6540 | if (sg) |
1871e52c | 6541 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6542 | return cpu; |
6543 | } | |
6d6bc0ad | 6544 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6545 | |
48f24c4d IM |
6546 | /* |
6547 | * multi-core sched-domains: | |
6548 | */ | |
1e9f28fa | 6549 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6550 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6551 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
6d6bc0ad | 6552 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa SS |
6553 | |
6554 | #if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT) | |
41a2d6cf | 6555 | static int |
96f874e2 RR |
6556 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6557 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6558 | { |
6711cab4 | 6559 | int group; |
7c16ec58 | 6560 | |
c69fc56d | 6561 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6562 | group = cpumask_first(mask); |
6711cab4 | 6563 | if (sg) |
6c99e9ad | 6564 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6565 | return group; |
1e9f28fa SS |
6566 | } |
6567 | #elif defined(CONFIG_SCHED_MC) | |
41a2d6cf | 6568 | static int |
96f874e2 RR |
6569 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6570 | struct sched_group **sg, struct cpumask *unused) | |
1e9f28fa | 6571 | { |
6711cab4 | 6572 | if (sg) |
6c99e9ad | 6573 | *sg = &per_cpu(sched_group_core, cpu).sg; |
1e9f28fa SS |
6574 | return cpu; |
6575 | } | |
6576 | #endif | |
6577 | ||
6c99e9ad RR |
6578 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6579 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6580 | |
41a2d6cf | 6581 | static int |
96f874e2 RR |
6582 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6583 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6584 | { |
6711cab4 | 6585 | int group; |
48f24c4d | 6586 | #ifdef CONFIG_SCHED_MC |
6ca09dfc | 6587 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6588 | group = cpumask_first(mask); |
1e9f28fa | 6589 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6590 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6591 | group = cpumask_first(mask); |
1da177e4 | 6592 | #else |
6711cab4 | 6593 | group = cpu; |
1da177e4 | 6594 | #endif |
6711cab4 | 6595 | if (sg) |
6c99e9ad | 6596 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6597 | return group; |
1da177e4 LT |
6598 | } |
6599 | ||
6600 | #ifdef CONFIG_NUMA | |
1da177e4 | 6601 | /* |
9c1cfda2 JH |
6602 | * The init_sched_build_groups can't handle what we want to do with node |
6603 | * groups, so roll our own. Now each node has its own list of groups which | |
6604 | * gets dynamically allocated. | |
1da177e4 | 6605 | */ |
62ea9ceb | 6606 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6607 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6608 | |
62ea9ceb | 6609 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6610 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6611 | |
96f874e2 RR |
6612 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6613 | struct sched_group **sg, | |
6614 | struct cpumask *nodemask) | |
9c1cfda2 | 6615 | { |
6711cab4 SS |
6616 | int group; |
6617 | ||
6ca09dfc | 6618 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6619 | group = cpumask_first(nodemask); |
6711cab4 SS |
6620 | |
6621 | if (sg) | |
6c99e9ad | 6622 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6623 | return group; |
1da177e4 | 6624 | } |
6711cab4 | 6625 | |
08069033 SS |
6626 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6627 | { | |
6628 | struct sched_group *sg = group_head; | |
6629 | int j; | |
6630 | ||
6631 | if (!sg) | |
6632 | return; | |
3a5c359a | 6633 | do { |
758b2cdc | 6634 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6635 | struct sched_domain *sd; |
08069033 | 6636 | |
6c99e9ad | 6637 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6638 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6639 | /* |
6640 | * Only add "power" once for each | |
6641 | * physical package. | |
6642 | */ | |
6643 | continue; | |
6644 | } | |
08069033 | 6645 | |
18a3885f | 6646 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6647 | } |
6648 | sg = sg->next; | |
6649 | } while (sg != group_head); | |
08069033 | 6650 | } |
0601a88d AH |
6651 | |
6652 | static int build_numa_sched_groups(struct s_data *d, | |
6653 | const struct cpumask *cpu_map, int num) | |
6654 | { | |
6655 | struct sched_domain *sd; | |
6656 | struct sched_group *sg, *prev; | |
6657 | int n, j; | |
6658 | ||
6659 | cpumask_clear(d->covered); | |
6660 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6661 | if (cpumask_empty(d->nodemask)) { | |
6662 | d->sched_group_nodes[num] = NULL; | |
6663 | goto out; | |
6664 | } | |
6665 | ||
6666 | sched_domain_node_span(num, d->domainspan); | |
6667 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6668 | ||
6669 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6670 | GFP_KERNEL, num); | |
6671 | if (!sg) { | |
3df0fc5b PZ |
6672 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6673 | num); | |
0601a88d AH |
6674 | return -ENOMEM; |
6675 | } | |
6676 | d->sched_group_nodes[num] = sg; | |
6677 | ||
6678 | for_each_cpu(j, d->nodemask) { | |
6679 | sd = &per_cpu(node_domains, j).sd; | |
6680 | sd->groups = sg; | |
6681 | } | |
6682 | ||
18a3885f | 6683 | sg->cpu_power = 0; |
0601a88d AH |
6684 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6685 | sg->next = sg; | |
6686 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6687 | ||
6688 | prev = sg; | |
6689 | for (j = 0; j < nr_node_ids; j++) { | |
6690 | n = (num + j) % nr_node_ids; | |
6691 | cpumask_complement(d->notcovered, d->covered); | |
6692 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6693 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6694 | if (cpumask_empty(d->tmpmask)) | |
6695 | break; | |
6696 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6697 | if (cpumask_empty(d->tmpmask)) | |
6698 | continue; | |
6699 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6700 | GFP_KERNEL, num); | |
6701 | if (!sg) { | |
3df0fc5b PZ |
6702 | printk(KERN_WARNING |
6703 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6704 | return -ENOMEM; |
6705 | } | |
18a3885f | 6706 | sg->cpu_power = 0; |
0601a88d AH |
6707 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6708 | sg->next = prev->next; | |
6709 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6710 | prev->next = sg; | |
6711 | prev = sg; | |
6712 | } | |
6713 | out: | |
6714 | return 0; | |
6715 | } | |
6d6bc0ad | 6716 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6717 | |
a616058b | 6718 | #ifdef CONFIG_NUMA |
51888ca2 | 6719 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6720 | static void free_sched_groups(const struct cpumask *cpu_map, |
6721 | struct cpumask *nodemask) | |
51888ca2 | 6722 | { |
a616058b | 6723 | int cpu, i; |
51888ca2 | 6724 | |
abcd083a | 6725 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6726 | struct sched_group **sched_group_nodes |
6727 | = sched_group_nodes_bycpu[cpu]; | |
6728 | ||
51888ca2 SV |
6729 | if (!sched_group_nodes) |
6730 | continue; | |
6731 | ||
076ac2af | 6732 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6733 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6734 | ||
6ca09dfc | 6735 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6736 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6737 | continue; |
6738 | ||
6739 | if (sg == NULL) | |
6740 | continue; | |
6741 | sg = sg->next; | |
6742 | next_sg: | |
6743 | oldsg = sg; | |
6744 | sg = sg->next; | |
6745 | kfree(oldsg); | |
6746 | if (oldsg != sched_group_nodes[i]) | |
6747 | goto next_sg; | |
6748 | } | |
6749 | kfree(sched_group_nodes); | |
6750 | sched_group_nodes_bycpu[cpu] = NULL; | |
6751 | } | |
51888ca2 | 6752 | } |
6d6bc0ad | 6753 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6754 | static void free_sched_groups(const struct cpumask *cpu_map, |
6755 | struct cpumask *nodemask) | |
a616058b SS |
6756 | { |
6757 | } | |
6d6bc0ad | 6758 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6759 | |
89c4710e SS |
6760 | /* |
6761 | * Initialize sched groups cpu_power. | |
6762 | * | |
6763 | * cpu_power indicates the capacity of sched group, which is used while | |
6764 | * distributing the load between different sched groups in a sched domain. | |
6765 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6766 | * there are asymmetries in the topology. If there are asymmetries, group | |
6767 | * having more cpu_power will pickup more load compared to the group having | |
6768 | * less cpu_power. | |
89c4710e SS |
6769 | */ |
6770 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6771 | { | |
6772 | struct sched_domain *child; | |
6773 | struct sched_group *group; | |
f93e65c1 PZ |
6774 | long power; |
6775 | int weight; | |
89c4710e SS |
6776 | |
6777 | WARN_ON(!sd || !sd->groups); | |
6778 | ||
13318a71 | 6779 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6780 | return; |
6781 | ||
6782 | child = sd->child; | |
6783 | ||
18a3885f | 6784 | sd->groups->cpu_power = 0; |
5517d86b | 6785 | |
f93e65c1 PZ |
6786 | if (!child) { |
6787 | power = SCHED_LOAD_SCALE; | |
6788 | weight = cpumask_weight(sched_domain_span(sd)); | |
6789 | /* | |
6790 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6791 | * Usually multiple threads get a better yield out of |
6792 | * that one core than a single thread would have, | |
6793 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6794 | */ |
a52bfd73 PZ |
6795 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6796 | power *= sd->smt_gain; | |
f93e65c1 | 6797 | power /= weight; |
a52bfd73 PZ |
6798 | power >>= SCHED_LOAD_SHIFT; |
6799 | } | |
18a3885f | 6800 | sd->groups->cpu_power += power; |
89c4710e SS |
6801 | return; |
6802 | } | |
6803 | ||
89c4710e | 6804 | /* |
f93e65c1 | 6805 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6806 | */ |
6807 | group = child->groups; | |
6808 | do { | |
18a3885f | 6809 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6810 | group = group->next; |
6811 | } while (group != child->groups); | |
6812 | } | |
6813 | ||
7c16ec58 MT |
6814 | /* |
6815 | * Initializers for schedule domains | |
6816 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6817 | */ | |
6818 | ||
a5d8c348 IM |
6819 | #ifdef CONFIG_SCHED_DEBUG |
6820 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6821 | #else | |
6822 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6823 | #endif | |
6824 | ||
7c16ec58 | 6825 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6826 | |
7c16ec58 MT |
6827 | #define SD_INIT_FUNC(type) \ |
6828 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6829 | { \ | |
6830 | memset(sd, 0, sizeof(*sd)); \ | |
6831 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6832 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6833 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6834 | } |
6835 | ||
6836 | SD_INIT_FUNC(CPU) | |
6837 | #ifdef CONFIG_NUMA | |
6838 | SD_INIT_FUNC(ALLNODES) | |
6839 | SD_INIT_FUNC(NODE) | |
6840 | #endif | |
6841 | #ifdef CONFIG_SCHED_SMT | |
6842 | SD_INIT_FUNC(SIBLING) | |
6843 | #endif | |
6844 | #ifdef CONFIG_SCHED_MC | |
6845 | SD_INIT_FUNC(MC) | |
6846 | #endif | |
6847 | ||
1d3504fc HS |
6848 | static int default_relax_domain_level = -1; |
6849 | ||
6850 | static int __init setup_relax_domain_level(char *str) | |
6851 | { | |
30e0e178 LZ |
6852 | unsigned long val; |
6853 | ||
6854 | val = simple_strtoul(str, NULL, 0); | |
6855 | if (val < SD_LV_MAX) | |
6856 | default_relax_domain_level = val; | |
6857 | ||
1d3504fc HS |
6858 | return 1; |
6859 | } | |
6860 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6861 | ||
6862 | static void set_domain_attribute(struct sched_domain *sd, | |
6863 | struct sched_domain_attr *attr) | |
6864 | { | |
6865 | int request; | |
6866 | ||
6867 | if (!attr || attr->relax_domain_level < 0) { | |
6868 | if (default_relax_domain_level < 0) | |
6869 | return; | |
6870 | else | |
6871 | request = default_relax_domain_level; | |
6872 | } else | |
6873 | request = attr->relax_domain_level; | |
6874 | if (request < sd->level) { | |
6875 | /* turn off idle balance on this domain */ | |
c88d5910 | 6876 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6877 | } else { |
6878 | /* turn on idle balance on this domain */ | |
c88d5910 | 6879 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6880 | } |
6881 | } | |
6882 | ||
2109b99e AH |
6883 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6884 | const struct cpumask *cpu_map) | |
6885 | { | |
6886 | switch (what) { | |
6887 | case sa_sched_groups: | |
6888 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6889 | d->sched_group_nodes = NULL; | |
6890 | case sa_rootdomain: | |
6891 | free_rootdomain(d->rd); /* fall through */ | |
6892 | case sa_tmpmask: | |
6893 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6894 | case sa_send_covered: | |
6895 | free_cpumask_var(d->send_covered); /* fall through */ | |
6896 | case sa_this_core_map: | |
6897 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6898 | case sa_this_sibling_map: | |
6899 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6900 | case sa_nodemask: | |
6901 | free_cpumask_var(d->nodemask); /* fall through */ | |
6902 | case sa_sched_group_nodes: | |
d1b55138 | 6903 | #ifdef CONFIG_NUMA |
2109b99e AH |
6904 | kfree(d->sched_group_nodes); /* fall through */ |
6905 | case sa_notcovered: | |
6906 | free_cpumask_var(d->notcovered); /* fall through */ | |
6907 | case sa_covered: | |
6908 | free_cpumask_var(d->covered); /* fall through */ | |
6909 | case sa_domainspan: | |
6910 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6911 | #endif |
2109b99e AH |
6912 | case sa_none: |
6913 | break; | |
6914 | } | |
6915 | } | |
3404c8d9 | 6916 | |
2109b99e AH |
6917 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6918 | const struct cpumask *cpu_map) | |
6919 | { | |
3404c8d9 | 6920 | #ifdef CONFIG_NUMA |
2109b99e AH |
6921 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6922 | return sa_none; | |
6923 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6924 | return sa_domainspan; | |
6925 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6926 | return sa_covered; | |
6927 | /* Allocate the per-node list of sched groups */ | |
6928 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6929 | sizeof(struct sched_group *), GFP_KERNEL); | |
6930 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6931 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6932 | return sa_notcovered; |
d1b55138 | 6933 | } |
2109b99e | 6934 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6935 | #endif |
2109b99e AH |
6936 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6937 | return sa_sched_group_nodes; | |
6938 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6939 | return sa_nodemask; | |
6940 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6941 | return sa_this_sibling_map; | |
6942 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) | |
6943 | return sa_this_core_map; | |
6944 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) | |
6945 | return sa_send_covered; | |
6946 | d->rd = alloc_rootdomain(); | |
6947 | if (!d->rd) { | |
3df0fc5b | 6948 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6949 | return sa_tmpmask; |
57d885fe | 6950 | } |
2109b99e AH |
6951 | return sa_rootdomain; |
6952 | } | |
57d885fe | 6953 | |
7f4588f3 AH |
6954 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6955 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6956 | { | |
6957 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6958 | #ifdef CONFIG_NUMA |
7f4588f3 | 6959 | struct sched_domain *parent; |
1da177e4 | 6960 | |
7f4588f3 AH |
6961 | d->sd_allnodes = 0; |
6962 | if (cpumask_weight(cpu_map) > | |
6963 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6964 | sd = &per_cpu(allnodes_domains, i).sd; | |
6965 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6966 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6967 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6968 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6969 | d->sd_allnodes = 1; | |
6970 | } | |
6971 | parent = sd; | |
6972 | ||
6973 | sd = &per_cpu(node_domains, i).sd; | |
6974 | SD_INIT(sd, NODE); | |
6975 | set_domain_attribute(sd, attr); | |
6976 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6977 | sd->parent = parent; | |
6978 | if (parent) | |
6979 | parent->child = sd; | |
6980 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6981 | #endif |
7f4588f3 AH |
6982 | return sd; |
6983 | } | |
1da177e4 | 6984 | |
87cce662 AH |
6985 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6986 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6987 | struct sched_domain *parent, int i) | |
6988 | { | |
6989 | struct sched_domain *sd; | |
6990 | sd = &per_cpu(phys_domains, i).sd; | |
6991 | SD_INIT(sd, CPU); | |
6992 | set_domain_attribute(sd, attr); | |
6993 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6994 | sd->parent = parent; | |
6995 | if (parent) | |
6996 | parent->child = sd; | |
6997 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6998 | return sd; | |
6999 | } | |
1da177e4 | 7000 | |
410c4081 AH |
7001 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7002 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7003 | struct sched_domain *parent, int i) | |
7004 | { | |
7005 | struct sched_domain *sd = parent; | |
1e9f28fa | 7006 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7007 | sd = &per_cpu(core_domains, i).sd; |
7008 | SD_INIT(sd, MC); | |
7009 | set_domain_attribute(sd, attr); | |
7010 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7011 | sd->parent = parent; | |
7012 | parent->child = sd; | |
7013 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7014 | #endif |
410c4081 AH |
7015 | return sd; |
7016 | } | |
1e9f28fa | 7017 | |
d8173535 AH |
7018 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7019 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7020 | struct sched_domain *parent, int i) | |
7021 | { | |
7022 | struct sched_domain *sd = parent; | |
1da177e4 | 7023 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7024 | sd = &per_cpu(cpu_domains, i).sd; |
7025 | SD_INIT(sd, SIBLING); | |
7026 | set_domain_attribute(sd, attr); | |
7027 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7028 | sd->parent = parent; | |
7029 | parent->child = sd; | |
7030 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7031 | #endif |
d8173535 AH |
7032 | return sd; |
7033 | } | |
1da177e4 | 7034 | |
0e8e85c9 AH |
7035 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7036 | const struct cpumask *cpu_map, int cpu) | |
7037 | { | |
7038 | switch (l) { | |
1da177e4 | 7039 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7040 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7041 | cpumask_and(d->this_sibling_map, cpu_map, | |
7042 | topology_thread_cpumask(cpu)); | |
7043 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7044 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7045 | &cpu_to_cpu_group, | |
7046 | d->send_covered, d->tmpmask); | |
7047 | break; | |
1da177e4 | 7048 | #endif |
1e9f28fa | 7049 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7050 | case SD_LV_MC: /* set up multi-core groups */ |
7051 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7052 | if (cpu == cpumask_first(d->this_core_map)) | |
7053 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7054 | &cpu_to_core_group, | |
7055 | d->send_covered, d->tmpmask); | |
7056 | break; | |
1e9f28fa | 7057 | #endif |
86548096 AH |
7058 | case SD_LV_CPU: /* set up physical groups */ |
7059 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7060 | if (!cpumask_empty(d->nodemask)) | |
7061 | init_sched_build_groups(d->nodemask, cpu_map, | |
7062 | &cpu_to_phys_group, | |
7063 | d->send_covered, d->tmpmask); | |
7064 | break; | |
1da177e4 | 7065 | #ifdef CONFIG_NUMA |
de616e36 AH |
7066 | case SD_LV_ALLNODES: |
7067 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7068 | d->send_covered, d->tmpmask); | |
7069 | break; | |
7070 | #endif | |
0e8e85c9 AH |
7071 | default: |
7072 | break; | |
7c16ec58 | 7073 | } |
0e8e85c9 | 7074 | } |
9c1cfda2 | 7075 | |
2109b99e AH |
7076 | /* |
7077 | * Build sched domains for a given set of cpus and attach the sched domains | |
7078 | * to the individual cpus | |
7079 | */ | |
7080 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7081 | struct sched_domain_attr *attr) | |
7082 | { | |
7083 | enum s_alloc alloc_state = sa_none; | |
7084 | struct s_data d; | |
294b0c96 | 7085 | struct sched_domain *sd; |
2109b99e | 7086 | int i; |
7c16ec58 | 7087 | #ifdef CONFIG_NUMA |
2109b99e | 7088 | d.sd_allnodes = 0; |
7c16ec58 | 7089 | #endif |
9c1cfda2 | 7090 | |
2109b99e AH |
7091 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7092 | if (alloc_state != sa_rootdomain) | |
7093 | goto error; | |
7094 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7095 | |
1da177e4 | 7096 | /* |
1a20ff27 | 7097 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7098 | */ |
abcd083a | 7099 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7100 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7101 | cpu_map); | |
9761eea8 | 7102 | |
7f4588f3 | 7103 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7104 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7105 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7106 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7107 | } |
9c1cfda2 | 7108 | |
abcd083a | 7109 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7110 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
a2af04cd | 7111 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7112 | } |
9c1cfda2 | 7113 | |
1da177e4 | 7114 | /* Set up physical groups */ |
86548096 AH |
7115 | for (i = 0; i < nr_node_ids; i++) |
7116 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7117 | |
1da177e4 LT |
7118 | #ifdef CONFIG_NUMA |
7119 | /* Set up node groups */ | |
de616e36 AH |
7120 | if (d.sd_allnodes) |
7121 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7122 | |
0601a88d AH |
7123 | for (i = 0; i < nr_node_ids; i++) |
7124 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7125 | goto error; |
1da177e4 LT |
7126 | #endif |
7127 | ||
7128 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7129 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7130 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7131 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7132 | init_sched_groups_power(i, sd); |
5c45bf27 | 7133 | } |
1da177e4 | 7134 | #endif |
1e9f28fa | 7135 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7136 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7137 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7138 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7139 | } |
7140 | #endif | |
1e9f28fa | 7141 | |
abcd083a | 7142 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7143 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7144 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7145 | } |
7146 | ||
9c1cfda2 | 7147 | #ifdef CONFIG_NUMA |
076ac2af | 7148 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7149 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7150 | |
49a02c51 | 7151 | if (d.sd_allnodes) { |
6711cab4 | 7152 | struct sched_group *sg; |
f712c0c7 | 7153 | |
96f874e2 | 7154 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7155 | d.tmpmask); |
f712c0c7 SS |
7156 | init_numa_sched_groups_power(sg); |
7157 | } | |
9c1cfda2 JH |
7158 | #endif |
7159 | ||
1da177e4 | 7160 | /* Attach the domains */ |
abcd083a | 7161 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7162 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7163 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7164 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7165 | sd = &per_cpu(core_domains, i).sd; |
1da177e4 | 7166 | #else |
6c99e9ad | 7167 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7168 | #endif |
49a02c51 | 7169 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7170 | } |
51888ca2 | 7171 | |
2109b99e AH |
7172 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7173 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7174 | return 0; | |
51888ca2 | 7175 | |
51888ca2 | 7176 | error: |
2109b99e AH |
7177 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7178 | return -ENOMEM; | |
1da177e4 | 7179 | } |
029190c5 | 7180 | |
96f874e2 | 7181 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7182 | { |
7183 | return __build_sched_domains(cpu_map, NULL); | |
7184 | } | |
7185 | ||
acc3f5d7 | 7186 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7187 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7188 | static struct sched_domain_attr *dattr_cur; |
7189 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7190 | |
7191 | /* | |
7192 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7193 | * cpumask) fails, then fallback to a single sched domain, |
7194 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7195 | */ |
4212823f | 7196 | static cpumask_var_t fallback_doms; |
029190c5 | 7197 | |
ee79d1bd HC |
7198 | /* |
7199 | * arch_update_cpu_topology lets virtualized architectures update the | |
7200 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7201 | * or 0 if it stayed the same. | |
7202 | */ | |
7203 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7204 | { |
ee79d1bd | 7205 | return 0; |
22e52b07 HC |
7206 | } |
7207 | ||
acc3f5d7 RR |
7208 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7209 | { | |
7210 | int i; | |
7211 | cpumask_var_t *doms; | |
7212 | ||
7213 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7214 | if (!doms) | |
7215 | return NULL; | |
7216 | for (i = 0; i < ndoms; i++) { | |
7217 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7218 | free_sched_domains(doms, i); | |
7219 | return NULL; | |
7220 | } | |
7221 | } | |
7222 | return doms; | |
7223 | } | |
7224 | ||
7225 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7226 | { | |
7227 | unsigned int i; | |
7228 | for (i = 0; i < ndoms; i++) | |
7229 | free_cpumask_var(doms[i]); | |
7230 | kfree(doms); | |
7231 | } | |
7232 | ||
1a20ff27 | 7233 | /* |
41a2d6cf | 7234 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7235 | * For now this just excludes isolated cpus, but could be used to |
7236 | * exclude other special cases in the future. | |
1a20ff27 | 7237 | */ |
96f874e2 | 7238 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7239 | { |
7378547f MM |
7240 | int err; |
7241 | ||
22e52b07 | 7242 | arch_update_cpu_topology(); |
029190c5 | 7243 | ndoms_cur = 1; |
acc3f5d7 | 7244 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7245 | if (!doms_cur) |
acc3f5d7 RR |
7246 | doms_cur = &fallback_doms; |
7247 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7248 | dattr_cur = NULL; |
acc3f5d7 | 7249 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7250 | register_sched_domain_sysctl(); |
7378547f MM |
7251 | |
7252 | return err; | |
1a20ff27 DG |
7253 | } |
7254 | ||
96f874e2 RR |
7255 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7256 | struct cpumask *tmpmask) | |
1da177e4 | 7257 | { |
7c16ec58 | 7258 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7259 | } |
1da177e4 | 7260 | |
1a20ff27 DG |
7261 | /* |
7262 | * Detach sched domains from a group of cpus specified in cpu_map | |
7263 | * These cpus will now be attached to the NULL domain | |
7264 | */ | |
96f874e2 | 7265 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7266 | { |
96f874e2 RR |
7267 | /* Save because hotplug lock held. */ |
7268 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7269 | int i; |
7270 | ||
abcd083a | 7271 | for_each_cpu(i, cpu_map) |
57d885fe | 7272 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7273 | synchronize_sched(); |
96f874e2 | 7274 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7275 | } |
7276 | ||
1d3504fc HS |
7277 | /* handle null as "default" */ |
7278 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7279 | struct sched_domain_attr *new, int idx_new) | |
7280 | { | |
7281 | struct sched_domain_attr tmp; | |
7282 | ||
7283 | /* fast path */ | |
7284 | if (!new && !cur) | |
7285 | return 1; | |
7286 | ||
7287 | tmp = SD_ATTR_INIT; | |
7288 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7289 | new ? (new + idx_new) : &tmp, | |
7290 | sizeof(struct sched_domain_attr)); | |
7291 | } | |
7292 | ||
029190c5 PJ |
7293 | /* |
7294 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7295 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7296 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7297 | * It destroys each deleted domain and builds each new domain. | |
7298 | * | |
acc3f5d7 | 7299 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7300 | * The masks don't intersect (don't overlap.) We should setup one |
7301 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7302 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7303 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7304 | * it as it is. | |
7305 | * | |
acc3f5d7 RR |
7306 | * The passed in 'doms_new' should be allocated using |
7307 | * alloc_sched_domains. This routine takes ownership of it and will | |
7308 | * free_sched_domains it when done with it. If the caller failed the | |
7309 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7310 | * and partition_sched_domains() will fallback to the single partition | |
7311 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7312 | * |
96f874e2 | 7313 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7314 | * ndoms_new == 0 is a special case for destroying existing domains, |
7315 | * and it will not create the default domain. | |
dfb512ec | 7316 | * |
029190c5 PJ |
7317 | * Call with hotplug lock held |
7318 | */ | |
acc3f5d7 | 7319 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7320 | struct sched_domain_attr *dattr_new) |
029190c5 | 7321 | { |
dfb512ec | 7322 | int i, j, n; |
d65bd5ec | 7323 | int new_topology; |
029190c5 | 7324 | |
712555ee | 7325 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7326 | |
7378547f MM |
7327 | /* always unregister in case we don't destroy any domains */ |
7328 | unregister_sched_domain_sysctl(); | |
7329 | ||
d65bd5ec HC |
7330 | /* Let architecture update cpu core mappings. */ |
7331 | new_topology = arch_update_cpu_topology(); | |
7332 | ||
dfb512ec | 7333 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7334 | |
7335 | /* Destroy deleted domains */ | |
7336 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7337 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7338 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7339 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7340 | goto match1; |
7341 | } | |
7342 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7343 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7344 | match1: |
7345 | ; | |
7346 | } | |
7347 | ||
e761b772 MK |
7348 | if (doms_new == NULL) { |
7349 | ndoms_cur = 0; | |
acc3f5d7 | 7350 | doms_new = &fallback_doms; |
6ad4c188 | 7351 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7352 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7353 | } |
7354 | ||
029190c5 PJ |
7355 | /* Build new domains */ |
7356 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7357 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7358 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7359 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7360 | goto match2; |
7361 | } | |
7362 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7363 | __build_sched_domains(doms_new[i], |
1d3504fc | 7364 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7365 | match2: |
7366 | ; | |
7367 | } | |
7368 | ||
7369 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7370 | if (doms_cur != &fallback_doms) |
7371 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7372 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7373 | doms_cur = doms_new; |
1d3504fc | 7374 | dattr_cur = dattr_new; |
029190c5 | 7375 | ndoms_cur = ndoms_new; |
7378547f MM |
7376 | |
7377 | register_sched_domain_sysctl(); | |
a1835615 | 7378 | |
712555ee | 7379 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7380 | } |
7381 | ||
5c45bf27 | 7382 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7383 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7384 | { |
95402b38 | 7385 | get_online_cpus(); |
dfb512ec MK |
7386 | |
7387 | /* Destroy domains first to force the rebuild */ | |
7388 | partition_sched_domains(0, NULL, NULL); | |
7389 | ||
e761b772 | 7390 | rebuild_sched_domains(); |
95402b38 | 7391 | put_online_cpus(); |
5c45bf27 SS |
7392 | } |
7393 | ||
7394 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7395 | { | |
afb8a9b7 | 7396 | unsigned int level = 0; |
5c45bf27 | 7397 | |
afb8a9b7 GS |
7398 | if (sscanf(buf, "%u", &level) != 1) |
7399 | return -EINVAL; | |
7400 | ||
7401 | /* | |
7402 | * level is always be positive so don't check for | |
7403 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7404 | * What happens on 0 or 1 byte write, | |
7405 | * need to check for count as well? | |
7406 | */ | |
7407 | ||
7408 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7409 | return -EINVAL; |
7410 | ||
7411 | if (smt) | |
afb8a9b7 | 7412 | sched_smt_power_savings = level; |
5c45bf27 | 7413 | else |
afb8a9b7 | 7414 | sched_mc_power_savings = level; |
5c45bf27 | 7415 | |
c70f22d2 | 7416 | arch_reinit_sched_domains(); |
5c45bf27 | 7417 | |
c70f22d2 | 7418 | return count; |
5c45bf27 SS |
7419 | } |
7420 | ||
5c45bf27 | 7421 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7422 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7423 | struct sysdev_class_attribute *attr, |
f718cd4a | 7424 | char *page) |
5c45bf27 SS |
7425 | { |
7426 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7427 | } | |
f718cd4a | 7428 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7429 | struct sysdev_class_attribute *attr, |
48f24c4d | 7430 | const char *buf, size_t count) |
5c45bf27 SS |
7431 | { |
7432 | return sched_power_savings_store(buf, count, 0); | |
7433 | } | |
f718cd4a AK |
7434 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7435 | sched_mc_power_savings_show, | |
7436 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7437 | #endif |
7438 | ||
7439 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7440 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7441 | struct sysdev_class_attribute *attr, |
f718cd4a | 7442 | char *page) |
5c45bf27 SS |
7443 | { |
7444 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7445 | } | |
f718cd4a | 7446 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7447 | struct sysdev_class_attribute *attr, |
48f24c4d | 7448 | const char *buf, size_t count) |
5c45bf27 SS |
7449 | { |
7450 | return sched_power_savings_store(buf, count, 1); | |
7451 | } | |
f718cd4a AK |
7452 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7453 | sched_smt_power_savings_show, | |
6707de00 AB |
7454 | sched_smt_power_savings_store); |
7455 | #endif | |
7456 | ||
39aac648 | 7457 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7458 | { |
7459 | int err = 0; | |
7460 | ||
7461 | #ifdef CONFIG_SCHED_SMT | |
7462 | if (smt_capable()) | |
7463 | err = sysfs_create_file(&cls->kset.kobj, | |
7464 | &attr_sched_smt_power_savings.attr); | |
7465 | #endif | |
7466 | #ifdef CONFIG_SCHED_MC | |
7467 | if (!err && mc_capable()) | |
7468 | err = sysfs_create_file(&cls->kset.kobj, | |
7469 | &attr_sched_mc_power_savings.attr); | |
7470 | #endif | |
7471 | return err; | |
7472 | } | |
6d6bc0ad | 7473 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7474 | |
1da177e4 | 7475 | /* |
3a101d05 TH |
7476 | * Update cpusets according to cpu_active mask. If cpusets are |
7477 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7478 | * around partition_sched_domains(). | |
1da177e4 | 7479 | */ |
3a101d05 TH |
7480 | static int __cpuexit cpuset_cpu_active(struct notifier_block *nfb, |
7481 | unsigned long action, void *hcpu) | |
e761b772 | 7482 | { |
3a101d05 | 7483 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7484 | case CPU_ONLINE: |
6ad4c188 | 7485 | case CPU_DOWN_FAILED: |
3a101d05 | 7486 | cpuset_update_active_cpus(); |
e761b772 | 7487 | return NOTIFY_OK; |
3a101d05 TH |
7488 | default: |
7489 | return NOTIFY_DONE; | |
7490 | } | |
7491 | } | |
e761b772 | 7492 | |
3a101d05 TH |
7493 | static int __cpuexit cpuset_cpu_inactive(struct notifier_block *nfb, |
7494 | unsigned long action, void *hcpu) | |
7495 | { | |
7496 | switch (action & ~CPU_TASKS_FROZEN) { | |
7497 | case CPU_DOWN_PREPARE: | |
7498 | cpuset_update_active_cpus(); | |
7499 | return NOTIFY_OK; | |
e761b772 MK |
7500 | default: |
7501 | return NOTIFY_DONE; | |
7502 | } | |
7503 | } | |
e761b772 MK |
7504 | |
7505 | static int update_runtime(struct notifier_block *nfb, | |
7506 | unsigned long action, void *hcpu) | |
1da177e4 | 7507 | { |
7def2be1 PZ |
7508 | int cpu = (int)(long)hcpu; |
7509 | ||
1da177e4 | 7510 | switch (action) { |
1da177e4 | 7511 | case CPU_DOWN_PREPARE: |
8bb78442 | 7512 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7513 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7514 | return NOTIFY_OK; |
7515 | ||
1da177e4 | 7516 | case CPU_DOWN_FAILED: |
8bb78442 | 7517 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7518 | case CPU_ONLINE: |
8bb78442 | 7519 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7520 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7521 | return NOTIFY_OK; |
7522 | ||
1da177e4 LT |
7523 | default: |
7524 | return NOTIFY_DONE; | |
7525 | } | |
1da177e4 | 7526 | } |
1da177e4 LT |
7527 | |
7528 | void __init sched_init_smp(void) | |
7529 | { | |
dcc30a35 RR |
7530 | cpumask_var_t non_isolated_cpus; |
7531 | ||
7532 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7533 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7534 | |
434d53b0 MT |
7535 | #if defined(CONFIG_NUMA) |
7536 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7537 | GFP_KERNEL); | |
7538 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7539 | #endif | |
95402b38 | 7540 | get_online_cpus(); |
712555ee | 7541 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7542 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7543 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7544 | if (cpumask_empty(non_isolated_cpus)) | |
7545 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7546 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7547 | put_online_cpus(); |
e761b772 | 7548 | |
3a101d05 TH |
7549 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7550 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7551 | |
7552 | /* RT runtime code needs to handle some hotplug events */ | |
7553 | hotcpu_notifier(update_runtime, 0); | |
7554 | ||
b328ca18 | 7555 | init_hrtick(); |
5c1e1767 NP |
7556 | |
7557 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7558 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7559 | BUG(); |
19978ca6 | 7560 | sched_init_granularity(); |
dcc30a35 | 7561 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7562 | |
0e3900e6 | 7563 | init_sched_rt_class(); |
1da177e4 LT |
7564 | } |
7565 | #else | |
7566 | void __init sched_init_smp(void) | |
7567 | { | |
19978ca6 | 7568 | sched_init_granularity(); |
1da177e4 LT |
7569 | } |
7570 | #endif /* CONFIG_SMP */ | |
7571 | ||
cd1bb94b AB |
7572 | const_debug unsigned int sysctl_timer_migration = 1; |
7573 | ||
1da177e4 LT |
7574 | int in_sched_functions(unsigned long addr) |
7575 | { | |
1da177e4 LT |
7576 | return in_lock_functions(addr) || |
7577 | (addr >= (unsigned long)__sched_text_start | |
7578 | && addr < (unsigned long)__sched_text_end); | |
7579 | } | |
7580 | ||
a9957449 | 7581 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7582 | { |
7583 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7584 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7585 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7586 | cfs_rq->rq = rq; | |
7587 | #endif | |
67e9fb2a | 7588 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7589 | } |
7590 | ||
fa85ae24 PZ |
7591 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7592 | { | |
7593 | struct rt_prio_array *array; | |
7594 | int i; | |
7595 | ||
7596 | array = &rt_rq->active; | |
7597 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7598 | INIT_LIST_HEAD(array->queue + i); | |
7599 | __clear_bit(i, array->bitmap); | |
7600 | } | |
7601 | /* delimiter for bitsearch: */ | |
7602 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7603 | ||
052f1dc7 | 7604 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7605 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7606 | #ifdef CONFIG_SMP |
e864c499 | 7607 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7608 | #endif |
48d5e258 | 7609 | #endif |
fa85ae24 PZ |
7610 | #ifdef CONFIG_SMP |
7611 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7612 | rt_rq->overloaded = 0; |
05fa785c | 7613 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7614 | #endif |
7615 | ||
7616 | rt_rq->rt_time = 0; | |
7617 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7618 | rt_rq->rt_runtime = 0; |
0986b11b | 7619 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7620 | |
052f1dc7 | 7621 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7622 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7623 | rt_rq->rq = rq; |
7624 | #endif | |
fa85ae24 PZ |
7625 | } |
7626 | ||
6f505b16 | 7627 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
7628 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
7629 | struct sched_entity *se, int cpu, int add, | |
7630 | struct sched_entity *parent) | |
6f505b16 | 7631 | { |
ec7dc8ac | 7632 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7633 | tg->cfs_rq[cpu] = cfs_rq; |
7634 | init_cfs_rq(cfs_rq, rq); | |
7635 | cfs_rq->tg = tg; | |
7636 | if (add) | |
7637 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
7638 | ||
7639 | tg->se[cpu] = se; | |
354d60c2 DG |
7640 | /* se could be NULL for init_task_group */ |
7641 | if (!se) | |
7642 | return; | |
7643 | ||
ec7dc8ac DG |
7644 | if (!parent) |
7645 | se->cfs_rq = &rq->cfs; | |
7646 | else | |
7647 | se->cfs_rq = parent->my_q; | |
7648 | ||
6f505b16 PZ |
7649 | se->my_q = cfs_rq; |
7650 | se->load.weight = tg->shares; | |
e05510d0 | 7651 | se->load.inv_weight = 0; |
ec7dc8ac | 7652 | se->parent = parent; |
6f505b16 | 7653 | } |
052f1dc7 | 7654 | #endif |
6f505b16 | 7655 | |
052f1dc7 | 7656 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
7657 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
7658 | struct sched_rt_entity *rt_se, int cpu, int add, | |
7659 | struct sched_rt_entity *parent) | |
6f505b16 | 7660 | { |
ec7dc8ac DG |
7661 | struct rq *rq = cpu_rq(cpu); |
7662 | ||
6f505b16 PZ |
7663 | tg->rt_rq[cpu] = rt_rq; |
7664 | init_rt_rq(rt_rq, rq); | |
7665 | rt_rq->tg = tg; | |
ac086bc2 | 7666 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7667 | if (add) |
7668 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
7669 | ||
7670 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7671 | if (!rt_se) |
7672 | return; | |
7673 | ||
ec7dc8ac DG |
7674 | if (!parent) |
7675 | rt_se->rt_rq = &rq->rt; | |
7676 | else | |
7677 | rt_se->rt_rq = parent->my_q; | |
7678 | ||
6f505b16 | 7679 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7680 | rt_se->parent = parent; |
6f505b16 PZ |
7681 | INIT_LIST_HEAD(&rt_se->run_list); |
7682 | } | |
7683 | #endif | |
7684 | ||
1da177e4 LT |
7685 | void __init sched_init(void) |
7686 | { | |
dd41f596 | 7687 | int i, j; |
434d53b0 MT |
7688 | unsigned long alloc_size = 0, ptr; |
7689 | ||
7690 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7691 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7692 | #endif | |
7693 | #ifdef CONFIG_RT_GROUP_SCHED | |
7694 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7695 | #endif |
df7c8e84 | 7696 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7697 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7698 | #endif |
434d53b0 | 7699 | if (alloc_size) { |
36b7b6d4 | 7700 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7701 | |
7702 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7703 | init_task_group.se = (struct sched_entity **)ptr; | |
7704 | ptr += nr_cpu_ids * sizeof(void **); | |
7705 | ||
7706 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7707 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7708 | |
6d6bc0ad | 7709 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7710 | #ifdef CONFIG_RT_GROUP_SCHED |
7711 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7712 | ptr += nr_cpu_ids * sizeof(void **); | |
7713 | ||
7714 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7715 | ptr += nr_cpu_ids * sizeof(void **); |
7716 | ||
6d6bc0ad | 7717 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7718 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7719 | for_each_possible_cpu(i) { | |
7720 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7721 | ptr += cpumask_size(); | |
7722 | } | |
7723 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7724 | } |
dd41f596 | 7725 | |
57d885fe GH |
7726 | #ifdef CONFIG_SMP |
7727 | init_defrootdomain(); | |
7728 | #endif | |
7729 | ||
d0b27fa7 PZ |
7730 | init_rt_bandwidth(&def_rt_bandwidth, |
7731 | global_rt_period(), global_rt_runtime()); | |
7732 | ||
7733 | #ifdef CONFIG_RT_GROUP_SCHED | |
7734 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7735 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7736 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7737 | |
7c941438 | 7738 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7739 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7740 | INIT_LIST_HEAD(&init_task_group.children); |
7741 | ||
7c941438 | 7742 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7743 | |
4a6cc4bd JK |
7744 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
7745 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
7746 | __alignof__(unsigned long)); | |
7747 | #endif | |
0a945022 | 7748 | for_each_possible_cpu(i) { |
70b97a7f | 7749 | struct rq *rq; |
1da177e4 LT |
7750 | |
7751 | rq = cpu_rq(i); | |
05fa785c | 7752 | raw_spin_lock_init(&rq->lock); |
7897986b | 7753 | rq->nr_running = 0; |
dce48a84 TG |
7754 | rq->calc_load_active = 0; |
7755 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7756 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7757 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7758 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7759 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7760 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7761 | #ifdef CONFIG_CGROUP_SCHED |
7762 | /* | |
7763 | * How much cpu bandwidth does init_task_group get? | |
7764 | * | |
7765 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7766 | * gets 100% of the cpu resources in the system. This overall | |
7767 | * system cpu resource is divided among the tasks of | |
7768 | * init_task_group and its child task-groups in a fair manner, | |
7769 | * based on each entity's (task or task-group's) weight | |
7770 | * (se->load.weight). | |
7771 | * | |
7772 | * In other words, if init_task_group has 10 tasks of weight | |
7773 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7774 | * then A0's share of the cpu resource is: | |
7775 | * | |
0d905bca | 7776 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7777 | * |
7778 | * We achieve this by letting init_task_group's tasks sit | |
7779 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7780 | */ | |
ec7dc8ac | 7781 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 7782 | #endif |
354d60c2 DG |
7783 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7784 | ||
7785 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7786 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7787 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7788 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 7789 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 7790 | #endif |
dd41f596 | 7791 | #endif |
1da177e4 | 7792 | |
dd41f596 IM |
7793 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7794 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7795 | |
7796 | rq->last_load_update_tick = jiffies; | |
7797 | ||
1da177e4 | 7798 | #ifdef CONFIG_SMP |
41c7ce9a | 7799 | rq->sd = NULL; |
57d885fe | 7800 | rq->rd = NULL; |
e51fd5e2 | 7801 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 7802 | rq->post_schedule = 0; |
1da177e4 | 7803 | rq->active_balance = 0; |
dd41f596 | 7804 | rq->next_balance = jiffies; |
1da177e4 | 7805 | rq->push_cpu = 0; |
0a2966b4 | 7806 | rq->cpu = i; |
1f11eb6a | 7807 | rq->online = 0; |
eae0c9df MG |
7808 | rq->idle_stamp = 0; |
7809 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7810 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
7811 | #ifdef CONFIG_NO_HZ |
7812 | rq->nohz_balance_kick = 0; | |
7813 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
7814 | #endif | |
1da177e4 | 7815 | #endif |
8f4d37ec | 7816 | init_rq_hrtick(rq); |
1da177e4 | 7817 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7818 | } |
7819 | ||
2dd73a4f | 7820 | set_load_weight(&init_task); |
b50f60ce | 7821 | |
e107be36 AK |
7822 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7823 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7824 | #endif | |
7825 | ||
c9819f45 | 7826 | #ifdef CONFIG_SMP |
962cf36c | 7827 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7828 | #endif |
7829 | ||
b50f60ce | 7830 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7831 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7832 | #endif |
7833 | ||
1da177e4 LT |
7834 | /* |
7835 | * The boot idle thread does lazy MMU switching as well: | |
7836 | */ | |
7837 | atomic_inc(&init_mm.mm_count); | |
7838 | enter_lazy_tlb(&init_mm, current); | |
7839 | ||
7840 | /* | |
7841 | * Make us the idle thread. Technically, schedule() should not be | |
7842 | * called from this thread, however somewhere below it might be, | |
7843 | * but because we are the idle thread, we just pick up running again | |
7844 | * when this runqueue becomes "idle". | |
7845 | */ | |
7846 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7847 | |
7848 | calc_load_update = jiffies + LOAD_FREQ; | |
7849 | ||
dd41f596 IM |
7850 | /* |
7851 | * During early bootup we pretend to be a normal task: | |
7852 | */ | |
7853 | current->sched_class = &fair_sched_class; | |
6892b75e | 7854 | |
6a7b3dc3 | 7855 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7856 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7857 | #ifdef CONFIG_SMP |
7d1e6a9b | 7858 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
7859 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
7860 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
7861 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
7862 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
7863 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 7864 | #endif |
bdddd296 RR |
7865 | /* May be allocated at isolcpus cmdline parse time */ |
7866 | if (cpu_isolated_map == NULL) | |
7867 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7868 | #endif /* SMP */ |
6a7b3dc3 | 7869 | |
cdd6c482 | 7870 | perf_event_init(); |
0d905bca | 7871 | |
6892b75e | 7872 | scheduler_running = 1; |
1da177e4 LT |
7873 | } |
7874 | ||
7875 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7876 | static inline int preempt_count_equals(int preempt_offset) |
7877 | { | |
234da7bc | 7878 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7879 | |
7880 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7881 | } | |
7882 | ||
d894837f | 7883 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7884 | { |
48f24c4d | 7885 | #ifdef in_atomic |
1da177e4 LT |
7886 | static unsigned long prev_jiffy; /* ratelimiting */ |
7887 | ||
e4aafea2 FW |
7888 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7889 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7890 | return; |
7891 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7892 | return; | |
7893 | prev_jiffy = jiffies; | |
7894 | ||
3df0fc5b PZ |
7895 | printk(KERN_ERR |
7896 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7897 | file, line); | |
7898 | printk(KERN_ERR | |
7899 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7900 | in_atomic(), irqs_disabled(), | |
7901 | current->pid, current->comm); | |
aef745fc IM |
7902 | |
7903 | debug_show_held_locks(current); | |
7904 | if (irqs_disabled()) | |
7905 | print_irqtrace_events(current); | |
7906 | dump_stack(); | |
1da177e4 LT |
7907 | #endif |
7908 | } | |
7909 | EXPORT_SYMBOL(__might_sleep); | |
7910 | #endif | |
7911 | ||
7912 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7913 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7914 | { | |
7915 | int on_rq; | |
3e51f33f | 7916 | |
3a5e4dc1 AK |
7917 | on_rq = p->se.on_rq; |
7918 | if (on_rq) | |
7919 | deactivate_task(rq, p, 0); | |
7920 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7921 | if (on_rq) { | |
7922 | activate_task(rq, p, 0); | |
7923 | resched_task(rq->curr); | |
7924 | } | |
7925 | } | |
7926 | ||
1da177e4 LT |
7927 | void normalize_rt_tasks(void) |
7928 | { | |
a0f98a1c | 7929 | struct task_struct *g, *p; |
1da177e4 | 7930 | unsigned long flags; |
70b97a7f | 7931 | struct rq *rq; |
1da177e4 | 7932 | |
4cf5d77a | 7933 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7934 | do_each_thread(g, p) { |
178be793 IM |
7935 | /* |
7936 | * Only normalize user tasks: | |
7937 | */ | |
7938 | if (!p->mm) | |
7939 | continue; | |
7940 | ||
6cfb0d5d | 7941 | p->se.exec_start = 0; |
6cfb0d5d | 7942 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7943 | p->se.statistics.wait_start = 0; |
7944 | p->se.statistics.sleep_start = 0; | |
7945 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7946 | #endif |
dd41f596 IM |
7947 | |
7948 | if (!rt_task(p)) { | |
7949 | /* | |
7950 | * Renice negative nice level userspace | |
7951 | * tasks back to 0: | |
7952 | */ | |
7953 | if (TASK_NICE(p) < 0 && p->mm) | |
7954 | set_user_nice(p, 0); | |
1da177e4 | 7955 | continue; |
dd41f596 | 7956 | } |
1da177e4 | 7957 | |
1d615482 | 7958 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7959 | rq = __task_rq_lock(p); |
1da177e4 | 7960 | |
178be793 | 7961 | normalize_task(rq, p); |
3a5e4dc1 | 7962 | |
b29739f9 | 7963 | __task_rq_unlock(rq); |
1d615482 | 7964 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7965 | } while_each_thread(g, p); |
7966 | ||
4cf5d77a | 7967 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7968 | } |
7969 | ||
7970 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7971 | |
67fc4e0c | 7972 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7973 | /* |
67fc4e0c | 7974 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7975 | * |
7976 | * They can only be called when the whole system has been | |
7977 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7978 | * activity can take place. Using them for anything else would | |
7979 | * be a serious bug, and as a result, they aren't even visible | |
7980 | * under any other configuration. | |
7981 | */ | |
7982 | ||
7983 | /** | |
7984 | * curr_task - return the current task for a given cpu. | |
7985 | * @cpu: the processor in question. | |
7986 | * | |
7987 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7988 | */ | |
36c8b586 | 7989 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
7990 | { |
7991 | return cpu_curr(cpu); | |
7992 | } | |
7993 | ||
67fc4e0c JW |
7994 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
7995 | ||
7996 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
7997 | /** |
7998 | * set_curr_task - set the current task for a given cpu. | |
7999 | * @cpu: the processor in question. | |
8000 | * @p: the task pointer to set. | |
8001 | * | |
8002 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8003 | * are serviced on a separate stack. It allows the architecture to switch the |
8004 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8005 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8006 | * and caller must save the original value of the current task (see | |
8007 | * curr_task() above) and restore that value before reenabling interrupts and | |
8008 | * re-starting the system. | |
8009 | * | |
8010 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8011 | */ | |
36c8b586 | 8012 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8013 | { |
8014 | cpu_curr(cpu) = p; | |
8015 | } | |
8016 | ||
8017 | #endif | |
29f59db3 | 8018 | |
bccbe08a PZ |
8019 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8020 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8021 | { |
8022 | int i; | |
8023 | ||
8024 | for_each_possible_cpu(i) { | |
8025 | if (tg->cfs_rq) | |
8026 | kfree(tg->cfs_rq[i]); | |
8027 | if (tg->se) | |
8028 | kfree(tg->se[i]); | |
6f505b16 PZ |
8029 | } |
8030 | ||
8031 | kfree(tg->cfs_rq); | |
8032 | kfree(tg->se); | |
6f505b16 PZ |
8033 | } |
8034 | ||
ec7dc8ac DG |
8035 | static |
8036 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8037 | { |
29f59db3 | 8038 | struct cfs_rq *cfs_rq; |
eab17229 | 8039 | struct sched_entity *se; |
9b5b7751 | 8040 | struct rq *rq; |
29f59db3 SV |
8041 | int i; |
8042 | ||
434d53b0 | 8043 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8044 | if (!tg->cfs_rq) |
8045 | goto err; | |
434d53b0 | 8046 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8047 | if (!tg->se) |
8048 | goto err; | |
052f1dc7 PZ |
8049 | |
8050 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8051 | |
8052 | for_each_possible_cpu(i) { | |
9b5b7751 | 8053 | rq = cpu_rq(i); |
29f59db3 | 8054 | |
eab17229 LZ |
8055 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8056 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8057 | if (!cfs_rq) |
8058 | goto err; | |
8059 | ||
eab17229 LZ |
8060 | se = kzalloc_node(sizeof(struct sched_entity), |
8061 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8062 | if (!se) |
dfc12eb2 | 8063 | goto err_free_rq; |
29f59db3 | 8064 | |
eab17229 | 8065 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
8066 | } |
8067 | ||
8068 | return 1; | |
8069 | ||
dfc12eb2 PC |
8070 | err_free_rq: |
8071 | kfree(cfs_rq); | |
bccbe08a PZ |
8072 | err: |
8073 | return 0; | |
8074 | } | |
8075 | ||
8076 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8077 | { | |
8078 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8079 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8080 | } | |
8081 | ||
8082 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8083 | { | |
8084 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8085 | } | |
6d6bc0ad | 8086 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8087 | static inline void free_fair_sched_group(struct task_group *tg) |
8088 | { | |
8089 | } | |
8090 | ||
ec7dc8ac DG |
8091 | static inline |
8092 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8093 | { |
8094 | return 1; | |
8095 | } | |
8096 | ||
8097 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8098 | { | |
8099 | } | |
8100 | ||
8101 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8102 | { | |
8103 | } | |
6d6bc0ad | 8104 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8105 | |
8106 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8107 | static void free_rt_sched_group(struct task_group *tg) |
8108 | { | |
8109 | int i; | |
8110 | ||
d0b27fa7 PZ |
8111 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8112 | ||
bccbe08a PZ |
8113 | for_each_possible_cpu(i) { |
8114 | if (tg->rt_rq) | |
8115 | kfree(tg->rt_rq[i]); | |
8116 | if (tg->rt_se) | |
8117 | kfree(tg->rt_se[i]); | |
8118 | } | |
8119 | ||
8120 | kfree(tg->rt_rq); | |
8121 | kfree(tg->rt_se); | |
8122 | } | |
8123 | ||
ec7dc8ac DG |
8124 | static |
8125 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8126 | { |
8127 | struct rt_rq *rt_rq; | |
eab17229 | 8128 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8129 | struct rq *rq; |
8130 | int i; | |
8131 | ||
434d53b0 | 8132 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8133 | if (!tg->rt_rq) |
8134 | goto err; | |
434d53b0 | 8135 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8136 | if (!tg->rt_se) |
8137 | goto err; | |
8138 | ||
d0b27fa7 PZ |
8139 | init_rt_bandwidth(&tg->rt_bandwidth, |
8140 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8141 | |
8142 | for_each_possible_cpu(i) { | |
8143 | rq = cpu_rq(i); | |
8144 | ||
eab17229 LZ |
8145 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8146 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8147 | if (!rt_rq) |
8148 | goto err; | |
29f59db3 | 8149 | |
eab17229 LZ |
8150 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8151 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8152 | if (!rt_se) |
dfc12eb2 | 8153 | goto err_free_rq; |
29f59db3 | 8154 | |
eab17229 | 8155 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8156 | } |
8157 | ||
bccbe08a PZ |
8158 | return 1; |
8159 | ||
dfc12eb2 PC |
8160 | err_free_rq: |
8161 | kfree(rt_rq); | |
bccbe08a PZ |
8162 | err: |
8163 | return 0; | |
8164 | } | |
8165 | ||
8166 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8167 | { | |
8168 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8169 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8170 | } | |
8171 | ||
8172 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8173 | { | |
8174 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8175 | } | |
6d6bc0ad | 8176 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8177 | static inline void free_rt_sched_group(struct task_group *tg) |
8178 | { | |
8179 | } | |
8180 | ||
ec7dc8ac DG |
8181 | static inline |
8182 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8183 | { |
8184 | return 1; | |
8185 | } | |
8186 | ||
8187 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8188 | { | |
8189 | } | |
8190 | ||
8191 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8192 | { | |
8193 | } | |
6d6bc0ad | 8194 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8195 | |
7c941438 | 8196 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8197 | static void free_sched_group(struct task_group *tg) |
8198 | { | |
8199 | free_fair_sched_group(tg); | |
8200 | free_rt_sched_group(tg); | |
8201 | kfree(tg); | |
8202 | } | |
8203 | ||
8204 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8205 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8206 | { |
8207 | struct task_group *tg; | |
8208 | unsigned long flags; | |
8209 | int i; | |
8210 | ||
8211 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8212 | if (!tg) | |
8213 | return ERR_PTR(-ENOMEM); | |
8214 | ||
ec7dc8ac | 8215 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8216 | goto err; |
8217 | ||
ec7dc8ac | 8218 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8219 | goto err; |
8220 | ||
8ed36996 | 8221 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8222 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8223 | register_fair_sched_group(tg, i); |
8224 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8225 | } |
6f505b16 | 8226 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8227 | |
8228 | WARN_ON(!parent); /* root should already exist */ | |
8229 | ||
8230 | tg->parent = parent; | |
f473aa5e | 8231 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8232 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8233 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8234 | |
9b5b7751 | 8235 | return tg; |
29f59db3 SV |
8236 | |
8237 | err: | |
6f505b16 | 8238 | free_sched_group(tg); |
29f59db3 SV |
8239 | return ERR_PTR(-ENOMEM); |
8240 | } | |
8241 | ||
9b5b7751 | 8242 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8243 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8244 | { |
29f59db3 | 8245 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8246 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8247 | } |
8248 | ||
9b5b7751 | 8249 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8250 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8251 | { |
8ed36996 | 8252 | unsigned long flags; |
9b5b7751 | 8253 | int i; |
29f59db3 | 8254 | |
8ed36996 | 8255 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8256 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8257 | unregister_fair_sched_group(tg, i); |
8258 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8259 | } |
6f505b16 | 8260 | list_del_rcu(&tg->list); |
f473aa5e | 8261 | list_del_rcu(&tg->siblings); |
8ed36996 | 8262 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8263 | |
9b5b7751 | 8264 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8265 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8266 | } |
8267 | ||
9b5b7751 | 8268 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8269 | * The caller of this function should have put the task in its new group |
8270 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8271 | * reflect its new group. | |
9b5b7751 SV |
8272 | */ |
8273 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8274 | { |
8275 | int on_rq, running; | |
8276 | unsigned long flags; | |
8277 | struct rq *rq; | |
8278 | ||
8279 | rq = task_rq_lock(tsk, &flags); | |
8280 | ||
051a1d1a | 8281 | running = task_current(rq, tsk); |
29f59db3 SV |
8282 | on_rq = tsk->se.on_rq; |
8283 | ||
0e1f3483 | 8284 | if (on_rq) |
29f59db3 | 8285 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8286 | if (unlikely(running)) |
8287 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8288 | |
6f505b16 | 8289 | set_task_rq(tsk, task_cpu(tsk)); |
29f59db3 | 8290 | |
810b3817 PZ |
8291 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8292 | if (tsk->sched_class->moved_group) | |
88ec22d3 | 8293 | tsk->sched_class->moved_group(tsk, on_rq); |
810b3817 PZ |
8294 | #endif |
8295 | ||
0e1f3483 HS |
8296 | if (unlikely(running)) |
8297 | tsk->sched_class->set_curr_task(rq); | |
8298 | if (on_rq) | |
371fd7e7 | 8299 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8300 | |
29f59db3 SV |
8301 | task_rq_unlock(rq, &flags); |
8302 | } | |
7c941438 | 8303 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8304 | |
052f1dc7 | 8305 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8306 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8307 | { |
8308 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8309 | int on_rq; |
8310 | ||
29f59db3 | 8311 | on_rq = se->on_rq; |
62fb1851 | 8312 | if (on_rq) |
29f59db3 SV |
8313 | dequeue_entity(cfs_rq, se, 0); |
8314 | ||
8315 | se->load.weight = shares; | |
e05510d0 | 8316 | se->load.inv_weight = 0; |
29f59db3 | 8317 | |
62fb1851 | 8318 | if (on_rq) |
29f59db3 | 8319 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8320 | } |
62fb1851 | 8321 | |
c09595f6 PZ |
8322 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8323 | { | |
8324 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8325 | struct rq *rq = cfs_rq->rq; | |
8326 | unsigned long flags; | |
8327 | ||
05fa785c | 8328 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8329 | __set_se_shares(se, shares); |
05fa785c | 8330 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8331 | } |
8332 | ||
8ed36996 PZ |
8333 | static DEFINE_MUTEX(shares_mutex); |
8334 | ||
4cf86d77 | 8335 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8336 | { |
8337 | int i; | |
8ed36996 | 8338 | unsigned long flags; |
c61935fd | 8339 | |
ec7dc8ac DG |
8340 | /* |
8341 | * We can't change the weight of the root cgroup. | |
8342 | */ | |
8343 | if (!tg->se[0]) | |
8344 | return -EINVAL; | |
8345 | ||
18d95a28 PZ |
8346 | if (shares < MIN_SHARES) |
8347 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8348 | else if (shares > MAX_SHARES) |
8349 | shares = MAX_SHARES; | |
62fb1851 | 8350 | |
8ed36996 | 8351 | mutex_lock(&shares_mutex); |
9b5b7751 | 8352 | if (tg->shares == shares) |
5cb350ba | 8353 | goto done; |
29f59db3 | 8354 | |
8ed36996 | 8355 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8356 | for_each_possible_cpu(i) |
8357 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8358 | list_del_rcu(&tg->siblings); |
8ed36996 | 8359 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8360 | |
8361 | /* wait for any ongoing reference to this group to finish */ | |
8362 | synchronize_sched(); | |
8363 | ||
8364 | /* | |
8365 | * Now we are free to modify the group's share on each cpu | |
8366 | * w/o tripping rebalance_share or load_balance_fair. | |
8367 | */ | |
9b5b7751 | 8368 | tg->shares = shares; |
c09595f6 PZ |
8369 | for_each_possible_cpu(i) { |
8370 | /* | |
8371 | * force a rebalance | |
8372 | */ | |
8373 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8374 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8375 | } |
29f59db3 | 8376 | |
6b2d7700 SV |
8377 | /* |
8378 | * Enable load balance activity on this group, by inserting it back on | |
8379 | * each cpu's rq->leaf_cfs_rq_list. | |
8380 | */ | |
8ed36996 | 8381 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8382 | for_each_possible_cpu(i) |
8383 | register_fair_sched_group(tg, i); | |
f473aa5e | 8384 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8385 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8386 | done: |
8ed36996 | 8387 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8388 | return 0; |
29f59db3 SV |
8389 | } |
8390 | ||
5cb350ba DG |
8391 | unsigned long sched_group_shares(struct task_group *tg) |
8392 | { | |
8393 | return tg->shares; | |
8394 | } | |
052f1dc7 | 8395 | #endif |
5cb350ba | 8396 | |
052f1dc7 | 8397 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8398 | /* |
9f0c1e56 | 8399 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8400 | */ |
9f0c1e56 PZ |
8401 | static DEFINE_MUTEX(rt_constraints_mutex); |
8402 | ||
8403 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8404 | { | |
8405 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8406 | return 1ULL << 20; |
9f0c1e56 | 8407 | |
9a7e0b18 | 8408 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8409 | } |
8410 | ||
9a7e0b18 PZ |
8411 | /* Must be called with tasklist_lock held */ |
8412 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8413 | { |
9a7e0b18 | 8414 | struct task_struct *g, *p; |
b40b2e8e | 8415 | |
9a7e0b18 PZ |
8416 | do_each_thread(g, p) { |
8417 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8418 | return 1; | |
8419 | } while_each_thread(g, p); | |
b40b2e8e | 8420 | |
9a7e0b18 PZ |
8421 | return 0; |
8422 | } | |
b40b2e8e | 8423 | |
9a7e0b18 PZ |
8424 | struct rt_schedulable_data { |
8425 | struct task_group *tg; | |
8426 | u64 rt_period; | |
8427 | u64 rt_runtime; | |
8428 | }; | |
b40b2e8e | 8429 | |
9a7e0b18 PZ |
8430 | static int tg_schedulable(struct task_group *tg, void *data) |
8431 | { | |
8432 | struct rt_schedulable_data *d = data; | |
8433 | struct task_group *child; | |
8434 | unsigned long total, sum = 0; | |
8435 | u64 period, runtime; | |
b40b2e8e | 8436 | |
9a7e0b18 PZ |
8437 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8438 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8439 | |
9a7e0b18 PZ |
8440 | if (tg == d->tg) { |
8441 | period = d->rt_period; | |
8442 | runtime = d->rt_runtime; | |
b40b2e8e | 8443 | } |
b40b2e8e | 8444 | |
4653f803 PZ |
8445 | /* |
8446 | * Cannot have more runtime than the period. | |
8447 | */ | |
8448 | if (runtime > period && runtime != RUNTIME_INF) | |
8449 | return -EINVAL; | |
6f505b16 | 8450 | |
4653f803 PZ |
8451 | /* |
8452 | * Ensure we don't starve existing RT tasks. | |
8453 | */ | |
9a7e0b18 PZ |
8454 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8455 | return -EBUSY; | |
6f505b16 | 8456 | |
9a7e0b18 | 8457 | total = to_ratio(period, runtime); |
6f505b16 | 8458 | |
4653f803 PZ |
8459 | /* |
8460 | * Nobody can have more than the global setting allows. | |
8461 | */ | |
8462 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8463 | return -EINVAL; | |
6f505b16 | 8464 | |
4653f803 PZ |
8465 | /* |
8466 | * The sum of our children's runtime should not exceed our own. | |
8467 | */ | |
9a7e0b18 PZ |
8468 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8469 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8470 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8471 | |
9a7e0b18 PZ |
8472 | if (child == d->tg) { |
8473 | period = d->rt_period; | |
8474 | runtime = d->rt_runtime; | |
8475 | } | |
6f505b16 | 8476 | |
9a7e0b18 | 8477 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8478 | } |
6f505b16 | 8479 | |
9a7e0b18 PZ |
8480 | if (sum > total) |
8481 | return -EINVAL; | |
8482 | ||
8483 | return 0; | |
6f505b16 PZ |
8484 | } |
8485 | ||
9a7e0b18 | 8486 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8487 | { |
9a7e0b18 PZ |
8488 | struct rt_schedulable_data data = { |
8489 | .tg = tg, | |
8490 | .rt_period = period, | |
8491 | .rt_runtime = runtime, | |
8492 | }; | |
8493 | ||
8494 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8495 | } |
8496 | ||
d0b27fa7 PZ |
8497 | static int tg_set_bandwidth(struct task_group *tg, |
8498 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8499 | { |
ac086bc2 | 8500 | int i, err = 0; |
9f0c1e56 | 8501 | |
9f0c1e56 | 8502 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8503 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8504 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8505 | if (err) | |
9f0c1e56 | 8506 | goto unlock; |
ac086bc2 | 8507 | |
0986b11b | 8508 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8509 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8510 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8511 | |
8512 | for_each_possible_cpu(i) { | |
8513 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8514 | ||
0986b11b | 8515 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8516 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8517 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8518 | } |
0986b11b | 8519 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
9f0c1e56 | 8520 | unlock: |
521f1a24 | 8521 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8522 | mutex_unlock(&rt_constraints_mutex); |
8523 | ||
8524 | return err; | |
6f505b16 PZ |
8525 | } |
8526 | ||
d0b27fa7 PZ |
8527 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8528 | { | |
8529 | u64 rt_runtime, rt_period; | |
8530 | ||
8531 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8532 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8533 | if (rt_runtime_us < 0) | |
8534 | rt_runtime = RUNTIME_INF; | |
8535 | ||
8536 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8537 | } | |
8538 | ||
9f0c1e56 PZ |
8539 | long sched_group_rt_runtime(struct task_group *tg) |
8540 | { | |
8541 | u64 rt_runtime_us; | |
8542 | ||
d0b27fa7 | 8543 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8544 | return -1; |
8545 | ||
d0b27fa7 | 8546 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8547 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8548 | return rt_runtime_us; | |
8549 | } | |
d0b27fa7 PZ |
8550 | |
8551 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8552 | { | |
8553 | u64 rt_runtime, rt_period; | |
8554 | ||
8555 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8556 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8557 | ||
619b0488 R |
8558 | if (rt_period == 0) |
8559 | return -EINVAL; | |
8560 | ||
d0b27fa7 PZ |
8561 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8562 | } | |
8563 | ||
8564 | long sched_group_rt_period(struct task_group *tg) | |
8565 | { | |
8566 | u64 rt_period_us; | |
8567 | ||
8568 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8569 | do_div(rt_period_us, NSEC_PER_USEC); | |
8570 | return rt_period_us; | |
8571 | } | |
8572 | ||
8573 | static int sched_rt_global_constraints(void) | |
8574 | { | |
4653f803 | 8575 | u64 runtime, period; |
d0b27fa7 PZ |
8576 | int ret = 0; |
8577 | ||
ec5d4989 HS |
8578 | if (sysctl_sched_rt_period <= 0) |
8579 | return -EINVAL; | |
8580 | ||
4653f803 PZ |
8581 | runtime = global_rt_runtime(); |
8582 | period = global_rt_period(); | |
8583 | ||
8584 | /* | |
8585 | * Sanity check on the sysctl variables. | |
8586 | */ | |
8587 | if (runtime > period && runtime != RUNTIME_INF) | |
8588 | return -EINVAL; | |
10b612f4 | 8589 | |
d0b27fa7 | 8590 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8591 | read_lock(&tasklist_lock); |
4653f803 | 8592 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8593 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8594 | mutex_unlock(&rt_constraints_mutex); |
8595 | ||
8596 | return ret; | |
8597 | } | |
54e99124 DG |
8598 | |
8599 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8600 | { | |
8601 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8602 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8603 | return 0; | |
8604 | ||
8605 | return 1; | |
8606 | } | |
8607 | ||
6d6bc0ad | 8608 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8609 | static int sched_rt_global_constraints(void) |
8610 | { | |
ac086bc2 PZ |
8611 | unsigned long flags; |
8612 | int i; | |
8613 | ||
ec5d4989 HS |
8614 | if (sysctl_sched_rt_period <= 0) |
8615 | return -EINVAL; | |
8616 | ||
60aa605d PZ |
8617 | /* |
8618 | * There's always some RT tasks in the root group | |
8619 | * -- migration, kstopmachine etc.. | |
8620 | */ | |
8621 | if (sysctl_sched_rt_runtime == 0) | |
8622 | return -EBUSY; | |
8623 | ||
0986b11b | 8624 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8625 | for_each_possible_cpu(i) { |
8626 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8627 | ||
0986b11b | 8628 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8629 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8630 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8631 | } |
0986b11b | 8632 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8633 | |
d0b27fa7 PZ |
8634 | return 0; |
8635 | } | |
6d6bc0ad | 8636 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8637 | |
8638 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8639 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8640 | loff_t *ppos) |
8641 | { | |
8642 | int ret; | |
8643 | int old_period, old_runtime; | |
8644 | static DEFINE_MUTEX(mutex); | |
8645 | ||
8646 | mutex_lock(&mutex); | |
8647 | old_period = sysctl_sched_rt_period; | |
8648 | old_runtime = sysctl_sched_rt_runtime; | |
8649 | ||
8d65af78 | 8650 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8651 | |
8652 | if (!ret && write) { | |
8653 | ret = sched_rt_global_constraints(); | |
8654 | if (ret) { | |
8655 | sysctl_sched_rt_period = old_period; | |
8656 | sysctl_sched_rt_runtime = old_runtime; | |
8657 | } else { | |
8658 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8659 | def_rt_bandwidth.rt_period = | |
8660 | ns_to_ktime(global_rt_period()); | |
8661 | } | |
8662 | } | |
8663 | mutex_unlock(&mutex); | |
8664 | ||
8665 | return ret; | |
8666 | } | |
68318b8e | 8667 | |
052f1dc7 | 8668 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8669 | |
8670 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8671 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8672 | { |
2b01dfe3 PM |
8673 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8674 | struct task_group, css); | |
68318b8e SV |
8675 | } |
8676 | ||
8677 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8678 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8679 | { |
ec7dc8ac | 8680 | struct task_group *tg, *parent; |
68318b8e | 8681 | |
2b01dfe3 | 8682 | if (!cgrp->parent) { |
68318b8e | 8683 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8684 | return &init_task_group.css; |
8685 | } | |
8686 | ||
ec7dc8ac DG |
8687 | parent = cgroup_tg(cgrp->parent); |
8688 | tg = sched_create_group(parent); | |
68318b8e SV |
8689 | if (IS_ERR(tg)) |
8690 | return ERR_PTR(-ENOMEM); | |
8691 | ||
68318b8e SV |
8692 | return &tg->css; |
8693 | } | |
8694 | ||
41a2d6cf IM |
8695 | static void |
8696 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8697 | { |
2b01dfe3 | 8698 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8699 | |
8700 | sched_destroy_group(tg); | |
8701 | } | |
8702 | ||
41a2d6cf | 8703 | static int |
be367d09 | 8704 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8705 | { |
b68aa230 | 8706 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8707 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8708 | return -EINVAL; |
8709 | #else | |
68318b8e SV |
8710 | /* We don't support RT-tasks being in separate groups */ |
8711 | if (tsk->sched_class != &fair_sched_class) | |
8712 | return -EINVAL; | |
b68aa230 | 8713 | #endif |
be367d09 BB |
8714 | return 0; |
8715 | } | |
68318b8e | 8716 | |
be367d09 BB |
8717 | static int |
8718 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8719 | struct task_struct *tsk, bool threadgroup) | |
8720 | { | |
8721 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8722 | if (retval) | |
8723 | return retval; | |
8724 | if (threadgroup) { | |
8725 | struct task_struct *c; | |
8726 | rcu_read_lock(); | |
8727 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8728 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8729 | if (retval) { | |
8730 | rcu_read_unlock(); | |
8731 | return retval; | |
8732 | } | |
8733 | } | |
8734 | rcu_read_unlock(); | |
8735 | } | |
68318b8e SV |
8736 | return 0; |
8737 | } | |
8738 | ||
8739 | static void | |
2b01dfe3 | 8740 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8741 | struct cgroup *old_cont, struct task_struct *tsk, |
8742 | bool threadgroup) | |
68318b8e SV |
8743 | { |
8744 | sched_move_task(tsk); | |
be367d09 BB |
8745 | if (threadgroup) { |
8746 | struct task_struct *c; | |
8747 | rcu_read_lock(); | |
8748 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8749 | sched_move_task(c); | |
8750 | } | |
8751 | rcu_read_unlock(); | |
8752 | } | |
68318b8e SV |
8753 | } |
8754 | ||
052f1dc7 | 8755 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8756 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8757 | u64 shareval) |
68318b8e | 8758 | { |
2b01dfe3 | 8759 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8760 | } |
8761 | ||
f4c753b7 | 8762 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8763 | { |
2b01dfe3 | 8764 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8765 | |
8766 | return (u64) tg->shares; | |
8767 | } | |
6d6bc0ad | 8768 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8769 | |
052f1dc7 | 8770 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8771 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8772 | s64 val) |
6f505b16 | 8773 | { |
06ecb27c | 8774 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8775 | } |
8776 | ||
06ecb27c | 8777 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8778 | { |
06ecb27c | 8779 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8780 | } |
d0b27fa7 PZ |
8781 | |
8782 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8783 | u64 rt_period_us) | |
8784 | { | |
8785 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8786 | } | |
8787 | ||
8788 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8789 | { | |
8790 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8791 | } | |
6d6bc0ad | 8792 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8793 | |
fe5c7cc2 | 8794 | static struct cftype cpu_files[] = { |
052f1dc7 | 8795 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8796 | { |
8797 | .name = "shares", | |
f4c753b7 PM |
8798 | .read_u64 = cpu_shares_read_u64, |
8799 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8800 | }, |
052f1dc7 PZ |
8801 | #endif |
8802 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8803 | { |
9f0c1e56 | 8804 | .name = "rt_runtime_us", |
06ecb27c PM |
8805 | .read_s64 = cpu_rt_runtime_read, |
8806 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8807 | }, |
d0b27fa7 PZ |
8808 | { |
8809 | .name = "rt_period_us", | |
f4c753b7 PM |
8810 | .read_u64 = cpu_rt_period_read_uint, |
8811 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8812 | }, |
052f1dc7 | 8813 | #endif |
68318b8e SV |
8814 | }; |
8815 | ||
8816 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8817 | { | |
fe5c7cc2 | 8818 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8819 | } |
8820 | ||
8821 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8822 | .name = "cpu", |
8823 | .create = cpu_cgroup_create, | |
8824 | .destroy = cpu_cgroup_destroy, | |
8825 | .can_attach = cpu_cgroup_can_attach, | |
8826 | .attach = cpu_cgroup_attach, | |
8827 | .populate = cpu_cgroup_populate, | |
8828 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8829 | .early_init = 1, |
8830 | }; | |
8831 | ||
052f1dc7 | 8832 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8833 | |
8834 | #ifdef CONFIG_CGROUP_CPUACCT | |
8835 | ||
8836 | /* | |
8837 | * CPU accounting code for task groups. | |
8838 | * | |
8839 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8840 | * (balbir@in.ibm.com). | |
8841 | */ | |
8842 | ||
934352f2 | 8843 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8844 | struct cpuacct { |
8845 | struct cgroup_subsys_state css; | |
8846 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8847 | u64 __percpu *cpuusage; |
ef12fefa | 8848 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8849 | struct cpuacct *parent; |
d842de87 SV |
8850 | }; |
8851 | ||
8852 | struct cgroup_subsys cpuacct_subsys; | |
8853 | ||
8854 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8855 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8856 | { |
32cd756a | 8857 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8858 | struct cpuacct, css); |
8859 | } | |
8860 | ||
8861 | /* return cpu accounting group to which this task belongs */ | |
8862 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8863 | { | |
8864 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8865 | struct cpuacct, css); | |
8866 | } | |
8867 | ||
8868 | /* create a new cpu accounting group */ | |
8869 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8870 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8871 | { |
8872 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8873 | int i; |
d842de87 SV |
8874 | |
8875 | if (!ca) | |
ef12fefa | 8876 | goto out; |
d842de87 SV |
8877 | |
8878 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8879 | if (!ca->cpuusage) |
8880 | goto out_free_ca; | |
8881 | ||
8882 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8883 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8884 | goto out_free_counters; | |
d842de87 | 8885 | |
934352f2 BR |
8886 | if (cgrp->parent) |
8887 | ca->parent = cgroup_ca(cgrp->parent); | |
8888 | ||
d842de87 | 8889 | return &ca->css; |
ef12fefa BR |
8890 | |
8891 | out_free_counters: | |
8892 | while (--i >= 0) | |
8893 | percpu_counter_destroy(&ca->cpustat[i]); | |
8894 | free_percpu(ca->cpuusage); | |
8895 | out_free_ca: | |
8896 | kfree(ca); | |
8897 | out: | |
8898 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8899 | } |
8900 | ||
8901 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8902 | static void |
32cd756a | 8903 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8904 | { |
32cd756a | 8905 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8906 | int i; |
d842de87 | 8907 | |
ef12fefa BR |
8908 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8909 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8910 | free_percpu(ca->cpuusage); |
8911 | kfree(ca); | |
8912 | } | |
8913 | ||
720f5498 KC |
8914 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8915 | { | |
b36128c8 | 8916 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8917 | u64 data; |
8918 | ||
8919 | #ifndef CONFIG_64BIT | |
8920 | /* | |
8921 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8922 | */ | |
05fa785c | 8923 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8924 | data = *cpuusage; |
05fa785c | 8925 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8926 | #else |
8927 | data = *cpuusage; | |
8928 | #endif | |
8929 | ||
8930 | return data; | |
8931 | } | |
8932 | ||
8933 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8934 | { | |
b36128c8 | 8935 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8936 | |
8937 | #ifndef CONFIG_64BIT | |
8938 | /* | |
8939 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8940 | */ | |
05fa785c | 8941 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8942 | *cpuusage = val; |
05fa785c | 8943 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8944 | #else |
8945 | *cpuusage = val; | |
8946 | #endif | |
8947 | } | |
8948 | ||
d842de87 | 8949 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8950 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8951 | { |
32cd756a | 8952 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8953 | u64 totalcpuusage = 0; |
8954 | int i; | |
8955 | ||
720f5498 KC |
8956 | for_each_present_cpu(i) |
8957 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8958 | |
8959 | return totalcpuusage; | |
8960 | } | |
8961 | ||
0297b803 DG |
8962 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8963 | u64 reset) | |
8964 | { | |
8965 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8966 | int err = 0; | |
8967 | int i; | |
8968 | ||
8969 | if (reset) { | |
8970 | err = -EINVAL; | |
8971 | goto out; | |
8972 | } | |
8973 | ||
720f5498 KC |
8974 | for_each_present_cpu(i) |
8975 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8976 | |
0297b803 DG |
8977 | out: |
8978 | return err; | |
8979 | } | |
8980 | ||
e9515c3c KC |
8981 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8982 | struct seq_file *m) | |
8983 | { | |
8984 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8985 | u64 percpu; | |
8986 | int i; | |
8987 | ||
8988 | for_each_present_cpu(i) { | |
8989 | percpu = cpuacct_cpuusage_read(ca, i); | |
8990 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8991 | } | |
8992 | seq_printf(m, "\n"); | |
8993 | return 0; | |
8994 | } | |
8995 | ||
ef12fefa BR |
8996 | static const char *cpuacct_stat_desc[] = { |
8997 | [CPUACCT_STAT_USER] = "user", | |
8998 | [CPUACCT_STAT_SYSTEM] = "system", | |
8999 | }; | |
9000 | ||
9001 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9002 | struct cgroup_map_cb *cb) | |
9003 | { | |
9004 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9005 | int i; | |
9006 | ||
9007 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9008 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9009 | val = cputime64_to_clock_t(val); | |
9010 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9011 | } | |
9012 | return 0; | |
9013 | } | |
9014 | ||
d842de87 SV |
9015 | static struct cftype files[] = { |
9016 | { | |
9017 | .name = "usage", | |
f4c753b7 PM |
9018 | .read_u64 = cpuusage_read, |
9019 | .write_u64 = cpuusage_write, | |
d842de87 | 9020 | }, |
e9515c3c KC |
9021 | { |
9022 | .name = "usage_percpu", | |
9023 | .read_seq_string = cpuacct_percpu_seq_read, | |
9024 | }, | |
ef12fefa BR |
9025 | { |
9026 | .name = "stat", | |
9027 | .read_map = cpuacct_stats_show, | |
9028 | }, | |
d842de87 SV |
9029 | }; |
9030 | ||
32cd756a | 9031 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9032 | { |
32cd756a | 9033 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9034 | } |
9035 | ||
9036 | /* | |
9037 | * charge this task's execution time to its accounting group. | |
9038 | * | |
9039 | * called with rq->lock held. | |
9040 | */ | |
9041 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9042 | { | |
9043 | struct cpuacct *ca; | |
934352f2 | 9044 | int cpu; |
d842de87 | 9045 | |
c40c6f85 | 9046 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9047 | return; |
9048 | ||
934352f2 | 9049 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9050 | |
9051 | rcu_read_lock(); | |
9052 | ||
d842de87 | 9053 | ca = task_ca(tsk); |
d842de87 | 9054 | |
934352f2 | 9055 | for (; ca; ca = ca->parent) { |
b36128c8 | 9056 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9057 | *cpuusage += cputime; |
9058 | } | |
a18b83b7 BR |
9059 | |
9060 | rcu_read_unlock(); | |
d842de87 SV |
9061 | } |
9062 | ||
fa535a77 AB |
9063 | /* |
9064 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9065 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9066 | * percpu_counter_add with values large enough to always overflow the | |
9067 | * per cpu batch limit causing bad SMP scalability. | |
9068 | * | |
9069 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9070 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9071 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9072 | */ | |
9073 | #ifdef CONFIG_SMP | |
9074 | #define CPUACCT_BATCH \ | |
9075 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9076 | #else | |
9077 | #define CPUACCT_BATCH 0 | |
9078 | #endif | |
9079 | ||
ef12fefa BR |
9080 | /* |
9081 | * Charge the system/user time to the task's accounting group. | |
9082 | */ | |
9083 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9084 | enum cpuacct_stat_index idx, cputime_t val) | |
9085 | { | |
9086 | struct cpuacct *ca; | |
fa535a77 | 9087 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9088 | |
9089 | if (unlikely(!cpuacct_subsys.active)) | |
9090 | return; | |
9091 | ||
9092 | rcu_read_lock(); | |
9093 | ca = task_ca(tsk); | |
9094 | ||
9095 | do { | |
fa535a77 | 9096 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9097 | ca = ca->parent; |
9098 | } while (ca); | |
9099 | rcu_read_unlock(); | |
9100 | } | |
9101 | ||
d842de87 SV |
9102 | struct cgroup_subsys cpuacct_subsys = { |
9103 | .name = "cpuacct", | |
9104 | .create = cpuacct_create, | |
9105 | .destroy = cpuacct_destroy, | |
9106 | .populate = cpuacct_populate, | |
9107 | .subsys_id = cpuacct_subsys_id, | |
9108 | }; | |
9109 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
9110 | |
9111 | #ifndef CONFIG_SMP | |
9112 | ||
03b042bf PM |
9113 | void synchronize_sched_expedited(void) |
9114 | { | |
fc390cde | 9115 | barrier(); |
03b042bf PM |
9116 | } |
9117 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9118 | ||
9119 | #else /* #ifndef CONFIG_SMP */ | |
9120 | ||
cc631fb7 | 9121 | static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0); |
03b042bf | 9122 | |
cc631fb7 | 9123 | static int synchronize_sched_expedited_cpu_stop(void *data) |
03b042bf | 9124 | { |
969c7921 TH |
9125 | /* |
9126 | * There must be a full memory barrier on each affected CPU | |
9127 | * between the time that try_stop_cpus() is called and the | |
9128 | * time that it returns. | |
9129 | * | |
9130 | * In the current initial implementation of cpu_stop, the | |
9131 | * above condition is already met when the control reaches | |
9132 | * this point and the following smp_mb() is not strictly | |
9133 | * necessary. Do smp_mb() anyway for documentation and | |
9134 | * robustness against future implementation changes. | |
9135 | */ | |
cc631fb7 | 9136 | smp_mb(); /* See above comment block. */ |
969c7921 | 9137 | return 0; |
03b042bf | 9138 | } |
03b042bf PM |
9139 | |
9140 | /* | |
9141 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9142 | * approach to force grace period to end quickly. This consumes | |
9143 | * significant time on all CPUs, and is thus not recommended for | |
9144 | * any sort of common-case code. | |
9145 | * | |
9146 | * Note that it is illegal to call this function while holding any | |
9147 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9148 | * observe this restriction will result in deadlock. | |
9149 | */ | |
9150 | void synchronize_sched_expedited(void) | |
9151 | { | |
969c7921 | 9152 | int snap, trycount = 0; |
03b042bf PM |
9153 | |
9154 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
969c7921 | 9155 | snap = atomic_read(&synchronize_sched_expedited_count) + 1; |
03b042bf | 9156 | get_online_cpus(); |
969c7921 TH |
9157 | while (try_stop_cpus(cpu_online_mask, |
9158 | synchronize_sched_expedited_cpu_stop, | |
94458d5e | 9159 | NULL) == -EAGAIN) { |
03b042bf PM |
9160 | put_online_cpus(); |
9161 | if (trycount++ < 10) | |
9162 | udelay(trycount * num_online_cpus()); | |
9163 | else { | |
9164 | synchronize_sched(); | |
9165 | return; | |
9166 | } | |
969c7921 | 9167 | if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) { |
03b042bf PM |
9168 | smp_mb(); /* ensure test happens before caller kfree */ |
9169 | return; | |
9170 | } | |
9171 | get_online_cpus(); | |
9172 | } | |
969c7921 | 9173 | atomic_inc(&synchronize_sched_expedited_count); |
cc631fb7 | 9174 | smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */ |
03b042bf | 9175 | put_online_cpus(); |
03b042bf PM |
9176 | } |
9177 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9178 | ||
9179 | #endif /* #else #ifndef CONFIG_SMP */ |