Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * kernel/sched.c | |
3 | * | |
4 | * Kernel scheduler and related syscalls | |
5 | * | |
6 | * Copyright (C) 1991-2002 Linus Torvalds | |
7 | * | |
8 | * 1996-12-23 Modified by Dave Grothe to fix bugs in semaphores and | |
9 | * make semaphores SMP safe | |
10 | * 1998-11-19 Implemented schedule_timeout() and related stuff | |
11 | * by Andrea Arcangeli | |
12 | * 2002-01-04 New ultra-scalable O(1) scheduler by Ingo Molnar: | |
13 | * hybrid priority-list and round-robin design with | |
14 | * an array-switch method of distributing timeslices | |
15 | * and per-CPU runqueues. Cleanups and useful suggestions | |
16 | * by Davide Libenzi, preemptible kernel bits by Robert Love. | |
17 | * 2003-09-03 Interactivity tuning by Con Kolivas. | |
18 | * 2004-04-02 Scheduler domains code by Nick Piggin | |
c31f2e8a IM |
19 | * 2007-04-15 Work begun on replacing all interactivity tuning with a |
20 | * fair scheduling design by Con Kolivas. | |
21 | * 2007-05-05 Load balancing (smp-nice) and other improvements | |
22 | * by Peter Williams | |
23 | * 2007-05-06 Interactivity improvements to CFS by Mike Galbraith | |
24 | * 2007-07-01 Group scheduling enhancements by Srivatsa Vaddagiri | |
b9131769 IM |
25 | * 2007-11-29 RT balancing improvements by Steven Rostedt, Gregory Haskins, |
26 | * Thomas Gleixner, Mike Kravetz | |
1da177e4 LT |
27 | */ |
28 | ||
29 | #include <linux/mm.h> | |
30 | #include <linux/module.h> | |
31 | #include <linux/nmi.h> | |
32 | #include <linux/init.h> | |
dff06c15 | 33 | #include <linux/uaccess.h> |
1da177e4 LT |
34 | #include <linux/highmem.h> |
35 | #include <linux/smp_lock.h> | |
36 | #include <asm/mmu_context.h> | |
37 | #include <linux/interrupt.h> | |
c59ede7b | 38 | #include <linux/capability.h> |
1da177e4 LT |
39 | #include <linux/completion.h> |
40 | #include <linux/kernel_stat.h> | |
9a11b49a | 41 | #include <linux/debug_locks.h> |
cdd6c482 | 42 | #include <linux/perf_event.h> |
1da177e4 LT |
43 | #include <linux/security.h> |
44 | #include <linux/notifier.h> | |
45 | #include <linux/profile.h> | |
7dfb7103 | 46 | #include <linux/freezer.h> |
198e2f18 | 47 | #include <linux/vmalloc.h> |
1da177e4 LT |
48 | #include <linux/blkdev.h> |
49 | #include <linux/delay.h> | |
b488893a | 50 | #include <linux/pid_namespace.h> |
1da177e4 LT |
51 | #include <linux/smp.h> |
52 | #include <linux/threads.h> | |
53 | #include <linux/timer.h> | |
54 | #include <linux/rcupdate.h> | |
55 | #include <linux/cpu.h> | |
56 | #include <linux/cpuset.h> | |
57 | #include <linux/percpu.h> | |
b5aadf7f | 58 | #include <linux/proc_fs.h> |
1da177e4 | 59 | #include <linux/seq_file.h> |
969c7921 | 60 | #include <linux/stop_machine.h> |
e692ab53 | 61 | #include <linux/sysctl.h> |
1da177e4 LT |
62 | #include <linux/syscalls.h> |
63 | #include <linux/times.h> | |
8f0ab514 | 64 | #include <linux/tsacct_kern.h> |
c6fd91f0 | 65 | #include <linux/kprobes.h> |
0ff92245 | 66 | #include <linux/delayacct.h> |
dff06c15 | 67 | #include <linux/unistd.h> |
f5ff8422 | 68 | #include <linux/pagemap.h> |
8f4d37ec | 69 | #include <linux/hrtimer.h> |
30914a58 | 70 | #include <linux/tick.h> |
f00b45c1 PZ |
71 | #include <linux/debugfs.h> |
72 | #include <linux/ctype.h> | |
6cd8a4bb | 73 | #include <linux/ftrace.h> |
5a0e3ad6 | 74 | #include <linux/slab.h> |
1da177e4 | 75 | |
5517d86b | 76 | #include <asm/tlb.h> |
838225b4 | 77 | #include <asm/irq_regs.h> |
1da177e4 | 78 | |
6e0534f2 | 79 | #include "sched_cpupri.h" |
21aa9af0 | 80 | #include "workqueue_sched.h" |
6e0534f2 | 81 | |
a8d154b0 | 82 | #define CREATE_TRACE_POINTS |
ad8d75ff | 83 | #include <trace/events/sched.h> |
a8d154b0 | 84 | |
1da177e4 LT |
85 | /* |
86 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
87 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
88 | * and back. | |
89 | */ | |
90 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
91 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
92 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
93 | ||
94 | /* | |
95 | * 'User priority' is the nice value converted to something we | |
96 | * can work with better when scaling various scheduler parameters, | |
97 | * it's a [ 0 ... 39 ] range. | |
98 | */ | |
99 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
100 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
101 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
102 | ||
103 | /* | |
d7876a08 | 104 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 105 | */ |
d6322faf | 106 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 107 | |
6aa645ea IM |
108 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
109 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
110 | ||
1da177e4 LT |
111 | /* |
112 | * These are the 'tuning knobs' of the scheduler: | |
113 | * | |
a4ec24b4 | 114 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
115 | * Timeslices get refilled after they expire. |
116 | */ | |
1da177e4 | 117 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 118 | |
d0b27fa7 PZ |
119 | /* |
120 | * single value that denotes runtime == period, ie unlimited time. | |
121 | */ | |
122 | #define RUNTIME_INF ((u64)~0ULL) | |
123 | ||
e05606d3 IM |
124 | static inline int rt_policy(int policy) |
125 | { | |
3f33a7ce | 126 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
127 | return 1; |
128 | return 0; | |
129 | } | |
130 | ||
131 | static inline int task_has_rt_policy(struct task_struct *p) | |
132 | { | |
133 | return rt_policy(p->policy); | |
134 | } | |
135 | ||
1da177e4 | 136 | /* |
6aa645ea | 137 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 138 | */ |
6aa645ea IM |
139 | struct rt_prio_array { |
140 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
141 | struct list_head queue[MAX_RT_PRIO]; | |
142 | }; | |
143 | ||
d0b27fa7 | 144 | struct rt_bandwidth { |
ea736ed5 | 145 | /* nests inside the rq lock: */ |
0986b11b | 146 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
147 | ktime_t rt_period; |
148 | u64 rt_runtime; | |
149 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
150 | }; |
151 | ||
152 | static struct rt_bandwidth def_rt_bandwidth; | |
153 | ||
154 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
155 | ||
156 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
157 | { | |
158 | struct rt_bandwidth *rt_b = | |
159 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
160 | ktime_t now; | |
161 | int overrun; | |
162 | int idle = 0; | |
163 | ||
164 | for (;;) { | |
165 | now = hrtimer_cb_get_time(timer); | |
166 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
167 | ||
168 | if (!overrun) | |
169 | break; | |
170 | ||
171 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
172 | } | |
173 | ||
174 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
175 | } | |
176 | ||
177 | static | |
178 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
179 | { | |
180 | rt_b->rt_period = ns_to_ktime(period); | |
181 | rt_b->rt_runtime = runtime; | |
182 | ||
0986b11b | 183 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 184 | |
d0b27fa7 PZ |
185 | hrtimer_init(&rt_b->rt_period_timer, |
186 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
187 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
188 | } |
189 | ||
c8bfff6d KH |
190 | static inline int rt_bandwidth_enabled(void) |
191 | { | |
192 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
193 | } |
194 | ||
195 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
196 | { | |
197 | ktime_t now; | |
198 | ||
cac64d00 | 199 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
200 | return; |
201 | ||
202 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
203 | return; | |
204 | ||
0986b11b | 205 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 206 | for (;;) { |
7f1e2ca9 PZ |
207 | unsigned long delta; |
208 | ktime_t soft, hard; | |
209 | ||
d0b27fa7 PZ |
210 | if (hrtimer_active(&rt_b->rt_period_timer)) |
211 | break; | |
212 | ||
213 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
214 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
215 | |
216 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
217 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
218 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
219 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 220 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 221 | } |
0986b11b | 222 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
223 | } |
224 | ||
225 | #ifdef CONFIG_RT_GROUP_SCHED | |
226 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
227 | { | |
228 | hrtimer_cancel(&rt_b->rt_period_timer); | |
229 | } | |
230 | #endif | |
231 | ||
712555ee HC |
232 | /* |
233 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
234 | * detach_destroy_domains and partition_sched_domains. | |
235 | */ | |
236 | static DEFINE_MUTEX(sched_domains_mutex); | |
237 | ||
7c941438 | 238 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 239 | |
68318b8e SV |
240 | #include <linux/cgroup.h> |
241 | ||
29f59db3 SV |
242 | struct cfs_rq; |
243 | ||
6f505b16 PZ |
244 | static LIST_HEAD(task_groups); |
245 | ||
29f59db3 | 246 | /* task group related information */ |
4cf86d77 | 247 | struct task_group { |
68318b8e | 248 | struct cgroup_subsys_state css; |
6c415b92 | 249 | |
052f1dc7 | 250 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
251 | /* schedulable entities of this group on each cpu */ |
252 | struct sched_entity **se; | |
253 | /* runqueue "owned" by this group on each cpu */ | |
254 | struct cfs_rq **cfs_rq; | |
255 | unsigned long shares; | |
052f1dc7 PZ |
256 | #endif |
257 | ||
258 | #ifdef CONFIG_RT_GROUP_SCHED | |
259 | struct sched_rt_entity **rt_se; | |
260 | struct rt_rq **rt_rq; | |
261 | ||
d0b27fa7 | 262 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 263 | #endif |
6b2d7700 | 264 | |
ae8393e5 | 265 | struct rcu_head rcu; |
6f505b16 | 266 | struct list_head list; |
f473aa5e PZ |
267 | |
268 | struct task_group *parent; | |
269 | struct list_head siblings; | |
270 | struct list_head children; | |
29f59db3 SV |
271 | }; |
272 | ||
eff766a6 | 273 | #define root_task_group init_task_group |
6f505b16 | 274 | |
8ed36996 | 275 | /* task_group_lock serializes add/remove of task groups and also changes to |
ec2c507f SV |
276 | * a task group's cpu shares. |
277 | */ | |
8ed36996 | 278 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 279 | |
e9036b36 CG |
280 | #ifdef CONFIG_FAIR_GROUP_SCHED |
281 | ||
57310a98 PZ |
282 | #ifdef CONFIG_SMP |
283 | static int root_task_group_empty(void) | |
284 | { | |
285 | return list_empty(&root_task_group.children); | |
286 | } | |
287 | #endif | |
288 | ||
052f1dc7 | 289 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 290 | |
cb4ad1ff | 291 | /* |
2e084786 LJ |
292 | * A weight of 0 or 1 can cause arithmetics problems. |
293 | * A weight of a cfs_rq is the sum of weights of which entities | |
294 | * are queued on this cfs_rq, so a weight of a entity should not be | |
295 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
296 | * (The default weight is 1024 - so there's no practical |
297 | * limitation from this.) | |
298 | */ | |
18d95a28 | 299 | #define MIN_SHARES 2 |
2e084786 | 300 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 301 | |
052f1dc7 PZ |
302 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
303 | #endif | |
304 | ||
29f59db3 | 305 | /* Default task group. |
3a252015 | 306 | * Every task in system belong to this group at bootup. |
29f59db3 | 307 | */ |
434d53b0 | 308 | struct task_group init_task_group; |
29f59db3 | 309 | |
7c941438 | 310 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 311 | |
6aa645ea IM |
312 | /* CFS-related fields in a runqueue */ |
313 | struct cfs_rq { | |
314 | struct load_weight load; | |
315 | unsigned long nr_running; | |
316 | ||
6aa645ea | 317 | u64 exec_clock; |
e9acbff6 | 318 | u64 min_vruntime; |
6aa645ea IM |
319 | |
320 | struct rb_root tasks_timeline; | |
321 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
322 | |
323 | struct list_head tasks; | |
324 | struct list_head *balance_iterator; | |
325 | ||
326 | /* | |
327 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
328 | * It is set to NULL otherwise (i.e when none are currently running). |
329 | */ | |
4793241b | 330 | struct sched_entity *curr, *next, *last; |
ddc97297 | 331 | |
5ac5c4d6 | 332 | unsigned int nr_spread_over; |
ddc97297 | 333 | |
62160e3f | 334 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
335 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
336 | ||
41a2d6cf IM |
337 | /* |
338 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
339 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
340 | * (like users, containers etc.) | |
341 | * | |
342 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
343 | * list is used during load balance. | |
344 | */ | |
41a2d6cf IM |
345 | struct list_head leaf_cfs_rq_list; |
346 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
347 | |
348 | #ifdef CONFIG_SMP | |
c09595f6 | 349 | /* |
c8cba857 | 350 | * the part of load.weight contributed by tasks |
c09595f6 | 351 | */ |
c8cba857 | 352 | unsigned long task_weight; |
c09595f6 | 353 | |
c8cba857 PZ |
354 | /* |
355 | * h_load = weight * f(tg) | |
356 | * | |
357 | * Where f(tg) is the recursive weight fraction assigned to | |
358 | * this group. | |
359 | */ | |
360 | unsigned long h_load; | |
c09595f6 | 361 | |
c8cba857 PZ |
362 | /* |
363 | * this cpu's part of tg->shares | |
364 | */ | |
365 | unsigned long shares; | |
f1d239f7 PZ |
366 | |
367 | /* | |
368 | * load.weight at the time we set shares | |
369 | */ | |
370 | unsigned long rq_weight; | |
c09595f6 | 371 | #endif |
6aa645ea IM |
372 | #endif |
373 | }; | |
1da177e4 | 374 | |
6aa645ea IM |
375 | /* Real-Time classes' related field in a runqueue: */ |
376 | struct rt_rq { | |
377 | struct rt_prio_array active; | |
63489e45 | 378 | unsigned long rt_nr_running; |
052f1dc7 | 379 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
380 | struct { |
381 | int curr; /* highest queued rt task prio */ | |
398a153b | 382 | #ifdef CONFIG_SMP |
e864c499 | 383 | int next; /* next highest */ |
398a153b | 384 | #endif |
e864c499 | 385 | } highest_prio; |
6f505b16 | 386 | #endif |
fa85ae24 | 387 | #ifdef CONFIG_SMP |
73fe6aae | 388 | unsigned long rt_nr_migratory; |
a1ba4d8b | 389 | unsigned long rt_nr_total; |
a22d7fc1 | 390 | int overloaded; |
917b627d | 391 | struct plist_head pushable_tasks; |
fa85ae24 | 392 | #endif |
6f505b16 | 393 | int rt_throttled; |
fa85ae24 | 394 | u64 rt_time; |
ac086bc2 | 395 | u64 rt_runtime; |
ea736ed5 | 396 | /* Nests inside the rq lock: */ |
0986b11b | 397 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 398 | |
052f1dc7 | 399 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
400 | unsigned long rt_nr_boosted; |
401 | ||
6f505b16 PZ |
402 | struct rq *rq; |
403 | struct list_head leaf_rt_rq_list; | |
404 | struct task_group *tg; | |
6f505b16 | 405 | #endif |
6aa645ea IM |
406 | }; |
407 | ||
57d885fe GH |
408 | #ifdef CONFIG_SMP |
409 | ||
410 | /* | |
411 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
412 | * variables. Each exclusive cpuset essentially defines an island domain by |
413 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
414 | * exclusive cpuset is created, we also create and attach a new root-domain |
415 | * object. | |
416 | * | |
57d885fe GH |
417 | */ |
418 | struct root_domain { | |
419 | atomic_t refcount; | |
c6c4927b RR |
420 | cpumask_var_t span; |
421 | cpumask_var_t online; | |
637f5085 | 422 | |
0eab9146 | 423 | /* |
637f5085 GH |
424 | * The "RT overload" flag: it gets set if a CPU has more than |
425 | * one runnable RT task. | |
426 | */ | |
c6c4927b | 427 | cpumask_var_t rto_mask; |
0eab9146 | 428 | atomic_t rto_count; |
6e0534f2 | 429 | struct cpupri cpupri; |
57d885fe GH |
430 | }; |
431 | ||
dc938520 GH |
432 | /* |
433 | * By default the system creates a single root-domain with all cpus as | |
434 | * members (mimicking the global state we have today). | |
435 | */ | |
57d885fe GH |
436 | static struct root_domain def_root_domain; |
437 | ||
ed2d372c | 438 | #endif /* CONFIG_SMP */ |
57d885fe | 439 | |
1da177e4 LT |
440 | /* |
441 | * This is the main, per-CPU runqueue data structure. | |
442 | * | |
443 | * Locking rule: those places that want to lock multiple runqueues | |
444 | * (such as the load balancing or the thread migration code), lock | |
445 | * acquire operations must be ordered by ascending &runqueue. | |
446 | */ | |
70b97a7f | 447 | struct rq { |
d8016491 | 448 | /* runqueue lock: */ |
05fa785c | 449 | raw_spinlock_t lock; |
1da177e4 LT |
450 | |
451 | /* | |
452 | * nr_running and cpu_load should be in the same cacheline because | |
453 | * remote CPUs use both these fields when doing load calculation. | |
454 | */ | |
455 | unsigned long nr_running; | |
6aa645ea IM |
456 | #define CPU_LOAD_IDX_MAX 5 |
457 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 458 | unsigned long last_load_update_tick; |
46cb4b7c | 459 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 460 | u64 nohz_stamp; |
83cd4fe2 | 461 | unsigned char nohz_balance_kick; |
46cb4b7c | 462 | #endif |
a64692a3 MG |
463 | unsigned int skip_clock_update; |
464 | ||
d8016491 IM |
465 | /* capture load from *all* tasks on this cpu: */ |
466 | struct load_weight load; | |
6aa645ea IM |
467 | unsigned long nr_load_updates; |
468 | u64 nr_switches; | |
469 | ||
470 | struct cfs_rq cfs; | |
6f505b16 | 471 | struct rt_rq rt; |
6f505b16 | 472 | |
6aa645ea | 473 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
474 | /* list of leaf cfs_rq on this cpu: */ |
475 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
476 | #endif |
477 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 478 | struct list_head leaf_rt_rq_list; |
1da177e4 | 479 | #endif |
1da177e4 LT |
480 | |
481 | /* | |
482 | * This is part of a global counter where only the total sum | |
483 | * over all CPUs matters. A task can increase this counter on | |
484 | * one CPU and if it got migrated afterwards it may decrease | |
485 | * it on another CPU. Always updated under the runqueue lock: | |
486 | */ | |
487 | unsigned long nr_uninterruptible; | |
488 | ||
34f971f6 | 489 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 490 | unsigned long next_balance; |
1da177e4 | 491 | struct mm_struct *prev_mm; |
6aa645ea | 492 | |
3e51f33f | 493 | u64 clock; |
305e6835 | 494 | u64 clock_task; |
6aa645ea | 495 | |
1da177e4 LT |
496 | atomic_t nr_iowait; |
497 | ||
498 | #ifdef CONFIG_SMP | |
0eab9146 | 499 | struct root_domain *rd; |
1da177e4 LT |
500 | struct sched_domain *sd; |
501 | ||
e51fd5e2 PZ |
502 | unsigned long cpu_power; |
503 | ||
a0a522ce | 504 | unsigned char idle_at_tick; |
1da177e4 | 505 | /* For active balancing */ |
3f029d3c | 506 | int post_schedule; |
1da177e4 LT |
507 | int active_balance; |
508 | int push_cpu; | |
969c7921 | 509 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
510 | /* cpu of this runqueue: */ |
511 | int cpu; | |
1f11eb6a | 512 | int online; |
1da177e4 | 513 | |
a8a51d5e | 514 | unsigned long avg_load_per_task; |
1da177e4 | 515 | |
e9e9250b PZ |
516 | u64 rt_avg; |
517 | u64 age_stamp; | |
1b9508f6 MG |
518 | u64 idle_stamp; |
519 | u64 avg_idle; | |
1da177e4 LT |
520 | #endif |
521 | ||
aa483808 VP |
522 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
523 | u64 prev_irq_time; | |
524 | #endif | |
525 | ||
dce48a84 TG |
526 | /* calc_load related fields */ |
527 | unsigned long calc_load_update; | |
528 | long calc_load_active; | |
529 | ||
8f4d37ec | 530 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
531 | #ifdef CONFIG_SMP |
532 | int hrtick_csd_pending; | |
533 | struct call_single_data hrtick_csd; | |
534 | #endif | |
8f4d37ec PZ |
535 | struct hrtimer hrtick_timer; |
536 | #endif | |
537 | ||
1da177e4 LT |
538 | #ifdef CONFIG_SCHEDSTATS |
539 | /* latency stats */ | |
540 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
541 | unsigned long long rq_cpu_time; |
542 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
543 | |
544 | /* sys_sched_yield() stats */ | |
480b9434 | 545 | unsigned int yld_count; |
1da177e4 LT |
546 | |
547 | /* schedule() stats */ | |
480b9434 KC |
548 | unsigned int sched_switch; |
549 | unsigned int sched_count; | |
550 | unsigned int sched_goidle; | |
1da177e4 LT |
551 | |
552 | /* try_to_wake_up() stats */ | |
480b9434 KC |
553 | unsigned int ttwu_count; |
554 | unsigned int ttwu_local; | |
b8efb561 IM |
555 | |
556 | /* BKL stats */ | |
480b9434 | 557 | unsigned int bkl_count; |
1da177e4 LT |
558 | #endif |
559 | }; | |
560 | ||
f34e3b61 | 561 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 562 | |
a64692a3 | 563 | |
1e5a7405 | 564 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 565 | |
0a2966b4 CL |
566 | static inline int cpu_of(struct rq *rq) |
567 | { | |
568 | #ifdef CONFIG_SMP | |
569 | return rq->cpu; | |
570 | #else | |
571 | return 0; | |
572 | #endif | |
573 | } | |
574 | ||
497f0ab3 | 575 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
576 | rcu_dereference_check((p), \ |
577 | rcu_read_lock_sched_held() || \ | |
578 | lockdep_is_held(&sched_domains_mutex)) | |
579 | ||
674311d5 NP |
580 | /* |
581 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 582 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
583 | * |
584 | * The domain tree of any CPU may only be accessed from within | |
585 | * preempt-disabled sections. | |
586 | */ | |
48f24c4d | 587 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 588 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
589 | |
590 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
591 | #define this_rq() (&__get_cpu_var(runqueues)) | |
592 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
593 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 594 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 595 | |
dc61b1d6 PZ |
596 | #ifdef CONFIG_CGROUP_SCHED |
597 | ||
598 | /* | |
599 | * Return the group to which this tasks belongs. | |
600 | * | |
601 | * We use task_subsys_state_check() and extend the RCU verification | |
602 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
603 | * holds that lock for each task it moves into the cgroup. Therefore | |
604 | * by holding that lock, we pin the task to the current cgroup. | |
605 | */ | |
606 | static inline struct task_group *task_group(struct task_struct *p) | |
607 | { | |
608 | struct cgroup_subsys_state *css; | |
609 | ||
610 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
611 | lockdep_is_held(&task_rq(p)->lock)); | |
612 | return container_of(css, struct task_group, css); | |
613 | } | |
614 | ||
615 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
616 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
617 | { | |
618 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
619 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
620 | p->se.parent = task_group(p)->se[cpu]; | |
621 | #endif | |
622 | ||
623 | #ifdef CONFIG_RT_GROUP_SCHED | |
624 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
625 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
626 | #endif | |
627 | } | |
628 | ||
629 | #else /* CONFIG_CGROUP_SCHED */ | |
630 | ||
631 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
632 | static inline struct task_group *task_group(struct task_struct *p) | |
633 | { | |
634 | return NULL; | |
635 | } | |
636 | ||
637 | #endif /* CONFIG_CGROUP_SCHED */ | |
638 | ||
fe44d621 | 639 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 640 | |
fe44d621 | 641 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 642 | { |
fe44d621 | 643 | s64 delta; |
305e6835 | 644 | |
f26f9aff MG |
645 | if (rq->skip_clock_update) |
646 | return; | |
aa483808 | 647 | |
fe44d621 PZ |
648 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
649 | rq->clock += delta; | |
650 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
651 | } |
652 | ||
bf5c91ba IM |
653 | /* |
654 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
655 | */ | |
656 | #ifdef CONFIG_SCHED_DEBUG | |
657 | # define const_debug __read_mostly | |
658 | #else | |
659 | # define const_debug static const | |
660 | #endif | |
661 | ||
017730c1 IM |
662 | /** |
663 | * runqueue_is_locked | |
e17b38bf | 664 | * @cpu: the processor in question. |
017730c1 IM |
665 | * |
666 | * Returns true if the current cpu runqueue is locked. | |
667 | * This interface allows printk to be called with the runqueue lock | |
668 | * held and know whether or not it is OK to wake up the klogd. | |
669 | */ | |
89f19f04 | 670 | int runqueue_is_locked(int cpu) |
017730c1 | 671 | { |
05fa785c | 672 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
673 | } |
674 | ||
bf5c91ba IM |
675 | /* |
676 | * Debugging: various feature bits | |
677 | */ | |
f00b45c1 PZ |
678 | |
679 | #define SCHED_FEAT(name, enabled) \ | |
680 | __SCHED_FEAT_##name , | |
681 | ||
bf5c91ba | 682 | enum { |
f00b45c1 | 683 | #include "sched_features.h" |
bf5c91ba IM |
684 | }; |
685 | ||
f00b45c1 PZ |
686 | #undef SCHED_FEAT |
687 | ||
688 | #define SCHED_FEAT(name, enabled) \ | |
689 | (1UL << __SCHED_FEAT_##name) * enabled | | |
690 | ||
bf5c91ba | 691 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
692 | #include "sched_features.h" |
693 | 0; | |
694 | ||
695 | #undef SCHED_FEAT | |
696 | ||
697 | #ifdef CONFIG_SCHED_DEBUG | |
698 | #define SCHED_FEAT(name, enabled) \ | |
699 | #name , | |
700 | ||
983ed7a6 | 701 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
702 | #include "sched_features.h" |
703 | NULL | |
704 | }; | |
705 | ||
706 | #undef SCHED_FEAT | |
707 | ||
34f3a814 | 708 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 709 | { |
f00b45c1 PZ |
710 | int i; |
711 | ||
712 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
713 | if (!(sysctl_sched_features & (1UL << i))) |
714 | seq_puts(m, "NO_"); | |
715 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 716 | } |
34f3a814 | 717 | seq_puts(m, "\n"); |
f00b45c1 | 718 | |
34f3a814 | 719 | return 0; |
f00b45c1 PZ |
720 | } |
721 | ||
722 | static ssize_t | |
723 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
724 | size_t cnt, loff_t *ppos) | |
725 | { | |
726 | char buf[64]; | |
7740191c | 727 | char *cmp; |
f00b45c1 PZ |
728 | int neg = 0; |
729 | int i; | |
730 | ||
731 | if (cnt > 63) | |
732 | cnt = 63; | |
733 | ||
734 | if (copy_from_user(&buf, ubuf, cnt)) | |
735 | return -EFAULT; | |
736 | ||
737 | buf[cnt] = 0; | |
7740191c | 738 | cmp = strstrip(buf); |
f00b45c1 | 739 | |
c24b7c52 | 740 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
741 | neg = 1; |
742 | cmp += 3; | |
743 | } | |
744 | ||
745 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 746 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
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 | 1257 | |
6d6bc0ad | 1258 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1259 | |
e9e9250b PZ |
1260 | static u64 sched_avg_period(void) |
1261 | { | |
1262 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1263 | } | |
1264 | ||
1265 | static void sched_avg_update(struct rq *rq) | |
1266 | { | |
1267 | s64 period = sched_avg_period(); | |
1268 | ||
1269 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1270 | /* |
1271 | * Inline assembly required to prevent the compiler | |
1272 | * optimising this loop into a divmod call. | |
1273 | * See __iter_div_u64_rem() for another example of this. | |
1274 | */ | |
1275 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1276 | rq->age_stamp += period; |
1277 | rq->rt_avg /= 2; | |
1278 | } | |
1279 | } | |
1280 | ||
1281 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1282 | { | |
1283 | rq->rt_avg += rt_delta; | |
1284 | sched_avg_update(rq); | |
1285 | } | |
1286 | ||
6d6bc0ad | 1287 | #else /* !CONFIG_SMP */ |
31656519 | 1288 | static void resched_task(struct task_struct *p) |
c24d20db | 1289 | { |
05fa785c | 1290 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1291 | set_tsk_need_resched(p); |
c24d20db | 1292 | } |
e9e9250b PZ |
1293 | |
1294 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1295 | { | |
1296 | } | |
da2b71ed SS |
1297 | |
1298 | static void sched_avg_update(struct rq *rq) | |
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 | { |
eb755805 | 1683 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1684 | } |
1685 | ||
c09595f6 PZ |
1686 | #else |
1687 | ||
c8cba857 | 1688 | static inline void update_shares(struct sched_domain *sd) |
4d8d595d PZ |
1689 | { |
1690 | } | |
1691 | ||
18d95a28 PZ |
1692 | #endif |
1693 | ||
8f45e2b5 GH |
1694 | #ifdef CONFIG_PREEMPT |
1695 | ||
b78bb868 PZ |
1696 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1697 | ||
70574a99 | 1698 | /* |
8f45e2b5 GH |
1699 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1700 | * way at the expense of forcing extra atomic operations in all | |
1701 | * invocations. This assures that the double_lock is acquired using the | |
1702 | * same underlying policy as the spinlock_t on this architecture, which | |
1703 | * reduces latency compared to the unfair variant below. However, it | |
1704 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1705 | */ |
8f45e2b5 GH |
1706 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1707 | __releases(this_rq->lock) | |
1708 | __acquires(busiest->lock) | |
1709 | __acquires(this_rq->lock) | |
1710 | { | |
05fa785c | 1711 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1712 | double_rq_lock(this_rq, busiest); |
1713 | ||
1714 | return 1; | |
1715 | } | |
1716 | ||
1717 | #else | |
1718 | /* | |
1719 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1720 | * latency by eliminating extra atomic operations when the locks are | |
1721 | * already in proper order on entry. This favors lower cpu-ids and will | |
1722 | * grant the double lock to lower cpus over higher ids under contention, | |
1723 | * regardless of entry order into the function. | |
1724 | */ | |
1725 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1726 | __releases(this_rq->lock) |
1727 | __acquires(busiest->lock) | |
1728 | __acquires(this_rq->lock) | |
1729 | { | |
1730 | int ret = 0; | |
1731 | ||
05fa785c | 1732 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1733 | if (busiest < this_rq) { |
05fa785c TG |
1734 | raw_spin_unlock(&this_rq->lock); |
1735 | raw_spin_lock(&busiest->lock); | |
1736 | raw_spin_lock_nested(&this_rq->lock, | |
1737 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1738 | ret = 1; |
1739 | } else | |
05fa785c TG |
1740 | raw_spin_lock_nested(&busiest->lock, |
1741 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1742 | } |
1743 | return ret; | |
1744 | } | |
1745 | ||
8f45e2b5 GH |
1746 | #endif /* CONFIG_PREEMPT */ |
1747 | ||
1748 | /* | |
1749 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1750 | */ | |
1751 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1752 | { | |
1753 | if (unlikely(!irqs_disabled())) { | |
1754 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1755 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1756 | BUG_ON(1); |
1757 | } | |
1758 | ||
1759 | return _double_lock_balance(this_rq, busiest); | |
1760 | } | |
1761 | ||
70574a99 AD |
1762 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1763 | __releases(busiest->lock) | |
1764 | { | |
05fa785c | 1765 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1766 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1767 | } | |
1e3c88bd PZ |
1768 | |
1769 | /* | |
1770 | * double_rq_lock - safely lock two runqueues | |
1771 | * | |
1772 | * Note this does not disable interrupts like task_rq_lock, | |
1773 | * you need to do so manually before calling. | |
1774 | */ | |
1775 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1776 | __acquires(rq1->lock) | |
1777 | __acquires(rq2->lock) | |
1778 | { | |
1779 | BUG_ON(!irqs_disabled()); | |
1780 | if (rq1 == rq2) { | |
1781 | raw_spin_lock(&rq1->lock); | |
1782 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1783 | } else { | |
1784 | if (rq1 < rq2) { | |
1785 | raw_spin_lock(&rq1->lock); | |
1786 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1787 | } else { | |
1788 | raw_spin_lock(&rq2->lock); | |
1789 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1790 | } | |
1791 | } | |
1e3c88bd PZ |
1792 | } |
1793 | ||
1794 | /* | |
1795 | * double_rq_unlock - safely unlock two runqueues | |
1796 | * | |
1797 | * Note this does not restore interrupts like task_rq_unlock, | |
1798 | * you need to do so manually after calling. | |
1799 | */ | |
1800 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1801 | __releases(rq1->lock) | |
1802 | __releases(rq2->lock) | |
1803 | { | |
1804 | raw_spin_unlock(&rq1->lock); | |
1805 | if (rq1 != rq2) | |
1806 | raw_spin_unlock(&rq2->lock); | |
1807 | else | |
1808 | __release(rq2->lock); | |
1809 | } | |
1810 | ||
18d95a28 PZ |
1811 | #endif |
1812 | ||
30432094 | 1813 | #ifdef CONFIG_FAIR_GROUP_SCHED |
34e83e85 IM |
1814 | static void cfs_rq_set_shares(struct cfs_rq *cfs_rq, unsigned long shares) |
1815 | { | |
30432094 | 1816 | #ifdef CONFIG_SMP |
34e83e85 IM |
1817 | cfs_rq->shares = shares; |
1818 | #endif | |
1819 | } | |
30432094 | 1820 | #endif |
e7693a36 | 1821 | |
74f5187a | 1822 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1823 | static void update_sysctl(void); |
acb4a848 | 1824 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1825 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1826 | |
cd29fe6f PZ |
1827 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1828 | { | |
1829 | set_task_rq(p, cpu); | |
1830 | #ifdef CONFIG_SMP | |
1831 | /* | |
1832 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1833 | * successfuly executed on another CPU. We must ensure that updates of | |
1834 | * per-task data have been completed by this moment. | |
1835 | */ | |
1836 | smp_wmb(); | |
1837 | task_thread_info(p)->cpu = cpu; | |
1838 | #endif | |
1839 | } | |
dce48a84 | 1840 | |
1e3c88bd | 1841 | static const struct sched_class rt_sched_class; |
dd41f596 | 1842 | |
34f971f6 | 1843 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1844 | #define for_each_class(class) \ |
1845 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1846 | |
1e3c88bd PZ |
1847 | #include "sched_stats.h" |
1848 | ||
c09595f6 | 1849 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1850 | { |
1851 | rq->nr_running++; | |
9c217245 IM |
1852 | } |
1853 | ||
c09595f6 | 1854 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1855 | { |
1856 | rq->nr_running--; | |
9c217245 IM |
1857 | } |
1858 | ||
45bf76df IM |
1859 | static void set_load_weight(struct task_struct *p) |
1860 | { | |
dd41f596 IM |
1861 | /* |
1862 | * SCHED_IDLE tasks get minimal weight: | |
1863 | */ | |
1864 | if (p->policy == SCHED_IDLE) { | |
1865 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1866 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1867 | return; | |
1868 | } | |
71f8bd46 | 1869 | |
dd41f596 IM |
1870 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1871 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1872 | } |
1873 | ||
371fd7e7 | 1874 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1875 | { |
a64692a3 | 1876 | update_rq_clock(rq); |
dd41f596 | 1877 | sched_info_queued(p); |
371fd7e7 | 1878 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1879 | p->se.on_rq = 1; |
71f8bd46 IM |
1880 | } |
1881 | ||
371fd7e7 | 1882 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1883 | { |
a64692a3 | 1884 | update_rq_clock(rq); |
46ac22ba | 1885 | sched_info_dequeued(p); |
371fd7e7 | 1886 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1887 | p->se.on_rq = 0; |
71f8bd46 IM |
1888 | } |
1889 | ||
1e3c88bd PZ |
1890 | /* |
1891 | * activate_task - move a task to the runqueue. | |
1892 | */ | |
371fd7e7 | 1893 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1894 | { |
1895 | if (task_contributes_to_load(p)) | |
1896 | rq->nr_uninterruptible--; | |
1897 | ||
371fd7e7 | 1898 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1899 | inc_nr_running(rq); |
1900 | } | |
1901 | ||
1902 | /* | |
1903 | * deactivate_task - remove a task from the runqueue. | |
1904 | */ | |
371fd7e7 | 1905 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1906 | { |
1907 | if (task_contributes_to_load(p)) | |
1908 | rq->nr_uninterruptible++; | |
1909 | ||
371fd7e7 | 1910 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1911 | dec_nr_running(rq); |
1912 | } | |
1913 | ||
b52bfee4 VP |
1914 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1915 | ||
305e6835 VP |
1916 | /* |
1917 | * There are no locks covering percpu hardirq/softirq time. | |
1918 | * They are only modified in account_system_vtime, on corresponding CPU | |
1919 | * with interrupts disabled. So, writes are safe. | |
1920 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1921 | * This may result in other CPU reading this CPU's irq time and can | |
1922 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1923 | * or new value with a side effect of accounting a slice of irq time to wrong |
1924 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1925 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1926 | */ |
b52bfee4 VP |
1927 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1928 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1929 | ||
1930 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1931 | static int sched_clock_irqtime; | |
1932 | ||
1933 | void enable_sched_clock_irqtime(void) | |
1934 | { | |
1935 | sched_clock_irqtime = 1; | |
1936 | } | |
1937 | ||
1938 | void disable_sched_clock_irqtime(void) | |
1939 | { | |
1940 | sched_clock_irqtime = 0; | |
1941 | } | |
1942 | ||
8e92c201 PZ |
1943 | #ifndef CONFIG_64BIT |
1944 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1945 | ||
1946 | static inline void irq_time_write_begin(void) | |
1947 | { | |
1948 | __this_cpu_inc(irq_time_seq.sequence); | |
1949 | smp_wmb(); | |
1950 | } | |
1951 | ||
1952 | static inline void irq_time_write_end(void) | |
1953 | { | |
1954 | smp_wmb(); | |
1955 | __this_cpu_inc(irq_time_seq.sequence); | |
1956 | } | |
1957 | ||
1958 | static inline u64 irq_time_read(int cpu) | |
1959 | { | |
1960 | u64 irq_time; | |
1961 | unsigned seq; | |
1962 | ||
1963 | do { | |
1964 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1965 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1966 | per_cpu(cpu_hardirq_time, cpu); | |
1967 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1968 | ||
1969 | return irq_time; | |
1970 | } | |
1971 | #else /* CONFIG_64BIT */ | |
1972 | static inline void irq_time_write_begin(void) | |
1973 | { | |
1974 | } | |
1975 | ||
1976 | static inline void irq_time_write_end(void) | |
1977 | { | |
1978 | } | |
1979 | ||
1980 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1981 | { |
305e6835 VP |
1982 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1983 | } | |
8e92c201 | 1984 | #endif /* CONFIG_64BIT */ |
305e6835 | 1985 | |
fe44d621 PZ |
1986 | /* |
1987 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1988 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1989 | */ | |
b52bfee4 VP |
1990 | void account_system_vtime(struct task_struct *curr) |
1991 | { | |
1992 | unsigned long flags; | |
fe44d621 | 1993 | s64 delta; |
b52bfee4 | 1994 | int cpu; |
b52bfee4 VP |
1995 | |
1996 | if (!sched_clock_irqtime) | |
1997 | return; | |
1998 | ||
1999 | local_irq_save(flags); | |
2000 | ||
b52bfee4 | 2001 | cpu = smp_processor_id(); |
fe44d621 PZ |
2002 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
2003 | __this_cpu_add(irq_start_time, delta); | |
2004 | ||
8e92c201 | 2005 | irq_time_write_begin(); |
b52bfee4 VP |
2006 | /* |
2007 | * We do not account for softirq time from ksoftirqd here. | |
2008 | * We want to continue accounting softirq time to ksoftirqd thread | |
2009 | * in that case, so as not to confuse scheduler with a special task | |
2010 | * that do not consume any time, but still wants to run. | |
2011 | */ | |
2012 | if (hardirq_count()) | |
fe44d621 | 2013 | __this_cpu_add(cpu_hardirq_time, delta); |
b52bfee4 | 2014 | else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD)) |
fe44d621 | 2015 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 2016 | |
8e92c201 | 2017 | irq_time_write_end(); |
b52bfee4 VP |
2018 | local_irq_restore(flags); |
2019 | } | |
b7dadc38 | 2020 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 2021 | |
fe44d621 | 2022 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 2023 | { |
fe44d621 PZ |
2024 | s64 irq_delta; |
2025 | ||
8e92c201 | 2026 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
2027 | |
2028 | /* | |
2029 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
2030 | * this case when a previous update_rq_clock() happened inside a | |
2031 | * {soft,}irq region. | |
2032 | * | |
2033 | * When this happens, we stop ->clock_task and only update the | |
2034 | * prev_irq_time stamp to account for the part that fit, so that a next | |
2035 | * update will consume the rest. This ensures ->clock_task is | |
2036 | * monotonic. | |
2037 | * | |
2038 | * It does however cause some slight miss-attribution of {soft,}irq | |
2039 | * time, a more accurate solution would be to update the irq_time using | |
2040 | * the current rq->clock timestamp, except that would require using | |
2041 | * atomic ops. | |
2042 | */ | |
2043 | if (irq_delta > delta) | |
2044 | irq_delta = delta; | |
2045 | ||
2046 | rq->prev_irq_time += irq_delta; | |
2047 | delta -= irq_delta; | |
2048 | rq->clock_task += delta; | |
2049 | ||
2050 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
2051 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
2052 | } |
2053 | ||
fe44d621 | 2054 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 2055 | |
fe44d621 | 2056 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 2057 | { |
fe44d621 | 2058 | rq->clock_task += delta; |
305e6835 VP |
2059 | } |
2060 | ||
fe44d621 | 2061 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 2062 | |
1e3c88bd PZ |
2063 | #include "sched_idletask.c" |
2064 | #include "sched_fair.c" | |
2065 | #include "sched_rt.c" | |
34f971f6 | 2066 | #include "sched_stoptask.c" |
1e3c88bd PZ |
2067 | #ifdef CONFIG_SCHED_DEBUG |
2068 | # include "sched_debug.c" | |
2069 | #endif | |
2070 | ||
34f971f6 PZ |
2071 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
2072 | { | |
2073 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
2074 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
2075 | ||
2076 | if (stop) { | |
2077 | /* | |
2078 | * Make it appear like a SCHED_FIFO task, its something | |
2079 | * userspace knows about and won't get confused about. | |
2080 | * | |
2081 | * Also, it will make PI more or less work without too | |
2082 | * much confusion -- but then, stop work should not | |
2083 | * rely on PI working anyway. | |
2084 | */ | |
2085 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
2086 | ||
2087 | stop->sched_class = &stop_sched_class; | |
2088 | } | |
2089 | ||
2090 | cpu_rq(cpu)->stop = stop; | |
2091 | ||
2092 | if (old_stop) { | |
2093 | /* | |
2094 | * Reset it back to a normal scheduling class so that | |
2095 | * it can die in pieces. | |
2096 | */ | |
2097 | old_stop->sched_class = &rt_sched_class; | |
2098 | } | |
2099 | } | |
2100 | ||
14531189 | 2101 | /* |
dd41f596 | 2102 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 2103 | */ |
14531189 IM |
2104 | static inline int __normal_prio(struct task_struct *p) |
2105 | { | |
dd41f596 | 2106 | return p->static_prio; |
14531189 IM |
2107 | } |
2108 | ||
b29739f9 IM |
2109 | /* |
2110 | * Calculate the expected normal priority: i.e. priority | |
2111 | * without taking RT-inheritance into account. Might be | |
2112 | * boosted by interactivity modifiers. Changes upon fork, | |
2113 | * setprio syscalls, and whenever the interactivity | |
2114 | * estimator recalculates. | |
2115 | */ | |
36c8b586 | 2116 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
2117 | { |
2118 | int prio; | |
2119 | ||
e05606d3 | 2120 | if (task_has_rt_policy(p)) |
b29739f9 IM |
2121 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
2122 | else | |
2123 | prio = __normal_prio(p); | |
2124 | return prio; | |
2125 | } | |
2126 | ||
2127 | /* | |
2128 | * Calculate the current priority, i.e. the priority | |
2129 | * taken into account by the scheduler. This value might | |
2130 | * be boosted by RT tasks, or might be boosted by | |
2131 | * interactivity modifiers. Will be RT if the task got | |
2132 | * RT-boosted. If not then it returns p->normal_prio. | |
2133 | */ | |
36c8b586 | 2134 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2135 | { |
2136 | p->normal_prio = normal_prio(p); | |
2137 | /* | |
2138 | * If we are RT tasks or we were boosted to RT priority, | |
2139 | * keep the priority unchanged. Otherwise, update priority | |
2140 | * to the normal priority: | |
2141 | */ | |
2142 | if (!rt_prio(p->prio)) | |
2143 | return p->normal_prio; | |
2144 | return p->prio; | |
2145 | } | |
2146 | ||
1da177e4 LT |
2147 | /** |
2148 | * task_curr - is this task currently executing on a CPU? | |
2149 | * @p: the task in question. | |
2150 | */ | |
36c8b586 | 2151 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2152 | { |
2153 | return cpu_curr(task_cpu(p)) == p; | |
2154 | } | |
2155 | ||
cb469845 SR |
2156 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2157 | const struct sched_class *prev_class, | |
2158 | int oldprio, int running) | |
2159 | { | |
2160 | if (prev_class != p->sched_class) { | |
2161 | if (prev_class->switched_from) | |
2162 | prev_class->switched_from(rq, p, running); | |
2163 | p->sched_class->switched_to(rq, p, running); | |
2164 | } else | |
2165 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
2166 | } | |
2167 | ||
1e5a7405 PZ |
2168 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2169 | { | |
2170 | const struct sched_class *class; | |
2171 | ||
2172 | if (p->sched_class == rq->curr->sched_class) { | |
2173 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2174 | } else { | |
2175 | for_each_class(class) { | |
2176 | if (class == rq->curr->sched_class) | |
2177 | break; | |
2178 | if (class == p->sched_class) { | |
2179 | resched_task(rq->curr); | |
2180 | break; | |
2181 | } | |
2182 | } | |
2183 | } | |
2184 | ||
2185 | /* | |
2186 | * A queue event has occurred, and we're going to schedule. In | |
2187 | * this case, we can save a useless back to back clock update. | |
2188 | */ | |
f26f9aff | 2189 | if (rq->curr->se.on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2190 | rq->skip_clock_update = 1; |
2191 | } | |
2192 | ||
1da177e4 | 2193 | #ifdef CONFIG_SMP |
cc367732 IM |
2194 | /* |
2195 | * Is this task likely cache-hot: | |
2196 | */ | |
e7693a36 | 2197 | static int |
cc367732 IM |
2198 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2199 | { | |
2200 | s64 delta; | |
2201 | ||
e6c8fba7 PZ |
2202 | if (p->sched_class != &fair_sched_class) |
2203 | return 0; | |
2204 | ||
ef8002f6 NR |
2205 | if (unlikely(p->policy == SCHED_IDLE)) |
2206 | return 0; | |
2207 | ||
f540a608 IM |
2208 | /* |
2209 | * Buddy candidates are cache hot: | |
2210 | */ | |
f685ceac | 2211 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2212 | (&p->se == cfs_rq_of(&p->se)->next || |
2213 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2214 | return 1; |
2215 | ||
6bc1665b IM |
2216 | if (sysctl_sched_migration_cost == -1) |
2217 | return 1; | |
2218 | if (sysctl_sched_migration_cost == 0) | |
2219 | return 0; | |
2220 | ||
cc367732 IM |
2221 | delta = now - p->se.exec_start; |
2222 | ||
2223 | return delta < (s64)sysctl_sched_migration_cost; | |
2224 | } | |
2225 | ||
dd41f596 | 2226 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2227 | { |
e2912009 PZ |
2228 | #ifdef CONFIG_SCHED_DEBUG |
2229 | /* | |
2230 | * We should never call set_task_cpu() on a blocked task, | |
2231 | * ttwu() will sort out the placement. | |
2232 | */ | |
077614ee PZ |
2233 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2234 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2235 | #endif |
2236 | ||
de1d7286 | 2237 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2238 | |
0c69774e PZ |
2239 | if (task_cpu(p) != new_cpu) { |
2240 | p->se.nr_migrations++; | |
2241 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2242 | } | |
dd41f596 IM |
2243 | |
2244 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2245 | } |
2246 | ||
969c7921 | 2247 | struct migration_arg { |
36c8b586 | 2248 | struct task_struct *task; |
1da177e4 | 2249 | int dest_cpu; |
70b97a7f | 2250 | }; |
1da177e4 | 2251 | |
969c7921 TH |
2252 | static int migration_cpu_stop(void *data); |
2253 | ||
1da177e4 LT |
2254 | /* |
2255 | * The task's runqueue lock must be held. | |
2256 | * Returns true if you have to wait for migration thread. | |
2257 | */ | |
969c7921 | 2258 | static bool migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2259 | { |
70b97a7f | 2260 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2261 | |
2262 | /* | |
2263 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2264 | * the next wake-up will properly place the task. |
1da177e4 | 2265 | */ |
969c7921 | 2266 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2267 | } |
2268 | ||
2269 | /* | |
2270 | * wait_task_inactive - wait for a thread to unschedule. | |
2271 | * | |
85ba2d86 RM |
2272 | * If @match_state is nonzero, it's the @p->state value just checked and |
2273 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2274 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2275 | * we return a positive number (its total switch count). If a second call | |
2276 | * a short while later returns the same number, the caller can be sure that | |
2277 | * @p has remained unscheduled the whole time. | |
2278 | * | |
1da177e4 LT |
2279 | * The caller must ensure that the task *will* unschedule sometime soon, |
2280 | * else this function might spin for a *long* time. This function can't | |
2281 | * be called with interrupts off, or it may introduce deadlock with | |
2282 | * smp_call_function() if an IPI is sent by the same process we are | |
2283 | * waiting to become inactive. | |
2284 | */ | |
85ba2d86 | 2285 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2286 | { |
2287 | unsigned long flags; | |
dd41f596 | 2288 | int running, on_rq; |
85ba2d86 | 2289 | unsigned long ncsw; |
70b97a7f | 2290 | struct rq *rq; |
1da177e4 | 2291 | |
3a5c359a AK |
2292 | for (;;) { |
2293 | /* | |
2294 | * We do the initial early heuristics without holding | |
2295 | * any task-queue locks at all. We'll only try to get | |
2296 | * the runqueue lock when things look like they will | |
2297 | * work out! | |
2298 | */ | |
2299 | rq = task_rq(p); | |
fa490cfd | 2300 | |
3a5c359a AK |
2301 | /* |
2302 | * If the task is actively running on another CPU | |
2303 | * still, just relax and busy-wait without holding | |
2304 | * any locks. | |
2305 | * | |
2306 | * NOTE! Since we don't hold any locks, it's not | |
2307 | * even sure that "rq" stays as the right runqueue! | |
2308 | * But we don't care, since "task_running()" will | |
2309 | * return false if the runqueue has changed and p | |
2310 | * is actually now running somewhere else! | |
2311 | */ | |
85ba2d86 RM |
2312 | while (task_running(rq, p)) { |
2313 | if (match_state && unlikely(p->state != match_state)) | |
2314 | return 0; | |
3a5c359a | 2315 | cpu_relax(); |
85ba2d86 | 2316 | } |
fa490cfd | 2317 | |
3a5c359a AK |
2318 | /* |
2319 | * Ok, time to look more closely! We need the rq | |
2320 | * lock now, to be *sure*. If we're wrong, we'll | |
2321 | * just go back and repeat. | |
2322 | */ | |
2323 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2324 | trace_sched_wait_task(p); |
3a5c359a AK |
2325 | running = task_running(rq, p); |
2326 | on_rq = p->se.on_rq; | |
85ba2d86 | 2327 | ncsw = 0; |
f31e11d8 | 2328 | if (!match_state || p->state == match_state) |
93dcf55f | 2329 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2330 | task_rq_unlock(rq, &flags); |
fa490cfd | 2331 | |
85ba2d86 RM |
2332 | /* |
2333 | * If it changed from the expected state, bail out now. | |
2334 | */ | |
2335 | if (unlikely(!ncsw)) | |
2336 | break; | |
2337 | ||
3a5c359a AK |
2338 | /* |
2339 | * Was it really running after all now that we | |
2340 | * checked with the proper locks actually held? | |
2341 | * | |
2342 | * Oops. Go back and try again.. | |
2343 | */ | |
2344 | if (unlikely(running)) { | |
2345 | cpu_relax(); | |
2346 | continue; | |
2347 | } | |
fa490cfd | 2348 | |
3a5c359a AK |
2349 | /* |
2350 | * It's not enough that it's not actively running, | |
2351 | * it must be off the runqueue _entirely_, and not | |
2352 | * preempted! | |
2353 | * | |
80dd99b3 | 2354 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2355 | * running right now), it's preempted, and we should |
2356 | * yield - it could be a while. | |
2357 | */ | |
2358 | if (unlikely(on_rq)) { | |
2359 | schedule_timeout_uninterruptible(1); | |
2360 | continue; | |
2361 | } | |
fa490cfd | 2362 | |
3a5c359a AK |
2363 | /* |
2364 | * Ahh, all good. It wasn't running, and it wasn't | |
2365 | * runnable, which means that it will never become | |
2366 | * running in the future either. We're all done! | |
2367 | */ | |
2368 | break; | |
2369 | } | |
85ba2d86 RM |
2370 | |
2371 | return ncsw; | |
1da177e4 LT |
2372 | } |
2373 | ||
2374 | /*** | |
2375 | * kick_process - kick a running thread to enter/exit the kernel | |
2376 | * @p: the to-be-kicked thread | |
2377 | * | |
2378 | * Cause a process which is running on another CPU to enter | |
2379 | * kernel-mode, without any delay. (to get signals handled.) | |
2380 | * | |
2381 | * NOTE: this function doesnt have to take the runqueue lock, | |
2382 | * because all it wants to ensure is that the remote task enters | |
2383 | * the kernel. If the IPI races and the task has been migrated | |
2384 | * to another CPU then no harm is done and the purpose has been | |
2385 | * achieved as well. | |
2386 | */ | |
36c8b586 | 2387 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2388 | { |
2389 | int cpu; | |
2390 | ||
2391 | preempt_disable(); | |
2392 | cpu = task_cpu(p); | |
2393 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2394 | smp_send_reschedule(cpu); | |
2395 | preempt_enable(); | |
2396 | } | |
b43e3521 | 2397 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2398 | #endif /* CONFIG_SMP */ |
1da177e4 | 2399 | |
0793a61d TG |
2400 | /** |
2401 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2402 | * @p: the task to evaluate | |
2403 | * @func: the function to be called | |
2404 | * @info: the function call argument | |
2405 | * | |
2406 | * Calls the function @func when the task is currently running. This might | |
2407 | * be on the current CPU, which just calls the function directly | |
2408 | */ | |
2409 | void task_oncpu_function_call(struct task_struct *p, | |
2410 | void (*func) (void *info), void *info) | |
2411 | { | |
2412 | int cpu; | |
2413 | ||
2414 | preempt_disable(); | |
2415 | cpu = task_cpu(p); | |
2416 | if (task_curr(p)) | |
2417 | smp_call_function_single(cpu, func, info, 1); | |
2418 | preempt_enable(); | |
2419 | } | |
2420 | ||
970b13ba | 2421 | #ifdef CONFIG_SMP |
30da688e ON |
2422 | /* |
2423 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2424 | */ | |
5da9a0fb PZ |
2425 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2426 | { | |
2427 | int dest_cpu; | |
2428 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2429 | ||
2430 | /* Look for allowed, online CPU in same node. */ | |
2431 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2432 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2433 | return dest_cpu; | |
2434 | ||
2435 | /* Any allowed, online CPU? */ | |
2436 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2437 | if (dest_cpu < nr_cpu_ids) | |
2438 | return dest_cpu; | |
2439 | ||
2440 | /* No more Mr. Nice Guy. */ | |
897f0b3c | 2441 | if (unlikely(dest_cpu >= nr_cpu_ids)) { |
9084bb82 | 2442 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
5da9a0fb PZ |
2443 | /* |
2444 | * Don't tell them about moving exiting tasks or | |
2445 | * kernel threads (both mm NULL), since they never | |
2446 | * leave kernel. | |
2447 | */ | |
2448 | if (p->mm && printk_ratelimit()) { | |
2449 | printk(KERN_INFO "process %d (%s) no " | |
2450 | "longer affine to cpu%d\n", | |
2451 | task_pid_nr(p), p->comm, cpu); | |
2452 | } | |
2453 | } | |
2454 | ||
2455 | return dest_cpu; | |
2456 | } | |
2457 | ||
e2912009 | 2458 | /* |
30da688e | 2459 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2460 | */ |
970b13ba | 2461 | static inline |
0017d735 | 2462 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2463 | { |
0017d735 | 2464 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2465 | |
2466 | /* | |
2467 | * In order not to call set_task_cpu() on a blocking task we need | |
2468 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2469 | * cpu. | |
2470 | * | |
2471 | * Since this is common to all placement strategies, this lives here. | |
2472 | * | |
2473 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2474 | * not worry about this generic constraint ] | |
2475 | */ | |
2476 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2477 | !cpu_online(cpu))) |
5da9a0fb | 2478 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2479 | |
2480 | return cpu; | |
970b13ba | 2481 | } |
09a40af5 MG |
2482 | |
2483 | static void update_avg(u64 *avg, u64 sample) | |
2484 | { | |
2485 | s64 diff = sample - *avg; | |
2486 | *avg += diff >> 3; | |
2487 | } | |
970b13ba PZ |
2488 | #endif |
2489 | ||
9ed3811a TH |
2490 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2491 | bool is_sync, bool is_migrate, bool is_local, | |
2492 | unsigned long en_flags) | |
2493 | { | |
2494 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2495 | if (is_sync) | |
2496 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2497 | if (is_migrate) | |
2498 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2499 | if (is_local) | |
2500 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2501 | else | |
2502 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2503 | ||
2504 | activate_task(rq, p, en_flags); | |
2505 | } | |
2506 | ||
2507 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2508 | int wake_flags, bool success) | |
2509 | { | |
2510 | trace_sched_wakeup(p, success); | |
2511 | check_preempt_curr(rq, p, wake_flags); | |
2512 | ||
2513 | p->state = TASK_RUNNING; | |
2514 | #ifdef CONFIG_SMP | |
2515 | if (p->sched_class->task_woken) | |
2516 | p->sched_class->task_woken(rq, p); | |
2517 | ||
2518 | if (unlikely(rq->idle_stamp)) { | |
2519 | u64 delta = rq->clock - rq->idle_stamp; | |
2520 | u64 max = 2*sysctl_sched_migration_cost; | |
2521 | ||
2522 | if (delta > max) | |
2523 | rq->avg_idle = max; | |
2524 | else | |
2525 | update_avg(&rq->avg_idle, delta); | |
2526 | rq->idle_stamp = 0; | |
2527 | } | |
2528 | #endif | |
21aa9af0 TH |
2529 | /* if a worker is waking up, notify workqueue */ |
2530 | if ((p->flags & PF_WQ_WORKER) && success) | |
2531 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2532 | } |
2533 | ||
2534 | /** | |
1da177e4 | 2535 | * try_to_wake_up - wake up a thread |
9ed3811a | 2536 | * @p: the thread to be awakened |
1da177e4 | 2537 | * @state: the mask of task states that can be woken |
9ed3811a | 2538 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2539 | * |
2540 | * Put it on the run-queue if it's not already there. The "current" | |
2541 | * thread is always on the run-queue (except when the actual | |
2542 | * re-schedule is in progress), and as such you're allowed to do | |
2543 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2544 | * runnable without the overhead of this. | |
2545 | * | |
9ed3811a TH |
2546 | * Returns %true if @p was woken up, %false if it was already running |
2547 | * or @state didn't match @p's state. | |
1da177e4 | 2548 | */ |
7d478721 PZ |
2549 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2550 | int wake_flags) | |
1da177e4 | 2551 | { |
cc367732 | 2552 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2553 | unsigned long flags; |
371fd7e7 | 2554 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2555 | struct rq *rq; |
1da177e4 | 2556 | |
e9c84311 | 2557 | this_cpu = get_cpu(); |
2398f2c6 | 2558 | |
04e2f174 | 2559 | smp_wmb(); |
ab3b3aa5 | 2560 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2561 | if (!(p->state & state)) |
1da177e4 LT |
2562 | goto out; |
2563 | ||
dd41f596 | 2564 | if (p->se.on_rq) |
1da177e4 LT |
2565 | goto out_running; |
2566 | ||
2567 | cpu = task_cpu(p); | |
cc367732 | 2568 | orig_cpu = cpu; |
1da177e4 LT |
2569 | |
2570 | #ifdef CONFIG_SMP | |
2571 | if (unlikely(task_running(rq, p))) | |
2572 | goto out_activate; | |
2573 | ||
e9c84311 PZ |
2574 | /* |
2575 | * In order to handle concurrent wakeups and release the rq->lock | |
2576 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2577 | * |
2578 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2579 | */ |
cc87f76a PZ |
2580 | if (task_contributes_to_load(p)) { |
2581 | if (likely(cpu_online(orig_cpu))) | |
2582 | rq->nr_uninterruptible--; | |
2583 | else | |
2584 | this_rq()->nr_uninterruptible--; | |
2585 | } | |
e9c84311 | 2586 | p->state = TASK_WAKING; |
efbbd05a | 2587 | |
371fd7e7 | 2588 | if (p->sched_class->task_waking) { |
efbbd05a | 2589 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2590 | en_flags |= ENQUEUE_WAKING; |
2591 | } | |
efbbd05a | 2592 | |
0017d735 PZ |
2593 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2594 | if (cpu != orig_cpu) | |
5d2f5a61 | 2595 | set_task_cpu(p, cpu); |
0017d735 | 2596 | __task_rq_unlock(rq); |
ab19cb23 | 2597 | |
0970d299 PZ |
2598 | rq = cpu_rq(cpu); |
2599 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2600 | |
0970d299 PZ |
2601 | /* |
2602 | * We migrated the task without holding either rq->lock, however | |
2603 | * since the task is not on the task list itself, nobody else | |
2604 | * will try and migrate the task, hence the rq should match the | |
2605 | * cpu we just moved it to. | |
2606 | */ | |
2607 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2608 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2609 | |
e7693a36 GH |
2610 | #ifdef CONFIG_SCHEDSTATS |
2611 | schedstat_inc(rq, ttwu_count); | |
2612 | if (cpu == this_cpu) | |
2613 | schedstat_inc(rq, ttwu_local); | |
2614 | else { | |
2615 | struct sched_domain *sd; | |
2616 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2617 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2618 | schedstat_inc(sd, ttwu_wake_remote); |
2619 | break; | |
2620 | } | |
2621 | } | |
2622 | } | |
6d6bc0ad | 2623 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2624 | |
1da177e4 LT |
2625 | out_activate: |
2626 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2627 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2628 | cpu == this_cpu, en_flags); | |
1da177e4 | 2629 | success = 1; |
1da177e4 | 2630 | out_running: |
9ed3811a | 2631 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2632 | out: |
2633 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2634 | put_cpu(); |
1da177e4 LT |
2635 | |
2636 | return success; | |
2637 | } | |
2638 | ||
21aa9af0 TH |
2639 | /** |
2640 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2641 | * @p: the thread to be awakened | |
2642 | * | |
2643 | * Put @p on the run-queue if it's not alredy there. The caller must | |
2644 | * ensure that this_rq() is locked, @p is bound to this_rq() and not | |
2645 | * the current task. this_rq() stays locked over invocation. | |
2646 | */ | |
2647 | static void try_to_wake_up_local(struct task_struct *p) | |
2648 | { | |
2649 | struct rq *rq = task_rq(p); | |
2650 | bool success = false; | |
2651 | ||
2652 | BUG_ON(rq != this_rq()); | |
2653 | BUG_ON(p == current); | |
2654 | lockdep_assert_held(&rq->lock); | |
2655 | ||
2656 | if (!(p->state & TASK_NORMAL)) | |
2657 | return; | |
2658 | ||
2659 | if (!p->se.on_rq) { | |
2660 | if (likely(!task_running(rq, p))) { | |
2661 | schedstat_inc(rq, ttwu_count); | |
2662 | schedstat_inc(rq, ttwu_local); | |
2663 | } | |
2664 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2665 | success = true; | |
2666 | } | |
2667 | ttwu_post_activation(p, rq, 0, success); | |
2668 | } | |
2669 | ||
50fa610a DH |
2670 | /** |
2671 | * wake_up_process - Wake up a specific process | |
2672 | * @p: The process to be woken up. | |
2673 | * | |
2674 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2675 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2676 | * running. | |
2677 | * | |
2678 | * It may be assumed that this function implies a write memory barrier before | |
2679 | * changing the task state if and only if any tasks are woken up. | |
2680 | */ | |
7ad5b3a5 | 2681 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2682 | { |
d9514f6c | 2683 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2684 | } |
1da177e4 LT |
2685 | EXPORT_SYMBOL(wake_up_process); |
2686 | ||
7ad5b3a5 | 2687 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2688 | { |
2689 | return try_to_wake_up(p, state, 0); | |
2690 | } | |
2691 | ||
1da177e4 LT |
2692 | /* |
2693 | * Perform scheduler related setup for a newly forked process p. | |
2694 | * p is forked by current. | |
dd41f596 IM |
2695 | * |
2696 | * __sched_fork() is basic setup used by init_idle() too: | |
2697 | */ | |
2698 | static void __sched_fork(struct task_struct *p) | |
2699 | { | |
dd41f596 IM |
2700 | p->se.exec_start = 0; |
2701 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2702 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2703 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2704 | |
2705 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2706 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2707 | #endif |
476d139c | 2708 | |
fa717060 | 2709 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2710 | p->se.on_rq = 0; |
4a55bd5e | 2711 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2712 | |
e107be36 AK |
2713 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2714 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2715 | #endif | |
dd41f596 IM |
2716 | } |
2717 | ||
2718 | /* | |
2719 | * fork()/clone()-time setup: | |
2720 | */ | |
2721 | void sched_fork(struct task_struct *p, int clone_flags) | |
2722 | { | |
2723 | int cpu = get_cpu(); | |
2724 | ||
2725 | __sched_fork(p); | |
06b83b5f | 2726 | /* |
0017d735 | 2727 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2728 | * nobody will actually run it, and a signal or other external |
2729 | * event cannot wake it up and insert it on the runqueue either. | |
2730 | */ | |
0017d735 | 2731 | p->state = TASK_RUNNING; |
dd41f596 | 2732 | |
b9dc29e7 MG |
2733 | /* |
2734 | * Revert to default priority/policy on fork if requested. | |
2735 | */ | |
2736 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2737 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2738 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2739 | p->normal_prio = p->static_prio; |
2740 | } | |
b9dc29e7 | 2741 | |
6c697bdf MG |
2742 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2743 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2744 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2745 | set_load_weight(p); |
2746 | } | |
2747 | ||
b9dc29e7 MG |
2748 | /* |
2749 | * We don't need the reset flag anymore after the fork. It has | |
2750 | * fulfilled its duty: | |
2751 | */ | |
2752 | p->sched_reset_on_fork = 0; | |
2753 | } | |
ca94c442 | 2754 | |
f83f9ac2 PW |
2755 | /* |
2756 | * Make sure we do not leak PI boosting priority to the child. | |
2757 | */ | |
2758 | p->prio = current->normal_prio; | |
2759 | ||
2ddbf952 HS |
2760 | if (!rt_prio(p->prio)) |
2761 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2762 | |
cd29fe6f PZ |
2763 | if (p->sched_class->task_fork) |
2764 | p->sched_class->task_fork(p); | |
2765 | ||
86951599 PZ |
2766 | /* |
2767 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2768 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2769 | * is ran before sched_fork(). | |
2770 | * | |
2771 | * Silence PROVE_RCU. | |
2772 | */ | |
2773 | rcu_read_lock(); | |
5f3edc1b | 2774 | set_task_cpu(p, cpu); |
86951599 | 2775 | rcu_read_unlock(); |
5f3edc1b | 2776 | |
52f17b6c | 2777 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2778 | if (likely(sched_info_on())) |
52f17b6c | 2779 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2780 | #endif |
d6077cb8 | 2781 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2782 | p->oncpu = 0; |
2783 | #endif | |
1da177e4 | 2784 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2785 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2786 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2787 | #endif |
917b627d GH |
2788 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2789 | ||
476d139c | 2790 | put_cpu(); |
1da177e4 LT |
2791 | } |
2792 | ||
2793 | /* | |
2794 | * wake_up_new_task - wake up a newly created task for the first time. | |
2795 | * | |
2796 | * This function will do some initial scheduler statistics housekeeping | |
2797 | * that must be done for every newly created context, then puts the task | |
2798 | * on the runqueue and wakes it. | |
2799 | */ | |
7ad5b3a5 | 2800 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2801 | { |
2802 | unsigned long flags; | |
dd41f596 | 2803 | struct rq *rq; |
c890692b | 2804 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2805 | |
2806 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2807 | rq = task_rq_lock(p, &flags); |
2808 | p->state = TASK_WAKING; | |
2809 | ||
fabf318e PZ |
2810 | /* |
2811 | * Fork balancing, do it here and not earlier because: | |
2812 | * - cpus_allowed can change in the fork path | |
2813 | * - any previously selected cpu might disappear through hotplug | |
2814 | * | |
0017d735 PZ |
2815 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2816 | * without people poking at ->cpus_allowed. | |
fabf318e | 2817 | */ |
0017d735 | 2818 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2819 | set_task_cpu(p, cpu); |
1da177e4 | 2820 | |
06b83b5f | 2821 | p->state = TASK_RUNNING; |
0017d735 PZ |
2822 | task_rq_unlock(rq, &flags); |
2823 | #endif | |
2824 | ||
2825 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2826 | activate_task(rq, p, 0); |
27a9da65 | 2827 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2828 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2829 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2830 | if (p->sched_class->task_woken) |
2831 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2832 | #endif |
dd41f596 | 2833 | task_rq_unlock(rq, &flags); |
fabf318e | 2834 | put_cpu(); |
1da177e4 LT |
2835 | } |
2836 | ||
e107be36 AK |
2837 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2838 | ||
2839 | /** | |
80dd99b3 | 2840 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2841 | * @notifier: notifier struct to register |
e107be36 AK |
2842 | */ |
2843 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2844 | { | |
2845 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2846 | } | |
2847 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2848 | ||
2849 | /** | |
2850 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2851 | * @notifier: notifier struct to unregister |
e107be36 AK |
2852 | * |
2853 | * This is safe to call from within a preemption notifier. | |
2854 | */ | |
2855 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2856 | { | |
2857 | hlist_del(¬ifier->link); | |
2858 | } | |
2859 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2860 | ||
2861 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2862 | { | |
2863 | struct preempt_notifier *notifier; | |
2864 | struct hlist_node *node; | |
2865 | ||
2866 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2867 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2868 | } | |
2869 | ||
2870 | static void | |
2871 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2872 | struct task_struct *next) | |
2873 | { | |
2874 | struct preempt_notifier *notifier; | |
2875 | struct hlist_node *node; | |
2876 | ||
2877 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2878 | notifier->ops->sched_out(notifier, next); | |
2879 | } | |
2880 | ||
6d6bc0ad | 2881 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2882 | |
2883 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2884 | { | |
2885 | } | |
2886 | ||
2887 | static void | |
2888 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2889 | struct task_struct *next) | |
2890 | { | |
2891 | } | |
2892 | ||
6d6bc0ad | 2893 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2894 | |
4866cde0 NP |
2895 | /** |
2896 | * prepare_task_switch - prepare to switch tasks | |
2897 | * @rq: the runqueue preparing to switch | |
421cee29 | 2898 | * @prev: the current task that is being switched out |
4866cde0 NP |
2899 | * @next: the task we are going to switch to. |
2900 | * | |
2901 | * This is called with the rq lock held and interrupts off. It must | |
2902 | * be paired with a subsequent finish_task_switch after the context | |
2903 | * switch. | |
2904 | * | |
2905 | * prepare_task_switch sets up locking and calls architecture specific | |
2906 | * hooks. | |
2907 | */ | |
e107be36 AK |
2908 | static inline void |
2909 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2910 | struct task_struct *next) | |
4866cde0 | 2911 | { |
e107be36 | 2912 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2913 | prepare_lock_switch(rq, next); |
2914 | prepare_arch_switch(next); | |
2915 | } | |
2916 | ||
1da177e4 LT |
2917 | /** |
2918 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2919 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2920 | * @prev: the thread we just switched away from. |
2921 | * | |
4866cde0 NP |
2922 | * finish_task_switch must be called after the context switch, paired |
2923 | * with a prepare_task_switch call before the context switch. | |
2924 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2925 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2926 | * |
2927 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2928 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2929 | * with the lock held can cause deadlocks; see schedule() for |
2930 | * details.) | |
2931 | */ | |
a9957449 | 2932 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2933 | __releases(rq->lock) |
2934 | { | |
1da177e4 | 2935 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2936 | long prev_state; |
1da177e4 LT |
2937 | |
2938 | rq->prev_mm = NULL; | |
2939 | ||
2940 | /* | |
2941 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2942 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2943 | * schedule one last time. The schedule call will never return, and |
2944 | * the scheduled task must drop that reference. | |
c394cc9f | 2945 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2946 | * still held, otherwise prev could be scheduled on another cpu, die |
2947 | * there before we look at prev->state, and then the reference would | |
2948 | * be dropped twice. | |
2949 | * Manfred Spraul <manfred@colorfullife.com> | |
2950 | */ | |
55a101f8 | 2951 | prev_state = prev->state; |
4866cde0 | 2952 | finish_arch_switch(prev); |
8381f65d JI |
2953 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2954 | local_irq_disable(); | |
2955 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2956 | perf_event_task_sched_in(current); |
8381f65d JI |
2957 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2958 | local_irq_enable(); | |
2959 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2960 | finish_lock_switch(rq, prev); |
e8fa1362 | 2961 | |
e107be36 | 2962 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2963 | if (mm) |
2964 | mmdrop(mm); | |
c394cc9f | 2965 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2966 | /* |
2967 | * Remove function-return probe instances associated with this | |
2968 | * task and put them back on the free list. | |
9761eea8 | 2969 | */ |
c6fd91f0 | 2970 | kprobe_flush_task(prev); |
1da177e4 | 2971 | put_task_struct(prev); |
c6fd91f0 | 2972 | } |
1da177e4 LT |
2973 | } |
2974 | ||
3f029d3c GH |
2975 | #ifdef CONFIG_SMP |
2976 | ||
2977 | /* assumes rq->lock is held */ | |
2978 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2979 | { | |
2980 | if (prev->sched_class->pre_schedule) | |
2981 | prev->sched_class->pre_schedule(rq, prev); | |
2982 | } | |
2983 | ||
2984 | /* rq->lock is NOT held, but preemption is disabled */ | |
2985 | static inline void post_schedule(struct rq *rq) | |
2986 | { | |
2987 | if (rq->post_schedule) { | |
2988 | unsigned long flags; | |
2989 | ||
05fa785c | 2990 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2991 | if (rq->curr->sched_class->post_schedule) |
2992 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2993 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2994 | |
2995 | rq->post_schedule = 0; | |
2996 | } | |
2997 | } | |
2998 | ||
2999 | #else | |
da19ab51 | 3000 | |
3f029d3c GH |
3001 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
3002 | { | |
3003 | } | |
3004 | ||
3005 | static inline void post_schedule(struct rq *rq) | |
3006 | { | |
1da177e4 LT |
3007 | } |
3008 | ||
3f029d3c GH |
3009 | #endif |
3010 | ||
1da177e4 LT |
3011 | /** |
3012 | * schedule_tail - first thing a freshly forked thread must call. | |
3013 | * @prev: the thread we just switched away from. | |
3014 | */ | |
36c8b586 | 3015 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
3016 | __releases(rq->lock) |
3017 | { | |
70b97a7f IM |
3018 | struct rq *rq = this_rq(); |
3019 | ||
4866cde0 | 3020 | finish_task_switch(rq, prev); |
da19ab51 | 3021 | |
3f029d3c GH |
3022 | /* |
3023 | * FIXME: do we need to worry about rq being invalidated by the | |
3024 | * task_switch? | |
3025 | */ | |
3026 | post_schedule(rq); | |
70b97a7f | 3027 | |
4866cde0 NP |
3028 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
3029 | /* In this case, finish_task_switch does not reenable preemption */ | |
3030 | preempt_enable(); | |
3031 | #endif | |
1da177e4 | 3032 | if (current->set_child_tid) |
b488893a | 3033 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
3034 | } |
3035 | ||
3036 | /* | |
3037 | * context_switch - switch to the new MM and the new | |
3038 | * thread's register state. | |
3039 | */ | |
dd41f596 | 3040 | static inline void |
70b97a7f | 3041 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 3042 | struct task_struct *next) |
1da177e4 | 3043 | { |
dd41f596 | 3044 | struct mm_struct *mm, *oldmm; |
1da177e4 | 3045 | |
e107be36 | 3046 | prepare_task_switch(rq, prev, next); |
27a9da65 | 3047 | trace_sched_switch(prev, next); |
dd41f596 IM |
3048 | mm = next->mm; |
3049 | oldmm = prev->active_mm; | |
9226d125 ZA |
3050 | /* |
3051 | * For paravirt, this is coupled with an exit in switch_to to | |
3052 | * combine the page table reload and the switch backend into | |
3053 | * one hypercall. | |
3054 | */ | |
224101ed | 3055 | arch_start_context_switch(prev); |
9226d125 | 3056 | |
31915ab4 | 3057 | if (!mm) { |
1da177e4 LT |
3058 | next->active_mm = oldmm; |
3059 | atomic_inc(&oldmm->mm_count); | |
3060 | enter_lazy_tlb(oldmm, next); | |
3061 | } else | |
3062 | switch_mm(oldmm, mm, next); | |
3063 | ||
31915ab4 | 3064 | if (!prev->mm) { |
1da177e4 | 3065 | prev->active_mm = NULL; |
1da177e4 LT |
3066 | rq->prev_mm = oldmm; |
3067 | } | |
3a5f5e48 IM |
3068 | /* |
3069 | * Since the runqueue lock will be released by the next | |
3070 | * task (which is an invalid locking op but in the case | |
3071 | * of the scheduler it's an obvious special-case), so we | |
3072 | * do an early lockdep release here: | |
3073 | */ | |
3074 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 3075 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 3076 | #endif |
1da177e4 LT |
3077 | |
3078 | /* Here we just switch the register state and the stack. */ | |
3079 | switch_to(prev, next, prev); | |
3080 | ||
dd41f596 IM |
3081 | barrier(); |
3082 | /* | |
3083 | * this_rq must be evaluated again because prev may have moved | |
3084 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
3085 | * frame will be invalid. | |
3086 | */ | |
3087 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
3088 | } |
3089 | ||
3090 | /* | |
3091 | * nr_running, nr_uninterruptible and nr_context_switches: | |
3092 | * | |
3093 | * externally visible scheduler statistics: current number of runnable | |
3094 | * threads, current number of uninterruptible-sleeping threads, total | |
3095 | * number of context switches performed since bootup. | |
3096 | */ | |
3097 | unsigned long nr_running(void) | |
3098 | { | |
3099 | unsigned long i, sum = 0; | |
3100 | ||
3101 | for_each_online_cpu(i) | |
3102 | sum += cpu_rq(i)->nr_running; | |
3103 | ||
3104 | return sum; | |
f711f609 | 3105 | } |
1da177e4 LT |
3106 | |
3107 | unsigned long nr_uninterruptible(void) | |
f711f609 | 3108 | { |
1da177e4 | 3109 | unsigned long i, sum = 0; |
f711f609 | 3110 | |
0a945022 | 3111 | for_each_possible_cpu(i) |
1da177e4 | 3112 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
3113 | |
3114 | /* | |
1da177e4 LT |
3115 | * Since we read the counters lockless, it might be slightly |
3116 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 3117 | */ |
1da177e4 LT |
3118 | if (unlikely((long)sum < 0)) |
3119 | sum = 0; | |
f711f609 | 3120 | |
1da177e4 | 3121 | return sum; |
f711f609 | 3122 | } |
f711f609 | 3123 | |
1da177e4 | 3124 | unsigned long long nr_context_switches(void) |
46cb4b7c | 3125 | { |
cc94abfc SR |
3126 | int i; |
3127 | unsigned long long sum = 0; | |
46cb4b7c | 3128 | |
0a945022 | 3129 | for_each_possible_cpu(i) |
1da177e4 | 3130 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 3131 | |
1da177e4 LT |
3132 | return sum; |
3133 | } | |
483b4ee6 | 3134 | |
1da177e4 LT |
3135 | unsigned long nr_iowait(void) |
3136 | { | |
3137 | unsigned long i, sum = 0; | |
483b4ee6 | 3138 | |
0a945022 | 3139 | for_each_possible_cpu(i) |
1da177e4 | 3140 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3141 | |
1da177e4 LT |
3142 | return sum; |
3143 | } | |
483b4ee6 | 3144 | |
8c215bd3 | 3145 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3146 | { |
8c215bd3 | 3147 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3148 | return atomic_read(&this->nr_iowait); |
3149 | } | |
46cb4b7c | 3150 | |
69d25870 AV |
3151 | unsigned long this_cpu_load(void) |
3152 | { | |
3153 | struct rq *this = this_rq(); | |
3154 | return this->cpu_load[0]; | |
3155 | } | |
e790fb0b | 3156 | |
46cb4b7c | 3157 | |
dce48a84 TG |
3158 | /* Variables and functions for calc_load */ |
3159 | static atomic_long_t calc_load_tasks; | |
3160 | static unsigned long calc_load_update; | |
3161 | unsigned long avenrun[3]; | |
3162 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3163 | |
74f5187a PZ |
3164 | static long calc_load_fold_active(struct rq *this_rq) |
3165 | { | |
3166 | long nr_active, delta = 0; | |
3167 | ||
3168 | nr_active = this_rq->nr_running; | |
3169 | nr_active += (long) this_rq->nr_uninterruptible; | |
3170 | ||
3171 | if (nr_active != this_rq->calc_load_active) { | |
3172 | delta = nr_active - this_rq->calc_load_active; | |
3173 | this_rq->calc_load_active = nr_active; | |
3174 | } | |
3175 | ||
3176 | return delta; | |
3177 | } | |
3178 | ||
0f004f5a PZ |
3179 | static unsigned long |
3180 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3181 | { | |
3182 | load *= exp; | |
3183 | load += active * (FIXED_1 - exp); | |
3184 | load += 1UL << (FSHIFT - 1); | |
3185 | return load >> FSHIFT; | |
3186 | } | |
3187 | ||
74f5187a PZ |
3188 | #ifdef CONFIG_NO_HZ |
3189 | /* | |
3190 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3191 | * | |
3192 | * When making the ILB scale, we should try to pull this in as well. | |
3193 | */ | |
3194 | static atomic_long_t calc_load_tasks_idle; | |
3195 | ||
3196 | static void calc_load_account_idle(struct rq *this_rq) | |
3197 | { | |
3198 | long delta; | |
3199 | ||
3200 | delta = calc_load_fold_active(this_rq); | |
3201 | if (delta) | |
3202 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3203 | } | |
3204 | ||
3205 | static long calc_load_fold_idle(void) | |
3206 | { | |
3207 | long delta = 0; | |
3208 | ||
3209 | /* | |
3210 | * Its got a race, we don't care... | |
3211 | */ | |
3212 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3213 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3214 | ||
3215 | return delta; | |
3216 | } | |
0f004f5a PZ |
3217 | |
3218 | /** | |
3219 | * fixed_power_int - compute: x^n, in O(log n) time | |
3220 | * | |
3221 | * @x: base of the power | |
3222 | * @frac_bits: fractional bits of @x | |
3223 | * @n: power to raise @x to. | |
3224 | * | |
3225 | * By exploiting the relation between the definition of the natural power | |
3226 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3227 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3228 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3229 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3230 | * of course trivially computable in O(log_2 n), the length of our binary | |
3231 | * vector. | |
3232 | */ | |
3233 | static unsigned long | |
3234 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3235 | { | |
3236 | unsigned long result = 1UL << frac_bits; | |
3237 | ||
3238 | if (n) for (;;) { | |
3239 | if (n & 1) { | |
3240 | result *= x; | |
3241 | result += 1UL << (frac_bits - 1); | |
3242 | result >>= frac_bits; | |
3243 | } | |
3244 | n >>= 1; | |
3245 | if (!n) | |
3246 | break; | |
3247 | x *= x; | |
3248 | x += 1UL << (frac_bits - 1); | |
3249 | x >>= frac_bits; | |
3250 | } | |
3251 | ||
3252 | return result; | |
3253 | } | |
3254 | ||
3255 | /* | |
3256 | * a1 = a0 * e + a * (1 - e) | |
3257 | * | |
3258 | * a2 = a1 * e + a * (1 - e) | |
3259 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3260 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3261 | * | |
3262 | * a3 = a2 * e + a * (1 - e) | |
3263 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3264 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3265 | * | |
3266 | * ... | |
3267 | * | |
3268 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3269 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3270 | * = a0 * e^n + a * (1 - e^n) | |
3271 | * | |
3272 | * [1] application of the geometric series: | |
3273 | * | |
3274 | * n 1 - x^(n+1) | |
3275 | * S_n := \Sum x^i = ------------- | |
3276 | * i=0 1 - x | |
3277 | */ | |
3278 | static unsigned long | |
3279 | calc_load_n(unsigned long load, unsigned long exp, | |
3280 | unsigned long active, unsigned int n) | |
3281 | { | |
3282 | ||
3283 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3284 | } | |
3285 | ||
3286 | /* | |
3287 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3288 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3289 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3290 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3291 | * | |
3292 | * Once we've updated the global active value, we need to apply the exponential | |
3293 | * weights adjusted to the number of cycles missed. | |
3294 | */ | |
3295 | static void calc_global_nohz(unsigned long ticks) | |
3296 | { | |
3297 | long delta, active, n; | |
3298 | ||
3299 | if (time_before(jiffies, calc_load_update)) | |
3300 | return; | |
3301 | ||
3302 | /* | |
3303 | * If we crossed a calc_load_update boundary, make sure to fold | |
3304 | * any pending idle changes, the respective CPUs might have | |
3305 | * missed the tick driven calc_load_account_active() update | |
3306 | * due to NO_HZ. | |
3307 | */ | |
3308 | delta = calc_load_fold_idle(); | |
3309 | if (delta) | |
3310 | atomic_long_add(delta, &calc_load_tasks); | |
3311 | ||
3312 | /* | |
3313 | * If we were idle for multiple load cycles, apply them. | |
3314 | */ | |
3315 | if (ticks >= LOAD_FREQ) { | |
3316 | n = ticks / LOAD_FREQ; | |
3317 | ||
3318 | active = atomic_long_read(&calc_load_tasks); | |
3319 | active = active > 0 ? active * FIXED_1 : 0; | |
3320 | ||
3321 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3322 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3323 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3324 | ||
3325 | calc_load_update += n * LOAD_FREQ; | |
3326 | } | |
3327 | ||
3328 | /* | |
3329 | * Its possible the remainder of the above division also crosses | |
3330 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3331 | * which comes after this will take care of that. | |
3332 | * | |
3333 | * Consider us being 11 ticks before a cycle completion, and us | |
3334 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3335 | * age us 4 cycles, and the test in calc_global_load() will | |
3336 | * pick up the final one. | |
3337 | */ | |
3338 | } | |
74f5187a PZ |
3339 | #else |
3340 | static void calc_load_account_idle(struct rq *this_rq) | |
3341 | { | |
3342 | } | |
3343 | ||
3344 | static inline long calc_load_fold_idle(void) | |
3345 | { | |
3346 | return 0; | |
3347 | } | |
0f004f5a PZ |
3348 | |
3349 | static void calc_global_nohz(unsigned long ticks) | |
3350 | { | |
3351 | } | |
74f5187a PZ |
3352 | #endif |
3353 | ||
2d02494f TG |
3354 | /** |
3355 | * get_avenrun - get the load average array | |
3356 | * @loads: pointer to dest load array | |
3357 | * @offset: offset to add | |
3358 | * @shift: shift count to shift the result left | |
3359 | * | |
3360 | * These values are estimates at best, so no need for locking. | |
3361 | */ | |
3362 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3363 | { | |
3364 | loads[0] = (avenrun[0] + offset) << shift; | |
3365 | loads[1] = (avenrun[1] + offset) << shift; | |
3366 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3367 | } |
46cb4b7c | 3368 | |
46cb4b7c | 3369 | /* |
dce48a84 TG |
3370 | * calc_load - update the avenrun load estimates 10 ticks after the |
3371 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3372 | */ |
0f004f5a | 3373 | void calc_global_load(unsigned long ticks) |
7835b98b | 3374 | { |
dce48a84 | 3375 | long active; |
1da177e4 | 3376 | |
0f004f5a PZ |
3377 | calc_global_nohz(ticks); |
3378 | ||
3379 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3380 | return; |
1da177e4 | 3381 | |
dce48a84 TG |
3382 | active = atomic_long_read(&calc_load_tasks); |
3383 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3384 | |
dce48a84 TG |
3385 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3386 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3387 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3388 | |
dce48a84 TG |
3389 | calc_load_update += LOAD_FREQ; |
3390 | } | |
1da177e4 | 3391 | |
dce48a84 | 3392 | /* |
74f5187a PZ |
3393 | * Called from update_cpu_load() to periodically update this CPU's |
3394 | * active count. | |
dce48a84 TG |
3395 | */ |
3396 | static void calc_load_account_active(struct rq *this_rq) | |
3397 | { | |
74f5187a | 3398 | long delta; |
08c183f3 | 3399 | |
74f5187a PZ |
3400 | if (time_before(jiffies, this_rq->calc_load_update)) |
3401 | return; | |
783609c6 | 3402 | |
74f5187a PZ |
3403 | delta = calc_load_fold_active(this_rq); |
3404 | delta += calc_load_fold_idle(); | |
3405 | if (delta) | |
dce48a84 | 3406 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3407 | |
3408 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3409 | } |
3410 | ||
fdf3e95d VP |
3411 | /* |
3412 | * The exact cpuload at various idx values, calculated at every tick would be | |
3413 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3414 | * | |
3415 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3416 | * on nth tick when cpu may be busy, then we have: | |
3417 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3418 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3419 | * | |
3420 | * decay_load_missed() below does efficient calculation of | |
3421 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3422 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3423 | * | |
3424 | * The calculation is approximated on a 128 point scale. | |
3425 | * degrade_zero_ticks is the number of ticks after which load at any | |
3426 | * particular idx is approximated to be zero. | |
3427 | * degrade_factor is a precomputed table, a row for each load idx. | |
3428 | * Each column corresponds to degradation factor for a power of two ticks, | |
3429 | * based on 128 point scale. | |
3430 | * Example: | |
3431 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3432 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3433 | * | |
3434 | * With this power of 2 load factors, we can degrade the load n times | |
3435 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3436 | * n mult/shifts needed by the exact degradation. | |
3437 | */ | |
3438 | #define DEGRADE_SHIFT 7 | |
3439 | static const unsigned char | |
3440 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3441 | static const unsigned char | |
3442 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3443 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3444 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3445 | {96, 72, 40, 12, 1, 0, 0}, | |
3446 | {112, 98, 75, 43, 15, 1, 0}, | |
3447 | {120, 112, 98, 76, 45, 16, 2} }; | |
3448 | ||
3449 | /* | |
3450 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3451 | * would be when CPU is idle and so we just decay the old load without | |
3452 | * adding any new load. | |
3453 | */ | |
3454 | static unsigned long | |
3455 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3456 | { | |
3457 | int j = 0; | |
3458 | ||
3459 | if (!missed_updates) | |
3460 | return load; | |
3461 | ||
3462 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3463 | return 0; | |
3464 | ||
3465 | if (idx == 1) | |
3466 | return load >> missed_updates; | |
3467 | ||
3468 | while (missed_updates) { | |
3469 | if (missed_updates % 2) | |
3470 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3471 | ||
3472 | missed_updates >>= 1; | |
3473 | j++; | |
3474 | } | |
3475 | return load; | |
3476 | } | |
3477 | ||
46cb4b7c | 3478 | /* |
dd41f596 | 3479 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3480 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3481 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3482 | */ |
dd41f596 | 3483 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3484 | { |
495eca49 | 3485 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3486 | unsigned long curr_jiffies = jiffies; |
3487 | unsigned long pending_updates; | |
dd41f596 | 3488 | int i, scale; |
46cb4b7c | 3489 | |
dd41f596 | 3490 | this_rq->nr_load_updates++; |
46cb4b7c | 3491 | |
fdf3e95d VP |
3492 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3493 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3494 | return; | |
3495 | ||
3496 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3497 | this_rq->last_load_update_tick = curr_jiffies; | |
3498 | ||
dd41f596 | 3499 | /* Update our load: */ |
fdf3e95d VP |
3500 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3501 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3502 | unsigned long old_load, new_load; |
7d1e6a9b | 3503 | |
dd41f596 | 3504 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3505 | |
dd41f596 | 3506 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3507 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3508 | new_load = this_load; |
a25707f3 IM |
3509 | /* |
3510 | * Round up the averaging division if load is increasing. This | |
3511 | * prevents us from getting stuck on 9 if the load is 10, for | |
3512 | * example. | |
3513 | */ | |
3514 | if (new_load > old_load) | |
fdf3e95d VP |
3515 | new_load += scale - 1; |
3516 | ||
3517 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3518 | } |
da2b71ed SS |
3519 | |
3520 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3521 | } |
3522 | ||
3523 | static void update_cpu_load_active(struct rq *this_rq) | |
3524 | { | |
3525 | update_cpu_load(this_rq); | |
46cb4b7c | 3526 | |
74f5187a | 3527 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3528 | } |
3529 | ||
dd41f596 | 3530 | #ifdef CONFIG_SMP |
8a0be9ef | 3531 | |
46cb4b7c | 3532 | /* |
38022906 PZ |
3533 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3534 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3535 | */ |
38022906 | 3536 | void sched_exec(void) |
46cb4b7c | 3537 | { |
38022906 | 3538 | struct task_struct *p = current; |
1da177e4 | 3539 | unsigned long flags; |
70b97a7f | 3540 | struct rq *rq; |
0017d735 | 3541 | int dest_cpu; |
46cb4b7c | 3542 | |
1da177e4 | 3543 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3544 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3545 | if (dest_cpu == smp_processor_id()) | |
3546 | goto unlock; | |
38022906 | 3547 | |
46cb4b7c | 3548 | /* |
38022906 | 3549 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3550 | */ |
30da688e | 3551 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
969c7921 TH |
3552 | likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) { |
3553 | struct migration_arg arg = { p, dest_cpu }; | |
46cb4b7c | 3554 | |
1da177e4 | 3555 | task_rq_unlock(rq, &flags); |
969c7921 | 3556 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3557 | return; |
3558 | } | |
0017d735 | 3559 | unlock: |
1da177e4 | 3560 | task_rq_unlock(rq, &flags); |
1da177e4 | 3561 | } |
dd41f596 | 3562 | |
1da177e4 LT |
3563 | #endif |
3564 | ||
1da177e4 LT |
3565 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3566 | ||
3567 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3568 | ||
3569 | /* | |
c5f8d995 | 3570 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3571 | * @p in case that task is currently running. |
c5f8d995 HS |
3572 | * |
3573 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3574 | */ |
c5f8d995 HS |
3575 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3576 | { | |
3577 | u64 ns = 0; | |
3578 | ||
3579 | if (task_current(rq, p)) { | |
3580 | update_rq_clock(rq); | |
305e6835 | 3581 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3582 | if ((s64)ns < 0) |
3583 | ns = 0; | |
3584 | } | |
3585 | ||
3586 | return ns; | |
3587 | } | |
3588 | ||
bb34d92f | 3589 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3590 | { |
1da177e4 | 3591 | unsigned long flags; |
41b86e9c | 3592 | struct rq *rq; |
bb34d92f | 3593 | u64 ns = 0; |
48f24c4d | 3594 | |
41b86e9c | 3595 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3596 | ns = do_task_delta_exec(p, rq); |
3597 | task_rq_unlock(rq, &flags); | |
1508487e | 3598 | |
c5f8d995 HS |
3599 | return ns; |
3600 | } | |
f06febc9 | 3601 | |
c5f8d995 HS |
3602 | /* |
3603 | * Return accounted runtime for the task. | |
3604 | * In case the task is currently running, return the runtime plus current's | |
3605 | * pending runtime that have not been accounted yet. | |
3606 | */ | |
3607 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3608 | { | |
3609 | unsigned long flags; | |
3610 | struct rq *rq; | |
3611 | u64 ns = 0; | |
3612 | ||
3613 | rq = task_rq_lock(p, &flags); | |
3614 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3615 | task_rq_unlock(rq, &flags); | |
3616 | ||
3617 | return ns; | |
3618 | } | |
48f24c4d | 3619 | |
c5f8d995 HS |
3620 | /* |
3621 | * Return sum_exec_runtime for the thread group. | |
3622 | * In case the task is currently running, return the sum plus current's | |
3623 | * pending runtime that have not been accounted yet. | |
3624 | * | |
3625 | * Note that the thread group might have other running tasks as well, | |
3626 | * so the return value not includes other pending runtime that other | |
3627 | * running tasks might have. | |
3628 | */ | |
3629 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3630 | { | |
3631 | struct task_cputime totals; | |
3632 | unsigned long flags; | |
3633 | struct rq *rq; | |
3634 | u64 ns; | |
3635 | ||
3636 | rq = task_rq_lock(p, &flags); | |
3637 | thread_group_cputime(p, &totals); | |
3638 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3639 | task_rq_unlock(rq, &flags); |
48f24c4d | 3640 | |
1da177e4 LT |
3641 | return ns; |
3642 | } | |
3643 | ||
1da177e4 LT |
3644 | /* |
3645 | * Account user cpu time to a process. | |
3646 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3647 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3648 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3649 | */ |
457533a7 MS |
3650 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3651 | cputime_t cputime_scaled) | |
1da177e4 LT |
3652 | { |
3653 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3654 | cputime64_t tmp; | |
3655 | ||
457533a7 | 3656 | /* Add user time to process. */ |
1da177e4 | 3657 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3658 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3659 | account_group_user_time(p, cputime); |
1da177e4 LT |
3660 | |
3661 | /* Add user time to cpustat. */ | |
3662 | tmp = cputime_to_cputime64(cputime); | |
3663 | if (TASK_NICE(p) > 0) | |
3664 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3665 | else | |
3666 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3667 | |
3668 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3669 | /* Account for user time used */ |
3670 | acct_update_integrals(p); | |
1da177e4 LT |
3671 | } |
3672 | ||
94886b84 LV |
3673 | /* |
3674 | * Account guest cpu time to a process. | |
3675 | * @p: the process that the cpu time gets accounted to | |
3676 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3677 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3678 | */ |
457533a7 MS |
3679 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3680 | cputime_t cputime_scaled) | |
94886b84 LV |
3681 | { |
3682 | cputime64_t tmp; | |
3683 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3684 | ||
3685 | tmp = cputime_to_cputime64(cputime); | |
3686 | ||
457533a7 | 3687 | /* Add guest time to process. */ |
94886b84 | 3688 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3689 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3690 | account_group_user_time(p, cputime); |
94886b84 LV |
3691 | p->gtime = cputime_add(p->gtime, cputime); |
3692 | ||
457533a7 | 3693 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3694 | if (TASK_NICE(p) > 0) { |
3695 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3696 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3697 | } else { | |
3698 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3699 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3700 | } | |
94886b84 LV |
3701 | } |
3702 | ||
1da177e4 LT |
3703 | /* |
3704 | * Account system cpu time to a process. | |
3705 | * @p: the process that the cpu time gets accounted to | |
3706 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3707 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3708 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3709 | */ |
3710 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3711 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3712 | { |
3713 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3714 | cputime64_t tmp; |
3715 | ||
983ed7a6 | 3716 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3717 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3718 | return; |
3719 | } | |
94886b84 | 3720 | |
457533a7 | 3721 | /* Add system time to process. */ |
1da177e4 | 3722 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3723 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3724 | account_group_system_time(p, cputime); |
1da177e4 LT |
3725 | |
3726 | /* Add system time to cpustat. */ | |
3727 | tmp = cputime_to_cputime64(cputime); | |
3728 | if (hardirq_count() - hardirq_offset) | |
3729 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
75e1056f | 3730 | else if (in_serving_softirq()) |
1da177e4 | 3731 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); |
1da177e4 | 3732 | else |
79741dd3 MS |
3733 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3734 | ||
ef12fefa BR |
3735 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3736 | ||
1da177e4 LT |
3737 | /* Account for system time used */ |
3738 | acct_update_integrals(p); | |
1da177e4 LT |
3739 | } |
3740 | ||
c66f08be | 3741 | /* |
1da177e4 | 3742 | * Account for involuntary wait time. |
1da177e4 | 3743 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3744 | */ |
79741dd3 | 3745 | void account_steal_time(cputime_t cputime) |
c66f08be | 3746 | { |
79741dd3 MS |
3747 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3748 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3749 | ||
3750 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3751 | } |
3752 | ||
1da177e4 | 3753 | /* |
79741dd3 MS |
3754 | * Account for idle time. |
3755 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3756 | */ |
79741dd3 | 3757 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3758 | { |
3759 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3760 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3761 | struct rq *rq = this_rq(); |
1da177e4 | 3762 | |
79741dd3 MS |
3763 | if (atomic_read(&rq->nr_iowait) > 0) |
3764 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3765 | else | |
3766 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3767 | } |
3768 | ||
79741dd3 MS |
3769 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3770 | ||
3771 | /* | |
3772 | * Account a single tick of cpu time. | |
3773 | * @p: the process that the cpu time gets accounted to | |
3774 | * @user_tick: indicates if the tick is a user or a system tick | |
3775 | */ | |
3776 | void account_process_tick(struct task_struct *p, int user_tick) | |
3777 | { | |
a42548a1 | 3778 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3779 | struct rq *rq = this_rq(); |
3780 | ||
3781 | if (user_tick) | |
a42548a1 | 3782 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3783 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3784 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3785 | one_jiffy_scaled); |
3786 | else | |
a42548a1 | 3787 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3788 | } |
3789 | ||
3790 | /* | |
3791 | * Account multiple ticks of steal time. | |
3792 | * @p: the process from which the cpu time has been stolen | |
3793 | * @ticks: number of stolen ticks | |
3794 | */ | |
3795 | void account_steal_ticks(unsigned long ticks) | |
3796 | { | |
3797 | account_steal_time(jiffies_to_cputime(ticks)); | |
3798 | } | |
3799 | ||
3800 | /* | |
3801 | * Account multiple ticks of idle time. | |
3802 | * @ticks: number of stolen ticks | |
3803 | */ | |
3804 | void account_idle_ticks(unsigned long ticks) | |
3805 | { | |
3806 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3807 | } |
3808 | ||
79741dd3 MS |
3809 | #endif |
3810 | ||
49048622 BS |
3811 | /* |
3812 | * Use precise platform statistics if available: | |
3813 | */ | |
3814 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3815 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3816 | { |
d99ca3b9 HS |
3817 | *ut = p->utime; |
3818 | *st = p->stime; | |
49048622 BS |
3819 | } |
3820 | ||
0cf55e1e | 3821 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3822 | { |
0cf55e1e HS |
3823 | struct task_cputime cputime; |
3824 | ||
3825 | thread_group_cputime(p, &cputime); | |
3826 | ||
3827 | *ut = cputime.utime; | |
3828 | *st = cputime.stime; | |
49048622 BS |
3829 | } |
3830 | #else | |
761b1d26 HS |
3831 | |
3832 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3833 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3834 | #endif |
3835 | ||
d180c5bc | 3836 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3837 | { |
d99ca3b9 | 3838 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3839 | |
3840 | /* | |
3841 | * Use CFS's precise accounting: | |
3842 | */ | |
d180c5bc | 3843 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3844 | |
3845 | if (total) { | |
e75e863d | 3846 | u64 temp = rtime; |
d180c5bc | 3847 | |
e75e863d | 3848 | temp *= utime; |
49048622 | 3849 | do_div(temp, total); |
d180c5bc HS |
3850 | utime = (cputime_t)temp; |
3851 | } else | |
3852 | utime = rtime; | |
49048622 | 3853 | |
d180c5bc HS |
3854 | /* |
3855 | * Compare with previous values, to keep monotonicity: | |
3856 | */ | |
761b1d26 | 3857 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3858 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3859 | |
d99ca3b9 HS |
3860 | *ut = p->prev_utime; |
3861 | *st = p->prev_stime; | |
49048622 BS |
3862 | } |
3863 | ||
0cf55e1e HS |
3864 | /* |
3865 | * Must be called with siglock held. | |
3866 | */ | |
3867 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3868 | { |
0cf55e1e HS |
3869 | struct signal_struct *sig = p->signal; |
3870 | struct task_cputime cputime; | |
3871 | cputime_t rtime, utime, total; | |
49048622 | 3872 | |
0cf55e1e | 3873 | thread_group_cputime(p, &cputime); |
49048622 | 3874 | |
0cf55e1e HS |
3875 | total = cputime_add(cputime.utime, cputime.stime); |
3876 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3877 | |
0cf55e1e | 3878 | if (total) { |
e75e863d | 3879 | u64 temp = rtime; |
49048622 | 3880 | |
e75e863d | 3881 | temp *= cputime.utime; |
0cf55e1e HS |
3882 | do_div(temp, total); |
3883 | utime = (cputime_t)temp; | |
3884 | } else | |
3885 | utime = rtime; | |
3886 | ||
3887 | sig->prev_utime = max(sig->prev_utime, utime); | |
3888 | sig->prev_stime = max(sig->prev_stime, | |
3889 | cputime_sub(rtime, sig->prev_utime)); | |
3890 | ||
3891 | *ut = sig->prev_utime; | |
3892 | *st = sig->prev_stime; | |
49048622 | 3893 | } |
49048622 | 3894 | #endif |
49048622 | 3895 | |
7835b98b CL |
3896 | /* |
3897 | * This function gets called by the timer code, with HZ frequency. | |
3898 | * We call it with interrupts disabled. | |
3899 | * | |
3900 | * It also gets called by the fork code, when changing the parent's | |
3901 | * timeslices. | |
3902 | */ | |
3903 | void scheduler_tick(void) | |
3904 | { | |
7835b98b CL |
3905 | int cpu = smp_processor_id(); |
3906 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3907 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3908 | |
3909 | sched_clock_tick(); | |
dd41f596 | 3910 | |
05fa785c | 3911 | raw_spin_lock(&rq->lock); |
3e51f33f | 3912 | update_rq_clock(rq); |
fdf3e95d | 3913 | update_cpu_load_active(rq); |
fa85ae24 | 3914 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3915 | raw_spin_unlock(&rq->lock); |
7835b98b | 3916 | |
e9d2b064 | 3917 | perf_event_task_tick(); |
e220d2dc | 3918 | |
e418e1c2 | 3919 | #ifdef CONFIG_SMP |
dd41f596 IM |
3920 | rq->idle_at_tick = idle_cpu(cpu); |
3921 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3922 | #endif |
1da177e4 LT |
3923 | } |
3924 | ||
132380a0 | 3925 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3926 | { |
3927 | if (in_lock_functions(addr)) { | |
3928 | addr = CALLER_ADDR2; | |
3929 | if (in_lock_functions(addr)) | |
3930 | addr = CALLER_ADDR3; | |
3931 | } | |
3932 | return addr; | |
3933 | } | |
1da177e4 | 3934 | |
7e49fcce SR |
3935 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3936 | defined(CONFIG_PREEMPT_TRACER)) | |
3937 | ||
43627582 | 3938 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3939 | { |
6cd8a4bb | 3940 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3941 | /* |
3942 | * Underflow? | |
3943 | */ | |
9a11b49a IM |
3944 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3945 | return; | |
6cd8a4bb | 3946 | #endif |
1da177e4 | 3947 | preempt_count() += val; |
6cd8a4bb | 3948 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3949 | /* |
3950 | * Spinlock count overflowing soon? | |
3951 | */ | |
33859f7f MOS |
3952 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3953 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3954 | #endif |
3955 | if (preempt_count() == val) | |
3956 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3957 | } |
3958 | EXPORT_SYMBOL(add_preempt_count); | |
3959 | ||
43627582 | 3960 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3961 | { |
6cd8a4bb | 3962 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3963 | /* |
3964 | * Underflow? | |
3965 | */ | |
01e3eb82 | 3966 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3967 | return; |
1da177e4 LT |
3968 | /* |
3969 | * Is the spinlock portion underflowing? | |
3970 | */ | |
9a11b49a IM |
3971 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3972 | !(preempt_count() & PREEMPT_MASK))) | |
3973 | return; | |
6cd8a4bb | 3974 | #endif |
9a11b49a | 3975 | |
6cd8a4bb SR |
3976 | if (preempt_count() == val) |
3977 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3978 | preempt_count() -= val; |
3979 | } | |
3980 | EXPORT_SYMBOL(sub_preempt_count); | |
3981 | ||
3982 | #endif | |
3983 | ||
3984 | /* | |
dd41f596 | 3985 | * Print scheduling while atomic bug: |
1da177e4 | 3986 | */ |
dd41f596 | 3987 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3988 | { |
838225b4 SS |
3989 | struct pt_regs *regs = get_irq_regs(); |
3990 | ||
3df0fc5b PZ |
3991 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3992 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3993 | |
dd41f596 | 3994 | debug_show_held_locks(prev); |
e21f5b15 | 3995 | print_modules(); |
dd41f596 IM |
3996 | if (irqs_disabled()) |
3997 | print_irqtrace_events(prev); | |
838225b4 SS |
3998 | |
3999 | if (regs) | |
4000 | show_regs(regs); | |
4001 | else | |
4002 | dump_stack(); | |
dd41f596 | 4003 | } |
1da177e4 | 4004 | |
dd41f596 IM |
4005 | /* |
4006 | * Various schedule()-time debugging checks and statistics: | |
4007 | */ | |
4008 | static inline void schedule_debug(struct task_struct *prev) | |
4009 | { | |
1da177e4 | 4010 | /* |
41a2d6cf | 4011 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
4012 | * schedule() atomically, we ignore that path for now. |
4013 | * Otherwise, whine if we are scheduling when we should not be. | |
4014 | */ | |
3f33a7ce | 4015 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
4016 | __schedule_bug(prev); |
4017 | ||
1da177e4 LT |
4018 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
4019 | ||
2d72376b | 4020 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
4021 | #ifdef CONFIG_SCHEDSTATS |
4022 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
4023 | schedstat_inc(this_rq(), bkl_count); |
4024 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
4025 | } |
4026 | #endif | |
dd41f596 IM |
4027 | } |
4028 | ||
6cecd084 | 4029 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 4030 | { |
a64692a3 MG |
4031 | if (prev->se.on_rq) |
4032 | update_rq_clock(rq); | |
6cecd084 | 4033 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
4034 | } |
4035 | ||
dd41f596 IM |
4036 | /* |
4037 | * Pick up the highest-prio task: | |
4038 | */ | |
4039 | static inline struct task_struct * | |
b67802ea | 4040 | pick_next_task(struct rq *rq) |
dd41f596 | 4041 | { |
5522d5d5 | 4042 | const struct sched_class *class; |
dd41f596 | 4043 | struct task_struct *p; |
1da177e4 LT |
4044 | |
4045 | /* | |
dd41f596 IM |
4046 | * Optimization: we know that if all tasks are in |
4047 | * the fair class we can call that function directly: | |
1da177e4 | 4048 | */ |
dd41f596 | 4049 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 4050 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
4051 | if (likely(p)) |
4052 | return p; | |
1da177e4 LT |
4053 | } |
4054 | ||
34f971f6 | 4055 | for_each_class(class) { |
fb8d4724 | 4056 | p = class->pick_next_task(rq); |
dd41f596 IM |
4057 | if (p) |
4058 | return p; | |
dd41f596 | 4059 | } |
34f971f6 PZ |
4060 | |
4061 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 4062 | } |
1da177e4 | 4063 | |
dd41f596 IM |
4064 | /* |
4065 | * schedule() is the main scheduler function. | |
4066 | */ | |
ff743345 | 4067 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
4068 | { |
4069 | struct task_struct *prev, *next; | |
67ca7bde | 4070 | unsigned long *switch_count; |
dd41f596 | 4071 | struct rq *rq; |
31656519 | 4072 | int cpu; |
dd41f596 | 4073 | |
ff743345 PZ |
4074 | need_resched: |
4075 | preempt_disable(); | |
dd41f596 IM |
4076 | cpu = smp_processor_id(); |
4077 | rq = cpu_rq(cpu); | |
25502a6c | 4078 | rcu_note_context_switch(cpu); |
dd41f596 | 4079 | prev = rq->curr; |
dd41f596 IM |
4080 | |
4081 | release_kernel_lock(prev); | |
4082 | need_resched_nonpreemptible: | |
4083 | ||
4084 | schedule_debug(prev); | |
1da177e4 | 4085 | |
31656519 | 4086 | if (sched_feat(HRTICK)) |
f333fdc9 | 4087 | hrtick_clear(rq); |
8f4d37ec | 4088 | |
05fa785c | 4089 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 4090 | |
246d86b5 | 4091 | switch_count = &prev->nivcsw; |
1da177e4 | 4092 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 4093 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 4094 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
4095 | } else { |
4096 | /* | |
4097 | * If a worker is going to sleep, notify and | |
4098 | * ask workqueue whether it wants to wake up a | |
4099 | * task to maintain concurrency. If so, wake | |
4100 | * up the task. | |
4101 | */ | |
4102 | if (prev->flags & PF_WQ_WORKER) { | |
4103 | struct task_struct *to_wakeup; | |
4104 | ||
4105 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
4106 | if (to_wakeup) | |
4107 | try_to_wake_up_local(to_wakeup); | |
4108 | } | |
371fd7e7 | 4109 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 4110 | } |
dd41f596 | 4111 | switch_count = &prev->nvcsw; |
1da177e4 LT |
4112 | } |
4113 | ||
3f029d3c | 4114 | pre_schedule(rq, prev); |
f65eda4f | 4115 | |
dd41f596 | 4116 | if (unlikely(!rq->nr_running)) |
1da177e4 | 4117 | idle_balance(cpu, rq); |
1da177e4 | 4118 | |
df1c99d4 | 4119 | put_prev_task(rq, prev); |
b67802ea | 4120 | next = pick_next_task(rq); |
f26f9aff MG |
4121 | clear_tsk_need_resched(prev); |
4122 | rq->skip_clock_update = 0; | |
1da177e4 | 4123 | |
1da177e4 | 4124 | if (likely(prev != next)) { |
673a90a1 | 4125 | sched_info_switch(prev, next); |
49f47433 | 4126 | perf_event_task_sched_out(prev, next); |
673a90a1 | 4127 | |
1da177e4 LT |
4128 | rq->nr_switches++; |
4129 | rq->curr = next; | |
4130 | ++*switch_count; | |
4131 | ||
dd41f596 | 4132 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 4133 | /* |
246d86b5 ON |
4134 | * The context switch have flipped the stack from under us |
4135 | * and restored the local variables which were saved when | |
4136 | * this task called schedule() in the past. prev == current | |
4137 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4138 | */ |
4139 | cpu = smp_processor_id(); | |
4140 | rq = cpu_rq(cpu); | |
1da177e4 | 4141 | } else |
05fa785c | 4142 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4143 | |
3f029d3c | 4144 | post_schedule(rq); |
1da177e4 | 4145 | |
246d86b5 | 4146 | if (unlikely(reacquire_kernel_lock(prev))) |
1da177e4 | 4147 | goto need_resched_nonpreemptible; |
8f4d37ec | 4148 | |
1da177e4 | 4149 | preempt_enable_no_resched(); |
ff743345 | 4150 | if (need_resched()) |
1da177e4 LT |
4151 | goto need_resched; |
4152 | } | |
1da177e4 LT |
4153 | EXPORT_SYMBOL(schedule); |
4154 | ||
c08f7829 | 4155 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
4156 | /* |
4157 | * Look out! "owner" is an entirely speculative pointer | |
4158 | * access and not reliable. | |
4159 | */ | |
4160 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
4161 | { | |
4162 | unsigned int cpu; | |
4163 | struct rq *rq; | |
4164 | ||
4165 | if (!sched_feat(OWNER_SPIN)) | |
4166 | return 0; | |
4167 | ||
4168 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
4169 | /* | |
4170 | * Need to access the cpu field knowing that | |
4171 | * DEBUG_PAGEALLOC could have unmapped it if | |
4172 | * the mutex owner just released it and exited. | |
4173 | */ | |
4174 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 4175 | return 0; |
0d66bf6d PZ |
4176 | #else |
4177 | cpu = owner->cpu; | |
4178 | #endif | |
4179 | ||
4180 | /* | |
4181 | * Even if the access succeeded (likely case), | |
4182 | * the cpu field may no longer be valid. | |
4183 | */ | |
4184 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 4185 | return 0; |
0d66bf6d PZ |
4186 | |
4187 | /* | |
4188 | * We need to validate that we can do a | |
4189 | * get_cpu() and that we have the percpu area. | |
4190 | */ | |
4191 | if (!cpu_online(cpu)) | |
4b402210 | 4192 | return 0; |
0d66bf6d PZ |
4193 | |
4194 | rq = cpu_rq(cpu); | |
4195 | ||
4196 | for (;;) { | |
4197 | /* | |
4198 | * Owner changed, break to re-assess state. | |
4199 | */ | |
9d0f4dcc TC |
4200 | if (lock->owner != owner) { |
4201 | /* | |
4202 | * If the lock has switched to a different owner, | |
4203 | * we likely have heavy contention. Return 0 to quit | |
4204 | * optimistic spinning and not contend further: | |
4205 | */ | |
4206 | if (lock->owner) | |
4207 | return 0; | |
0d66bf6d | 4208 | break; |
9d0f4dcc | 4209 | } |
0d66bf6d PZ |
4210 | |
4211 | /* | |
4212 | * Is that owner really running on that cpu? | |
4213 | */ | |
4214 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
4215 | return 0; | |
4216 | ||
4217 | cpu_relax(); | |
4218 | } | |
4b402210 | 4219 | |
0d66bf6d PZ |
4220 | return 1; |
4221 | } | |
4222 | #endif | |
4223 | ||
1da177e4 LT |
4224 | #ifdef CONFIG_PREEMPT |
4225 | /* | |
2ed6e34f | 4226 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4227 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4228 | * occur there and call schedule directly. |
4229 | */ | |
d1f74e20 | 4230 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4231 | { |
4232 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4233 | |
1da177e4 LT |
4234 | /* |
4235 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4236 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4237 | */ |
beed33a8 | 4238 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4239 | return; |
4240 | ||
3a5c359a | 4241 | do { |
d1f74e20 | 4242 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4243 | schedule(); |
d1f74e20 | 4244 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4245 | |
3a5c359a AK |
4246 | /* |
4247 | * Check again in case we missed a preemption opportunity | |
4248 | * between schedule and now. | |
4249 | */ | |
4250 | barrier(); | |
5ed0cec0 | 4251 | } while (need_resched()); |
1da177e4 | 4252 | } |
1da177e4 LT |
4253 | EXPORT_SYMBOL(preempt_schedule); |
4254 | ||
4255 | /* | |
2ed6e34f | 4256 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4257 | * off of irq context. |
4258 | * Note, that this is called and return with irqs disabled. This will | |
4259 | * protect us against recursive calling from irq. | |
4260 | */ | |
4261 | asmlinkage void __sched preempt_schedule_irq(void) | |
4262 | { | |
4263 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4264 | |
2ed6e34f | 4265 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4266 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4267 | ||
3a5c359a AK |
4268 | do { |
4269 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4270 | local_irq_enable(); |
4271 | schedule(); | |
4272 | local_irq_disable(); | |
3a5c359a | 4273 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4274 | |
3a5c359a AK |
4275 | /* |
4276 | * Check again in case we missed a preemption opportunity | |
4277 | * between schedule and now. | |
4278 | */ | |
4279 | barrier(); | |
5ed0cec0 | 4280 | } while (need_resched()); |
1da177e4 LT |
4281 | } |
4282 | ||
4283 | #endif /* CONFIG_PREEMPT */ | |
4284 | ||
63859d4f | 4285 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4286 | void *key) |
1da177e4 | 4287 | { |
63859d4f | 4288 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4289 | } |
1da177e4 LT |
4290 | EXPORT_SYMBOL(default_wake_function); |
4291 | ||
4292 | /* | |
41a2d6cf IM |
4293 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4294 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4295 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4296 | * | |
4297 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4298 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4299 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4300 | */ | |
78ddb08f | 4301 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4302 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4303 | { |
2e45874c | 4304 | wait_queue_t *curr, *next; |
1da177e4 | 4305 | |
2e45874c | 4306 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4307 | unsigned flags = curr->flags; |
4308 | ||
63859d4f | 4309 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4310 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4311 | break; |
4312 | } | |
4313 | } | |
4314 | ||
4315 | /** | |
4316 | * __wake_up - wake up threads blocked on a waitqueue. | |
4317 | * @q: the waitqueue | |
4318 | * @mode: which threads | |
4319 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4320 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4321 | * |
4322 | * It may be assumed that this function implies a write memory barrier before | |
4323 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4324 | */ |
7ad5b3a5 | 4325 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4326 | int nr_exclusive, void *key) |
1da177e4 LT |
4327 | { |
4328 | unsigned long flags; | |
4329 | ||
4330 | spin_lock_irqsave(&q->lock, flags); | |
4331 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4332 | spin_unlock_irqrestore(&q->lock, flags); | |
4333 | } | |
1da177e4 LT |
4334 | EXPORT_SYMBOL(__wake_up); |
4335 | ||
4336 | /* | |
4337 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4338 | */ | |
7ad5b3a5 | 4339 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4340 | { |
4341 | __wake_up_common(q, mode, 1, 0, NULL); | |
4342 | } | |
22c43c81 | 4343 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4344 | |
4ede816a DL |
4345 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4346 | { | |
4347 | __wake_up_common(q, mode, 1, 0, key); | |
4348 | } | |
4349 | ||
1da177e4 | 4350 | /** |
4ede816a | 4351 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4352 | * @q: the waitqueue |
4353 | * @mode: which threads | |
4354 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4355 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4356 | * |
4357 | * The sync wakeup differs that the waker knows that it will schedule | |
4358 | * away soon, so while the target thread will be woken up, it will not | |
4359 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4360 | * with each other. This can prevent needless bouncing between CPUs. | |
4361 | * | |
4362 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4363 | * |
4364 | * It may be assumed that this function implies a write memory barrier before | |
4365 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4366 | */ |
4ede816a DL |
4367 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4368 | int nr_exclusive, void *key) | |
1da177e4 LT |
4369 | { |
4370 | unsigned long flags; | |
7d478721 | 4371 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4372 | |
4373 | if (unlikely(!q)) | |
4374 | return; | |
4375 | ||
4376 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4377 | wake_flags = 0; |
1da177e4 LT |
4378 | |
4379 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4380 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4381 | spin_unlock_irqrestore(&q->lock, flags); |
4382 | } | |
4ede816a DL |
4383 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4384 | ||
4385 | /* | |
4386 | * __wake_up_sync - see __wake_up_sync_key() | |
4387 | */ | |
4388 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4389 | { | |
4390 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4391 | } | |
1da177e4 LT |
4392 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4393 | ||
65eb3dc6 KD |
4394 | /** |
4395 | * complete: - signals a single thread waiting on this completion | |
4396 | * @x: holds the state of this particular completion | |
4397 | * | |
4398 | * This will wake up a single thread waiting on this completion. Threads will be | |
4399 | * awakened in the same order in which they were queued. | |
4400 | * | |
4401 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4402 | * |
4403 | * It may be assumed that this function implies a write memory barrier before | |
4404 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4405 | */ |
b15136e9 | 4406 | void complete(struct completion *x) |
1da177e4 LT |
4407 | { |
4408 | unsigned long flags; | |
4409 | ||
4410 | spin_lock_irqsave(&x->wait.lock, flags); | |
4411 | x->done++; | |
d9514f6c | 4412 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4413 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4414 | } | |
4415 | EXPORT_SYMBOL(complete); | |
4416 | ||
65eb3dc6 KD |
4417 | /** |
4418 | * complete_all: - signals all threads waiting on this completion | |
4419 | * @x: holds the state of this particular completion | |
4420 | * | |
4421 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4422 | * |
4423 | * It may be assumed that this function implies a write memory barrier before | |
4424 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4425 | */ |
b15136e9 | 4426 | void complete_all(struct completion *x) |
1da177e4 LT |
4427 | { |
4428 | unsigned long flags; | |
4429 | ||
4430 | spin_lock_irqsave(&x->wait.lock, flags); | |
4431 | x->done += UINT_MAX/2; | |
d9514f6c | 4432 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4433 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4434 | } | |
4435 | EXPORT_SYMBOL(complete_all); | |
4436 | ||
8cbbe86d AK |
4437 | static inline long __sched |
4438 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4439 | { |
1da177e4 LT |
4440 | if (!x->done) { |
4441 | DECLARE_WAITQUEUE(wait, current); | |
4442 | ||
a93d2f17 | 4443 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4444 | do { |
94d3d824 | 4445 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4446 | timeout = -ERESTARTSYS; |
4447 | break; | |
8cbbe86d AK |
4448 | } |
4449 | __set_current_state(state); | |
1da177e4 LT |
4450 | spin_unlock_irq(&x->wait.lock); |
4451 | timeout = schedule_timeout(timeout); | |
4452 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4453 | } while (!x->done && timeout); |
1da177e4 | 4454 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4455 | if (!x->done) |
4456 | return timeout; | |
1da177e4 LT |
4457 | } |
4458 | x->done--; | |
ea71a546 | 4459 | return timeout ?: 1; |
1da177e4 | 4460 | } |
1da177e4 | 4461 | |
8cbbe86d AK |
4462 | static long __sched |
4463 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4464 | { |
1da177e4 LT |
4465 | might_sleep(); |
4466 | ||
4467 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4468 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4469 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4470 | return timeout; |
4471 | } | |
1da177e4 | 4472 | |
65eb3dc6 KD |
4473 | /** |
4474 | * wait_for_completion: - waits for completion of a task | |
4475 | * @x: holds the state of this particular completion | |
4476 | * | |
4477 | * This waits to be signaled for completion of a specific task. It is NOT | |
4478 | * interruptible and there is no timeout. | |
4479 | * | |
4480 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4481 | * and interrupt capability. Also see complete(). | |
4482 | */ | |
b15136e9 | 4483 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4484 | { |
4485 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4486 | } |
8cbbe86d | 4487 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4488 | |
65eb3dc6 KD |
4489 | /** |
4490 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4491 | * @x: holds the state of this particular completion | |
4492 | * @timeout: timeout value in jiffies | |
4493 | * | |
4494 | * This waits for either a completion of a specific task to be signaled or for a | |
4495 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4496 | * interruptible. | |
4497 | */ | |
b15136e9 | 4498 | unsigned long __sched |
8cbbe86d | 4499 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4500 | { |
8cbbe86d | 4501 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4502 | } |
8cbbe86d | 4503 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4504 | |
65eb3dc6 KD |
4505 | /** |
4506 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4507 | * @x: holds the state of this particular completion | |
4508 | * | |
4509 | * This waits for completion of a specific task to be signaled. It is | |
4510 | * interruptible. | |
4511 | */ | |
8cbbe86d | 4512 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4513 | { |
51e97990 AK |
4514 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4515 | if (t == -ERESTARTSYS) | |
4516 | return t; | |
4517 | return 0; | |
0fec171c | 4518 | } |
8cbbe86d | 4519 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4520 | |
65eb3dc6 KD |
4521 | /** |
4522 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4523 | * @x: holds the state of this particular completion | |
4524 | * @timeout: timeout value in jiffies | |
4525 | * | |
4526 | * This waits for either a completion of a specific task to be signaled or for a | |
4527 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4528 | */ | |
b15136e9 | 4529 | unsigned long __sched |
8cbbe86d AK |
4530 | wait_for_completion_interruptible_timeout(struct completion *x, |
4531 | unsigned long timeout) | |
0fec171c | 4532 | { |
8cbbe86d | 4533 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4534 | } |
8cbbe86d | 4535 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4536 | |
65eb3dc6 KD |
4537 | /** |
4538 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4539 | * @x: holds the state of this particular completion | |
4540 | * | |
4541 | * This waits to be signaled for completion of a specific task. It can be | |
4542 | * interrupted by a kill signal. | |
4543 | */ | |
009e577e MW |
4544 | int __sched wait_for_completion_killable(struct completion *x) |
4545 | { | |
4546 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4547 | if (t == -ERESTARTSYS) | |
4548 | return t; | |
4549 | return 0; | |
4550 | } | |
4551 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4552 | ||
0aa12fb4 SW |
4553 | /** |
4554 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4555 | * @x: holds the state of this particular completion | |
4556 | * @timeout: timeout value in jiffies | |
4557 | * | |
4558 | * This waits for either a completion of a specific task to be | |
4559 | * signaled or for a specified timeout to expire. It can be | |
4560 | * interrupted by a kill signal. The timeout is in jiffies. | |
4561 | */ | |
4562 | unsigned long __sched | |
4563 | wait_for_completion_killable_timeout(struct completion *x, | |
4564 | unsigned long timeout) | |
4565 | { | |
4566 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4567 | } | |
4568 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4569 | ||
be4de352 DC |
4570 | /** |
4571 | * try_wait_for_completion - try to decrement a completion without blocking | |
4572 | * @x: completion structure | |
4573 | * | |
4574 | * Returns: 0 if a decrement cannot be done without blocking | |
4575 | * 1 if a decrement succeeded. | |
4576 | * | |
4577 | * If a completion is being used as a counting completion, | |
4578 | * attempt to decrement the counter without blocking. This | |
4579 | * enables us to avoid waiting if the resource the completion | |
4580 | * is protecting is not available. | |
4581 | */ | |
4582 | bool try_wait_for_completion(struct completion *x) | |
4583 | { | |
7539a3b3 | 4584 | unsigned long flags; |
be4de352 DC |
4585 | int ret = 1; |
4586 | ||
7539a3b3 | 4587 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4588 | if (!x->done) |
4589 | ret = 0; | |
4590 | else | |
4591 | x->done--; | |
7539a3b3 | 4592 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4593 | return ret; |
4594 | } | |
4595 | EXPORT_SYMBOL(try_wait_for_completion); | |
4596 | ||
4597 | /** | |
4598 | * completion_done - Test to see if a completion has any waiters | |
4599 | * @x: completion structure | |
4600 | * | |
4601 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4602 | * 1 if there are no waiters. | |
4603 | * | |
4604 | */ | |
4605 | bool completion_done(struct completion *x) | |
4606 | { | |
7539a3b3 | 4607 | unsigned long flags; |
be4de352 DC |
4608 | int ret = 1; |
4609 | ||
7539a3b3 | 4610 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4611 | if (!x->done) |
4612 | ret = 0; | |
7539a3b3 | 4613 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4614 | return ret; |
4615 | } | |
4616 | EXPORT_SYMBOL(completion_done); | |
4617 | ||
8cbbe86d AK |
4618 | static long __sched |
4619 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4620 | { |
0fec171c IM |
4621 | unsigned long flags; |
4622 | wait_queue_t wait; | |
4623 | ||
4624 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4625 | |
8cbbe86d | 4626 | __set_current_state(state); |
1da177e4 | 4627 | |
8cbbe86d AK |
4628 | spin_lock_irqsave(&q->lock, flags); |
4629 | __add_wait_queue(q, &wait); | |
4630 | spin_unlock(&q->lock); | |
4631 | timeout = schedule_timeout(timeout); | |
4632 | spin_lock_irq(&q->lock); | |
4633 | __remove_wait_queue(q, &wait); | |
4634 | spin_unlock_irqrestore(&q->lock, flags); | |
4635 | ||
4636 | return timeout; | |
4637 | } | |
4638 | ||
4639 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4640 | { | |
4641 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4642 | } |
1da177e4 LT |
4643 | EXPORT_SYMBOL(interruptible_sleep_on); |
4644 | ||
0fec171c | 4645 | long __sched |
95cdf3b7 | 4646 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4647 | { |
8cbbe86d | 4648 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4649 | } |
1da177e4 LT |
4650 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4651 | ||
0fec171c | 4652 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4653 | { |
8cbbe86d | 4654 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4655 | } |
1da177e4 LT |
4656 | EXPORT_SYMBOL(sleep_on); |
4657 | ||
0fec171c | 4658 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4659 | { |
8cbbe86d | 4660 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4661 | } |
1da177e4 LT |
4662 | EXPORT_SYMBOL(sleep_on_timeout); |
4663 | ||
b29739f9 IM |
4664 | #ifdef CONFIG_RT_MUTEXES |
4665 | ||
4666 | /* | |
4667 | * rt_mutex_setprio - set the current priority of a task | |
4668 | * @p: task | |
4669 | * @prio: prio value (kernel-internal form) | |
4670 | * | |
4671 | * This function changes the 'effective' priority of a task. It does | |
4672 | * not touch ->normal_prio like __setscheduler(). | |
4673 | * | |
4674 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4675 | */ | |
36c8b586 | 4676 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4677 | { |
4678 | unsigned long flags; | |
83b699ed | 4679 | int oldprio, on_rq, running; |
70b97a7f | 4680 | struct rq *rq; |
83ab0aa0 | 4681 | const struct sched_class *prev_class; |
b29739f9 IM |
4682 | |
4683 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4684 | ||
4685 | rq = task_rq_lock(p, &flags); | |
4686 | ||
a8027073 | 4687 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4688 | oldprio = p->prio; |
83ab0aa0 | 4689 | prev_class = p->sched_class; |
dd41f596 | 4690 | on_rq = p->se.on_rq; |
051a1d1a | 4691 | running = task_current(rq, p); |
0e1f3483 | 4692 | if (on_rq) |
69be72c1 | 4693 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4694 | if (running) |
4695 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4696 | |
4697 | if (rt_prio(prio)) | |
4698 | p->sched_class = &rt_sched_class; | |
4699 | else | |
4700 | p->sched_class = &fair_sched_class; | |
4701 | ||
b29739f9 IM |
4702 | p->prio = prio; |
4703 | ||
0e1f3483 HS |
4704 | if (running) |
4705 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4706 | if (on_rq) { |
371fd7e7 | 4707 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4708 | |
4709 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4710 | } |
4711 | task_rq_unlock(rq, &flags); | |
4712 | } | |
4713 | ||
4714 | #endif | |
4715 | ||
36c8b586 | 4716 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4717 | { |
dd41f596 | 4718 | int old_prio, delta, on_rq; |
1da177e4 | 4719 | unsigned long flags; |
70b97a7f | 4720 | struct rq *rq; |
1da177e4 LT |
4721 | |
4722 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4723 | return; | |
4724 | /* | |
4725 | * We have to be careful, if called from sys_setpriority(), | |
4726 | * the task might be in the middle of scheduling on another CPU. | |
4727 | */ | |
4728 | rq = task_rq_lock(p, &flags); | |
4729 | /* | |
4730 | * The RT priorities are set via sched_setscheduler(), but we still | |
4731 | * allow the 'normal' nice value to be set - but as expected | |
4732 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4733 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4734 | */ |
e05606d3 | 4735 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4736 | p->static_prio = NICE_TO_PRIO(nice); |
4737 | goto out_unlock; | |
4738 | } | |
dd41f596 | 4739 | on_rq = p->se.on_rq; |
c09595f6 | 4740 | if (on_rq) |
69be72c1 | 4741 | dequeue_task(rq, p, 0); |
1da177e4 | 4742 | |
1da177e4 | 4743 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4744 | set_load_weight(p); |
b29739f9 IM |
4745 | old_prio = p->prio; |
4746 | p->prio = effective_prio(p); | |
4747 | delta = p->prio - old_prio; | |
1da177e4 | 4748 | |
dd41f596 | 4749 | if (on_rq) { |
371fd7e7 | 4750 | enqueue_task(rq, p, 0); |
1da177e4 | 4751 | /* |
d5f9f942 AM |
4752 | * If the task increased its priority or is running and |
4753 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4754 | */ |
d5f9f942 | 4755 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4756 | resched_task(rq->curr); |
4757 | } | |
4758 | out_unlock: | |
4759 | task_rq_unlock(rq, &flags); | |
4760 | } | |
1da177e4 LT |
4761 | EXPORT_SYMBOL(set_user_nice); |
4762 | ||
e43379f1 MM |
4763 | /* |
4764 | * can_nice - check if a task can reduce its nice value | |
4765 | * @p: task | |
4766 | * @nice: nice value | |
4767 | */ | |
36c8b586 | 4768 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4769 | { |
024f4747 MM |
4770 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4771 | int nice_rlim = 20 - nice; | |
48f24c4d | 4772 | |
78d7d407 | 4773 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4774 | capable(CAP_SYS_NICE)); |
4775 | } | |
4776 | ||
1da177e4 LT |
4777 | #ifdef __ARCH_WANT_SYS_NICE |
4778 | ||
4779 | /* | |
4780 | * sys_nice - change the priority of the current process. | |
4781 | * @increment: priority increment | |
4782 | * | |
4783 | * sys_setpriority is a more generic, but much slower function that | |
4784 | * does similar things. | |
4785 | */ | |
5add95d4 | 4786 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4787 | { |
48f24c4d | 4788 | long nice, retval; |
1da177e4 LT |
4789 | |
4790 | /* | |
4791 | * Setpriority might change our priority at the same moment. | |
4792 | * We don't have to worry. Conceptually one call occurs first | |
4793 | * and we have a single winner. | |
4794 | */ | |
e43379f1 MM |
4795 | if (increment < -40) |
4796 | increment = -40; | |
1da177e4 LT |
4797 | if (increment > 40) |
4798 | increment = 40; | |
4799 | ||
2b8f836f | 4800 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4801 | if (nice < -20) |
4802 | nice = -20; | |
4803 | if (nice > 19) | |
4804 | nice = 19; | |
4805 | ||
e43379f1 MM |
4806 | if (increment < 0 && !can_nice(current, nice)) |
4807 | return -EPERM; | |
4808 | ||
1da177e4 LT |
4809 | retval = security_task_setnice(current, nice); |
4810 | if (retval) | |
4811 | return retval; | |
4812 | ||
4813 | set_user_nice(current, nice); | |
4814 | return 0; | |
4815 | } | |
4816 | ||
4817 | #endif | |
4818 | ||
4819 | /** | |
4820 | * task_prio - return the priority value of a given task. | |
4821 | * @p: the task in question. | |
4822 | * | |
4823 | * This is the priority value as seen by users in /proc. | |
4824 | * RT tasks are offset by -200. Normal tasks are centered | |
4825 | * around 0, value goes from -16 to +15. | |
4826 | */ | |
36c8b586 | 4827 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4828 | { |
4829 | return p->prio - MAX_RT_PRIO; | |
4830 | } | |
4831 | ||
4832 | /** | |
4833 | * task_nice - return the nice value of a given task. | |
4834 | * @p: the task in question. | |
4835 | */ | |
36c8b586 | 4836 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4837 | { |
4838 | return TASK_NICE(p); | |
4839 | } | |
150d8bed | 4840 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4841 | |
4842 | /** | |
4843 | * idle_cpu - is a given cpu idle currently? | |
4844 | * @cpu: the processor in question. | |
4845 | */ | |
4846 | int idle_cpu(int cpu) | |
4847 | { | |
4848 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4849 | } | |
4850 | ||
1da177e4 LT |
4851 | /** |
4852 | * idle_task - return the idle task for a given cpu. | |
4853 | * @cpu: the processor in question. | |
4854 | */ | |
36c8b586 | 4855 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4856 | { |
4857 | return cpu_rq(cpu)->idle; | |
4858 | } | |
4859 | ||
4860 | /** | |
4861 | * find_process_by_pid - find a process with a matching PID value. | |
4862 | * @pid: the pid in question. | |
4863 | */ | |
a9957449 | 4864 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4865 | { |
228ebcbe | 4866 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4867 | } |
4868 | ||
4869 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4870 | static void |
4871 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4872 | { |
dd41f596 | 4873 | BUG_ON(p->se.on_rq); |
48f24c4d | 4874 | |
1da177e4 LT |
4875 | p->policy = policy; |
4876 | p->rt_priority = prio; | |
b29739f9 IM |
4877 | p->normal_prio = normal_prio(p); |
4878 | /* we are holding p->pi_lock already */ | |
4879 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4880 | if (rt_prio(p->prio)) |
4881 | p->sched_class = &rt_sched_class; | |
4882 | else | |
4883 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4884 | set_load_weight(p); |
1da177e4 LT |
4885 | } |
4886 | ||
c69e8d9c DH |
4887 | /* |
4888 | * check the target process has a UID that matches the current process's | |
4889 | */ | |
4890 | static bool check_same_owner(struct task_struct *p) | |
4891 | { | |
4892 | const struct cred *cred = current_cred(), *pcred; | |
4893 | bool match; | |
4894 | ||
4895 | rcu_read_lock(); | |
4896 | pcred = __task_cred(p); | |
4897 | match = (cred->euid == pcred->euid || | |
4898 | cred->euid == pcred->uid); | |
4899 | rcu_read_unlock(); | |
4900 | return match; | |
4901 | } | |
4902 | ||
961ccddd RR |
4903 | static int __sched_setscheduler(struct task_struct *p, int policy, |
4904 | struct sched_param *param, bool user) | |
1da177e4 | 4905 | { |
83b699ed | 4906 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4907 | unsigned long flags; |
83ab0aa0 | 4908 | const struct sched_class *prev_class; |
70b97a7f | 4909 | struct rq *rq; |
ca94c442 | 4910 | int reset_on_fork; |
1da177e4 | 4911 | |
66e5393a SR |
4912 | /* may grab non-irq protected spin_locks */ |
4913 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4914 | recheck: |
4915 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4916 | if (policy < 0) { |
4917 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4918 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4919 | } else { |
4920 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4921 | policy &= ~SCHED_RESET_ON_FORK; | |
4922 | ||
4923 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4924 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4925 | policy != SCHED_IDLE) | |
4926 | return -EINVAL; | |
4927 | } | |
4928 | ||
1da177e4 LT |
4929 | /* |
4930 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4931 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4932 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4933 | */ |
4934 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4935 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4936 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4937 | return -EINVAL; |
e05606d3 | 4938 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4939 | return -EINVAL; |
4940 | ||
37e4ab3f OC |
4941 | /* |
4942 | * Allow unprivileged RT tasks to decrease priority: | |
4943 | */ | |
961ccddd | 4944 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4945 | if (rt_policy(policy)) { |
a44702e8 ON |
4946 | unsigned long rlim_rtprio = |
4947 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4948 | |
4949 | /* can't set/change the rt policy */ | |
4950 | if (policy != p->policy && !rlim_rtprio) | |
4951 | return -EPERM; | |
4952 | ||
4953 | /* can't increase priority */ | |
4954 | if (param->sched_priority > p->rt_priority && | |
4955 | param->sched_priority > rlim_rtprio) | |
4956 | return -EPERM; | |
4957 | } | |
dd41f596 IM |
4958 | /* |
4959 | * Like positive nice levels, dont allow tasks to | |
4960 | * move out of SCHED_IDLE either: | |
4961 | */ | |
4962 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4963 | return -EPERM; | |
5fe1d75f | 4964 | |
37e4ab3f | 4965 | /* can't change other user's priorities */ |
c69e8d9c | 4966 | if (!check_same_owner(p)) |
37e4ab3f | 4967 | return -EPERM; |
ca94c442 LP |
4968 | |
4969 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4970 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4971 | return -EPERM; | |
37e4ab3f | 4972 | } |
1da177e4 | 4973 | |
725aad24 | 4974 | if (user) { |
b0ae1981 | 4975 | retval = security_task_setscheduler(p); |
725aad24 JF |
4976 | if (retval) |
4977 | return retval; | |
4978 | } | |
4979 | ||
b29739f9 IM |
4980 | /* |
4981 | * make sure no PI-waiters arrive (or leave) while we are | |
4982 | * changing the priority of the task: | |
4983 | */ | |
1d615482 | 4984 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4985 | /* |
4986 | * To be able to change p->policy safely, the apropriate | |
4987 | * runqueue lock must be held. | |
4988 | */ | |
b29739f9 | 4989 | rq = __task_rq_lock(p); |
dc61b1d6 | 4990 | |
34f971f6 PZ |
4991 | /* |
4992 | * Changing the policy of the stop threads its a very bad idea | |
4993 | */ | |
4994 | if (p == rq->stop) { | |
4995 | __task_rq_unlock(rq); | |
4996 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4997 | return -EINVAL; | |
4998 | } | |
4999 | ||
dc61b1d6 PZ |
5000 | #ifdef CONFIG_RT_GROUP_SCHED |
5001 | if (user) { | |
5002 | /* | |
5003 | * Do not allow realtime tasks into groups that have no runtime | |
5004 | * assigned. | |
5005 | */ | |
5006 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
5007 | task_group(p)->rt_bandwidth.rt_runtime == 0) { | |
5008 | __task_rq_unlock(rq); | |
5009 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
5010 | return -EPERM; | |
5011 | } | |
5012 | } | |
5013 | #endif | |
5014 | ||
1da177e4 LT |
5015 | /* recheck policy now with rq lock held */ |
5016 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
5017 | policy = oldpolicy = -1; | |
b29739f9 | 5018 | __task_rq_unlock(rq); |
1d615482 | 5019 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
5020 | goto recheck; |
5021 | } | |
dd41f596 | 5022 | on_rq = p->se.on_rq; |
051a1d1a | 5023 | running = task_current(rq, p); |
0e1f3483 | 5024 | if (on_rq) |
2e1cb74a | 5025 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
5026 | if (running) |
5027 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 5028 | |
ca94c442 LP |
5029 | p->sched_reset_on_fork = reset_on_fork; |
5030 | ||
1da177e4 | 5031 | oldprio = p->prio; |
83ab0aa0 | 5032 | prev_class = p->sched_class; |
dd41f596 | 5033 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 5034 | |
0e1f3483 HS |
5035 | if (running) |
5036 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
5037 | if (on_rq) { |
5038 | activate_task(rq, p, 0); | |
cb469845 SR |
5039 | |
5040 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 5041 | } |
b29739f9 | 5042 | __task_rq_unlock(rq); |
1d615482 | 5043 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 5044 | |
95e02ca9 TG |
5045 | rt_mutex_adjust_pi(p); |
5046 | ||
1da177e4 LT |
5047 | return 0; |
5048 | } | |
961ccddd RR |
5049 | |
5050 | /** | |
5051 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
5052 | * @p: the task in question. | |
5053 | * @policy: new policy. | |
5054 | * @param: structure containing the new RT priority. | |
5055 | * | |
5056 | * NOTE that the task may be already dead. | |
5057 | */ | |
5058 | int sched_setscheduler(struct task_struct *p, int policy, | |
5059 | struct sched_param *param) | |
5060 | { | |
5061 | return __sched_setscheduler(p, policy, param, true); | |
5062 | } | |
1da177e4 LT |
5063 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
5064 | ||
961ccddd RR |
5065 | /** |
5066 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
5067 | * @p: the task in question. | |
5068 | * @policy: new policy. | |
5069 | * @param: structure containing the new RT priority. | |
5070 | * | |
5071 | * Just like sched_setscheduler, only don't bother checking if the | |
5072 | * current context has permission. For example, this is needed in | |
5073 | * stop_machine(): we create temporary high priority worker threads, | |
5074 | * but our caller might not have that capability. | |
5075 | */ | |
5076 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
5077 | struct sched_param *param) | |
5078 | { | |
5079 | return __sched_setscheduler(p, policy, param, false); | |
5080 | } | |
5081 | ||
95cdf3b7 IM |
5082 | static int |
5083 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 5084 | { |
1da177e4 LT |
5085 | struct sched_param lparam; |
5086 | struct task_struct *p; | |
36c8b586 | 5087 | int retval; |
1da177e4 LT |
5088 | |
5089 | if (!param || pid < 0) | |
5090 | return -EINVAL; | |
5091 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
5092 | return -EFAULT; | |
5fe1d75f ON |
5093 | |
5094 | rcu_read_lock(); | |
5095 | retval = -ESRCH; | |
1da177e4 | 5096 | p = find_process_by_pid(pid); |
5fe1d75f ON |
5097 | if (p != NULL) |
5098 | retval = sched_setscheduler(p, policy, &lparam); | |
5099 | rcu_read_unlock(); | |
36c8b586 | 5100 | |
1da177e4 LT |
5101 | return retval; |
5102 | } | |
5103 | ||
5104 | /** | |
5105 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
5106 | * @pid: the pid in question. | |
5107 | * @policy: new policy. | |
5108 | * @param: structure containing the new RT priority. | |
5109 | */ | |
5add95d4 HC |
5110 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
5111 | struct sched_param __user *, param) | |
1da177e4 | 5112 | { |
c21761f1 JB |
5113 | /* negative values for policy are not valid */ |
5114 | if (policy < 0) | |
5115 | return -EINVAL; | |
5116 | ||
1da177e4 LT |
5117 | return do_sched_setscheduler(pid, policy, param); |
5118 | } | |
5119 | ||
5120 | /** | |
5121 | * sys_sched_setparam - set/change the RT priority of a thread | |
5122 | * @pid: the pid in question. | |
5123 | * @param: structure containing the new RT priority. | |
5124 | */ | |
5add95d4 | 5125 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5126 | { |
5127 | return do_sched_setscheduler(pid, -1, param); | |
5128 | } | |
5129 | ||
5130 | /** | |
5131 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
5132 | * @pid: the pid in question. | |
5133 | */ | |
5add95d4 | 5134 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5135 | { |
36c8b586 | 5136 | struct task_struct *p; |
3a5c359a | 5137 | int retval; |
1da177e4 LT |
5138 | |
5139 | if (pid < 0) | |
3a5c359a | 5140 | return -EINVAL; |
1da177e4 LT |
5141 | |
5142 | retval = -ESRCH; | |
5fe85be0 | 5143 | rcu_read_lock(); |
1da177e4 LT |
5144 | p = find_process_by_pid(pid); |
5145 | if (p) { | |
5146 | retval = security_task_getscheduler(p); | |
5147 | if (!retval) | |
ca94c442 LP |
5148 | retval = p->policy |
5149 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5150 | } |
5fe85be0 | 5151 | rcu_read_unlock(); |
1da177e4 LT |
5152 | return retval; |
5153 | } | |
5154 | ||
5155 | /** | |
ca94c442 | 5156 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5157 | * @pid: the pid in question. |
5158 | * @param: structure containing the RT priority. | |
5159 | */ | |
5add95d4 | 5160 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5161 | { |
5162 | struct sched_param lp; | |
36c8b586 | 5163 | struct task_struct *p; |
3a5c359a | 5164 | int retval; |
1da177e4 LT |
5165 | |
5166 | if (!param || pid < 0) | |
3a5c359a | 5167 | return -EINVAL; |
1da177e4 | 5168 | |
5fe85be0 | 5169 | rcu_read_lock(); |
1da177e4 LT |
5170 | p = find_process_by_pid(pid); |
5171 | retval = -ESRCH; | |
5172 | if (!p) | |
5173 | goto out_unlock; | |
5174 | ||
5175 | retval = security_task_getscheduler(p); | |
5176 | if (retval) | |
5177 | goto out_unlock; | |
5178 | ||
5179 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5180 | rcu_read_unlock(); |
1da177e4 LT |
5181 | |
5182 | /* | |
5183 | * This one might sleep, we cannot do it with a spinlock held ... | |
5184 | */ | |
5185 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5186 | ||
1da177e4 LT |
5187 | return retval; |
5188 | ||
5189 | out_unlock: | |
5fe85be0 | 5190 | rcu_read_unlock(); |
1da177e4 LT |
5191 | return retval; |
5192 | } | |
5193 | ||
96f874e2 | 5194 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5195 | { |
5a16f3d3 | 5196 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5197 | struct task_struct *p; |
5198 | int retval; | |
1da177e4 | 5199 | |
95402b38 | 5200 | get_online_cpus(); |
23f5d142 | 5201 | rcu_read_lock(); |
1da177e4 LT |
5202 | |
5203 | p = find_process_by_pid(pid); | |
5204 | if (!p) { | |
23f5d142 | 5205 | rcu_read_unlock(); |
95402b38 | 5206 | put_online_cpus(); |
1da177e4 LT |
5207 | return -ESRCH; |
5208 | } | |
5209 | ||
23f5d142 | 5210 | /* Prevent p going away */ |
1da177e4 | 5211 | get_task_struct(p); |
23f5d142 | 5212 | rcu_read_unlock(); |
1da177e4 | 5213 | |
5a16f3d3 RR |
5214 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5215 | retval = -ENOMEM; | |
5216 | goto out_put_task; | |
5217 | } | |
5218 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5219 | retval = -ENOMEM; | |
5220 | goto out_free_cpus_allowed; | |
5221 | } | |
1da177e4 | 5222 | retval = -EPERM; |
c69e8d9c | 5223 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
5224 | goto out_unlock; |
5225 | ||
b0ae1981 | 5226 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5227 | if (retval) |
5228 | goto out_unlock; | |
5229 | ||
5a16f3d3 RR |
5230 | cpuset_cpus_allowed(p, cpus_allowed); |
5231 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5232 | again: |
5a16f3d3 | 5233 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5234 | |
8707d8b8 | 5235 | if (!retval) { |
5a16f3d3 RR |
5236 | cpuset_cpus_allowed(p, cpus_allowed); |
5237 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5238 | /* |
5239 | * We must have raced with a concurrent cpuset | |
5240 | * update. Just reset the cpus_allowed to the | |
5241 | * cpuset's cpus_allowed | |
5242 | */ | |
5a16f3d3 | 5243 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5244 | goto again; |
5245 | } | |
5246 | } | |
1da177e4 | 5247 | out_unlock: |
5a16f3d3 RR |
5248 | free_cpumask_var(new_mask); |
5249 | out_free_cpus_allowed: | |
5250 | free_cpumask_var(cpus_allowed); | |
5251 | out_put_task: | |
1da177e4 | 5252 | put_task_struct(p); |
95402b38 | 5253 | put_online_cpus(); |
1da177e4 LT |
5254 | return retval; |
5255 | } | |
5256 | ||
5257 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5258 | struct cpumask *new_mask) |
1da177e4 | 5259 | { |
96f874e2 RR |
5260 | if (len < cpumask_size()) |
5261 | cpumask_clear(new_mask); | |
5262 | else if (len > cpumask_size()) | |
5263 | len = cpumask_size(); | |
5264 | ||
1da177e4 LT |
5265 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5266 | } | |
5267 | ||
5268 | /** | |
5269 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5270 | * @pid: pid of the process | |
5271 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5272 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5273 | */ | |
5add95d4 HC |
5274 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5275 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5276 | { |
5a16f3d3 | 5277 | cpumask_var_t new_mask; |
1da177e4 LT |
5278 | int retval; |
5279 | ||
5a16f3d3 RR |
5280 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5281 | return -ENOMEM; | |
1da177e4 | 5282 | |
5a16f3d3 RR |
5283 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5284 | if (retval == 0) | |
5285 | retval = sched_setaffinity(pid, new_mask); | |
5286 | free_cpumask_var(new_mask); | |
5287 | return retval; | |
1da177e4 LT |
5288 | } |
5289 | ||
96f874e2 | 5290 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5291 | { |
36c8b586 | 5292 | struct task_struct *p; |
31605683 TG |
5293 | unsigned long flags; |
5294 | struct rq *rq; | |
1da177e4 | 5295 | int retval; |
1da177e4 | 5296 | |
95402b38 | 5297 | get_online_cpus(); |
23f5d142 | 5298 | rcu_read_lock(); |
1da177e4 LT |
5299 | |
5300 | retval = -ESRCH; | |
5301 | p = find_process_by_pid(pid); | |
5302 | if (!p) | |
5303 | goto out_unlock; | |
5304 | ||
e7834f8f DQ |
5305 | retval = security_task_getscheduler(p); |
5306 | if (retval) | |
5307 | goto out_unlock; | |
5308 | ||
31605683 | 5309 | rq = task_rq_lock(p, &flags); |
96f874e2 | 5310 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 5311 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5312 | |
5313 | out_unlock: | |
23f5d142 | 5314 | rcu_read_unlock(); |
95402b38 | 5315 | put_online_cpus(); |
1da177e4 | 5316 | |
9531b62f | 5317 | return retval; |
1da177e4 LT |
5318 | } |
5319 | ||
5320 | /** | |
5321 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5322 | * @pid: pid of the process | |
5323 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5324 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5325 | */ | |
5add95d4 HC |
5326 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5327 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5328 | { |
5329 | int ret; | |
f17c8607 | 5330 | cpumask_var_t mask; |
1da177e4 | 5331 | |
84fba5ec | 5332 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5333 | return -EINVAL; |
5334 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5335 | return -EINVAL; |
5336 | ||
f17c8607 RR |
5337 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5338 | return -ENOMEM; | |
1da177e4 | 5339 | |
f17c8607 RR |
5340 | ret = sched_getaffinity(pid, mask); |
5341 | if (ret == 0) { | |
8bc037fb | 5342 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5343 | |
5344 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5345 | ret = -EFAULT; |
5346 | else | |
cd3d8031 | 5347 | ret = retlen; |
f17c8607 RR |
5348 | } |
5349 | free_cpumask_var(mask); | |
1da177e4 | 5350 | |
f17c8607 | 5351 | return ret; |
1da177e4 LT |
5352 | } |
5353 | ||
5354 | /** | |
5355 | * sys_sched_yield - yield the current processor to other threads. | |
5356 | * | |
dd41f596 IM |
5357 | * This function yields the current CPU to other tasks. If there are no |
5358 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5359 | */ |
5add95d4 | 5360 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5361 | { |
70b97a7f | 5362 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5363 | |
2d72376b | 5364 | schedstat_inc(rq, yld_count); |
4530d7ab | 5365 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5366 | |
5367 | /* | |
5368 | * Since we are going to call schedule() anyway, there's | |
5369 | * no need to preempt or enable interrupts: | |
5370 | */ | |
5371 | __release(rq->lock); | |
8a25d5de | 5372 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5373 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5374 | preempt_enable_no_resched(); |
5375 | ||
5376 | schedule(); | |
5377 | ||
5378 | return 0; | |
5379 | } | |
5380 | ||
d86ee480 PZ |
5381 | static inline int should_resched(void) |
5382 | { | |
5383 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5384 | } | |
5385 | ||
e7b38404 | 5386 | static void __cond_resched(void) |
1da177e4 | 5387 | { |
e7aaaa69 FW |
5388 | add_preempt_count(PREEMPT_ACTIVE); |
5389 | schedule(); | |
5390 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5391 | } |
5392 | ||
02b67cc3 | 5393 | int __sched _cond_resched(void) |
1da177e4 | 5394 | { |
d86ee480 | 5395 | if (should_resched()) { |
1da177e4 LT |
5396 | __cond_resched(); |
5397 | return 1; | |
5398 | } | |
5399 | return 0; | |
5400 | } | |
02b67cc3 | 5401 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5402 | |
5403 | /* | |
613afbf8 | 5404 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5405 | * call schedule, and on return reacquire the lock. |
5406 | * | |
41a2d6cf | 5407 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5408 | * operations here to prevent schedule() from being called twice (once via |
5409 | * spin_unlock(), once by hand). | |
5410 | */ | |
613afbf8 | 5411 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5412 | { |
d86ee480 | 5413 | int resched = should_resched(); |
6df3cecb JK |
5414 | int ret = 0; |
5415 | ||
f607c668 PZ |
5416 | lockdep_assert_held(lock); |
5417 | ||
95c354fe | 5418 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5419 | spin_unlock(lock); |
d86ee480 | 5420 | if (resched) |
95c354fe NP |
5421 | __cond_resched(); |
5422 | else | |
5423 | cpu_relax(); | |
6df3cecb | 5424 | ret = 1; |
1da177e4 | 5425 | spin_lock(lock); |
1da177e4 | 5426 | } |
6df3cecb | 5427 | return ret; |
1da177e4 | 5428 | } |
613afbf8 | 5429 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5430 | |
613afbf8 | 5431 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5432 | { |
5433 | BUG_ON(!in_softirq()); | |
5434 | ||
d86ee480 | 5435 | if (should_resched()) { |
98d82567 | 5436 | local_bh_enable(); |
1da177e4 LT |
5437 | __cond_resched(); |
5438 | local_bh_disable(); | |
5439 | return 1; | |
5440 | } | |
5441 | return 0; | |
5442 | } | |
613afbf8 | 5443 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5444 | |
1da177e4 LT |
5445 | /** |
5446 | * yield - yield the current processor to other threads. | |
5447 | * | |
72fd4a35 | 5448 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5449 | * thread runnable and calls sys_sched_yield(). |
5450 | */ | |
5451 | void __sched yield(void) | |
5452 | { | |
5453 | set_current_state(TASK_RUNNING); | |
5454 | sys_sched_yield(); | |
5455 | } | |
1da177e4 LT |
5456 | EXPORT_SYMBOL(yield); |
5457 | ||
5458 | /* | |
41a2d6cf | 5459 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5460 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5461 | */ |
5462 | void __sched io_schedule(void) | |
5463 | { | |
54d35f29 | 5464 | struct rq *rq = raw_rq(); |
1da177e4 | 5465 | |
0ff92245 | 5466 | delayacct_blkio_start(); |
1da177e4 | 5467 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5468 | current->in_iowait = 1; |
1da177e4 | 5469 | schedule(); |
8f0dfc34 | 5470 | current->in_iowait = 0; |
1da177e4 | 5471 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5472 | delayacct_blkio_end(); |
1da177e4 | 5473 | } |
1da177e4 LT |
5474 | EXPORT_SYMBOL(io_schedule); |
5475 | ||
5476 | long __sched io_schedule_timeout(long timeout) | |
5477 | { | |
54d35f29 | 5478 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5479 | long ret; |
5480 | ||
0ff92245 | 5481 | delayacct_blkio_start(); |
1da177e4 | 5482 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5483 | current->in_iowait = 1; |
1da177e4 | 5484 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5485 | current->in_iowait = 0; |
1da177e4 | 5486 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5487 | delayacct_blkio_end(); |
1da177e4 LT |
5488 | return ret; |
5489 | } | |
5490 | ||
5491 | /** | |
5492 | * sys_sched_get_priority_max - return maximum RT priority. | |
5493 | * @policy: scheduling class. | |
5494 | * | |
5495 | * this syscall returns the maximum rt_priority that can be used | |
5496 | * by a given scheduling class. | |
5497 | */ | |
5add95d4 | 5498 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5499 | { |
5500 | int ret = -EINVAL; | |
5501 | ||
5502 | switch (policy) { | |
5503 | case SCHED_FIFO: | |
5504 | case SCHED_RR: | |
5505 | ret = MAX_USER_RT_PRIO-1; | |
5506 | break; | |
5507 | case SCHED_NORMAL: | |
b0a9499c | 5508 | case SCHED_BATCH: |
dd41f596 | 5509 | case SCHED_IDLE: |
1da177e4 LT |
5510 | ret = 0; |
5511 | break; | |
5512 | } | |
5513 | return ret; | |
5514 | } | |
5515 | ||
5516 | /** | |
5517 | * sys_sched_get_priority_min - return minimum RT priority. | |
5518 | * @policy: scheduling class. | |
5519 | * | |
5520 | * this syscall returns the minimum rt_priority that can be used | |
5521 | * by a given scheduling class. | |
5522 | */ | |
5add95d4 | 5523 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5524 | { |
5525 | int ret = -EINVAL; | |
5526 | ||
5527 | switch (policy) { | |
5528 | case SCHED_FIFO: | |
5529 | case SCHED_RR: | |
5530 | ret = 1; | |
5531 | break; | |
5532 | case SCHED_NORMAL: | |
b0a9499c | 5533 | case SCHED_BATCH: |
dd41f596 | 5534 | case SCHED_IDLE: |
1da177e4 LT |
5535 | ret = 0; |
5536 | } | |
5537 | return ret; | |
5538 | } | |
5539 | ||
5540 | /** | |
5541 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5542 | * @pid: pid of the process. | |
5543 | * @interval: userspace pointer to the timeslice value. | |
5544 | * | |
5545 | * this syscall writes the default timeslice value of a given process | |
5546 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5547 | */ | |
17da2bd9 | 5548 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5549 | struct timespec __user *, interval) |
1da177e4 | 5550 | { |
36c8b586 | 5551 | struct task_struct *p; |
a4ec24b4 | 5552 | unsigned int time_slice; |
dba091b9 TG |
5553 | unsigned long flags; |
5554 | struct rq *rq; | |
3a5c359a | 5555 | int retval; |
1da177e4 | 5556 | struct timespec t; |
1da177e4 LT |
5557 | |
5558 | if (pid < 0) | |
3a5c359a | 5559 | return -EINVAL; |
1da177e4 LT |
5560 | |
5561 | retval = -ESRCH; | |
1a551ae7 | 5562 | rcu_read_lock(); |
1da177e4 LT |
5563 | p = find_process_by_pid(pid); |
5564 | if (!p) | |
5565 | goto out_unlock; | |
5566 | ||
5567 | retval = security_task_getscheduler(p); | |
5568 | if (retval) | |
5569 | goto out_unlock; | |
5570 | ||
dba091b9 TG |
5571 | rq = task_rq_lock(p, &flags); |
5572 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5573 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5574 | |
1a551ae7 | 5575 | rcu_read_unlock(); |
a4ec24b4 | 5576 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5577 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5578 | return retval; |
3a5c359a | 5579 | |
1da177e4 | 5580 | out_unlock: |
1a551ae7 | 5581 | rcu_read_unlock(); |
1da177e4 LT |
5582 | return retval; |
5583 | } | |
5584 | ||
7c731e0a | 5585 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5586 | |
82a1fcb9 | 5587 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5588 | { |
1da177e4 | 5589 | unsigned long free = 0; |
36c8b586 | 5590 | unsigned state; |
1da177e4 | 5591 | |
1da177e4 | 5592 | state = p->state ? __ffs(p->state) + 1 : 0; |
3df0fc5b | 5593 | printk(KERN_INFO "%-13.13s %c", p->comm, |
2ed6e34f | 5594 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5595 | #if BITS_PER_LONG == 32 |
1da177e4 | 5596 | if (state == TASK_RUNNING) |
3df0fc5b | 5597 | printk(KERN_CONT " running "); |
1da177e4 | 5598 | else |
3df0fc5b | 5599 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5600 | #else |
5601 | if (state == TASK_RUNNING) | |
3df0fc5b | 5602 | printk(KERN_CONT " running task "); |
1da177e4 | 5603 | else |
3df0fc5b | 5604 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5605 | #endif |
5606 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5607 | free = stack_not_used(p); |
1da177e4 | 5608 | #endif |
3df0fc5b | 5609 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5610 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5611 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5612 | |
5fb5e6de | 5613 | show_stack(p, NULL); |
1da177e4 LT |
5614 | } |
5615 | ||
e59e2ae2 | 5616 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5617 | { |
36c8b586 | 5618 | struct task_struct *g, *p; |
1da177e4 | 5619 | |
4bd77321 | 5620 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5621 | printk(KERN_INFO |
5622 | " task PC stack pid father\n"); | |
1da177e4 | 5623 | #else |
3df0fc5b PZ |
5624 | printk(KERN_INFO |
5625 | " task PC stack pid father\n"); | |
1da177e4 LT |
5626 | #endif |
5627 | read_lock(&tasklist_lock); | |
5628 | do_each_thread(g, p) { | |
5629 | /* | |
5630 | * reset the NMI-timeout, listing all files on a slow | |
5631 | * console might take alot of time: | |
5632 | */ | |
5633 | touch_nmi_watchdog(); | |
39bc89fd | 5634 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5635 | sched_show_task(p); |
1da177e4 LT |
5636 | } while_each_thread(g, p); |
5637 | ||
04c9167f JF |
5638 | touch_all_softlockup_watchdogs(); |
5639 | ||
dd41f596 IM |
5640 | #ifdef CONFIG_SCHED_DEBUG |
5641 | sysrq_sched_debug_show(); | |
5642 | #endif | |
1da177e4 | 5643 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5644 | /* |
5645 | * Only show locks if all tasks are dumped: | |
5646 | */ | |
93335a21 | 5647 | if (!state_filter) |
e59e2ae2 | 5648 | debug_show_all_locks(); |
1da177e4 LT |
5649 | } |
5650 | ||
1df21055 IM |
5651 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5652 | { | |
dd41f596 | 5653 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5654 | } |
5655 | ||
f340c0d1 IM |
5656 | /** |
5657 | * init_idle - set up an idle thread for a given CPU | |
5658 | * @idle: task in question | |
5659 | * @cpu: cpu the idle task belongs to | |
5660 | * | |
5661 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5662 | * flag, to make booting more robust. | |
5663 | */ | |
5c1e1767 | 5664 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5665 | { |
70b97a7f | 5666 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5667 | unsigned long flags; |
5668 | ||
05fa785c | 5669 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5670 | |
dd41f596 | 5671 | __sched_fork(idle); |
06b83b5f | 5672 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5673 | idle->se.exec_start = sched_clock(); |
5674 | ||
96f874e2 | 5675 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5676 | /* |
5677 | * We're having a chicken and egg problem, even though we are | |
5678 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5679 | * lockdep check in task_group() will fail. | |
5680 | * | |
5681 | * Similar case to sched_fork(). / Alternatively we could | |
5682 | * use task_rq_lock() here and obtain the other rq->lock. | |
5683 | * | |
5684 | * Silence PROVE_RCU | |
5685 | */ | |
5686 | rcu_read_lock(); | |
dd41f596 | 5687 | __set_task_cpu(idle, cpu); |
6506cf6c | 5688 | rcu_read_unlock(); |
1da177e4 | 5689 | |
1da177e4 | 5690 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5691 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5692 | idle->oncpu = 1; | |
5693 | #endif | |
05fa785c | 5694 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5695 | |
5696 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5697 | #if defined(CONFIG_PREEMPT) |
5698 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5699 | #else | |
a1261f54 | 5700 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5701 | #endif |
dd41f596 IM |
5702 | /* |
5703 | * The idle tasks have their own, simple scheduling class: | |
5704 | */ | |
5705 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5706 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5707 | } |
5708 | ||
5709 | /* | |
5710 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5711 | * indicates which cpus entered this state. This is used | |
5712 | * in the rcu update to wait only for active cpus. For system | |
5713 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5714 | * always be CPU_BITS_NONE. |
1da177e4 | 5715 | */ |
6a7b3dc3 | 5716 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5717 | |
19978ca6 IM |
5718 | /* |
5719 | * Increase the granularity value when there are more CPUs, | |
5720 | * because with more CPUs the 'effective latency' as visible | |
5721 | * to users decreases. But the relationship is not linear, | |
5722 | * so pick a second-best guess by going with the log2 of the | |
5723 | * number of CPUs. | |
5724 | * | |
5725 | * This idea comes from the SD scheduler of Con Kolivas: | |
5726 | */ | |
acb4a848 | 5727 | static int get_update_sysctl_factor(void) |
19978ca6 | 5728 | { |
4ca3ef71 | 5729 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5730 | unsigned int factor; |
5731 | ||
5732 | switch (sysctl_sched_tunable_scaling) { | |
5733 | case SCHED_TUNABLESCALING_NONE: | |
5734 | factor = 1; | |
5735 | break; | |
5736 | case SCHED_TUNABLESCALING_LINEAR: | |
5737 | factor = cpus; | |
5738 | break; | |
5739 | case SCHED_TUNABLESCALING_LOG: | |
5740 | default: | |
5741 | factor = 1 + ilog2(cpus); | |
5742 | break; | |
5743 | } | |
19978ca6 | 5744 | |
acb4a848 CE |
5745 | return factor; |
5746 | } | |
19978ca6 | 5747 | |
acb4a848 CE |
5748 | static void update_sysctl(void) |
5749 | { | |
5750 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5751 | |
0bcdcf28 CE |
5752 | #define SET_SYSCTL(name) \ |
5753 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5754 | SET_SYSCTL(sched_min_granularity); | |
5755 | SET_SYSCTL(sched_latency); | |
5756 | SET_SYSCTL(sched_wakeup_granularity); | |
5757 | SET_SYSCTL(sched_shares_ratelimit); | |
5758 | #undef SET_SYSCTL | |
5759 | } | |
55cd5340 | 5760 | |
0bcdcf28 CE |
5761 | static inline void sched_init_granularity(void) |
5762 | { | |
5763 | update_sysctl(); | |
19978ca6 IM |
5764 | } |
5765 | ||
1da177e4 LT |
5766 | #ifdef CONFIG_SMP |
5767 | /* | |
5768 | * This is how migration works: | |
5769 | * | |
969c7921 TH |
5770 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5771 | * stop_one_cpu(). | |
5772 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5773 | * off the CPU) | |
5774 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5775 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5776 | * it and puts it into the right queue. |
969c7921 TH |
5777 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5778 | * is done. | |
1da177e4 LT |
5779 | */ |
5780 | ||
5781 | /* | |
5782 | * Change a given task's CPU affinity. Migrate the thread to a | |
5783 | * proper CPU and schedule it away if the CPU it's executing on | |
5784 | * is removed from the allowed bitmask. | |
5785 | * | |
5786 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5787 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5788 | * call is not atomic; no spinlocks may be held. |
5789 | */ | |
96f874e2 | 5790 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5791 | { |
5792 | unsigned long flags; | |
70b97a7f | 5793 | struct rq *rq; |
969c7921 | 5794 | unsigned int dest_cpu; |
48f24c4d | 5795 | int ret = 0; |
1da177e4 | 5796 | |
65cc8e48 PZ |
5797 | /* |
5798 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5799 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5800 | */ | |
5801 | again: | |
5802 | while (task_is_waking(p)) | |
5803 | cpu_relax(); | |
1da177e4 | 5804 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5805 | if (task_is_waking(p)) { |
5806 | task_rq_unlock(rq, &flags); | |
5807 | goto again; | |
5808 | } | |
e2912009 | 5809 | |
6ad4c188 | 5810 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5811 | ret = -EINVAL; |
5812 | goto out; | |
5813 | } | |
5814 | ||
9985b0ba | 5815 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5816 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5817 | ret = -EINVAL; |
5818 | goto out; | |
5819 | } | |
5820 | ||
73fe6aae | 5821 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5822 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5823 | else { |
96f874e2 RR |
5824 | cpumask_copy(&p->cpus_allowed, new_mask); |
5825 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5826 | } |
5827 | ||
1da177e4 | 5828 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5829 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5830 | goto out; |
5831 | ||
969c7921 TH |
5832 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
5833 | if (migrate_task(p, dest_cpu)) { | |
5834 | struct migration_arg arg = { p, dest_cpu }; | |
1da177e4 LT |
5835 | /* Need help from migration thread: drop lock and wait. */ |
5836 | task_rq_unlock(rq, &flags); | |
969c7921 | 5837 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5838 | tlb_migrate_finish(p->mm); |
5839 | return 0; | |
5840 | } | |
5841 | out: | |
5842 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5843 | |
1da177e4 LT |
5844 | return ret; |
5845 | } | |
cd8ba7cd | 5846 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5847 | |
5848 | /* | |
41a2d6cf | 5849 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5850 | * this because either it can't run here any more (set_cpus_allowed() |
5851 | * away from this CPU, or CPU going down), or because we're | |
5852 | * attempting to rebalance this task on exec (sched_exec). | |
5853 | * | |
5854 | * So we race with normal scheduler movements, but that's OK, as long | |
5855 | * as the task is no longer on this CPU. | |
efc30814 KK |
5856 | * |
5857 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5858 | */ |
efc30814 | 5859 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5860 | { |
70b97a7f | 5861 | struct rq *rq_dest, *rq_src; |
e2912009 | 5862 | int ret = 0; |
1da177e4 | 5863 | |
e761b772 | 5864 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5865 | return ret; |
1da177e4 LT |
5866 | |
5867 | rq_src = cpu_rq(src_cpu); | |
5868 | rq_dest = cpu_rq(dest_cpu); | |
5869 | ||
5870 | double_rq_lock(rq_src, rq_dest); | |
5871 | /* Already moved. */ | |
5872 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5873 | goto done; |
1da177e4 | 5874 | /* Affinity changed (again). */ |
96f874e2 | 5875 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5876 | goto fail; |
1da177e4 | 5877 | |
e2912009 PZ |
5878 | /* |
5879 | * If we're not on a rq, the next wake-up will ensure we're | |
5880 | * placed properly. | |
5881 | */ | |
5882 | if (p->se.on_rq) { | |
2e1cb74a | 5883 | deactivate_task(rq_src, p, 0); |
e2912009 | 5884 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5885 | activate_task(rq_dest, p, 0); |
15afe09b | 5886 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5887 | } |
b1e38734 | 5888 | done: |
efc30814 | 5889 | ret = 1; |
b1e38734 | 5890 | fail: |
1da177e4 | 5891 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5892 | return ret; |
1da177e4 LT |
5893 | } |
5894 | ||
5895 | /* | |
969c7921 TH |
5896 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5897 | * and performs thread migration by bumping thread off CPU then | |
5898 | * 'pushing' onto another runqueue. | |
1da177e4 | 5899 | */ |
969c7921 | 5900 | static int migration_cpu_stop(void *data) |
1da177e4 | 5901 | { |
969c7921 | 5902 | struct migration_arg *arg = data; |
f7b4cddc | 5903 | |
969c7921 TH |
5904 | /* |
5905 | * The original target cpu might have gone down and we might | |
5906 | * be on another cpu but it doesn't matter. | |
5907 | */ | |
f7b4cddc | 5908 | local_irq_disable(); |
969c7921 | 5909 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5910 | local_irq_enable(); |
1da177e4 | 5911 | return 0; |
f7b4cddc ON |
5912 | } |
5913 | ||
1da177e4 | 5914 | #ifdef CONFIG_HOTPLUG_CPU |
054b9108 | 5915 | /* |
3a4fa0a2 | 5916 | * Figure out where task on dead CPU should go, use force if necessary. |
054b9108 | 5917 | */ |
6a1bdc1b | 5918 | void move_task_off_dead_cpu(int dead_cpu, struct task_struct *p) |
1da177e4 | 5919 | { |
1445c08d ON |
5920 | struct rq *rq = cpu_rq(dead_cpu); |
5921 | int needs_cpu, uninitialized_var(dest_cpu); | |
5922 | unsigned long flags; | |
e76bd8d9 | 5923 | |
1445c08d | 5924 | local_irq_save(flags); |
e76bd8d9 | 5925 | |
1445c08d ON |
5926 | raw_spin_lock(&rq->lock); |
5927 | needs_cpu = (task_cpu(p) == dead_cpu) && (p->state != TASK_WAKING); | |
5928 | if (needs_cpu) | |
5929 | dest_cpu = select_fallback_rq(dead_cpu, p); | |
5930 | raw_spin_unlock(&rq->lock); | |
c1804d54 ON |
5931 | /* |
5932 | * It can only fail if we race with set_cpus_allowed(), | |
5933 | * in the racer should migrate the task anyway. | |
5934 | */ | |
1445c08d | 5935 | if (needs_cpu) |
c1804d54 | 5936 | __migrate_task(p, dead_cpu, dest_cpu); |
1445c08d | 5937 | local_irq_restore(flags); |
1da177e4 LT |
5938 | } |
5939 | ||
5940 | /* | |
5941 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5942 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5943 | * for performance reasons the counter is not stricly tracking tasks to | |
5944 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5945 | * to keep the global sum constant after CPU-down: | |
5946 | */ | |
70b97a7f | 5947 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5948 | { |
6ad4c188 | 5949 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 LT |
5950 | unsigned long flags; |
5951 | ||
5952 | local_irq_save(flags); | |
5953 | double_rq_lock(rq_src, rq_dest); | |
5954 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; | |
5955 | rq_src->nr_uninterruptible = 0; | |
5956 | double_rq_unlock(rq_src, rq_dest); | |
5957 | local_irq_restore(flags); | |
5958 | } | |
5959 | ||
5960 | /* Run through task list and migrate tasks from the dead cpu. */ | |
5961 | static void migrate_live_tasks(int src_cpu) | |
5962 | { | |
48f24c4d | 5963 | struct task_struct *p, *t; |
1da177e4 | 5964 | |
f7b4cddc | 5965 | read_lock(&tasklist_lock); |
1da177e4 | 5966 | |
48f24c4d IM |
5967 | do_each_thread(t, p) { |
5968 | if (p == current) | |
1da177e4 LT |
5969 | continue; |
5970 | ||
48f24c4d IM |
5971 | if (task_cpu(p) == src_cpu) |
5972 | move_task_off_dead_cpu(src_cpu, p); | |
5973 | } while_each_thread(t, p); | |
1da177e4 | 5974 | |
f7b4cddc | 5975 | read_unlock(&tasklist_lock); |
1da177e4 LT |
5976 | } |
5977 | ||
dd41f596 IM |
5978 | /* |
5979 | * Schedules idle task to be the next runnable task on current CPU. | |
94bc9a7b DA |
5980 | * It does so by boosting its priority to highest possible. |
5981 | * Used by CPU offline code. | |
1da177e4 LT |
5982 | */ |
5983 | void sched_idle_next(void) | |
5984 | { | |
48f24c4d | 5985 | int this_cpu = smp_processor_id(); |
70b97a7f | 5986 | struct rq *rq = cpu_rq(this_cpu); |
1da177e4 LT |
5987 | struct task_struct *p = rq->idle; |
5988 | unsigned long flags; | |
5989 | ||
5990 | /* cpu has to be offline */ | |
48f24c4d | 5991 | BUG_ON(cpu_online(this_cpu)); |
1da177e4 | 5992 | |
48f24c4d IM |
5993 | /* |
5994 | * Strictly not necessary since rest of the CPUs are stopped by now | |
5995 | * and interrupts disabled on the current cpu. | |
1da177e4 | 5996 | */ |
05fa785c | 5997 | raw_spin_lock_irqsave(&rq->lock, flags); |
1da177e4 | 5998 | |
dd41f596 | 5999 | __setscheduler(rq, p, SCHED_FIFO, MAX_RT_PRIO-1); |
48f24c4d | 6000 | |
94bc9a7b | 6001 | activate_task(rq, p, 0); |
1da177e4 | 6002 | |
05fa785c | 6003 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
6004 | } |
6005 | ||
48f24c4d IM |
6006 | /* |
6007 | * Ensures that the idle task is using init_mm right before its cpu goes | |
1da177e4 LT |
6008 | * offline. |
6009 | */ | |
6010 | void idle_task_exit(void) | |
6011 | { | |
6012 | struct mm_struct *mm = current->active_mm; | |
6013 | ||
6014 | BUG_ON(cpu_online(smp_processor_id())); | |
6015 | ||
6016 | if (mm != &init_mm) | |
6017 | switch_mm(mm, &init_mm, current); | |
6018 | mmdrop(mm); | |
6019 | } | |
6020 | ||
054b9108 | 6021 | /* called under rq->lock with disabled interrupts */ |
36c8b586 | 6022 | static void migrate_dead(unsigned int dead_cpu, struct task_struct *p) |
1da177e4 | 6023 | { |
70b97a7f | 6024 | struct rq *rq = cpu_rq(dead_cpu); |
1da177e4 LT |
6025 | |
6026 | /* Must be exiting, otherwise would be on tasklist. */ | |
270f722d | 6027 | BUG_ON(!p->exit_state); |
1da177e4 LT |
6028 | |
6029 | /* Cannot have done final schedule yet: would have vanished. */ | |
c394cc9f | 6030 | BUG_ON(p->state == TASK_DEAD); |
1da177e4 | 6031 | |
48f24c4d | 6032 | get_task_struct(p); |
1da177e4 LT |
6033 | |
6034 | /* | |
6035 | * Drop lock around migration; if someone else moves it, | |
41a2d6cf | 6036 | * that's OK. No task can be added to this CPU, so iteration is |
1da177e4 LT |
6037 | * fine. |
6038 | */ | |
05fa785c | 6039 | raw_spin_unlock_irq(&rq->lock); |
48f24c4d | 6040 | move_task_off_dead_cpu(dead_cpu, p); |
05fa785c | 6041 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 6042 | |
48f24c4d | 6043 | put_task_struct(p); |
1da177e4 LT |
6044 | } |
6045 | ||
6046 | /* release_task() removes task from tasklist, so we won't find dead tasks. */ | |
6047 | static void migrate_dead_tasks(unsigned int dead_cpu) | |
6048 | { | |
70b97a7f | 6049 | struct rq *rq = cpu_rq(dead_cpu); |
dd41f596 | 6050 | struct task_struct *next; |
48f24c4d | 6051 | |
dd41f596 IM |
6052 | for ( ; ; ) { |
6053 | if (!rq->nr_running) | |
6054 | break; | |
b67802ea | 6055 | next = pick_next_task(rq); |
dd41f596 IM |
6056 | if (!next) |
6057 | break; | |
79c53799 | 6058 | next->sched_class->put_prev_task(rq, next); |
dd41f596 | 6059 | migrate_dead(dead_cpu, next); |
e692ab53 | 6060 | |
1da177e4 LT |
6061 | } |
6062 | } | |
dce48a84 TG |
6063 | |
6064 | /* | |
6065 | * remove the tasks which were accounted by rq from calc_load_tasks. | |
6066 | */ | |
6067 | static void calc_global_load_remove(struct rq *rq) | |
6068 | { | |
6069 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); | |
a468d389 | 6070 | rq->calc_load_active = 0; |
dce48a84 | 6071 | } |
1da177e4 LT |
6072 | #endif /* CONFIG_HOTPLUG_CPU */ |
6073 | ||
e692ab53 NP |
6074 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
6075 | ||
6076 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
6077 | { |
6078 | .procname = "sched_domain", | |
c57baf1e | 6079 | .mode = 0555, |
e0361851 | 6080 | }, |
56992309 | 6081 | {} |
e692ab53 NP |
6082 | }; |
6083 | ||
6084 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
6085 | { |
6086 | .procname = "kernel", | |
c57baf1e | 6087 | .mode = 0555, |
e0361851 AD |
6088 | .child = sd_ctl_dir, |
6089 | }, | |
56992309 | 6090 | {} |
e692ab53 NP |
6091 | }; |
6092 | ||
6093 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
6094 | { | |
6095 | struct ctl_table *entry = | |
5cf9f062 | 6096 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 6097 | |
e692ab53 NP |
6098 | return entry; |
6099 | } | |
6100 | ||
6382bc90 MM |
6101 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
6102 | { | |
cd790076 | 6103 | struct ctl_table *entry; |
6382bc90 | 6104 | |
cd790076 MM |
6105 | /* |
6106 | * In the intermediate directories, both the child directory and | |
6107 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 6108 | * will always be set. In the lowest directory the names are |
cd790076 MM |
6109 | * static strings and all have proc handlers. |
6110 | */ | |
6111 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
6112 | if (entry->child) |
6113 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
6114 | if (entry->proc_handler == NULL) |
6115 | kfree(entry->procname); | |
6116 | } | |
6382bc90 MM |
6117 | |
6118 | kfree(*tablep); | |
6119 | *tablep = NULL; | |
6120 | } | |
6121 | ||
e692ab53 | 6122 | static void |
e0361851 | 6123 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
6124 | const char *procname, void *data, int maxlen, |
6125 | mode_t mode, proc_handler *proc_handler) | |
6126 | { | |
e692ab53 NP |
6127 | entry->procname = procname; |
6128 | entry->data = data; | |
6129 | entry->maxlen = maxlen; | |
6130 | entry->mode = mode; | |
6131 | entry->proc_handler = proc_handler; | |
6132 | } | |
6133 | ||
6134 | static struct ctl_table * | |
6135 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
6136 | { | |
a5d8c348 | 6137 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 6138 | |
ad1cdc1d MM |
6139 | if (table == NULL) |
6140 | return NULL; | |
6141 | ||
e0361851 | 6142 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 6143 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6144 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 6145 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 6146 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 6147 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6148 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 6149 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6150 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 6151 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6152 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 6153 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6154 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 6155 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6156 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 6157 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 6158 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 6159 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 6160 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
6161 | &sd->cache_nice_tries, |
6162 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 6163 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 6164 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
6165 | set_table_entry(&table[11], "name", sd->name, |
6166 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
6167 | /* &table[12] is terminator */ | |
e692ab53 NP |
6168 | |
6169 | return table; | |
6170 | } | |
6171 | ||
9a4e7159 | 6172 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
6173 | { |
6174 | struct ctl_table *entry, *table; | |
6175 | struct sched_domain *sd; | |
6176 | int domain_num = 0, i; | |
6177 | char buf[32]; | |
6178 | ||
6179 | for_each_domain(cpu, sd) | |
6180 | domain_num++; | |
6181 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
6182 | if (table == NULL) |
6183 | return NULL; | |
e692ab53 NP |
6184 | |
6185 | i = 0; | |
6186 | for_each_domain(cpu, sd) { | |
6187 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 6188 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6189 | entry->mode = 0555; |
e692ab53 NP |
6190 | entry->child = sd_alloc_ctl_domain_table(sd); |
6191 | entry++; | |
6192 | i++; | |
6193 | } | |
6194 | return table; | |
6195 | } | |
6196 | ||
6197 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 6198 | static void register_sched_domain_sysctl(void) |
e692ab53 | 6199 | { |
6ad4c188 | 6200 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
6201 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
6202 | char buf[32]; | |
6203 | ||
7378547f MM |
6204 | WARN_ON(sd_ctl_dir[0].child); |
6205 | sd_ctl_dir[0].child = entry; | |
6206 | ||
ad1cdc1d MM |
6207 | if (entry == NULL) |
6208 | return; | |
6209 | ||
6ad4c188 | 6210 | for_each_possible_cpu(i) { |
e692ab53 | 6211 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6212 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6213 | entry->mode = 0555; |
e692ab53 | 6214 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6215 | entry++; |
e692ab53 | 6216 | } |
7378547f MM |
6217 | |
6218 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6219 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6220 | } | |
6382bc90 | 6221 | |
7378547f | 6222 | /* may be called multiple times per register */ |
6382bc90 MM |
6223 | static void unregister_sched_domain_sysctl(void) |
6224 | { | |
7378547f MM |
6225 | if (sd_sysctl_header) |
6226 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6227 | sd_sysctl_header = NULL; |
7378547f MM |
6228 | if (sd_ctl_dir[0].child) |
6229 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6230 | } |
e692ab53 | 6231 | #else |
6382bc90 MM |
6232 | static void register_sched_domain_sysctl(void) |
6233 | { | |
6234 | } | |
6235 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6236 | { |
6237 | } | |
6238 | #endif | |
6239 | ||
1f11eb6a GH |
6240 | static void set_rq_online(struct rq *rq) |
6241 | { | |
6242 | if (!rq->online) { | |
6243 | const struct sched_class *class; | |
6244 | ||
c6c4927b | 6245 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6246 | rq->online = 1; |
6247 | ||
6248 | for_each_class(class) { | |
6249 | if (class->rq_online) | |
6250 | class->rq_online(rq); | |
6251 | } | |
6252 | } | |
6253 | } | |
6254 | ||
6255 | static void set_rq_offline(struct rq *rq) | |
6256 | { | |
6257 | if (rq->online) { | |
6258 | const struct sched_class *class; | |
6259 | ||
6260 | for_each_class(class) { | |
6261 | if (class->rq_offline) | |
6262 | class->rq_offline(rq); | |
6263 | } | |
6264 | ||
c6c4927b | 6265 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6266 | rq->online = 0; |
6267 | } | |
6268 | } | |
6269 | ||
1da177e4 LT |
6270 | /* |
6271 | * migration_call - callback that gets triggered when a CPU is added. | |
6272 | * Here we can start up the necessary migration thread for the new CPU. | |
6273 | */ | |
48f24c4d IM |
6274 | static int __cpuinit |
6275 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6276 | { |
48f24c4d | 6277 | int cpu = (long)hcpu; |
1da177e4 | 6278 | unsigned long flags; |
969c7921 | 6279 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
6280 | |
6281 | switch (action) { | |
5be9361c | 6282 | |
1da177e4 | 6283 | case CPU_UP_PREPARE: |
8bb78442 | 6284 | case CPU_UP_PREPARE_FROZEN: |
a468d389 | 6285 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6286 | break; |
48f24c4d | 6287 | |
1da177e4 | 6288 | case CPU_ONLINE: |
8bb78442 | 6289 | case CPU_ONLINE_FROZEN: |
1f94ef59 | 6290 | /* Update our root-domain */ |
05fa785c | 6291 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6292 | if (rq->rd) { |
c6c4927b | 6293 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6294 | |
6295 | set_rq_online(rq); | |
1f94ef59 | 6296 | } |
05fa785c | 6297 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6298 | break; |
48f24c4d | 6299 | |
1da177e4 | 6300 | #ifdef CONFIG_HOTPLUG_CPU |
1da177e4 | 6301 | case CPU_DEAD: |
8bb78442 | 6302 | case CPU_DEAD_FROZEN: |
1da177e4 | 6303 | migrate_live_tasks(cpu); |
1da177e4 | 6304 | /* Idle task back to normal (off runqueue, low prio) */ |
05fa785c | 6305 | raw_spin_lock_irq(&rq->lock); |
2e1cb74a | 6306 | deactivate_task(rq, rq->idle, 0); |
dd41f596 IM |
6307 | __setscheduler(rq, rq->idle, SCHED_NORMAL, 0); |
6308 | rq->idle->sched_class = &idle_sched_class; | |
1da177e4 | 6309 | migrate_dead_tasks(cpu); |
05fa785c | 6310 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 LT |
6311 | migrate_nr_uninterruptible(rq); |
6312 | BUG_ON(rq->nr_running != 0); | |
dce48a84 | 6313 | calc_global_load_remove(rq); |
1da177e4 | 6314 | break; |
57d885fe | 6315 | |
08f503b0 GH |
6316 | case CPU_DYING: |
6317 | case CPU_DYING_FROZEN: | |
57d885fe | 6318 | /* Update our root-domain */ |
05fa785c | 6319 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6320 | if (rq->rd) { |
c6c4927b | 6321 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6322 | set_rq_offline(rq); |
57d885fe | 6323 | } |
05fa785c | 6324 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
57d885fe | 6325 | break; |
1da177e4 LT |
6326 | #endif |
6327 | } | |
6328 | return NOTIFY_OK; | |
6329 | } | |
6330 | ||
f38b0820 PM |
6331 | /* |
6332 | * Register at high priority so that task migration (migrate_all_tasks) | |
6333 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6334 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6335 | */ |
26c2143b | 6336 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6337 | .notifier_call = migration_call, |
50a323b7 | 6338 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6339 | }; |
6340 | ||
3a101d05 TH |
6341 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6342 | unsigned long action, void *hcpu) | |
6343 | { | |
6344 | switch (action & ~CPU_TASKS_FROZEN) { | |
6345 | case CPU_ONLINE: | |
6346 | case CPU_DOWN_FAILED: | |
6347 | set_cpu_active((long)hcpu, true); | |
6348 | return NOTIFY_OK; | |
6349 | default: | |
6350 | return NOTIFY_DONE; | |
6351 | } | |
6352 | } | |
6353 | ||
6354 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6355 | unsigned long action, void *hcpu) | |
6356 | { | |
6357 | switch (action & ~CPU_TASKS_FROZEN) { | |
6358 | case CPU_DOWN_PREPARE: | |
6359 | set_cpu_active((long)hcpu, false); | |
6360 | return NOTIFY_OK; | |
6361 | default: | |
6362 | return NOTIFY_DONE; | |
6363 | } | |
6364 | } | |
6365 | ||
7babe8db | 6366 | static int __init migration_init(void) |
1da177e4 LT |
6367 | { |
6368 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6369 | int err; |
48f24c4d | 6370 | |
3a101d05 | 6371 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6372 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6373 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6374 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6375 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6376 | |
3a101d05 TH |
6377 | /* Register cpu active notifiers */ |
6378 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6379 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6380 | ||
a004cd42 | 6381 | return 0; |
1da177e4 | 6382 | } |
7babe8db | 6383 | early_initcall(migration_init); |
1da177e4 LT |
6384 | #endif |
6385 | ||
6386 | #ifdef CONFIG_SMP | |
476f3534 | 6387 | |
3e9830dc | 6388 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6389 | |
f6630114 MT |
6390 | static __read_mostly int sched_domain_debug_enabled; |
6391 | ||
6392 | static int __init sched_domain_debug_setup(char *str) | |
6393 | { | |
6394 | sched_domain_debug_enabled = 1; | |
6395 | ||
6396 | return 0; | |
6397 | } | |
6398 | early_param("sched_debug", sched_domain_debug_setup); | |
6399 | ||
7c16ec58 | 6400 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6401 | struct cpumask *groupmask) |
1da177e4 | 6402 | { |
4dcf6aff | 6403 | struct sched_group *group = sd->groups; |
434d53b0 | 6404 | char str[256]; |
1da177e4 | 6405 | |
968ea6d8 | 6406 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6407 | cpumask_clear(groupmask); |
4dcf6aff IM |
6408 | |
6409 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6410 | ||
6411 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6412 | printk("does not load-balance\n"); |
4dcf6aff | 6413 | if (sd->parent) |
3df0fc5b PZ |
6414 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6415 | " has parent"); | |
4dcf6aff | 6416 | return -1; |
41c7ce9a NP |
6417 | } |
6418 | ||
3df0fc5b | 6419 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6420 | |
758b2cdc | 6421 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6422 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6423 | "CPU%d\n", cpu); | |
4dcf6aff | 6424 | } |
758b2cdc | 6425 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6426 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6427 | " CPU%d\n", cpu); | |
4dcf6aff | 6428 | } |
1da177e4 | 6429 | |
4dcf6aff | 6430 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6431 | do { |
4dcf6aff | 6432 | if (!group) { |
3df0fc5b PZ |
6433 | printk("\n"); |
6434 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6435 | break; |
6436 | } | |
6437 | ||
18a3885f | 6438 | if (!group->cpu_power) { |
3df0fc5b PZ |
6439 | printk(KERN_CONT "\n"); |
6440 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6441 | "set\n"); | |
4dcf6aff IM |
6442 | break; |
6443 | } | |
1da177e4 | 6444 | |
758b2cdc | 6445 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6446 | printk(KERN_CONT "\n"); |
6447 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6448 | break; |
6449 | } | |
1da177e4 | 6450 | |
758b2cdc | 6451 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6452 | printk(KERN_CONT "\n"); |
6453 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6454 | break; |
6455 | } | |
1da177e4 | 6456 | |
758b2cdc | 6457 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6458 | |
968ea6d8 | 6459 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6460 | |
3df0fc5b | 6461 | printk(KERN_CONT " %s", str); |
18a3885f | 6462 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6463 | printk(KERN_CONT " (cpu_power = %d)", |
6464 | group->cpu_power); | |
381512cf | 6465 | } |
1da177e4 | 6466 | |
4dcf6aff IM |
6467 | group = group->next; |
6468 | } while (group != sd->groups); | |
3df0fc5b | 6469 | printk(KERN_CONT "\n"); |
1da177e4 | 6470 | |
758b2cdc | 6471 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6472 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6473 | |
758b2cdc RR |
6474 | if (sd->parent && |
6475 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6476 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6477 | "of domain->span\n"); | |
4dcf6aff IM |
6478 | return 0; |
6479 | } | |
1da177e4 | 6480 | |
4dcf6aff IM |
6481 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6482 | { | |
d5dd3db1 | 6483 | cpumask_var_t groupmask; |
4dcf6aff | 6484 | int level = 0; |
1da177e4 | 6485 | |
f6630114 MT |
6486 | if (!sched_domain_debug_enabled) |
6487 | return; | |
6488 | ||
4dcf6aff IM |
6489 | if (!sd) { |
6490 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6491 | return; | |
6492 | } | |
1da177e4 | 6493 | |
4dcf6aff IM |
6494 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6495 | ||
d5dd3db1 | 6496 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6497 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6498 | return; | |
6499 | } | |
6500 | ||
4dcf6aff | 6501 | for (;;) { |
7c16ec58 | 6502 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6503 | break; |
1da177e4 LT |
6504 | level++; |
6505 | sd = sd->parent; | |
33859f7f | 6506 | if (!sd) |
4dcf6aff IM |
6507 | break; |
6508 | } | |
d5dd3db1 | 6509 | free_cpumask_var(groupmask); |
1da177e4 | 6510 | } |
6d6bc0ad | 6511 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6512 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6513 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6514 | |
1a20ff27 | 6515 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6516 | { |
758b2cdc | 6517 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6518 | return 1; |
6519 | ||
6520 | /* Following flags need at least 2 groups */ | |
6521 | if (sd->flags & (SD_LOAD_BALANCE | | |
6522 | SD_BALANCE_NEWIDLE | | |
6523 | SD_BALANCE_FORK | | |
89c4710e SS |
6524 | SD_BALANCE_EXEC | |
6525 | SD_SHARE_CPUPOWER | | |
6526 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6527 | if (sd->groups != sd->groups->next) |
6528 | return 0; | |
6529 | } | |
6530 | ||
6531 | /* Following flags don't use groups */ | |
c88d5910 | 6532 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6533 | return 0; |
6534 | ||
6535 | return 1; | |
6536 | } | |
6537 | ||
48f24c4d IM |
6538 | static int |
6539 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6540 | { |
6541 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6542 | ||
6543 | if (sd_degenerate(parent)) | |
6544 | return 1; | |
6545 | ||
758b2cdc | 6546 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6547 | return 0; |
6548 | ||
245af2c7 SS |
6549 | /* Flags needing groups don't count if only 1 group in parent */ |
6550 | if (parent->groups == parent->groups->next) { | |
6551 | pflags &= ~(SD_LOAD_BALANCE | | |
6552 | SD_BALANCE_NEWIDLE | | |
6553 | SD_BALANCE_FORK | | |
89c4710e SS |
6554 | SD_BALANCE_EXEC | |
6555 | SD_SHARE_CPUPOWER | | |
6556 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6557 | if (nr_node_ids == 1) |
6558 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6559 | } |
6560 | if (~cflags & pflags) | |
6561 | return 0; | |
6562 | ||
6563 | return 1; | |
6564 | } | |
6565 | ||
c6c4927b RR |
6566 | static void free_rootdomain(struct root_domain *rd) |
6567 | { | |
047106ad PZ |
6568 | synchronize_sched(); |
6569 | ||
68e74568 RR |
6570 | cpupri_cleanup(&rd->cpupri); |
6571 | ||
c6c4927b RR |
6572 | free_cpumask_var(rd->rto_mask); |
6573 | free_cpumask_var(rd->online); | |
6574 | free_cpumask_var(rd->span); | |
6575 | kfree(rd); | |
6576 | } | |
6577 | ||
57d885fe GH |
6578 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6579 | { | |
a0490fa3 | 6580 | struct root_domain *old_rd = NULL; |
57d885fe | 6581 | unsigned long flags; |
57d885fe | 6582 | |
05fa785c | 6583 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6584 | |
6585 | if (rq->rd) { | |
a0490fa3 | 6586 | old_rd = rq->rd; |
57d885fe | 6587 | |
c6c4927b | 6588 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6589 | set_rq_offline(rq); |
57d885fe | 6590 | |
c6c4927b | 6591 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6592 | |
a0490fa3 IM |
6593 | /* |
6594 | * If we dont want to free the old_rt yet then | |
6595 | * set old_rd to NULL to skip the freeing later | |
6596 | * in this function: | |
6597 | */ | |
6598 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6599 | old_rd = NULL; | |
57d885fe GH |
6600 | } |
6601 | ||
6602 | atomic_inc(&rd->refcount); | |
6603 | rq->rd = rd; | |
6604 | ||
c6c4927b | 6605 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6606 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6607 | set_rq_online(rq); |
57d885fe | 6608 | |
05fa785c | 6609 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6610 | |
6611 | if (old_rd) | |
6612 | free_rootdomain(old_rd); | |
57d885fe GH |
6613 | } |
6614 | ||
68c38fc3 | 6615 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6616 | { |
6617 | memset(rd, 0, sizeof(*rd)); | |
6618 | ||
68c38fc3 | 6619 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6620 | goto out; |
68c38fc3 | 6621 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6622 | goto free_span; |
68c38fc3 | 6623 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6624 | goto free_online; |
6e0534f2 | 6625 | |
68c38fc3 | 6626 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6627 | goto free_rto_mask; |
c6c4927b | 6628 | return 0; |
6e0534f2 | 6629 | |
68e74568 RR |
6630 | free_rto_mask: |
6631 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6632 | free_online: |
6633 | free_cpumask_var(rd->online); | |
6634 | free_span: | |
6635 | free_cpumask_var(rd->span); | |
0c910d28 | 6636 | out: |
c6c4927b | 6637 | return -ENOMEM; |
57d885fe GH |
6638 | } |
6639 | ||
6640 | static void init_defrootdomain(void) | |
6641 | { | |
68c38fc3 | 6642 | init_rootdomain(&def_root_domain); |
c6c4927b | 6643 | |
57d885fe GH |
6644 | atomic_set(&def_root_domain.refcount, 1); |
6645 | } | |
6646 | ||
dc938520 | 6647 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6648 | { |
6649 | struct root_domain *rd; | |
6650 | ||
6651 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6652 | if (!rd) | |
6653 | return NULL; | |
6654 | ||
68c38fc3 | 6655 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6656 | kfree(rd); |
6657 | return NULL; | |
6658 | } | |
57d885fe GH |
6659 | |
6660 | return rd; | |
6661 | } | |
6662 | ||
1da177e4 | 6663 | /* |
0eab9146 | 6664 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6665 | * hold the hotplug lock. |
6666 | */ | |
0eab9146 IM |
6667 | static void |
6668 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6669 | { |
70b97a7f | 6670 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6671 | struct sched_domain *tmp; |
6672 | ||
669c55e9 PZ |
6673 | for (tmp = sd; tmp; tmp = tmp->parent) |
6674 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6675 | ||
245af2c7 | 6676 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6677 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6678 | struct sched_domain *parent = tmp->parent; |
6679 | if (!parent) | |
6680 | break; | |
f29c9b1c | 6681 | |
1a848870 | 6682 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6683 | tmp->parent = parent->parent; |
1a848870 SS |
6684 | if (parent->parent) |
6685 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6686 | } else |
6687 | tmp = tmp->parent; | |
245af2c7 SS |
6688 | } |
6689 | ||
1a848870 | 6690 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6691 | sd = sd->parent; |
1a848870 SS |
6692 | if (sd) |
6693 | sd->child = NULL; | |
6694 | } | |
1da177e4 LT |
6695 | |
6696 | sched_domain_debug(sd, cpu); | |
6697 | ||
57d885fe | 6698 | rq_attach_root(rq, rd); |
674311d5 | 6699 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6700 | } |
6701 | ||
6702 | /* cpus with isolated domains */ | |
dcc30a35 | 6703 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6704 | |
6705 | /* Setup the mask of cpus configured for isolated domains */ | |
6706 | static int __init isolated_cpu_setup(char *str) | |
6707 | { | |
bdddd296 | 6708 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6709 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6710 | return 1; |
6711 | } | |
6712 | ||
8927f494 | 6713 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6714 | |
6715 | /* | |
6711cab4 SS |
6716 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6717 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6718 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6719 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6720 | * |
6721 | * init_sched_build_groups will build a circular linked list of the groups | |
6722 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6723 | * and ->cpu_power to 0. | |
6724 | */ | |
a616058b | 6725 | static void |
96f874e2 RR |
6726 | init_sched_build_groups(const struct cpumask *span, |
6727 | const struct cpumask *cpu_map, | |
6728 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6729 | struct sched_group **sg, |
96f874e2 RR |
6730 | struct cpumask *tmpmask), |
6731 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6732 | { |
6733 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6734 | int i; |
6735 | ||
96f874e2 | 6736 | cpumask_clear(covered); |
7c16ec58 | 6737 | |
abcd083a | 6738 | for_each_cpu(i, span) { |
6711cab4 | 6739 | struct sched_group *sg; |
7c16ec58 | 6740 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6741 | int j; |
6742 | ||
758b2cdc | 6743 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6744 | continue; |
6745 | ||
758b2cdc | 6746 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6747 | sg->cpu_power = 0; |
1da177e4 | 6748 | |
abcd083a | 6749 | for_each_cpu(j, span) { |
7c16ec58 | 6750 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6751 | continue; |
6752 | ||
96f874e2 | 6753 | cpumask_set_cpu(j, covered); |
758b2cdc | 6754 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6755 | } |
6756 | if (!first) | |
6757 | first = sg; | |
6758 | if (last) | |
6759 | last->next = sg; | |
6760 | last = sg; | |
6761 | } | |
6762 | last->next = first; | |
6763 | } | |
6764 | ||
9c1cfda2 | 6765 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6766 | |
9c1cfda2 | 6767 | #ifdef CONFIG_NUMA |
198e2f18 | 6768 | |
9c1cfda2 JH |
6769 | /** |
6770 | * find_next_best_node - find the next node to include in a sched_domain | |
6771 | * @node: node whose sched_domain we're building | |
6772 | * @used_nodes: nodes already in the sched_domain | |
6773 | * | |
41a2d6cf | 6774 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6775 | * finds the closest node not already in the @used_nodes map. |
6776 | * | |
6777 | * Should use nodemask_t. | |
6778 | */ | |
c5f59f08 | 6779 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6780 | { |
6781 | int i, n, val, min_val, best_node = 0; | |
6782 | ||
6783 | min_val = INT_MAX; | |
6784 | ||
076ac2af | 6785 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6786 | /* Start at @node */ |
076ac2af | 6787 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6788 | |
6789 | if (!nr_cpus_node(n)) | |
6790 | continue; | |
6791 | ||
6792 | /* Skip already used nodes */ | |
c5f59f08 | 6793 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6794 | continue; |
6795 | ||
6796 | /* Simple min distance search */ | |
6797 | val = node_distance(node, n); | |
6798 | ||
6799 | if (val < min_val) { | |
6800 | min_val = val; | |
6801 | best_node = n; | |
6802 | } | |
6803 | } | |
6804 | ||
c5f59f08 | 6805 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6806 | return best_node; |
6807 | } | |
6808 | ||
6809 | /** | |
6810 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6811 | * @node: node whose cpumask we're constructing | |
73486722 | 6812 | * @span: resulting cpumask |
9c1cfda2 | 6813 | * |
41a2d6cf | 6814 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6815 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6816 | * out optimally. | |
6817 | */ | |
96f874e2 | 6818 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6819 | { |
c5f59f08 | 6820 | nodemask_t used_nodes; |
48f24c4d | 6821 | int i; |
9c1cfda2 | 6822 | |
6ca09dfc | 6823 | cpumask_clear(span); |
c5f59f08 | 6824 | nodes_clear(used_nodes); |
9c1cfda2 | 6825 | |
6ca09dfc | 6826 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6827 | node_set(node, used_nodes); |
9c1cfda2 JH |
6828 | |
6829 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6830 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6831 | |
6ca09dfc | 6832 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6833 | } |
9c1cfda2 | 6834 | } |
6d6bc0ad | 6835 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6836 | |
5c45bf27 | 6837 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6838 | |
6c99e9ad RR |
6839 | /* |
6840 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6841 | * |
6842 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6843 | * and struct sched_domain. ) | |
6c99e9ad RR |
6844 | */ |
6845 | struct static_sched_group { | |
6846 | struct sched_group sg; | |
6847 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6848 | }; | |
6849 | ||
6850 | struct static_sched_domain { | |
6851 | struct sched_domain sd; | |
6852 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6853 | }; | |
6854 | ||
49a02c51 AH |
6855 | struct s_data { |
6856 | #ifdef CONFIG_NUMA | |
6857 | int sd_allnodes; | |
6858 | cpumask_var_t domainspan; | |
6859 | cpumask_var_t covered; | |
6860 | cpumask_var_t notcovered; | |
6861 | #endif | |
6862 | cpumask_var_t nodemask; | |
6863 | cpumask_var_t this_sibling_map; | |
6864 | cpumask_var_t this_core_map; | |
01a08546 | 6865 | cpumask_var_t this_book_map; |
49a02c51 AH |
6866 | cpumask_var_t send_covered; |
6867 | cpumask_var_t tmpmask; | |
6868 | struct sched_group **sched_group_nodes; | |
6869 | struct root_domain *rd; | |
6870 | }; | |
6871 | ||
2109b99e AH |
6872 | enum s_alloc { |
6873 | sa_sched_groups = 0, | |
6874 | sa_rootdomain, | |
6875 | sa_tmpmask, | |
6876 | sa_send_covered, | |
01a08546 | 6877 | sa_this_book_map, |
2109b99e AH |
6878 | sa_this_core_map, |
6879 | sa_this_sibling_map, | |
6880 | sa_nodemask, | |
6881 | sa_sched_group_nodes, | |
6882 | #ifdef CONFIG_NUMA | |
6883 | sa_notcovered, | |
6884 | sa_covered, | |
6885 | sa_domainspan, | |
6886 | #endif | |
6887 | sa_none, | |
6888 | }; | |
6889 | ||
9c1cfda2 | 6890 | /* |
48f24c4d | 6891 | * SMT sched-domains: |
9c1cfda2 | 6892 | */ |
1da177e4 | 6893 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6894 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6895 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6896 | |
41a2d6cf | 6897 | static int |
96f874e2 RR |
6898 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6899 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6900 | { |
6711cab4 | 6901 | if (sg) |
1871e52c | 6902 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6903 | return cpu; |
6904 | } | |
6d6bc0ad | 6905 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6906 | |
48f24c4d IM |
6907 | /* |
6908 | * multi-core sched-domains: | |
6909 | */ | |
1e9f28fa | 6910 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6911 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6912 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6913 | |
41a2d6cf | 6914 | static int |
96f874e2 RR |
6915 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6916 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6917 | { |
6711cab4 | 6918 | int group; |
f269893c | 6919 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 6920 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6921 | group = cpumask_first(mask); |
f269893c HC |
6922 | #else |
6923 | group = cpu; | |
6924 | #endif | |
6711cab4 | 6925 | if (sg) |
6c99e9ad | 6926 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6927 | return group; |
1e9f28fa | 6928 | } |
f269893c | 6929 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 6930 | |
01a08546 HC |
6931 | /* |
6932 | * book sched-domains: | |
6933 | */ | |
6934 | #ifdef CONFIG_SCHED_BOOK | |
6935 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
6936 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
6937 | ||
41a2d6cf | 6938 | static int |
01a08546 HC |
6939 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
6940 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6941 | { |
01a08546 HC |
6942 | int group = cpu; |
6943 | #ifdef CONFIG_SCHED_MC | |
6944 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
6945 | group = cpumask_first(mask); | |
6946 | #elif defined(CONFIG_SCHED_SMT) | |
6947 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
6948 | group = cpumask_first(mask); | |
6949 | #endif | |
6711cab4 | 6950 | if (sg) |
01a08546 HC |
6951 | *sg = &per_cpu(sched_group_book, group).sg; |
6952 | return group; | |
1e9f28fa | 6953 | } |
01a08546 | 6954 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 6955 | |
6c99e9ad RR |
6956 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6957 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6958 | |
41a2d6cf | 6959 | static int |
96f874e2 RR |
6960 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6961 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6962 | { |
6711cab4 | 6963 | int group; |
01a08546 HC |
6964 | #ifdef CONFIG_SCHED_BOOK |
6965 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
6966 | group = cpumask_first(mask); | |
6967 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 6968 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6969 | group = cpumask_first(mask); |
1e9f28fa | 6970 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6971 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6972 | group = cpumask_first(mask); |
1da177e4 | 6973 | #else |
6711cab4 | 6974 | group = cpu; |
1da177e4 | 6975 | #endif |
6711cab4 | 6976 | if (sg) |
6c99e9ad | 6977 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6978 | return group; |
1da177e4 LT |
6979 | } |
6980 | ||
6981 | #ifdef CONFIG_NUMA | |
1da177e4 | 6982 | /* |
9c1cfda2 JH |
6983 | * The init_sched_build_groups can't handle what we want to do with node |
6984 | * groups, so roll our own. Now each node has its own list of groups which | |
6985 | * gets dynamically allocated. | |
1da177e4 | 6986 | */ |
62ea9ceb | 6987 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6988 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6989 | |
62ea9ceb | 6990 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6991 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6992 | |
96f874e2 RR |
6993 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6994 | struct sched_group **sg, | |
6995 | struct cpumask *nodemask) | |
9c1cfda2 | 6996 | { |
6711cab4 SS |
6997 | int group; |
6998 | ||
6ca09dfc | 6999 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 7000 | group = cpumask_first(nodemask); |
6711cab4 SS |
7001 | |
7002 | if (sg) | |
6c99e9ad | 7003 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 7004 | return group; |
1da177e4 | 7005 | } |
6711cab4 | 7006 | |
08069033 SS |
7007 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
7008 | { | |
7009 | struct sched_group *sg = group_head; | |
7010 | int j; | |
7011 | ||
7012 | if (!sg) | |
7013 | return; | |
3a5c359a | 7014 | do { |
758b2cdc | 7015 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 7016 | struct sched_domain *sd; |
08069033 | 7017 | |
6c99e9ad | 7018 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 7019 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
7020 | /* |
7021 | * Only add "power" once for each | |
7022 | * physical package. | |
7023 | */ | |
7024 | continue; | |
7025 | } | |
08069033 | 7026 | |
18a3885f | 7027 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
7028 | } |
7029 | sg = sg->next; | |
7030 | } while (sg != group_head); | |
08069033 | 7031 | } |
0601a88d AH |
7032 | |
7033 | static int build_numa_sched_groups(struct s_data *d, | |
7034 | const struct cpumask *cpu_map, int num) | |
7035 | { | |
7036 | struct sched_domain *sd; | |
7037 | struct sched_group *sg, *prev; | |
7038 | int n, j; | |
7039 | ||
7040 | cpumask_clear(d->covered); | |
7041 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
7042 | if (cpumask_empty(d->nodemask)) { | |
7043 | d->sched_group_nodes[num] = NULL; | |
7044 | goto out; | |
7045 | } | |
7046 | ||
7047 | sched_domain_node_span(num, d->domainspan); | |
7048 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
7049 | ||
7050 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7051 | GFP_KERNEL, num); | |
7052 | if (!sg) { | |
3df0fc5b PZ |
7053 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
7054 | num); | |
0601a88d AH |
7055 | return -ENOMEM; |
7056 | } | |
7057 | d->sched_group_nodes[num] = sg; | |
7058 | ||
7059 | for_each_cpu(j, d->nodemask) { | |
7060 | sd = &per_cpu(node_domains, j).sd; | |
7061 | sd->groups = sg; | |
7062 | } | |
7063 | ||
18a3885f | 7064 | sg->cpu_power = 0; |
0601a88d AH |
7065 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
7066 | sg->next = sg; | |
7067 | cpumask_or(d->covered, d->covered, d->nodemask); | |
7068 | ||
7069 | prev = sg; | |
7070 | for (j = 0; j < nr_node_ids; j++) { | |
7071 | n = (num + j) % nr_node_ids; | |
7072 | cpumask_complement(d->notcovered, d->covered); | |
7073 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
7074 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
7075 | if (cpumask_empty(d->tmpmask)) | |
7076 | break; | |
7077 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
7078 | if (cpumask_empty(d->tmpmask)) | |
7079 | continue; | |
7080 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
7081 | GFP_KERNEL, num); | |
7082 | if (!sg) { | |
3df0fc5b PZ |
7083 | printk(KERN_WARNING |
7084 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
7085 | return -ENOMEM; |
7086 | } | |
18a3885f | 7087 | sg->cpu_power = 0; |
0601a88d AH |
7088 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
7089 | sg->next = prev->next; | |
7090 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
7091 | prev->next = sg; | |
7092 | prev = sg; | |
7093 | } | |
7094 | out: | |
7095 | return 0; | |
7096 | } | |
6d6bc0ad | 7097 | #endif /* CONFIG_NUMA */ |
1da177e4 | 7098 | |
a616058b | 7099 | #ifdef CONFIG_NUMA |
51888ca2 | 7100 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
7101 | static void free_sched_groups(const struct cpumask *cpu_map, |
7102 | struct cpumask *nodemask) | |
51888ca2 | 7103 | { |
a616058b | 7104 | int cpu, i; |
51888ca2 | 7105 | |
abcd083a | 7106 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
7107 | struct sched_group **sched_group_nodes |
7108 | = sched_group_nodes_bycpu[cpu]; | |
7109 | ||
51888ca2 SV |
7110 | if (!sched_group_nodes) |
7111 | continue; | |
7112 | ||
076ac2af | 7113 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
7114 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
7115 | ||
6ca09dfc | 7116 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 7117 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
7118 | continue; |
7119 | ||
7120 | if (sg == NULL) | |
7121 | continue; | |
7122 | sg = sg->next; | |
7123 | next_sg: | |
7124 | oldsg = sg; | |
7125 | sg = sg->next; | |
7126 | kfree(oldsg); | |
7127 | if (oldsg != sched_group_nodes[i]) | |
7128 | goto next_sg; | |
7129 | } | |
7130 | kfree(sched_group_nodes); | |
7131 | sched_group_nodes_bycpu[cpu] = NULL; | |
7132 | } | |
51888ca2 | 7133 | } |
6d6bc0ad | 7134 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
7135 | static void free_sched_groups(const struct cpumask *cpu_map, |
7136 | struct cpumask *nodemask) | |
a616058b SS |
7137 | { |
7138 | } | |
6d6bc0ad | 7139 | #endif /* CONFIG_NUMA */ |
51888ca2 | 7140 | |
89c4710e SS |
7141 | /* |
7142 | * Initialize sched groups cpu_power. | |
7143 | * | |
7144 | * cpu_power indicates the capacity of sched group, which is used while | |
7145 | * distributing the load between different sched groups in a sched domain. | |
7146 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
7147 | * there are asymmetries in the topology. If there are asymmetries, group | |
7148 | * having more cpu_power will pickup more load compared to the group having | |
7149 | * less cpu_power. | |
89c4710e SS |
7150 | */ |
7151 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
7152 | { | |
7153 | struct sched_domain *child; | |
7154 | struct sched_group *group; | |
f93e65c1 PZ |
7155 | long power; |
7156 | int weight; | |
89c4710e SS |
7157 | |
7158 | WARN_ON(!sd || !sd->groups); | |
7159 | ||
13318a71 | 7160 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
7161 | return; |
7162 | ||
aae6d3dd SS |
7163 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
7164 | ||
89c4710e SS |
7165 | child = sd->child; |
7166 | ||
18a3885f | 7167 | sd->groups->cpu_power = 0; |
5517d86b | 7168 | |
f93e65c1 PZ |
7169 | if (!child) { |
7170 | power = SCHED_LOAD_SCALE; | |
7171 | weight = cpumask_weight(sched_domain_span(sd)); | |
7172 | /* | |
7173 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
7174 | * Usually multiple threads get a better yield out of |
7175 | * that one core than a single thread would have, | |
7176 | * reflect that in sd->smt_gain. | |
f93e65c1 | 7177 | */ |
a52bfd73 PZ |
7178 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
7179 | power *= sd->smt_gain; | |
f93e65c1 | 7180 | power /= weight; |
a52bfd73 PZ |
7181 | power >>= SCHED_LOAD_SHIFT; |
7182 | } | |
18a3885f | 7183 | sd->groups->cpu_power += power; |
89c4710e SS |
7184 | return; |
7185 | } | |
7186 | ||
89c4710e | 7187 | /* |
f93e65c1 | 7188 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
7189 | */ |
7190 | group = child->groups; | |
7191 | do { | |
18a3885f | 7192 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
7193 | group = group->next; |
7194 | } while (group != child->groups); | |
7195 | } | |
7196 | ||
7c16ec58 MT |
7197 | /* |
7198 | * Initializers for schedule domains | |
7199 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
7200 | */ | |
7201 | ||
a5d8c348 IM |
7202 | #ifdef CONFIG_SCHED_DEBUG |
7203 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
7204 | #else | |
7205 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
7206 | #endif | |
7207 | ||
7c16ec58 | 7208 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 7209 | |
7c16ec58 MT |
7210 | #define SD_INIT_FUNC(type) \ |
7211 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
7212 | { \ | |
7213 | memset(sd, 0, sizeof(*sd)); \ | |
7214 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 7215 | sd->level = SD_LV_##type; \ |
a5d8c348 | 7216 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7217 | } |
7218 | ||
7219 | SD_INIT_FUNC(CPU) | |
7220 | #ifdef CONFIG_NUMA | |
7221 | SD_INIT_FUNC(ALLNODES) | |
7222 | SD_INIT_FUNC(NODE) | |
7223 | #endif | |
7224 | #ifdef CONFIG_SCHED_SMT | |
7225 | SD_INIT_FUNC(SIBLING) | |
7226 | #endif | |
7227 | #ifdef CONFIG_SCHED_MC | |
7228 | SD_INIT_FUNC(MC) | |
7229 | #endif | |
01a08546 HC |
7230 | #ifdef CONFIG_SCHED_BOOK |
7231 | SD_INIT_FUNC(BOOK) | |
7232 | #endif | |
7c16ec58 | 7233 | |
1d3504fc HS |
7234 | static int default_relax_domain_level = -1; |
7235 | ||
7236 | static int __init setup_relax_domain_level(char *str) | |
7237 | { | |
30e0e178 LZ |
7238 | unsigned long val; |
7239 | ||
7240 | val = simple_strtoul(str, NULL, 0); | |
7241 | if (val < SD_LV_MAX) | |
7242 | default_relax_domain_level = val; | |
7243 | ||
1d3504fc HS |
7244 | return 1; |
7245 | } | |
7246 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7247 | ||
7248 | static void set_domain_attribute(struct sched_domain *sd, | |
7249 | struct sched_domain_attr *attr) | |
7250 | { | |
7251 | int request; | |
7252 | ||
7253 | if (!attr || attr->relax_domain_level < 0) { | |
7254 | if (default_relax_domain_level < 0) | |
7255 | return; | |
7256 | else | |
7257 | request = default_relax_domain_level; | |
7258 | } else | |
7259 | request = attr->relax_domain_level; | |
7260 | if (request < sd->level) { | |
7261 | /* turn off idle balance on this domain */ | |
c88d5910 | 7262 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7263 | } else { |
7264 | /* turn on idle balance on this domain */ | |
c88d5910 | 7265 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7266 | } |
7267 | } | |
7268 | ||
2109b99e AH |
7269 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7270 | const struct cpumask *cpu_map) | |
7271 | { | |
7272 | switch (what) { | |
7273 | case sa_sched_groups: | |
7274 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
7275 | d->sched_group_nodes = NULL; | |
7276 | case sa_rootdomain: | |
7277 | free_rootdomain(d->rd); /* fall through */ | |
7278 | case sa_tmpmask: | |
7279 | free_cpumask_var(d->tmpmask); /* fall through */ | |
7280 | case sa_send_covered: | |
7281 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
7282 | case sa_this_book_map: |
7283 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
7284 | case sa_this_core_map: |
7285 | free_cpumask_var(d->this_core_map); /* fall through */ | |
7286 | case sa_this_sibling_map: | |
7287 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
7288 | case sa_nodemask: | |
7289 | free_cpumask_var(d->nodemask); /* fall through */ | |
7290 | case sa_sched_group_nodes: | |
d1b55138 | 7291 | #ifdef CONFIG_NUMA |
2109b99e AH |
7292 | kfree(d->sched_group_nodes); /* fall through */ |
7293 | case sa_notcovered: | |
7294 | free_cpumask_var(d->notcovered); /* fall through */ | |
7295 | case sa_covered: | |
7296 | free_cpumask_var(d->covered); /* fall through */ | |
7297 | case sa_domainspan: | |
7298 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 7299 | #endif |
2109b99e AH |
7300 | case sa_none: |
7301 | break; | |
7302 | } | |
7303 | } | |
3404c8d9 | 7304 | |
2109b99e AH |
7305 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7306 | const struct cpumask *cpu_map) | |
7307 | { | |
3404c8d9 | 7308 | #ifdef CONFIG_NUMA |
2109b99e AH |
7309 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
7310 | return sa_none; | |
7311 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
7312 | return sa_domainspan; | |
7313 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
7314 | return sa_covered; | |
7315 | /* Allocate the per-node list of sched groups */ | |
7316 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
7317 | sizeof(struct sched_group *), GFP_KERNEL); | |
7318 | if (!d->sched_group_nodes) { | |
3df0fc5b | 7319 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 7320 | return sa_notcovered; |
d1b55138 | 7321 | } |
2109b99e | 7322 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 7323 | #endif |
2109b99e AH |
7324 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
7325 | return sa_sched_group_nodes; | |
7326 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
7327 | return sa_nodemask; | |
7328 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
7329 | return sa_this_sibling_map; | |
01a08546 | 7330 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 7331 | return sa_this_core_map; |
01a08546 HC |
7332 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
7333 | return sa_this_book_map; | |
2109b99e AH |
7334 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
7335 | return sa_send_covered; | |
7336 | d->rd = alloc_rootdomain(); | |
7337 | if (!d->rd) { | |
3df0fc5b | 7338 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 7339 | return sa_tmpmask; |
57d885fe | 7340 | } |
2109b99e AH |
7341 | return sa_rootdomain; |
7342 | } | |
57d885fe | 7343 | |
7f4588f3 AH |
7344 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
7345 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
7346 | { | |
7347 | struct sched_domain *sd = NULL; | |
7c16ec58 | 7348 | #ifdef CONFIG_NUMA |
7f4588f3 | 7349 | struct sched_domain *parent; |
1da177e4 | 7350 | |
7f4588f3 AH |
7351 | d->sd_allnodes = 0; |
7352 | if (cpumask_weight(cpu_map) > | |
7353 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
7354 | sd = &per_cpu(allnodes_domains, i).sd; | |
7355 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 7356 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
7357 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7358 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7359 | d->sd_allnodes = 1; | |
7360 | } | |
7361 | parent = sd; | |
7362 | ||
7363 | sd = &per_cpu(node_domains, i).sd; | |
7364 | SD_INIT(sd, NODE); | |
7365 | set_domain_attribute(sd, attr); | |
7366 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
7367 | sd->parent = parent; | |
7368 | if (parent) | |
7369 | parent->child = sd; | |
7370 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 7371 | #endif |
7f4588f3 AH |
7372 | return sd; |
7373 | } | |
1da177e4 | 7374 | |
87cce662 AH |
7375 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7376 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7377 | struct sched_domain *parent, int i) | |
7378 | { | |
7379 | struct sched_domain *sd; | |
7380 | sd = &per_cpu(phys_domains, i).sd; | |
7381 | SD_INIT(sd, CPU); | |
7382 | set_domain_attribute(sd, attr); | |
7383 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
7384 | sd->parent = parent; | |
7385 | if (parent) | |
7386 | parent->child = sd; | |
7387 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7388 | return sd; | |
7389 | } | |
1da177e4 | 7390 | |
01a08546 HC |
7391 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
7392 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7393 | struct sched_domain *parent, int i) | |
7394 | { | |
7395 | struct sched_domain *sd = parent; | |
7396 | #ifdef CONFIG_SCHED_BOOK | |
7397 | sd = &per_cpu(book_domains, i).sd; | |
7398 | SD_INIT(sd, BOOK); | |
7399 | set_domain_attribute(sd, attr); | |
7400 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
7401 | sd->parent = parent; | |
7402 | parent->child = sd; | |
7403 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7404 | #endif | |
7405 | return sd; | |
7406 | } | |
7407 | ||
410c4081 AH |
7408 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7409 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7410 | struct sched_domain *parent, int i) | |
7411 | { | |
7412 | struct sched_domain *sd = parent; | |
1e9f28fa | 7413 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7414 | sd = &per_cpu(core_domains, i).sd; |
7415 | SD_INIT(sd, MC); | |
7416 | set_domain_attribute(sd, attr); | |
7417 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7418 | sd->parent = parent; | |
7419 | parent->child = sd; | |
7420 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7421 | #endif |
410c4081 AH |
7422 | return sd; |
7423 | } | |
1e9f28fa | 7424 | |
d8173535 AH |
7425 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7426 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7427 | struct sched_domain *parent, int i) | |
7428 | { | |
7429 | struct sched_domain *sd = parent; | |
1da177e4 | 7430 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7431 | sd = &per_cpu(cpu_domains, i).sd; |
7432 | SD_INIT(sd, SIBLING); | |
7433 | set_domain_attribute(sd, attr); | |
7434 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7435 | sd->parent = parent; | |
7436 | parent->child = sd; | |
7437 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7438 | #endif |
d8173535 AH |
7439 | return sd; |
7440 | } | |
1da177e4 | 7441 | |
0e8e85c9 AH |
7442 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7443 | const struct cpumask *cpu_map, int cpu) | |
7444 | { | |
7445 | switch (l) { | |
1da177e4 | 7446 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7447 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7448 | cpumask_and(d->this_sibling_map, cpu_map, | |
7449 | topology_thread_cpumask(cpu)); | |
7450 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7451 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7452 | &cpu_to_cpu_group, | |
7453 | d->send_covered, d->tmpmask); | |
7454 | break; | |
1da177e4 | 7455 | #endif |
1e9f28fa | 7456 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7457 | case SD_LV_MC: /* set up multi-core groups */ |
7458 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7459 | if (cpu == cpumask_first(d->this_core_map)) | |
7460 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7461 | &cpu_to_core_group, | |
7462 | d->send_covered, d->tmpmask); | |
7463 | break; | |
01a08546 HC |
7464 | #endif |
7465 | #ifdef CONFIG_SCHED_BOOK | |
7466 | case SD_LV_BOOK: /* set up book groups */ | |
7467 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7468 | if (cpu == cpumask_first(d->this_book_map)) | |
7469 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7470 | &cpu_to_book_group, | |
7471 | d->send_covered, d->tmpmask); | |
7472 | break; | |
1e9f28fa | 7473 | #endif |
86548096 AH |
7474 | case SD_LV_CPU: /* set up physical groups */ |
7475 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7476 | if (!cpumask_empty(d->nodemask)) | |
7477 | init_sched_build_groups(d->nodemask, cpu_map, | |
7478 | &cpu_to_phys_group, | |
7479 | d->send_covered, d->tmpmask); | |
7480 | break; | |
1da177e4 | 7481 | #ifdef CONFIG_NUMA |
de616e36 AH |
7482 | case SD_LV_ALLNODES: |
7483 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7484 | d->send_covered, d->tmpmask); | |
7485 | break; | |
7486 | #endif | |
0e8e85c9 AH |
7487 | default: |
7488 | break; | |
7c16ec58 | 7489 | } |
0e8e85c9 | 7490 | } |
9c1cfda2 | 7491 | |
2109b99e AH |
7492 | /* |
7493 | * Build sched domains for a given set of cpus and attach the sched domains | |
7494 | * to the individual cpus | |
7495 | */ | |
7496 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7497 | struct sched_domain_attr *attr) | |
7498 | { | |
7499 | enum s_alloc alloc_state = sa_none; | |
7500 | struct s_data d; | |
294b0c96 | 7501 | struct sched_domain *sd; |
2109b99e | 7502 | int i; |
7c16ec58 | 7503 | #ifdef CONFIG_NUMA |
2109b99e | 7504 | d.sd_allnodes = 0; |
7c16ec58 | 7505 | #endif |
9c1cfda2 | 7506 | |
2109b99e AH |
7507 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7508 | if (alloc_state != sa_rootdomain) | |
7509 | goto error; | |
7510 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7511 | |
1da177e4 | 7512 | /* |
1a20ff27 | 7513 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7514 | */ |
abcd083a | 7515 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7516 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7517 | cpu_map); | |
9761eea8 | 7518 | |
7f4588f3 | 7519 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7520 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7521 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7522 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7523 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7524 | } |
9c1cfda2 | 7525 | |
abcd083a | 7526 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7527 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7528 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7529 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7530 | } |
9c1cfda2 | 7531 | |
1da177e4 | 7532 | /* Set up physical groups */ |
86548096 AH |
7533 | for (i = 0; i < nr_node_ids; i++) |
7534 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7535 | |
1da177e4 LT |
7536 | #ifdef CONFIG_NUMA |
7537 | /* Set up node groups */ | |
de616e36 AH |
7538 | if (d.sd_allnodes) |
7539 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7540 | |
0601a88d AH |
7541 | for (i = 0; i < nr_node_ids; i++) |
7542 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7543 | goto error; |
1da177e4 LT |
7544 | #endif |
7545 | ||
7546 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7547 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7548 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7549 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7550 | init_sched_groups_power(i, sd); |
5c45bf27 | 7551 | } |
1da177e4 | 7552 | #endif |
1e9f28fa | 7553 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7554 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7555 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7556 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7557 | } |
7558 | #endif | |
01a08546 HC |
7559 | #ifdef CONFIG_SCHED_BOOK |
7560 | for_each_cpu(i, cpu_map) { | |
7561 | sd = &per_cpu(book_domains, i).sd; | |
7562 | init_sched_groups_power(i, sd); | |
7563 | } | |
7564 | #endif | |
1e9f28fa | 7565 | |
abcd083a | 7566 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7567 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7568 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7569 | } |
7570 | ||
9c1cfda2 | 7571 | #ifdef CONFIG_NUMA |
076ac2af | 7572 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7573 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7574 | |
49a02c51 | 7575 | if (d.sd_allnodes) { |
6711cab4 | 7576 | struct sched_group *sg; |
f712c0c7 | 7577 | |
96f874e2 | 7578 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7579 | d.tmpmask); |
f712c0c7 SS |
7580 | init_numa_sched_groups_power(sg); |
7581 | } | |
9c1cfda2 JH |
7582 | #endif |
7583 | ||
1da177e4 | 7584 | /* Attach the domains */ |
abcd083a | 7585 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7586 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7587 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7588 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7589 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7590 | #elif defined(CONFIG_SCHED_BOOK) |
7591 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7592 | #else |
6c99e9ad | 7593 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7594 | #endif |
49a02c51 | 7595 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7596 | } |
51888ca2 | 7597 | |
2109b99e AH |
7598 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7599 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7600 | return 0; | |
51888ca2 | 7601 | |
51888ca2 | 7602 | error: |
2109b99e AH |
7603 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7604 | return -ENOMEM; | |
1da177e4 | 7605 | } |
029190c5 | 7606 | |
96f874e2 | 7607 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7608 | { |
7609 | return __build_sched_domains(cpu_map, NULL); | |
7610 | } | |
7611 | ||
acc3f5d7 | 7612 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7613 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7614 | static struct sched_domain_attr *dattr_cur; |
7615 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7616 | |
7617 | /* | |
7618 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7619 | * cpumask) fails, then fallback to a single sched domain, |
7620 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7621 | */ |
4212823f | 7622 | static cpumask_var_t fallback_doms; |
029190c5 | 7623 | |
ee79d1bd HC |
7624 | /* |
7625 | * arch_update_cpu_topology lets virtualized architectures update the | |
7626 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7627 | * or 0 if it stayed the same. | |
7628 | */ | |
7629 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7630 | { |
ee79d1bd | 7631 | return 0; |
22e52b07 HC |
7632 | } |
7633 | ||
acc3f5d7 RR |
7634 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7635 | { | |
7636 | int i; | |
7637 | cpumask_var_t *doms; | |
7638 | ||
7639 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7640 | if (!doms) | |
7641 | return NULL; | |
7642 | for (i = 0; i < ndoms; i++) { | |
7643 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7644 | free_sched_domains(doms, i); | |
7645 | return NULL; | |
7646 | } | |
7647 | } | |
7648 | return doms; | |
7649 | } | |
7650 | ||
7651 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7652 | { | |
7653 | unsigned int i; | |
7654 | for (i = 0; i < ndoms; i++) | |
7655 | free_cpumask_var(doms[i]); | |
7656 | kfree(doms); | |
7657 | } | |
7658 | ||
1a20ff27 | 7659 | /* |
41a2d6cf | 7660 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7661 | * For now this just excludes isolated cpus, but could be used to |
7662 | * exclude other special cases in the future. | |
1a20ff27 | 7663 | */ |
96f874e2 | 7664 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7665 | { |
7378547f MM |
7666 | int err; |
7667 | ||
22e52b07 | 7668 | arch_update_cpu_topology(); |
029190c5 | 7669 | ndoms_cur = 1; |
acc3f5d7 | 7670 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7671 | if (!doms_cur) |
acc3f5d7 RR |
7672 | doms_cur = &fallback_doms; |
7673 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7674 | dattr_cur = NULL; |
acc3f5d7 | 7675 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7676 | register_sched_domain_sysctl(); |
7378547f MM |
7677 | |
7678 | return err; | |
1a20ff27 DG |
7679 | } |
7680 | ||
96f874e2 RR |
7681 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7682 | struct cpumask *tmpmask) | |
1da177e4 | 7683 | { |
7c16ec58 | 7684 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7685 | } |
1da177e4 | 7686 | |
1a20ff27 DG |
7687 | /* |
7688 | * Detach sched domains from a group of cpus specified in cpu_map | |
7689 | * These cpus will now be attached to the NULL domain | |
7690 | */ | |
96f874e2 | 7691 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7692 | { |
96f874e2 RR |
7693 | /* Save because hotplug lock held. */ |
7694 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7695 | int i; |
7696 | ||
abcd083a | 7697 | for_each_cpu(i, cpu_map) |
57d885fe | 7698 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7699 | synchronize_sched(); |
96f874e2 | 7700 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7701 | } |
7702 | ||
1d3504fc HS |
7703 | /* handle null as "default" */ |
7704 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7705 | struct sched_domain_attr *new, int idx_new) | |
7706 | { | |
7707 | struct sched_domain_attr tmp; | |
7708 | ||
7709 | /* fast path */ | |
7710 | if (!new && !cur) | |
7711 | return 1; | |
7712 | ||
7713 | tmp = SD_ATTR_INIT; | |
7714 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7715 | new ? (new + idx_new) : &tmp, | |
7716 | sizeof(struct sched_domain_attr)); | |
7717 | } | |
7718 | ||
029190c5 PJ |
7719 | /* |
7720 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7721 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7722 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7723 | * It destroys each deleted domain and builds each new domain. | |
7724 | * | |
acc3f5d7 | 7725 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7726 | * The masks don't intersect (don't overlap.) We should setup one |
7727 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7728 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7729 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7730 | * it as it is. | |
7731 | * | |
acc3f5d7 RR |
7732 | * The passed in 'doms_new' should be allocated using |
7733 | * alloc_sched_domains. This routine takes ownership of it and will | |
7734 | * free_sched_domains it when done with it. If the caller failed the | |
7735 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7736 | * and partition_sched_domains() will fallback to the single partition | |
7737 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7738 | * |
96f874e2 | 7739 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7740 | * ndoms_new == 0 is a special case for destroying existing domains, |
7741 | * and it will not create the default domain. | |
dfb512ec | 7742 | * |
029190c5 PJ |
7743 | * Call with hotplug lock held |
7744 | */ | |
acc3f5d7 | 7745 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7746 | struct sched_domain_attr *dattr_new) |
029190c5 | 7747 | { |
dfb512ec | 7748 | int i, j, n; |
d65bd5ec | 7749 | int new_topology; |
029190c5 | 7750 | |
712555ee | 7751 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7752 | |
7378547f MM |
7753 | /* always unregister in case we don't destroy any domains */ |
7754 | unregister_sched_domain_sysctl(); | |
7755 | ||
d65bd5ec HC |
7756 | /* Let architecture update cpu core mappings. */ |
7757 | new_topology = arch_update_cpu_topology(); | |
7758 | ||
dfb512ec | 7759 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7760 | |
7761 | /* Destroy deleted domains */ | |
7762 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7763 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7764 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7765 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7766 | goto match1; |
7767 | } | |
7768 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7769 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7770 | match1: |
7771 | ; | |
7772 | } | |
7773 | ||
e761b772 MK |
7774 | if (doms_new == NULL) { |
7775 | ndoms_cur = 0; | |
acc3f5d7 | 7776 | doms_new = &fallback_doms; |
6ad4c188 | 7777 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7778 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7779 | } |
7780 | ||
029190c5 PJ |
7781 | /* Build new domains */ |
7782 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7783 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7784 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7785 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7786 | goto match2; |
7787 | } | |
7788 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7789 | __build_sched_domains(doms_new[i], |
1d3504fc | 7790 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7791 | match2: |
7792 | ; | |
7793 | } | |
7794 | ||
7795 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7796 | if (doms_cur != &fallback_doms) |
7797 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7798 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7799 | doms_cur = doms_new; |
1d3504fc | 7800 | dattr_cur = dattr_new; |
029190c5 | 7801 | ndoms_cur = ndoms_new; |
7378547f MM |
7802 | |
7803 | register_sched_domain_sysctl(); | |
a1835615 | 7804 | |
712555ee | 7805 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7806 | } |
7807 | ||
5c45bf27 | 7808 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7809 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7810 | { |
95402b38 | 7811 | get_online_cpus(); |
dfb512ec MK |
7812 | |
7813 | /* Destroy domains first to force the rebuild */ | |
7814 | partition_sched_domains(0, NULL, NULL); | |
7815 | ||
e761b772 | 7816 | rebuild_sched_domains(); |
95402b38 | 7817 | put_online_cpus(); |
5c45bf27 SS |
7818 | } |
7819 | ||
7820 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7821 | { | |
afb8a9b7 | 7822 | unsigned int level = 0; |
5c45bf27 | 7823 | |
afb8a9b7 GS |
7824 | if (sscanf(buf, "%u", &level) != 1) |
7825 | return -EINVAL; | |
7826 | ||
7827 | /* | |
7828 | * level is always be positive so don't check for | |
7829 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7830 | * What happens on 0 or 1 byte write, | |
7831 | * need to check for count as well? | |
7832 | */ | |
7833 | ||
7834 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7835 | return -EINVAL; |
7836 | ||
7837 | if (smt) | |
afb8a9b7 | 7838 | sched_smt_power_savings = level; |
5c45bf27 | 7839 | else |
afb8a9b7 | 7840 | sched_mc_power_savings = level; |
5c45bf27 | 7841 | |
c70f22d2 | 7842 | arch_reinit_sched_domains(); |
5c45bf27 | 7843 | |
c70f22d2 | 7844 | return count; |
5c45bf27 SS |
7845 | } |
7846 | ||
5c45bf27 | 7847 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7848 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7849 | struct sysdev_class_attribute *attr, |
f718cd4a | 7850 | char *page) |
5c45bf27 SS |
7851 | { |
7852 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7853 | } | |
f718cd4a | 7854 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7855 | struct sysdev_class_attribute *attr, |
48f24c4d | 7856 | const char *buf, size_t count) |
5c45bf27 SS |
7857 | { |
7858 | return sched_power_savings_store(buf, count, 0); | |
7859 | } | |
f718cd4a AK |
7860 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7861 | sched_mc_power_savings_show, | |
7862 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7863 | #endif |
7864 | ||
7865 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7866 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7867 | struct sysdev_class_attribute *attr, |
f718cd4a | 7868 | char *page) |
5c45bf27 SS |
7869 | { |
7870 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7871 | } | |
f718cd4a | 7872 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7873 | struct sysdev_class_attribute *attr, |
48f24c4d | 7874 | const char *buf, size_t count) |
5c45bf27 SS |
7875 | { |
7876 | return sched_power_savings_store(buf, count, 1); | |
7877 | } | |
f718cd4a AK |
7878 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7879 | sched_smt_power_savings_show, | |
6707de00 AB |
7880 | sched_smt_power_savings_store); |
7881 | #endif | |
7882 | ||
39aac648 | 7883 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7884 | { |
7885 | int err = 0; | |
7886 | ||
7887 | #ifdef CONFIG_SCHED_SMT | |
7888 | if (smt_capable()) | |
7889 | err = sysfs_create_file(&cls->kset.kobj, | |
7890 | &attr_sched_smt_power_savings.attr); | |
7891 | #endif | |
7892 | #ifdef CONFIG_SCHED_MC | |
7893 | if (!err && mc_capable()) | |
7894 | err = sysfs_create_file(&cls->kset.kobj, | |
7895 | &attr_sched_mc_power_savings.attr); | |
7896 | #endif | |
7897 | return err; | |
7898 | } | |
6d6bc0ad | 7899 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7900 | |
1da177e4 | 7901 | /* |
3a101d05 TH |
7902 | * Update cpusets according to cpu_active mask. If cpusets are |
7903 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7904 | * around partition_sched_domains(). | |
1da177e4 | 7905 | */ |
0b2e918a TH |
7906 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7907 | void *hcpu) | |
e761b772 | 7908 | { |
3a101d05 | 7909 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7910 | case CPU_ONLINE: |
6ad4c188 | 7911 | case CPU_DOWN_FAILED: |
3a101d05 | 7912 | cpuset_update_active_cpus(); |
e761b772 | 7913 | return NOTIFY_OK; |
3a101d05 TH |
7914 | default: |
7915 | return NOTIFY_DONE; | |
7916 | } | |
7917 | } | |
e761b772 | 7918 | |
0b2e918a TH |
7919 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7920 | void *hcpu) | |
3a101d05 TH |
7921 | { |
7922 | switch (action & ~CPU_TASKS_FROZEN) { | |
7923 | case CPU_DOWN_PREPARE: | |
7924 | cpuset_update_active_cpus(); | |
7925 | return NOTIFY_OK; | |
e761b772 MK |
7926 | default: |
7927 | return NOTIFY_DONE; | |
7928 | } | |
7929 | } | |
e761b772 MK |
7930 | |
7931 | static int update_runtime(struct notifier_block *nfb, | |
7932 | unsigned long action, void *hcpu) | |
1da177e4 | 7933 | { |
7def2be1 PZ |
7934 | int cpu = (int)(long)hcpu; |
7935 | ||
1da177e4 | 7936 | switch (action) { |
1da177e4 | 7937 | case CPU_DOWN_PREPARE: |
8bb78442 | 7938 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7939 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7940 | return NOTIFY_OK; |
7941 | ||
1da177e4 | 7942 | case CPU_DOWN_FAILED: |
8bb78442 | 7943 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7944 | case CPU_ONLINE: |
8bb78442 | 7945 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7946 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7947 | return NOTIFY_OK; |
7948 | ||
1da177e4 LT |
7949 | default: |
7950 | return NOTIFY_DONE; | |
7951 | } | |
1da177e4 | 7952 | } |
1da177e4 LT |
7953 | |
7954 | void __init sched_init_smp(void) | |
7955 | { | |
dcc30a35 RR |
7956 | cpumask_var_t non_isolated_cpus; |
7957 | ||
7958 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7959 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7960 | |
434d53b0 MT |
7961 | #if defined(CONFIG_NUMA) |
7962 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7963 | GFP_KERNEL); | |
7964 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7965 | #endif | |
95402b38 | 7966 | get_online_cpus(); |
712555ee | 7967 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7968 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7969 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7970 | if (cpumask_empty(non_isolated_cpus)) | |
7971 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7972 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7973 | put_online_cpus(); |
e761b772 | 7974 | |
3a101d05 TH |
7975 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7976 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7977 | |
7978 | /* RT runtime code needs to handle some hotplug events */ | |
7979 | hotcpu_notifier(update_runtime, 0); | |
7980 | ||
b328ca18 | 7981 | init_hrtick(); |
5c1e1767 NP |
7982 | |
7983 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7984 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7985 | BUG(); |
19978ca6 | 7986 | sched_init_granularity(); |
dcc30a35 | 7987 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7988 | |
0e3900e6 | 7989 | init_sched_rt_class(); |
1da177e4 LT |
7990 | } |
7991 | #else | |
7992 | void __init sched_init_smp(void) | |
7993 | { | |
19978ca6 | 7994 | sched_init_granularity(); |
1da177e4 LT |
7995 | } |
7996 | #endif /* CONFIG_SMP */ | |
7997 | ||
cd1bb94b AB |
7998 | const_debug unsigned int sysctl_timer_migration = 1; |
7999 | ||
1da177e4 LT |
8000 | int in_sched_functions(unsigned long addr) |
8001 | { | |
1da177e4 LT |
8002 | return in_lock_functions(addr) || |
8003 | (addr >= (unsigned long)__sched_text_start | |
8004 | && addr < (unsigned long)__sched_text_end); | |
8005 | } | |
8006 | ||
a9957449 | 8007 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
8008 | { |
8009 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 8010 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
8011 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8012 | cfs_rq->rq = rq; | |
8013 | #endif | |
67e9fb2a | 8014 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
8015 | } |
8016 | ||
fa85ae24 PZ |
8017 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
8018 | { | |
8019 | struct rt_prio_array *array; | |
8020 | int i; | |
8021 | ||
8022 | array = &rt_rq->active; | |
8023 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
8024 | INIT_LIST_HEAD(array->queue + i); | |
8025 | __clear_bit(i, array->bitmap); | |
8026 | } | |
8027 | /* delimiter for bitsearch: */ | |
8028 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
8029 | ||
052f1dc7 | 8030 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 8031 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 8032 | #ifdef CONFIG_SMP |
e864c499 | 8033 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 8034 | #endif |
48d5e258 | 8035 | #endif |
fa85ae24 PZ |
8036 | #ifdef CONFIG_SMP |
8037 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 8038 | rt_rq->overloaded = 0; |
05fa785c | 8039 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
8040 | #endif |
8041 | ||
8042 | rt_rq->rt_time = 0; | |
8043 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 8044 | rt_rq->rt_runtime = 0; |
0986b11b | 8045 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 8046 | |
052f1dc7 | 8047 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 8048 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
8049 | rt_rq->rq = rq; |
8050 | #endif | |
fa85ae24 PZ |
8051 | } |
8052 | ||
6f505b16 | 8053 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac DG |
8054 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
8055 | struct sched_entity *se, int cpu, int add, | |
8056 | struct sched_entity *parent) | |
6f505b16 | 8057 | { |
ec7dc8ac | 8058 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
8059 | tg->cfs_rq[cpu] = cfs_rq; |
8060 | init_cfs_rq(cfs_rq, rq); | |
8061 | cfs_rq->tg = tg; | |
8062 | if (add) | |
8063 | list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list); | |
8064 | ||
8065 | tg->se[cpu] = se; | |
354d60c2 DG |
8066 | /* se could be NULL for init_task_group */ |
8067 | if (!se) | |
8068 | return; | |
8069 | ||
ec7dc8ac DG |
8070 | if (!parent) |
8071 | se->cfs_rq = &rq->cfs; | |
8072 | else | |
8073 | se->cfs_rq = parent->my_q; | |
8074 | ||
6f505b16 PZ |
8075 | se->my_q = cfs_rq; |
8076 | se->load.weight = tg->shares; | |
e05510d0 | 8077 | se->load.inv_weight = 0; |
ec7dc8ac | 8078 | se->parent = parent; |
6f505b16 | 8079 | } |
052f1dc7 | 8080 | #endif |
6f505b16 | 8081 | |
052f1dc7 | 8082 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac DG |
8083 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
8084 | struct sched_rt_entity *rt_se, int cpu, int add, | |
8085 | struct sched_rt_entity *parent) | |
6f505b16 | 8086 | { |
ec7dc8ac DG |
8087 | struct rq *rq = cpu_rq(cpu); |
8088 | ||
6f505b16 PZ |
8089 | tg->rt_rq[cpu] = rt_rq; |
8090 | init_rt_rq(rt_rq, rq); | |
8091 | rt_rq->tg = tg; | |
ac086bc2 | 8092 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
8093 | if (add) |
8094 | list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list); | |
8095 | ||
8096 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
8097 | if (!rt_se) |
8098 | return; | |
8099 | ||
ec7dc8ac DG |
8100 | if (!parent) |
8101 | rt_se->rt_rq = &rq->rt; | |
8102 | else | |
8103 | rt_se->rt_rq = parent->my_q; | |
8104 | ||
6f505b16 | 8105 | rt_se->my_q = rt_rq; |
ec7dc8ac | 8106 | rt_se->parent = parent; |
6f505b16 PZ |
8107 | INIT_LIST_HEAD(&rt_se->run_list); |
8108 | } | |
8109 | #endif | |
8110 | ||
1da177e4 LT |
8111 | void __init sched_init(void) |
8112 | { | |
dd41f596 | 8113 | int i, j; |
434d53b0 MT |
8114 | unsigned long alloc_size = 0, ptr; |
8115 | ||
8116 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8117 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
8118 | #endif | |
8119 | #ifdef CONFIG_RT_GROUP_SCHED | |
8120 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8121 | #endif |
df7c8e84 | 8122 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 8123 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 8124 | #endif |
434d53b0 | 8125 | if (alloc_size) { |
36b7b6d4 | 8126 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
8127 | |
8128 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
8129 | init_task_group.se = (struct sched_entity **)ptr; | |
8130 | ptr += nr_cpu_ids * sizeof(void **); | |
8131 | ||
8132 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
8133 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 8134 | |
6d6bc0ad | 8135 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
8136 | #ifdef CONFIG_RT_GROUP_SCHED |
8137 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
8138 | ptr += nr_cpu_ids * sizeof(void **); | |
8139 | ||
8140 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
8141 | ptr += nr_cpu_ids * sizeof(void **); |
8142 | ||
6d6bc0ad | 8143 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
8144 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8145 | for_each_possible_cpu(i) { | |
8146 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
8147 | ptr += cpumask_size(); | |
8148 | } | |
8149 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 8150 | } |
dd41f596 | 8151 | |
57d885fe GH |
8152 | #ifdef CONFIG_SMP |
8153 | init_defrootdomain(); | |
8154 | #endif | |
8155 | ||
d0b27fa7 PZ |
8156 | init_rt_bandwidth(&def_rt_bandwidth, |
8157 | global_rt_period(), global_rt_runtime()); | |
8158 | ||
8159 | #ifdef CONFIG_RT_GROUP_SCHED | |
8160 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
8161 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 8162 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 8163 | |
7c941438 | 8164 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 8165 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
8166 | INIT_LIST_HEAD(&init_task_group.children); |
8167 | ||
7c941438 | 8168 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 8169 | |
4a6cc4bd JK |
8170 | #if defined CONFIG_FAIR_GROUP_SCHED && defined CONFIG_SMP |
8171 | update_shares_data = __alloc_percpu(nr_cpu_ids * sizeof(unsigned long), | |
8172 | __alignof__(unsigned long)); | |
8173 | #endif | |
0a945022 | 8174 | for_each_possible_cpu(i) { |
70b97a7f | 8175 | struct rq *rq; |
1da177e4 LT |
8176 | |
8177 | rq = cpu_rq(i); | |
05fa785c | 8178 | raw_spin_lock_init(&rq->lock); |
7897986b | 8179 | rq->nr_running = 0; |
dce48a84 TG |
8180 | rq->calc_load_active = 0; |
8181 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 8182 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 8183 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 8184 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 8185 | init_task_group.shares = init_task_group_load; |
6f505b16 | 8186 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
8187 | #ifdef CONFIG_CGROUP_SCHED |
8188 | /* | |
8189 | * How much cpu bandwidth does init_task_group get? | |
8190 | * | |
8191 | * In case of task-groups formed thr' the cgroup filesystem, it | |
8192 | * gets 100% of the cpu resources in the system. This overall | |
8193 | * system cpu resource is divided among the tasks of | |
8194 | * init_task_group and its child task-groups in a fair manner, | |
8195 | * based on each entity's (task or task-group's) weight | |
8196 | * (se->load.weight). | |
8197 | * | |
8198 | * In other words, if init_task_group has 10 tasks of weight | |
8199 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
8200 | * then A0's share of the cpu resource is: | |
8201 | * | |
0d905bca | 8202 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
8203 | * |
8204 | * We achieve this by letting init_task_group's tasks sit | |
8205 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
8206 | */ | |
ec7dc8ac | 8207 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, 1, NULL); |
052f1dc7 | 8208 | #endif |
354d60c2 DG |
8209 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
8210 | ||
8211 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 8212 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8213 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 8214 | #ifdef CONFIG_CGROUP_SCHED |
ec7dc8ac | 8215 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, 1, NULL); |
354d60c2 | 8216 | #endif |
dd41f596 | 8217 | #endif |
1da177e4 | 8218 | |
dd41f596 IM |
8219 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
8220 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
8221 | |
8222 | rq->last_load_update_tick = jiffies; | |
8223 | ||
1da177e4 | 8224 | #ifdef CONFIG_SMP |
41c7ce9a | 8225 | rq->sd = NULL; |
57d885fe | 8226 | rq->rd = NULL; |
e51fd5e2 | 8227 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 8228 | rq->post_schedule = 0; |
1da177e4 | 8229 | rq->active_balance = 0; |
dd41f596 | 8230 | rq->next_balance = jiffies; |
1da177e4 | 8231 | rq->push_cpu = 0; |
0a2966b4 | 8232 | rq->cpu = i; |
1f11eb6a | 8233 | rq->online = 0; |
eae0c9df MG |
8234 | rq->idle_stamp = 0; |
8235 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 8236 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
8237 | #ifdef CONFIG_NO_HZ |
8238 | rq->nohz_balance_kick = 0; | |
8239 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
8240 | #endif | |
1da177e4 | 8241 | #endif |
8f4d37ec | 8242 | init_rq_hrtick(rq); |
1da177e4 | 8243 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8244 | } |
8245 | ||
2dd73a4f | 8246 | set_load_weight(&init_task); |
b50f60ce | 8247 | |
e107be36 AK |
8248 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8249 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8250 | #endif | |
8251 | ||
c9819f45 | 8252 | #ifdef CONFIG_SMP |
962cf36c | 8253 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8254 | #endif |
8255 | ||
b50f60ce | 8256 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 8257 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
8258 | #endif |
8259 | ||
1da177e4 LT |
8260 | /* |
8261 | * The boot idle thread does lazy MMU switching as well: | |
8262 | */ | |
8263 | atomic_inc(&init_mm.mm_count); | |
8264 | enter_lazy_tlb(&init_mm, current); | |
8265 | ||
8266 | /* | |
8267 | * Make us the idle thread. Technically, schedule() should not be | |
8268 | * called from this thread, however somewhere below it might be, | |
8269 | * but because we are the idle thread, we just pick up running again | |
8270 | * when this runqueue becomes "idle". | |
8271 | */ | |
8272 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8273 | |
8274 | calc_load_update = jiffies + LOAD_FREQ; | |
8275 | ||
dd41f596 IM |
8276 | /* |
8277 | * During early bootup we pretend to be a normal task: | |
8278 | */ | |
8279 | current->sched_class = &fair_sched_class; | |
6892b75e | 8280 | |
6a7b3dc3 | 8281 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8282 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8283 | #ifdef CONFIG_SMP |
7d1e6a9b | 8284 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8285 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8286 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8287 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8288 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8289 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8290 | #endif |
bdddd296 RR |
8291 | /* May be allocated at isolcpus cmdline parse time */ |
8292 | if (cpu_isolated_map == NULL) | |
8293 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8294 | #endif /* SMP */ |
6a7b3dc3 | 8295 | |
cdd6c482 | 8296 | perf_event_init(); |
0d905bca | 8297 | |
6892b75e | 8298 | scheduler_running = 1; |
1da177e4 LT |
8299 | } |
8300 | ||
8301 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
8302 | static inline int preempt_count_equals(int preempt_offset) |
8303 | { | |
234da7bc | 8304 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
8305 | |
8306 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
8307 | } | |
8308 | ||
d894837f | 8309 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8310 | { |
48f24c4d | 8311 | #ifdef in_atomic |
1da177e4 LT |
8312 | static unsigned long prev_jiffy; /* ratelimiting */ |
8313 | ||
e4aafea2 FW |
8314 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8315 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8316 | return; |
8317 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8318 | return; | |
8319 | prev_jiffy = jiffies; | |
8320 | ||
3df0fc5b PZ |
8321 | printk(KERN_ERR |
8322 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8323 | file, line); | |
8324 | printk(KERN_ERR | |
8325 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8326 | in_atomic(), irqs_disabled(), | |
8327 | current->pid, current->comm); | |
aef745fc IM |
8328 | |
8329 | debug_show_held_locks(current); | |
8330 | if (irqs_disabled()) | |
8331 | print_irqtrace_events(current); | |
8332 | dump_stack(); | |
1da177e4 LT |
8333 | #endif |
8334 | } | |
8335 | EXPORT_SYMBOL(__might_sleep); | |
8336 | #endif | |
8337 | ||
8338 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8339 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8340 | { | |
8341 | int on_rq; | |
3e51f33f | 8342 | |
3a5e4dc1 AK |
8343 | on_rq = p->se.on_rq; |
8344 | if (on_rq) | |
8345 | deactivate_task(rq, p, 0); | |
8346 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8347 | if (on_rq) { | |
8348 | activate_task(rq, p, 0); | |
8349 | resched_task(rq->curr); | |
8350 | } | |
8351 | } | |
8352 | ||
1da177e4 LT |
8353 | void normalize_rt_tasks(void) |
8354 | { | |
a0f98a1c | 8355 | struct task_struct *g, *p; |
1da177e4 | 8356 | unsigned long flags; |
70b97a7f | 8357 | struct rq *rq; |
1da177e4 | 8358 | |
4cf5d77a | 8359 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8360 | do_each_thread(g, p) { |
178be793 IM |
8361 | /* |
8362 | * Only normalize user tasks: | |
8363 | */ | |
8364 | if (!p->mm) | |
8365 | continue; | |
8366 | ||
6cfb0d5d | 8367 | p->se.exec_start = 0; |
6cfb0d5d | 8368 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8369 | p->se.statistics.wait_start = 0; |
8370 | p->se.statistics.sleep_start = 0; | |
8371 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8372 | #endif |
dd41f596 IM |
8373 | |
8374 | if (!rt_task(p)) { | |
8375 | /* | |
8376 | * Renice negative nice level userspace | |
8377 | * tasks back to 0: | |
8378 | */ | |
8379 | if (TASK_NICE(p) < 0 && p->mm) | |
8380 | set_user_nice(p, 0); | |
1da177e4 | 8381 | continue; |
dd41f596 | 8382 | } |
1da177e4 | 8383 | |
1d615482 | 8384 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8385 | rq = __task_rq_lock(p); |
1da177e4 | 8386 | |
178be793 | 8387 | normalize_task(rq, p); |
3a5e4dc1 | 8388 | |
b29739f9 | 8389 | __task_rq_unlock(rq); |
1d615482 | 8390 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8391 | } while_each_thread(g, p); |
8392 | ||
4cf5d77a | 8393 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8394 | } |
8395 | ||
8396 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8397 | |
67fc4e0c | 8398 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8399 | /* |
67fc4e0c | 8400 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8401 | * |
8402 | * They can only be called when the whole system has been | |
8403 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8404 | * activity can take place. Using them for anything else would | |
8405 | * be a serious bug, and as a result, they aren't even visible | |
8406 | * under any other configuration. | |
8407 | */ | |
8408 | ||
8409 | /** | |
8410 | * curr_task - return the current task for a given cpu. | |
8411 | * @cpu: the processor in question. | |
8412 | * | |
8413 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8414 | */ | |
36c8b586 | 8415 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8416 | { |
8417 | return cpu_curr(cpu); | |
8418 | } | |
8419 | ||
67fc4e0c JW |
8420 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8421 | ||
8422 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8423 | /** |
8424 | * set_curr_task - set the current task for a given cpu. | |
8425 | * @cpu: the processor in question. | |
8426 | * @p: the task pointer to set. | |
8427 | * | |
8428 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8429 | * are serviced on a separate stack. It allows the architecture to switch the |
8430 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8431 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8432 | * and caller must save the original value of the current task (see | |
8433 | * curr_task() above) and restore that value before reenabling interrupts and | |
8434 | * re-starting the system. | |
8435 | * | |
8436 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8437 | */ | |
36c8b586 | 8438 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8439 | { |
8440 | cpu_curr(cpu) = p; | |
8441 | } | |
8442 | ||
8443 | #endif | |
29f59db3 | 8444 | |
bccbe08a PZ |
8445 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8446 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8447 | { |
8448 | int i; | |
8449 | ||
8450 | for_each_possible_cpu(i) { | |
8451 | if (tg->cfs_rq) | |
8452 | kfree(tg->cfs_rq[i]); | |
8453 | if (tg->se) | |
8454 | kfree(tg->se[i]); | |
6f505b16 PZ |
8455 | } |
8456 | ||
8457 | kfree(tg->cfs_rq); | |
8458 | kfree(tg->se); | |
6f505b16 PZ |
8459 | } |
8460 | ||
ec7dc8ac DG |
8461 | static |
8462 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8463 | { |
29f59db3 | 8464 | struct cfs_rq *cfs_rq; |
eab17229 | 8465 | struct sched_entity *se; |
9b5b7751 | 8466 | struct rq *rq; |
29f59db3 SV |
8467 | int i; |
8468 | ||
434d53b0 | 8469 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8470 | if (!tg->cfs_rq) |
8471 | goto err; | |
434d53b0 | 8472 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8473 | if (!tg->se) |
8474 | goto err; | |
052f1dc7 PZ |
8475 | |
8476 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8477 | |
8478 | for_each_possible_cpu(i) { | |
9b5b7751 | 8479 | rq = cpu_rq(i); |
29f59db3 | 8480 | |
eab17229 LZ |
8481 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8482 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8483 | if (!cfs_rq) |
8484 | goto err; | |
8485 | ||
eab17229 LZ |
8486 | se = kzalloc_node(sizeof(struct sched_entity), |
8487 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8488 | if (!se) |
dfc12eb2 | 8489 | goto err_free_rq; |
29f59db3 | 8490 | |
eab17229 | 8491 | init_tg_cfs_entry(tg, cfs_rq, se, i, 0, parent->se[i]); |
bccbe08a PZ |
8492 | } |
8493 | ||
8494 | return 1; | |
8495 | ||
49246274 | 8496 | err_free_rq: |
dfc12eb2 | 8497 | kfree(cfs_rq); |
49246274 | 8498 | err: |
bccbe08a PZ |
8499 | return 0; |
8500 | } | |
8501 | ||
8502 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8503 | { | |
8504 | list_add_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list, | |
8505 | &cpu_rq(cpu)->leaf_cfs_rq_list); | |
8506 | } | |
8507 | ||
8508 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8509 | { | |
8510 | list_del_rcu(&tg->cfs_rq[cpu]->leaf_cfs_rq_list); | |
8511 | } | |
6d6bc0ad | 8512 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8513 | static inline void free_fair_sched_group(struct task_group *tg) |
8514 | { | |
8515 | } | |
8516 | ||
ec7dc8ac DG |
8517 | static inline |
8518 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8519 | { |
8520 | return 1; | |
8521 | } | |
8522 | ||
8523 | static inline void register_fair_sched_group(struct task_group *tg, int cpu) | |
8524 | { | |
8525 | } | |
8526 | ||
8527 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) | |
8528 | { | |
8529 | } | |
6d6bc0ad | 8530 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8531 | |
8532 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8533 | static void free_rt_sched_group(struct task_group *tg) |
8534 | { | |
8535 | int i; | |
8536 | ||
d0b27fa7 PZ |
8537 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8538 | ||
bccbe08a PZ |
8539 | for_each_possible_cpu(i) { |
8540 | if (tg->rt_rq) | |
8541 | kfree(tg->rt_rq[i]); | |
8542 | if (tg->rt_se) | |
8543 | kfree(tg->rt_se[i]); | |
8544 | } | |
8545 | ||
8546 | kfree(tg->rt_rq); | |
8547 | kfree(tg->rt_se); | |
8548 | } | |
8549 | ||
ec7dc8ac DG |
8550 | static |
8551 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8552 | { |
8553 | struct rt_rq *rt_rq; | |
eab17229 | 8554 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8555 | struct rq *rq; |
8556 | int i; | |
8557 | ||
434d53b0 | 8558 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8559 | if (!tg->rt_rq) |
8560 | goto err; | |
434d53b0 | 8561 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8562 | if (!tg->rt_se) |
8563 | goto err; | |
8564 | ||
d0b27fa7 PZ |
8565 | init_rt_bandwidth(&tg->rt_bandwidth, |
8566 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8567 | |
8568 | for_each_possible_cpu(i) { | |
8569 | rq = cpu_rq(i); | |
8570 | ||
eab17229 LZ |
8571 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8572 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8573 | if (!rt_rq) |
8574 | goto err; | |
29f59db3 | 8575 | |
eab17229 LZ |
8576 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8577 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8578 | if (!rt_se) |
dfc12eb2 | 8579 | goto err_free_rq; |
29f59db3 | 8580 | |
eab17229 | 8581 | init_tg_rt_entry(tg, rt_rq, rt_se, i, 0, parent->rt_se[i]); |
29f59db3 SV |
8582 | } |
8583 | ||
bccbe08a PZ |
8584 | return 1; |
8585 | ||
49246274 | 8586 | err_free_rq: |
dfc12eb2 | 8587 | kfree(rt_rq); |
49246274 | 8588 | err: |
bccbe08a PZ |
8589 | return 0; |
8590 | } | |
8591 | ||
8592 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8593 | { | |
8594 | list_add_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list, | |
8595 | &cpu_rq(cpu)->leaf_rt_rq_list); | |
8596 | } | |
8597 | ||
8598 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8599 | { | |
8600 | list_del_rcu(&tg->rt_rq[cpu]->leaf_rt_rq_list); | |
8601 | } | |
6d6bc0ad | 8602 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8603 | static inline void free_rt_sched_group(struct task_group *tg) |
8604 | { | |
8605 | } | |
8606 | ||
ec7dc8ac DG |
8607 | static inline |
8608 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8609 | { |
8610 | return 1; | |
8611 | } | |
8612 | ||
8613 | static inline void register_rt_sched_group(struct task_group *tg, int cpu) | |
8614 | { | |
8615 | } | |
8616 | ||
8617 | static inline void unregister_rt_sched_group(struct task_group *tg, int cpu) | |
8618 | { | |
8619 | } | |
6d6bc0ad | 8620 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8621 | |
7c941438 | 8622 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8623 | static void free_sched_group(struct task_group *tg) |
8624 | { | |
8625 | free_fair_sched_group(tg); | |
8626 | free_rt_sched_group(tg); | |
8627 | kfree(tg); | |
8628 | } | |
8629 | ||
8630 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8631 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8632 | { |
8633 | struct task_group *tg; | |
8634 | unsigned long flags; | |
8635 | int i; | |
8636 | ||
8637 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8638 | if (!tg) | |
8639 | return ERR_PTR(-ENOMEM); | |
8640 | ||
ec7dc8ac | 8641 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8642 | goto err; |
8643 | ||
ec7dc8ac | 8644 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8645 | goto err; |
8646 | ||
8ed36996 | 8647 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8648 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8649 | register_fair_sched_group(tg, i); |
8650 | register_rt_sched_group(tg, i); | |
9b5b7751 | 8651 | } |
6f505b16 | 8652 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8653 | |
8654 | WARN_ON(!parent); /* root should already exist */ | |
8655 | ||
8656 | tg->parent = parent; | |
f473aa5e | 8657 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8658 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8659 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8660 | |
9b5b7751 | 8661 | return tg; |
29f59db3 SV |
8662 | |
8663 | err: | |
6f505b16 | 8664 | free_sched_group(tg); |
29f59db3 SV |
8665 | return ERR_PTR(-ENOMEM); |
8666 | } | |
8667 | ||
9b5b7751 | 8668 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8669 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8670 | { |
29f59db3 | 8671 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8672 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8673 | } |
8674 | ||
9b5b7751 | 8675 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8676 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8677 | { |
8ed36996 | 8678 | unsigned long flags; |
9b5b7751 | 8679 | int i; |
29f59db3 | 8680 | |
8ed36996 | 8681 | spin_lock_irqsave(&task_group_lock, flags); |
9b5b7751 | 8682 | for_each_possible_cpu(i) { |
bccbe08a PZ |
8683 | unregister_fair_sched_group(tg, i); |
8684 | unregister_rt_sched_group(tg, i); | |
9b5b7751 | 8685 | } |
6f505b16 | 8686 | list_del_rcu(&tg->list); |
f473aa5e | 8687 | list_del_rcu(&tg->siblings); |
8ed36996 | 8688 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8689 | |
9b5b7751 | 8690 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8691 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8692 | } |
8693 | ||
9b5b7751 | 8694 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8695 | * The caller of this function should have put the task in its new group |
8696 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8697 | * reflect its new group. | |
9b5b7751 SV |
8698 | */ |
8699 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8700 | { |
8701 | int on_rq, running; | |
8702 | unsigned long flags; | |
8703 | struct rq *rq; | |
8704 | ||
8705 | rq = task_rq_lock(tsk, &flags); | |
8706 | ||
051a1d1a | 8707 | running = task_current(rq, tsk); |
29f59db3 SV |
8708 | on_rq = tsk->se.on_rq; |
8709 | ||
0e1f3483 | 8710 | if (on_rq) |
29f59db3 | 8711 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8712 | if (unlikely(running)) |
8713 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8714 | |
810b3817 | 8715 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8716 | if (tsk->sched_class->task_move_group) |
8717 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8718 | else | |
810b3817 | 8719 | #endif |
b2b5ce02 | 8720 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8721 | |
0e1f3483 HS |
8722 | if (unlikely(running)) |
8723 | tsk->sched_class->set_curr_task(rq); | |
8724 | if (on_rq) | |
371fd7e7 | 8725 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8726 | |
29f59db3 SV |
8727 | task_rq_unlock(rq, &flags); |
8728 | } | |
7c941438 | 8729 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8730 | |
052f1dc7 | 8731 | #ifdef CONFIG_FAIR_GROUP_SCHED |
c09595f6 | 8732 | static void __set_se_shares(struct sched_entity *se, unsigned long shares) |
29f59db3 SV |
8733 | { |
8734 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
29f59db3 SV |
8735 | int on_rq; |
8736 | ||
29f59db3 | 8737 | on_rq = se->on_rq; |
62fb1851 | 8738 | if (on_rq) |
29f59db3 SV |
8739 | dequeue_entity(cfs_rq, se, 0); |
8740 | ||
8741 | se->load.weight = shares; | |
e05510d0 | 8742 | se->load.inv_weight = 0; |
29f59db3 | 8743 | |
62fb1851 | 8744 | if (on_rq) |
29f59db3 | 8745 | enqueue_entity(cfs_rq, se, 0); |
c09595f6 | 8746 | } |
62fb1851 | 8747 | |
c09595f6 PZ |
8748 | static void set_se_shares(struct sched_entity *se, unsigned long shares) |
8749 | { | |
8750 | struct cfs_rq *cfs_rq = se->cfs_rq; | |
8751 | struct rq *rq = cfs_rq->rq; | |
8752 | unsigned long flags; | |
8753 | ||
05fa785c | 8754 | raw_spin_lock_irqsave(&rq->lock, flags); |
c09595f6 | 8755 | __set_se_shares(se, shares); |
05fa785c | 8756 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
29f59db3 SV |
8757 | } |
8758 | ||
8ed36996 PZ |
8759 | static DEFINE_MUTEX(shares_mutex); |
8760 | ||
4cf86d77 | 8761 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8762 | { |
8763 | int i; | |
8ed36996 | 8764 | unsigned long flags; |
c61935fd | 8765 | |
ec7dc8ac DG |
8766 | /* |
8767 | * We can't change the weight of the root cgroup. | |
8768 | */ | |
8769 | if (!tg->se[0]) | |
8770 | return -EINVAL; | |
8771 | ||
18d95a28 PZ |
8772 | if (shares < MIN_SHARES) |
8773 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8774 | else if (shares > MAX_SHARES) |
8775 | shares = MAX_SHARES; | |
62fb1851 | 8776 | |
8ed36996 | 8777 | mutex_lock(&shares_mutex); |
9b5b7751 | 8778 | if (tg->shares == shares) |
5cb350ba | 8779 | goto done; |
29f59db3 | 8780 | |
8ed36996 | 8781 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8782 | for_each_possible_cpu(i) |
8783 | unregister_fair_sched_group(tg, i); | |
f473aa5e | 8784 | list_del_rcu(&tg->siblings); |
8ed36996 | 8785 | spin_unlock_irqrestore(&task_group_lock, flags); |
6b2d7700 SV |
8786 | |
8787 | /* wait for any ongoing reference to this group to finish */ | |
8788 | synchronize_sched(); | |
8789 | ||
8790 | /* | |
8791 | * Now we are free to modify the group's share on each cpu | |
8792 | * w/o tripping rebalance_share or load_balance_fair. | |
8793 | */ | |
9b5b7751 | 8794 | tg->shares = shares; |
c09595f6 PZ |
8795 | for_each_possible_cpu(i) { |
8796 | /* | |
8797 | * force a rebalance | |
8798 | */ | |
8799 | cfs_rq_set_shares(tg->cfs_rq[i], 0); | |
cb4ad1ff | 8800 | set_se_shares(tg->se[i], shares); |
c09595f6 | 8801 | } |
29f59db3 | 8802 | |
6b2d7700 SV |
8803 | /* |
8804 | * Enable load balance activity on this group, by inserting it back on | |
8805 | * each cpu's rq->leaf_cfs_rq_list. | |
8806 | */ | |
8ed36996 | 8807 | spin_lock_irqsave(&task_group_lock, flags); |
bccbe08a PZ |
8808 | for_each_possible_cpu(i) |
8809 | register_fair_sched_group(tg, i); | |
f473aa5e | 8810 | list_add_rcu(&tg->siblings, &tg->parent->children); |
8ed36996 | 8811 | spin_unlock_irqrestore(&task_group_lock, flags); |
5cb350ba | 8812 | done: |
8ed36996 | 8813 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8814 | return 0; |
29f59db3 SV |
8815 | } |
8816 | ||
5cb350ba DG |
8817 | unsigned long sched_group_shares(struct task_group *tg) |
8818 | { | |
8819 | return tg->shares; | |
8820 | } | |
052f1dc7 | 8821 | #endif |
5cb350ba | 8822 | |
052f1dc7 | 8823 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8824 | /* |
9f0c1e56 | 8825 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8826 | */ |
9f0c1e56 PZ |
8827 | static DEFINE_MUTEX(rt_constraints_mutex); |
8828 | ||
8829 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8830 | { | |
8831 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8832 | return 1ULL << 20; |
9f0c1e56 | 8833 | |
9a7e0b18 | 8834 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8835 | } |
8836 | ||
9a7e0b18 PZ |
8837 | /* Must be called with tasklist_lock held */ |
8838 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8839 | { |
9a7e0b18 | 8840 | struct task_struct *g, *p; |
b40b2e8e | 8841 | |
9a7e0b18 PZ |
8842 | do_each_thread(g, p) { |
8843 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8844 | return 1; | |
8845 | } while_each_thread(g, p); | |
b40b2e8e | 8846 | |
9a7e0b18 PZ |
8847 | return 0; |
8848 | } | |
b40b2e8e | 8849 | |
9a7e0b18 PZ |
8850 | struct rt_schedulable_data { |
8851 | struct task_group *tg; | |
8852 | u64 rt_period; | |
8853 | u64 rt_runtime; | |
8854 | }; | |
b40b2e8e | 8855 | |
9a7e0b18 PZ |
8856 | static int tg_schedulable(struct task_group *tg, void *data) |
8857 | { | |
8858 | struct rt_schedulable_data *d = data; | |
8859 | struct task_group *child; | |
8860 | unsigned long total, sum = 0; | |
8861 | u64 period, runtime; | |
b40b2e8e | 8862 | |
9a7e0b18 PZ |
8863 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8864 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8865 | |
9a7e0b18 PZ |
8866 | if (tg == d->tg) { |
8867 | period = d->rt_period; | |
8868 | runtime = d->rt_runtime; | |
b40b2e8e | 8869 | } |
b40b2e8e | 8870 | |
4653f803 PZ |
8871 | /* |
8872 | * Cannot have more runtime than the period. | |
8873 | */ | |
8874 | if (runtime > period && runtime != RUNTIME_INF) | |
8875 | return -EINVAL; | |
6f505b16 | 8876 | |
4653f803 PZ |
8877 | /* |
8878 | * Ensure we don't starve existing RT tasks. | |
8879 | */ | |
9a7e0b18 PZ |
8880 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8881 | return -EBUSY; | |
6f505b16 | 8882 | |
9a7e0b18 | 8883 | total = to_ratio(period, runtime); |
6f505b16 | 8884 | |
4653f803 PZ |
8885 | /* |
8886 | * Nobody can have more than the global setting allows. | |
8887 | */ | |
8888 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8889 | return -EINVAL; | |
6f505b16 | 8890 | |
4653f803 PZ |
8891 | /* |
8892 | * The sum of our children's runtime should not exceed our own. | |
8893 | */ | |
9a7e0b18 PZ |
8894 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8895 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8896 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8897 | |
9a7e0b18 PZ |
8898 | if (child == d->tg) { |
8899 | period = d->rt_period; | |
8900 | runtime = d->rt_runtime; | |
8901 | } | |
6f505b16 | 8902 | |
9a7e0b18 | 8903 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8904 | } |
6f505b16 | 8905 | |
9a7e0b18 PZ |
8906 | if (sum > total) |
8907 | return -EINVAL; | |
8908 | ||
8909 | return 0; | |
6f505b16 PZ |
8910 | } |
8911 | ||
9a7e0b18 | 8912 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8913 | { |
9a7e0b18 PZ |
8914 | struct rt_schedulable_data data = { |
8915 | .tg = tg, | |
8916 | .rt_period = period, | |
8917 | .rt_runtime = runtime, | |
8918 | }; | |
8919 | ||
8920 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8921 | } |
8922 | ||
d0b27fa7 PZ |
8923 | static int tg_set_bandwidth(struct task_group *tg, |
8924 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8925 | { |
ac086bc2 | 8926 | int i, err = 0; |
9f0c1e56 | 8927 | |
9f0c1e56 | 8928 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8929 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8930 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8931 | if (err) | |
9f0c1e56 | 8932 | goto unlock; |
ac086bc2 | 8933 | |
0986b11b | 8934 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8935 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8936 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8937 | |
8938 | for_each_possible_cpu(i) { | |
8939 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8940 | ||
0986b11b | 8941 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8942 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8943 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8944 | } |
0986b11b | 8945 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8946 | unlock: |
521f1a24 | 8947 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8948 | mutex_unlock(&rt_constraints_mutex); |
8949 | ||
8950 | return err; | |
6f505b16 PZ |
8951 | } |
8952 | ||
d0b27fa7 PZ |
8953 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8954 | { | |
8955 | u64 rt_runtime, rt_period; | |
8956 | ||
8957 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8958 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8959 | if (rt_runtime_us < 0) | |
8960 | rt_runtime = RUNTIME_INF; | |
8961 | ||
8962 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8963 | } | |
8964 | ||
9f0c1e56 PZ |
8965 | long sched_group_rt_runtime(struct task_group *tg) |
8966 | { | |
8967 | u64 rt_runtime_us; | |
8968 | ||
d0b27fa7 | 8969 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8970 | return -1; |
8971 | ||
d0b27fa7 | 8972 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8973 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8974 | return rt_runtime_us; | |
8975 | } | |
d0b27fa7 PZ |
8976 | |
8977 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8978 | { | |
8979 | u64 rt_runtime, rt_period; | |
8980 | ||
8981 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8982 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8983 | ||
619b0488 R |
8984 | if (rt_period == 0) |
8985 | return -EINVAL; | |
8986 | ||
d0b27fa7 PZ |
8987 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8988 | } | |
8989 | ||
8990 | long sched_group_rt_period(struct task_group *tg) | |
8991 | { | |
8992 | u64 rt_period_us; | |
8993 | ||
8994 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8995 | do_div(rt_period_us, NSEC_PER_USEC); | |
8996 | return rt_period_us; | |
8997 | } | |
8998 | ||
8999 | static int sched_rt_global_constraints(void) | |
9000 | { | |
4653f803 | 9001 | u64 runtime, period; |
d0b27fa7 PZ |
9002 | int ret = 0; |
9003 | ||
ec5d4989 HS |
9004 | if (sysctl_sched_rt_period <= 0) |
9005 | return -EINVAL; | |
9006 | ||
4653f803 PZ |
9007 | runtime = global_rt_runtime(); |
9008 | period = global_rt_period(); | |
9009 | ||
9010 | /* | |
9011 | * Sanity check on the sysctl variables. | |
9012 | */ | |
9013 | if (runtime > period && runtime != RUNTIME_INF) | |
9014 | return -EINVAL; | |
10b612f4 | 9015 | |
d0b27fa7 | 9016 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 9017 | read_lock(&tasklist_lock); |
4653f803 | 9018 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 9019 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
9020 | mutex_unlock(&rt_constraints_mutex); |
9021 | ||
9022 | return ret; | |
9023 | } | |
54e99124 DG |
9024 | |
9025 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
9026 | { | |
9027 | /* Don't accept realtime tasks when there is no way for them to run */ | |
9028 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
9029 | return 0; | |
9030 | ||
9031 | return 1; | |
9032 | } | |
9033 | ||
6d6bc0ad | 9034 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9035 | static int sched_rt_global_constraints(void) |
9036 | { | |
ac086bc2 PZ |
9037 | unsigned long flags; |
9038 | int i; | |
9039 | ||
ec5d4989 HS |
9040 | if (sysctl_sched_rt_period <= 0) |
9041 | return -EINVAL; | |
9042 | ||
60aa605d PZ |
9043 | /* |
9044 | * There's always some RT tasks in the root group | |
9045 | * -- migration, kstopmachine etc.. | |
9046 | */ | |
9047 | if (sysctl_sched_rt_runtime == 0) | |
9048 | return -EBUSY; | |
9049 | ||
0986b11b | 9050 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
9051 | for_each_possible_cpu(i) { |
9052 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
9053 | ||
0986b11b | 9054 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 9055 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 9056 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 9057 | } |
0986b11b | 9058 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 9059 | |
d0b27fa7 PZ |
9060 | return 0; |
9061 | } | |
6d6bc0ad | 9062 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
9063 | |
9064 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 9065 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
9066 | loff_t *ppos) |
9067 | { | |
9068 | int ret; | |
9069 | int old_period, old_runtime; | |
9070 | static DEFINE_MUTEX(mutex); | |
9071 | ||
9072 | mutex_lock(&mutex); | |
9073 | old_period = sysctl_sched_rt_period; | |
9074 | old_runtime = sysctl_sched_rt_runtime; | |
9075 | ||
8d65af78 | 9076 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
9077 | |
9078 | if (!ret && write) { | |
9079 | ret = sched_rt_global_constraints(); | |
9080 | if (ret) { | |
9081 | sysctl_sched_rt_period = old_period; | |
9082 | sysctl_sched_rt_runtime = old_runtime; | |
9083 | } else { | |
9084 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
9085 | def_rt_bandwidth.rt_period = | |
9086 | ns_to_ktime(global_rt_period()); | |
9087 | } | |
9088 | } | |
9089 | mutex_unlock(&mutex); | |
9090 | ||
9091 | return ret; | |
9092 | } | |
68318b8e | 9093 | |
052f1dc7 | 9094 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
9095 | |
9096 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 9097 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 9098 | { |
2b01dfe3 PM |
9099 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
9100 | struct task_group, css); | |
68318b8e SV |
9101 | } |
9102 | ||
9103 | static struct cgroup_subsys_state * | |
2b01dfe3 | 9104 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 9105 | { |
ec7dc8ac | 9106 | struct task_group *tg, *parent; |
68318b8e | 9107 | |
2b01dfe3 | 9108 | if (!cgrp->parent) { |
68318b8e | 9109 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
9110 | return &init_task_group.css; |
9111 | } | |
9112 | ||
ec7dc8ac DG |
9113 | parent = cgroup_tg(cgrp->parent); |
9114 | tg = sched_create_group(parent); | |
68318b8e SV |
9115 | if (IS_ERR(tg)) |
9116 | return ERR_PTR(-ENOMEM); | |
9117 | ||
68318b8e SV |
9118 | return &tg->css; |
9119 | } | |
9120 | ||
41a2d6cf IM |
9121 | static void |
9122 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 9123 | { |
2b01dfe3 | 9124 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9125 | |
9126 | sched_destroy_group(tg); | |
9127 | } | |
9128 | ||
41a2d6cf | 9129 | static int |
be367d09 | 9130 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 9131 | { |
b68aa230 | 9132 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 9133 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
9134 | return -EINVAL; |
9135 | #else | |
68318b8e SV |
9136 | /* We don't support RT-tasks being in separate groups */ |
9137 | if (tsk->sched_class != &fair_sched_class) | |
9138 | return -EINVAL; | |
b68aa230 | 9139 | #endif |
be367d09 BB |
9140 | return 0; |
9141 | } | |
68318b8e | 9142 | |
be367d09 BB |
9143 | static int |
9144 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
9145 | struct task_struct *tsk, bool threadgroup) | |
9146 | { | |
9147 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
9148 | if (retval) | |
9149 | return retval; | |
9150 | if (threadgroup) { | |
9151 | struct task_struct *c; | |
9152 | rcu_read_lock(); | |
9153 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9154 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
9155 | if (retval) { | |
9156 | rcu_read_unlock(); | |
9157 | return retval; | |
9158 | } | |
9159 | } | |
9160 | rcu_read_unlock(); | |
9161 | } | |
68318b8e SV |
9162 | return 0; |
9163 | } | |
9164 | ||
9165 | static void | |
2b01dfe3 | 9166 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
9167 | struct cgroup *old_cont, struct task_struct *tsk, |
9168 | bool threadgroup) | |
68318b8e SV |
9169 | { |
9170 | sched_move_task(tsk); | |
be367d09 BB |
9171 | if (threadgroup) { |
9172 | struct task_struct *c; | |
9173 | rcu_read_lock(); | |
9174 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
9175 | sched_move_task(c); | |
9176 | } | |
9177 | rcu_read_unlock(); | |
9178 | } | |
68318b8e SV |
9179 | } |
9180 | ||
052f1dc7 | 9181 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 9182 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 9183 | u64 shareval) |
68318b8e | 9184 | { |
2b01dfe3 | 9185 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
9186 | } |
9187 | ||
f4c753b7 | 9188 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 9189 | { |
2b01dfe3 | 9190 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
9191 | |
9192 | return (u64) tg->shares; | |
9193 | } | |
6d6bc0ad | 9194 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 9195 | |
052f1dc7 | 9196 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 9197 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 9198 | s64 val) |
6f505b16 | 9199 | { |
06ecb27c | 9200 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
9201 | } |
9202 | ||
06ecb27c | 9203 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 9204 | { |
06ecb27c | 9205 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 9206 | } |
d0b27fa7 PZ |
9207 | |
9208 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
9209 | u64 rt_period_us) | |
9210 | { | |
9211 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
9212 | } | |
9213 | ||
9214 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
9215 | { | |
9216 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
9217 | } | |
6d6bc0ad | 9218 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 9219 | |
fe5c7cc2 | 9220 | static struct cftype cpu_files[] = { |
052f1dc7 | 9221 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
9222 | { |
9223 | .name = "shares", | |
f4c753b7 PM |
9224 | .read_u64 = cpu_shares_read_u64, |
9225 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 9226 | }, |
052f1dc7 PZ |
9227 | #endif |
9228 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 9229 | { |
9f0c1e56 | 9230 | .name = "rt_runtime_us", |
06ecb27c PM |
9231 | .read_s64 = cpu_rt_runtime_read, |
9232 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 9233 | }, |
d0b27fa7 PZ |
9234 | { |
9235 | .name = "rt_period_us", | |
f4c753b7 PM |
9236 | .read_u64 = cpu_rt_period_read_uint, |
9237 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 9238 | }, |
052f1dc7 | 9239 | #endif |
68318b8e SV |
9240 | }; |
9241 | ||
9242 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
9243 | { | |
fe5c7cc2 | 9244 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
9245 | } |
9246 | ||
9247 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
9248 | .name = "cpu", |
9249 | .create = cpu_cgroup_create, | |
9250 | .destroy = cpu_cgroup_destroy, | |
9251 | .can_attach = cpu_cgroup_can_attach, | |
9252 | .attach = cpu_cgroup_attach, | |
9253 | .populate = cpu_cgroup_populate, | |
9254 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
9255 | .early_init = 1, |
9256 | }; | |
9257 | ||
052f1dc7 | 9258 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
9259 | |
9260 | #ifdef CONFIG_CGROUP_CPUACCT | |
9261 | ||
9262 | /* | |
9263 | * CPU accounting code for task groups. | |
9264 | * | |
9265 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
9266 | * (balbir@in.ibm.com). | |
9267 | */ | |
9268 | ||
934352f2 | 9269 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
9270 | struct cpuacct { |
9271 | struct cgroup_subsys_state css; | |
9272 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 9273 | u64 __percpu *cpuusage; |
ef12fefa | 9274 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 9275 | struct cpuacct *parent; |
d842de87 SV |
9276 | }; |
9277 | ||
9278 | struct cgroup_subsys cpuacct_subsys; | |
9279 | ||
9280 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 9281 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 9282 | { |
32cd756a | 9283 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
9284 | struct cpuacct, css); |
9285 | } | |
9286 | ||
9287 | /* return cpu accounting group to which this task belongs */ | |
9288 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
9289 | { | |
9290 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
9291 | struct cpuacct, css); | |
9292 | } | |
9293 | ||
9294 | /* create a new cpu accounting group */ | |
9295 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 9296 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
9297 | { |
9298 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 9299 | int i; |
d842de87 SV |
9300 | |
9301 | if (!ca) | |
ef12fefa | 9302 | goto out; |
d842de87 SV |
9303 | |
9304 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9305 | if (!ca->cpuusage) |
9306 | goto out_free_ca; | |
9307 | ||
9308 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9309 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9310 | goto out_free_counters; | |
d842de87 | 9311 | |
934352f2 BR |
9312 | if (cgrp->parent) |
9313 | ca->parent = cgroup_ca(cgrp->parent); | |
9314 | ||
d842de87 | 9315 | return &ca->css; |
ef12fefa BR |
9316 | |
9317 | out_free_counters: | |
9318 | while (--i >= 0) | |
9319 | percpu_counter_destroy(&ca->cpustat[i]); | |
9320 | free_percpu(ca->cpuusage); | |
9321 | out_free_ca: | |
9322 | kfree(ca); | |
9323 | out: | |
9324 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9325 | } |
9326 | ||
9327 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9328 | static void |
32cd756a | 9329 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9330 | { |
32cd756a | 9331 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9332 | int i; |
d842de87 | 9333 | |
ef12fefa BR |
9334 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9335 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9336 | free_percpu(ca->cpuusage); |
9337 | kfree(ca); | |
9338 | } | |
9339 | ||
720f5498 KC |
9340 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9341 | { | |
b36128c8 | 9342 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9343 | u64 data; |
9344 | ||
9345 | #ifndef CONFIG_64BIT | |
9346 | /* | |
9347 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9348 | */ | |
05fa785c | 9349 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9350 | data = *cpuusage; |
05fa785c | 9351 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9352 | #else |
9353 | data = *cpuusage; | |
9354 | #endif | |
9355 | ||
9356 | return data; | |
9357 | } | |
9358 | ||
9359 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9360 | { | |
b36128c8 | 9361 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9362 | |
9363 | #ifndef CONFIG_64BIT | |
9364 | /* | |
9365 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9366 | */ | |
05fa785c | 9367 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9368 | *cpuusage = val; |
05fa785c | 9369 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9370 | #else |
9371 | *cpuusage = val; | |
9372 | #endif | |
9373 | } | |
9374 | ||
d842de87 | 9375 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9376 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9377 | { |
32cd756a | 9378 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9379 | u64 totalcpuusage = 0; |
9380 | int i; | |
9381 | ||
720f5498 KC |
9382 | for_each_present_cpu(i) |
9383 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9384 | |
9385 | return totalcpuusage; | |
9386 | } | |
9387 | ||
0297b803 DG |
9388 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9389 | u64 reset) | |
9390 | { | |
9391 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9392 | int err = 0; | |
9393 | int i; | |
9394 | ||
9395 | if (reset) { | |
9396 | err = -EINVAL; | |
9397 | goto out; | |
9398 | } | |
9399 | ||
720f5498 KC |
9400 | for_each_present_cpu(i) |
9401 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9402 | |
0297b803 DG |
9403 | out: |
9404 | return err; | |
9405 | } | |
9406 | ||
e9515c3c KC |
9407 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9408 | struct seq_file *m) | |
9409 | { | |
9410 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9411 | u64 percpu; | |
9412 | int i; | |
9413 | ||
9414 | for_each_present_cpu(i) { | |
9415 | percpu = cpuacct_cpuusage_read(ca, i); | |
9416 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9417 | } | |
9418 | seq_printf(m, "\n"); | |
9419 | return 0; | |
9420 | } | |
9421 | ||
ef12fefa BR |
9422 | static const char *cpuacct_stat_desc[] = { |
9423 | [CPUACCT_STAT_USER] = "user", | |
9424 | [CPUACCT_STAT_SYSTEM] = "system", | |
9425 | }; | |
9426 | ||
9427 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9428 | struct cgroup_map_cb *cb) | |
9429 | { | |
9430 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9431 | int i; | |
9432 | ||
9433 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9434 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9435 | val = cputime64_to_clock_t(val); | |
9436 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9437 | } | |
9438 | return 0; | |
9439 | } | |
9440 | ||
d842de87 SV |
9441 | static struct cftype files[] = { |
9442 | { | |
9443 | .name = "usage", | |
f4c753b7 PM |
9444 | .read_u64 = cpuusage_read, |
9445 | .write_u64 = cpuusage_write, | |
d842de87 | 9446 | }, |
e9515c3c KC |
9447 | { |
9448 | .name = "usage_percpu", | |
9449 | .read_seq_string = cpuacct_percpu_seq_read, | |
9450 | }, | |
ef12fefa BR |
9451 | { |
9452 | .name = "stat", | |
9453 | .read_map = cpuacct_stats_show, | |
9454 | }, | |
d842de87 SV |
9455 | }; |
9456 | ||
32cd756a | 9457 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9458 | { |
32cd756a | 9459 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9460 | } |
9461 | ||
9462 | /* | |
9463 | * charge this task's execution time to its accounting group. | |
9464 | * | |
9465 | * called with rq->lock held. | |
9466 | */ | |
9467 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9468 | { | |
9469 | struct cpuacct *ca; | |
934352f2 | 9470 | int cpu; |
d842de87 | 9471 | |
c40c6f85 | 9472 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9473 | return; |
9474 | ||
934352f2 | 9475 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9476 | |
9477 | rcu_read_lock(); | |
9478 | ||
d842de87 | 9479 | ca = task_ca(tsk); |
d842de87 | 9480 | |
934352f2 | 9481 | for (; ca; ca = ca->parent) { |
b36128c8 | 9482 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9483 | *cpuusage += cputime; |
9484 | } | |
a18b83b7 BR |
9485 | |
9486 | rcu_read_unlock(); | |
d842de87 SV |
9487 | } |
9488 | ||
fa535a77 AB |
9489 | /* |
9490 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9491 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9492 | * percpu_counter_add with values large enough to always overflow the | |
9493 | * per cpu batch limit causing bad SMP scalability. | |
9494 | * | |
9495 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9496 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9497 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9498 | */ | |
9499 | #ifdef CONFIG_SMP | |
9500 | #define CPUACCT_BATCH \ | |
9501 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9502 | #else | |
9503 | #define CPUACCT_BATCH 0 | |
9504 | #endif | |
9505 | ||
ef12fefa BR |
9506 | /* |
9507 | * Charge the system/user time to the task's accounting group. | |
9508 | */ | |
9509 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9510 | enum cpuacct_stat_index idx, cputime_t val) | |
9511 | { | |
9512 | struct cpuacct *ca; | |
fa535a77 | 9513 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9514 | |
9515 | if (unlikely(!cpuacct_subsys.active)) | |
9516 | return; | |
9517 | ||
9518 | rcu_read_lock(); | |
9519 | ca = task_ca(tsk); | |
9520 | ||
9521 | do { | |
fa535a77 | 9522 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9523 | ca = ca->parent; |
9524 | } while (ca); | |
9525 | rcu_read_unlock(); | |
9526 | } | |
9527 | ||
d842de87 SV |
9528 | struct cgroup_subsys cpuacct_subsys = { |
9529 | .name = "cpuacct", | |
9530 | .create = cpuacct_create, | |
9531 | .destroy = cpuacct_destroy, | |
9532 | .populate = cpuacct_populate, | |
9533 | .subsys_id = cpuacct_subsys_id, | |
9534 | }; | |
9535 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
9536 | |
9537 | #ifndef CONFIG_SMP | |
9538 | ||
03b042bf PM |
9539 | void synchronize_sched_expedited(void) |
9540 | { | |
fc390cde | 9541 | barrier(); |
03b042bf PM |
9542 | } |
9543 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9544 | ||
9545 | #else /* #ifndef CONFIG_SMP */ | |
9546 | ||
cc631fb7 | 9547 | static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0); |
03b042bf | 9548 | |
cc631fb7 | 9549 | static int synchronize_sched_expedited_cpu_stop(void *data) |
03b042bf | 9550 | { |
969c7921 TH |
9551 | /* |
9552 | * There must be a full memory barrier on each affected CPU | |
9553 | * between the time that try_stop_cpus() is called and the | |
9554 | * time that it returns. | |
9555 | * | |
9556 | * In the current initial implementation of cpu_stop, the | |
9557 | * above condition is already met when the control reaches | |
9558 | * this point and the following smp_mb() is not strictly | |
9559 | * necessary. Do smp_mb() anyway for documentation and | |
9560 | * robustness against future implementation changes. | |
9561 | */ | |
cc631fb7 | 9562 | smp_mb(); /* See above comment block. */ |
969c7921 | 9563 | return 0; |
03b042bf | 9564 | } |
03b042bf PM |
9565 | |
9566 | /* | |
9567 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9568 | * approach to force grace period to end quickly. This consumes | |
9569 | * significant time on all CPUs, and is thus not recommended for | |
9570 | * any sort of common-case code. | |
9571 | * | |
9572 | * Note that it is illegal to call this function while holding any | |
9573 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9574 | * observe this restriction will result in deadlock. | |
9575 | */ | |
9576 | void synchronize_sched_expedited(void) | |
9577 | { | |
969c7921 | 9578 | int snap, trycount = 0; |
03b042bf PM |
9579 | |
9580 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
969c7921 | 9581 | snap = atomic_read(&synchronize_sched_expedited_count) + 1; |
03b042bf | 9582 | get_online_cpus(); |
969c7921 TH |
9583 | while (try_stop_cpus(cpu_online_mask, |
9584 | synchronize_sched_expedited_cpu_stop, | |
94458d5e | 9585 | NULL) == -EAGAIN) { |
03b042bf PM |
9586 | put_online_cpus(); |
9587 | if (trycount++ < 10) | |
9588 | udelay(trycount * num_online_cpus()); | |
9589 | else { | |
9590 | synchronize_sched(); | |
9591 | return; | |
9592 | } | |
969c7921 | 9593 | if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) { |
03b042bf PM |
9594 | smp_mb(); /* ensure test happens before caller kfree */ |
9595 | return; | |
9596 | } | |
9597 | get_online_cpus(); | |
9598 | } | |
969c7921 | 9599 | atomic_inc(&synchronize_sched_expedited_count); |
cc631fb7 | 9600 | smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */ |
03b042bf | 9601 | put_online_cpus(); |
03b042bf PM |
9602 | } |
9603 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9604 | ||
9605 | #endif /* #else #ifndef CONFIG_SMP */ |