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> |
335d7afb | 78 | #include <asm/mutex.h> |
1da177e4 | 79 | |
6e0534f2 | 80 | #include "sched_cpupri.h" |
21aa9af0 | 81 | #include "workqueue_sched.h" |
6e0534f2 | 82 | |
a8d154b0 | 83 | #define CREATE_TRACE_POINTS |
ad8d75ff | 84 | #include <trace/events/sched.h> |
a8d154b0 | 85 | |
1da177e4 LT |
86 | /* |
87 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
88 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
89 | * and back. | |
90 | */ | |
91 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
92 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
93 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
94 | ||
95 | /* | |
96 | * 'User priority' is the nice value converted to something we | |
97 | * can work with better when scaling various scheduler parameters, | |
98 | * it's a [ 0 ... 39 ] range. | |
99 | */ | |
100 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
101 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
102 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
103 | ||
104 | /* | |
d7876a08 | 105 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 106 | */ |
d6322faf | 107 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 108 | |
6aa645ea IM |
109 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
110 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
111 | ||
1da177e4 LT |
112 | /* |
113 | * These are the 'tuning knobs' of the scheduler: | |
114 | * | |
a4ec24b4 | 115 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
116 | * Timeslices get refilled after they expire. |
117 | */ | |
1da177e4 | 118 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 119 | |
d0b27fa7 PZ |
120 | /* |
121 | * single value that denotes runtime == period, ie unlimited time. | |
122 | */ | |
123 | #define RUNTIME_INF ((u64)~0ULL) | |
124 | ||
e05606d3 IM |
125 | static inline int rt_policy(int policy) |
126 | { | |
3f33a7ce | 127 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
128 | return 1; |
129 | return 0; | |
130 | } | |
131 | ||
132 | static inline int task_has_rt_policy(struct task_struct *p) | |
133 | { | |
134 | return rt_policy(p->policy); | |
135 | } | |
136 | ||
1da177e4 | 137 | /* |
6aa645ea | 138 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 139 | */ |
6aa645ea IM |
140 | struct rt_prio_array { |
141 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
142 | struct list_head queue[MAX_RT_PRIO]; | |
143 | }; | |
144 | ||
d0b27fa7 | 145 | struct rt_bandwidth { |
ea736ed5 | 146 | /* nests inside the rq lock: */ |
0986b11b | 147 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
148 | ktime_t rt_period; |
149 | u64 rt_runtime; | |
150 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
151 | }; |
152 | ||
153 | static struct rt_bandwidth def_rt_bandwidth; | |
154 | ||
155 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
156 | ||
157 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
158 | { | |
159 | struct rt_bandwidth *rt_b = | |
160 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
161 | ktime_t now; | |
162 | int overrun; | |
163 | int idle = 0; | |
164 | ||
165 | for (;;) { | |
166 | now = hrtimer_cb_get_time(timer); | |
167 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
168 | ||
169 | if (!overrun) | |
170 | break; | |
171 | ||
172 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
173 | } | |
174 | ||
175 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
176 | } | |
177 | ||
178 | static | |
179 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
180 | { | |
181 | rt_b->rt_period = ns_to_ktime(period); | |
182 | rt_b->rt_runtime = runtime; | |
183 | ||
0986b11b | 184 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 185 | |
d0b27fa7 PZ |
186 | hrtimer_init(&rt_b->rt_period_timer, |
187 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
188 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
189 | } |
190 | ||
c8bfff6d KH |
191 | static inline int rt_bandwidth_enabled(void) |
192 | { | |
193 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
194 | } |
195 | ||
196 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
197 | { | |
198 | ktime_t now; | |
199 | ||
cac64d00 | 200 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
201 | return; |
202 | ||
203 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
204 | return; | |
205 | ||
0986b11b | 206 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 207 | for (;;) { |
7f1e2ca9 PZ |
208 | unsigned long delta; |
209 | ktime_t soft, hard; | |
210 | ||
d0b27fa7 PZ |
211 | if (hrtimer_active(&rt_b->rt_period_timer)) |
212 | break; | |
213 | ||
214 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
215 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
216 | |
217 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
218 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
219 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
220 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 221 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 222 | } |
0986b11b | 223 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
224 | } |
225 | ||
226 | #ifdef CONFIG_RT_GROUP_SCHED | |
227 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
228 | { | |
229 | hrtimer_cancel(&rt_b->rt_period_timer); | |
230 | } | |
231 | #endif | |
232 | ||
712555ee HC |
233 | /* |
234 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
235 | * detach_destroy_domains and partition_sched_domains. | |
236 | */ | |
237 | static DEFINE_MUTEX(sched_domains_mutex); | |
238 | ||
7c941438 | 239 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 240 | |
68318b8e SV |
241 | #include <linux/cgroup.h> |
242 | ||
29f59db3 SV |
243 | struct cfs_rq; |
244 | ||
6f505b16 PZ |
245 | static LIST_HEAD(task_groups); |
246 | ||
29f59db3 | 247 | /* task group related information */ |
4cf86d77 | 248 | struct task_group { |
68318b8e | 249 | struct cgroup_subsys_state css; |
6c415b92 | 250 | |
052f1dc7 | 251 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
252 | /* schedulable entities of this group on each cpu */ |
253 | struct sched_entity **se; | |
254 | /* runqueue "owned" by this group on each cpu */ | |
255 | struct cfs_rq **cfs_rq; | |
256 | unsigned long shares; | |
2069dd75 PZ |
257 | |
258 | atomic_t load_weight; | |
052f1dc7 PZ |
259 | #endif |
260 | ||
261 | #ifdef CONFIG_RT_GROUP_SCHED | |
262 | struct sched_rt_entity **rt_se; | |
263 | struct rt_rq **rt_rq; | |
264 | ||
d0b27fa7 | 265 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 266 | #endif |
6b2d7700 | 267 | |
ae8393e5 | 268 | struct rcu_head rcu; |
6f505b16 | 269 | struct list_head list; |
f473aa5e PZ |
270 | |
271 | struct task_group *parent; | |
272 | struct list_head siblings; | |
273 | struct list_head children; | |
29f59db3 SV |
274 | }; |
275 | ||
eff766a6 | 276 | #define root_task_group init_task_group |
6f505b16 | 277 | |
3d4b47b4 | 278 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 279 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 280 | |
e9036b36 CG |
281 | #ifdef CONFIG_FAIR_GROUP_SCHED |
282 | ||
052f1dc7 | 283 | # define INIT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 284 | |
cb4ad1ff | 285 | /* |
2e084786 LJ |
286 | * A weight of 0 or 1 can cause arithmetics problems. |
287 | * A weight of a cfs_rq is the sum of weights of which entities | |
288 | * are queued on this cfs_rq, so a weight of a entity should not be | |
289 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
290 | * (The default weight is 1024 - so there's no practical |
291 | * limitation from this.) | |
292 | */ | |
18d95a28 | 293 | #define MIN_SHARES 2 |
2e084786 | 294 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 295 | |
052f1dc7 PZ |
296 | static int init_task_group_load = INIT_TASK_GROUP_LOAD; |
297 | #endif | |
298 | ||
29f59db3 | 299 | /* Default task group. |
3a252015 | 300 | * Every task in system belong to this group at bootup. |
29f59db3 | 301 | */ |
434d53b0 | 302 | struct task_group init_task_group; |
29f59db3 | 303 | |
7c941438 | 304 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 305 | |
6aa645ea IM |
306 | /* CFS-related fields in a runqueue */ |
307 | struct cfs_rq { | |
308 | struct load_weight load; | |
309 | unsigned long nr_running; | |
310 | ||
6aa645ea | 311 | u64 exec_clock; |
e9acbff6 | 312 | u64 min_vruntime; |
6aa645ea IM |
313 | |
314 | struct rb_root tasks_timeline; | |
315 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
316 | |
317 | struct list_head tasks; | |
318 | struct list_head *balance_iterator; | |
319 | ||
320 | /* | |
321 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
322 | * It is set to NULL otherwise (i.e when none are currently running). |
323 | */ | |
4793241b | 324 | struct sched_entity *curr, *next, *last; |
ddc97297 | 325 | |
5ac5c4d6 | 326 | unsigned int nr_spread_over; |
ddc97297 | 327 | |
62160e3f | 328 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
329 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
330 | ||
41a2d6cf IM |
331 | /* |
332 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
333 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
334 | * (like users, containers etc.) | |
335 | * | |
336 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
337 | * list is used during load balance. | |
338 | */ | |
3d4b47b4 | 339 | int on_list; |
41a2d6cf IM |
340 | struct list_head leaf_cfs_rq_list; |
341 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
342 | |
343 | #ifdef CONFIG_SMP | |
c09595f6 | 344 | /* |
c8cba857 | 345 | * the part of load.weight contributed by tasks |
c09595f6 | 346 | */ |
c8cba857 | 347 | unsigned long task_weight; |
c09595f6 | 348 | |
c8cba857 PZ |
349 | /* |
350 | * h_load = weight * f(tg) | |
351 | * | |
352 | * Where f(tg) is the recursive weight fraction assigned to | |
353 | * this group. | |
354 | */ | |
355 | unsigned long h_load; | |
c09595f6 | 356 | |
3b3d190e PT |
357 | /* |
358 | * Maintaining per-cpu shares distribution for group scheduling | |
359 | * | |
360 | * load_stamp is the last time we updated the load average | |
361 | * load_last is the last time we updated the load average and saw load | |
362 | * load_unacc_exec_time is currently unaccounted execution time | |
363 | */ | |
2069dd75 PZ |
364 | u64 load_avg; |
365 | u64 load_period; | |
3b3d190e | 366 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 367 | |
2069dd75 | 368 | unsigned long load_contribution; |
c09595f6 | 369 | #endif |
6aa645ea IM |
370 | #endif |
371 | }; | |
1da177e4 | 372 | |
6aa645ea IM |
373 | /* Real-Time classes' related field in a runqueue: */ |
374 | struct rt_rq { | |
375 | struct rt_prio_array active; | |
63489e45 | 376 | unsigned long rt_nr_running; |
052f1dc7 | 377 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
378 | struct { |
379 | int curr; /* highest queued rt task prio */ | |
398a153b | 380 | #ifdef CONFIG_SMP |
e864c499 | 381 | int next; /* next highest */ |
398a153b | 382 | #endif |
e864c499 | 383 | } highest_prio; |
6f505b16 | 384 | #endif |
fa85ae24 | 385 | #ifdef CONFIG_SMP |
73fe6aae | 386 | unsigned long rt_nr_migratory; |
a1ba4d8b | 387 | unsigned long rt_nr_total; |
a22d7fc1 | 388 | int overloaded; |
917b627d | 389 | struct plist_head pushable_tasks; |
fa85ae24 | 390 | #endif |
6f505b16 | 391 | int rt_throttled; |
fa85ae24 | 392 | u64 rt_time; |
ac086bc2 | 393 | u64 rt_runtime; |
ea736ed5 | 394 | /* Nests inside the rq lock: */ |
0986b11b | 395 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 396 | |
052f1dc7 | 397 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
398 | unsigned long rt_nr_boosted; |
399 | ||
6f505b16 PZ |
400 | struct rq *rq; |
401 | struct list_head leaf_rt_rq_list; | |
402 | struct task_group *tg; | |
6f505b16 | 403 | #endif |
6aa645ea IM |
404 | }; |
405 | ||
57d885fe GH |
406 | #ifdef CONFIG_SMP |
407 | ||
408 | /* | |
409 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
410 | * variables. Each exclusive cpuset essentially defines an island domain by |
411 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
412 | * exclusive cpuset is created, we also create and attach a new root-domain |
413 | * object. | |
414 | * | |
57d885fe GH |
415 | */ |
416 | struct root_domain { | |
417 | atomic_t refcount; | |
c6c4927b RR |
418 | cpumask_var_t span; |
419 | cpumask_var_t online; | |
637f5085 | 420 | |
0eab9146 | 421 | /* |
637f5085 GH |
422 | * The "RT overload" flag: it gets set if a CPU has more than |
423 | * one runnable RT task. | |
424 | */ | |
c6c4927b | 425 | cpumask_var_t rto_mask; |
0eab9146 | 426 | atomic_t rto_count; |
6e0534f2 | 427 | struct cpupri cpupri; |
57d885fe GH |
428 | }; |
429 | ||
dc938520 GH |
430 | /* |
431 | * By default the system creates a single root-domain with all cpus as | |
432 | * members (mimicking the global state we have today). | |
433 | */ | |
57d885fe GH |
434 | static struct root_domain def_root_domain; |
435 | ||
ed2d372c | 436 | #endif /* CONFIG_SMP */ |
57d885fe | 437 | |
1da177e4 LT |
438 | /* |
439 | * This is the main, per-CPU runqueue data structure. | |
440 | * | |
441 | * Locking rule: those places that want to lock multiple runqueues | |
442 | * (such as the load balancing or the thread migration code), lock | |
443 | * acquire operations must be ordered by ascending &runqueue. | |
444 | */ | |
70b97a7f | 445 | struct rq { |
d8016491 | 446 | /* runqueue lock: */ |
05fa785c | 447 | raw_spinlock_t lock; |
1da177e4 LT |
448 | |
449 | /* | |
450 | * nr_running and cpu_load should be in the same cacheline because | |
451 | * remote CPUs use both these fields when doing load calculation. | |
452 | */ | |
453 | unsigned long nr_running; | |
6aa645ea IM |
454 | #define CPU_LOAD_IDX_MAX 5 |
455 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 456 | unsigned long last_load_update_tick; |
46cb4b7c | 457 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 458 | u64 nohz_stamp; |
83cd4fe2 | 459 | unsigned char nohz_balance_kick; |
46cb4b7c | 460 | #endif |
a64692a3 MG |
461 | unsigned int skip_clock_update; |
462 | ||
d8016491 IM |
463 | /* capture load from *all* tasks on this cpu: */ |
464 | struct load_weight load; | |
6aa645ea IM |
465 | unsigned long nr_load_updates; |
466 | u64 nr_switches; | |
467 | ||
468 | struct cfs_rq cfs; | |
6f505b16 | 469 | struct rt_rq rt; |
6f505b16 | 470 | |
6aa645ea | 471 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
472 | /* list of leaf cfs_rq on this cpu: */ |
473 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
474 | #endif |
475 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 476 | struct list_head leaf_rt_rq_list; |
1da177e4 | 477 | #endif |
1da177e4 LT |
478 | |
479 | /* | |
480 | * This is part of a global counter where only the total sum | |
481 | * over all CPUs matters. A task can increase this counter on | |
482 | * one CPU and if it got migrated afterwards it may decrease | |
483 | * it on another CPU. Always updated under the runqueue lock: | |
484 | */ | |
485 | unsigned long nr_uninterruptible; | |
486 | ||
34f971f6 | 487 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 488 | unsigned long next_balance; |
1da177e4 | 489 | struct mm_struct *prev_mm; |
6aa645ea | 490 | |
3e51f33f | 491 | u64 clock; |
305e6835 | 492 | u64 clock_task; |
6aa645ea | 493 | |
1da177e4 LT |
494 | atomic_t nr_iowait; |
495 | ||
496 | #ifdef CONFIG_SMP | |
0eab9146 | 497 | struct root_domain *rd; |
1da177e4 LT |
498 | struct sched_domain *sd; |
499 | ||
e51fd5e2 PZ |
500 | unsigned long cpu_power; |
501 | ||
a0a522ce | 502 | unsigned char idle_at_tick; |
1da177e4 | 503 | /* For active balancing */ |
3f029d3c | 504 | int post_schedule; |
1da177e4 LT |
505 | int active_balance; |
506 | int push_cpu; | |
969c7921 | 507 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
508 | /* cpu of this runqueue: */ |
509 | int cpu; | |
1f11eb6a | 510 | int online; |
1da177e4 | 511 | |
a8a51d5e | 512 | unsigned long avg_load_per_task; |
1da177e4 | 513 | |
e9e9250b PZ |
514 | u64 rt_avg; |
515 | u64 age_stamp; | |
1b9508f6 MG |
516 | u64 idle_stamp; |
517 | u64 avg_idle; | |
1da177e4 LT |
518 | #endif |
519 | ||
aa483808 VP |
520 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
521 | u64 prev_irq_time; | |
522 | #endif | |
523 | ||
dce48a84 TG |
524 | /* calc_load related fields */ |
525 | unsigned long calc_load_update; | |
526 | long calc_load_active; | |
527 | ||
8f4d37ec | 528 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
529 | #ifdef CONFIG_SMP |
530 | int hrtick_csd_pending; | |
531 | struct call_single_data hrtick_csd; | |
532 | #endif | |
8f4d37ec PZ |
533 | struct hrtimer hrtick_timer; |
534 | #endif | |
535 | ||
1da177e4 LT |
536 | #ifdef CONFIG_SCHEDSTATS |
537 | /* latency stats */ | |
538 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
539 | unsigned long long rq_cpu_time; |
540 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
541 | |
542 | /* sys_sched_yield() stats */ | |
480b9434 | 543 | unsigned int yld_count; |
1da177e4 LT |
544 | |
545 | /* schedule() stats */ | |
480b9434 KC |
546 | unsigned int sched_switch; |
547 | unsigned int sched_count; | |
548 | unsigned int sched_goidle; | |
1da177e4 LT |
549 | |
550 | /* try_to_wake_up() stats */ | |
480b9434 KC |
551 | unsigned int ttwu_count; |
552 | unsigned int ttwu_local; | |
b8efb561 IM |
553 | |
554 | /* BKL stats */ | |
480b9434 | 555 | unsigned int bkl_count; |
1da177e4 LT |
556 | #endif |
557 | }; | |
558 | ||
f34e3b61 | 559 | static DEFINE_PER_CPU_SHARED_ALIGNED(struct rq, runqueues); |
1da177e4 | 560 | |
a64692a3 | 561 | |
1e5a7405 | 562 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags); |
dd41f596 | 563 | |
0a2966b4 CL |
564 | static inline int cpu_of(struct rq *rq) |
565 | { | |
566 | #ifdef CONFIG_SMP | |
567 | return rq->cpu; | |
568 | #else | |
569 | return 0; | |
570 | #endif | |
571 | } | |
572 | ||
497f0ab3 | 573 | #define rcu_dereference_check_sched_domain(p) \ |
d11c563d PM |
574 | rcu_dereference_check((p), \ |
575 | rcu_read_lock_sched_held() || \ | |
576 | lockdep_is_held(&sched_domains_mutex)) | |
577 | ||
674311d5 NP |
578 | /* |
579 | * The domain tree (rq->sd) is protected by RCU's quiescent state transition. | |
1a20ff27 | 580 | * See detach_destroy_domains: synchronize_sched for details. |
674311d5 NP |
581 | * |
582 | * The domain tree of any CPU may only be accessed from within | |
583 | * preempt-disabled sections. | |
584 | */ | |
48f24c4d | 585 | #define for_each_domain(cpu, __sd) \ |
497f0ab3 | 586 | for (__sd = rcu_dereference_check_sched_domain(cpu_rq(cpu)->sd); __sd; __sd = __sd->parent) |
1da177e4 LT |
587 | |
588 | #define cpu_rq(cpu) (&per_cpu(runqueues, (cpu))) | |
589 | #define this_rq() (&__get_cpu_var(runqueues)) | |
590 | #define task_rq(p) cpu_rq(task_cpu(p)) | |
591 | #define cpu_curr(cpu) (cpu_rq(cpu)->curr) | |
54d35f29 | 592 | #define raw_rq() (&__raw_get_cpu_var(runqueues)) |
1da177e4 | 593 | |
dc61b1d6 PZ |
594 | #ifdef CONFIG_CGROUP_SCHED |
595 | ||
596 | /* | |
597 | * Return the group to which this tasks belongs. | |
598 | * | |
599 | * We use task_subsys_state_check() and extend the RCU verification | |
600 | * with lockdep_is_held(&task_rq(p)->lock) because cpu_cgroup_attach() | |
601 | * holds that lock for each task it moves into the cgroup. Therefore | |
602 | * by holding that lock, we pin the task to the current cgroup. | |
603 | */ | |
604 | static inline struct task_group *task_group(struct task_struct *p) | |
605 | { | |
606 | struct cgroup_subsys_state *css; | |
607 | ||
608 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
609 | lockdep_is_held(&task_rq(p)->lock)); | |
610 | return container_of(css, struct task_group, css); | |
611 | } | |
612 | ||
613 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
614 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
615 | { | |
616 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
617 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
618 | p->se.parent = task_group(p)->se[cpu]; | |
619 | #endif | |
620 | ||
621 | #ifdef CONFIG_RT_GROUP_SCHED | |
622 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
623 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
624 | #endif | |
625 | } | |
626 | ||
627 | #else /* CONFIG_CGROUP_SCHED */ | |
628 | ||
629 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
630 | static inline struct task_group *task_group(struct task_struct *p) | |
631 | { | |
632 | return NULL; | |
633 | } | |
634 | ||
635 | #endif /* CONFIG_CGROUP_SCHED */ | |
636 | ||
305e6835 | 637 | static u64 irq_time_cpu(int cpu); |
aa483808 | 638 | static void sched_irq_time_avg_update(struct rq *rq, u64 irq_time); |
305e6835 | 639 | |
aa9c4c0f | 640 | inline void update_rq_clock(struct rq *rq) |
3e51f33f | 641 | { |
305e6835 VP |
642 | if (!rq->skip_clock_update) { |
643 | int cpu = cpu_of(rq); | |
644 | u64 irq_time; | |
645 | ||
646 | rq->clock = sched_clock_cpu(cpu); | |
647 | irq_time = irq_time_cpu(cpu); | |
648 | if (rq->clock - irq_time > rq->clock_task) | |
649 | rq->clock_task = rq->clock - irq_time; | |
aa483808 VP |
650 | |
651 | sched_irq_time_avg_update(rq, irq_time); | |
305e6835 | 652 | } |
3e51f33f PZ |
653 | } |
654 | ||
bf5c91ba IM |
655 | /* |
656 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
657 | */ | |
658 | #ifdef CONFIG_SCHED_DEBUG | |
659 | # define const_debug __read_mostly | |
660 | #else | |
661 | # define const_debug static const | |
662 | #endif | |
663 | ||
017730c1 IM |
664 | /** |
665 | * runqueue_is_locked | |
e17b38bf | 666 | * @cpu: the processor in question. |
017730c1 IM |
667 | * |
668 | * Returns true if the current cpu runqueue is locked. | |
669 | * This interface allows printk to be called with the runqueue lock | |
670 | * held and know whether or not it is OK to wake up the klogd. | |
671 | */ | |
89f19f04 | 672 | int runqueue_is_locked(int cpu) |
017730c1 | 673 | { |
05fa785c | 674 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
675 | } |
676 | ||
bf5c91ba IM |
677 | /* |
678 | * Debugging: various feature bits | |
679 | */ | |
f00b45c1 PZ |
680 | |
681 | #define SCHED_FEAT(name, enabled) \ | |
682 | __SCHED_FEAT_##name , | |
683 | ||
bf5c91ba | 684 | enum { |
f00b45c1 | 685 | #include "sched_features.h" |
bf5c91ba IM |
686 | }; |
687 | ||
f00b45c1 PZ |
688 | #undef SCHED_FEAT |
689 | ||
690 | #define SCHED_FEAT(name, enabled) \ | |
691 | (1UL << __SCHED_FEAT_##name) * enabled | | |
692 | ||
bf5c91ba | 693 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
694 | #include "sched_features.h" |
695 | 0; | |
696 | ||
697 | #undef SCHED_FEAT | |
698 | ||
699 | #ifdef CONFIG_SCHED_DEBUG | |
700 | #define SCHED_FEAT(name, enabled) \ | |
701 | #name , | |
702 | ||
983ed7a6 | 703 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
704 | #include "sched_features.h" |
705 | NULL | |
706 | }; | |
707 | ||
708 | #undef SCHED_FEAT | |
709 | ||
34f3a814 | 710 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 711 | { |
f00b45c1 PZ |
712 | int i; |
713 | ||
714 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
715 | if (!(sysctl_sched_features & (1UL << i))) |
716 | seq_puts(m, "NO_"); | |
717 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 718 | } |
34f3a814 | 719 | seq_puts(m, "\n"); |
f00b45c1 | 720 | |
34f3a814 | 721 | return 0; |
f00b45c1 PZ |
722 | } |
723 | ||
724 | static ssize_t | |
725 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
726 | size_t cnt, loff_t *ppos) | |
727 | { | |
728 | char buf[64]; | |
7740191c | 729 | char *cmp; |
f00b45c1 PZ |
730 | int neg = 0; |
731 | int i; | |
732 | ||
733 | if (cnt > 63) | |
734 | cnt = 63; | |
735 | ||
736 | if (copy_from_user(&buf, ubuf, cnt)) | |
737 | return -EFAULT; | |
738 | ||
739 | buf[cnt] = 0; | |
7740191c | 740 | cmp = strstrip(buf); |
f00b45c1 | 741 | |
c24b7c52 | 742 | if (strncmp(buf, "NO_", 3) == 0) { |
f00b45c1 PZ |
743 | neg = 1; |
744 | cmp += 3; | |
745 | } | |
746 | ||
747 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 748 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
749 | if (neg) |
750 | sysctl_sched_features &= ~(1UL << i); | |
751 | else | |
752 | sysctl_sched_features |= (1UL << i); | |
753 | break; | |
754 | } | |
755 | } | |
756 | ||
757 | if (!sched_feat_names[i]) | |
758 | return -EINVAL; | |
759 | ||
42994724 | 760 | *ppos += cnt; |
f00b45c1 PZ |
761 | |
762 | return cnt; | |
763 | } | |
764 | ||
34f3a814 LZ |
765 | static int sched_feat_open(struct inode *inode, struct file *filp) |
766 | { | |
767 | return single_open(filp, sched_feat_show, NULL); | |
768 | } | |
769 | ||
828c0950 | 770 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
771 | .open = sched_feat_open, |
772 | .write = sched_feat_write, | |
773 | .read = seq_read, | |
774 | .llseek = seq_lseek, | |
775 | .release = single_release, | |
f00b45c1 PZ |
776 | }; |
777 | ||
778 | static __init int sched_init_debug(void) | |
779 | { | |
f00b45c1 PZ |
780 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
781 | &sched_feat_fops); | |
782 | ||
783 | return 0; | |
784 | } | |
785 | late_initcall(sched_init_debug); | |
786 | ||
787 | #endif | |
788 | ||
789 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 790 | |
b82d9fdd PZ |
791 | /* |
792 | * Number of tasks to iterate in a single balance run. | |
793 | * Limited because this is done with IRQs disabled. | |
794 | */ | |
795 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
796 | ||
e9e9250b PZ |
797 | /* |
798 | * period over which we average the RT time consumption, measured | |
799 | * in ms. | |
800 | * | |
801 | * default: 1s | |
802 | */ | |
803 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
804 | ||
fa85ae24 | 805 | /* |
9f0c1e56 | 806 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
807 | * default: 1s |
808 | */ | |
9f0c1e56 | 809 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 810 | |
6892b75e IM |
811 | static __read_mostly int scheduler_running; |
812 | ||
9f0c1e56 PZ |
813 | /* |
814 | * part of the period that we allow rt tasks to run in us. | |
815 | * default: 0.95s | |
816 | */ | |
817 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 818 | |
d0b27fa7 PZ |
819 | static inline u64 global_rt_period(void) |
820 | { | |
821 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
822 | } | |
823 | ||
824 | static inline u64 global_rt_runtime(void) | |
825 | { | |
e26873bb | 826 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
827 | return RUNTIME_INF; |
828 | ||
829 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
830 | } | |
fa85ae24 | 831 | |
1da177e4 | 832 | #ifndef prepare_arch_switch |
4866cde0 NP |
833 | # define prepare_arch_switch(next) do { } while (0) |
834 | #endif | |
835 | #ifndef finish_arch_switch | |
836 | # define finish_arch_switch(prev) do { } while (0) | |
837 | #endif | |
838 | ||
051a1d1a DA |
839 | static inline int task_current(struct rq *rq, struct task_struct *p) |
840 | { | |
841 | return rq->curr == p; | |
842 | } | |
843 | ||
4866cde0 | 844 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 845 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 846 | { |
051a1d1a | 847 | return task_current(rq, p); |
4866cde0 NP |
848 | } |
849 | ||
70b97a7f | 850 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
851 | { |
852 | } | |
853 | ||
70b97a7f | 854 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 855 | { |
da04c035 IM |
856 | #ifdef CONFIG_DEBUG_SPINLOCK |
857 | /* this is a valid case when another task releases the spinlock */ | |
858 | rq->lock.owner = current; | |
859 | #endif | |
8a25d5de IM |
860 | /* |
861 | * If we are tracking spinlock dependencies then we have to | |
862 | * fix up the runqueue lock - which gets 'carried over' from | |
863 | * prev into current: | |
864 | */ | |
865 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
866 | ||
05fa785c | 867 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
868 | } |
869 | ||
870 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 871 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
872 | { |
873 | #ifdef CONFIG_SMP | |
874 | return p->oncpu; | |
875 | #else | |
051a1d1a | 876 | return task_current(rq, p); |
4866cde0 NP |
877 | #endif |
878 | } | |
879 | ||
70b97a7f | 880 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
881 | { |
882 | #ifdef CONFIG_SMP | |
883 | /* | |
884 | * We can optimise this out completely for !SMP, because the | |
885 | * SMP rebalancing from interrupt is the only thing that cares | |
886 | * here. | |
887 | */ | |
888 | next->oncpu = 1; | |
889 | #endif | |
890 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 891 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 892 | #else |
05fa785c | 893 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
894 | #endif |
895 | } | |
896 | ||
70b97a7f | 897 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
898 | { |
899 | #ifdef CONFIG_SMP | |
900 | /* | |
901 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
902 | * We must ensure this doesn't happen until the switch is completely | |
903 | * finished. | |
904 | */ | |
905 | smp_wmb(); | |
906 | prev->oncpu = 0; | |
907 | #endif | |
908 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
909 | local_irq_enable(); | |
1da177e4 | 910 | #endif |
4866cde0 NP |
911 | } |
912 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 913 | |
0970d299 | 914 | /* |
65cc8e48 PZ |
915 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
916 | * against ttwu(). | |
0970d299 PZ |
917 | */ |
918 | static inline int task_is_waking(struct task_struct *p) | |
919 | { | |
0017d735 | 920 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
921 | } |
922 | ||
b29739f9 IM |
923 | /* |
924 | * __task_rq_lock - lock the runqueue a given task resides on. | |
925 | * Must be called interrupts disabled. | |
926 | */ | |
70b97a7f | 927 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
928 | __acquires(rq->lock) |
929 | { | |
0970d299 PZ |
930 | struct rq *rq; |
931 | ||
3a5c359a | 932 | for (;;) { |
0970d299 | 933 | rq = task_rq(p); |
05fa785c | 934 | raw_spin_lock(&rq->lock); |
65cc8e48 | 935 | if (likely(rq == task_rq(p))) |
3a5c359a | 936 | return rq; |
05fa785c | 937 | raw_spin_unlock(&rq->lock); |
b29739f9 | 938 | } |
b29739f9 IM |
939 | } |
940 | ||
1da177e4 LT |
941 | /* |
942 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 943 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
944 | * explicitly disabling preemption. |
945 | */ | |
70b97a7f | 946 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
947 | __acquires(rq->lock) |
948 | { | |
70b97a7f | 949 | struct rq *rq; |
1da177e4 | 950 | |
3a5c359a AK |
951 | for (;;) { |
952 | local_irq_save(*flags); | |
953 | rq = task_rq(p); | |
05fa785c | 954 | raw_spin_lock(&rq->lock); |
65cc8e48 | 955 | if (likely(rq == task_rq(p))) |
3a5c359a | 956 | return rq; |
05fa785c | 957 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 958 | } |
1da177e4 LT |
959 | } |
960 | ||
a9957449 | 961 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
962 | __releases(rq->lock) |
963 | { | |
05fa785c | 964 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
965 | } |
966 | ||
70b97a7f | 967 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
968 | __releases(rq->lock) |
969 | { | |
05fa785c | 970 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
971 | } |
972 | ||
1da177e4 | 973 | /* |
cc2a73b5 | 974 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 975 | */ |
a9957449 | 976 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
977 | __acquires(rq->lock) |
978 | { | |
70b97a7f | 979 | struct rq *rq; |
1da177e4 LT |
980 | |
981 | local_irq_disable(); | |
982 | rq = this_rq(); | |
05fa785c | 983 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
984 | |
985 | return rq; | |
986 | } | |
987 | ||
8f4d37ec PZ |
988 | #ifdef CONFIG_SCHED_HRTICK |
989 | /* | |
990 | * Use HR-timers to deliver accurate preemption points. | |
991 | * | |
992 | * Its all a bit involved since we cannot program an hrt while holding the | |
993 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
994 | * reschedule event. | |
995 | * | |
996 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
997 | * rq->lock. | |
998 | */ | |
8f4d37ec PZ |
999 | |
1000 | /* | |
1001 | * Use hrtick when: | |
1002 | * - enabled by features | |
1003 | * - hrtimer is actually high res | |
1004 | */ | |
1005 | static inline int hrtick_enabled(struct rq *rq) | |
1006 | { | |
1007 | if (!sched_feat(HRTICK)) | |
1008 | return 0; | |
ba42059f | 1009 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1010 | return 0; |
8f4d37ec PZ |
1011 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1012 | } | |
1013 | ||
8f4d37ec PZ |
1014 | static void hrtick_clear(struct rq *rq) |
1015 | { | |
1016 | if (hrtimer_active(&rq->hrtick_timer)) | |
1017 | hrtimer_cancel(&rq->hrtick_timer); | |
1018 | } | |
1019 | ||
8f4d37ec PZ |
1020 | /* |
1021 | * High-resolution timer tick. | |
1022 | * Runs from hardirq context with interrupts disabled. | |
1023 | */ | |
1024 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1025 | { | |
1026 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1027 | ||
1028 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1029 | ||
05fa785c | 1030 | raw_spin_lock(&rq->lock); |
3e51f33f | 1031 | update_rq_clock(rq); |
8f4d37ec | 1032 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1033 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1034 | |
1035 | return HRTIMER_NORESTART; | |
1036 | } | |
1037 | ||
95e904c7 | 1038 | #ifdef CONFIG_SMP |
31656519 PZ |
1039 | /* |
1040 | * called from hardirq (IPI) context | |
1041 | */ | |
1042 | static void __hrtick_start(void *arg) | |
b328ca18 | 1043 | { |
31656519 | 1044 | struct rq *rq = arg; |
b328ca18 | 1045 | |
05fa785c | 1046 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1047 | hrtimer_restart(&rq->hrtick_timer); |
1048 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1049 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1050 | } |
1051 | ||
31656519 PZ |
1052 | /* |
1053 | * Called to set the hrtick timer state. | |
1054 | * | |
1055 | * called with rq->lock held and irqs disabled | |
1056 | */ | |
1057 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1058 | { |
31656519 PZ |
1059 | struct hrtimer *timer = &rq->hrtick_timer; |
1060 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1061 | |
cc584b21 | 1062 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1063 | |
1064 | if (rq == this_rq()) { | |
1065 | hrtimer_restart(timer); | |
1066 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1067 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1068 | rq->hrtick_csd_pending = 1; |
1069 | } | |
b328ca18 PZ |
1070 | } |
1071 | ||
1072 | static int | |
1073 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1074 | { | |
1075 | int cpu = (int)(long)hcpu; | |
1076 | ||
1077 | switch (action) { | |
1078 | case CPU_UP_CANCELED: | |
1079 | case CPU_UP_CANCELED_FROZEN: | |
1080 | case CPU_DOWN_PREPARE: | |
1081 | case CPU_DOWN_PREPARE_FROZEN: | |
1082 | case CPU_DEAD: | |
1083 | case CPU_DEAD_FROZEN: | |
31656519 | 1084 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1085 | return NOTIFY_OK; |
1086 | } | |
1087 | ||
1088 | return NOTIFY_DONE; | |
1089 | } | |
1090 | ||
fa748203 | 1091 | static __init void init_hrtick(void) |
b328ca18 PZ |
1092 | { |
1093 | hotcpu_notifier(hotplug_hrtick, 0); | |
1094 | } | |
31656519 PZ |
1095 | #else |
1096 | /* | |
1097 | * Called to set the hrtick timer state. | |
1098 | * | |
1099 | * called with rq->lock held and irqs disabled | |
1100 | */ | |
1101 | static void hrtick_start(struct rq *rq, u64 delay) | |
1102 | { | |
7f1e2ca9 | 1103 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1104 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1105 | } |
b328ca18 | 1106 | |
006c75f1 | 1107 | static inline void init_hrtick(void) |
8f4d37ec | 1108 | { |
8f4d37ec | 1109 | } |
31656519 | 1110 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1111 | |
31656519 | 1112 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1113 | { |
31656519 PZ |
1114 | #ifdef CONFIG_SMP |
1115 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1116 | |
31656519 PZ |
1117 | rq->hrtick_csd.flags = 0; |
1118 | rq->hrtick_csd.func = __hrtick_start; | |
1119 | rq->hrtick_csd.info = rq; | |
1120 | #endif | |
8f4d37ec | 1121 | |
31656519 PZ |
1122 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1123 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1124 | } |
006c75f1 | 1125 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1126 | static inline void hrtick_clear(struct rq *rq) |
1127 | { | |
1128 | } | |
1129 | ||
8f4d37ec PZ |
1130 | static inline void init_rq_hrtick(struct rq *rq) |
1131 | { | |
1132 | } | |
1133 | ||
b328ca18 PZ |
1134 | static inline void init_hrtick(void) |
1135 | { | |
1136 | } | |
006c75f1 | 1137 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1138 | |
c24d20db IM |
1139 | /* |
1140 | * resched_task - mark a task 'to be rescheduled now'. | |
1141 | * | |
1142 | * On UP this means the setting of the need_resched flag, on SMP it | |
1143 | * might also involve a cross-CPU call to trigger the scheduler on | |
1144 | * the target CPU. | |
1145 | */ | |
1146 | #ifdef CONFIG_SMP | |
1147 | ||
1148 | #ifndef tsk_is_polling | |
1149 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1150 | #endif | |
1151 | ||
31656519 | 1152 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1153 | { |
1154 | int cpu; | |
1155 | ||
05fa785c | 1156 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1157 | |
5ed0cec0 | 1158 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1159 | return; |
1160 | ||
5ed0cec0 | 1161 | set_tsk_need_resched(p); |
c24d20db IM |
1162 | |
1163 | cpu = task_cpu(p); | |
1164 | if (cpu == smp_processor_id()) | |
1165 | return; | |
1166 | ||
1167 | /* NEED_RESCHED must be visible before we test polling */ | |
1168 | smp_mb(); | |
1169 | if (!tsk_is_polling(p)) | |
1170 | smp_send_reschedule(cpu); | |
1171 | } | |
1172 | ||
1173 | static void resched_cpu(int cpu) | |
1174 | { | |
1175 | struct rq *rq = cpu_rq(cpu); | |
1176 | unsigned long flags; | |
1177 | ||
05fa785c | 1178 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1179 | return; |
1180 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1181 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1182 | } |
06d8308c TG |
1183 | |
1184 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1185 | /* |
1186 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1187 | * from an idle cpu. This is good for power-savings. | |
1188 | * | |
1189 | * We don't do similar optimization for completely idle system, as | |
1190 | * selecting an idle cpu will add more delays to the timers than intended | |
1191 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1192 | */ | |
1193 | int get_nohz_timer_target(void) | |
1194 | { | |
1195 | int cpu = smp_processor_id(); | |
1196 | int i; | |
1197 | struct sched_domain *sd; | |
1198 | ||
1199 | for_each_domain(cpu, sd) { | |
1200 | for_each_cpu(i, sched_domain_span(sd)) | |
1201 | if (!idle_cpu(i)) | |
1202 | return i; | |
1203 | } | |
1204 | return cpu; | |
1205 | } | |
06d8308c TG |
1206 | /* |
1207 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1208 | * idle CPU then this timer might expire before the next timer event | |
1209 | * which is scheduled to wake up that CPU. In case of a completely | |
1210 | * idle system the next event might even be infinite time into the | |
1211 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1212 | * leaves the inner idle loop so the newly added timer is taken into | |
1213 | * account when the CPU goes back to idle and evaluates the timer | |
1214 | * wheel for the next timer event. | |
1215 | */ | |
1216 | void wake_up_idle_cpu(int cpu) | |
1217 | { | |
1218 | struct rq *rq = cpu_rq(cpu); | |
1219 | ||
1220 | if (cpu == smp_processor_id()) | |
1221 | return; | |
1222 | ||
1223 | /* | |
1224 | * This is safe, as this function is called with the timer | |
1225 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1226 | * to idle and has not yet set rq->curr to idle then it will | |
1227 | * be serialized on the timer wheel base lock and take the new | |
1228 | * timer into account automatically. | |
1229 | */ | |
1230 | if (rq->curr != rq->idle) | |
1231 | return; | |
1232 | ||
1233 | /* | |
1234 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1235 | * lockless. The worst case is that the other CPU runs the | |
1236 | * idle task through an additional NOOP schedule() | |
1237 | */ | |
5ed0cec0 | 1238 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1239 | |
1240 | /* NEED_RESCHED must be visible before we test polling */ | |
1241 | smp_mb(); | |
1242 | if (!tsk_is_polling(rq->idle)) | |
1243 | smp_send_reschedule(cpu); | |
1244 | } | |
39c0cbe2 | 1245 | |
6d6bc0ad | 1246 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1247 | |
e9e9250b PZ |
1248 | static u64 sched_avg_period(void) |
1249 | { | |
1250 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1251 | } | |
1252 | ||
1253 | static void sched_avg_update(struct rq *rq) | |
1254 | { | |
1255 | s64 period = sched_avg_period(); | |
1256 | ||
1257 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1258 | /* |
1259 | * Inline assembly required to prevent the compiler | |
1260 | * optimising this loop into a divmod call. | |
1261 | * See __iter_div_u64_rem() for another example of this. | |
1262 | */ | |
1263 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1264 | rq->age_stamp += period; |
1265 | rq->rt_avg /= 2; | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1270 | { | |
1271 | rq->rt_avg += rt_delta; | |
1272 | sched_avg_update(rq); | |
1273 | } | |
1274 | ||
6d6bc0ad | 1275 | #else /* !CONFIG_SMP */ |
31656519 | 1276 | static void resched_task(struct task_struct *p) |
c24d20db | 1277 | { |
05fa785c | 1278 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1279 | set_tsk_need_resched(p); |
c24d20db | 1280 | } |
e9e9250b PZ |
1281 | |
1282 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1283 | { | |
1284 | } | |
da2b71ed SS |
1285 | |
1286 | static void sched_avg_update(struct rq *rq) | |
1287 | { | |
1288 | } | |
6d6bc0ad | 1289 | #endif /* CONFIG_SMP */ |
c24d20db | 1290 | |
45bf76df IM |
1291 | #if BITS_PER_LONG == 32 |
1292 | # define WMULT_CONST (~0UL) | |
1293 | #else | |
1294 | # define WMULT_CONST (1UL << 32) | |
1295 | #endif | |
1296 | ||
1297 | #define WMULT_SHIFT 32 | |
1298 | ||
194081eb IM |
1299 | /* |
1300 | * Shift right and round: | |
1301 | */ | |
cf2ab469 | 1302 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1303 | |
a7be37ac PZ |
1304 | /* |
1305 | * delta *= weight / lw | |
1306 | */ | |
cb1c4fc9 | 1307 | static unsigned long |
45bf76df IM |
1308 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1309 | struct load_weight *lw) | |
1310 | { | |
1311 | u64 tmp; | |
1312 | ||
7a232e03 LJ |
1313 | if (!lw->inv_weight) { |
1314 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1315 | lw->inv_weight = 1; | |
1316 | else | |
1317 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1318 | / (lw->weight+1); | |
1319 | } | |
45bf76df IM |
1320 | |
1321 | tmp = (u64)delta_exec * weight; | |
1322 | /* | |
1323 | * Check whether we'd overflow the 64-bit multiplication: | |
1324 | */ | |
194081eb | 1325 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1326 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1327 | WMULT_SHIFT/2); |
1328 | else | |
cf2ab469 | 1329 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1330 | |
ecf691da | 1331 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1332 | } |
1333 | ||
1091985b | 1334 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1335 | { |
1336 | lw->weight += inc; | |
e89996ae | 1337 | lw->inv_weight = 0; |
45bf76df IM |
1338 | } |
1339 | ||
1091985b | 1340 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1341 | { |
1342 | lw->weight -= dec; | |
e89996ae | 1343 | lw->inv_weight = 0; |
45bf76df IM |
1344 | } |
1345 | ||
2069dd75 PZ |
1346 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1347 | { | |
1348 | lw->weight = w; | |
1349 | lw->inv_weight = 0; | |
1350 | } | |
1351 | ||
2dd73a4f PW |
1352 | /* |
1353 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1354 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1355 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1356 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1357 | * scaled version of the new time slice allocation that they receive on time |
1358 | * slice expiry etc. | |
1359 | */ | |
1360 | ||
cce7ade8 PZ |
1361 | #define WEIGHT_IDLEPRIO 3 |
1362 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1363 | |
1364 | /* | |
1365 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1366 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1367 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1368 | * that remained on nice 0. | |
1369 | * | |
1370 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1371 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1372 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1373 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1374 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1375 | */ |
1376 | static const int prio_to_weight[40] = { | |
254753dc IM |
1377 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1378 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1379 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1380 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1381 | /* 0 */ 1024, 820, 655, 526, 423, | |
1382 | /* 5 */ 335, 272, 215, 172, 137, | |
1383 | /* 10 */ 110, 87, 70, 56, 45, | |
1384 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1385 | }; |
1386 | ||
5714d2de IM |
1387 | /* |
1388 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1389 | * | |
1390 | * In cases where the weight does not change often, we can use the | |
1391 | * precalculated inverse to speed up arithmetics by turning divisions | |
1392 | * into multiplications: | |
1393 | */ | |
dd41f596 | 1394 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1395 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1396 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1397 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1398 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1399 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1400 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1401 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1402 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1403 | }; |
2dd73a4f | 1404 | |
ef12fefa BR |
1405 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1406 | enum cpuacct_stat_index { | |
1407 | CPUACCT_STAT_USER, /* ... user mode */ | |
1408 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1409 | ||
1410 | CPUACCT_STAT_NSTATS, | |
1411 | }; | |
1412 | ||
d842de87 SV |
1413 | #ifdef CONFIG_CGROUP_CPUACCT |
1414 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1415 | static void cpuacct_update_stats(struct task_struct *tsk, |
1416 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1417 | #else |
1418 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1419 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1420 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1421 | #endif |
1422 | ||
18d95a28 PZ |
1423 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1424 | { | |
1425 | update_load_add(&rq->load, load); | |
1426 | } | |
1427 | ||
1428 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1429 | { | |
1430 | update_load_sub(&rq->load, load); | |
1431 | } | |
1432 | ||
7940ca36 | 1433 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1434 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1435 | |
1436 | /* | |
1437 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1438 | * leaving it for the final time. | |
1439 | */ | |
eb755805 | 1440 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1441 | { |
1442 | struct task_group *parent, *child; | |
eb755805 | 1443 | int ret; |
c09595f6 PZ |
1444 | |
1445 | rcu_read_lock(); | |
1446 | parent = &root_task_group; | |
1447 | down: | |
eb755805 PZ |
1448 | ret = (*down)(parent, data); |
1449 | if (ret) | |
1450 | goto out_unlock; | |
c09595f6 PZ |
1451 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1452 | parent = child; | |
1453 | goto down; | |
1454 | ||
1455 | up: | |
1456 | continue; | |
1457 | } | |
eb755805 PZ |
1458 | ret = (*up)(parent, data); |
1459 | if (ret) | |
1460 | goto out_unlock; | |
c09595f6 PZ |
1461 | |
1462 | child = parent; | |
1463 | parent = parent->parent; | |
1464 | if (parent) | |
1465 | goto up; | |
eb755805 | 1466 | out_unlock: |
c09595f6 | 1467 | rcu_read_unlock(); |
eb755805 PZ |
1468 | |
1469 | return ret; | |
c09595f6 PZ |
1470 | } |
1471 | ||
eb755805 PZ |
1472 | static int tg_nop(struct task_group *tg, void *data) |
1473 | { | |
1474 | return 0; | |
c09595f6 | 1475 | } |
eb755805 PZ |
1476 | #endif |
1477 | ||
1478 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1479 | /* Used instead of source_load when we know the type == 0 */ |
1480 | static unsigned long weighted_cpuload(const int cpu) | |
1481 | { | |
1482 | return cpu_rq(cpu)->load.weight; | |
1483 | } | |
1484 | ||
1485 | /* | |
1486 | * Return a low guess at the load of a migration-source cpu weighted | |
1487 | * according to the scheduling class and "nice" value. | |
1488 | * | |
1489 | * We want to under-estimate the load of migration sources, to | |
1490 | * balance conservatively. | |
1491 | */ | |
1492 | static unsigned long source_load(int cpu, int type) | |
1493 | { | |
1494 | struct rq *rq = cpu_rq(cpu); | |
1495 | unsigned long total = weighted_cpuload(cpu); | |
1496 | ||
1497 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1498 | return total; | |
1499 | ||
1500 | return min(rq->cpu_load[type-1], total); | |
1501 | } | |
1502 | ||
1503 | /* | |
1504 | * Return a high guess at the load of a migration-target cpu weighted | |
1505 | * according to the scheduling class and "nice" value. | |
1506 | */ | |
1507 | static unsigned long target_load(int cpu, int type) | |
1508 | { | |
1509 | struct rq *rq = cpu_rq(cpu); | |
1510 | unsigned long total = weighted_cpuload(cpu); | |
1511 | ||
1512 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1513 | return total; | |
1514 | ||
1515 | return max(rq->cpu_load[type-1], total); | |
1516 | } | |
1517 | ||
ae154be1 PZ |
1518 | static unsigned long power_of(int cpu) |
1519 | { | |
e51fd5e2 | 1520 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1521 | } |
1522 | ||
eb755805 PZ |
1523 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1524 | ||
1525 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1526 | { | |
1527 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1528 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1529 | |
4cd42620 SR |
1530 | if (nr_running) |
1531 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1532 | else |
1533 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1534 | |
1535 | return rq->avg_load_per_task; | |
1536 | } | |
1537 | ||
1538 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1539 | |
c09595f6 | 1540 | /* |
c8cba857 PZ |
1541 | * Compute the cpu's hierarchical load factor for each task group. |
1542 | * This needs to be done in a top-down fashion because the load of a child | |
1543 | * group is a fraction of its parents load. | |
c09595f6 | 1544 | */ |
eb755805 | 1545 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1546 | { |
c8cba857 | 1547 | unsigned long load; |
eb755805 | 1548 | long cpu = (long)data; |
c09595f6 | 1549 | |
c8cba857 PZ |
1550 | if (!tg->parent) { |
1551 | load = cpu_rq(cpu)->load.weight; | |
1552 | } else { | |
1553 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1554 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1555 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1556 | } | |
c09595f6 | 1557 | |
c8cba857 | 1558 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1559 | |
eb755805 | 1560 | return 0; |
c09595f6 PZ |
1561 | } |
1562 | ||
eb755805 | 1563 | static void update_h_load(long cpu) |
c09595f6 | 1564 | { |
eb755805 | 1565 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1566 | } |
1567 | ||
18d95a28 PZ |
1568 | #endif |
1569 | ||
8f45e2b5 GH |
1570 | #ifdef CONFIG_PREEMPT |
1571 | ||
b78bb868 PZ |
1572 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1573 | ||
70574a99 | 1574 | /* |
8f45e2b5 GH |
1575 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1576 | * way at the expense of forcing extra atomic operations in all | |
1577 | * invocations. This assures that the double_lock is acquired using the | |
1578 | * same underlying policy as the spinlock_t on this architecture, which | |
1579 | * reduces latency compared to the unfair variant below. However, it | |
1580 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1581 | */ |
8f45e2b5 GH |
1582 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1583 | __releases(this_rq->lock) | |
1584 | __acquires(busiest->lock) | |
1585 | __acquires(this_rq->lock) | |
1586 | { | |
05fa785c | 1587 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1588 | double_rq_lock(this_rq, busiest); |
1589 | ||
1590 | return 1; | |
1591 | } | |
1592 | ||
1593 | #else | |
1594 | /* | |
1595 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1596 | * latency by eliminating extra atomic operations when the locks are | |
1597 | * already in proper order on entry. This favors lower cpu-ids and will | |
1598 | * grant the double lock to lower cpus over higher ids under contention, | |
1599 | * regardless of entry order into the function. | |
1600 | */ | |
1601 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1602 | __releases(this_rq->lock) |
1603 | __acquires(busiest->lock) | |
1604 | __acquires(this_rq->lock) | |
1605 | { | |
1606 | int ret = 0; | |
1607 | ||
05fa785c | 1608 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1609 | if (busiest < this_rq) { |
05fa785c TG |
1610 | raw_spin_unlock(&this_rq->lock); |
1611 | raw_spin_lock(&busiest->lock); | |
1612 | raw_spin_lock_nested(&this_rq->lock, | |
1613 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1614 | ret = 1; |
1615 | } else | |
05fa785c TG |
1616 | raw_spin_lock_nested(&busiest->lock, |
1617 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1618 | } |
1619 | return ret; | |
1620 | } | |
1621 | ||
8f45e2b5 GH |
1622 | #endif /* CONFIG_PREEMPT */ |
1623 | ||
1624 | /* | |
1625 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1626 | */ | |
1627 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1628 | { | |
1629 | if (unlikely(!irqs_disabled())) { | |
1630 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1631 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1632 | BUG_ON(1); |
1633 | } | |
1634 | ||
1635 | return _double_lock_balance(this_rq, busiest); | |
1636 | } | |
1637 | ||
70574a99 AD |
1638 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1639 | __releases(busiest->lock) | |
1640 | { | |
05fa785c | 1641 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1642 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1643 | } | |
1e3c88bd PZ |
1644 | |
1645 | /* | |
1646 | * double_rq_lock - safely lock two runqueues | |
1647 | * | |
1648 | * Note this does not disable interrupts like task_rq_lock, | |
1649 | * you need to do so manually before calling. | |
1650 | */ | |
1651 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1652 | __acquires(rq1->lock) | |
1653 | __acquires(rq2->lock) | |
1654 | { | |
1655 | BUG_ON(!irqs_disabled()); | |
1656 | if (rq1 == rq2) { | |
1657 | raw_spin_lock(&rq1->lock); | |
1658 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1659 | } else { | |
1660 | if (rq1 < rq2) { | |
1661 | raw_spin_lock(&rq1->lock); | |
1662 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1663 | } else { | |
1664 | raw_spin_lock(&rq2->lock); | |
1665 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1666 | } | |
1667 | } | |
1e3c88bd PZ |
1668 | } |
1669 | ||
1670 | /* | |
1671 | * double_rq_unlock - safely unlock two runqueues | |
1672 | * | |
1673 | * Note this does not restore interrupts like task_rq_unlock, | |
1674 | * you need to do so manually after calling. | |
1675 | */ | |
1676 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1677 | __releases(rq1->lock) | |
1678 | __releases(rq2->lock) | |
1679 | { | |
1680 | raw_spin_unlock(&rq1->lock); | |
1681 | if (rq1 != rq2) | |
1682 | raw_spin_unlock(&rq2->lock); | |
1683 | else | |
1684 | __release(rq2->lock); | |
1685 | } | |
1686 | ||
18d95a28 PZ |
1687 | #endif |
1688 | ||
74f5187a | 1689 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1690 | static void update_sysctl(void); |
acb4a848 | 1691 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1692 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1693 | |
cd29fe6f PZ |
1694 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1695 | { | |
1696 | set_task_rq(p, cpu); | |
1697 | #ifdef CONFIG_SMP | |
1698 | /* | |
1699 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1700 | * successfuly executed on another CPU. We must ensure that updates of | |
1701 | * per-task data have been completed by this moment. | |
1702 | */ | |
1703 | smp_wmb(); | |
1704 | task_thread_info(p)->cpu = cpu; | |
1705 | #endif | |
1706 | } | |
dce48a84 | 1707 | |
1e3c88bd | 1708 | static const struct sched_class rt_sched_class; |
dd41f596 | 1709 | |
34f971f6 | 1710 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1711 | #define for_each_class(class) \ |
1712 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1713 | |
1e3c88bd PZ |
1714 | #include "sched_stats.h" |
1715 | ||
c09595f6 | 1716 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1717 | { |
1718 | rq->nr_running++; | |
9c217245 IM |
1719 | } |
1720 | ||
c09595f6 | 1721 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1722 | { |
1723 | rq->nr_running--; | |
9c217245 IM |
1724 | } |
1725 | ||
45bf76df IM |
1726 | static void set_load_weight(struct task_struct *p) |
1727 | { | |
dd41f596 IM |
1728 | /* |
1729 | * SCHED_IDLE tasks get minimal weight: | |
1730 | */ | |
1731 | if (p->policy == SCHED_IDLE) { | |
1732 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1733 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1734 | return; | |
1735 | } | |
71f8bd46 | 1736 | |
dd41f596 IM |
1737 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1738 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1739 | } |
1740 | ||
371fd7e7 | 1741 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1742 | { |
a64692a3 | 1743 | update_rq_clock(rq); |
dd41f596 | 1744 | sched_info_queued(p); |
371fd7e7 | 1745 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1746 | p->se.on_rq = 1; |
71f8bd46 IM |
1747 | } |
1748 | ||
371fd7e7 | 1749 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1750 | { |
a64692a3 | 1751 | update_rq_clock(rq); |
46ac22ba | 1752 | sched_info_dequeued(p); |
371fd7e7 | 1753 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1754 | p->se.on_rq = 0; |
71f8bd46 IM |
1755 | } |
1756 | ||
1e3c88bd PZ |
1757 | /* |
1758 | * activate_task - move a task to the runqueue. | |
1759 | */ | |
371fd7e7 | 1760 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1761 | { |
1762 | if (task_contributes_to_load(p)) | |
1763 | rq->nr_uninterruptible--; | |
1764 | ||
371fd7e7 | 1765 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1766 | inc_nr_running(rq); |
1767 | } | |
1768 | ||
1769 | /* | |
1770 | * deactivate_task - remove a task from the runqueue. | |
1771 | */ | |
371fd7e7 | 1772 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1773 | { |
1774 | if (task_contributes_to_load(p)) | |
1775 | rq->nr_uninterruptible++; | |
1776 | ||
371fd7e7 | 1777 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1778 | dec_nr_running(rq); |
1779 | } | |
1780 | ||
b52bfee4 VP |
1781 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1782 | ||
305e6835 VP |
1783 | /* |
1784 | * There are no locks covering percpu hardirq/softirq time. | |
1785 | * They are only modified in account_system_vtime, on corresponding CPU | |
1786 | * with interrupts disabled. So, writes are safe. | |
1787 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1788 | * This may result in other CPU reading this CPU's irq time and can | |
1789 | * race with irq/account_system_vtime on this CPU. We would either get old | |
1790 | * or new value (or semi updated value on 32 bit) with a side effect of | |
1791 | * accounting a slice of irq time to wrong task when irq is in progress | |
1792 | * while we read rq->clock. That is a worthy compromise in place of having | |
1793 | * locks on each irq in account_system_time. | |
1794 | */ | |
b52bfee4 VP |
1795 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1796 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1797 | ||
1798 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1799 | static int sched_clock_irqtime; | |
1800 | ||
1801 | void enable_sched_clock_irqtime(void) | |
1802 | { | |
1803 | sched_clock_irqtime = 1; | |
1804 | } | |
1805 | ||
1806 | void disable_sched_clock_irqtime(void) | |
1807 | { | |
1808 | sched_clock_irqtime = 0; | |
1809 | } | |
1810 | ||
305e6835 VP |
1811 | static u64 irq_time_cpu(int cpu) |
1812 | { | |
1813 | if (!sched_clock_irqtime) | |
1814 | return 0; | |
1815 | ||
1816 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); | |
1817 | } | |
1818 | ||
b52bfee4 VP |
1819 | void account_system_vtime(struct task_struct *curr) |
1820 | { | |
1821 | unsigned long flags; | |
1822 | int cpu; | |
1823 | u64 now, delta; | |
1824 | ||
1825 | if (!sched_clock_irqtime) | |
1826 | return; | |
1827 | ||
1828 | local_irq_save(flags); | |
1829 | ||
b52bfee4 | 1830 | cpu = smp_processor_id(); |
d267f87f | 1831 | now = sched_clock_cpu(cpu); |
b52bfee4 VP |
1832 | delta = now - per_cpu(irq_start_time, cpu); |
1833 | per_cpu(irq_start_time, cpu) = now; | |
1834 | /* | |
1835 | * We do not account for softirq time from ksoftirqd here. | |
1836 | * We want to continue accounting softirq time to ksoftirqd thread | |
1837 | * in that case, so as not to confuse scheduler with a special task | |
1838 | * that do not consume any time, but still wants to run. | |
1839 | */ | |
1840 | if (hardirq_count()) | |
1841 | per_cpu(cpu_hardirq_time, cpu) += delta; | |
1842 | else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD)) | |
1843 | per_cpu(cpu_softirq_time, cpu) += delta; | |
1844 | ||
1845 | local_irq_restore(flags); | |
1846 | } | |
b7dadc38 | 1847 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1848 | |
aa483808 VP |
1849 | static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) |
1850 | { | |
1851 | if (sched_clock_irqtime && sched_feat(NONIRQ_POWER)) { | |
1852 | u64 delta_irq = curr_irq_time - rq->prev_irq_time; | |
1853 | rq->prev_irq_time = curr_irq_time; | |
1854 | sched_rt_avg_update(rq, delta_irq); | |
1855 | } | |
1856 | } | |
1857 | ||
305e6835 VP |
1858 | #else |
1859 | ||
1860 | static u64 irq_time_cpu(int cpu) | |
1861 | { | |
1862 | return 0; | |
1863 | } | |
1864 | ||
aa483808 VP |
1865 | static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) { } |
1866 | ||
b52bfee4 VP |
1867 | #endif |
1868 | ||
1e3c88bd PZ |
1869 | #include "sched_idletask.c" |
1870 | #include "sched_fair.c" | |
1871 | #include "sched_rt.c" | |
34f971f6 | 1872 | #include "sched_stoptask.c" |
1e3c88bd PZ |
1873 | #ifdef CONFIG_SCHED_DEBUG |
1874 | # include "sched_debug.c" | |
1875 | #endif | |
1876 | ||
34f971f6 PZ |
1877 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
1878 | { | |
1879 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
1880 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
1881 | ||
1882 | if (stop) { | |
1883 | /* | |
1884 | * Make it appear like a SCHED_FIFO task, its something | |
1885 | * userspace knows about and won't get confused about. | |
1886 | * | |
1887 | * Also, it will make PI more or less work without too | |
1888 | * much confusion -- but then, stop work should not | |
1889 | * rely on PI working anyway. | |
1890 | */ | |
1891 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
1892 | ||
1893 | stop->sched_class = &stop_sched_class; | |
1894 | } | |
1895 | ||
1896 | cpu_rq(cpu)->stop = stop; | |
1897 | ||
1898 | if (old_stop) { | |
1899 | /* | |
1900 | * Reset it back to a normal scheduling class so that | |
1901 | * it can die in pieces. | |
1902 | */ | |
1903 | old_stop->sched_class = &rt_sched_class; | |
1904 | } | |
1905 | } | |
1906 | ||
14531189 | 1907 | /* |
dd41f596 | 1908 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1909 | */ |
14531189 IM |
1910 | static inline int __normal_prio(struct task_struct *p) |
1911 | { | |
dd41f596 | 1912 | return p->static_prio; |
14531189 IM |
1913 | } |
1914 | ||
b29739f9 IM |
1915 | /* |
1916 | * Calculate the expected normal priority: i.e. priority | |
1917 | * without taking RT-inheritance into account. Might be | |
1918 | * boosted by interactivity modifiers. Changes upon fork, | |
1919 | * setprio syscalls, and whenever the interactivity | |
1920 | * estimator recalculates. | |
1921 | */ | |
36c8b586 | 1922 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1923 | { |
1924 | int prio; | |
1925 | ||
e05606d3 | 1926 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1927 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1928 | else | |
1929 | prio = __normal_prio(p); | |
1930 | return prio; | |
1931 | } | |
1932 | ||
1933 | /* | |
1934 | * Calculate the current priority, i.e. the priority | |
1935 | * taken into account by the scheduler. This value might | |
1936 | * be boosted by RT tasks, or might be boosted by | |
1937 | * interactivity modifiers. Will be RT if the task got | |
1938 | * RT-boosted. If not then it returns p->normal_prio. | |
1939 | */ | |
36c8b586 | 1940 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
1941 | { |
1942 | p->normal_prio = normal_prio(p); | |
1943 | /* | |
1944 | * If we are RT tasks or we were boosted to RT priority, | |
1945 | * keep the priority unchanged. Otherwise, update priority | |
1946 | * to the normal priority: | |
1947 | */ | |
1948 | if (!rt_prio(p->prio)) | |
1949 | return p->normal_prio; | |
1950 | return p->prio; | |
1951 | } | |
1952 | ||
1da177e4 LT |
1953 | /** |
1954 | * task_curr - is this task currently executing on a CPU? | |
1955 | * @p: the task in question. | |
1956 | */ | |
36c8b586 | 1957 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
1958 | { |
1959 | return cpu_curr(task_cpu(p)) == p; | |
1960 | } | |
1961 | ||
cb469845 SR |
1962 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
1963 | const struct sched_class *prev_class, | |
1964 | int oldprio, int running) | |
1965 | { | |
1966 | if (prev_class != p->sched_class) { | |
1967 | if (prev_class->switched_from) | |
1968 | prev_class->switched_from(rq, p, running); | |
1969 | p->sched_class->switched_to(rq, p, running); | |
1970 | } else | |
1971 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
1972 | } | |
1973 | ||
1e5a7405 PZ |
1974 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
1975 | { | |
1976 | const struct sched_class *class; | |
1977 | ||
1978 | if (p->sched_class == rq->curr->sched_class) { | |
1979 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
1980 | } else { | |
1981 | for_each_class(class) { | |
1982 | if (class == rq->curr->sched_class) | |
1983 | break; | |
1984 | if (class == p->sched_class) { | |
1985 | resched_task(rq->curr); | |
1986 | break; | |
1987 | } | |
1988 | } | |
1989 | } | |
1990 | ||
1991 | /* | |
1992 | * A queue event has occurred, and we're going to schedule. In | |
1993 | * this case, we can save a useless back to back clock update. | |
1994 | */ | |
1995 | if (test_tsk_need_resched(rq->curr)) | |
1996 | rq->skip_clock_update = 1; | |
1997 | } | |
1998 | ||
1da177e4 | 1999 | #ifdef CONFIG_SMP |
cc367732 IM |
2000 | /* |
2001 | * Is this task likely cache-hot: | |
2002 | */ | |
e7693a36 | 2003 | static int |
cc367732 IM |
2004 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2005 | { | |
2006 | s64 delta; | |
2007 | ||
e6c8fba7 PZ |
2008 | if (p->sched_class != &fair_sched_class) |
2009 | return 0; | |
2010 | ||
ef8002f6 NR |
2011 | if (unlikely(p->policy == SCHED_IDLE)) |
2012 | return 0; | |
2013 | ||
f540a608 IM |
2014 | /* |
2015 | * Buddy candidates are cache hot: | |
2016 | */ | |
f685ceac | 2017 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2018 | (&p->se == cfs_rq_of(&p->se)->next || |
2019 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2020 | return 1; |
2021 | ||
6bc1665b IM |
2022 | if (sysctl_sched_migration_cost == -1) |
2023 | return 1; | |
2024 | if (sysctl_sched_migration_cost == 0) | |
2025 | return 0; | |
2026 | ||
cc367732 IM |
2027 | delta = now - p->se.exec_start; |
2028 | ||
2029 | return delta < (s64)sysctl_sched_migration_cost; | |
2030 | } | |
2031 | ||
dd41f596 | 2032 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2033 | { |
e2912009 PZ |
2034 | #ifdef CONFIG_SCHED_DEBUG |
2035 | /* | |
2036 | * We should never call set_task_cpu() on a blocked task, | |
2037 | * ttwu() will sort out the placement. | |
2038 | */ | |
077614ee PZ |
2039 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2040 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2041 | #endif |
2042 | ||
de1d7286 | 2043 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2044 | |
0c69774e PZ |
2045 | if (task_cpu(p) != new_cpu) { |
2046 | p->se.nr_migrations++; | |
2047 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2048 | } | |
dd41f596 IM |
2049 | |
2050 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2051 | } |
2052 | ||
969c7921 | 2053 | struct migration_arg { |
36c8b586 | 2054 | struct task_struct *task; |
1da177e4 | 2055 | int dest_cpu; |
70b97a7f | 2056 | }; |
1da177e4 | 2057 | |
969c7921 TH |
2058 | static int migration_cpu_stop(void *data); |
2059 | ||
1da177e4 LT |
2060 | /* |
2061 | * The task's runqueue lock must be held. | |
2062 | * Returns true if you have to wait for migration thread. | |
2063 | */ | |
969c7921 | 2064 | static bool migrate_task(struct task_struct *p, int dest_cpu) |
1da177e4 | 2065 | { |
70b97a7f | 2066 | struct rq *rq = task_rq(p); |
1da177e4 LT |
2067 | |
2068 | /* | |
2069 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2070 | * the next wake-up will properly place the task. |
1da177e4 | 2071 | */ |
969c7921 | 2072 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2073 | } |
2074 | ||
2075 | /* | |
2076 | * wait_task_inactive - wait for a thread to unschedule. | |
2077 | * | |
85ba2d86 RM |
2078 | * If @match_state is nonzero, it's the @p->state value just checked and |
2079 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2080 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2081 | * we return a positive number (its total switch count). If a second call | |
2082 | * a short while later returns the same number, the caller can be sure that | |
2083 | * @p has remained unscheduled the whole time. | |
2084 | * | |
1da177e4 LT |
2085 | * The caller must ensure that the task *will* unschedule sometime soon, |
2086 | * else this function might spin for a *long* time. This function can't | |
2087 | * be called with interrupts off, or it may introduce deadlock with | |
2088 | * smp_call_function() if an IPI is sent by the same process we are | |
2089 | * waiting to become inactive. | |
2090 | */ | |
85ba2d86 | 2091 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2092 | { |
2093 | unsigned long flags; | |
dd41f596 | 2094 | int running, on_rq; |
85ba2d86 | 2095 | unsigned long ncsw; |
70b97a7f | 2096 | struct rq *rq; |
1da177e4 | 2097 | |
3a5c359a AK |
2098 | for (;;) { |
2099 | /* | |
2100 | * We do the initial early heuristics without holding | |
2101 | * any task-queue locks at all. We'll only try to get | |
2102 | * the runqueue lock when things look like they will | |
2103 | * work out! | |
2104 | */ | |
2105 | rq = task_rq(p); | |
fa490cfd | 2106 | |
3a5c359a AK |
2107 | /* |
2108 | * If the task is actively running on another CPU | |
2109 | * still, just relax and busy-wait without holding | |
2110 | * any locks. | |
2111 | * | |
2112 | * NOTE! Since we don't hold any locks, it's not | |
2113 | * even sure that "rq" stays as the right runqueue! | |
2114 | * But we don't care, since "task_running()" will | |
2115 | * return false if the runqueue has changed and p | |
2116 | * is actually now running somewhere else! | |
2117 | */ | |
85ba2d86 RM |
2118 | while (task_running(rq, p)) { |
2119 | if (match_state && unlikely(p->state != match_state)) | |
2120 | return 0; | |
3a5c359a | 2121 | cpu_relax(); |
85ba2d86 | 2122 | } |
fa490cfd | 2123 | |
3a5c359a AK |
2124 | /* |
2125 | * Ok, time to look more closely! We need the rq | |
2126 | * lock now, to be *sure*. If we're wrong, we'll | |
2127 | * just go back and repeat. | |
2128 | */ | |
2129 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2130 | trace_sched_wait_task(p); |
3a5c359a AK |
2131 | running = task_running(rq, p); |
2132 | on_rq = p->se.on_rq; | |
85ba2d86 | 2133 | ncsw = 0; |
f31e11d8 | 2134 | if (!match_state || p->state == match_state) |
93dcf55f | 2135 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2136 | task_rq_unlock(rq, &flags); |
fa490cfd | 2137 | |
85ba2d86 RM |
2138 | /* |
2139 | * If it changed from the expected state, bail out now. | |
2140 | */ | |
2141 | if (unlikely(!ncsw)) | |
2142 | break; | |
2143 | ||
3a5c359a AK |
2144 | /* |
2145 | * Was it really running after all now that we | |
2146 | * checked with the proper locks actually held? | |
2147 | * | |
2148 | * Oops. Go back and try again.. | |
2149 | */ | |
2150 | if (unlikely(running)) { | |
2151 | cpu_relax(); | |
2152 | continue; | |
2153 | } | |
fa490cfd | 2154 | |
3a5c359a AK |
2155 | /* |
2156 | * It's not enough that it's not actively running, | |
2157 | * it must be off the runqueue _entirely_, and not | |
2158 | * preempted! | |
2159 | * | |
80dd99b3 | 2160 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2161 | * running right now), it's preempted, and we should |
2162 | * yield - it could be a while. | |
2163 | */ | |
2164 | if (unlikely(on_rq)) { | |
2165 | schedule_timeout_uninterruptible(1); | |
2166 | continue; | |
2167 | } | |
fa490cfd | 2168 | |
3a5c359a AK |
2169 | /* |
2170 | * Ahh, all good. It wasn't running, and it wasn't | |
2171 | * runnable, which means that it will never become | |
2172 | * running in the future either. We're all done! | |
2173 | */ | |
2174 | break; | |
2175 | } | |
85ba2d86 RM |
2176 | |
2177 | return ncsw; | |
1da177e4 LT |
2178 | } |
2179 | ||
2180 | /*** | |
2181 | * kick_process - kick a running thread to enter/exit the kernel | |
2182 | * @p: the to-be-kicked thread | |
2183 | * | |
2184 | * Cause a process which is running on another CPU to enter | |
2185 | * kernel-mode, without any delay. (to get signals handled.) | |
2186 | * | |
2187 | * NOTE: this function doesnt have to take the runqueue lock, | |
2188 | * because all it wants to ensure is that the remote task enters | |
2189 | * the kernel. If the IPI races and the task has been migrated | |
2190 | * to another CPU then no harm is done and the purpose has been | |
2191 | * achieved as well. | |
2192 | */ | |
36c8b586 | 2193 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2194 | { |
2195 | int cpu; | |
2196 | ||
2197 | preempt_disable(); | |
2198 | cpu = task_cpu(p); | |
2199 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2200 | smp_send_reschedule(cpu); | |
2201 | preempt_enable(); | |
2202 | } | |
b43e3521 | 2203 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2204 | #endif /* CONFIG_SMP */ |
1da177e4 | 2205 | |
0793a61d TG |
2206 | /** |
2207 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2208 | * @p: the task to evaluate | |
2209 | * @func: the function to be called | |
2210 | * @info: the function call argument | |
2211 | * | |
2212 | * Calls the function @func when the task is currently running. This might | |
2213 | * be on the current CPU, which just calls the function directly | |
2214 | */ | |
2215 | void task_oncpu_function_call(struct task_struct *p, | |
2216 | void (*func) (void *info), void *info) | |
2217 | { | |
2218 | int cpu; | |
2219 | ||
2220 | preempt_disable(); | |
2221 | cpu = task_cpu(p); | |
2222 | if (task_curr(p)) | |
2223 | smp_call_function_single(cpu, func, info, 1); | |
2224 | preempt_enable(); | |
2225 | } | |
2226 | ||
970b13ba | 2227 | #ifdef CONFIG_SMP |
30da688e ON |
2228 | /* |
2229 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2230 | */ | |
5da9a0fb PZ |
2231 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2232 | { | |
2233 | int dest_cpu; | |
2234 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2235 | ||
2236 | /* Look for allowed, online CPU in same node. */ | |
2237 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2238 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2239 | return dest_cpu; | |
2240 | ||
2241 | /* Any allowed, online CPU? */ | |
2242 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2243 | if (dest_cpu < nr_cpu_ids) | |
2244 | return dest_cpu; | |
2245 | ||
2246 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2247 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2248 | /* | |
2249 | * Don't tell them about moving exiting tasks or | |
2250 | * kernel threads (both mm NULL), since they never | |
2251 | * leave kernel. | |
2252 | */ | |
2253 | if (p->mm && printk_ratelimit()) { | |
2254 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2255 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2256 | } |
2257 | ||
2258 | return dest_cpu; | |
2259 | } | |
2260 | ||
e2912009 | 2261 | /* |
30da688e | 2262 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2263 | */ |
970b13ba | 2264 | static inline |
0017d735 | 2265 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2266 | { |
0017d735 | 2267 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2268 | |
2269 | /* | |
2270 | * In order not to call set_task_cpu() on a blocking task we need | |
2271 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2272 | * cpu. | |
2273 | * | |
2274 | * Since this is common to all placement strategies, this lives here. | |
2275 | * | |
2276 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2277 | * not worry about this generic constraint ] | |
2278 | */ | |
2279 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2280 | !cpu_online(cpu))) |
5da9a0fb | 2281 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2282 | |
2283 | return cpu; | |
970b13ba | 2284 | } |
09a40af5 MG |
2285 | |
2286 | static void update_avg(u64 *avg, u64 sample) | |
2287 | { | |
2288 | s64 diff = sample - *avg; | |
2289 | *avg += diff >> 3; | |
2290 | } | |
970b13ba PZ |
2291 | #endif |
2292 | ||
9ed3811a TH |
2293 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2294 | bool is_sync, bool is_migrate, bool is_local, | |
2295 | unsigned long en_flags) | |
2296 | { | |
2297 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2298 | if (is_sync) | |
2299 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2300 | if (is_migrate) | |
2301 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2302 | if (is_local) | |
2303 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2304 | else | |
2305 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2306 | ||
2307 | activate_task(rq, p, en_flags); | |
2308 | } | |
2309 | ||
2310 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2311 | int wake_flags, bool success) | |
2312 | { | |
2313 | trace_sched_wakeup(p, success); | |
2314 | check_preempt_curr(rq, p, wake_flags); | |
2315 | ||
2316 | p->state = TASK_RUNNING; | |
2317 | #ifdef CONFIG_SMP | |
2318 | if (p->sched_class->task_woken) | |
2319 | p->sched_class->task_woken(rq, p); | |
2320 | ||
2321 | if (unlikely(rq->idle_stamp)) { | |
2322 | u64 delta = rq->clock - rq->idle_stamp; | |
2323 | u64 max = 2*sysctl_sched_migration_cost; | |
2324 | ||
2325 | if (delta > max) | |
2326 | rq->avg_idle = max; | |
2327 | else | |
2328 | update_avg(&rq->avg_idle, delta); | |
2329 | rq->idle_stamp = 0; | |
2330 | } | |
2331 | #endif | |
21aa9af0 TH |
2332 | /* if a worker is waking up, notify workqueue */ |
2333 | if ((p->flags & PF_WQ_WORKER) && success) | |
2334 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2335 | } |
2336 | ||
2337 | /** | |
1da177e4 | 2338 | * try_to_wake_up - wake up a thread |
9ed3811a | 2339 | * @p: the thread to be awakened |
1da177e4 | 2340 | * @state: the mask of task states that can be woken |
9ed3811a | 2341 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2342 | * |
2343 | * Put it on the run-queue if it's not already there. The "current" | |
2344 | * thread is always on the run-queue (except when the actual | |
2345 | * re-schedule is in progress), and as such you're allowed to do | |
2346 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2347 | * runnable without the overhead of this. | |
2348 | * | |
9ed3811a TH |
2349 | * Returns %true if @p was woken up, %false if it was already running |
2350 | * or @state didn't match @p's state. | |
1da177e4 | 2351 | */ |
7d478721 PZ |
2352 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2353 | int wake_flags) | |
1da177e4 | 2354 | { |
cc367732 | 2355 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2356 | unsigned long flags; |
371fd7e7 | 2357 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2358 | struct rq *rq; |
1da177e4 | 2359 | |
e9c84311 | 2360 | this_cpu = get_cpu(); |
2398f2c6 | 2361 | |
04e2f174 | 2362 | smp_wmb(); |
ab3b3aa5 | 2363 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2364 | if (!(p->state & state)) |
1da177e4 LT |
2365 | goto out; |
2366 | ||
dd41f596 | 2367 | if (p->se.on_rq) |
1da177e4 LT |
2368 | goto out_running; |
2369 | ||
2370 | cpu = task_cpu(p); | |
cc367732 | 2371 | orig_cpu = cpu; |
1da177e4 LT |
2372 | |
2373 | #ifdef CONFIG_SMP | |
2374 | if (unlikely(task_running(rq, p))) | |
2375 | goto out_activate; | |
2376 | ||
e9c84311 PZ |
2377 | /* |
2378 | * In order to handle concurrent wakeups and release the rq->lock | |
2379 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2380 | * |
2381 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2382 | */ |
cc87f76a PZ |
2383 | if (task_contributes_to_load(p)) { |
2384 | if (likely(cpu_online(orig_cpu))) | |
2385 | rq->nr_uninterruptible--; | |
2386 | else | |
2387 | this_rq()->nr_uninterruptible--; | |
2388 | } | |
e9c84311 | 2389 | p->state = TASK_WAKING; |
efbbd05a | 2390 | |
371fd7e7 | 2391 | if (p->sched_class->task_waking) { |
efbbd05a | 2392 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2393 | en_flags |= ENQUEUE_WAKING; |
2394 | } | |
efbbd05a | 2395 | |
0017d735 PZ |
2396 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2397 | if (cpu != orig_cpu) | |
5d2f5a61 | 2398 | set_task_cpu(p, cpu); |
0017d735 | 2399 | __task_rq_unlock(rq); |
ab19cb23 | 2400 | |
0970d299 PZ |
2401 | rq = cpu_rq(cpu); |
2402 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2403 | |
0970d299 PZ |
2404 | /* |
2405 | * We migrated the task without holding either rq->lock, however | |
2406 | * since the task is not on the task list itself, nobody else | |
2407 | * will try and migrate the task, hence the rq should match the | |
2408 | * cpu we just moved it to. | |
2409 | */ | |
2410 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2411 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2412 | |
e7693a36 GH |
2413 | #ifdef CONFIG_SCHEDSTATS |
2414 | schedstat_inc(rq, ttwu_count); | |
2415 | if (cpu == this_cpu) | |
2416 | schedstat_inc(rq, ttwu_local); | |
2417 | else { | |
2418 | struct sched_domain *sd; | |
2419 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2420 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2421 | schedstat_inc(sd, ttwu_wake_remote); |
2422 | break; | |
2423 | } | |
2424 | } | |
2425 | } | |
6d6bc0ad | 2426 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2427 | |
1da177e4 LT |
2428 | out_activate: |
2429 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2430 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2431 | cpu == this_cpu, en_flags); | |
1da177e4 | 2432 | success = 1; |
1da177e4 | 2433 | out_running: |
9ed3811a | 2434 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2435 | out: |
2436 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2437 | put_cpu(); |
1da177e4 LT |
2438 | |
2439 | return success; | |
2440 | } | |
2441 | ||
21aa9af0 TH |
2442 | /** |
2443 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2444 | * @p: the thread to be awakened | |
2445 | * | |
2446 | * Put @p on the run-queue if it's not alredy there. The caller must | |
2447 | * ensure that this_rq() is locked, @p is bound to this_rq() and not | |
2448 | * the current task. this_rq() stays locked over invocation. | |
2449 | */ | |
2450 | static void try_to_wake_up_local(struct task_struct *p) | |
2451 | { | |
2452 | struct rq *rq = task_rq(p); | |
2453 | bool success = false; | |
2454 | ||
2455 | BUG_ON(rq != this_rq()); | |
2456 | BUG_ON(p == current); | |
2457 | lockdep_assert_held(&rq->lock); | |
2458 | ||
2459 | if (!(p->state & TASK_NORMAL)) | |
2460 | return; | |
2461 | ||
2462 | if (!p->se.on_rq) { | |
2463 | if (likely(!task_running(rq, p))) { | |
2464 | schedstat_inc(rq, ttwu_count); | |
2465 | schedstat_inc(rq, ttwu_local); | |
2466 | } | |
2467 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2468 | success = true; | |
2469 | } | |
2470 | ttwu_post_activation(p, rq, 0, success); | |
2471 | } | |
2472 | ||
50fa610a DH |
2473 | /** |
2474 | * wake_up_process - Wake up a specific process | |
2475 | * @p: The process to be woken up. | |
2476 | * | |
2477 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2478 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2479 | * running. | |
2480 | * | |
2481 | * It may be assumed that this function implies a write memory barrier before | |
2482 | * changing the task state if and only if any tasks are woken up. | |
2483 | */ | |
7ad5b3a5 | 2484 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2485 | { |
d9514f6c | 2486 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2487 | } |
1da177e4 LT |
2488 | EXPORT_SYMBOL(wake_up_process); |
2489 | ||
7ad5b3a5 | 2490 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2491 | { |
2492 | return try_to_wake_up(p, state, 0); | |
2493 | } | |
2494 | ||
1da177e4 LT |
2495 | /* |
2496 | * Perform scheduler related setup for a newly forked process p. | |
2497 | * p is forked by current. | |
dd41f596 IM |
2498 | * |
2499 | * __sched_fork() is basic setup used by init_idle() too: | |
2500 | */ | |
2501 | static void __sched_fork(struct task_struct *p) | |
2502 | { | |
dd41f596 IM |
2503 | p->se.exec_start = 0; |
2504 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2505 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2506 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2507 | |
2508 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2509 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2510 | #endif |
476d139c | 2511 | |
fa717060 | 2512 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2513 | p->se.on_rq = 0; |
4a55bd5e | 2514 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2515 | |
e107be36 AK |
2516 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2517 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2518 | #endif | |
dd41f596 IM |
2519 | } |
2520 | ||
2521 | /* | |
2522 | * fork()/clone()-time setup: | |
2523 | */ | |
2524 | void sched_fork(struct task_struct *p, int clone_flags) | |
2525 | { | |
2526 | int cpu = get_cpu(); | |
2527 | ||
2528 | __sched_fork(p); | |
06b83b5f | 2529 | /* |
0017d735 | 2530 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2531 | * nobody will actually run it, and a signal or other external |
2532 | * event cannot wake it up and insert it on the runqueue either. | |
2533 | */ | |
0017d735 | 2534 | p->state = TASK_RUNNING; |
dd41f596 | 2535 | |
b9dc29e7 MG |
2536 | /* |
2537 | * Revert to default priority/policy on fork if requested. | |
2538 | */ | |
2539 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2540 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2541 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2542 | p->normal_prio = p->static_prio; |
2543 | } | |
b9dc29e7 | 2544 | |
6c697bdf MG |
2545 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2546 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2547 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2548 | set_load_weight(p); |
2549 | } | |
2550 | ||
b9dc29e7 MG |
2551 | /* |
2552 | * We don't need the reset flag anymore after the fork. It has | |
2553 | * fulfilled its duty: | |
2554 | */ | |
2555 | p->sched_reset_on_fork = 0; | |
2556 | } | |
ca94c442 | 2557 | |
f83f9ac2 PW |
2558 | /* |
2559 | * Make sure we do not leak PI boosting priority to the child. | |
2560 | */ | |
2561 | p->prio = current->normal_prio; | |
2562 | ||
2ddbf952 HS |
2563 | if (!rt_prio(p->prio)) |
2564 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2565 | |
cd29fe6f PZ |
2566 | if (p->sched_class->task_fork) |
2567 | p->sched_class->task_fork(p); | |
2568 | ||
86951599 PZ |
2569 | /* |
2570 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2571 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2572 | * is ran before sched_fork(). | |
2573 | * | |
2574 | * Silence PROVE_RCU. | |
2575 | */ | |
2576 | rcu_read_lock(); | |
5f3edc1b | 2577 | set_task_cpu(p, cpu); |
86951599 | 2578 | rcu_read_unlock(); |
5f3edc1b | 2579 | |
52f17b6c | 2580 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2581 | if (likely(sched_info_on())) |
52f17b6c | 2582 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2583 | #endif |
d6077cb8 | 2584 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2585 | p->oncpu = 0; |
2586 | #endif | |
1da177e4 | 2587 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2588 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2589 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2590 | #endif |
917b627d GH |
2591 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
2592 | ||
476d139c | 2593 | put_cpu(); |
1da177e4 LT |
2594 | } |
2595 | ||
2596 | /* | |
2597 | * wake_up_new_task - wake up a newly created task for the first time. | |
2598 | * | |
2599 | * This function will do some initial scheduler statistics housekeeping | |
2600 | * that must be done for every newly created context, then puts the task | |
2601 | * on the runqueue and wakes it. | |
2602 | */ | |
7ad5b3a5 | 2603 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2604 | { |
2605 | unsigned long flags; | |
dd41f596 | 2606 | struct rq *rq; |
c890692b | 2607 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2608 | |
2609 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2610 | rq = task_rq_lock(p, &flags); |
2611 | p->state = TASK_WAKING; | |
2612 | ||
fabf318e PZ |
2613 | /* |
2614 | * Fork balancing, do it here and not earlier because: | |
2615 | * - cpus_allowed can change in the fork path | |
2616 | * - any previously selected cpu might disappear through hotplug | |
2617 | * | |
0017d735 PZ |
2618 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2619 | * without people poking at ->cpus_allowed. | |
fabf318e | 2620 | */ |
0017d735 | 2621 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2622 | set_task_cpu(p, cpu); |
1da177e4 | 2623 | |
06b83b5f | 2624 | p->state = TASK_RUNNING; |
0017d735 PZ |
2625 | task_rq_unlock(rq, &flags); |
2626 | #endif | |
2627 | ||
2628 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2629 | activate_task(rq, p, 0); |
27a9da65 | 2630 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2631 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2632 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2633 | if (p->sched_class->task_woken) |
2634 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2635 | #endif |
dd41f596 | 2636 | task_rq_unlock(rq, &flags); |
fabf318e | 2637 | put_cpu(); |
1da177e4 LT |
2638 | } |
2639 | ||
e107be36 AK |
2640 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2641 | ||
2642 | /** | |
80dd99b3 | 2643 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2644 | * @notifier: notifier struct to register |
e107be36 AK |
2645 | */ |
2646 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2647 | { | |
2648 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2649 | } | |
2650 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2651 | ||
2652 | /** | |
2653 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2654 | * @notifier: notifier struct to unregister |
e107be36 AK |
2655 | * |
2656 | * This is safe to call from within a preemption notifier. | |
2657 | */ | |
2658 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2659 | { | |
2660 | hlist_del(¬ifier->link); | |
2661 | } | |
2662 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2663 | ||
2664 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2665 | { | |
2666 | struct preempt_notifier *notifier; | |
2667 | struct hlist_node *node; | |
2668 | ||
2669 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2670 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2671 | } | |
2672 | ||
2673 | static void | |
2674 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2675 | struct task_struct *next) | |
2676 | { | |
2677 | struct preempt_notifier *notifier; | |
2678 | struct hlist_node *node; | |
2679 | ||
2680 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2681 | notifier->ops->sched_out(notifier, next); | |
2682 | } | |
2683 | ||
6d6bc0ad | 2684 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2685 | |
2686 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2687 | { | |
2688 | } | |
2689 | ||
2690 | static void | |
2691 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2692 | struct task_struct *next) | |
2693 | { | |
2694 | } | |
2695 | ||
6d6bc0ad | 2696 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2697 | |
4866cde0 NP |
2698 | /** |
2699 | * prepare_task_switch - prepare to switch tasks | |
2700 | * @rq: the runqueue preparing to switch | |
421cee29 | 2701 | * @prev: the current task that is being switched out |
4866cde0 NP |
2702 | * @next: the task we are going to switch to. |
2703 | * | |
2704 | * This is called with the rq lock held and interrupts off. It must | |
2705 | * be paired with a subsequent finish_task_switch after the context | |
2706 | * switch. | |
2707 | * | |
2708 | * prepare_task_switch sets up locking and calls architecture specific | |
2709 | * hooks. | |
2710 | */ | |
e107be36 AK |
2711 | static inline void |
2712 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2713 | struct task_struct *next) | |
4866cde0 | 2714 | { |
e107be36 | 2715 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2716 | prepare_lock_switch(rq, next); |
2717 | prepare_arch_switch(next); | |
2718 | } | |
2719 | ||
1da177e4 LT |
2720 | /** |
2721 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2722 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2723 | * @prev: the thread we just switched away from. |
2724 | * | |
4866cde0 NP |
2725 | * finish_task_switch must be called after the context switch, paired |
2726 | * with a prepare_task_switch call before the context switch. | |
2727 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2728 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2729 | * |
2730 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2731 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2732 | * with the lock held can cause deadlocks; see schedule() for |
2733 | * details.) | |
2734 | */ | |
a9957449 | 2735 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2736 | __releases(rq->lock) |
2737 | { | |
1da177e4 | 2738 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2739 | long prev_state; |
1da177e4 LT |
2740 | |
2741 | rq->prev_mm = NULL; | |
2742 | ||
2743 | /* | |
2744 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2745 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2746 | * schedule one last time. The schedule call will never return, and |
2747 | * the scheduled task must drop that reference. | |
c394cc9f | 2748 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2749 | * still held, otherwise prev could be scheduled on another cpu, die |
2750 | * there before we look at prev->state, and then the reference would | |
2751 | * be dropped twice. | |
2752 | * Manfred Spraul <manfred@colorfullife.com> | |
2753 | */ | |
55a101f8 | 2754 | prev_state = prev->state; |
4866cde0 | 2755 | finish_arch_switch(prev); |
8381f65d JI |
2756 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2757 | local_irq_disable(); | |
2758 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2759 | perf_event_task_sched_in(current); |
8381f65d JI |
2760 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2761 | local_irq_enable(); | |
2762 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2763 | finish_lock_switch(rq, prev); |
e8fa1362 | 2764 | |
e107be36 | 2765 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2766 | if (mm) |
2767 | mmdrop(mm); | |
c394cc9f | 2768 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2769 | /* |
2770 | * Remove function-return probe instances associated with this | |
2771 | * task and put them back on the free list. | |
9761eea8 | 2772 | */ |
c6fd91f0 | 2773 | kprobe_flush_task(prev); |
1da177e4 | 2774 | put_task_struct(prev); |
c6fd91f0 | 2775 | } |
1da177e4 LT |
2776 | } |
2777 | ||
3f029d3c GH |
2778 | #ifdef CONFIG_SMP |
2779 | ||
2780 | /* assumes rq->lock is held */ | |
2781 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2782 | { | |
2783 | if (prev->sched_class->pre_schedule) | |
2784 | prev->sched_class->pre_schedule(rq, prev); | |
2785 | } | |
2786 | ||
2787 | /* rq->lock is NOT held, but preemption is disabled */ | |
2788 | static inline void post_schedule(struct rq *rq) | |
2789 | { | |
2790 | if (rq->post_schedule) { | |
2791 | unsigned long flags; | |
2792 | ||
05fa785c | 2793 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2794 | if (rq->curr->sched_class->post_schedule) |
2795 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2796 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2797 | |
2798 | rq->post_schedule = 0; | |
2799 | } | |
2800 | } | |
2801 | ||
2802 | #else | |
da19ab51 | 2803 | |
3f029d3c GH |
2804 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2805 | { | |
2806 | } | |
2807 | ||
2808 | static inline void post_schedule(struct rq *rq) | |
2809 | { | |
1da177e4 LT |
2810 | } |
2811 | ||
3f029d3c GH |
2812 | #endif |
2813 | ||
1da177e4 LT |
2814 | /** |
2815 | * schedule_tail - first thing a freshly forked thread must call. | |
2816 | * @prev: the thread we just switched away from. | |
2817 | */ | |
36c8b586 | 2818 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2819 | __releases(rq->lock) |
2820 | { | |
70b97a7f IM |
2821 | struct rq *rq = this_rq(); |
2822 | ||
4866cde0 | 2823 | finish_task_switch(rq, prev); |
da19ab51 | 2824 | |
3f029d3c GH |
2825 | /* |
2826 | * FIXME: do we need to worry about rq being invalidated by the | |
2827 | * task_switch? | |
2828 | */ | |
2829 | post_schedule(rq); | |
70b97a7f | 2830 | |
4866cde0 NP |
2831 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2832 | /* In this case, finish_task_switch does not reenable preemption */ | |
2833 | preempt_enable(); | |
2834 | #endif | |
1da177e4 | 2835 | if (current->set_child_tid) |
b488893a | 2836 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2837 | } |
2838 | ||
2839 | /* | |
2840 | * context_switch - switch to the new MM and the new | |
2841 | * thread's register state. | |
2842 | */ | |
dd41f596 | 2843 | static inline void |
70b97a7f | 2844 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2845 | struct task_struct *next) |
1da177e4 | 2846 | { |
dd41f596 | 2847 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2848 | |
e107be36 | 2849 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2850 | trace_sched_switch(prev, next); |
dd41f596 IM |
2851 | mm = next->mm; |
2852 | oldmm = prev->active_mm; | |
9226d125 ZA |
2853 | /* |
2854 | * For paravirt, this is coupled with an exit in switch_to to | |
2855 | * combine the page table reload and the switch backend into | |
2856 | * one hypercall. | |
2857 | */ | |
224101ed | 2858 | arch_start_context_switch(prev); |
9226d125 | 2859 | |
31915ab4 | 2860 | if (!mm) { |
1da177e4 LT |
2861 | next->active_mm = oldmm; |
2862 | atomic_inc(&oldmm->mm_count); | |
2863 | enter_lazy_tlb(oldmm, next); | |
2864 | } else | |
2865 | switch_mm(oldmm, mm, next); | |
2866 | ||
31915ab4 | 2867 | if (!prev->mm) { |
1da177e4 | 2868 | prev->active_mm = NULL; |
1da177e4 LT |
2869 | rq->prev_mm = oldmm; |
2870 | } | |
3a5f5e48 IM |
2871 | /* |
2872 | * Since the runqueue lock will be released by the next | |
2873 | * task (which is an invalid locking op but in the case | |
2874 | * of the scheduler it's an obvious special-case), so we | |
2875 | * do an early lockdep release here: | |
2876 | */ | |
2877 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2878 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2879 | #endif |
1da177e4 LT |
2880 | |
2881 | /* Here we just switch the register state and the stack. */ | |
2882 | switch_to(prev, next, prev); | |
2883 | ||
dd41f596 IM |
2884 | barrier(); |
2885 | /* | |
2886 | * this_rq must be evaluated again because prev may have moved | |
2887 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2888 | * frame will be invalid. | |
2889 | */ | |
2890 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2891 | } |
2892 | ||
2893 | /* | |
2894 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2895 | * | |
2896 | * externally visible scheduler statistics: current number of runnable | |
2897 | * threads, current number of uninterruptible-sleeping threads, total | |
2898 | * number of context switches performed since bootup. | |
2899 | */ | |
2900 | unsigned long nr_running(void) | |
2901 | { | |
2902 | unsigned long i, sum = 0; | |
2903 | ||
2904 | for_each_online_cpu(i) | |
2905 | sum += cpu_rq(i)->nr_running; | |
2906 | ||
2907 | return sum; | |
f711f609 | 2908 | } |
1da177e4 LT |
2909 | |
2910 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2911 | { |
1da177e4 | 2912 | unsigned long i, sum = 0; |
f711f609 | 2913 | |
0a945022 | 2914 | for_each_possible_cpu(i) |
1da177e4 | 2915 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2916 | |
2917 | /* | |
1da177e4 LT |
2918 | * Since we read the counters lockless, it might be slightly |
2919 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2920 | */ |
1da177e4 LT |
2921 | if (unlikely((long)sum < 0)) |
2922 | sum = 0; | |
f711f609 | 2923 | |
1da177e4 | 2924 | return sum; |
f711f609 | 2925 | } |
f711f609 | 2926 | |
1da177e4 | 2927 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2928 | { |
cc94abfc SR |
2929 | int i; |
2930 | unsigned long long sum = 0; | |
46cb4b7c | 2931 | |
0a945022 | 2932 | for_each_possible_cpu(i) |
1da177e4 | 2933 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2934 | |
1da177e4 LT |
2935 | return sum; |
2936 | } | |
483b4ee6 | 2937 | |
1da177e4 LT |
2938 | unsigned long nr_iowait(void) |
2939 | { | |
2940 | unsigned long i, sum = 0; | |
483b4ee6 | 2941 | |
0a945022 | 2942 | for_each_possible_cpu(i) |
1da177e4 | 2943 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2944 | |
1da177e4 LT |
2945 | return sum; |
2946 | } | |
483b4ee6 | 2947 | |
8c215bd3 | 2948 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2949 | { |
8c215bd3 | 2950 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2951 | return atomic_read(&this->nr_iowait); |
2952 | } | |
46cb4b7c | 2953 | |
69d25870 AV |
2954 | unsigned long this_cpu_load(void) |
2955 | { | |
2956 | struct rq *this = this_rq(); | |
2957 | return this->cpu_load[0]; | |
2958 | } | |
e790fb0b | 2959 | |
46cb4b7c | 2960 | |
dce48a84 TG |
2961 | /* Variables and functions for calc_load */ |
2962 | static atomic_long_t calc_load_tasks; | |
2963 | static unsigned long calc_load_update; | |
2964 | unsigned long avenrun[3]; | |
2965 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 2966 | |
74f5187a PZ |
2967 | static long calc_load_fold_active(struct rq *this_rq) |
2968 | { | |
2969 | long nr_active, delta = 0; | |
2970 | ||
2971 | nr_active = this_rq->nr_running; | |
2972 | nr_active += (long) this_rq->nr_uninterruptible; | |
2973 | ||
2974 | if (nr_active != this_rq->calc_load_active) { | |
2975 | delta = nr_active - this_rq->calc_load_active; | |
2976 | this_rq->calc_load_active = nr_active; | |
2977 | } | |
2978 | ||
2979 | return delta; | |
2980 | } | |
2981 | ||
2982 | #ifdef CONFIG_NO_HZ | |
2983 | /* | |
2984 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
2985 | * | |
2986 | * When making the ILB scale, we should try to pull this in as well. | |
2987 | */ | |
2988 | static atomic_long_t calc_load_tasks_idle; | |
2989 | ||
2990 | static void calc_load_account_idle(struct rq *this_rq) | |
2991 | { | |
2992 | long delta; | |
2993 | ||
2994 | delta = calc_load_fold_active(this_rq); | |
2995 | if (delta) | |
2996 | atomic_long_add(delta, &calc_load_tasks_idle); | |
2997 | } | |
2998 | ||
2999 | static long calc_load_fold_idle(void) | |
3000 | { | |
3001 | long delta = 0; | |
3002 | ||
3003 | /* | |
3004 | * Its got a race, we don't care... | |
3005 | */ | |
3006 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3007 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3008 | ||
3009 | return delta; | |
3010 | } | |
3011 | #else | |
3012 | static void calc_load_account_idle(struct rq *this_rq) | |
3013 | { | |
3014 | } | |
3015 | ||
3016 | static inline long calc_load_fold_idle(void) | |
3017 | { | |
3018 | return 0; | |
3019 | } | |
3020 | #endif | |
3021 | ||
2d02494f TG |
3022 | /** |
3023 | * get_avenrun - get the load average array | |
3024 | * @loads: pointer to dest load array | |
3025 | * @offset: offset to add | |
3026 | * @shift: shift count to shift the result left | |
3027 | * | |
3028 | * These values are estimates at best, so no need for locking. | |
3029 | */ | |
3030 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3031 | { | |
3032 | loads[0] = (avenrun[0] + offset) << shift; | |
3033 | loads[1] = (avenrun[1] + offset) << shift; | |
3034 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3035 | } |
46cb4b7c | 3036 | |
dce48a84 TG |
3037 | static unsigned long |
3038 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
db1b1fef | 3039 | { |
dce48a84 TG |
3040 | load *= exp; |
3041 | load += active * (FIXED_1 - exp); | |
3042 | return load >> FSHIFT; | |
3043 | } | |
46cb4b7c SS |
3044 | |
3045 | /* | |
dce48a84 TG |
3046 | * calc_load - update the avenrun load estimates 10 ticks after the |
3047 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3048 | */ |
dce48a84 | 3049 | void calc_global_load(void) |
7835b98b | 3050 | { |
dce48a84 TG |
3051 | unsigned long upd = calc_load_update + 10; |
3052 | long active; | |
1da177e4 | 3053 | |
dce48a84 TG |
3054 | if (time_before(jiffies, upd)) |
3055 | return; | |
1da177e4 | 3056 | |
dce48a84 TG |
3057 | active = atomic_long_read(&calc_load_tasks); |
3058 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3059 | |
dce48a84 TG |
3060 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3061 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3062 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3063 | |
dce48a84 TG |
3064 | calc_load_update += LOAD_FREQ; |
3065 | } | |
1da177e4 | 3066 | |
dce48a84 | 3067 | /* |
74f5187a PZ |
3068 | * Called from update_cpu_load() to periodically update this CPU's |
3069 | * active count. | |
dce48a84 TG |
3070 | */ |
3071 | static void calc_load_account_active(struct rq *this_rq) | |
3072 | { | |
74f5187a | 3073 | long delta; |
08c183f3 | 3074 | |
74f5187a PZ |
3075 | if (time_before(jiffies, this_rq->calc_load_update)) |
3076 | return; | |
783609c6 | 3077 | |
74f5187a PZ |
3078 | delta = calc_load_fold_active(this_rq); |
3079 | delta += calc_load_fold_idle(); | |
3080 | if (delta) | |
dce48a84 | 3081 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3082 | |
3083 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3084 | } |
3085 | ||
fdf3e95d VP |
3086 | /* |
3087 | * The exact cpuload at various idx values, calculated at every tick would be | |
3088 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3089 | * | |
3090 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3091 | * on nth tick when cpu may be busy, then we have: | |
3092 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3093 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3094 | * | |
3095 | * decay_load_missed() below does efficient calculation of | |
3096 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3097 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3098 | * | |
3099 | * The calculation is approximated on a 128 point scale. | |
3100 | * degrade_zero_ticks is the number of ticks after which load at any | |
3101 | * particular idx is approximated to be zero. | |
3102 | * degrade_factor is a precomputed table, a row for each load idx. | |
3103 | * Each column corresponds to degradation factor for a power of two ticks, | |
3104 | * based on 128 point scale. | |
3105 | * Example: | |
3106 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3107 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3108 | * | |
3109 | * With this power of 2 load factors, we can degrade the load n times | |
3110 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3111 | * n mult/shifts needed by the exact degradation. | |
3112 | */ | |
3113 | #define DEGRADE_SHIFT 7 | |
3114 | static const unsigned char | |
3115 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3116 | static const unsigned char | |
3117 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3118 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3119 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3120 | {96, 72, 40, 12, 1, 0, 0}, | |
3121 | {112, 98, 75, 43, 15, 1, 0}, | |
3122 | {120, 112, 98, 76, 45, 16, 2} }; | |
3123 | ||
3124 | /* | |
3125 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3126 | * would be when CPU is idle and so we just decay the old load without | |
3127 | * adding any new load. | |
3128 | */ | |
3129 | static unsigned long | |
3130 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3131 | { | |
3132 | int j = 0; | |
3133 | ||
3134 | if (!missed_updates) | |
3135 | return load; | |
3136 | ||
3137 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3138 | return 0; | |
3139 | ||
3140 | if (idx == 1) | |
3141 | return load >> missed_updates; | |
3142 | ||
3143 | while (missed_updates) { | |
3144 | if (missed_updates % 2) | |
3145 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3146 | ||
3147 | missed_updates >>= 1; | |
3148 | j++; | |
3149 | } | |
3150 | return load; | |
3151 | } | |
3152 | ||
46cb4b7c | 3153 | /* |
dd41f596 | 3154 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3155 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3156 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3157 | */ |
dd41f596 | 3158 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3159 | { |
495eca49 | 3160 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3161 | unsigned long curr_jiffies = jiffies; |
3162 | unsigned long pending_updates; | |
dd41f596 | 3163 | int i, scale; |
46cb4b7c | 3164 | |
dd41f596 | 3165 | this_rq->nr_load_updates++; |
46cb4b7c | 3166 | |
fdf3e95d VP |
3167 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3168 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3169 | return; | |
3170 | ||
3171 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3172 | this_rq->last_load_update_tick = curr_jiffies; | |
3173 | ||
dd41f596 | 3174 | /* Update our load: */ |
fdf3e95d VP |
3175 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3176 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3177 | unsigned long old_load, new_load; |
7d1e6a9b | 3178 | |
dd41f596 | 3179 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3180 | |
dd41f596 | 3181 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3182 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3183 | new_load = this_load; |
a25707f3 IM |
3184 | /* |
3185 | * Round up the averaging division if load is increasing. This | |
3186 | * prevents us from getting stuck on 9 if the load is 10, for | |
3187 | * example. | |
3188 | */ | |
3189 | if (new_load > old_load) | |
fdf3e95d VP |
3190 | new_load += scale - 1; |
3191 | ||
3192 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3193 | } |
da2b71ed SS |
3194 | |
3195 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3196 | } |
3197 | ||
3198 | static void update_cpu_load_active(struct rq *this_rq) | |
3199 | { | |
3200 | update_cpu_load(this_rq); | |
46cb4b7c | 3201 | |
74f5187a | 3202 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3203 | } |
3204 | ||
dd41f596 | 3205 | #ifdef CONFIG_SMP |
8a0be9ef | 3206 | |
46cb4b7c | 3207 | /* |
38022906 PZ |
3208 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3209 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3210 | */ |
38022906 | 3211 | void sched_exec(void) |
46cb4b7c | 3212 | { |
38022906 | 3213 | struct task_struct *p = current; |
1da177e4 | 3214 | unsigned long flags; |
70b97a7f | 3215 | struct rq *rq; |
0017d735 | 3216 | int dest_cpu; |
46cb4b7c | 3217 | |
1da177e4 | 3218 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3219 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3220 | if (dest_cpu == smp_processor_id()) | |
3221 | goto unlock; | |
38022906 | 3222 | |
46cb4b7c | 3223 | /* |
38022906 | 3224 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3225 | */ |
30da688e | 3226 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
969c7921 TH |
3227 | likely(cpu_active(dest_cpu)) && migrate_task(p, dest_cpu)) { |
3228 | struct migration_arg arg = { p, dest_cpu }; | |
46cb4b7c | 3229 | |
1da177e4 | 3230 | task_rq_unlock(rq, &flags); |
969c7921 | 3231 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3232 | return; |
3233 | } | |
0017d735 | 3234 | unlock: |
1da177e4 | 3235 | task_rq_unlock(rq, &flags); |
1da177e4 | 3236 | } |
dd41f596 | 3237 | |
1da177e4 LT |
3238 | #endif |
3239 | ||
1da177e4 LT |
3240 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3241 | ||
3242 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3243 | ||
3244 | /* | |
c5f8d995 | 3245 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3246 | * @p in case that task is currently running. |
c5f8d995 HS |
3247 | * |
3248 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3249 | */ |
c5f8d995 HS |
3250 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3251 | { | |
3252 | u64 ns = 0; | |
3253 | ||
3254 | if (task_current(rq, p)) { | |
3255 | update_rq_clock(rq); | |
305e6835 | 3256 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3257 | if ((s64)ns < 0) |
3258 | ns = 0; | |
3259 | } | |
3260 | ||
3261 | return ns; | |
3262 | } | |
3263 | ||
bb34d92f | 3264 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3265 | { |
1da177e4 | 3266 | unsigned long flags; |
41b86e9c | 3267 | struct rq *rq; |
bb34d92f | 3268 | u64 ns = 0; |
48f24c4d | 3269 | |
41b86e9c | 3270 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3271 | ns = do_task_delta_exec(p, rq); |
3272 | task_rq_unlock(rq, &flags); | |
1508487e | 3273 | |
c5f8d995 HS |
3274 | return ns; |
3275 | } | |
f06febc9 | 3276 | |
c5f8d995 HS |
3277 | /* |
3278 | * Return accounted runtime for the task. | |
3279 | * In case the task is currently running, return the runtime plus current's | |
3280 | * pending runtime that have not been accounted yet. | |
3281 | */ | |
3282 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3283 | { | |
3284 | unsigned long flags; | |
3285 | struct rq *rq; | |
3286 | u64 ns = 0; | |
3287 | ||
3288 | rq = task_rq_lock(p, &flags); | |
3289 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3290 | task_rq_unlock(rq, &flags); | |
3291 | ||
3292 | return ns; | |
3293 | } | |
48f24c4d | 3294 | |
c5f8d995 HS |
3295 | /* |
3296 | * Return sum_exec_runtime for the thread group. | |
3297 | * In case the task is currently running, return the sum plus current's | |
3298 | * pending runtime that have not been accounted yet. | |
3299 | * | |
3300 | * Note that the thread group might have other running tasks as well, | |
3301 | * so the return value not includes other pending runtime that other | |
3302 | * running tasks might have. | |
3303 | */ | |
3304 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3305 | { | |
3306 | struct task_cputime totals; | |
3307 | unsigned long flags; | |
3308 | struct rq *rq; | |
3309 | u64 ns; | |
3310 | ||
3311 | rq = task_rq_lock(p, &flags); | |
3312 | thread_group_cputime(p, &totals); | |
3313 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3314 | task_rq_unlock(rq, &flags); |
48f24c4d | 3315 | |
1da177e4 LT |
3316 | return ns; |
3317 | } | |
3318 | ||
1da177e4 LT |
3319 | /* |
3320 | * Account user cpu time to a process. | |
3321 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3322 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3323 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3324 | */ |
457533a7 MS |
3325 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3326 | cputime_t cputime_scaled) | |
1da177e4 LT |
3327 | { |
3328 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3329 | cputime64_t tmp; | |
3330 | ||
457533a7 | 3331 | /* Add user time to process. */ |
1da177e4 | 3332 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3333 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3334 | account_group_user_time(p, cputime); |
1da177e4 LT |
3335 | |
3336 | /* Add user time to cpustat. */ | |
3337 | tmp = cputime_to_cputime64(cputime); | |
3338 | if (TASK_NICE(p) > 0) | |
3339 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3340 | else | |
3341 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3342 | |
3343 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3344 | /* Account for user time used */ |
3345 | acct_update_integrals(p); | |
1da177e4 LT |
3346 | } |
3347 | ||
94886b84 LV |
3348 | /* |
3349 | * Account guest cpu time to a process. | |
3350 | * @p: the process that the cpu time gets accounted to | |
3351 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3352 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3353 | */ |
457533a7 MS |
3354 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3355 | cputime_t cputime_scaled) | |
94886b84 LV |
3356 | { |
3357 | cputime64_t tmp; | |
3358 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3359 | ||
3360 | tmp = cputime_to_cputime64(cputime); | |
3361 | ||
457533a7 | 3362 | /* Add guest time to process. */ |
94886b84 | 3363 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3364 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3365 | account_group_user_time(p, cputime); |
94886b84 LV |
3366 | p->gtime = cputime_add(p->gtime, cputime); |
3367 | ||
457533a7 | 3368 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3369 | if (TASK_NICE(p) > 0) { |
3370 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3371 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3372 | } else { | |
3373 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3374 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3375 | } | |
94886b84 LV |
3376 | } |
3377 | ||
1da177e4 LT |
3378 | /* |
3379 | * Account system cpu time to a process. | |
3380 | * @p: the process that the cpu time gets accounted to | |
3381 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3382 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3383 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3384 | */ |
3385 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3386 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3387 | { |
3388 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3389 | cputime64_t tmp; |
3390 | ||
983ed7a6 | 3391 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3392 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3393 | return; |
3394 | } | |
94886b84 | 3395 | |
457533a7 | 3396 | /* Add system time to process. */ |
1da177e4 | 3397 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3398 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3399 | account_group_system_time(p, cputime); |
1da177e4 LT |
3400 | |
3401 | /* Add system time to cpustat. */ | |
3402 | tmp = cputime_to_cputime64(cputime); | |
3403 | if (hardirq_count() - hardirq_offset) | |
3404 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
75e1056f | 3405 | else if (in_serving_softirq()) |
1da177e4 | 3406 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); |
1da177e4 | 3407 | else |
79741dd3 MS |
3408 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3409 | ||
ef12fefa BR |
3410 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3411 | ||
1da177e4 LT |
3412 | /* Account for system time used */ |
3413 | acct_update_integrals(p); | |
1da177e4 LT |
3414 | } |
3415 | ||
c66f08be | 3416 | /* |
1da177e4 | 3417 | * Account for involuntary wait time. |
1da177e4 | 3418 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3419 | */ |
79741dd3 | 3420 | void account_steal_time(cputime_t cputime) |
c66f08be | 3421 | { |
79741dd3 MS |
3422 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3423 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3424 | ||
3425 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3426 | } |
3427 | ||
1da177e4 | 3428 | /* |
79741dd3 MS |
3429 | * Account for idle time. |
3430 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3431 | */ |
79741dd3 | 3432 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3433 | { |
3434 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3435 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3436 | struct rq *rq = this_rq(); |
1da177e4 | 3437 | |
79741dd3 MS |
3438 | if (atomic_read(&rq->nr_iowait) > 0) |
3439 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3440 | else | |
3441 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3442 | } |
3443 | ||
79741dd3 MS |
3444 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3445 | ||
3446 | /* | |
3447 | * Account a single tick of cpu time. | |
3448 | * @p: the process that the cpu time gets accounted to | |
3449 | * @user_tick: indicates if the tick is a user or a system tick | |
3450 | */ | |
3451 | void account_process_tick(struct task_struct *p, int user_tick) | |
3452 | { | |
a42548a1 | 3453 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3454 | struct rq *rq = this_rq(); |
3455 | ||
3456 | if (user_tick) | |
a42548a1 | 3457 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3458 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3459 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3460 | one_jiffy_scaled); |
3461 | else | |
a42548a1 | 3462 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3463 | } |
3464 | ||
3465 | /* | |
3466 | * Account multiple ticks of steal time. | |
3467 | * @p: the process from which the cpu time has been stolen | |
3468 | * @ticks: number of stolen ticks | |
3469 | */ | |
3470 | void account_steal_ticks(unsigned long ticks) | |
3471 | { | |
3472 | account_steal_time(jiffies_to_cputime(ticks)); | |
3473 | } | |
3474 | ||
3475 | /* | |
3476 | * Account multiple ticks of idle time. | |
3477 | * @ticks: number of stolen ticks | |
3478 | */ | |
3479 | void account_idle_ticks(unsigned long ticks) | |
3480 | { | |
3481 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3482 | } |
3483 | ||
79741dd3 MS |
3484 | #endif |
3485 | ||
49048622 BS |
3486 | /* |
3487 | * Use precise platform statistics if available: | |
3488 | */ | |
3489 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3490 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3491 | { |
d99ca3b9 HS |
3492 | *ut = p->utime; |
3493 | *st = p->stime; | |
49048622 BS |
3494 | } |
3495 | ||
0cf55e1e | 3496 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3497 | { |
0cf55e1e HS |
3498 | struct task_cputime cputime; |
3499 | ||
3500 | thread_group_cputime(p, &cputime); | |
3501 | ||
3502 | *ut = cputime.utime; | |
3503 | *st = cputime.stime; | |
49048622 BS |
3504 | } |
3505 | #else | |
761b1d26 HS |
3506 | |
3507 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3508 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3509 | #endif |
3510 | ||
d180c5bc | 3511 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3512 | { |
d99ca3b9 | 3513 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3514 | |
3515 | /* | |
3516 | * Use CFS's precise accounting: | |
3517 | */ | |
d180c5bc | 3518 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3519 | |
3520 | if (total) { | |
e75e863d | 3521 | u64 temp = rtime; |
d180c5bc | 3522 | |
e75e863d | 3523 | temp *= utime; |
49048622 | 3524 | do_div(temp, total); |
d180c5bc HS |
3525 | utime = (cputime_t)temp; |
3526 | } else | |
3527 | utime = rtime; | |
49048622 | 3528 | |
d180c5bc HS |
3529 | /* |
3530 | * Compare with previous values, to keep monotonicity: | |
3531 | */ | |
761b1d26 | 3532 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3533 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3534 | |
d99ca3b9 HS |
3535 | *ut = p->prev_utime; |
3536 | *st = p->prev_stime; | |
49048622 BS |
3537 | } |
3538 | ||
0cf55e1e HS |
3539 | /* |
3540 | * Must be called with siglock held. | |
3541 | */ | |
3542 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3543 | { |
0cf55e1e HS |
3544 | struct signal_struct *sig = p->signal; |
3545 | struct task_cputime cputime; | |
3546 | cputime_t rtime, utime, total; | |
49048622 | 3547 | |
0cf55e1e | 3548 | thread_group_cputime(p, &cputime); |
49048622 | 3549 | |
0cf55e1e HS |
3550 | total = cputime_add(cputime.utime, cputime.stime); |
3551 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3552 | |
0cf55e1e | 3553 | if (total) { |
e75e863d | 3554 | u64 temp = rtime; |
49048622 | 3555 | |
e75e863d | 3556 | temp *= cputime.utime; |
0cf55e1e HS |
3557 | do_div(temp, total); |
3558 | utime = (cputime_t)temp; | |
3559 | } else | |
3560 | utime = rtime; | |
3561 | ||
3562 | sig->prev_utime = max(sig->prev_utime, utime); | |
3563 | sig->prev_stime = max(sig->prev_stime, | |
3564 | cputime_sub(rtime, sig->prev_utime)); | |
3565 | ||
3566 | *ut = sig->prev_utime; | |
3567 | *st = sig->prev_stime; | |
49048622 | 3568 | } |
49048622 | 3569 | #endif |
49048622 | 3570 | |
7835b98b CL |
3571 | /* |
3572 | * This function gets called by the timer code, with HZ frequency. | |
3573 | * We call it with interrupts disabled. | |
3574 | * | |
3575 | * It also gets called by the fork code, when changing the parent's | |
3576 | * timeslices. | |
3577 | */ | |
3578 | void scheduler_tick(void) | |
3579 | { | |
7835b98b CL |
3580 | int cpu = smp_processor_id(); |
3581 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3582 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3583 | |
3584 | sched_clock_tick(); | |
dd41f596 | 3585 | |
05fa785c | 3586 | raw_spin_lock(&rq->lock); |
3e51f33f | 3587 | update_rq_clock(rq); |
fdf3e95d | 3588 | update_cpu_load_active(rq); |
fa85ae24 | 3589 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3590 | raw_spin_unlock(&rq->lock); |
7835b98b | 3591 | |
e9d2b064 | 3592 | perf_event_task_tick(); |
e220d2dc | 3593 | |
e418e1c2 | 3594 | #ifdef CONFIG_SMP |
dd41f596 IM |
3595 | rq->idle_at_tick = idle_cpu(cpu); |
3596 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3597 | #endif |
1da177e4 LT |
3598 | } |
3599 | ||
132380a0 | 3600 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3601 | { |
3602 | if (in_lock_functions(addr)) { | |
3603 | addr = CALLER_ADDR2; | |
3604 | if (in_lock_functions(addr)) | |
3605 | addr = CALLER_ADDR3; | |
3606 | } | |
3607 | return addr; | |
3608 | } | |
1da177e4 | 3609 | |
7e49fcce SR |
3610 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3611 | defined(CONFIG_PREEMPT_TRACER)) | |
3612 | ||
43627582 | 3613 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3614 | { |
6cd8a4bb | 3615 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3616 | /* |
3617 | * Underflow? | |
3618 | */ | |
9a11b49a IM |
3619 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3620 | return; | |
6cd8a4bb | 3621 | #endif |
1da177e4 | 3622 | preempt_count() += val; |
6cd8a4bb | 3623 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3624 | /* |
3625 | * Spinlock count overflowing soon? | |
3626 | */ | |
33859f7f MOS |
3627 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3628 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3629 | #endif |
3630 | if (preempt_count() == val) | |
3631 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3632 | } |
3633 | EXPORT_SYMBOL(add_preempt_count); | |
3634 | ||
43627582 | 3635 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3636 | { |
6cd8a4bb | 3637 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3638 | /* |
3639 | * Underflow? | |
3640 | */ | |
01e3eb82 | 3641 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3642 | return; |
1da177e4 LT |
3643 | /* |
3644 | * Is the spinlock portion underflowing? | |
3645 | */ | |
9a11b49a IM |
3646 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3647 | !(preempt_count() & PREEMPT_MASK))) | |
3648 | return; | |
6cd8a4bb | 3649 | #endif |
9a11b49a | 3650 | |
6cd8a4bb SR |
3651 | if (preempt_count() == val) |
3652 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3653 | preempt_count() -= val; |
3654 | } | |
3655 | EXPORT_SYMBOL(sub_preempt_count); | |
3656 | ||
3657 | #endif | |
3658 | ||
3659 | /* | |
dd41f596 | 3660 | * Print scheduling while atomic bug: |
1da177e4 | 3661 | */ |
dd41f596 | 3662 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3663 | { |
838225b4 SS |
3664 | struct pt_regs *regs = get_irq_regs(); |
3665 | ||
3df0fc5b PZ |
3666 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3667 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3668 | |
dd41f596 | 3669 | debug_show_held_locks(prev); |
e21f5b15 | 3670 | print_modules(); |
dd41f596 IM |
3671 | if (irqs_disabled()) |
3672 | print_irqtrace_events(prev); | |
838225b4 SS |
3673 | |
3674 | if (regs) | |
3675 | show_regs(regs); | |
3676 | else | |
3677 | dump_stack(); | |
dd41f596 | 3678 | } |
1da177e4 | 3679 | |
dd41f596 IM |
3680 | /* |
3681 | * Various schedule()-time debugging checks and statistics: | |
3682 | */ | |
3683 | static inline void schedule_debug(struct task_struct *prev) | |
3684 | { | |
1da177e4 | 3685 | /* |
41a2d6cf | 3686 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3687 | * schedule() atomically, we ignore that path for now. |
3688 | * Otherwise, whine if we are scheduling when we should not be. | |
3689 | */ | |
3f33a7ce | 3690 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3691 | __schedule_bug(prev); |
3692 | ||
1da177e4 LT |
3693 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3694 | ||
2d72376b | 3695 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3696 | #ifdef CONFIG_SCHEDSTATS |
3697 | if (unlikely(prev->lock_depth >= 0)) { | |
2d72376b IM |
3698 | schedstat_inc(this_rq(), bkl_count); |
3699 | schedstat_inc(prev, sched_info.bkl_count); | |
b8efb561 IM |
3700 | } |
3701 | #endif | |
dd41f596 IM |
3702 | } |
3703 | ||
6cecd084 | 3704 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3705 | { |
a64692a3 MG |
3706 | if (prev->se.on_rq) |
3707 | update_rq_clock(rq); | |
3708 | rq->skip_clock_update = 0; | |
6cecd084 | 3709 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3710 | } |
3711 | ||
dd41f596 IM |
3712 | /* |
3713 | * Pick up the highest-prio task: | |
3714 | */ | |
3715 | static inline struct task_struct * | |
b67802ea | 3716 | pick_next_task(struct rq *rq) |
dd41f596 | 3717 | { |
5522d5d5 | 3718 | const struct sched_class *class; |
dd41f596 | 3719 | struct task_struct *p; |
1da177e4 LT |
3720 | |
3721 | /* | |
dd41f596 IM |
3722 | * Optimization: we know that if all tasks are in |
3723 | * the fair class we can call that function directly: | |
1da177e4 | 3724 | */ |
dd41f596 | 3725 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3726 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3727 | if (likely(p)) |
3728 | return p; | |
1da177e4 LT |
3729 | } |
3730 | ||
34f971f6 | 3731 | for_each_class(class) { |
fb8d4724 | 3732 | p = class->pick_next_task(rq); |
dd41f596 IM |
3733 | if (p) |
3734 | return p; | |
dd41f596 | 3735 | } |
34f971f6 PZ |
3736 | |
3737 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3738 | } |
1da177e4 | 3739 | |
dd41f596 IM |
3740 | /* |
3741 | * schedule() is the main scheduler function. | |
3742 | */ | |
ff743345 | 3743 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3744 | { |
3745 | struct task_struct *prev, *next; | |
67ca7bde | 3746 | unsigned long *switch_count; |
dd41f596 | 3747 | struct rq *rq; |
31656519 | 3748 | int cpu; |
dd41f596 | 3749 | |
ff743345 PZ |
3750 | need_resched: |
3751 | preempt_disable(); | |
dd41f596 IM |
3752 | cpu = smp_processor_id(); |
3753 | rq = cpu_rq(cpu); | |
25502a6c | 3754 | rcu_note_context_switch(cpu); |
dd41f596 | 3755 | prev = rq->curr; |
dd41f596 IM |
3756 | |
3757 | release_kernel_lock(prev); | |
3758 | need_resched_nonpreemptible: | |
3759 | ||
3760 | schedule_debug(prev); | |
1da177e4 | 3761 | |
31656519 | 3762 | if (sched_feat(HRTICK)) |
f333fdc9 | 3763 | hrtick_clear(rq); |
8f4d37ec | 3764 | |
05fa785c | 3765 | raw_spin_lock_irq(&rq->lock); |
1e819950 | 3766 | clear_tsk_need_resched(prev); |
1da177e4 | 3767 | |
246d86b5 | 3768 | switch_count = &prev->nivcsw; |
1da177e4 | 3769 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3770 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3771 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
3772 | } else { |
3773 | /* | |
3774 | * If a worker is going to sleep, notify and | |
3775 | * ask workqueue whether it wants to wake up a | |
3776 | * task to maintain concurrency. If so, wake | |
3777 | * up the task. | |
3778 | */ | |
3779 | if (prev->flags & PF_WQ_WORKER) { | |
3780 | struct task_struct *to_wakeup; | |
3781 | ||
3782 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3783 | if (to_wakeup) | |
3784 | try_to_wake_up_local(to_wakeup); | |
3785 | } | |
371fd7e7 | 3786 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 3787 | } |
dd41f596 | 3788 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3789 | } |
3790 | ||
3f029d3c | 3791 | pre_schedule(rq, prev); |
f65eda4f | 3792 | |
dd41f596 | 3793 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3794 | idle_balance(cpu, rq); |
1da177e4 | 3795 | |
df1c99d4 | 3796 | put_prev_task(rq, prev); |
b67802ea | 3797 | next = pick_next_task(rq); |
1da177e4 | 3798 | |
1da177e4 | 3799 | if (likely(prev != next)) { |
673a90a1 | 3800 | sched_info_switch(prev, next); |
49f47433 | 3801 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3802 | |
1da177e4 LT |
3803 | rq->nr_switches++; |
3804 | rq->curr = next; | |
3805 | ++*switch_count; | |
3806 | ||
dd41f596 | 3807 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3808 | /* |
246d86b5 ON |
3809 | * The context switch have flipped the stack from under us |
3810 | * and restored the local variables which were saved when | |
3811 | * this task called schedule() in the past. prev == current | |
3812 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3813 | */ |
3814 | cpu = smp_processor_id(); | |
3815 | rq = cpu_rq(cpu); | |
1da177e4 | 3816 | } else |
05fa785c | 3817 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3818 | |
3f029d3c | 3819 | post_schedule(rq); |
1da177e4 | 3820 | |
246d86b5 | 3821 | if (unlikely(reacquire_kernel_lock(prev))) |
1da177e4 | 3822 | goto need_resched_nonpreemptible; |
8f4d37ec | 3823 | |
1da177e4 | 3824 | preempt_enable_no_resched(); |
ff743345 | 3825 | if (need_resched()) |
1da177e4 LT |
3826 | goto need_resched; |
3827 | } | |
1da177e4 LT |
3828 | EXPORT_SYMBOL(schedule); |
3829 | ||
c08f7829 | 3830 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
3831 | /* |
3832 | * Look out! "owner" is an entirely speculative pointer | |
3833 | * access and not reliable. | |
3834 | */ | |
3835 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
3836 | { | |
3837 | unsigned int cpu; | |
3838 | struct rq *rq; | |
3839 | ||
3840 | if (!sched_feat(OWNER_SPIN)) | |
3841 | return 0; | |
3842 | ||
3843 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
3844 | /* | |
3845 | * Need to access the cpu field knowing that | |
3846 | * DEBUG_PAGEALLOC could have unmapped it if | |
3847 | * the mutex owner just released it and exited. | |
3848 | */ | |
3849 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 3850 | return 0; |
0d66bf6d PZ |
3851 | #else |
3852 | cpu = owner->cpu; | |
3853 | #endif | |
3854 | ||
3855 | /* | |
3856 | * Even if the access succeeded (likely case), | |
3857 | * the cpu field may no longer be valid. | |
3858 | */ | |
3859 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 3860 | return 0; |
0d66bf6d PZ |
3861 | |
3862 | /* | |
3863 | * We need to validate that we can do a | |
3864 | * get_cpu() and that we have the percpu area. | |
3865 | */ | |
3866 | if (!cpu_online(cpu)) | |
4b402210 | 3867 | return 0; |
0d66bf6d PZ |
3868 | |
3869 | rq = cpu_rq(cpu); | |
3870 | ||
3871 | for (;;) { | |
3872 | /* | |
3873 | * Owner changed, break to re-assess state. | |
3874 | */ | |
9d0f4dcc TC |
3875 | if (lock->owner != owner) { |
3876 | /* | |
3877 | * If the lock has switched to a different owner, | |
3878 | * we likely have heavy contention. Return 0 to quit | |
3879 | * optimistic spinning and not contend further: | |
3880 | */ | |
3881 | if (lock->owner) | |
3882 | return 0; | |
0d66bf6d | 3883 | break; |
9d0f4dcc | 3884 | } |
0d66bf6d PZ |
3885 | |
3886 | /* | |
3887 | * Is that owner really running on that cpu? | |
3888 | */ | |
3889 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
3890 | return 0; | |
3891 | ||
335d7afb | 3892 | arch_mutex_cpu_relax(); |
0d66bf6d | 3893 | } |
4b402210 | 3894 | |
0d66bf6d PZ |
3895 | return 1; |
3896 | } | |
3897 | #endif | |
3898 | ||
1da177e4 LT |
3899 | #ifdef CONFIG_PREEMPT |
3900 | /* | |
2ed6e34f | 3901 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 3902 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
3903 | * occur there and call schedule directly. |
3904 | */ | |
d1f74e20 | 3905 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
3906 | { |
3907 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3908 | |
1da177e4 LT |
3909 | /* |
3910 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 3911 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 3912 | */ |
beed33a8 | 3913 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
3914 | return; |
3915 | ||
3a5c359a | 3916 | do { |
d1f74e20 | 3917 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 3918 | schedule(); |
d1f74e20 | 3919 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 3920 | |
3a5c359a AK |
3921 | /* |
3922 | * Check again in case we missed a preemption opportunity | |
3923 | * between schedule and now. | |
3924 | */ | |
3925 | barrier(); | |
5ed0cec0 | 3926 | } while (need_resched()); |
1da177e4 | 3927 | } |
1da177e4 LT |
3928 | EXPORT_SYMBOL(preempt_schedule); |
3929 | ||
3930 | /* | |
2ed6e34f | 3931 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
3932 | * off of irq context. |
3933 | * Note, that this is called and return with irqs disabled. This will | |
3934 | * protect us against recursive calling from irq. | |
3935 | */ | |
3936 | asmlinkage void __sched preempt_schedule_irq(void) | |
3937 | { | |
3938 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 3939 | |
2ed6e34f | 3940 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
3941 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
3942 | ||
3a5c359a AK |
3943 | do { |
3944 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
3945 | local_irq_enable(); |
3946 | schedule(); | |
3947 | local_irq_disable(); | |
3a5c359a | 3948 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 3949 | |
3a5c359a AK |
3950 | /* |
3951 | * Check again in case we missed a preemption opportunity | |
3952 | * between schedule and now. | |
3953 | */ | |
3954 | barrier(); | |
5ed0cec0 | 3955 | } while (need_resched()); |
1da177e4 LT |
3956 | } |
3957 | ||
3958 | #endif /* CONFIG_PREEMPT */ | |
3959 | ||
63859d4f | 3960 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 3961 | void *key) |
1da177e4 | 3962 | { |
63859d4f | 3963 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 3964 | } |
1da177e4 LT |
3965 | EXPORT_SYMBOL(default_wake_function); |
3966 | ||
3967 | /* | |
41a2d6cf IM |
3968 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
3969 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
3970 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
3971 | * | |
3972 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 3973 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
3974 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
3975 | */ | |
78ddb08f | 3976 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 3977 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 3978 | { |
2e45874c | 3979 | wait_queue_t *curr, *next; |
1da177e4 | 3980 | |
2e45874c | 3981 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
3982 | unsigned flags = curr->flags; |
3983 | ||
63859d4f | 3984 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 3985 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
3986 | break; |
3987 | } | |
3988 | } | |
3989 | ||
3990 | /** | |
3991 | * __wake_up - wake up threads blocked on a waitqueue. | |
3992 | * @q: the waitqueue | |
3993 | * @mode: which threads | |
3994 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 3995 | * @key: is directly passed to the wakeup function |
50fa610a DH |
3996 | * |
3997 | * It may be assumed that this function implies a write memory barrier before | |
3998 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 3999 | */ |
7ad5b3a5 | 4000 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4001 | int nr_exclusive, void *key) |
1da177e4 LT |
4002 | { |
4003 | unsigned long flags; | |
4004 | ||
4005 | spin_lock_irqsave(&q->lock, flags); | |
4006 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4007 | spin_unlock_irqrestore(&q->lock, flags); | |
4008 | } | |
1da177e4 LT |
4009 | EXPORT_SYMBOL(__wake_up); |
4010 | ||
4011 | /* | |
4012 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4013 | */ | |
7ad5b3a5 | 4014 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4015 | { |
4016 | __wake_up_common(q, mode, 1, 0, NULL); | |
4017 | } | |
22c43c81 | 4018 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4019 | |
4ede816a DL |
4020 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4021 | { | |
4022 | __wake_up_common(q, mode, 1, 0, key); | |
4023 | } | |
4024 | ||
1da177e4 | 4025 | /** |
4ede816a | 4026 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4027 | * @q: the waitqueue |
4028 | * @mode: which threads | |
4029 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4030 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4031 | * |
4032 | * The sync wakeup differs that the waker knows that it will schedule | |
4033 | * away soon, so while the target thread will be woken up, it will not | |
4034 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4035 | * with each other. This can prevent needless bouncing between CPUs. | |
4036 | * | |
4037 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4038 | * |
4039 | * It may be assumed that this function implies a write memory barrier before | |
4040 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4041 | */ |
4ede816a DL |
4042 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4043 | int nr_exclusive, void *key) | |
1da177e4 LT |
4044 | { |
4045 | unsigned long flags; | |
7d478721 | 4046 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4047 | |
4048 | if (unlikely(!q)) | |
4049 | return; | |
4050 | ||
4051 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4052 | wake_flags = 0; |
1da177e4 LT |
4053 | |
4054 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4055 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4056 | spin_unlock_irqrestore(&q->lock, flags); |
4057 | } | |
4ede816a DL |
4058 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4059 | ||
4060 | /* | |
4061 | * __wake_up_sync - see __wake_up_sync_key() | |
4062 | */ | |
4063 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4064 | { | |
4065 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4066 | } | |
1da177e4 LT |
4067 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4068 | ||
65eb3dc6 KD |
4069 | /** |
4070 | * complete: - signals a single thread waiting on this completion | |
4071 | * @x: holds the state of this particular completion | |
4072 | * | |
4073 | * This will wake up a single thread waiting on this completion. Threads will be | |
4074 | * awakened in the same order in which they were queued. | |
4075 | * | |
4076 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4077 | * |
4078 | * It may be assumed that this function implies a write memory barrier before | |
4079 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4080 | */ |
b15136e9 | 4081 | void complete(struct completion *x) |
1da177e4 LT |
4082 | { |
4083 | unsigned long flags; | |
4084 | ||
4085 | spin_lock_irqsave(&x->wait.lock, flags); | |
4086 | x->done++; | |
d9514f6c | 4087 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4088 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4089 | } | |
4090 | EXPORT_SYMBOL(complete); | |
4091 | ||
65eb3dc6 KD |
4092 | /** |
4093 | * complete_all: - signals all threads waiting on this completion | |
4094 | * @x: holds the state of this particular completion | |
4095 | * | |
4096 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4097 | * |
4098 | * It may be assumed that this function implies a write memory barrier before | |
4099 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4100 | */ |
b15136e9 | 4101 | void complete_all(struct completion *x) |
1da177e4 LT |
4102 | { |
4103 | unsigned long flags; | |
4104 | ||
4105 | spin_lock_irqsave(&x->wait.lock, flags); | |
4106 | x->done += UINT_MAX/2; | |
d9514f6c | 4107 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4108 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4109 | } | |
4110 | EXPORT_SYMBOL(complete_all); | |
4111 | ||
8cbbe86d AK |
4112 | static inline long __sched |
4113 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4114 | { |
1da177e4 LT |
4115 | if (!x->done) { |
4116 | DECLARE_WAITQUEUE(wait, current); | |
4117 | ||
a93d2f17 | 4118 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4119 | do { |
94d3d824 | 4120 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4121 | timeout = -ERESTARTSYS; |
4122 | break; | |
8cbbe86d AK |
4123 | } |
4124 | __set_current_state(state); | |
1da177e4 LT |
4125 | spin_unlock_irq(&x->wait.lock); |
4126 | timeout = schedule_timeout(timeout); | |
4127 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4128 | } while (!x->done && timeout); |
1da177e4 | 4129 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4130 | if (!x->done) |
4131 | return timeout; | |
1da177e4 LT |
4132 | } |
4133 | x->done--; | |
ea71a546 | 4134 | return timeout ?: 1; |
1da177e4 | 4135 | } |
1da177e4 | 4136 | |
8cbbe86d AK |
4137 | static long __sched |
4138 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4139 | { |
1da177e4 LT |
4140 | might_sleep(); |
4141 | ||
4142 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4143 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4144 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4145 | return timeout; |
4146 | } | |
1da177e4 | 4147 | |
65eb3dc6 KD |
4148 | /** |
4149 | * wait_for_completion: - waits for completion of a task | |
4150 | * @x: holds the state of this particular completion | |
4151 | * | |
4152 | * This waits to be signaled for completion of a specific task. It is NOT | |
4153 | * interruptible and there is no timeout. | |
4154 | * | |
4155 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4156 | * and interrupt capability. Also see complete(). | |
4157 | */ | |
b15136e9 | 4158 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4159 | { |
4160 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4161 | } |
8cbbe86d | 4162 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4163 | |
65eb3dc6 KD |
4164 | /** |
4165 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4166 | * @x: holds the state of this particular completion | |
4167 | * @timeout: timeout value in jiffies | |
4168 | * | |
4169 | * This waits for either a completion of a specific task to be signaled or for a | |
4170 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4171 | * interruptible. | |
4172 | */ | |
b15136e9 | 4173 | unsigned long __sched |
8cbbe86d | 4174 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4175 | { |
8cbbe86d | 4176 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4177 | } |
8cbbe86d | 4178 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4179 | |
65eb3dc6 KD |
4180 | /** |
4181 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4182 | * @x: holds the state of this particular completion | |
4183 | * | |
4184 | * This waits for completion of a specific task to be signaled. It is | |
4185 | * interruptible. | |
4186 | */ | |
8cbbe86d | 4187 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4188 | { |
51e97990 AK |
4189 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4190 | if (t == -ERESTARTSYS) | |
4191 | return t; | |
4192 | return 0; | |
0fec171c | 4193 | } |
8cbbe86d | 4194 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4195 | |
65eb3dc6 KD |
4196 | /** |
4197 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4198 | * @x: holds the state of this particular completion | |
4199 | * @timeout: timeout value in jiffies | |
4200 | * | |
4201 | * This waits for either a completion of a specific task to be signaled or for a | |
4202 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4203 | */ | |
b15136e9 | 4204 | unsigned long __sched |
8cbbe86d AK |
4205 | wait_for_completion_interruptible_timeout(struct completion *x, |
4206 | unsigned long timeout) | |
0fec171c | 4207 | { |
8cbbe86d | 4208 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4209 | } |
8cbbe86d | 4210 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4211 | |
65eb3dc6 KD |
4212 | /** |
4213 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4214 | * @x: holds the state of this particular completion | |
4215 | * | |
4216 | * This waits to be signaled for completion of a specific task. It can be | |
4217 | * interrupted by a kill signal. | |
4218 | */ | |
009e577e MW |
4219 | int __sched wait_for_completion_killable(struct completion *x) |
4220 | { | |
4221 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4222 | if (t == -ERESTARTSYS) | |
4223 | return t; | |
4224 | return 0; | |
4225 | } | |
4226 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4227 | ||
0aa12fb4 SW |
4228 | /** |
4229 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4230 | * @x: holds the state of this particular completion | |
4231 | * @timeout: timeout value in jiffies | |
4232 | * | |
4233 | * This waits for either a completion of a specific task to be | |
4234 | * signaled or for a specified timeout to expire. It can be | |
4235 | * interrupted by a kill signal. The timeout is in jiffies. | |
4236 | */ | |
4237 | unsigned long __sched | |
4238 | wait_for_completion_killable_timeout(struct completion *x, | |
4239 | unsigned long timeout) | |
4240 | { | |
4241 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4242 | } | |
4243 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4244 | ||
be4de352 DC |
4245 | /** |
4246 | * try_wait_for_completion - try to decrement a completion without blocking | |
4247 | * @x: completion structure | |
4248 | * | |
4249 | * Returns: 0 if a decrement cannot be done without blocking | |
4250 | * 1 if a decrement succeeded. | |
4251 | * | |
4252 | * If a completion is being used as a counting completion, | |
4253 | * attempt to decrement the counter without blocking. This | |
4254 | * enables us to avoid waiting if the resource the completion | |
4255 | * is protecting is not available. | |
4256 | */ | |
4257 | bool try_wait_for_completion(struct completion *x) | |
4258 | { | |
7539a3b3 | 4259 | unsigned long flags; |
be4de352 DC |
4260 | int ret = 1; |
4261 | ||
7539a3b3 | 4262 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4263 | if (!x->done) |
4264 | ret = 0; | |
4265 | else | |
4266 | x->done--; | |
7539a3b3 | 4267 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4268 | return ret; |
4269 | } | |
4270 | EXPORT_SYMBOL(try_wait_for_completion); | |
4271 | ||
4272 | /** | |
4273 | * completion_done - Test to see if a completion has any waiters | |
4274 | * @x: completion structure | |
4275 | * | |
4276 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4277 | * 1 if there are no waiters. | |
4278 | * | |
4279 | */ | |
4280 | bool completion_done(struct completion *x) | |
4281 | { | |
7539a3b3 | 4282 | unsigned long flags; |
be4de352 DC |
4283 | int ret = 1; |
4284 | ||
7539a3b3 | 4285 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4286 | if (!x->done) |
4287 | ret = 0; | |
7539a3b3 | 4288 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4289 | return ret; |
4290 | } | |
4291 | EXPORT_SYMBOL(completion_done); | |
4292 | ||
8cbbe86d AK |
4293 | static long __sched |
4294 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4295 | { |
0fec171c IM |
4296 | unsigned long flags; |
4297 | wait_queue_t wait; | |
4298 | ||
4299 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4300 | |
8cbbe86d | 4301 | __set_current_state(state); |
1da177e4 | 4302 | |
8cbbe86d AK |
4303 | spin_lock_irqsave(&q->lock, flags); |
4304 | __add_wait_queue(q, &wait); | |
4305 | spin_unlock(&q->lock); | |
4306 | timeout = schedule_timeout(timeout); | |
4307 | spin_lock_irq(&q->lock); | |
4308 | __remove_wait_queue(q, &wait); | |
4309 | spin_unlock_irqrestore(&q->lock, flags); | |
4310 | ||
4311 | return timeout; | |
4312 | } | |
4313 | ||
4314 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4315 | { | |
4316 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4317 | } |
1da177e4 LT |
4318 | EXPORT_SYMBOL(interruptible_sleep_on); |
4319 | ||
0fec171c | 4320 | long __sched |
95cdf3b7 | 4321 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4322 | { |
8cbbe86d | 4323 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4324 | } |
1da177e4 LT |
4325 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4326 | ||
0fec171c | 4327 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4328 | { |
8cbbe86d | 4329 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4330 | } |
1da177e4 LT |
4331 | EXPORT_SYMBOL(sleep_on); |
4332 | ||
0fec171c | 4333 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4334 | { |
8cbbe86d | 4335 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4336 | } |
1da177e4 LT |
4337 | EXPORT_SYMBOL(sleep_on_timeout); |
4338 | ||
b29739f9 IM |
4339 | #ifdef CONFIG_RT_MUTEXES |
4340 | ||
4341 | /* | |
4342 | * rt_mutex_setprio - set the current priority of a task | |
4343 | * @p: task | |
4344 | * @prio: prio value (kernel-internal form) | |
4345 | * | |
4346 | * This function changes the 'effective' priority of a task. It does | |
4347 | * not touch ->normal_prio like __setscheduler(). | |
4348 | * | |
4349 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4350 | */ | |
36c8b586 | 4351 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4352 | { |
4353 | unsigned long flags; | |
83b699ed | 4354 | int oldprio, on_rq, running; |
70b97a7f | 4355 | struct rq *rq; |
83ab0aa0 | 4356 | const struct sched_class *prev_class; |
b29739f9 IM |
4357 | |
4358 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4359 | ||
4360 | rq = task_rq_lock(p, &flags); | |
4361 | ||
a8027073 | 4362 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4363 | oldprio = p->prio; |
83ab0aa0 | 4364 | prev_class = p->sched_class; |
dd41f596 | 4365 | on_rq = p->se.on_rq; |
051a1d1a | 4366 | running = task_current(rq, p); |
0e1f3483 | 4367 | if (on_rq) |
69be72c1 | 4368 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4369 | if (running) |
4370 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4371 | |
4372 | if (rt_prio(prio)) | |
4373 | p->sched_class = &rt_sched_class; | |
4374 | else | |
4375 | p->sched_class = &fair_sched_class; | |
4376 | ||
b29739f9 IM |
4377 | p->prio = prio; |
4378 | ||
0e1f3483 HS |
4379 | if (running) |
4380 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4381 | if (on_rq) { |
371fd7e7 | 4382 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4383 | |
4384 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4385 | } |
4386 | task_rq_unlock(rq, &flags); | |
4387 | } | |
4388 | ||
4389 | #endif | |
4390 | ||
36c8b586 | 4391 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4392 | { |
dd41f596 | 4393 | int old_prio, delta, on_rq; |
1da177e4 | 4394 | unsigned long flags; |
70b97a7f | 4395 | struct rq *rq; |
1da177e4 LT |
4396 | |
4397 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4398 | return; | |
4399 | /* | |
4400 | * We have to be careful, if called from sys_setpriority(), | |
4401 | * the task might be in the middle of scheduling on another CPU. | |
4402 | */ | |
4403 | rq = task_rq_lock(p, &flags); | |
4404 | /* | |
4405 | * The RT priorities are set via sched_setscheduler(), but we still | |
4406 | * allow the 'normal' nice value to be set - but as expected | |
4407 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4408 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4409 | */ |
e05606d3 | 4410 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4411 | p->static_prio = NICE_TO_PRIO(nice); |
4412 | goto out_unlock; | |
4413 | } | |
dd41f596 | 4414 | on_rq = p->se.on_rq; |
c09595f6 | 4415 | if (on_rq) |
69be72c1 | 4416 | dequeue_task(rq, p, 0); |
1da177e4 | 4417 | |
1da177e4 | 4418 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4419 | set_load_weight(p); |
b29739f9 IM |
4420 | old_prio = p->prio; |
4421 | p->prio = effective_prio(p); | |
4422 | delta = p->prio - old_prio; | |
1da177e4 | 4423 | |
dd41f596 | 4424 | if (on_rq) { |
371fd7e7 | 4425 | enqueue_task(rq, p, 0); |
1da177e4 | 4426 | /* |
d5f9f942 AM |
4427 | * If the task increased its priority or is running and |
4428 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4429 | */ |
d5f9f942 | 4430 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4431 | resched_task(rq->curr); |
4432 | } | |
4433 | out_unlock: | |
4434 | task_rq_unlock(rq, &flags); | |
4435 | } | |
1da177e4 LT |
4436 | EXPORT_SYMBOL(set_user_nice); |
4437 | ||
e43379f1 MM |
4438 | /* |
4439 | * can_nice - check if a task can reduce its nice value | |
4440 | * @p: task | |
4441 | * @nice: nice value | |
4442 | */ | |
36c8b586 | 4443 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4444 | { |
024f4747 MM |
4445 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4446 | int nice_rlim = 20 - nice; | |
48f24c4d | 4447 | |
78d7d407 | 4448 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4449 | capable(CAP_SYS_NICE)); |
4450 | } | |
4451 | ||
1da177e4 LT |
4452 | #ifdef __ARCH_WANT_SYS_NICE |
4453 | ||
4454 | /* | |
4455 | * sys_nice - change the priority of the current process. | |
4456 | * @increment: priority increment | |
4457 | * | |
4458 | * sys_setpriority is a more generic, but much slower function that | |
4459 | * does similar things. | |
4460 | */ | |
5add95d4 | 4461 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4462 | { |
48f24c4d | 4463 | long nice, retval; |
1da177e4 LT |
4464 | |
4465 | /* | |
4466 | * Setpriority might change our priority at the same moment. | |
4467 | * We don't have to worry. Conceptually one call occurs first | |
4468 | * and we have a single winner. | |
4469 | */ | |
e43379f1 MM |
4470 | if (increment < -40) |
4471 | increment = -40; | |
1da177e4 LT |
4472 | if (increment > 40) |
4473 | increment = 40; | |
4474 | ||
2b8f836f | 4475 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4476 | if (nice < -20) |
4477 | nice = -20; | |
4478 | if (nice > 19) | |
4479 | nice = 19; | |
4480 | ||
e43379f1 MM |
4481 | if (increment < 0 && !can_nice(current, nice)) |
4482 | return -EPERM; | |
4483 | ||
1da177e4 LT |
4484 | retval = security_task_setnice(current, nice); |
4485 | if (retval) | |
4486 | return retval; | |
4487 | ||
4488 | set_user_nice(current, nice); | |
4489 | return 0; | |
4490 | } | |
4491 | ||
4492 | #endif | |
4493 | ||
4494 | /** | |
4495 | * task_prio - return the priority value of a given task. | |
4496 | * @p: the task in question. | |
4497 | * | |
4498 | * This is the priority value as seen by users in /proc. | |
4499 | * RT tasks are offset by -200. Normal tasks are centered | |
4500 | * around 0, value goes from -16 to +15. | |
4501 | */ | |
36c8b586 | 4502 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4503 | { |
4504 | return p->prio - MAX_RT_PRIO; | |
4505 | } | |
4506 | ||
4507 | /** | |
4508 | * task_nice - return the nice value of a given task. | |
4509 | * @p: the task in question. | |
4510 | */ | |
36c8b586 | 4511 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4512 | { |
4513 | return TASK_NICE(p); | |
4514 | } | |
150d8bed | 4515 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4516 | |
4517 | /** | |
4518 | * idle_cpu - is a given cpu idle currently? | |
4519 | * @cpu: the processor in question. | |
4520 | */ | |
4521 | int idle_cpu(int cpu) | |
4522 | { | |
4523 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4524 | } | |
4525 | ||
1da177e4 LT |
4526 | /** |
4527 | * idle_task - return the idle task for a given cpu. | |
4528 | * @cpu: the processor in question. | |
4529 | */ | |
36c8b586 | 4530 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4531 | { |
4532 | return cpu_rq(cpu)->idle; | |
4533 | } | |
4534 | ||
4535 | /** | |
4536 | * find_process_by_pid - find a process with a matching PID value. | |
4537 | * @pid: the pid in question. | |
4538 | */ | |
a9957449 | 4539 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4540 | { |
228ebcbe | 4541 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4542 | } |
4543 | ||
4544 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4545 | static void |
4546 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4547 | { |
dd41f596 | 4548 | BUG_ON(p->se.on_rq); |
48f24c4d | 4549 | |
1da177e4 LT |
4550 | p->policy = policy; |
4551 | p->rt_priority = prio; | |
b29739f9 IM |
4552 | p->normal_prio = normal_prio(p); |
4553 | /* we are holding p->pi_lock already */ | |
4554 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4555 | if (rt_prio(p->prio)) |
4556 | p->sched_class = &rt_sched_class; | |
4557 | else | |
4558 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4559 | set_load_weight(p); |
1da177e4 LT |
4560 | } |
4561 | ||
c69e8d9c DH |
4562 | /* |
4563 | * check the target process has a UID that matches the current process's | |
4564 | */ | |
4565 | static bool check_same_owner(struct task_struct *p) | |
4566 | { | |
4567 | const struct cred *cred = current_cred(), *pcred; | |
4568 | bool match; | |
4569 | ||
4570 | rcu_read_lock(); | |
4571 | pcred = __task_cred(p); | |
4572 | match = (cred->euid == pcred->euid || | |
4573 | cred->euid == pcred->uid); | |
4574 | rcu_read_unlock(); | |
4575 | return match; | |
4576 | } | |
4577 | ||
961ccddd | 4578 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4579 | const struct sched_param *param, bool user) |
1da177e4 | 4580 | { |
83b699ed | 4581 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4582 | unsigned long flags; |
83ab0aa0 | 4583 | const struct sched_class *prev_class; |
70b97a7f | 4584 | struct rq *rq; |
ca94c442 | 4585 | int reset_on_fork; |
1da177e4 | 4586 | |
66e5393a SR |
4587 | /* may grab non-irq protected spin_locks */ |
4588 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4589 | recheck: |
4590 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4591 | if (policy < 0) { |
4592 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4593 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4594 | } else { |
4595 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4596 | policy &= ~SCHED_RESET_ON_FORK; | |
4597 | ||
4598 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4599 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4600 | policy != SCHED_IDLE) | |
4601 | return -EINVAL; | |
4602 | } | |
4603 | ||
1da177e4 LT |
4604 | /* |
4605 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4606 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4607 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4608 | */ |
4609 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4610 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4611 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4612 | return -EINVAL; |
e05606d3 | 4613 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4614 | return -EINVAL; |
4615 | ||
37e4ab3f OC |
4616 | /* |
4617 | * Allow unprivileged RT tasks to decrease priority: | |
4618 | */ | |
961ccddd | 4619 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4620 | if (rt_policy(policy)) { |
a44702e8 ON |
4621 | unsigned long rlim_rtprio = |
4622 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4623 | |
4624 | /* can't set/change the rt policy */ | |
4625 | if (policy != p->policy && !rlim_rtprio) | |
4626 | return -EPERM; | |
4627 | ||
4628 | /* can't increase priority */ | |
4629 | if (param->sched_priority > p->rt_priority && | |
4630 | param->sched_priority > rlim_rtprio) | |
4631 | return -EPERM; | |
4632 | } | |
dd41f596 IM |
4633 | /* |
4634 | * Like positive nice levels, dont allow tasks to | |
4635 | * move out of SCHED_IDLE either: | |
4636 | */ | |
4637 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4638 | return -EPERM; | |
5fe1d75f | 4639 | |
37e4ab3f | 4640 | /* can't change other user's priorities */ |
c69e8d9c | 4641 | if (!check_same_owner(p)) |
37e4ab3f | 4642 | return -EPERM; |
ca94c442 LP |
4643 | |
4644 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4645 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4646 | return -EPERM; | |
37e4ab3f | 4647 | } |
1da177e4 | 4648 | |
725aad24 | 4649 | if (user) { |
b0ae1981 | 4650 | retval = security_task_setscheduler(p); |
725aad24 JF |
4651 | if (retval) |
4652 | return retval; | |
4653 | } | |
4654 | ||
b29739f9 IM |
4655 | /* |
4656 | * make sure no PI-waiters arrive (or leave) while we are | |
4657 | * changing the priority of the task: | |
4658 | */ | |
1d615482 | 4659 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4660 | /* |
4661 | * To be able to change p->policy safely, the apropriate | |
4662 | * runqueue lock must be held. | |
4663 | */ | |
b29739f9 | 4664 | rq = __task_rq_lock(p); |
dc61b1d6 | 4665 | |
34f971f6 PZ |
4666 | /* |
4667 | * Changing the policy of the stop threads its a very bad idea | |
4668 | */ | |
4669 | if (p == rq->stop) { | |
4670 | __task_rq_unlock(rq); | |
4671 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4672 | return -EINVAL; | |
4673 | } | |
4674 | ||
dc61b1d6 PZ |
4675 | #ifdef CONFIG_RT_GROUP_SCHED |
4676 | if (user) { | |
4677 | /* | |
4678 | * Do not allow realtime tasks into groups that have no runtime | |
4679 | * assigned. | |
4680 | */ | |
4681 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
4682 | task_group(p)->rt_bandwidth.rt_runtime == 0) { | |
4683 | __task_rq_unlock(rq); | |
4684 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4685 | return -EPERM; | |
4686 | } | |
4687 | } | |
4688 | #endif | |
4689 | ||
1da177e4 LT |
4690 | /* recheck policy now with rq lock held */ |
4691 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4692 | policy = oldpolicy = -1; | |
b29739f9 | 4693 | __task_rq_unlock(rq); |
1d615482 | 4694 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4695 | goto recheck; |
4696 | } | |
dd41f596 | 4697 | on_rq = p->se.on_rq; |
051a1d1a | 4698 | running = task_current(rq, p); |
0e1f3483 | 4699 | if (on_rq) |
2e1cb74a | 4700 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4701 | if (running) |
4702 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4703 | |
ca94c442 LP |
4704 | p->sched_reset_on_fork = reset_on_fork; |
4705 | ||
1da177e4 | 4706 | oldprio = p->prio; |
83ab0aa0 | 4707 | prev_class = p->sched_class; |
dd41f596 | 4708 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4709 | |
0e1f3483 HS |
4710 | if (running) |
4711 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4712 | if (on_rq) { |
4713 | activate_task(rq, p, 0); | |
cb469845 SR |
4714 | |
4715 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4716 | } |
b29739f9 | 4717 | __task_rq_unlock(rq); |
1d615482 | 4718 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4719 | |
95e02ca9 TG |
4720 | rt_mutex_adjust_pi(p); |
4721 | ||
1da177e4 LT |
4722 | return 0; |
4723 | } | |
961ccddd RR |
4724 | |
4725 | /** | |
4726 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4727 | * @p: the task in question. | |
4728 | * @policy: new policy. | |
4729 | * @param: structure containing the new RT priority. | |
4730 | * | |
4731 | * NOTE that the task may be already dead. | |
4732 | */ | |
4733 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4734 | const struct sched_param *param) |
961ccddd RR |
4735 | { |
4736 | return __sched_setscheduler(p, policy, param, true); | |
4737 | } | |
1da177e4 LT |
4738 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4739 | ||
961ccddd RR |
4740 | /** |
4741 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4742 | * @p: the task in question. | |
4743 | * @policy: new policy. | |
4744 | * @param: structure containing the new RT priority. | |
4745 | * | |
4746 | * Just like sched_setscheduler, only don't bother checking if the | |
4747 | * current context has permission. For example, this is needed in | |
4748 | * stop_machine(): we create temporary high priority worker threads, | |
4749 | * but our caller might not have that capability. | |
4750 | */ | |
4751 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4752 | const struct sched_param *param) |
961ccddd RR |
4753 | { |
4754 | return __sched_setscheduler(p, policy, param, false); | |
4755 | } | |
4756 | ||
95cdf3b7 IM |
4757 | static int |
4758 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4759 | { |
1da177e4 LT |
4760 | struct sched_param lparam; |
4761 | struct task_struct *p; | |
36c8b586 | 4762 | int retval; |
1da177e4 LT |
4763 | |
4764 | if (!param || pid < 0) | |
4765 | return -EINVAL; | |
4766 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4767 | return -EFAULT; | |
5fe1d75f ON |
4768 | |
4769 | rcu_read_lock(); | |
4770 | retval = -ESRCH; | |
1da177e4 | 4771 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4772 | if (p != NULL) |
4773 | retval = sched_setscheduler(p, policy, &lparam); | |
4774 | rcu_read_unlock(); | |
36c8b586 | 4775 | |
1da177e4 LT |
4776 | return retval; |
4777 | } | |
4778 | ||
4779 | /** | |
4780 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4781 | * @pid: the pid in question. | |
4782 | * @policy: new policy. | |
4783 | * @param: structure containing the new RT priority. | |
4784 | */ | |
5add95d4 HC |
4785 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4786 | struct sched_param __user *, param) | |
1da177e4 | 4787 | { |
c21761f1 JB |
4788 | /* negative values for policy are not valid */ |
4789 | if (policy < 0) | |
4790 | return -EINVAL; | |
4791 | ||
1da177e4 LT |
4792 | return do_sched_setscheduler(pid, policy, param); |
4793 | } | |
4794 | ||
4795 | /** | |
4796 | * sys_sched_setparam - set/change the RT priority of a thread | |
4797 | * @pid: the pid in question. | |
4798 | * @param: structure containing the new RT priority. | |
4799 | */ | |
5add95d4 | 4800 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4801 | { |
4802 | return do_sched_setscheduler(pid, -1, param); | |
4803 | } | |
4804 | ||
4805 | /** | |
4806 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4807 | * @pid: the pid in question. | |
4808 | */ | |
5add95d4 | 4809 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4810 | { |
36c8b586 | 4811 | struct task_struct *p; |
3a5c359a | 4812 | int retval; |
1da177e4 LT |
4813 | |
4814 | if (pid < 0) | |
3a5c359a | 4815 | return -EINVAL; |
1da177e4 LT |
4816 | |
4817 | retval = -ESRCH; | |
5fe85be0 | 4818 | rcu_read_lock(); |
1da177e4 LT |
4819 | p = find_process_by_pid(pid); |
4820 | if (p) { | |
4821 | retval = security_task_getscheduler(p); | |
4822 | if (!retval) | |
ca94c442 LP |
4823 | retval = p->policy |
4824 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4825 | } |
5fe85be0 | 4826 | rcu_read_unlock(); |
1da177e4 LT |
4827 | return retval; |
4828 | } | |
4829 | ||
4830 | /** | |
ca94c442 | 4831 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
4832 | * @pid: the pid in question. |
4833 | * @param: structure containing the RT priority. | |
4834 | */ | |
5add95d4 | 4835 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4836 | { |
4837 | struct sched_param lp; | |
36c8b586 | 4838 | struct task_struct *p; |
3a5c359a | 4839 | int retval; |
1da177e4 LT |
4840 | |
4841 | if (!param || pid < 0) | |
3a5c359a | 4842 | return -EINVAL; |
1da177e4 | 4843 | |
5fe85be0 | 4844 | rcu_read_lock(); |
1da177e4 LT |
4845 | p = find_process_by_pid(pid); |
4846 | retval = -ESRCH; | |
4847 | if (!p) | |
4848 | goto out_unlock; | |
4849 | ||
4850 | retval = security_task_getscheduler(p); | |
4851 | if (retval) | |
4852 | goto out_unlock; | |
4853 | ||
4854 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 4855 | rcu_read_unlock(); |
1da177e4 LT |
4856 | |
4857 | /* | |
4858 | * This one might sleep, we cannot do it with a spinlock held ... | |
4859 | */ | |
4860 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
4861 | ||
1da177e4 LT |
4862 | return retval; |
4863 | ||
4864 | out_unlock: | |
5fe85be0 | 4865 | rcu_read_unlock(); |
1da177e4 LT |
4866 | return retval; |
4867 | } | |
4868 | ||
96f874e2 | 4869 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 4870 | { |
5a16f3d3 | 4871 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
4872 | struct task_struct *p; |
4873 | int retval; | |
1da177e4 | 4874 | |
95402b38 | 4875 | get_online_cpus(); |
23f5d142 | 4876 | rcu_read_lock(); |
1da177e4 LT |
4877 | |
4878 | p = find_process_by_pid(pid); | |
4879 | if (!p) { | |
23f5d142 | 4880 | rcu_read_unlock(); |
95402b38 | 4881 | put_online_cpus(); |
1da177e4 LT |
4882 | return -ESRCH; |
4883 | } | |
4884 | ||
23f5d142 | 4885 | /* Prevent p going away */ |
1da177e4 | 4886 | get_task_struct(p); |
23f5d142 | 4887 | rcu_read_unlock(); |
1da177e4 | 4888 | |
5a16f3d3 RR |
4889 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
4890 | retval = -ENOMEM; | |
4891 | goto out_put_task; | |
4892 | } | |
4893 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
4894 | retval = -ENOMEM; | |
4895 | goto out_free_cpus_allowed; | |
4896 | } | |
1da177e4 | 4897 | retval = -EPERM; |
c69e8d9c | 4898 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
4899 | goto out_unlock; |
4900 | ||
b0ae1981 | 4901 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
4902 | if (retval) |
4903 | goto out_unlock; | |
4904 | ||
5a16f3d3 RR |
4905 | cpuset_cpus_allowed(p, cpus_allowed); |
4906 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 4907 | again: |
5a16f3d3 | 4908 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 4909 | |
8707d8b8 | 4910 | if (!retval) { |
5a16f3d3 RR |
4911 | cpuset_cpus_allowed(p, cpus_allowed); |
4912 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
4913 | /* |
4914 | * We must have raced with a concurrent cpuset | |
4915 | * update. Just reset the cpus_allowed to the | |
4916 | * cpuset's cpus_allowed | |
4917 | */ | |
5a16f3d3 | 4918 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
4919 | goto again; |
4920 | } | |
4921 | } | |
1da177e4 | 4922 | out_unlock: |
5a16f3d3 RR |
4923 | free_cpumask_var(new_mask); |
4924 | out_free_cpus_allowed: | |
4925 | free_cpumask_var(cpus_allowed); | |
4926 | out_put_task: | |
1da177e4 | 4927 | put_task_struct(p); |
95402b38 | 4928 | put_online_cpus(); |
1da177e4 LT |
4929 | return retval; |
4930 | } | |
4931 | ||
4932 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 4933 | struct cpumask *new_mask) |
1da177e4 | 4934 | { |
96f874e2 RR |
4935 | if (len < cpumask_size()) |
4936 | cpumask_clear(new_mask); | |
4937 | else if (len > cpumask_size()) | |
4938 | len = cpumask_size(); | |
4939 | ||
1da177e4 LT |
4940 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
4941 | } | |
4942 | ||
4943 | /** | |
4944 | * sys_sched_setaffinity - set the cpu affinity of a process | |
4945 | * @pid: pid of the process | |
4946 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4947 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
4948 | */ | |
5add95d4 HC |
4949 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
4950 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 4951 | { |
5a16f3d3 | 4952 | cpumask_var_t new_mask; |
1da177e4 LT |
4953 | int retval; |
4954 | ||
5a16f3d3 RR |
4955 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
4956 | return -ENOMEM; | |
1da177e4 | 4957 | |
5a16f3d3 RR |
4958 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
4959 | if (retval == 0) | |
4960 | retval = sched_setaffinity(pid, new_mask); | |
4961 | free_cpumask_var(new_mask); | |
4962 | return retval; | |
1da177e4 LT |
4963 | } |
4964 | ||
96f874e2 | 4965 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 4966 | { |
36c8b586 | 4967 | struct task_struct *p; |
31605683 TG |
4968 | unsigned long flags; |
4969 | struct rq *rq; | |
1da177e4 | 4970 | int retval; |
1da177e4 | 4971 | |
95402b38 | 4972 | get_online_cpus(); |
23f5d142 | 4973 | rcu_read_lock(); |
1da177e4 LT |
4974 | |
4975 | retval = -ESRCH; | |
4976 | p = find_process_by_pid(pid); | |
4977 | if (!p) | |
4978 | goto out_unlock; | |
4979 | ||
e7834f8f DQ |
4980 | retval = security_task_getscheduler(p); |
4981 | if (retval) | |
4982 | goto out_unlock; | |
4983 | ||
31605683 | 4984 | rq = task_rq_lock(p, &flags); |
96f874e2 | 4985 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 4986 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
4987 | |
4988 | out_unlock: | |
23f5d142 | 4989 | rcu_read_unlock(); |
95402b38 | 4990 | put_online_cpus(); |
1da177e4 | 4991 | |
9531b62f | 4992 | return retval; |
1da177e4 LT |
4993 | } |
4994 | ||
4995 | /** | |
4996 | * sys_sched_getaffinity - get the cpu affinity of a process | |
4997 | * @pid: pid of the process | |
4998 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
4999 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5000 | */ | |
5add95d4 HC |
5001 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5002 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5003 | { |
5004 | int ret; | |
f17c8607 | 5005 | cpumask_var_t mask; |
1da177e4 | 5006 | |
84fba5ec | 5007 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5008 | return -EINVAL; |
5009 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5010 | return -EINVAL; |
5011 | ||
f17c8607 RR |
5012 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5013 | return -ENOMEM; | |
1da177e4 | 5014 | |
f17c8607 RR |
5015 | ret = sched_getaffinity(pid, mask); |
5016 | if (ret == 0) { | |
8bc037fb | 5017 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5018 | |
5019 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5020 | ret = -EFAULT; |
5021 | else | |
cd3d8031 | 5022 | ret = retlen; |
f17c8607 RR |
5023 | } |
5024 | free_cpumask_var(mask); | |
1da177e4 | 5025 | |
f17c8607 | 5026 | return ret; |
1da177e4 LT |
5027 | } |
5028 | ||
5029 | /** | |
5030 | * sys_sched_yield - yield the current processor to other threads. | |
5031 | * | |
dd41f596 IM |
5032 | * This function yields the current CPU to other tasks. If there are no |
5033 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5034 | */ |
5add95d4 | 5035 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5036 | { |
70b97a7f | 5037 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5038 | |
2d72376b | 5039 | schedstat_inc(rq, yld_count); |
4530d7ab | 5040 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5041 | |
5042 | /* | |
5043 | * Since we are going to call schedule() anyway, there's | |
5044 | * no need to preempt or enable interrupts: | |
5045 | */ | |
5046 | __release(rq->lock); | |
8a25d5de | 5047 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5048 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5049 | preempt_enable_no_resched(); |
5050 | ||
5051 | schedule(); | |
5052 | ||
5053 | return 0; | |
5054 | } | |
5055 | ||
d86ee480 PZ |
5056 | static inline int should_resched(void) |
5057 | { | |
5058 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5059 | } | |
5060 | ||
e7b38404 | 5061 | static void __cond_resched(void) |
1da177e4 | 5062 | { |
e7aaaa69 FW |
5063 | add_preempt_count(PREEMPT_ACTIVE); |
5064 | schedule(); | |
5065 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5066 | } |
5067 | ||
02b67cc3 | 5068 | int __sched _cond_resched(void) |
1da177e4 | 5069 | { |
d86ee480 | 5070 | if (should_resched()) { |
1da177e4 LT |
5071 | __cond_resched(); |
5072 | return 1; | |
5073 | } | |
5074 | return 0; | |
5075 | } | |
02b67cc3 | 5076 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5077 | |
5078 | /* | |
613afbf8 | 5079 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5080 | * call schedule, and on return reacquire the lock. |
5081 | * | |
41a2d6cf | 5082 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5083 | * operations here to prevent schedule() from being called twice (once via |
5084 | * spin_unlock(), once by hand). | |
5085 | */ | |
613afbf8 | 5086 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5087 | { |
d86ee480 | 5088 | int resched = should_resched(); |
6df3cecb JK |
5089 | int ret = 0; |
5090 | ||
f607c668 PZ |
5091 | lockdep_assert_held(lock); |
5092 | ||
95c354fe | 5093 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5094 | spin_unlock(lock); |
d86ee480 | 5095 | if (resched) |
95c354fe NP |
5096 | __cond_resched(); |
5097 | else | |
5098 | cpu_relax(); | |
6df3cecb | 5099 | ret = 1; |
1da177e4 | 5100 | spin_lock(lock); |
1da177e4 | 5101 | } |
6df3cecb | 5102 | return ret; |
1da177e4 | 5103 | } |
613afbf8 | 5104 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5105 | |
613afbf8 | 5106 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5107 | { |
5108 | BUG_ON(!in_softirq()); | |
5109 | ||
d86ee480 | 5110 | if (should_resched()) { |
98d82567 | 5111 | local_bh_enable(); |
1da177e4 LT |
5112 | __cond_resched(); |
5113 | local_bh_disable(); | |
5114 | return 1; | |
5115 | } | |
5116 | return 0; | |
5117 | } | |
613afbf8 | 5118 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5119 | |
1da177e4 LT |
5120 | /** |
5121 | * yield - yield the current processor to other threads. | |
5122 | * | |
72fd4a35 | 5123 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5124 | * thread runnable and calls sys_sched_yield(). |
5125 | */ | |
5126 | void __sched yield(void) | |
5127 | { | |
5128 | set_current_state(TASK_RUNNING); | |
5129 | sys_sched_yield(); | |
5130 | } | |
1da177e4 LT |
5131 | EXPORT_SYMBOL(yield); |
5132 | ||
5133 | /* | |
41a2d6cf | 5134 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5135 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5136 | */ |
5137 | void __sched io_schedule(void) | |
5138 | { | |
54d35f29 | 5139 | struct rq *rq = raw_rq(); |
1da177e4 | 5140 | |
0ff92245 | 5141 | delayacct_blkio_start(); |
1da177e4 | 5142 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5143 | current->in_iowait = 1; |
1da177e4 | 5144 | schedule(); |
8f0dfc34 | 5145 | current->in_iowait = 0; |
1da177e4 | 5146 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5147 | delayacct_blkio_end(); |
1da177e4 | 5148 | } |
1da177e4 LT |
5149 | EXPORT_SYMBOL(io_schedule); |
5150 | ||
5151 | long __sched io_schedule_timeout(long timeout) | |
5152 | { | |
54d35f29 | 5153 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5154 | long ret; |
5155 | ||
0ff92245 | 5156 | delayacct_blkio_start(); |
1da177e4 | 5157 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5158 | current->in_iowait = 1; |
1da177e4 | 5159 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5160 | current->in_iowait = 0; |
1da177e4 | 5161 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5162 | delayacct_blkio_end(); |
1da177e4 LT |
5163 | return ret; |
5164 | } | |
5165 | ||
5166 | /** | |
5167 | * sys_sched_get_priority_max - return maximum RT priority. | |
5168 | * @policy: scheduling class. | |
5169 | * | |
5170 | * this syscall returns the maximum rt_priority that can be used | |
5171 | * by a given scheduling class. | |
5172 | */ | |
5add95d4 | 5173 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5174 | { |
5175 | int ret = -EINVAL; | |
5176 | ||
5177 | switch (policy) { | |
5178 | case SCHED_FIFO: | |
5179 | case SCHED_RR: | |
5180 | ret = MAX_USER_RT_PRIO-1; | |
5181 | break; | |
5182 | case SCHED_NORMAL: | |
b0a9499c | 5183 | case SCHED_BATCH: |
dd41f596 | 5184 | case SCHED_IDLE: |
1da177e4 LT |
5185 | ret = 0; |
5186 | break; | |
5187 | } | |
5188 | return ret; | |
5189 | } | |
5190 | ||
5191 | /** | |
5192 | * sys_sched_get_priority_min - return minimum RT priority. | |
5193 | * @policy: scheduling class. | |
5194 | * | |
5195 | * this syscall returns the minimum rt_priority that can be used | |
5196 | * by a given scheduling class. | |
5197 | */ | |
5add95d4 | 5198 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5199 | { |
5200 | int ret = -EINVAL; | |
5201 | ||
5202 | switch (policy) { | |
5203 | case SCHED_FIFO: | |
5204 | case SCHED_RR: | |
5205 | ret = 1; | |
5206 | break; | |
5207 | case SCHED_NORMAL: | |
b0a9499c | 5208 | case SCHED_BATCH: |
dd41f596 | 5209 | case SCHED_IDLE: |
1da177e4 LT |
5210 | ret = 0; |
5211 | } | |
5212 | return ret; | |
5213 | } | |
5214 | ||
5215 | /** | |
5216 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5217 | * @pid: pid of the process. | |
5218 | * @interval: userspace pointer to the timeslice value. | |
5219 | * | |
5220 | * this syscall writes the default timeslice value of a given process | |
5221 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5222 | */ | |
17da2bd9 | 5223 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5224 | struct timespec __user *, interval) |
1da177e4 | 5225 | { |
36c8b586 | 5226 | struct task_struct *p; |
a4ec24b4 | 5227 | unsigned int time_slice; |
dba091b9 TG |
5228 | unsigned long flags; |
5229 | struct rq *rq; | |
3a5c359a | 5230 | int retval; |
1da177e4 | 5231 | struct timespec t; |
1da177e4 LT |
5232 | |
5233 | if (pid < 0) | |
3a5c359a | 5234 | return -EINVAL; |
1da177e4 LT |
5235 | |
5236 | retval = -ESRCH; | |
1a551ae7 | 5237 | rcu_read_lock(); |
1da177e4 LT |
5238 | p = find_process_by_pid(pid); |
5239 | if (!p) | |
5240 | goto out_unlock; | |
5241 | ||
5242 | retval = security_task_getscheduler(p); | |
5243 | if (retval) | |
5244 | goto out_unlock; | |
5245 | ||
dba091b9 TG |
5246 | rq = task_rq_lock(p, &flags); |
5247 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5248 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5249 | |
1a551ae7 | 5250 | rcu_read_unlock(); |
a4ec24b4 | 5251 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5252 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5253 | return retval; |
3a5c359a | 5254 | |
1da177e4 | 5255 | out_unlock: |
1a551ae7 | 5256 | rcu_read_unlock(); |
1da177e4 LT |
5257 | return retval; |
5258 | } | |
5259 | ||
7c731e0a | 5260 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5261 | |
82a1fcb9 | 5262 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5263 | { |
1da177e4 | 5264 | unsigned long free = 0; |
36c8b586 | 5265 | unsigned state; |
1da177e4 | 5266 | |
1da177e4 | 5267 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5268 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5269 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5270 | #if BITS_PER_LONG == 32 |
1da177e4 | 5271 | if (state == TASK_RUNNING) |
3df0fc5b | 5272 | printk(KERN_CONT " running "); |
1da177e4 | 5273 | else |
3df0fc5b | 5274 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5275 | #else |
5276 | if (state == TASK_RUNNING) | |
3df0fc5b | 5277 | printk(KERN_CONT " running task "); |
1da177e4 | 5278 | else |
3df0fc5b | 5279 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5280 | #endif |
5281 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5282 | free = stack_not_used(p); |
1da177e4 | 5283 | #endif |
3df0fc5b | 5284 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5285 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5286 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5287 | |
5fb5e6de | 5288 | show_stack(p, NULL); |
1da177e4 LT |
5289 | } |
5290 | ||
e59e2ae2 | 5291 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5292 | { |
36c8b586 | 5293 | struct task_struct *g, *p; |
1da177e4 | 5294 | |
4bd77321 | 5295 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5296 | printk(KERN_INFO |
5297 | " task PC stack pid father\n"); | |
1da177e4 | 5298 | #else |
3df0fc5b PZ |
5299 | printk(KERN_INFO |
5300 | " task PC stack pid father\n"); | |
1da177e4 LT |
5301 | #endif |
5302 | read_lock(&tasklist_lock); | |
5303 | do_each_thread(g, p) { | |
5304 | /* | |
5305 | * reset the NMI-timeout, listing all files on a slow | |
5306 | * console might take alot of time: | |
5307 | */ | |
5308 | touch_nmi_watchdog(); | |
39bc89fd | 5309 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5310 | sched_show_task(p); |
1da177e4 LT |
5311 | } while_each_thread(g, p); |
5312 | ||
04c9167f JF |
5313 | touch_all_softlockup_watchdogs(); |
5314 | ||
dd41f596 IM |
5315 | #ifdef CONFIG_SCHED_DEBUG |
5316 | sysrq_sched_debug_show(); | |
5317 | #endif | |
1da177e4 | 5318 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5319 | /* |
5320 | * Only show locks if all tasks are dumped: | |
5321 | */ | |
93335a21 | 5322 | if (!state_filter) |
e59e2ae2 | 5323 | debug_show_all_locks(); |
1da177e4 LT |
5324 | } |
5325 | ||
1df21055 IM |
5326 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5327 | { | |
dd41f596 | 5328 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5329 | } |
5330 | ||
f340c0d1 IM |
5331 | /** |
5332 | * init_idle - set up an idle thread for a given CPU | |
5333 | * @idle: task in question | |
5334 | * @cpu: cpu the idle task belongs to | |
5335 | * | |
5336 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5337 | * flag, to make booting more robust. | |
5338 | */ | |
5c1e1767 | 5339 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5340 | { |
70b97a7f | 5341 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5342 | unsigned long flags; |
5343 | ||
05fa785c | 5344 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5345 | |
dd41f596 | 5346 | __sched_fork(idle); |
06b83b5f | 5347 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5348 | idle->se.exec_start = sched_clock(); |
5349 | ||
96f874e2 | 5350 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5351 | /* |
5352 | * We're having a chicken and egg problem, even though we are | |
5353 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5354 | * lockdep check in task_group() will fail. | |
5355 | * | |
5356 | * Similar case to sched_fork(). / Alternatively we could | |
5357 | * use task_rq_lock() here and obtain the other rq->lock. | |
5358 | * | |
5359 | * Silence PROVE_RCU | |
5360 | */ | |
5361 | rcu_read_lock(); | |
dd41f596 | 5362 | __set_task_cpu(idle, cpu); |
6506cf6c | 5363 | rcu_read_unlock(); |
1da177e4 | 5364 | |
1da177e4 | 5365 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5366 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5367 | idle->oncpu = 1; | |
5368 | #endif | |
05fa785c | 5369 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5370 | |
5371 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5372 | #if defined(CONFIG_PREEMPT) |
5373 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5374 | #else | |
a1261f54 | 5375 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5376 | #endif |
dd41f596 IM |
5377 | /* |
5378 | * The idle tasks have their own, simple scheduling class: | |
5379 | */ | |
5380 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5381 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5382 | } |
5383 | ||
5384 | /* | |
5385 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5386 | * indicates which cpus entered this state. This is used | |
5387 | * in the rcu update to wait only for active cpus. For system | |
5388 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5389 | * always be CPU_BITS_NONE. |
1da177e4 | 5390 | */ |
6a7b3dc3 | 5391 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5392 | |
19978ca6 IM |
5393 | /* |
5394 | * Increase the granularity value when there are more CPUs, | |
5395 | * because with more CPUs the 'effective latency' as visible | |
5396 | * to users decreases. But the relationship is not linear, | |
5397 | * so pick a second-best guess by going with the log2 of the | |
5398 | * number of CPUs. | |
5399 | * | |
5400 | * This idea comes from the SD scheduler of Con Kolivas: | |
5401 | */ | |
acb4a848 | 5402 | static int get_update_sysctl_factor(void) |
19978ca6 | 5403 | { |
4ca3ef71 | 5404 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5405 | unsigned int factor; |
5406 | ||
5407 | switch (sysctl_sched_tunable_scaling) { | |
5408 | case SCHED_TUNABLESCALING_NONE: | |
5409 | factor = 1; | |
5410 | break; | |
5411 | case SCHED_TUNABLESCALING_LINEAR: | |
5412 | factor = cpus; | |
5413 | break; | |
5414 | case SCHED_TUNABLESCALING_LOG: | |
5415 | default: | |
5416 | factor = 1 + ilog2(cpus); | |
5417 | break; | |
5418 | } | |
19978ca6 | 5419 | |
acb4a848 CE |
5420 | return factor; |
5421 | } | |
19978ca6 | 5422 | |
acb4a848 CE |
5423 | static void update_sysctl(void) |
5424 | { | |
5425 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5426 | |
0bcdcf28 CE |
5427 | #define SET_SYSCTL(name) \ |
5428 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5429 | SET_SYSCTL(sched_min_granularity); | |
5430 | SET_SYSCTL(sched_latency); | |
5431 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5432 | #undef SET_SYSCTL |
5433 | } | |
55cd5340 | 5434 | |
0bcdcf28 CE |
5435 | static inline void sched_init_granularity(void) |
5436 | { | |
5437 | update_sysctl(); | |
19978ca6 IM |
5438 | } |
5439 | ||
1da177e4 LT |
5440 | #ifdef CONFIG_SMP |
5441 | /* | |
5442 | * This is how migration works: | |
5443 | * | |
969c7921 TH |
5444 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5445 | * stop_one_cpu(). | |
5446 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5447 | * off the CPU) | |
5448 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5449 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5450 | * it and puts it into the right queue. |
969c7921 TH |
5451 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5452 | * is done. | |
1da177e4 LT |
5453 | */ |
5454 | ||
5455 | /* | |
5456 | * Change a given task's CPU affinity. Migrate the thread to a | |
5457 | * proper CPU and schedule it away if the CPU it's executing on | |
5458 | * is removed from the allowed bitmask. | |
5459 | * | |
5460 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5461 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5462 | * call is not atomic; no spinlocks may be held. |
5463 | */ | |
96f874e2 | 5464 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5465 | { |
5466 | unsigned long flags; | |
70b97a7f | 5467 | struct rq *rq; |
969c7921 | 5468 | unsigned int dest_cpu; |
48f24c4d | 5469 | int ret = 0; |
1da177e4 | 5470 | |
65cc8e48 PZ |
5471 | /* |
5472 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5473 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5474 | */ | |
5475 | again: | |
5476 | while (task_is_waking(p)) | |
5477 | cpu_relax(); | |
1da177e4 | 5478 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5479 | if (task_is_waking(p)) { |
5480 | task_rq_unlock(rq, &flags); | |
5481 | goto again; | |
5482 | } | |
e2912009 | 5483 | |
6ad4c188 | 5484 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5485 | ret = -EINVAL; |
5486 | goto out; | |
5487 | } | |
5488 | ||
9985b0ba | 5489 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5490 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5491 | ret = -EINVAL; |
5492 | goto out; | |
5493 | } | |
5494 | ||
73fe6aae | 5495 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5496 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5497 | else { |
96f874e2 RR |
5498 | cpumask_copy(&p->cpus_allowed, new_mask); |
5499 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5500 | } |
5501 | ||
1da177e4 | 5502 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5503 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5504 | goto out; |
5505 | ||
969c7921 TH |
5506 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
5507 | if (migrate_task(p, dest_cpu)) { | |
5508 | struct migration_arg arg = { p, dest_cpu }; | |
1da177e4 LT |
5509 | /* Need help from migration thread: drop lock and wait. */ |
5510 | task_rq_unlock(rq, &flags); | |
969c7921 | 5511 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5512 | tlb_migrate_finish(p->mm); |
5513 | return 0; | |
5514 | } | |
5515 | out: | |
5516 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5517 | |
1da177e4 LT |
5518 | return ret; |
5519 | } | |
cd8ba7cd | 5520 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5521 | |
5522 | /* | |
41a2d6cf | 5523 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5524 | * this because either it can't run here any more (set_cpus_allowed() |
5525 | * away from this CPU, or CPU going down), or because we're | |
5526 | * attempting to rebalance this task on exec (sched_exec). | |
5527 | * | |
5528 | * So we race with normal scheduler movements, but that's OK, as long | |
5529 | * as the task is no longer on this CPU. | |
efc30814 KK |
5530 | * |
5531 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5532 | */ |
efc30814 | 5533 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5534 | { |
70b97a7f | 5535 | struct rq *rq_dest, *rq_src; |
e2912009 | 5536 | int ret = 0; |
1da177e4 | 5537 | |
e761b772 | 5538 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5539 | return ret; |
1da177e4 LT |
5540 | |
5541 | rq_src = cpu_rq(src_cpu); | |
5542 | rq_dest = cpu_rq(dest_cpu); | |
5543 | ||
5544 | double_rq_lock(rq_src, rq_dest); | |
5545 | /* Already moved. */ | |
5546 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5547 | goto done; |
1da177e4 | 5548 | /* Affinity changed (again). */ |
96f874e2 | 5549 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5550 | goto fail; |
1da177e4 | 5551 | |
e2912009 PZ |
5552 | /* |
5553 | * If we're not on a rq, the next wake-up will ensure we're | |
5554 | * placed properly. | |
5555 | */ | |
5556 | if (p->se.on_rq) { | |
2e1cb74a | 5557 | deactivate_task(rq_src, p, 0); |
e2912009 | 5558 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5559 | activate_task(rq_dest, p, 0); |
15afe09b | 5560 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5561 | } |
b1e38734 | 5562 | done: |
efc30814 | 5563 | ret = 1; |
b1e38734 | 5564 | fail: |
1da177e4 | 5565 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5566 | return ret; |
1da177e4 LT |
5567 | } |
5568 | ||
5569 | /* | |
969c7921 TH |
5570 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5571 | * and performs thread migration by bumping thread off CPU then | |
5572 | * 'pushing' onto another runqueue. | |
1da177e4 | 5573 | */ |
969c7921 | 5574 | static int migration_cpu_stop(void *data) |
1da177e4 | 5575 | { |
969c7921 | 5576 | struct migration_arg *arg = data; |
f7b4cddc | 5577 | |
969c7921 TH |
5578 | /* |
5579 | * The original target cpu might have gone down and we might | |
5580 | * be on another cpu but it doesn't matter. | |
5581 | */ | |
f7b4cddc | 5582 | local_irq_disable(); |
969c7921 | 5583 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5584 | local_irq_enable(); |
1da177e4 | 5585 | return 0; |
f7b4cddc ON |
5586 | } |
5587 | ||
1da177e4 | 5588 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5589 | |
054b9108 | 5590 | /* |
48c5ccae PZ |
5591 | * Ensures that the idle task is using init_mm right before its cpu goes |
5592 | * offline. | |
054b9108 | 5593 | */ |
48c5ccae | 5594 | void idle_task_exit(void) |
1da177e4 | 5595 | { |
48c5ccae | 5596 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5597 | |
48c5ccae | 5598 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5599 | |
48c5ccae PZ |
5600 | if (mm != &init_mm) |
5601 | switch_mm(mm, &init_mm, current); | |
5602 | mmdrop(mm); | |
1da177e4 LT |
5603 | } |
5604 | ||
5605 | /* | |
5606 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5607 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5608 | * for performance reasons the counter is not stricly tracking tasks to | |
5609 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5610 | * to keep the global sum constant after CPU-down: | |
5611 | */ | |
70b97a7f | 5612 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5613 | { |
6ad4c188 | 5614 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 5615 | |
1da177e4 LT |
5616 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
5617 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
5618 | } |
5619 | ||
dd41f596 | 5620 | /* |
48c5ccae | 5621 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 5622 | */ |
48c5ccae | 5623 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 5624 | { |
48c5ccae PZ |
5625 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
5626 | rq->calc_load_active = 0; | |
1da177e4 LT |
5627 | } |
5628 | ||
48f24c4d | 5629 | /* |
48c5ccae PZ |
5630 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5631 | * try_to_wake_up()->select_task_rq(). | |
5632 | * | |
5633 | * Called with rq->lock held even though we'er in stop_machine() and | |
5634 | * there's no concurrency possible, we hold the required locks anyway | |
5635 | * because of lock validation efforts. | |
1da177e4 | 5636 | */ |
48c5ccae | 5637 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5638 | { |
70b97a7f | 5639 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5640 | struct task_struct *next, *stop = rq->stop; |
5641 | int dest_cpu; | |
1da177e4 LT |
5642 | |
5643 | /* | |
48c5ccae PZ |
5644 | * Fudge the rq selection such that the below task selection loop |
5645 | * doesn't get stuck on the currently eligible stop task. | |
5646 | * | |
5647 | * We're currently inside stop_machine() and the rq is either stuck | |
5648 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5649 | * either way we should never end up calling schedule() until we're | |
5650 | * done here. | |
1da177e4 | 5651 | */ |
48c5ccae | 5652 | rq->stop = NULL; |
48f24c4d | 5653 | |
dd41f596 | 5654 | for ( ; ; ) { |
48c5ccae PZ |
5655 | /* |
5656 | * There's this thread running, bail when that's the only | |
5657 | * remaining thread. | |
5658 | */ | |
5659 | if (rq->nr_running == 1) | |
dd41f596 | 5660 | break; |
48c5ccae | 5661 | |
b67802ea | 5662 | next = pick_next_task(rq); |
48c5ccae | 5663 | BUG_ON(!next); |
79c53799 | 5664 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5665 | |
48c5ccae PZ |
5666 | /* Find suitable destination for @next, with force if needed. */ |
5667 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5668 | raw_spin_unlock(&rq->lock); | |
5669 | ||
5670 | __migrate_task(next, dead_cpu, dest_cpu); | |
5671 | ||
5672 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5673 | } |
dce48a84 | 5674 | |
48c5ccae | 5675 | rq->stop = stop; |
dce48a84 | 5676 | } |
48c5ccae | 5677 | |
1da177e4 LT |
5678 | #endif /* CONFIG_HOTPLUG_CPU */ |
5679 | ||
e692ab53 NP |
5680 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5681 | ||
5682 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5683 | { |
5684 | .procname = "sched_domain", | |
c57baf1e | 5685 | .mode = 0555, |
e0361851 | 5686 | }, |
56992309 | 5687 | {} |
e692ab53 NP |
5688 | }; |
5689 | ||
5690 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5691 | { |
5692 | .procname = "kernel", | |
c57baf1e | 5693 | .mode = 0555, |
e0361851 AD |
5694 | .child = sd_ctl_dir, |
5695 | }, | |
56992309 | 5696 | {} |
e692ab53 NP |
5697 | }; |
5698 | ||
5699 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5700 | { | |
5701 | struct ctl_table *entry = | |
5cf9f062 | 5702 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5703 | |
e692ab53 NP |
5704 | return entry; |
5705 | } | |
5706 | ||
6382bc90 MM |
5707 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5708 | { | |
cd790076 | 5709 | struct ctl_table *entry; |
6382bc90 | 5710 | |
cd790076 MM |
5711 | /* |
5712 | * In the intermediate directories, both the child directory and | |
5713 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5714 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5715 | * static strings and all have proc handlers. |
5716 | */ | |
5717 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5718 | if (entry->child) |
5719 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5720 | if (entry->proc_handler == NULL) |
5721 | kfree(entry->procname); | |
5722 | } | |
6382bc90 MM |
5723 | |
5724 | kfree(*tablep); | |
5725 | *tablep = NULL; | |
5726 | } | |
5727 | ||
e692ab53 | 5728 | static void |
e0361851 | 5729 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5730 | const char *procname, void *data, int maxlen, |
5731 | mode_t mode, proc_handler *proc_handler) | |
5732 | { | |
e692ab53 NP |
5733 | entry->procname = procname; |
5734 | entry->data = data; | |
5735 | entry->maxlen = maxlen; | |
5736 | entry->mode = mode; | |
5737 | entry->proc_handler = proc_handler; | |
5738 | } | |
5739 | ||
5740 | static struct ctl_table * | |
5741 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5742 | { | |
a5d8c348 | 5743 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5744 | |
ad1cdc1d MM |
5745 | if (table == NULL) |
5746 | return NULL; | |
5747 | ||
e0361851 | 5748 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5749 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5750 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5751 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5752 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5753 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5754 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5755 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5756 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5757 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5758 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5759 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5760 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5761 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5762 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5763 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5764 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5765 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5766 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5767 | &sd->cache_nice_tries, |
5768 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5769 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5770 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5771 | set_table_entry(&table[11], "name", sd->name, |
5772 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5773 | /* &table[12] is terminator */ | |
e692ab53 NP |
5774 | |
5775 | return table; | |
5776 | } | |
5777 | ||
9a4e7159 | 5778 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5779 | { |
5780 | struct ctl_table *entry, *table; | |
5781 | struct sched_domain *sd; | |
5782 | int domain_num = 0, i; | |
5783 | char buf[32]; | |
5784 | ||
5785 | for_each_domain(cpu, sd) | |
5786 | domain_num++; | |
5787 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5788 | if (table == NULL) |
5789 | return NULL; | |
e692ab53 NP |
5790 | |
5791 | i = 0; | |
5792 | for_each_domain(cpu, sd) { | |
5793 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5794 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5795 | entry->mode = 0555; |
e692ab53 NP |
5796 | entry->child = sd_alloc_ctl_domain_table(sd); |
5797 | entry++; | |
5798 | i++; | |
5799 | } | |
5800 | return table; | |
5801 | } | |
5802 | ||
5803 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5804 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5805 | { |
6ad4c188 | 5806 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5807 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5808 | char buf[32]; | |
5809 | ||
7378547f MM |
5810 | WARN_ON(sd_ctl_dir[0].child); |
5811 | sd_ctl_dir[0].child = entry; | |
5812 | ||
ad1cdc1d MM |
5813 | if (entry == NULL) |
5814 | return; | |
5815 | ||
6ad4c188 | 5816 | for_each_possible_cpu(i) { |
e692ab53 | 5817 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5818 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5819 | entry->mode = 0555; |
e692ab53 | 5820 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5821 | entry++; |
e692ab53 | 5822 | } |
7378547f MM |
5823 | |
5824 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5825 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5826 | } | |
6382bc90 | 5827 | |
7378547f | 5828 | /* may be called multiple times per register */ |
6382bc90 MM |
5829 | static void unregister_sched_domain_sysctl(void) |
5830 | { | |
7378547f MM |
5831 | if (sd_sysctl_header) |
5832 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 5833 | sd_sysctl_header = NULL; |
7378547f MM |
5834 | if (sd_ctl_dir[0].child) |
5835 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 5836 | } |
e692ab53 | 5837 | #else |
6382bc90 MM |
5838 | static void register_sched_domain_sysctl(void) |
5839 | { | |
5840 | } | |
5841 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
5842 | { |
5843 | } | |
5844 | #endif | |
5845 | ||
1f11eb6a GH |
5846 | static void set_rq_online(struct rq *rq) |
5847 | { | |
5848 | if (!rq->online) { | |
5849 | const struct sched_class *class; | |
5850 | ||
c6c4927b | 5851 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5852 | rq->online = 1; |
5853 | ||
5854 | for_each_class(class) { | |
5855 | if (class->rq_online) | |
5856 | class->rq_online(rq); | |
5857 | } | |
5858 | } | |
5859 | } | |
5860 | ||
5861 | static void set_rq_offline(struct rq *rq) | |
5862 | { | |
5863 | if (rq->online) { | |
5864 | const struct sched_class *class; | |
5865 | ||
5866 | for_each_class(class) { | |
5867 | if (class->rq_offline) | |
5868 | class->rq_offline(rq); | |
5869 | } | |
5870 | ||
c6c4927b | 5871 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
5872 | rq->online = 0; |
5873 | } | |
5874 | } | |
5875 | ||
1da177e4 LT |
5876 | /* |
5877 | * migration_call - callback that gets triggered when a CPU is added. | |
5878 | * Here we can start up the necessary migration thread for the new CPU. | |
5879 | */ | |
48f24c4d IM |
5880 | static int __cpuinit |
5881 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 5882 | { |
48f24c4d | 5883 | int cpu = (long)hcpu; |
1da177e4 | 5884 | unsigned long flags; |
969c7921 | 5885 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 5886 | |
48c5ccae | 5887 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 5888 | |
1da177e4 | 5889 | case CPU_UP_PREPARE: |
a468d389 | 5890 | rq->calc_load_update = calc_load_update; |
1da177e4 | 5891 | break; |
48f24c4d | 5892 | |
1da177e4 | 5893 | case CPU_ONLINE: |
1f94ef59 | 5894 | /* Update our root-domain */ |
05fa785c | 5895 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 5896 | if (rq->rd) { |
c6c4927b | 5897 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
5898 | |
5899 | set_rq_online(rq); | |
1f94ef59 | 5900 | } |
05fa785c | 5901 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 5902 | break; |
48f24c4d | 5903 | |
1da177e4 | 5904 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 5905 | case CPU_DYING: |
57d885fe | 5906 | /* Update our root-domain */ |
05fa785c | 5907 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 5908 | if (rq->rd) { |
c6c4927b | 5909 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 5910 | set_rq_offline(rq); |
57d885fe | 5911 | } |
48c5ccae PZ |
5912 | migrate_tasks(cpu); |
5913 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 5914 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
5915 | |
5916 | migrate_nr_uninterruptible(rq); | |
5917 | calc_global_load_remove(rq); | |
57d885fe | 5918 | break; |
1da177e4 LT |
5919 | #endif |
5920 | } | |
5921 | return NOTIFY_OK; | |
5922 | } | |
5923 | ||
f38b0820 PM |
5924 | /* |
5925 | * Register at high priority so that task migration (migrate_all_tasks) | |
5926 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 5927 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 5928 | */ |
26c2143b | 5929 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 5930 | .notifier_call = migration_call, |
50a323b7 | 5931 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
5932 | }; |
5933 | ||
3a101d05 TH |
5934 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
5935 | unsigned long action, void *hcpu) | |
5936 | { | |
5937 | switch (action & ~CPU_TASKS_FROZEN) { | |
5938 | case CPU_ONLINE: | |
5939 | case CPU_DOWN_FAILED: | |
5940 | set_cpu_active((long)hcpu, true); | |
5941 | return NOTIFY_OK; | |
5942 | default: | |
5943 | return NOTIFY_DONE; | |
5944 | } | |
5945 | } | |
5946 | ||
5947 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
5948 | unsigned long action, void *hcpu) | |
5949 | { | |
5950 | switch (action & ~CPU_TASKS_FROZEN) { | |
5951 | case CPU_DOWN_PREPARE: | |
5952 | set_cpu_active((long)hcpu, false); | |
5953 | return NOTIFY_OK; | |
5954 | default: | |
5955 | return NOTIFY_DONE; | |
5956 | } | |
5957 | } | |
5958 | ||
7babe8db | 5959 | static int __init migration_init(void) |
1da177e4 LT |
5960 | { |
5961 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 5962 | int err; |
48f24c4d | 5963 | |
3a101d05 | 5964 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
5965 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
5966 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
5967 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
5968 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 5969 | |
3a101d05 TH |
5970 | /* Register cpu active notifiers */ |
5971 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
5972 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
5973 | ||
a004cd42 | 5974 | return 0; |
1da177e4 | 5975 | } |
7babe8db | 5976 | early_initcall(migration_init); |
1da177e4 LT |
5977 | #endif |
5978 | ||
5979 | #ifdef CONFIG_SMP | |
476f3534 | 5980 | |
3e9830dc | 5981 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 5982 | |
f6630114 MT |
5983 | static __read_mostly int sched_domain_debug_enabled; |
5984 | ||
5985 | static int __init sched_domain_debug_setup(char *str) | |
5986 | { | |
5987 | sched_domain_debug_enabled = 1; | |
5988 | ||
5989 | return 0; | |
5990 | } | |
5991 | early_param("sched_debug", sched_domain_debug_setup); | |
5992 | ||
7c16ec58 | 5993 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 5994 | struct cpumask *groupmask) |
1da177e4 | 5995 | { |
4dcf6aff | 5996 | struct sched_group *group = sd->groups; |
434d53b0 | 5997 | char str[256]; |
1da177e4 | 5998 | |
968ea6d8 | 5999 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6000 | cpumask_clear(groupmask); |
4dcf6aff IM |
6001 | |
6002 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6003 | ||
6004 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6005 | printk("does not load-balance\n"); |
4dcf6aff | 6006 | if (sd->parent) |
3df0fc5b PZ |
6007 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6008 | " has parent"); | |
4dcf6aff | 6009 | return -1; |
41c7ce9a NP |
6010 | } |
6011 | ||
3df0fc5b | 6012 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6013 | |
758b2cdc | 6014 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6015 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6016 | "CPU%d\n", cpu); | |
4dcf6aff | 6017 | } |
758b2cdc | 6018 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6019 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6020 | " CPU%d\n", cpu); | |
4dcf6aff | 6021 | } |
1da177e4 | 6022 | |
4dcf6aff | 6023 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6024 | do { |
4dcf6aff | 6025 | if (!group) { |
3df0fc5b PZ |
6026 | printk("\n"); |
6027 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6028 | break; |
6029 | } | |
6030 | ||
18a3885f | 6031 | if (!group->cpu_power) { |
3df0fc5b PZ |
6032 | printk(KERN_CONT "\n"); |
6033 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6034 | "set\n"); | |
4dcf6aff IM |
6035 | break; |
6036 | } | |
1da177e4 | 6037 | |
758b2cdc | 6038 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6039 | printk(KERN_CONT "\n"); |
6040 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6041 | break; |
6042 | } | |
1da177e4 | 6043 | |
758b2cdc | 6044 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6045 | printk(KERN_CONT "\n"); |
6046 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6047 | break; |
6048 | } | |
1da177e4 | 6049 | |
758b2cdc | 6050 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6051 | |
968ea6d8 | 6052 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6053 | |
3df0fc5b | 6054 | printk(KERN_CONT " %s", str); |
18a3885f | 6055 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6056 | printk(KERN_CONT " (cpu_power = %d)", |
6057 | group->cpu_power); | |
381512cf | 6058 | } |
1da177e4 | 6059 | |
4dcf6aff IM |
6060 | group = group->next; |
6061 | } while (group != sd->groups); | |
3df0fc5b | 6062 | printk(KERN_CONT "\n"); |
1da177e4 | 6063 | |
758b2cdc | 6064 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6065 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6066 | |
758b2cdc RR |
6067 | if (sd->parent && |
6068 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6069 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6070 | "of domain->span\n"); | |
4dcf6aff IM |
6071 | return 0; |
6072 | } | |
1da177e4 | 6073 | |
4dcf6aff IM |
6074 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6075 | { | |
d5dd3db1 | 6076 | cpumask_var_t groupmask; |
4dcf6aff | 6077 | int level = 0; |
1da177e4 | 6078 | |
f6630114 MT |
6079 | if (!sched_domain_debug_enabled) |
6080 | return; | |
6081 | ||
4dcf6aff IM |
6082 | if (!sd) { |
6083 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6084 | return; | |
6085 | } | |
1da177e4 | 6086 | |
4dcf6aff IM |
6087 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6088 | ||
d5dd3db1 | 6089 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6090 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6091 | return; | |
6092 | } | |
6093 | ||
4dcf6aff | 6094 | for (;;) { |
7c16ec58 | 6095 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6096 | break; |
1da177e4 LT |
6097 | level++; |
6098 | sd = sd->parent; | |
33859f7f | 6099 | if (!sd) |
4dcf6aff IM |
6100 | break; |
6101 | } | |
d5dd3db1 | 6102 | free_cpumask_var(groupmask); |
1da177e4 | 6103 | } |
6d6bc0ad | 6104 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6105 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6106 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6107 | |
1a20ff27 | 6108 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6109 | { |
758b2cdc | 6110 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6111 | return 1; |
6112 | ||
6113 | /* Following flags need at least 2 groups */ | |
6114 | if (sd->flags & (SD_LOAD_BALANCE | | |
6115 | SD_BALANCE_NEWIDLE | | |
6116 | SD_BALANCE_FORK | | |
89c4710e SS |
6117 | SD_BALANCE_EXEC | |
6118 | SD_SHARE_CPUPOWER | | |
6119 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6120 | if (sd->groups != sd->groups->next) |
6121 | return 0; | |
6122 | } | |
6123 | ||
6124 | /* Following flags don't use groups */ | |
c88d5910 | 6125 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6126 | return 0; |
6127 | ||
6128 | return 1; | |
6129 | } | |
6130 | ||
48f24c4d IM |
6131 | static int |
6132 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6133 | { |
6134 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6135 | ||
6136 | if (sd_degenerate(parent)) | |
6137 | return 1; | |
6138 | ||
758b2cdc | 6139 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6140 | return 0; |
6141 | ||
245af2c7 SS |
6142 | /* Flags needing groups don't count if only 1 group in parent */ |
6143 | if (parent->groups == parent->groups->next) { | |
6144 | pflags &= ~(SD_LOAD_BALANCE | | |
6145 | SD_BALANCE_NEWIDLE | | |
6146 | SD_BALANCE_FORK | | |
89c4710e SS |
6147 | SD_BALANCE_EXEC | |
6148 | SD_SHARE_CPUPOWER | | |
6149 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6150 | if (nr_node_ids == 1) |
6151 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6152 | } |
6153 | if (~cflags & pflags) | |
6154 | return 0; | |
6155 | ||
6156 | return 1; | |
6157 | } | |
6158 | ||
c6c4927b RR |
6159 | static void free_rootdomain(struct root_domain *rd) |
6160 | { | |
047106ad PZ |
6161 | synchronize_sched(); |
6162 | ||
68e74568 RR |
6163 | cpupri_cleanup(&rd->cpupri); |
6164 | ||
c6c4927b RR |
6165 | free_cpumask_var(rd->rto_mask); |
6166 | free_cpumask_var(rd->online); | |
6167 | free_cpumask_var(rd->span); | |
6168 | kfree(rd); | |
6169 | } | |
6170 | ||
57d885fe GH |
6171 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6172 | { | |
a0490fa3 | 6173 | struct root_domain *old_rd = NULL; |
57d885fe | 6174 | unsigned long flags; |
57d885fe | 6175 | |
05fa785c | 6176 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6177 | |
6178 | if (rq->rd) { | |
a0490fa3 | 6179 | old_rd = rq->rd; |
57d885fe | 6180 | |
c6c4927b | 6181 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6182 | set_rq_offline(rq); |
57d885fe | 6183 | |
c6c4927b | 6184 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6185 | |
a0490fa3 IM |
6186 | /* |
6187 | * If we dont want to free the old_rt yet then | |
6188 | * set old_rd to NULL to skip the freeing later | |
6189 | * in this function: | |
6190 | */ | |
6191 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6192 | old_rd = NULL; | |
57d885fe GH |
6193 | } |
6194 | ||
6195 | atomic_inc(&rd->refcount); | |
6196 | rq->rd = rd; | |
6197 | ||
c6c4927b | 6198 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6199 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6200 | set_rq_online(rq); |
57d885fe | 6201 | |
05fa785c | 6202 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6203 | |
6204 | if (old_rd) | |
6205 | free_rootdomain(old_rd); | |
57d885fe GH |
6206 | } |
6207 | ||
68c38fc3 | 6208 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6209 | { |
6210 | memset(rd, 0, sizeof(*rd)); | |
6211 | ||
68c38fc3 | 6212 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6213 | goto out; |
68c38fc3 | 6214 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6215 | goto free_span; |
68c38fc3 | 6216 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6217 | goto free_online; |
6e0534f2 | 6218 | |
68c38fc3 | 6219 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6220 | goto free_rto_mask; |
c6c4927b | 6221 | return 0; |
6e0534f2 | 6222 | |
68e74568 RR |
6223 | free_rto_mask: |
6224 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6225 | free_online: |
6226 | free_cpumask_var(rd->online); | |
6227 | free_span: | |
6228 | free_cpumask_var(rd->span); | |
0c910d28 | 6229 | out: |
c6c4927b | 6230 | return -ENOMEM; |
57d885fe GH |
6231 | } |
6232 | ||
6233 | static void init_defrootdomain(void) | |
6234 | { | |
68c38fc3 | 6235 | init_rootdomain(&def_root_domain); |
c6c4927b | 6236 | |
57d885fe GH |
6237 | atomic_set(&def_root_domain.refcount, 1); |
6238 | } | |
6239 | ||
dc938520 | 6240 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6241 | { |
6242 | struct root_domain *rd; | |
6243 | ||
6244 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6245 | if (!rd) | |
6246 | return NULL; | |
6247 | ||
68c38fc3 | 6248 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6249 | kfree(rd); |
6250 | return NULL; | |
6251 | } | |
57d885fe GH |
6252 | |
6253 | return rd; | |
6254 | } | |
6255 | ||
1da177e4 | 6256 | /* |
0eab9146 | 6257 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6258 | * hold the hotplug lock. |
6259 | */ | |
0eab9146 IM |
6260 | static void |
6261 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6262 | { |
70b97a7f | 6263 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6264 | struct sched_domain *tmp; |
6265 | ||
669c55e9 PZ |
6266 | for (tmp = sd; tmp; tmp = tmp->parent) |
6267 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6268 | ||
245af2c7 | 6269 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6270 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6271 | struct sched_domain *parent = tmp->parent; |
6272 | if (!parent) | |
6273 | break; | |
f29c9b1c | 6274 | |
1a848870 | 6275 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6276 | tmp->parent = parent->parent; |
1a848870 SS |
6277 | if (parent->parent) |
6278 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6279 | } else |
6280 | tmp = tmp->parent; | |
245af2c7 SS |
6281 | } |
6282 | ||
1a848870 | 6283 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6284 | sd = sd->parent; |
1a848870 SS |
6285 | if (sd) |
6286 | sd->child = NULL; | |
6287 | } | |
1da177e4 LT |
6288 | |
6289 | sched_domain_debug(sd, cpu); | |
6290 | ||
57d885fe | 6291 | rq_attach_root(rq, rd); |
674311d5 | 6292 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6293 | } |
6294 | ||
6295 | /* cpus with isolated domains */ | |
dcc30a35 | 6296 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6297 | |
6298 | /* Setup the mask of cpus configured for isolated domains */ | |
6299 | static int __init isolated_cpu_setup(char *str) | |
6300 | { | |
bdddd296 | 6301 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6302 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6303 | return 1; |
6304 | } | |
6305 | ||
8927f494 | 6306 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6307 | |
6308 | /* | |
6711cab4 SS |
6309 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6310 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6311 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6312 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6313 | * |
6314 | * init_sched_build_groups will build a circular linked list of the groups | |
6315 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6316 | * and ->cpu_power to 0. | |
6317 | */ | |
a616058b | 6318 | static void |
96f874e2 RR |
6319 | init_sched_build_groups(const struct cpumask *span, |
6320 | const struct cpumask *cpu_map, | |
6321 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6322 | struct sched_group **sg, |
96f874e2 RR |
6323 | struct cpumask *tmpmask), |
6324 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6325 | { |
6326 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6327 | int i; |
6328 | ||
96f874e2 | 6329 | cpumask_clear(covered); |
7c16ec58 | 6330 | |
abcd083a | 6331 | for_each_cpu(i, span) { |
6711cab4 | 6332 | struct sched_group *sg; |
7c16ec58 | 6333 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6334 | int j; |
6335 | ||
758b2cdc | 6336 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6337 | continue; |
6338 | ||
758b2cdc | 6339 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6340 | sg->cpu_power = 0; |
1da177e4 | 6341 | |
abcd083a | 6342 | for_each_cpu(j, span) { |
7c16ec58 | 6343 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6344 | continue; |
6345 | ||
96f874e2 | 6346 | cpumask_set_cpu(j, covered); |
758b2cdc | 6347 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6348 | } |
6349 | if (!first) | |
6350 | first = sg; | |
6351 | if (last) | |
6352 | last->next = sg; | |
6353 | last = sg; | |
6354 | } | |
6355 | last->next = first; | |
6356 | } | |
6357 | ||
9c1cfda2 | 6358 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6359 | |
9c1cfda2 | 6360 | #ifdef CONFIG_NUMA |
198e2f18 | 6361 | |
9c1cfda2 JH |
6362 | /** |
6363 | * find_next_best_node - find the next node to include in a sched_domain | |
6364 | * @node: node whose sched_domain we're building | |
6365 | * @used_nodes: nodes already in the sched_domain | |
6366 | * | |
41a2d6cf | 6367 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6368 | * finds the closest node not already in the @used_nodes map. |
6369 | * | |
6370 | * Should use nodemask_t. | |
6371 | */ | |
c5f59f08 | 6372 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6373 | { |
6374 | int i, n, val, min_val, best_node = 0; | |
6375 | ||
6376 | min_val = INT_MAX; | |
6377 | ||
076ac2af | 6378 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6379 | /* Start at @node */ |
076ac2af | 6380 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6381 | |
6382 | if (!nr_cpus_node(n)) | |
6383 | continue; | |
6384 | ||
6385 | /* Skip already used nodes */ | |
c5f59f08 | 6386 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6387 | continue; |
6388 | ||
6389 | /* Simple min distance search */ | |
6390 | val = node_distance(node, n); | |
6391 | ||
6392 | if (val < min_val) { | |
6393 | min_val = val; | |
6394 | best_node = n; | |
6395 | } | |
6396 | } | |
6397 | ||
c5f59f08 | 6398 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6399 | return best_node; |
6400 | } | |
6401 | ||
6402 | /** | |
6403 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6404 | * @node: node whose cpumask we're constructing | |
73486722 | 6405 | * @span: resulting cpumask |
9c1cfda2 | 6406 | * |
41a2d6cf | 6407 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6408 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6409 | * out optimally. | |
6410 | */ | |
96f874e2 | 6411 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6412 | { |
c5f59f08 | 6413 | nodemask_t used_nodes; |
48f24c4d | 6414 | int i; |
9c1cfda2 | 6415 | |
6ca09dfc | 6416 | cpumask_clear(span); |
c5f59f08 | 6417 | nodes_clear(used_nodes); |
9c1cfda2 | 6418 | |
6ca09dfc | 6419 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6420 | node_set(node, used_nodes); |
9c1cfda2 JH |
6421 | |
6422 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6423 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6424 | |
6ca09dfc | 6425 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6426 | } |
9c1cfda2 | 6427 | } |
6d6bc0ad | 6428 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6429 | |
5c45bf27 | 6430 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6431 | |
6c99e9ad RR |
6432 | /* |
6433 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6434 | * |
6435 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6436 | * and struct sched_domain. ) | |
6c99e9ad RR |
6437 | */ |
6438 | struct static_sched_group { | |
6439 | struct sched_group sg; | |
6440 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6441 | }; | |
6442 | ||
6443 | struct static_sched_domain { | |
6444 | struct sched_domain sd; | |
6445 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6446 | }; | |
6447 | ||
49a02c51 AH |
6448 | struct s_data { |
6449 | #ifdef CONFIG_NUMA | |
6450 | int sd_allnodes; | |
6451 | cpumask_var_t domainspan; | |
6452 | cpumask_var_t covered; | |
6453 | cpumask_var_t notcovered; | |
6454 | #endif | |
6455 | cpumask_var_t nodemask; | |
6456 | cpumask_var_t this_sibling_map; | |
6457 | cpumask_var_t this_core_map; | |
01a08546 | 6458 | cpumask_var_t this_book_map; |
49a02c51 AH |
6459 | cpumask_var_t send_covered; |
6460 | cpumask_var_t tmpmask; | |
6461 | struct sched_group **sched_group_nodes; | |
6462 | struct root_domain *rd; | |
6463 | }; | |
6464 | ||
2109b99e AH |
6465 | enum s_alloc { |
6466 | sa_sched_groups = 0, | |
6467 | sa_rootdomain, | |
6468 | sa_tmpmask, | |
6469 | sa_send_covered, | |
01a08546 | 6470 | sa_this_book_map, |
2109b99e AH |
6471 | sa_this_core_map, |
6472 | sa_this_sibling_map, | |
6473 | sa_nodemask, | |
6474 | sa_sched_group_nodes, | |
6475 | #ifdef CONFIG_NUMA | |
6476 | sa_notcovered, | |
6477 | sa_covered, | |
6478 | sa_domainspan, | |
6479 | #endif | |
6480 | sa_none, | |
6481 | }; | |
6482 | ||
9c1cfda2 | 6483 | /* |
48f24c4d | 6484 | * SMT sched-domains: |
9c1cfda2 | 6485 | */ |
1da177e4 | 6486 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6487 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6488 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6489 | |
41a2d6cf | 6490 | static int |
96f874e2 RR |
6491 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6492 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6493 | { |
6711cab4 | 6494 | if (sg) |
1871e52c | 6495 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6496 | return cpu; |
6497 | } | |
6d6bc0ad | 6498 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6499 | |
48f24c4d IM |
6500 | /* |
6501 | * multi-core sched-domains: | |
6502 | */ | |
1e9f28fa | 6503 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6504 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6505 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6506 | |
41a2d6cf | 6507 | static int |
96f874e2 RR |
6508 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6509 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6510 | { |
6711cab4 | 6511 | int group; |
f269893c | 6512 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 6513 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6514 | group = cpumask_first(mask); |
f269893c HC |
6515 | #else |
6516 | group = cpu; | |
6517 | #endif | |
6711cab4 | 6518 | if (sg) |
6c99e9ad | 6519 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6520 | return group; |
1e9f28fa | 6521 | } |
f269893c | 6522 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 6523 | |
01a08546 HC |
6524 | /* |
6525 | * book sched-domains: | |
6526 | */ | |
6527 | #ifdef CONFIG_SCHED_BOOK | |
6528 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
6529 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
6530 | ||
41a2d6cf | 6531 | static int |
01a08546 HC |
6532 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
6533 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6534 | { |
01a08546 HC |
6535 | int group = cpu; |
6536 | #ifdef CONFIG_SCHED_MC | |
6537 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
6538 | group = cpumask_first(mask); | |
6539 | #elif defined(CONFIG_SCHED_SMT) | |
6540 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
6541 | group = cpumask_first(mask); | |
6542 | #endif | |
6711cab4 | 6543 | if (sg) |
01a08546 HC |
6544 | *sg = &per_cpu(sched_group_book, group).sg; |
6545 | return group; | |
1e9f28fa | 6546 | } |
01a08546 | 6547 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 6548 | |
6c99e9ad RR |
6549 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6550 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6551 | |
41a2d6cf | 6552 | static int |
96f874e2 RR |
6553 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6554 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6555 | { |
6711cab4 | 6556 | int group; |
01a08546 HC |
6557 | #ifdef CONFIG_SCHED_BOOK |
6558 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
6559 | group = cpumask_first(mask); | |
6560 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 6561 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6562 | group = cpumask_first(mask); |
1e9f28fa | 6563 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6564 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6565 | group = cpumask_first(mask); |
1da177e4 | 6566 | #else |
6711cab4 | 6567 | group = cpu; |
1da177e4 | 6568 | #endif |
6711cab4 | 6569 | if (sg) |
6c99e9ad | 6570 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6571 | return group; |
1da177e4 LT |
6572 | } |
6573 | ||
6574 | #ifdef CONFIG_NUMA | |
1da177e4 | 6575 | /* |
9c1cfda2 JH |
6576 | * The init_sched_build_groups can't handle what we want to do with node |
6577 | * groups, so roll our own. Now each node has its own list of groups which | |
6578 | * gets dynamically allocated. | |
1da177e4 | 6579 | */ |
62ea9ceb | 6580 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6581 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6582 | |
62ea9ceb | 6583 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6584 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6585 | |
96f874e2 RR |
6586 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6587 | struct sched_group **sg, | |
6588 | struct cpumask *nodemask) | |
9c1cfda2 | 6589 | { |
6711cab4 SS |
6590 | int group; |
6591 | ||
6ca09dfc | 6592 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6593 | group = cpumask_first(nodemask); |
6711cab4 SS |
6594 | |
6595 | if (sg) | |
6c99e9ad | 6596 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6597 | return group; |
1da177e4 | 6598 | } |
6711cab4 | 6599 | |
08069033 SS |
6600 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6601 | { | |
6602 | struct sched_group *sg = group_head; | |
6603 | int j; | |
6604 | ||
6605 | if (!sg) | |
6606 | return; | |
3a5c359a | 6607 | do { |
758b2cdc | 6608 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6609 | struct sched_domain *sd; |
08069033 | 6610 | |
6c99e9ad | 6611 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6612 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6613 | /* |
6614 | * Only add "power" once for each | |
6615 | * physical package. | |
6616 | */ | |
6617 | continue; | |
6618 | } | |
08069033 | 6619 | |
18a3885f | 6620 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6621 | } |
6622 | sg = sg->next; | |
6623 | } while (sg != group_head); | |
08069033 | 6624 | } |
0601a88d AH |
6625 | |
6626 | static int build_numa_sched_groups(struct s_data *d, | |
6627 | const struct cpumask *cpu_map, int num) | |
6628 | { | |
6629 | struct sched_domain *sd; | |
6630 | struct sched_group *sg, *prev; | |
6631 | int n, j; | |
6632 | ||
6633 | cpumask_clear(d->covered); | |
6634 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6635 | if (cpumask_empty(d->nodemask)) { | |
6636 | d->sched_group_nodes[num] = NULL; | |
6637 | goto out; | |
6638 | } | |
6639 | ||
6640 | sched_domain_node_span(num, d->domainspan); | |
6641 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6642 | ||
6643 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6644 | GFP_KERNEL, num); | |
6645 | if (!sg) { | |
3df0fc5b PZ |
6646 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6647 | num); | |
0601a88d AH |
6648 | return -ENOMEM; |
6649 | } | |
6650 | d->sched_group_nodes[num] = sg; | |
6651 | ||
6652 | for_each_cpu(j, d->nodemask) { | |
6653 | sd = &per_cpu(node_domains, j).sd; | |
6654 | sd->groups = sg; | |
6655 | } | |
6656 | ||
18a3885f | 6657 | sg->cpu_power = 0; |
0601a88d AH |
6658 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6659 | sg->next = sg; | |
6660 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6661 | ||
6662 | prev = sg; | |
6663 | for (j = 0; j < nr_node_ids; j++) { | |
6664 | n = (num + j) % nr_node_ids; | |
6665 | cpumask_complement(d->notcovered, d->covered); | |
6666 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6667 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6668 | if (cpumask_empty(d->tmpmask)) | |
6669 | break; | |
6670 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6671 | if (cpumask_empty(d->tmpmask)) | |
6672 | continue; | |
6673 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6674 | GFP_KERNEL, num); | |
6675 | if (!sg) { | |
3df0fc5b PZ |
6676 | printk(KERN_WARNING |
6677 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6678 | return -ENOMEM; |
6679 | } | |
18a3885f | 6680 | sg->cpu_power = 0; |
0601a88d AH |
6681 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6682 | sg->next = prev->next; | |
6683 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6684 | prev->next = sg; | |
6685 | prev = sg; | |
6686 | } | |
6687 | out: | |
6688 | return 0; | |
6689 | } | |
6d6bc0ad | 6690 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6691 | |
a616058b | 6692 | #ifdef CONFIG_NUMA |
51888ca2 | 6693 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6694 | static void free_sched_groups(const struct cpumask *cpu_map, |
6695 | struct cpumask *nodemask) | |
51888ca2 | 6696 | { |
a616058b | 6697 | int cpu, i; |
51888ca2 | 6698 | |
abcd083a | 6699 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6700 | struct sched_group **sched_group_nodes |
6701 | = sched_group_nodes_bycpu[cpu]; | |
6702 | ||
51888ca2 SV |
6703 | if (!sched_group_nodes) |
6704 | continue; | |
6705 | ||
076ac2af | 6706 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6707 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6708 | ||
6ca09dfc | 6709 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6710 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6711 | continue; |
6712 | ||
6713 | if (sg == NULL) | |
6714 | continue; | |
6715 | sg = sg->next; | |
6716 | next_sg: | |
6717 | oldsg = sg; | |
6718 | sg = sg->next; | |
6719 | kfree(oldsg); | |
6720 | if (oldsg != sched_group_nodes[i]) | |
6721 | goto next_sg; | |
6722 | } | |
6723 | kfree(sched_group_nodes); | |
6724 | sched_group_nodes_bycpu[cpu] = NULL; | |
6725 | } | |
51888ca2 | 6726 | } |
6d6bc0ad | 6727 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6728 | static void free_sched_groups(const struct cpumask *cpu_map, |
6729 | struct cpumask *nodemask) | |
a616058b SS |
6730 | { |
6731 | } | |
6d6bc0ad | 6732 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6733 | |
89c4710e SS |
6734 | /* |
6735 | * Initialize sched groups cpu_power. | |
6736 | * | |
6737 | * cpu_power indicates the capacity of sched group, which is used while | |
6738 | * distributing the load between different sched groups in a sched domain. | |
6739 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6740 | * there are asymmetries in the topology. If there are asymmetries, group | |
6741 | * having more cpu_power will pickup more load compared to the group having | |
6742 | * less cpu_power. | |
89c4710e SS |
6743 | */ |
6744 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6745 | { | |
6746 | struct sched_domain *child; | |
6747 | struct sched_group *group; | |
f93e65c1 PZ |
6748 | long power; |
6749 | int weight; | |
89c4710e SS |
6750 | |
6751 | WARN_ON(!sd || !sd->groups); | |
6752 | ||
13318a71 | 6753 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6754 | return; |
6755 | ||
aae6d3dd SS |
6756 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
6757 | ||
89c4710e SS |
6758 | child = sd->child; |
6759 | ||
18a3885f | 6760 | sd->groups->cpu_power = 0; |
5517d86b | 6761 | |
f93e65c1 PZ |
6762 | if (!child) { |
6763 | power = SCHED_LOAD_SCALE; | |
6764 | weight = cpumask_weight(sched_domain_span(sd)); | |
6765 | /* | |
6766 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6767 | * Usually multiple threads get a better yield out of |
6768 | * that one core than a single thread would have, | |
6769 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6770 | */ |
a52bfd73 PZ |
6771 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6772 | power *= sd->smt_gain; | |
f93e65c1 | 6773 | power /= weight; |
a52bfd73 PZ |
6774 | power >>= SCHED_LOAD_SHIFT; |
6775 | } | |
18a3885f | 6776 | sd->groups->cpu_power += power; |
89c4710e SS |
6777 | return; |
6778 | } | |
6779 | ||
89c4710e | 6780 | /* |
f93e65c1 | 6781 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6782 | */ |
6783 | group = child->groups; | |
6784 | do { | |
18a3885f | 6785 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6786 | group = group->next; |
6787 | } while (group != child->groups); | |
6788 | } | |
6789 | ||
7c16ec58 MT |
6790 | /* |
6791 | * Initializers for schedule domains | |
6792 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6793 | */ | |
6794 | ||
a5d8c348 IM |
6795 | #ifdef CONFIG_SCHED_DEBUG |
6796 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6797 | #else | |
6798 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6799 | #endif | |
6800 | ||
7c16ec58 | 6801 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6802 | |
7c16ec58 MT |
6803 | #define SD_INIT_FUNC(type) \ |
6804 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6805 | { \ | |
6806 | memset(sd, 0, sizeof(*sd)); \ | |
6807 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6808 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6809 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6810 | } |
6811 | ||
6812 | SD_INIT_FUNC(CPU) | |
6813 | #ifdef CONFIG_NUMA | |
6814 | SD_INIT_FUNC(ALLNODES) | |
6815 | SD_INIT_FUNC(NODE) | |
6816 | #endif | |
6817 | #ifdef CONFIG_SCHED_SMT | |
6818 | SD_INIT_FUNC(SIBLING) | |
6819 | #endif | |
6820 | #ifdef CONFIG_SCHED_MC | |
6821 | SD_INIT_FUNC(MC) | |
6822 | #endif | |
01a08546 HC |
6823 | #ifdef CONFIG_SCHED_BOOK |
6824 | SD_INIT_FUNC(BOOK) | |
6825 | #endif | |
7c16ec58 | 6826 | |
1d3504fc HS |
6827 | static int default_relax_domain_level = -1; |
6828 | ||
6829 | static int __init setup_relax_domain_level(char *str) | |
6830 | { | |
30e0e178 LZ |
6831 | unsigned long val; |
6832 | ||
6833 | val = simple_strtoul(str, NULL, 0); | |
6834 | if (val < SD_LV_MAX) | |
6835 | default_relax_domain_level = val; | |
6836 | ||
1d3504fc HS |
6837 | return 1; |
6838 | } | |
6839 | __setup("relax_domain_level=", setup_relax_domain_level); | |
6840 | ||
6841 | static void set_domain_attribute(struct sched_domain *sd, | |
6842 | struct sched_domain_attr *attr) | |
6843 | { | |
6844 | int request; | |
6845 | ||
6846 | if (!attr || attr->relax_domain_level < 0) { | |
6847 | if (default_relax_domain_level < 0) | |
6848 | return; | |
6849 | else | |
6850 | request = default_relax_domain_level; | |
6851 | } else | |
6852 | request = attr->relax_domain_level; | |
6853 | if (request < sd->level) { | |
6854 | /* turn off idle balance on this domain */ | |
c88d5910 | 6855 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6856 | } else { |
6857 | /* turn on idle balance on this domain */ | |
c88d5910 | 6858 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
6859 | } |
6860 | } | |
6861 | ||
2109b99e AH |
6862 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
6863 | const struct cpumask *cpu_map) | |
6864 | { | |
6865 | switch (what) { | |
6866 | case sa_sched_groups: | |
6867 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
6868 | d->sched_group_nodes = NULL; | |
6869 | case sa_rootdomain: | |
6870 | free_rootdomain(d->rd); /* fall through */ | |
6871 | case sa_tmpmask: | |
6872 | free_cpumask_var(d->tmpmask); /* fall through */ | |
6873 | case sa_send_covered: | |
6874 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
6875 | case sa_this_book_map: |
6876 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
6877 | case sa_this_core_map: |
6878 | free_cpumask_var(d->this_core_map); /* fall through */ | |
6879 | case sa_this_sibling_map: | |
6880 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
6881 | case sa_nodemask: | |
6882 | free_cpumask_var(d->nodemask); /* fall through */ | |
6883 | case sa_sched_group_nodes: | |
d1b55138 | 6884 | #ifdef CONFIG_NUMA |
2109b99e AH |
6885 | kfree(d->sched_group_nodes); /* fall through */ |
6886 | case sa_notcovered: | |
6887 | free_cpumask_var(d->notcovered); /* fall through */ | |
6888 | case sa_covered: | |
6889 | free_cpumask_var(d->covered); /* fall through */ | |
6890 | case sa_domainspan: | |
6891 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 6892 | #endif |
2109b99e AH |
6893 | case sa_none: |
6894 | break; | |
6895 | } | |
6896 | } | |
3404c8d9 | 6897 | |
2109b99e AH |
6898 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
6899 | const struct cpumask *cpu_map) | |
6900 | { | |
3404c8d9 | 6901 | #ifdef CONFIG_NUMA |
2109b99e AH |
6902 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
6903 | return sa_none; | |
6904 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
6905 | return sa_domainspan; | |
6906 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
6907 | return sa_covered; | |
6908 | /* Allocate the per-node list of sched groups */ | |
6909 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
6910 | sizeof(struct sched_group *), GFP_KERNEL); | |
6911 | if (!d->sched_group_nodes) { | |
3df0fc5b | 6912 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 6913 | return sa_notcovered; |
d1b55138 | 6914 | } |
2109b99e | 6915 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 6916 | #endif |
2109b99e AH |
6917 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
6918 | return sa_sched_group_nodes; | |
6919 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
6920 | return sa_nodemask; | |
6921 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
6922 | return sa_this_sibling_map; | |
01a08546 | 6923 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 6924 | return sa_this_core_map; |
01a08546 HC |
6925 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
6926 | return sa_this_book_map; | |
2109b99e AH |
6927 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
6928 | return sa_send_covered; | |
6929 | d->rd = alloc_rootdomain(); | |
6930 | if (!d->rd) { | |
3df0fc5b | 6931 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 6932 | return sa_tmpmask; |
57d885fe | 6933 | } |
2109b99e AH |
6934 | return sa_rootdomain; |
6935 | } | |
57d885fe | 6936 | |
7f4588f3 AH |
6937 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
6938 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
6939 | { | |
6940 | struct sched_domain *sd = NULL; | |
7c16ec58 | 6941 | #ifdef CONFIG_NUMA |
7f4588f3 | 6942 | struct sched_domain *parent; |
1da177e4 | 6943 | |
7f4588f3 AH |
6944 | d->sd_allnodes = 0; |
6945 | if (cpumask_weight(cpu_map) > | |
6946 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
6947 | sd = &per_cpu(allnodes_domains, i).sd; | |
6948 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 6949 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
6950 | cpumask_copy(sched_domain_span(sd), cpu_map); |
6951 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6952 | d->sd_allnodes = 1; | |
6953 | } | |
6954 | parent = sd; | |
6955 | ||
6956 | sd = &per_cpu(node_domains, i).sd; | |
6957 | SD_INIT(sd, NODE); | |
6958 | set_domain_attribute(sd, attr); | |
6959 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
6960 | sd->parent = parent; | |
6961 | if (parent) | |
6962 | parent->child = sd; | |
6963 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 6964 | #endif |
7f4588f3 AH |
6965 | return sd; |
6966 | } | |
1da177e4 | 6967 | |
87cce662 AH |
6968 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
6969 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6970 | struct sched_domain *parent, int i) | |
6971 | { | |
6972 | struct sched_domain *sd; | |
6973 | sd = &per_cpu(phys_domains, i).sd; | |
6974 | SD_INIT(sd, CPU); | |
6975 | set_domain_attribute(sd, attr); | |
6976 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
6977 | sd->parent = parent; | |
6978 | if (parent) | |
6979 | parent->child = sd; | |
6980 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6981 | return sd; | |
6982 | } | |
1da177e4 | 6983 | |
01a08546 HC |
6984 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
6985 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
6986 | struct sched_domain *parent, int i) | |
6987 | { | |
6988 | struct sched_domain *sd = parent; | |
6989 | #ifdef CONFIG_SCHED_BOOK | |
6990 | sd = &per_cpu(book_domains, i).sd; | |
6991 | SD_INIT(sd, BOOK); | |
6992 | set_domain_attribute(sd, attr); | |
6993 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
6994 | sd->parent = parent; | |
6995 | parent->child = sd; | |
6996 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
6997 | #endif | |
6998 | return sd; | |
6999 | } | |
7000 | ||
410c4081 AH |
7001 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7002 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7003 | struct sched_domain *parent, int i) | |
7004 | { | |
7005 | struct sched_domain *sd = parent; | |
1e9f28fa | 7006 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7007 | sd = &per_cpu(core_domains, i).sd; |
7008 | SD_INIT(sd, MC); | |
7009 | set_domain_attribute(sd, attr); | |
7010 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7011 | sd->parent = parent; | |
7012 | parent->child = sd; | |
7013 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7014 | #endif |
410c4081 AH |
7015 | return sd; |
7016 | } | |
1e9f28fa | 7017 | |
d8173535 AH |
7018 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7019 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7020 | struct sched_domain *parent, int i) | |
7021 | { | |
7022 | struct sched_domain *sd = parent; | |
1da177e4 | 7023 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7024 | sd = &per_cpu(cpu_domains, i).sd; |
7025 | SD_INIT(sd, SIBLING); | |
7026 | set_domain_attribute(sd, attr); | |
7027 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7028 | sd->parent = parent; | |
7029 | parent->child = sd; | |
7030 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7031 | #endif |
d8173535 AH |
7032 | return sd; |
7033 | } | |
1da177e4 | 7034 | |
0e8e85c9 AH |
7035 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7036 | const struct cpumask *cpu_map, int cpu) | |
7037 | { | |
7038 | switch (l) { | |
1da177e4 | 7039 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7040 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7041 | cpumask_and(d->this_sibling_map, cpu_map, | |
7042 | topology_thread_cpumask(cpu)); | |
7043 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7044 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7045 | &cpu_to_cpu_group, | |
7046 | d->send_covered, d->tmpmask); | |
7047 | break; | |
1da177e4 | 7048 | #endif |
1e9f28fa | 7049 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7050 | case SD_LV_MC: /* set up multi-core groups */ |
7051 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7052 | if (cpu == cpumask_first(d->this_core_map)) | |
7053 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7054 | &cpu_to_core_group, | |
7055 | d->send_covered, d->tmpmask); | |
7056 | break; | |
01a08546 HC |
7057 | #endif |
7058 | #ifdef CONFIG_SCHED_BOOK | |
7059 | case SD_LV_BOOK: /* set up book groups */ | |
7060 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7061 | if (cpu == cpumask_first(d->this_book_map)) | |
7062 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7063 | &cpu_to_book_group, | |
7064 | d->send_covered, d->tmpmask); | |
7065 | break; | |
1e9f28fa | 7066 | #endif |
86548096 AH |
7067 | case SD_LV_CPU: /* set up physical groups */ |
7068 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7069 | if (!cpumask_empty(d->nodemask)) | |
7070 | init_sched_build_groups(d->nodemask, cpu_map, | |
7071 | &cpu_to_phys_group, | |
7072 | d->send_covered, d->tmpmask); | |
7073 | break; | |
1da177e4 | 7074 | #ifdef CONFIG_NUMA |
de616e36 AH |
7075 | case SD_LV_ALLNODES: |
7076 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7077 | d->send_covered, d->tmpmask); | |
7078 | break; | |
7079 | #endif | |
0e8e85c9 AH |
7080 | default: |
7081 | break; | |
7c16ec58 | 7082 | } |
0e8e85c9 | 7083 | } |
9c1cfda2 | 7084 | |
2109b99e AH |
7085 | /* |
7086 | * Build sched domains for a given set of cpus and attach the sched domains | |
7087 | * to the individual cpus | |
7088 | */ | |
7089 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7090 | struct sched_domain_attr *attr) | |
7091 | { | |
7092 | enum s_alloc alloc_state = sa_none; | |
7093 | struct s_data d; | |
294b0c96 | 7094 | struct sched_domain *sd; |
2109b99e | 7095 | int i; |
7c16ec58 | 7096 | #ifdef CONFIG_NUMA |
2109b99e | 7097 | d.sd_allnodes = 0; |
7c16ec58 | 7098 | #endif |
9c1cfda2 | 7099 | |
2109b99e AH |
7100 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7101 | if (alloc_state != sa_rootdomain) | |
7102 | goto error; | |
7103 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7104 | |
1da177e4 | 7105 | /* |
1a20ff27 | 7106 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7107 | */ |
abcd083a | 7108 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7109 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7110 | cpu_map); | |
9761eea8 | 7111 | |
7f4588f3 | 7112 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7113 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7114 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7115 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7116 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7117 | } |
9c1cfda2 | 7118 | |
abcd083a | 7119 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7120 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7121 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7122 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7123 | } |
9c1cfda2 | 7124 | |
1da177e4 | 7125 | /* Set up physical groups */ |
86548096 AH |
7126 | for (i = 0; i < nr_node_ids; i++) |
7127 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7128 | |
1da177e4 LT |
7129 | #ifdef CONFIG_NUMA |
7130 | /* Set up node groups */ | |
de616e36 AH |
7131 | if (d.sd_allnodes) |
7132 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7133 | |
0601a88d AH |
7134 | for (i = 0; i < nr_node_ids; i++) |
7135 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7136 | goto error; |
1da177e4 LT |
7137 | #endif |
7138 | ||
7139 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7140 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7141 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7142 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7143 | init_sched_groups_power(i, sd); |
5c45bf27 | 7144 | } |
1da177e4 | 7145 | #endif |
1e9f28fa | 7146 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7147 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7148 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7149 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7150 | } |
7151 | #endif | |
01a08546 HC |
7152 | #ifdef CONFIG_SCHED_BOOK |
7153 | for_each_cpu(i, cpu_map) { | |
7154 | sd = &per_cpu(book_domains, i).sd; | |
7155 | init_sched_groups_power(i, sd); | |
7156 | } | |
7157 | #endif | |
1e9f28fa | 7158 | |
abcd083a | 7159 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7160 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7161 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7162 | } |
7163 | ||
9c1cfda2 | 7164 | #ifdef CONFIG_NUMA |
076ac2af | 7165 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7166 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7167 | |
49a02c51 | 7168 | if (d.sd_allnodes) { |
6711cab4 | 7169 | struct sched_group *sg; |
f712c0c7 | 7170 | |
96f874e2 | 7171 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7172 | d.tmpmask); |
f712c0c7 SS |
7173 | init_numa_sched_groups_power(sg); |
7174 | } | |
9c1cfda2 JH |
7175 | #endif |
7176 | ||
1da177e4 | 7177 | /* Attach the domains */ |
abcd083a | 7178 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7179 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7180 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7181 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7182 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7183 | #elif defined(CONFIG_SCHED_BOOK) |
7184 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7185 | #else |
6c99e9ad | 7186 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7187 | #endif |
49a02c51 | 7188 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7189 | } |
51888ca2 | 7190 | |
2109b99e AH |
7191 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7192 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7193 | return 0; | |
51888ca2 | 7194 | |
51888ca2 | 7195 | error: |
2109b99e AH |
7196 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7197 | return -ENOMEM; | |
1da177e4 | 7198 | } |
029190c5 | 7199 | |
96f874e2 | 7200 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7201 | { |
7202 | return __build_sched_domains(cpu_map, NULL); | |
7203 | } | |
7204 | ||
acc3f5d7 | 7205 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7206 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7207 | static struct sched_domain_attr *dattr_cur; |
7208 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7209 | |
7210 | /* | |
7211 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7212 | * cpumask) fails, then fallback to a single sched domain, |
7213 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7214 | */ |
4212823f | 7215 | static cpumask_var_t fallback_doms; |
029190c5 | 7216 | |
ee79d1bd HC |
7217 | /* |
7218 | * arch_update_cpu_topology lets virtualized architectures update the | |
7219 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7220 | * or 0 if it stayed the same. | |
7221 | */ | |
7222 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7223 | { |
ee79d1bd | 7224 | return 0; |
22e52b07 HC |
7225 | } |
7226 | ||
acc3f5d7 RR |
7227 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7228 | { | |
7229 | int i; | |
7230 | cpumask_var_t *doms; | |
7231 | ||
7232 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7233 | if (!doms) | |
7234 | return NULL; | |
7235 | for (i = 0; i < ndoms; i++) { | |
7236 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7237 | free_sched_domains(doms, i); | |
7238 | return NULL; | |
7239 | } | |
7240 | } | |
7241 | return doms; | |
7242 | } | |
7243 | ||
7244 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7245 | { | |
7246 | unsigned int i; | |
7247 | for (i = 0; i < ndoms; i++) | |
7248 | free_cpumask_var(doms[i]); | |
7249 | kfree(doms); | |
7250 | } | |
7251 | ||
1a20ff27 | 7252 | /* |
41a2d6cf | 7253 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7254 | * For now this just excludes isolated cpus, but could be used to |
7255 | * exclude other special cases in the future. | |
1a20ff27 | 7256 | */ |
96f874e2 | 7257 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7258 | { |
7378547f MM |
7259 | int err; |
7260 | ||
22e52b07 | 7261 | arch_update_cpu_topology(); |
029190c5 | 7262 | ndoms_cur = 1; |
acc3f5d7 | 7263 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7264 | if (!doms_cur) |
acc3f5d7 RR |
7265 | doms_cur = &fallback_doms; |
7266 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7267 | dattr_cur = NULL; |
acc3f5d7 | 7268 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7269 | register_sched_domain_sysctl(); |
7378547f MM |
7270 | |
7271 | return err; | |
1a20ff27 DG |
7272 | } |
7273 | ||
96f874e2 RR |
7274 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7275 | struct cpumask *tmpmask) | |
1da177e4 | 7276 | { |
7c16ec58 | 7277 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7278 | } |
1da177e4 | 7279 | |
1a20ff27 DG |
7280 | /* |
7281 | * Detach sched domains from a group of cpus specified in cpu_map | |
7282 | * These cpus will now be attached to the NULL domain | |
7283 | */ | |
96f874e2 | 7284 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7285 | { |
96f874e2 RR |
7286 | /* Save because hotplug lock held. */ |
7287 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7288 | int i; |
7289 | ||
abcd083a | 7290 | for_each_cpu(i, cpu_map) |
57d885fe | 7291 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7292 | synchronize_sched(); |
96f874e2 | 7293 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7294 | } |
7295 | ||
1d3504fc HS |
7296 | /* handle null as "default" */ |
7297 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7298 | struct sched_domain_attr *new, int idx_new) | |
7299 | { | |
7300 | struct sched_domain_attr tmp; | |
7301 | ||
7302 | /* fast path */ | |
7303 | if (!new && !cur) | |
7304 | return 1; | |
7305 | ||
7306 | tmp = SD_ATTR_INIT; | |
7307 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7308 | new ? (new + idx_new) : &tmp, | |
7309 | sizeof(struct sched_domain_attr)); | |
7310 | } | |
7311 | ||
029190c5 PJ |
7312 | /* |
7313 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7314 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7315 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7316 | * It destroys each deleted domain and builds each new domain. | |
7317 | * | |
acc3f5d7 | 7318 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7319 | * The masks don't intersect (don't overlap.) We should setup one |
7320 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7321 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7322 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7323 | * it as it is. | |
7324 | * | |
acc3f5d7 RR |
7325 | * The passed in 'doms_new' should be allocated using |
7326 | * alloc_sched_domains. This routine takes ownership of it and will | |
7327 | * free_sched_domains it when done with it. If the caller failed the | |
7328 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7329 | * and partition_sched_domains() will fallback to the single partition | |
7330 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7331 | * |
96f874e2 | 7332 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7333 | * ndoms_new == 0 is a special case for destroying existing domains, |
7334 | * and it will not create the default domain. | |
dfb512ec | 7335 | * |
029190c5 PJ |
7336 | * Call with hotplug lock held |
7337 | */ | |
acc3f5d7 | 7338 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7339 | struct sched_domain_attr *dattr_new) |
029190c5 | 7340 | { |
dfb512ec | 7341 | int i, j, n; |
d65bd5ec | 7342 | int new_topology; |
029190c5 | 7343 | |
712555ee | 7344 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7345 | |
7378547f MM |
7346 | /* always unregister in case we don't destroy any domains */ |
7347 | unregister_sched_domain_sysctl(); | |
7348 | ||
d65bd5ec HC |
7349 | /* Let architecture update cpu core mappings. */ |
7350 | new_topology = arch_update_cpu_topology(); | |
7351 | ||
dfb512ec | 7352 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7353 | |
7354 | /* Destroy deleted domains */ | |
7355 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7356 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7357 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7358 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7359 | goto match1; |
7360 | } | |
7361 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7362 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7363 | match1: |
7364 | ; | |
7365 | } | |
7366 | ||
e761b772 MK |
7367 | if (doms_new == NULL) { |
7368 | ndoms_cur = 0; | |
acc3f5d7 | 7369 | doms_new = &fallback_doms; |
6ad4c188 | 7370 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7371 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7372 | } |
7373 | ||
029190c5 PJ |
7374 | /* Build new domains */ |
7375 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7376 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7377 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7378 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7379 | goto match2; |
7380 | } | |
7381 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7382 | __build_sched_domains(doms_new[i], |
1d3504fc | 7383 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7384 | match2: |
7385 | ; | |
7386 | } | |
7387 | ||
7388 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7389 | if (doms_cur != &fallback_doms) |
7390 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7391 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7392 | doms_cur = doms_new; |
1d3504fc | 7393 | dattr_cur = dattr_new; |
029190c5 | 7394 | ndoms_cur = ndoms_new; |
7378547f MM |
7395 | |
7396 | register_sched_domain_sysctl(); | |
a1835615 | 7397 | |
712555ee | 7398 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7399 | } |
7400 | ||
5c45bf27 | 7401 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7402 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7403 | { |
95402b38 | 7404 | get_online_cpus(); |
dfb512ec MK |
7405 | |
7406 | /* Destroy domains first to force the rebuild */ | |
7407 | partition_sched_domains(0, NULL, NULL); | |
7408 | ||
e761b772 | 7409 | rebuild_sched_domains(); |
95402b38 | 7410 | put_online_cpus(); |
5c45bf27 SS |
7411 | } |
7412 | ||
7413 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7414 | { | |
afb8a9b7 | 7415 | unsigned int level = 0; |
5c45bf27 | 7416 | |
afb8a9b7 GS |
7417 | if (sscanf(buf, "%u", &level) != 1) |
7418 | return -EINVAL; | |
7419 | ||
7420 | /* | |
7421 | * level is always be positive so don't check for | |
7422 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7423 | * What happens on 0 or 1 byte write, | |
7424 | * need to check for count as well? | |
7425 | */ | |
7426 | ||
7427 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7428 | return -EINVAL; |
7429 | ||
7430 | if (smt) | |
afb8a9b7 | 7431 | sched_smt_power_savings = level; |
5c45bf27 | 7432 | else |
afb8a9b7 | 7433 | sched_mc_power_savings = level; |
5c45bf27 | 7434 | |
c70f22d2 | 7435 | arch_reinit_sched_domains(); |
5c45bf27 | 7436 | |
c70f22d2 | 7437 | return count; |
5c45bf27 SS |
7438 | } |
7439 | ||
5c45bf27 | 7440 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7441 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7442 | struct sysdev_class_attribute *attr, |
f718cd4a | 7443 | char *page) |
5c45bf27 SS |
7444 | { |
7445 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7446 | } | |
f718cd4a | 7447 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7448 | struct sysdev_class_attribute *attr, |
48f24c4d | 7449 | const char *buf, size_t count) |
5c45bf27 SS |
7450 | { |
7451 | return sched_power_savings_store(buf, count, 0); | |
7452 | } | |
f718cd4a AK |
7453 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7454 | sched_mc_power_savings_show, | |
7455 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7456 | #endif |
7457 | ||
7458 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7459 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7460 | struct sysdev_class_attribute *attr, |
f718cd4a | 7461 | char *page) |
5c45bf27 SS |
7462 | { |
7463 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7464 | } | |
f718cd4a | 7465 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7466 | struct sysdev_class_attribute *attr, |
48f24c4d | 7467 | const char *buf, size_t count) |
5c45bf27 SS |
7468 | { |
7469 | return sched_power_savings_store(buf, count, 1); | |
7470 | } | |
f718cd4a AK |
7471 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7472 | sched_smt_power_savings_show, | |
6707de00 AB |
7473 | sched_smt_power_savings_store); |
7474 | #endif | |
7475 | ||
39aac648 | 7476 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7477 | { |
7478 | int err = 0; | |
7479 | ||
7480 | #ifdef CONFIG_SCHED_SMT | |
7481 | if (smt_capable()) | |
7482 | err = sysfs_create_file(&cls->kset.kobj, | |
7483 | &attr_sched_smt_power_savings.attr); | |
7484 | #endif | |
7485 | #ifdef CONFIG_SCHED_MC | |
7486 | if (!err && mc_capable()) | |
7487 | err = sysfs_create_file(&cls->kset.kobj, | |
7488 | &attr_sched_mc_power_savings.attr); | |
7489 | #endif | |
7490 | return err; | |
7491 | } | |
6d6bc0ad | 7492 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7493 | |
1da177e4 | 7494 | /* |
3a101d05 TH |
7495 | * Update cpusets according to cpu_active mask. If cpusets are |
7496 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7497 | * around partition_sched_domains(). | |
1da177e4 | 7498 | */ |
0b2e918a TH |
7499 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7500 | void *hcpu) | |
e761b772 | 7501 | { |
3a101d05 | 7502 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7503 | case CPU_ONLINE: |
6ad4c188 | 7504 | case CPU_DOWN_FAILED: |
3a101d05 | 7505 | cpuset_update_active_cpus(); |
e761b772 | 7506 | return NOTIFY_OK; |
3a101d05 TH |
7507 | default: |
7508 | return NOTIFY_DONE; | |
7509 | } | |
7510 | } | |
e761b772 | 7511 | |
0b2e918a TH |
7512 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7513 | void *hcpu) | |
3a101d05 TH |
7514 | { |
7515 | switch (action & ~CPU_TASKS_FROZEN) { | |
7516 | case CPU_DOWN_PREPARE: | |
7517 | cpuset_update_active_cpus(); | |
7518 | return NOTIFY_OK; | |
e761b772 MK |
7519 | default: |
7520 | return NOTIFY_DONE; | |
7521 | } | |
7522 | } | |
e761b772 MK |
7523 | |
7524 | static int update_runtime(struct notifier_block *nfb, | |
7525 | unsigned long action, void *hcpu) | |
1da177e4 | 7526 | { |
7def2be1 PZ |
7527 | int cpu = (int)(long)hcpu; |
7528 | ||
1da177e4 | 7529 | switch (action) { |
1da177e4 | 7530 | case CPU_DOWN_PREPARE: |
8bb78442 | 7531 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7532 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7533 | return NOTIFY_OK; |
7534 | ||
1da177e4 | 7535 | case CPU_DOWN_FAILED: |
8bb78442 | 7536 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7537 | case CPU_ONLINE: |
8bb78442 | 7538 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7539 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7540 | return NOTIFY_OK; |
7541 | ||
1da177e4 LT |
7542 | default: |
7543 | return NOTIFY_DONE; | |
7544 | } | |
1da177e4 | 7545 | } |
1da177e4 LT |
7546 | |
7547 | void __init sched_init_smp(void) | |
7548 | { | |
dcc30a35 RR |
7549 | cpumask_var_t non_isolated_cpus; |
7550 | ||
7551 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7552 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7553 | |
434d53b0 MT |
7554 | #if defined(CONFIG_NUMA) |
7555 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7556 | GFP_KERNEL); | |
7557 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7558 | #endif | |
95402b38 | 7559 | get_online_cpus(); |
712555ee | 7560 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7561 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7562 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7563 | if (cpumask_empty(non_isolated_cpus)) | |
7564 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7565 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7566 | put_online_cpus(); |
e761b772 | 7567 | |
3a101d05 TH |
7568 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7569 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7570 | |
7571 | /* RT runtime code needs to handle some hotplug events */ | |
7572 | hotcpu_notifier(update_runtime, 0); | |
7573 | ||
b328ca18 | 7574 | init_hrtick(); |
5c1e1767 NP |
7575 | |
7576 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7577 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7578 | BUG(); |
19978ca6 | 7579 | sched_init_granularity(); |
dcc30a35 | 7580 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7581 | |
0e3900e6 | 7582 | init_sched_rt_class(); |
1da177e4 LT |
7583 | } |
7584 | #else | |
7585 | void __init sched_init_smp(void) | |
7586 | { | |
19978ca6 | 7587 | sched_init_granularity(); |
1da177e4 LT |
7588 | } |
7589 | #endif /* CONFIG_SMP */ | |
7590 | ||
cd1bb94b AB |
7591 | const_debug unsigned int sysctl_timer_migration = 1; |
7592 | ||
1da177e4 LT |
7593 | int in_sched_functions(unsigned long addr) |
7594 | { | |
1da177e4 LT |
7595 | return in_lock_functions(addr) || |
7596 | (addr >= (unsigned long)__sched_text_start | |
7597 | && addr < (unsigned long)__sched_text_end); | |
7598 | } | |
7599 | ||
a9957449 | 7600 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7601 | { |
7602 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7603 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7604 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7605 | cfs_rq->rq = rq; | |
7606 | #endif | |
67e9fb2a | 7607 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7608 | } |
7609 | ||
fa85ae24 PZ |
7610 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7611 | { | |
7612 | struct rt_prio_array *array; | |
7613 | int i; | |
7614 | ||
7615 | array = &rt_rq->active; | |
7616 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7617 | INIT_LIST_HEAD(array->queue + i); | |
7618 | __clear_bit(i, array->bitmap); | |
7619 | } | |
7620 | /* delimiter for bitsearch: */ | |
7621 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7622 | ||
052f1dc7 | 7623 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7624 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7625 | #ifdef CONFIG_SMP |
e864c499 | 7626 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7627 | #endif |
48d5e258 | 7628 | #endif |
fa85ae24 PZ |
7629 | #ifdef CONFIG_SMP |
7630 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7631 | rt_rq->overloaded = 0; |
05fa785c | 7632 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7633 | #endif |
7634 | ||
7635 | rt_rq->rt_time = 0; | |
7636 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7637 | rt_rq->rt_runtime = 0; |
0986b11b | 7638 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7639 | |
052f1dc7 | 7640 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7641 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7642 | rt_rq->rq = rq; |
7643 | #endif | |
fa85ae24 PZ |
7644 | } |
7645 | ||
6f505b16 | 7646 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 7647 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 7648 | struct sched_entity *se, int cpu, |
ec7dc8ac | 7649 | struct sched_entity *parent) |
6f505b16 | 7650 | { |
ec7dc8ac | 7651 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7652 | tg->cfs_rq[cpu] = cfs_rq; |
7653 | init_cfs_rq(cfs_rq, rq); | |
7654 | cfs_rq->tg = tg; | |
6f505b16 PZ |
7655 | |
7656 | tg->se[cpu] = se; | |
354d60c2 DG |
7657 | /* se could be NULL for init_task_group */ |
7658 | if (!se) | |
7659 | return; | |
7660 | ||
ec7dc8ac DG |
7661 | if (!parent) |
7662 | se->cfs_rq = &rq->cfs; | |
7663 | else | |
7664 | se->cfs_rq = parent->my_q; | |
7665 | ||
6f505b16 | 7666 | se->my_q = cfs_rq; |
9437178f | 7667 | update_load_set(&se->load, 0); |
ec7dc8ac | 7668 | se->parent = parent; |
6f505b16 | 7669 | } |
052f1dc7 | 7670 | #endif |
6f505b16 | 7671 | |
052f1dc7 | 7672 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 7673 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 7674 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 7675 | struct sched_rt_entity *parent) |
6f505b16 | 7676 | { |
ec7dc8ac DG |
7677 | struct rq *rq = cpu_rq(cpu); |
7678 | ||
6f505b16 PZ |
7679 | tg->rt_rq[cpu] = rt_rq; |
7680 | init_rt_rq(rt_rq, rq); | |
7681 | rt_rq->tg = tg; | |
ac086bc2 | 7682 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7683 | |
7684 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7685 | if (!rt_se) |
7686 | return; | |
7687 | ||
ec7dc8ac DG |
7688 | if (!parent) |
7689 | rt_se->rt_rq = &rq->rt; | |
7690 | else | |
7691 | rt_se->rt_rq = parent->my_q; | |
7692 | ||
6f505b16 | 7693 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7694 | rt_se->parent = parent; |
6f505b16 PZ |
7695 | INIT_LIST_HEAD(&rt_se->run_list); |
7696 | } | |
7697 | #endif | |
7698 | ||
1da177e4 LT |
7699 | void __init sched_init(void) |
7700 | { | |
dd41f596 | 7701 | int i, j; |
434d53b0 MT |
7702 | unsigned long alloc_size = 0, ptr; |
7703 | ||
7704 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7705 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7706 | #endif | |
7707 | #ifdef CONFIG_RT_GROUP_SCHED | |
7708 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7709 | #endif |
df7c8e84 | 7710 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7711 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7712 | #endif |
434d53b0 | 7713 | if (alloc_size) { |
36b7b6d4 | 7714 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7715 | |
7716 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7717 | init_task_group.se = (struct sched_entity **)ptr; | |
7718 | ptr += nr_cpu_ids * sizeof(void **); | |
7719 | ||
7720 | init_task_group.cfs_rq = (struct cfs_rq **)ptr; | |
7721 | ptr += nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7722 | |
6d6bc0ad | 7723 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 MT |
7724 | #ifdef CONFIG_RT_GROUP_SCHED |
7725 | init_task_group.rt_se = (struct sched_rt_entity **)ptr; | |
7726 | ptr += nr_cpu_ids * sizeof(void **); | |
7727 | ||
7728 | init_task_group.rt_rq = (struct rt_rq **)ptr; | |
eff766a6 PZ |
7729 | ptr += nr_cpu_ids * sizeof(void **); |
7730 | ||
6d6bc0ad | 7731 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7732 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7733 | for_each_possible_cpu(i) { | |
7734 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7735 | ptr += cpumask_size(); | |
7736 | } | |
7737 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7738 | } |
dd41f596 | 7739 | |
57d885fe GH |
7740 | #ifdef CONFIG_SMP |
7741 | init_defrootdomain(); | |
7742 | #endif | |
7743 | ||
d0b27fa7 PZ |
7744 | init_rt_bandwidth(&def_rt_bandwidth, |
7745 | global_rt_period(), global_rt_runtime()); | |
7746 | ||
7747 | #ifdef CONFIG_RT_GROUP_SCHED | |
7748 | init_rt_bandwidth(&init_task_group.rt_bandwidth, | |
7749 | global_rt_period(), global_rt_runtime()); | |
6d6bc0ad | 7750 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7751 | |
7c941438 | 7752 | #ifdef CONFIG_CGROUP_SCHED |
6f505b16 | 7753 | list_add(&init_task_group.list, &task_groups); |
f473aa5e PZ |
7754 | INIT_LIST_HEAD(&init_task_group.children); |
7755 | ||
7c941438 | 7756 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7757 | |
0a945022 | 7758 | for_each_possible_cpu(i) { |
70b97a7f | 7759 | struct rq *rq; |
1da177e4 LT |
7760 | |
7761 | rq = cpu_rq(i); | |
05fa785c | 7762 | raw_spin_lock_init(&rq->lock); |
7897986b | 7763 | rq->nr_running = 0; |
dce48a84 TG |
7764 | rq->calc_load_active = 0; |
7765 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7766 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7767 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7768 | #ifdef CONFIG_FAIR_GROUP_SCHED |
4cf86d77 | 7769 | init_task_group.shares = init_task_group_load; |
6f505b16 | 7770 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 DG |
7771 | #ifdef CONFIG_CGROUP_SCHED |
7772 | /* | |
7773 | * How much cpu bandwidth does init_task_group get? | |
7774 | * | |
7775 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7776 | * gets 100% of the cpu resources in the system. This overall | |
7777 | * system cpu resource is divided among the tasks of | |
7778 | * init_task_group and its child task-groups in a fair manner, | |
7779 | * based on each entity's (task or task-group's) weight | |
7780 | * (se->load.weight). | |
7781 | * | |
7782 | * In other words, if init_task_group has 10 tasks of weight | |
7783 | * 1024) and two child groups A0 and A1 (of weight 1024 each), | |
7784 | * then A0's share of the cpu resource is: | |
7785 | * | |
0d905bca | 7786 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 DG |
7787 | * |
7788 | * We achieve this by letting init_task_group's tasks sit | |
7789 | * directly in rq->cfs (i.e init_task_group->se[] = NULL). | |
7790 | */ | |
3d4b47b4 | 7791 | init_tg_cfs_entry(&init_task_group, &rq->cfs, NULL, i, NULL); |
052f1dc7 | 7792 | #endif |
354d60c2 DG |
7793 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7794 | ||
7795 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7796 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7797 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
354d60c2 | 7798 | #ifdef CONFIG_CGROUP_SCHED |
3d4b47b4 | 7799 | init_tg_rt_entry(&init_task_group, &rq->rt, NULL, i, NULL); |
354d60c2 | 7800 | #endif |
dd41f596 | 7801 | #endif |
1da177e4 | 7802 | |
dd41f596 IM |
7803 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7804 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7805 | |
7806 | rq->last_load_update_tick = jiffies; | |
7807 | ||
1da177e4 | 7808 | #ifdef CONFIG_SMP |
41c7ce9a | 7809 | rq->sd = NULL; |
57d885fe | 7810 | rq->rd = NULL; |
e51fd5e2 | 7811 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 7812 | rq->post_schedule = 0; |
1da177e4 | 7813 | rq->active_balance = 0; |
dd41f596 | 7814 | rq->next_balance = jiffies; |
1da177e4 | 7815 | rq->push_cpu = 0; |
0a2966b4 | 7816 | rq->cpu = i; |
1f11eb6a | 7817 | rq->online = 0; |
eae0c9df MG |
7818 | rq->idle_stamp = 0; |
7819 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7820 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
7821 | #ifdef CONFIG_NO_HZ |
7822 | rq->nohz_balance_kick = 0; | |
7823 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
7824 | #endif | |
1da177e4 | 7825 | #endif |
8f4d37ec | 7826 | init_rq_hrtick(rq); |
1da177e4 | 7827 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7828 | } |
7829 | ||
2dd73a4f | 7830 | set_load_weight(&init_task); |
b50f60ce | 7831 | |
e107be36 AK |
7832 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
7833 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
7834 | #endif | |
7835 | ||
c9819f45 | 7836 | #ifdef CONFIG_SMP |
962cf36c | 7837 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
7838 | #endif |
7839 | ||
b50f60ce | 7840 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 7841 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
7842 | #endif |
7843 | ||
1da177e4 LT |
7844 | /* |
7845 | * The boot idle thread does lazy MMU switching as well: | |
7846 | */ | |
7847 | atomic_inc(&init_mm.mm_count); | |
7848 | enter_lazy_tlb(&init_mm, current); | |
7849 | ||
7850 | /* | |
7851 | * Make us the idle thread. Technically, schedule() should not be | |
7852 | * called from this thread, however somewhere below it might be, | |
7853 | * but because we are the idle thread, we just pick up running again | |
7854 | * when this runqueue becomes "idle". | |
7855 | */ | |
7856 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
7857 | |
7858 | calc_load_update = jiffies + LOAD_FREQ; | |
7859 | ||
dd41f596 IM |
7860 | /* |
7861 | * During early bootup we pretend to be a normal task: | |
7862 | */ | |
7863 | current->sched_class = &fair_sched_class; | |
6892b75e | 7864 | |
6a7b3dc3 | 7865 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 7866 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 7867 | #ifdef CONFIG_SMP |
7d1e6a9b | 7868 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
7869 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
7870 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
7871 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
7872 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
7873 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 7874 | #endif |
bdddd296 RR |
7875 | /* May be allocated at isolcpus cmdline parse time */ |
7876 | if (cpu_isolated_map == NULL) | |
7877 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 7878 | #endif /* SMP */ |
6a7b3dc3 | 7879 | |
cdd6c482 | 7880 | perf_event_init(); |
0d905bca | 7881 | |
6892b75e | 7882 | scheduler_running = 1; |
1da177e4 LT |
7883 | } |
7884 | ||
7885 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
7886 | static inline int preempt_count_equals(int preempt_offset) |
7887 | { | |
234da7bc | 7888 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
7889 | |
7890 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
7891 | } | |
7892 | ||
d894837f | 7893 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 7894 | { |
48f24c4d | 7895 | #ifdef in_atomic |
1da177e4 LT |
7896 | static unsigned long prev_jiffy; /* ratelimiting */ |
7897 | ||
e4aafea2 FW |
7898 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
7899 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
7900 | return; |
7901 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
7902 | return; | |
7903 | prev_jiffy = jiffies; | |
7904 | ||
3df0fc5b PZ |
7905 | printk(KERN_ERR |
7906 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
7907 | file, line); | |
7908 | printk(KERN_ERR | |
7909 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
7910 | in_atomic(), irqs_disabled(), | |
7911 | current->pid, current->comm); | |
aef745fc IM |
7912 | |
7913 | debug_show_held_locks(current); | |
7914 | if (irqs_disabled()) | |
7915 | print_irqtrace_events(current); | |
7916 | dump_stack(); | |
1da177e4 LT |
7917 | #endif |
7918 | } | |
7919 | EXPORT_SYMBOL(__might_sleep); | |
7920 | #endif | |
7921 | ||
7922 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
7923 | static void normalize_task(struct rq *rq, struct task_struct *p) |
7924 | { | |
7925 | int on_rq; | |
3e51f33f | 7926 | |
3a5e4dc1 AK |
7927 | on_rq = p->se.on_rq; |
7928 | if (on_rq) | |
7929 | deactivate_task(rq, p, 0); | |
7930 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
7931 | if (on_rq) { | |
7932 | activate_task(rq, p, 0); | |
7933 | resched_task(rq->curr); | |
7934 | } | |
7935 | } | |
7936 | ||
1da177e4 LT |
7937 | void normalize_rt_tasks(void) |
7938 | { | |
a0f98a1c | 7939 | struct task_struct *g, *p; |
1da177e4 | 7940 | unsigned long flags; |
70b97a7f | 7941 | struct rq *rq; |
1da177e4 | 7942 | |
4cf5d77a | 7943 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 7944 | do_each_thread(g, p) { |
178be793 IM |
7945 | /* |
7946 | * Only normalize user tasks: | |
7947 | */ | |
7948 | if (!p->mm) | |
7949 | continue; | |
7950 | ||
6cfb0d5d | 7951 | p->se.exec_start = 0; |
6cfb0d5d | 7952 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
7953 | p->se.statistics.wait_start = 0; |
7954 | p->se.statistics.sleep_start = 0; | |
7955 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 7956 | #endif |
dd41f596 IM |
7957 | |
7958 | if (!rt_task(p)) { | |
7959 | /* | |
7960 | * Renice negative nice level userspace | |
7961 | * tasks back to 0: | |
7962 | */ | |
7963 | if (TASK_NICE(p) < 0 && p->mm) | |
7964 | set_user_nice(p, 0); | |
1da177e4 | 7965 | continue; |
dd41f596 | 7966 | } |
1da177e4 | 7967 | |
1d615482 | 7968 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 7969 | rq = __task_rq_lock(p); |
1da177e4 | 7970 | |
178be793 | 7971 | normalize_task(rq, p); |
3a5e4dc1 | 7972 | |
b29739f9 | 7973 | __task_rq_unlock(rq); |
1d615482 | 7974 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
7975 | } while_each_thread(g, p); |
7976 | ||
4cf5d77a | 7977 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
7978 | } |
7979 | ||
7980 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 7981 | |
67fc4e0c | 7982 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 7983 | /* |
67fc4e0c | 7984 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
7985 | * |
7986 | * They can only be called when the whole system has been | |
7987 | * stopped - every CPU needs to be quiescent, and no scheduling | |
7988 | * activity can take place. Using them for anything else would | |
7989 | * be a serious bug, and as a result, they aren't even visible | |
7990 | * under any other configuration. | |
7991 | */ | |
7992 | ||
7993 | /** | |
7994 | * curr_task - return the current task for a given cpu. | |
7995 | * @cpu: the processor in question. | |
7996 | * | |
7997 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
7998 | */ | |
36c8b586 | 7999 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8000 | { |
8001 | return cpu_curr(cpu); | |
8002 | } | |
8003 | ||
67fc4e0c JW |
8004 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8005 | ||
8006 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8007 | /** |
8008 | * set_curr_task - set the current task for a given cpu. | |
8009 | * @cpu: the processor in question. | |
8010 | * @p: the task pointer to set. | |
8011 | * | |
8012 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8013 | * are serviced on a separate stack. It allows the architecture to switch the |
8014 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8015 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8016 | * and caller must save the original value of the current task (see | |
8017 | * curr_task() above) and restore that value before reenabling interrupts and | |
8018 | * re-starting the system. | |
8019 | * | |
8020 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8021 | */ | |
36c8b586 | 8022 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8023 | { |
8024 | cpu_curr(cpu) = p; | |
8025 | } | |
8026 | ||
8027 | #endif | |
29f59db3 | 8028 | |
bccbe08a PZ |
8029 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8030 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8031 | { |
8032 | int i; | |
8033 | ||
8034 | for_each_possible_cpu(i) { | |
8035 | if (tg->cfs_rq) | |
8036 | kfree(tg->cfs_rq[i]); | |
8037 | if (tg->se) | |
8038 | kfree(tg->se[i]); | |
6f505b16 PZ |
8039 | } |
8040 | ||
8041 | kfree(tg->cfs_rq); | |
8042 | kfree(tg->se); | |
6f505b16 PZ |
8043 | } |
8044 | ||
ec7dc8ac DG |
8045 | static |
8046 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8047 | { |
29f59db3 | 8048 | struct cfs_rq *cfs_rq; |
eab17229 | 8049 | struct sched_entity *se; |
9b5b7751 | 8050 | struct rq *rq; |
29f59db3 SV |
8051 | int i; |
8052 | ||
434d53b0 | 8053 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8054 | if (!tg->cfs_rq) |
8055 | goto err; | |
434d53b0 | 8056 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8057 | if (!tg->se) |
8058 | goto err; | |
052f1dc7 PZ |
8059 | |
8060 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8061 | |
8062 | for_each_possible_cpu(i) { | |
9b5b7751 | 8063 | rq = cpu_rq(i); |
29f59db3 | 8064 | |
eab17229 LZ |
8065 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8066 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8067 | if (!cfs_rq) |
8068 | goto err; | |
8069 | ||
eab17229 LZ |
8070 | se = kzalloc_node(sizeof(struct sched_entity), |
8071 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8072 | if (!se) |
dfc12eb2 | 8073 | goto err_free_rq; |
29f59db3 | 8074 | |
3d4b47b4 | 8075 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8076 | } |
8077 | ||
8078 | return 1; | |
8079 | ||
49246274 | 8080 | err_free_rq: |
dfc12eb2 | 8081 | kfree(cfs_rq); |
49246274 | 8082 | err: |
bccbe08a PZ |
8083 | return 0; |
8084 | } | |
8085 | ||
bccbe08a PZ |
8086 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8087 | { | |
3d4b47b4 PZ |
8088 | struct rq *rq = cpu_rq(cpu); |
8089 | unsigned long flags; | |
8090 | int i; | |
8091 | ||
8092 | /* | |
8093 | * Only empty task groups can be destroyed; so we can speculatively | |
8094 | * check on_list without danger of it being re-added. | |
8095 | */ | |
8096 | if (!tg->cfs_rq[cpu]->on_list) | |
8097 | return; | |
8098 | ||
8099 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8100 | list_del_leaf_cfs_rq(tg->cfs_rq[i]); | |
8101 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
bccbe08a | 8102 | } |
6d6bc0ad | 8103 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8104 | static inline void free_fair_sched_group(struct task_group *tg) |
8105 | { | |
8106 | } | |
8107 | ||
ec7dc8ac DG |
8108 | static inline |
8109 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8110 | { |
8111 | return 1; | |
8112 | } | |
8113 | ||
bccbe08a PZ |
8114 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8115 | { | |
8116 | } | |
6d6bc0ad | 8117 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8118 | |
8119 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8120 | static void free_rt_sched_group(struct task_group *tg) |
8121 | { | |
8122 | int i; | |
8123 | ||
d0b27fa7 PZ |
8124 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8125 | ||
bccbe08a PZ |
8126 | for_each_possible_cpu(i) { |
8127 | if (tg->rt_rq) | |
8128 | kfree(tg->rt_rq[i]); | |
8129 | if (tg->rt_se) | |
8130 | kfree(tg->rt_se[i]); | |
8131 | } | |
8132 | ||
8133 | kfree(tg->rt_rq); | |
8134 | kfree(tg->rt_se); | |
8135 | } | |
8136 | ||
ec7dc8ac DG |
8137 | static |
8138 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8139 | { |
8140 | struct rt_rq *rt_rq; | |
eab17229 | 8141 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8142 | struct rq *rq; |
8143 | int i; | |
8144 | ||
434d53b0 | 8145 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8146 | if (!tg->rt_rq) |
8147 | goto err; | |
434d53b0 | 8148 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8149 | if (!tg->rt_se) |
8150 | goto err; | |
8151 | ||
d0b27fa7 PZ |
8152 | init_rt_bandwidth(&tg->rt_bandwidth, |
8153 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8154 | |
8155 | for_each_possible_cpu(i) { | |
8156 | rq = cpu_rq(i); | |
8157 | ||
eab17229 LZ |
8158 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8159 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8160 | if (!rt_rq) |
8161 | goto err; | |
29f59db3 | 8162 | |
eab17229 LZ |
8163 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8164 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8165 | if (!rt_se) |
dfc12eb2 | 8166 | goto err_free_rq; |
29f59db3 | 8167 | |
3d4b47b4 | 8168 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8169 | } |
8170 | ||
bccbe08a PZ |
8171 | return 1; |
8172 | ||
49246274 | 8173 | err_free_rq: |
dfc12eb2 | 8174 | kfree(rt_rq); |
49246274 | 8175 | err: |
bccbe08a PZ |
8176 | return 0; |
8177 | } | |
6d6bc0ad | 8178 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8179 | static inline void free_rt_sched_group(struct task_group *tg) |
8180 | { | |
8181 | } | |
8182 | ||
ec7dc8ac DG |
8183 | static inline |
8184 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8185 | { |
8186 | return 1; | |
8187 | } | |
6d6bc0ad | 8188 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8189 | |
7c941438 | 8190 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8191 | static void free_sched_group(struct task_group *tg) |
8192 | { | |
8193 | free_fair_sched_group(tg); | |
8194 | free_rt_sched_group(tg); | |
8195 | kfree(tg); | |
8196 | } | |
8197 | ||
8198 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8199 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8200 | { |
8201 | struct task_group *tg; | |
8202 | unsigned long flags; | |
bccbe08a PZ |
8203 | |
8204 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8205 | if (!tg) | |
8206 | return ERR_PTR(-ENOMEM); | |
8207 | ||
ec7dc8ac | 8208 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8209 | goto err; |
8210 | ||
ec7dc8ac | 8211 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8212 | goto err; |
8213 | ||
8ed36996 | 8214 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8215 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8216 | |
8217 | WARN_ON(!parent); /* root should already exist */ | |
8218 | ||
8219 | tg->parent = parent; | |
f473aa5e | 8220 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8221 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8222 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8223 | |
9b5b7751 | 8224 | return tg; |
29f59db3 SV |
8225 | |
8226 | err: | |
6f505b16 | 8227 | free_sched_group(tg); |
29f59db3 SV |
8228 | return ERR_PTR(-ENOMEM); |
8229 | } | |
8230 | ||
9b5b7751 | 8231 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8232 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8233 | { |
29f59db3 | 8234 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8235 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8236 | } |
8237 | ||
9b5b7751 | 8238 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8239 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8240 | { |
8ed36996 | 8241 | unsigned long flags; |
9b5b7751 | 8242 | int i; |
29f59db3 | 8243 | |
3d4b47b4 PZ |
8244 | /* end participation in shares distribution */ |
8245 | for_each_possible_cpu(i) | |
bccbe08a | 8246 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8247 | |
8248 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8249 | list_del_rcu(&tg->list); |
f473aa5e | 8250 | list_del_rcu(&tg->siblings); |
8ed36996 | 8251 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8252 | |
9b5b7751 | 8253 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8254 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8255 | } |
8256 | ||
9b5b7751 | 8257 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8258 | * The caller of this function should have put the task in its new group |
8259 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8260 | * reflect its new group. | |
9b5b7751 SV |
8261 | */ |
8262 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8263 | { |
8264 | int on_rq, running; | |
8265 | unsigned long flags; | |
8266 | struct rq *rq; | |
8267 | ||
8268 | rq = task_rq_lock(tsk, &flags); | |
8269 | ||
051a1d1a | 8270 | running = task_current(rq, tsk); |
29f59db3 SV |
8271 | on_rq = tsk->se.on_rq; |
8272 | ||
0e1f3483 | 8273 | if (on_rq) |
29f59db3 | 8274 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8275 | if (unlikely(running)) |
8276 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8277 | |
810b3817 | 8278 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8279 | if (tsk->sched_class->task_move_group) |
8280 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8281 | else | |
810b3817 | 8282 | #endif |
b2b5ce02 | 8283 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8284 | |
0e1f3483 HS |
8285 | if (unlikely(running)) |
8286 | tsk->sched_class->set_curr_task(rq); | |
8287 | if (on_rq) | |
371fd7e7 | 8288 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8289 | |
29f59db3 SV |
8290 | task_rq_unlock(rq, &flags); |
8291 | } | |
7c941438 | 8292 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8293 | |
052f1dc7 | 8294 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8295 | static DEFINE_MUTEX(shares_mutex); |
8296 | ||
4cf86d77 | 8297 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8298 | { |
8299 | int i; | |
9437178f | 8300 | unsigned long flags; |
c61935fd | 8301 | |
ec7dc8ac DG |
8302 | /* |
8303 | * We can't change the weight of the root cgroup. | |
8304 | */ | |
8305 | if (!tg->se[0]) | |
8306 | return -EINVAL; | |
8307 | ||
18d95a28 PZ |
8308 | if (shares < MIN_SHARES) |
8309 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8310 | else if (shares > MAX_SHARES) |
8311 | shares = MAX_SHARES; | |
62fb1851 | 8312 | |
8ed36996 | 8313 | mutex_lock(&shares_mutex); |
9b5b7751 | 8314 | if (tg->shares == shares) |
5cb350ba | 8315 | goto done; |
29f59db3 | 8316 | |
9b5b7751 | 8317 | tg->shares = shares; |
c09595f6 | 8318 | for_each_possible_cpu(i) { |
9437178f PT |
8319 | struct rq *rq = cpu_rq(i); |
8320 | struct sched_entity *se; | |
8321 | ||
8322 | se = tg->se[i]; | |
8323 | /* Propagate contribution to hierarchy */ | |
8324 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8325 | for_each_sched_entity(se) | |
8326 | update_cfs_shares(group_cfs_rq(se), 0); | |
8327 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
c09595f6 | 8328 | } |
29f59db3 | 8329 | |
5cb350ba | 8330 | done: |
8ed36996 | 8331 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8332 | return 0; |
29f59db3 SV |
8333 | } |
8334 | ||
5cb350ba DG |
8335 | unsigned long sched_group_shares(struct task_group *tg) |
8336 | { | |
8337 | return tg->shares; | |
8338 | } | |
052f1dc7 | 8339 | #endif |
5cb350ba | 8340 | |
052f1dc7 | 8341 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8342 | /* |
9f0c1e56 | 8343 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8344 | */ |
9f0c1e56 PZ |
8345 | static DEFINE_MUTEX(rt_constraints_mutex); |
8346 | ||
8347 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8348 | { | |
8349 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8350 | return 1ULL << 20; |
9f0c1e56 | 8351 | |
9a7e0b18 | 8352 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8353 | } |
8354 | ||
9a7e0b18 PZ |
8355 | /* Must be called with tasklist_lock held */ |
8356 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8357 | { |
9a7e0b18 | 8358 | struct task_struct *g, *p; |
b40b2e8e | 8359 | |
9a7e0b18 PZ |
8360 | do_each_thread(g, p) { |
8361 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8362 | return 1; | |
8363 | } while_each_thread(g, p); | |
b40b2e8e | 8364 | |
9a7e0b18 PZ |
8365 | return 0; |
8366 | } | |
b40b2e8e | 8367 | |
9a7e0b18 PZ |
8368 | struct rt_schedulable_data { |
8369 | struct task_group *tg; | |
8370 | u64 rt_period; | |
8371 | u64 rt_runtime; | |
8372 | }; | |
b40b2e8e | 8373 | |
9a7e0b18 PZ |
8374 | static int tg_schedulable(struct task_group *tg, void *data) |
8375 | { | |
8376 | struct rt_schedulable_data *d = data; | |
8377 | struct task_group *child; | |
8378 | unsigned long total, sum = 0; | |
8379 | u64 period, runtime; | |
b40b2e8e | 8380 | |
9a7e0b18 PZ |
8381 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8382 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8383 | |
9a7e0b18 PZ |
8384 | if (tg == d->tg) { |
8385 | period = d->rt_period; | |
8386 | runtime = d->rt_runtime; | |
b40b2e8e | 8387 | } |
b40b2e8e | 8388 | |
4653f803 PZ |
8389 | /* |
8390 | * Cannot have more runtime than the period. | |
8391 | */ | |
8392 | if (runtime > period && runtime != RUNTIME_INF) | |
8393 | return -EINVAL; | |
6f505b16 | 8394 | |
4653f803 PZ |
8395 | /* |
8396 | * Ensure we don't starve existing RT tasks. | |
8397 | */ | |
9a7e0b18 PZ |
8398 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8399 | return -EBUSY; | |
6f505b16 | 8400 | |
9a7e0b18 | 8401 | total = to_ratio(period, runtime); |
6f505b16 | 8402 | |
4653f803 PZ |
8403 | /* |
8404 | * Nobody can have more than the global setting allows. | |
8405 | */ | |
8406 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8407 | return -EINVAL; | |
6f505b16 | 8408 | |
4653f803 PZ |
8409 | /* |
8410 | * The sum of our children's runtime should not exceed our own. | |
8411 | */ | |
9a7e0b18 PZ |
8412 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8413 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8414 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8415 | |
9a7e0b18 PZ |
8416 | if (child == d->tg) { |
8417 | period = d->rt_period; | |
8418 | runtime = d->rt_runtime; | |
8419 | } | |
6f505b16 | 8420 | |
9a7e0b18 | 8421 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8422 | } |
6f505b16 | 8423 | |
9a7e0b18 PZ |
8424 | if (sum > total) |
8425 | return -EINVAL; | |
8426 | ||
8427 | return 0; | |
6f505b16 PZ |
8428 | } |
8429 | ||
9a7e0b18 | 8430 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8431 | { |
9a7e0b18 PZ |
8432 | struct rt_schedulable_data data = { |
8433 | .tg = tg, | |
8434 | .rt_period = period, | |
8435 | .rt_runtime = runtime, | |
8436 | }; | |
8437 | ||
8438 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8439 | } |
8440 | ||
d0b27fa7 PZ |
8441 | static int tg_set_bandwidth(struct task_group *tg, |
8442 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8443 | { |
ac086bc2 | 8444 | int i, err = 0; |
9f0c1e56 | 8445 | |
9f0c1e56 | 8446 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8447 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8448 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8449 | if (err) | |
9f0c1e56 | 8450 | goto unlock; |
ac086bc2 | 8451 | |
0986b11b | 8452 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8453 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8454 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8455 | |
8456 | for_each_possible_cpu(i) { | |
8457 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8458 | ||
0986b11b | 8459 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8460 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8461 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8462 | } |
0986b11b | 8463 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8464 | unlock: |
521f1a24 | 8465 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8466 | mutex_unlock(&rt_constraints_mutex); |
8467 | ||
8468 | return err; | |
6f505b16 PZ |
8469 | } |
8470 | ||
d0b27fa7 PZ |
8471 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8472 | { | |
8473 | u64 rt_runtime, rt_period; | |
8474 | ||
8475 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8476 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8477 | if (rt_runtime_us < 0) | |
8478 | rt_runtime = RUNTIME_INF; | |
8479 | ||
8480 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8481 | } | |
8482 | ||
9f0c1e56 PZ |
8483 | long sched_group_rt_runtime(struct task_group *tg) |
8484 | { | |
8485 | u64 rt_runtime_us; | |
8486 | ||
d0b27fa7 | 8487 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8488 | return -1; |
8489 | ||
d0b27fa7 | 8490 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8491 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8492 | return rt_runtime_us; | |
8493 | } | |
d0b27fa7 PZ |
8494 | |
8495 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8496 | { | |
8497 | u64 rt_runtime, rt_period; | |
8498 | ||
8499 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8500 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8501 | ||
619b0488 R |
8502 | if (rt_period == 0) |
8503 | return -EINVAL; | |
8504 | ||
d0b27fa7 PZ |
8505 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8506 | } | |
8507 | ||
8508 | long sched_group_rt_period(struct task_group *tg) | |
8509 | { | |
8510 | u64 rt_period_us; | |
8511 | ||
8512 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8513 | do_div(rt_period_us, NSEC_PER_USEC); | |
8514 | return rt_period_us; | |
8515 | } | |
8516 | ||
8517 | static int sched_rt_global_constraints(void) | |
8518 | { | |
4653f803 | 8519 | u64 runtime, period; |
d0b27fa7 PZ |
8520 | int ret = 0; |
8521 | ||
ec5d4989 HS |
8522 | if (sysctl_sched_rt_period <= 0) |
8523 | return -EINVAL; | |
8524 | ||
4653f803 PZ |
8525 | runtime = global_rt_runtime(); |
8526 | period = global_rt_period(); | |
8527 | ||
8528 | /* | |
8529 | * Sanity check on the sysctl variables. | |
8530 | */ | |
8531 | if (runtime > period && runtime != RUNTIME_INF) | |
8532 | return -EINVAL; | |
10b612f4 | 8533 | |
d0b27fa7 | 8534 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8535 | read_lock(&tasklist_lock); |
4653f803 | 8536 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8537 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8538 | mutex_unlock(&rt_constraints_mutex); |
8539 | ||
8540 | return ret; | |
8541 | } | |
54e99124 DG |
8542 | |
8543 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8544 | { | |
8545 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8546 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8547 | return 0; | |
8548 | ||
8549 | return 1; | |
8550 | } | |
8551 | ||
6d6bc0ad | 8552 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8553 | static int sched_rt_global_constraints(void) |
8554 | { | |
ac086bc2 PZ |
8555 | unsigned long flags; |
8556 | int i; | |
8557 | ||
ec5d4989 HS |
8558 | if (sysctl_sched_rt_period <= 0) |
8559 | return -EINVAL; | |
8560 | ||
60aa605d PZ |
8561 | /* |
8562 | * There's always some RT tasks in the root group | |
8563 | * -- migration, kstopmachine etc.. | |
8564 | */ | |
8565 | if (sysctl_sched_rt_runtime == 0) | |
8566 | return -EBUSY; | |
8567 | ||
0986b11b | 8568 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8569 | for_each_possible_cpu(i) { |
8570 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8571 | ||
0986b11b | 8572 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8573 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8574 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8575 | } |
0986b11b | 8576 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8577 | |
d0b27fa7 PZ |
8578 | return 0; |
8579 | } | |
6d6bc0ad | 8580 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8581 | |
8582 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8583 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8584 | loff_t *ppos) |
8585 | { | |
8586 | int ret; | |
8587 | int old_period, old_runtime; | |
8588 | static DEFINE_MUTEX(mutex); | |
8589 | ||
8590 | mutex_lock(&mutex); | |
8591 | old_period = sysctl_sched_rt_period; | |
8592 | old_runtime = sysctl_sched_rt_runtime; | |
8593 | ||
8d65af78 | 8594 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8595 | |
8596 | if (!ret && write) { | |
8597 | ret = sched_rt_global_constraints(); | |
8598 | if (ret) { | |
8599 | sysctl_sched_rt_period = old_period; | |
8600 | sysctl_sched_rt_runtime = old_runtime; | |
8601 | } else { | |
8602 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8603 | def_rt_bandwidth.rt_period = | |
8604 | ns_to_ktime(global_rt_period()); | |
8605 | } | |
8606 | } | |
8607 | mutex_unlock(&mutex); | |
8608 | ||
8609 | return ret; | |
8610 | } | |
68318b8e | 8611 | |
052f1dc7 | 8612 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8613 | |
8614 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8615 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8616 | { |
2b01dfe3 PM |
8617 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8618 | struct task_group, css); | |
68318b8e SV |
8619 | } |
8620 | ||
8621 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8622 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8623 | { |
ec7dc8ac | 8624 | struct task_group *tg, *parent; |
68318b8e | 8625 | |
2b01dfe3 | 8626 | if (!cgrp->parent) { |
68318b8e | 8627 | /* This is early initialization for the top cgroup */ |
68318b8e SV |
8628 | return &init_task_group.css; |
8629 | } | |
8630 | ||
ec7dc8ac DG |
8631 | parent = cgroup_tg(cgrp->parent); |
8632 | tg = sched_create_group(parent); | |
68318b8e SV |
8633 | if (IS_ERR(tg)) |
8634 | return ERR_PTR(-ENOMEM); | |
8635 | ||
68318b8e SV |
8636 | return &tg->css; |
8637 | } | |
8638 | ||
41a2d6cf IM |
8639 | static void |
8640 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8641 | { |
2b01dfe3 | 8642 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8643 | |
8644 | sched_destroy_group(tg); | |
8645 | } | |
8646 | ||
41a2d6cf | 8647 | static int |
be367d09 | 8648 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8649 | { |
b68aa230 | 8650 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8651 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8652 | return -EINVAL; |
8653 | #else | |
68318b8e SV |
8654 | /* We don't support RT-tasks being in separate groups */ |
8655 | if (tsk->sched_class != &fair_sched_class) | |
8656 | return -EINVAL; | |
b68aa230 | 8657 | #endif |
be367d09 BB |
8658 | return 0; |
8659 | } | |
68318b8e | 8660 | |
be367d09 BB |
8661 | static int |
8662 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8663 | struct task_struct *tsk, bool threadgroup) | |
8664 | { | |
8665 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8666 | if (retval) | |
8667 | return retval; | |
8668 | if (threadgroup) { | |
8669 | struct task_struct *c; | |
8670 | rcu_read_lock(); | |
8671 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8672 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8673 | if (retval) { | |
8674 | rcu_read_unlock(); | |
8675 | return retval; | |
8676 | } | |
8677 | } | |
8678 | rcu_read_unlock(); | |
8679 | } | |
68318b8e SV |
8680 | return 0; |
8681 | } | |
8682 | ||
8683 | static void | |
2b01dfe3 | 8684 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8685 | struct cgroup *old_cont, struct task_struct *tsk, |
8686 | bool threadgroup) | |
68318b8e SV |
8687 | { |
8688 | sched_move_task(tsk); | |
be367d09 BB |
8689 | if (threadgroup) { |
8690 | struct task_struct *c; | |
8691 | rcu_read_lock(); | |
8692 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8693 | sched_move_task(c); | |
8694 | } | |
8695 | rcu_read_unlock(); | |
8696 | } | |
68318b8e SV |
8697 | } |
8698 | ||
052f1dc7 | 8699 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8700 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8701 | u64 shareval) |
68318b8e | 8702 | { |
2b01dfe3 | 8703 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8704 | } |
8705 | ||
f4c753b7 | 8706 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8707 | { |
2b01dfe3 | 8708 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8709 | |
8710 | return (u64) tg->shares; | |
8711 | } | |
6d6bc0ad | 8712 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8713 | |
052f1dc7 | 8714 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8715 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8716 | s64 val) |
6f505b16 | 8717 | { |
06ecb27c | 8718 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8719 | } |
8720 | ||
06ecb27c | 8721 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8722 | { |
06ecb27c | 8723 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8724 | } |
d0b27fa7 PZ |
8725 | |
8726 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8727 | u64 rt_period_us) | |
8728 | { | |
8729 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8730 | } | |
8731 | ||
8732 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8733 | { | |
8734 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8735 | } | |
6d6bc0ad | 8736 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8737 | |
fe5c7cc2 | 8738 | static struct cftype cpu_files[] = { |
052f1dc7 | 8739 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8740 | { |
8741 | .name = "shares", | |
f4c753b7 PM |
8742 | .read_u64 = cpu_shares_read_u64, |
8743 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8744 | }, |
052f1dc7 PZ |
8745 | #endif |
8746 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8747 | { |
9f0c1e56 | 8748 | .name = "rt_runtime_us", |
06ecb27c PM |
8749 | .read_s64 = cpu_rt_runtime_read, |
8750 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8751 | }, |
d0b27fa7 PZ |
8752 | { |
8753 | .name = "rt_period_us", | |
f4c753b7 PM |
8754 | .read_u64 = cpu_rt_period_read_uint, |
8755 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8756 | }, |
052f1dc7 | 8757 | #endif |
68318b8e SV |
8758 | }; |
8759 | ||
8760 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8761 | { | |
fe5c7cc2 | 8762 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8763 | } |
8764 | ||
8765 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8766 | .name = "cpu", |
8767 | .create = cpu_cgroup_create, | |
8768 | .destroy = cpu_cgroup_destroy, | |
8769 | .can_attach = cpu_cgroup_can_attach, | |
8770 | .attach = cpu_cgroup_attach, | |
8771 | .populate = cpu_cgroup_populate, | |
8772 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8773 | .early_init = 1, |
8774 | }; | |
8775 | ||
052f1dc7 | 8776 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8777 | |
8778 | #ifdef CONFIG_CGROUP_CPUACCT | |
8779 | ||
8780 | /* | |
8781 | * CPU accounting code for task groups. | |
8782 | * | |
8783 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8784 | * (balbir@in.ibm.com). | |
8785 | */ | |
8786 | ||
934352f2 | 8787 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8788 | struct cpuacct { |
8789 | struct cgroup_subsys_state css; | |
8790 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8791 | u64 __percpu *cpuusage; |
ef12fefa | 8792 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8793 | struct cpuacct *parent; |
d842de87 SV |
8794 | }; |
8795 | ||
8796 | struct cgroup_subsys cpuacct_subsys; | |
8797 | ||
8798 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8799 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8800 | { |
32cd756a | 8801 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8802 | struct cpuacct, css); |
8803 | } | |
8804 | ||
8805 | /* return cpu accounting group to which this task belongs */ | |
8806 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8807 | { | |
8808 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8809 | struct cpuacct, css); | |
8810 | } | |
8811 | ||
8812 | /* create a new cpu accounting group */ | |
8813 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8814 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8815 | { |
8816 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8817 | int i; |
d842de87 SV |
8818 | |
8819 | if (!ca) | |
ef12fefa | 8820 | goto out; |
d842de87 SV |
8821 | |
8822 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
8823 | if (!ca->cpuusage) |
8824 | goto out_free_ca; | |
8825 | ||
8826 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
8827 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
8828 | goto out_free_counters; | |
d842de87 | 8829 | |
934352f2 BR |
8830 | if (cgrp->parent) |
8831 | ca->parent = cgroup_ca(cgrp->parent); | |
8832 | ||
d842de87 | 8833 | return &ca->css; |
ef12fefa BR |
8834 | |
8835 | out_free_counters: | |
8836 | while (--i >= 0) | |
8837 | percpu_counter_destroy(&ca->cpustat[i]); | |
8838 | free_percpu(ca->cpuusage); | |
8839 | out_free_ca: | |
8840 | kfree(ca); | |
8841 | out: | |
8842 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
8843 | } |
8844 | ||
8845 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 8846 | static void |
32cd756a | 8847 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8848 | { |
32cd756a | 8849 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 8850 | int i; |
d842de87 | 8851 | |
ef12fefa BR |
8852 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
8853 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
8854 | free_percpu(ca->cpuusage); |
8855 | kfree(ca); | |
8856 | } | |
8857 | ||
720f5498 KC |
8858 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
8859 | { | |
b36128c8 | 8860 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8861 | u64 data; |
8862 | ||
8863 | #ifndef CONFIG_64BIT | |
8864 | /* | |
8865 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
8866 | */ | |
05fa785c | 8867 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8868 | data = *cpuusage; |
05fa785c | 8869 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8870 | #else |
8871 | data = *cpuusage; | |
8872 | #endif | |
8873 | ||
8874 | return data; | |
8875 | } | |
8876 | ||
8877 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
8878 | { | |
b36128c8 | 8879 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
8880 | |
8881 | #ifndef CONFIG_64BIT | |
8882 | /* | |
8883 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
8884 | */ | |
05fa785c | 8885 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 8886 | *cpuusage = val; |
05fa785c | 8887 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
8888 | #else |
8889 | *cpuusage = val; | |
8890 | #endif | |
8891 | } | |
8892 | ||
d842de87 | 8893 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 8894 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 8895 | { |
32cd756a | 8896 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
8897 | u64 totalcpuusage = 0; |
8898 | int i; | |
8899 | ||
720f5498 KC |
8900 | for_each_present_cpu(i) |
8901 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
8902 | |
8903 | return totalcpuusage; | |
8904 | } | |
8905 | ||
0297b803 DG |
8906 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
8907 | u64 reset) | |
8908 | { | |
8909 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8910 | int err = 0; | |
8911 | int i; | |
8912 | ||
8913 | if (reset) { | |
8914 | err = -EINVAL; | |
8915 | goto out; | |
8916 | } | |
8917 | ||
720f5498 KC |
8918 | for_each_present_cpu(i) |
8919 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 8920 | |
0297b803 DG |
8921 | out: |
8922 | return err; | |
8923 | } | |
8924 | ||
e9515c3c KC |
8925 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
8926 | struct seq_file *m) | |
8927 | { | |
8928 | struct cpuacct *ca = cgroup_ca(cgroup); | |
8929 | u64 percpu; | |
8930 | int i; | |
8931 | ||
8932 | for_each_present_cpu(i) { | |
8933 | percpu = cpuacct_cpuusage_read(ca, i); | |
8934 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
8935 | } | |
8936 | seq_printf(m, "\n"); | |
8937 | return 0; | |
8938 | } | |
8939 | ||
ef12fefa BR |
8940 | static const char *cpuacct_stat_desc[] = { |
8941 | [CPUACCT_STAT_USER] = "user", | |
8942 | [CPUACCT_STAT_SYSTEM] = "system", | |
8943 | }; | |
8944 | ||
8945 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
8946 | struct cgroup_map_cb *cb) | |
8947 | { | |
8948 | struct cpuacct *ca = cgroup_ca(cgrp); | |
8949 | int i; | |
8950 | ||
8951 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
8952 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
8953 | val = cputime64_to_clock_t(val); | |
8954 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
8955 | } | |
8956 | return 0; | |
8957 | } | |
8958 | ||
d842de87 SV |
8959 | static struct cftype files[] = { |
8960 | { | |
8961 | .name = "usage", | |
f4c753b7 PM |
8962 | .read_u64 = cpuusage_read, |
8963 | .write_u64 = cpuusage_write, | |
d842de87 | 8964 | }, |
e9515c3c KC |
8965 | { |
8966 | .name = "usage_percpu", | |
8967 | .read_seq_string = cpuacct_percpu_seq_read, | |
8968 | }, | |
ef12fefa BR |
8969 | { |
8970 | .name = "stat", | |
8971 | .read_map = cpuacct_stats_show, | |
8972 | }, | |
d842de87 SV |
8973 | }; |
8974 | ||
32cd756a | 8975 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 8976 | { |
32cd756a | 8977 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
8978 | } |
8979 | ||
8980 | /* | |
8981 | * charge this task's execution time to its accounting group. | |
8982 | * | |
8983 | * called with rq->lock held. | |
8984 | */ | |
8985 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
8986 | { | |
8987 | struct cpuacct *ca; | |
934352f2 | 8988 | int cpu; |
d842de87 | 8989 | |
c40c6f85 | 8990 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
8991 | return; |
8992 | ||
934352f2 | 8993 | cpu = task_cpu(tsk); |
a18b83b7 BR |
8994 | |
8995 | rcu_read_lock(); | |
8996 | ||
d842de87 | 8997 | ca = task_ca(tsk); |
d842de87 | 8998 | |
934352f2 | 8999 | for (; ca; ca = ca->parent) { |
b36128c8 | 9000 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9001 | *cpuusage += cputime; |
9002 | } | |
a18b83b7 BR |
9003 | |
9004 | rcu_read_unlock(); | |
d842de87 SV |
9005 | } |
9006 | ||
fa535a77 AB |
9007 | /* |
9008 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9009 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9010 | * percpu_counter_add with values large enough to always overflow the | |
9011 | * per cpu batch limit causing bad SMP scalability. | |
9012 | * | |
9013 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9014 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9015 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9016 | */ | |
9017 | #ifdef CONFIG_SMP | |
9018 | #define CPUACCT_BATCH \ | |
9019 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9020 | #else | |
9021 | #define CPUACCT_BATCH 0 | |
9022 | #endif | |
9023 | ||
ef12fefa BR |
9024 | /* |
9025 | * Charge the system/user time to the task's accounting group. | |
9026 | */ | |
9027 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9028 | enum cpuacct_stat_index idx, cputime_t val) | |
9029 | { | |
9030 | struct cpuacct *ca; | |
fa535a77 | 9031 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9032 | |
9033 | if (unlikely(!cpuacct_subsys.active)) | |
9034 | return; | |
9035 | ||
9036 | rcu_read_lock(); | |
9037 | ca = task_ca(tsk); | |
9038 | ||
9039 | do { | |
fa535a77 | 9040 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9041 | ca = ca->parent; |
9042 | } while (ca); | |
9043 | rcu_read_unlock(); | |
9044 | } | |
9045 | ||
d842de87 SV |
9046 | struct cgroup_subsys cpuacct_subsys = { |
9047 | .name = "cpuacct", | |
9048 | .create = cpuacct_create, | |
9049 | .destroy = cpuacct_destroy, | |
9050 | .populate = cpuacct_populate, | |
9051 | .subsys_id = cpuacct_subsys_id, | |
9052 | }; | |
9053 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf PM |
9054 | |
9055 | #ifndef CONFIG_SMP | |
9056 | ||
03b042bf PM |
9057 | void synchronize_sched_expedited(void) |
9058 | { | |
fc390cde | 9059 | barrier(); |
03b042bf PM |
9060 | } |
9061 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9062 | ||
9063 | #else /* #ifndef CONFIG_SMP */ | |
9064 | ||
cc631fb7 | 9065 | static atomic_t synchronize_sched_expedited_count = ATOMIC_INIT(0); |
03b042bf | 9066 | |
cc631fb7 | 9067 | static int synchronize_sched_expedited_cpu_stop(void *data) |
03b042bf | 9068 | { |
969c7921 TH |
9069 | /* |
9070 | * There must be a full memory barrier on each affected CPU | |
9071 | * between the time that try_stop_cpus() is called and the | |
9072 | * time that it returns. | |
9073 | * | |
9074 | * In the current initial implementation of cpu_stop, the | |
9075 | * above condition is already met when the control reaches | |
9076 | * this point and the following smp_mb() is not strictly | |
9077 | * necessary. Do smp_mb() anyway for documentation and | |
9078 | * robustness against future implementation changes. | |
9079 | */ | |
cc631fb7 | 9080 | smp_mb(); /* See above comment block. */ |
969c7921 | 9081 | return 0; |
03b042bf | 9082 | } |
03b042bf PM |
9083 | |
9084 | /* | |
9085 | * Wait for an rcu-sched grace period to elapse, but use "big hammer" | |
9086 | * approach to force grace period to end quickly. This consumes | |
9087 | * significant time on all CPUs, and is thus not recommended for | |
9088 | * any sort of common-case code. | |
9089 | * | |
9090 | * Note that it is illegal to call this function while holding any | |
9091 | * lock that is acquired by a CPU-hotplug notifier. Failing to | |
9092 | * observe this restriction will result in deadlock. | |
9093 | */ | |
9094 | void synchronize_sched_expedited(void) | |
9095 | { | |
969c7921 | 9096 | int snap, trycount = 0; |
03b042bf PM |
9097 | |
9098 | smp_mb(); /* ensure prior mod happens before capturing snap. */ | |
969c7921 | 9099 | snap = atomic_read(&synchronize_sched_expedited_count) + 1; |
03b042bf | 9100 | get_online_cpus(); |
969c7921 TH |
9101 | while (try_stop_cpus(cpu_online_mask, |
9102 | synchronize_sched_expedited_cpu_stop, | |
94458d5e | 9103 | NULL) == -EAGAIN) { |
03b042bf PM |
9104 | put_online_cpus(); |
9105 | if (trycount++ < 10) | |
9106 | udelay(trycount * num_online_cpus()); | |
9107 | else { | |
9108 | synchronize_sched(); | |
9109 | return; | |
9110 | } | |
969c7921 | 9111 | if (atomic_read(&synchronize_sched_expedited_count) - snap > 0) { |
03b042bf PM |
9112 | smp_mb(); /* ensure test happens before caller kfree */ |
9113 | return; | |
9114 | } | |
9115 | get_online_cpus(); | |
9116 | } | |
969c7921 | 9117 | atomic_inc(&synchronize_sched_expedited_count); |
cc631fb7 | 9118 | smp_mb__after_atomic_inc(); /* ensure post-GP actions seen after GP. */ |
03b042bf | 9119 | put_online_cpus(); |
03b042bf PM |
9120 | } |
9121 | EXPORT_SYMBOL_GPL(synchronize_sched_expedited); | |
9122 | ||
9123 | #endif /* #else #ifndef CONFIG_SMP */ |