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" |
5091faa4 | 82 | #include "sched_autogroup.h" |
6e0534f2 | 83 | |
a8d154b0 | 84 | #define CREATE_TRACE_POINTS |
ad8d75ff | 85 | #include <trace/events/sched.h> |
a8d154b0 | 86 | |
1da177e4 LT |
87 | /* |
88 | * Convert user-nice values [ -20 ... 0 ... 19 ] | |
89 | * to static priority [ MAX_RT_PRIO..MAX_PRIO-1 ], | |
90 | * and back. | |
91 | */ | |
92 | #define NICE_TO_PRIO(nice) (MAX_RT_PRIO + (nice) + 20) | |
93 | #define PRIO_TO_NICE(prio) ((prio) - MAX_RT_PRIO - 20) | |
94 | #define TASK_NICE(p) PRIO_TO_NICE((p)->static_prio) | |
95 | ||
96 | /* | |
97 | * 'User priority' is the nice value converted to something we | |
98 | * can work with better when scaling various scheduler parameters, | |
99 | * it's a [ 0 ... 39 ] range. | |
100 | */ | |
101 | #define USER_PRIO(p) ((p)-MAX_RT_PRIO) | |
102 | #define TASK_USER_PRIO(p) USER_PRIO((p)->static_prio) | |
103 | #define MAX_USER_PRIO (USER_PRIO(MAX_PRIO)) | |
104 | ||
105 | /* | |
d7876a08 | 106 | * Helpers for converting nanosecond timing to jiffy resolution |
1da177e4 | 107 | */ |
d6322faf | 108 | #define NS_TO_JIFFIES(TIME) ((unsigned long)(TIME) / (NSEC_PER_SEC / HZ)) |
1da177e4 | 109 | |
6aa645ea IM |
110 | #define NICE_0_LOAD SCHED_LOAD_SCALE |
111 | #define NICE_0_SHIFT SCHED_LOAD_SHIFT | |
112 | ||
1da177e4 LT |
113 | /* |
114 | * These are the 'tuning knobs' of the scheduler: | |
115 | * | |
a4ec24b4 | 116 | * default timeslice is 100 msecs (used only for SCHED_RR tasks). |
1da177e4 LT |
117 | * Timeslices get refilled after they expire. |
118 | */ | |
1da177e4 | 119 | #define DEF_TIMESLICE (100 * HZ / 1000) |
2dd73a4f | 120 | |
d0b27fa7 PZ |
121 | /* |
122 | * single value that denotes runtime == period, ie unlimited time. | |
123 | */ | |
124 | #define RUNTIME_INF ((u64)~0ULL) | |
125 | ||
e05606d3 IM |
126 | static inline int rt_policy(int policy) |
127 | { | |
3f33a7ce | 128 | if (unlikely(policy == SCHED_FIFO || policy == SCHED_RR)) |
e05606d3 IM |
129 | return 1; |
130 | return 0; | |
131 | } | |
132 | ||
133 | static inline int task_has_rt_policy(struct task_struct *p) | |
134 | { | |
135 | return rt_policy(p->policy); | |
136 | } | |
137 | ||
1da177e4 | 138 | /* |
6aa645ea | 139 | * This is the priority-queue data structure of the RT scheduling class: |
1da177e4 | 140 | */ |
6aa645ea IM |
141 | struct rt_prio_array { |
142 | DECLARE_BITMAP(bitmap, MAX_RT_PRIO+1); /* include 1 bit for delimiter */ | |
143 | struct list_head queue[MAX_RT_PRIO]; | |
144 | }; | |
145 | ||
d0b27fa7 | 146 | struct rt_bandwidth { |
ea736ed5 | 147 | /* nests inside the rq lock: */ |
0986b11b | 148 | raw_spinlock_t rt_runtime_lock; |
ea736ed5 IM |
149 | ktime_t rt_period; |
150 | u64 rt_runtime; | |
151 | struct hrtimer rt_period_timer; | |
d0b27fa7 PZ |
152 | }; |
153 | ||
154 | static struct rt_bandwidth def_rt_bandwidth; | |
155 | ||
156 | static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun); | |
157 | ||
158 | static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer) | |
159 | { | |
160 | struct rt_bandwidth *rt_b = | |
161 | container_of(timer, struct rt_bandwidth, rt_period_timer); | |
162 | ktime_t now; | |
163 | int overrun; | |
164 | int idle = 0; | |
165 | ||
166 | for (;;) { | |
167 | now = hrtimer_cb_get_time(timer); | |
168 | overrun = hrtimer_forward(timer, now, rt_b->rt_period); | |
169 | ||
170 | if (!overrun) | |
171 | break; | |
172 | ||
173 | idle = do_sched_rt_period_timer(rt_b, overrun); | |
174 | } | |
175 | ||
176 | return idle ? HRTIMER_NORESTART : HRTIMER_RESTART; | |
177 | } | |
178 | ||
179 | static | |
180 | void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime) | |
181 | { | |
182 | rt_b->rt_period = ns_to_ktime(period); | |
183 | rt_b->rt_runtime = runtime; | |
184 | ||
0986b11b | 185 | raw_spin_lock_init(&rt_b->rt_runtime_lock); |
ac086bc2 | 186 | |
d0b27fa7 PZ |
187 | hrtimer_init(&rt_b->rt_period_timer, |
188 | CLOCK_MONOTONIC, HRTIMER_MODE_REL); | |
189 | rt_b->rt_period_timer.function = sched_rt_period_timer; | |
d0b27fa7 PZ |
190 | } |
191 | ||
c8bfff6d KH |
192 | static inline int rt_bandwidth_enabled(void) |
193 | { | |
194 | return sysctl_sched_rt_runtime >= 0; | |
d0b27fa7 PZ |
195 | } |
196 | ||
197 | static void start_rt_bandwidth(struct rt_bandwidth *rt_b) | |
198 | { | |
199 | ktime_t now; | |
200 | ||
cac64d00 | 201 | if (!rt_bandwidth_enabled() || rt_b->rt_runtime == RUNTIME_INF) |
d0b27fa7 PZ |
202 | return; |
203 | ||
204 | if (hrtimer_active(&rt_b->rt_period_timer)) | |
205 | return; | |
206 | ||
0986b11b | 207 | raw_spin_lock(&rt_b->rt_runtime_lock); |
d0b27fa7 | 208 | for (;;) { |
7f1e2ca9 PZ |
209 | unsigned long delta; |
210 | ktime_t soft, hard; | |
211 | ||
d0b27fa7 PZ |
212 | if (hrtimer_active(&rt_b->rt_period_timer)) |
213 | break; | |
214 | ||
215 | now = hrtimer_cb_get_time(&rt_b->rt_period_timer); | |
216 | hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period); | |
7f1e2ca9 PZ |
217 | |
218 | soft = hrtimer_get_softexpires(&rt_b->rt_period_timer); | |
219 | hard = hrtimer_get_expires(&rt_b->rt_period_timer); | |
220 | delta = ktime_to_ns(ktime_sub(hard, soft)); | |
221 | __hrtimer_start_range_ns(&rt_b->rt_period_timer, soft, delta, | |
5c333864 | 222 | HRTIMER_MODE_ABS_PINNED, 0); |
d0b27fa7 | 223 | } |
0986b11b | 224 | raw_spin_unlock(&rt_b->rt_runtime_lock); |
d0b27fa7 PZ |
225 | } |
226 | ||
227 | #ifdef CONFIG_RT_GROUP_SCHED | |
228 | static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b) | |
229 | { | |
230 | hrtimer_cancel(&rt_b->rt_period_timer); | |
231 | } | |
232 | #endif | |
233 | ||
712555ee HC |
234 | /* |
235 | * sched_domains_mutex serializes calls to arch_init_sched_domains, | |
236 | * detach_destroy_domains and partition_sched_domains. | |
237 | */ | |
238 | static DEFINE_MUTEX(sched_domains_mutex); | |
239 | ||
7c941438 | 240 | #ifdef CONFIG_CGROUP_SCHED |
29f59db3 | 241 | |
68318b8e SV |
242 | #include <linux/cgroup.h> |
243 | ||
29f59db3 SV |
244 | struct cfs_rq; |
245 | ||
6f505b16 PZ |
246 | static LIST_HEAD(task_groups); |
247 | ||
29f59db3 | 248 | /* task group related information */ |
4cf86d77 | 249 | struct task_group { |
68318b8e | 250 | struct cgroup_subsys_state css; |
6c415b92 | 251 | |
052f1dc7 | 252 | #ifdef CONFIG_FAIR_GROUP_SCHED |
29f59db3 SV |
253 | /* schedulable entities of this group on each cpu */ |
254 | struct sched_entity **se; | |
255 | /* runqueue "owned" by this group on each cpu */ | |
256 | struct cfs_rq **cfs_rq; | |
257 | unsigned long shares; | |
2069dd75 PZ |
258 | |
259 | atomic_t load_weight; | |
052f1dc7 PZ |
260 | #endif |
261 | ||
262 | #ifdef CONFIG_RT_GROUP_SCHED | |
263 | struct sched_rt_entity **rt_se; | |
264 | struct rt_rq **rt_rq; | |
265 | ||
d0b27fa7 | 266 | struct rt_bandwidth rt_bandwidth; |
052f1dc7 | 267 | #endif |
6b2d7700 | 268 | |
ae8393e5 | 269 | struct rcu_head rcu; |
6f505b16 | 270 | struct list_head list; |
f473aa5e PZ |
271 | |
272 | struct task_group *parent; | |
273 | struct list_head siblings; | |
274 | struct list_head children; | |
5091faa4 MG |
275 | |
276 | #ifdef CONFIG_SCHED_AUTOGROUP | |
277 | struct autogroup *autogroup; | |
278 | #endif | |
29f59db3 SV |
279 | }; |
280 | ||
3d4b47b4 | 281 | /* task_group_lock serializes the addition/removal of task groups */ |
8ed36996 | 282 | static DEFINE_SPINLOCK(task_group_lock); |
ec2c507f | 283 | |
e9036b36 CG |
284 | #ifdef CONFIG_FAIR_GROUP_SCHED |
285 | ||
07e06b01 | 286 | # define ROOT_TASK_GROUP_LOAD NICE_0_LOAD |
052f1dc7 | 287 | |
cb4ad1ff | 288 | /* |
2e084786 LJ |
289 | * A weight of 0 or 1 can cause arithmetics problems. |
290 | * A weight of a cfs_rq is the sum of weights of which entities | |
291 | * are queued on this cfs_rq, so a weight of a entity should not be | |
292 | * too large, so as the shares value of a task group. | |
cb4ad1ff MX |
293 | * (The default weight is 1024 - so there's no practical |
294 | * limitation from this.) | |
295 | */ | |
18d95a28 | 296 | #define MIN_SHARES 2 |
2e084786 | 297 | #define MAX_SHARES (1UL << 18) |
18d95a28 | 298 | |
07e06b01 | 299 | static int root_task_group_load = ROOT_TASK_GROUP_LOAD; |
052f1dc7 PZ |
300 | #endif |
301 | ||
29f59db3 | 302 | /* Default task group. |
3a252015 | 303 | * Every task in system belong to this group at bootup. |
29f59db3 | 304 | */ |
07e06b01 | 305 | struct task_group root_task_group; |
29f59db3 | 306 | |
7c941438 | 307 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 308 | |
6aa645ea IM |
309 | /* CFS-related fields in a runqueue */ |
310 | struct cfs_rq { | |
311 | struct load_weight load; | |
312 | unsigned long nr_running; | |
313 | ||
6aa645ea | 314 | u64 exec_clock; |
e9acbff6 | 315 | u64 min_vruntime; |
6aa645ea IM |
316 | |
317 | struct rb_root tasks_timeline; | |
318 | struct rb_node *rb_leftmost; | |
4a55bd5e PZ |
319 | |
320 | struct list_head tasks; | |
321 | struct list_head *balance_iterator; | |
322 | ||
323 | /* | |
324 | * 'curr' points to currently running entity on this cfs_rq. | |
6aa645ea IM |
325 | * It is set to NULL otherwise (i.e when none are currently running). |
326 | */ | |
4793241b | 327 | struct sched_entity *curr, *next, *last; |
ddc97297 | 328 | |
5ac5c4d6 | 329 | unsigned int nr_spread_over; |
ddc97297 | 330 | |
62160e3f | 331 | #ifdef CONFIG_FAIR_GROUP_SCHED |
6aa645ea IM |
332 | struct rq *rq; /* cpu runqueue to which this cfs_rq is attached */ |
333 | ||
41a2d6cf IM |
334 | /* |
335 | * leaf cfs_rqs are those that hold tasks (lowest schedulable entity in | |
6aa645ea IM |
336 | * a hierarchy). Non-leaf lrqs hold other higher schedulable entities |
337 | * (like users, containers etc.) | |
338 | * | |
339 | * leaf_cfs_rq_list ties together list of leaf cfs_rq's in a cpu. This | |
340 | * list is used during load balance. | |
341 | */ | |
3d4b47b4 | 342 | int on_list; |
41a2d6cf IM |
343 | struct list_head leaf_cfs_rq_list; |
344 | struct task_group *tg; /* group that "owns" this runqueue */ | |
c09595f6 PZ |
345 | |
346 | #ifdef CONFIG_SMP | |
c09595f6 | 347 | /* |
c8cba857 | 348 | * the part of load.weight contributed by tasks |
c09595f6 | 349 | */ |
c8cba857 | 350 | unsigned long task_weight; |
c09595f6 | 351 | |
c8cba857 PZ |
352 | /* |
353 | * h_load = weight * f(tg) | |
354 | * | |
355 | * Where f(tg) is the recursive weight fraction assigned to | |
356 | * this group. | |
357 | */ | |
358 | unsigned long h_load; | |
c09595f6 | 359 | |
c8cba857 | 360 | /* |
3b3d190e PT |
361 | * Maintaining per-cpu shares distribution for group scheduling |
362 | * | |
363 | * load_stamp is the last time we updated the load average | |
364 | * load_last is the last time we updated the load average and saw load | |
365 | * load_unacc_exec_time is currently unaccounted execution time | |
c8cba857 | 366 | */ |
2069dd75 PZ |
367 | u64 load_avg; |
368 | u64 load_period; | |
3b3d190e | 369 | u64 load_stamp, load_last, load_unacc_exec_time; |
f1d239f7 | 370 | |
2069dd75 | 371 | unsigned long load_contribution; |
c09595f6 | 372 | #endif |
6aa645ea IM |
373 | #endif |
374 | }; | |
1da177e4 | 375 | |
6aa645ea IM |
376 | /* Real-Time classes' related field in a runqueue: */ |
377 | struct rt_rq { | |
378 | struct rt_prio_array active; | |
63489e45 | 379 | unsigned long rt_nr_running; |
052f1dc7 | 380 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 GH |
381 | struct { |
382 | int curr; /* highest queued rt task prio */ | |
398a153b | 383 | #ifdef CONFIG_SMP |
e864c499 | 384 | int next; /* next highest */ |
398a153b | 385 | #endif |
e864c499 | 386 | } highest_prio; |
6f505b16 | 387 | #endif |
fa85ae24 | 388 | #ifdef CONFIG_SMP |
73fe6aae | 389 | unsigned long rt_nr_migratory; |
a1ba4d8b | 390 | unsigned long rt_nr_total; |
a22d7fc1 | 391 | int overloaded; |
917b627d | 392 | struct plist_head pushable_tasks; |
fa85ae24 | 393 | #endif |
6f505b16 | 394 | int rt_throttled; |
fa85ae24 | 395 | u64 rt_time; |
ac086bc2 | 396 | u64 rt_runtime; |
ea736ed5 | 397 | /* Nests inside the rq lock: */ |
0986b11b | 398 | raw_spinlock_t rt_runtime_lock; |
6f505b16 | 399 | |
052f1dc7 | 400 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc PZ |
401 | unsigned long rt_nr_boosted; |
402 | ||
6f505b16 PZ |
403 | struct rq *rq; |
404 | struct list_head leaf_rt_rq_list; | |
405 | struct task_group *tg; | |
6f505b16 | 406 | #endif |
6aa645ea IM |
407 | }; |
408 | ||
57d885fe GH |
409 | #ifdef CONFIG_SMP |
410 | ||
411 | /* | |
412 | * We add the notion of a root-domain which will be used to define per-domain | |
0eab9146 IM |
413 | * variables. Each exclusive cpuset essentially defines an island domain by |
414 | * fully partitioning the member cpus from any other cpuset. Whenever a new | |
57d885fe GH |
415 | * exclusive cpuset is created, we also create and attach a new root-domain |
416 | * object. | |
417 | * | |
57d885fe GH |
418 | */ |
419 | struct root_domain { | |
420 | atomic_t refcount; | |
c6c4927b RR |
421 | cpumask_var_t span; |
422 | cpumask_var_t online; | |
637f5085 | 423 | |
0eab9146 | 424 | /* |
637f5085 GH |
425 | * The "RT overload" flag: it gets set if a CPU has more than |
426 | * one runnable RT task. | |
427 | */ | |
c6c4927b | 428 | cpumask_var_t rto_mask; |
0eab9146 | 429 | atomic_t rto_count; |
6e0534f2 | 430 | struct cpupri cpupri; |
57d885fe GH |
431 | }; |
432 | ||
dc938520 GH |
433 | /* |
434 | * By default the system creates a single root-domain with all cpus as | |
435 | * members (mimicking the global state we have today). | |
436 | */ | |
57d885fe GH |
437 | static struct root_domain def_root_domain; |
438 | ||
ed2d372c | 439 | #endif /* CONFIG_SMP */ |
57d885fe | 440 | |
1da177e4 LT |
441 | /* |
442 | * This is the main, per-CPU runqueue data structure. | |
443 | * | |
444 | * Locking rule: those places that want to lock multiple runqueues | |
445 | * (such as the load balancing or the thread migration code), lock | |
446 | * acquire operations must be ordered by ascending &runqueue. | |
447 | */ | |
70b97a7f | 448 | struct rq { |
d8016491 | 449 | /* runqueue lock: */ |
05fa785c | 450 | raw_spinlock_t lock; |
1da177e4 LT |
451 | |
452 | /* | |
453 | * nr_running and cpu_load should be in the same cacheline because | |
454 | * remote CPUs use both these fields when doing load calculation. | |
455 | */ | |
456 | unsigned long nr_running; | |
6aa645ea IM |
457 | #define CPU_LOAD_IDX_MAX 5 |
458 | unsigned long cpu_load[CPU_LOAD_IDX_MAX]; | |
fdf3e95d | 459 | unsigned long last_load_update_tick; |
46cb4b7c | 460 | #ifdef CONFIG_NO_HZ |
39c0cbe2 | 461 | u64 nohz_stamp; |
83cd4fe2 | 462 | unsigned char nohz_balance_kick; |
46cb4b7c | 463 | #endif |
a64692a3 MG |
464 | unsigned int skip_clock_update; |
465 | ||
d8016491 IM |
466 | /* capture load from *all* tasks on this cpu: */ |
467 | struct load_weight load; | |
6aa645ea IM |
468 | unsigned long nr_load_updates; |
469 | u64 nr_switches; | |
470 | ||
471 | struct cfs_rq cfs; | |
6f505b16 | 472 | struct rt_rq rt; |
6f505b16 | 473 | |
6aa645ea | 474 | #ifdef CONFIG_FAIR_GROUP_SCHED |
d8016491 IM |
475 | /* list of leaf cfs_rq on this cpu: */ |
476 | struct list_head leaf_cfs_rq_list; | |
052f1dc7 PZ |
477 | #endif |
478 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 479 | struct list_head leaf_rt_rq_list; |
1da177e4 | 480 | #endif |
1da177e4 LT |
481 | |
482 | /* | |
483 | * This is part of a global counter where only the total sum | |
484 | * over all CPUs matters. A task can increase this counter on | |
485 | * one CPU and if it got migrated afterwards it may decrease | |
486 | * it on another CPU. Always updated under the runqueue lock: | |
487 | */ | |
488 | unsigned long nr_uninterruptible; | |
489 | ||
34f971f6 | 490 | struct task_struct *curr, *idle, *stop; |
c9819f45 | 491 | unsigned long next_balance; |
1da177e4 | 492 | struct mm_struct *prev_mm; |
6aa645ea | 493 | |
3e51f33f | 494 | u64 clock; |
305e6835 | 495 | u64 clock_task; |
6aa645ea | 496 | |
1da177e4 LT |
497 | atomic_t nr_iowait; |
498 | ||
499 | #ifdef CONFIG_SMP | |
0eab9146 | 500 | struct root_domain *rd; |
1da177e4 LT |
501 | struct sched_domain *sd; |
502 | ||
e51fd5e2 PZ |
503 | unsigned long cpu_power; |
504 | ||
a0a522ce | 505 | unsigned char idle_at_tick; |
1da177e4 | 506 | /* For active balancing */ |
3f029d3c | 507 | int post_schedule; |
1da177e4 LT |
508 | int active_balance; |
509 | int push_cpu; | |
969c7921 | 510 | struct cpu_stop_work active_balance_work; |
d8016491 IM |
511 | /* cpu of this runqueue: */ |
512 | int cpu; | |
1f11eb6a | 513 | int online; |
1da177e4 | 514 | |
a8a51d5e | 515 | unsigned long avg_load_per_task; |
1da177e4 | 516 | |
e9e9250b PZ |
517 | u64 rt_avg; |
518 | u64 age_stamp; | |
1b9508f6 MG |
519 | u64 idle_stamp; |
520 | u64 avg_idle; | |
1da177e4 LT |
521 | #endif |
522 | ||
aa483808 VP |
523 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
524 | u64 prev_irq_time; | |
525 | #endif | |
526 | ||
dce48a84 TG |
527 | /* calc_load related fields */ |
528 | unsigned long calc_load_update; | |
529 | long calc_load_active; | |
530 | ||
8f4d37ec | 531 | #ifdef CONFIG_SCHED_HRTICK |
31656519 PZ |
532 | #ifdef CONFIG_SMP |
533 | int hrtick_csd_pending; | |
534 | struct call_single_data hrtick_csd; | |
535 | #endif | |
8f4d37ec PZ |
536 | struct hrtimer hrtick_timer; |
537 | #endif | |
538 | ||
1da177e4 LT |
539 | #ifdef CONFIG_SCHEDSTATS |
540 | /* latency stats */ | |
541 | struct sched_info rq_sched_info; | |
9c2c4802 KC |
542 | unsigned long long rq_cpu_time; |
543 | /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ | |
1da177e4 LT |
544 | |
545 | /* sys_sched_yield() stats */ | |
480b9434 | 546 | unsigned int yld_count; |
1da177e4 LT |
547 | |
548 | /* schedule() stats */ | |
480b9434 KC |
549 | unsigned int sched_switch; |
550 | unsigned int sched_count; | |
551 | unsigned int sched_goidle; | |
1da177e4 LT |
552 | |
553 | /* try_to_wake_up() stats */ | |
480b9434 KC |
554 | unsigned int ttwu_count; |
555 | unsigned int ttwu_local; | |
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 | { | |
5091faa4 | 606 | struct task_group *tg; |
dc61b1d6 PZ |
607 | struct cgroup_subsys_state *css; |
608 | ||
609 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, | |
610 | lockdep_is_held(&task_rq(p)->lock)); | |
5091faa4 MG |
611 | tg = container_of(css, struct task_group, css); |
612 | ||
613 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
614 | } |
615 | ||
616 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
617 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
618 | { | |
619 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
620 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
621 | p->se.parent = task_group(p)->se[cpu]; | |
622 | #endif | |
623 | ||
624 | #ifdef CONFIG_RT_GROUP_SCHED | |
625 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
626 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
627 | #endif | |
628 | } | |
629 | ||
630 | #else /* CONFIG_CGROUP_SCHED */ | |
631 | ||
632 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
633 | static inline struct task_group *task_group(struct task_struct *p) | |
634 | { | |
635 | return NULL; | |
636 | } | |
637 | ||
638 | #endif /* CONFIG_CGROUP_SCHED */ | |
639 | ||
fe44d621 | 640 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 641 | |
fe44d621 | 642 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 643 | { |
fe44d621 | 644 | s64 delta; |
305e6835 | 645 | |
f26f9aff MG |
646 | if (rq->skip_clock_update) |
647 | return; | |
aa483808 | 648 | |
fe44d621 PZ |
649 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
650 | rq->clock += delta; | |
651 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
652 | } |
653 | ||
bf5c91ba IM |
654 | /* |
655 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
656 | */ | |
657 | #ifdef CONFIG_SCHED_DEBUG | |
658 | # define const_debug __read_mostly | |
659 | #else | |
660 | # define const_debug static const | |
661 | #endif | |
662 | ||
017730c1 IM |
663 | /** |
664 | * runqueue_is_locked | |
e17b38bf | 665 | * @cpu: the processor in question. |
017730c1 IM |
666 | * |
667 | * Returns true if the current cpu runqueue is locked. | |
668 | * This interface allows printk to be called with the runqueue lock | |
669 | * held and know whether or not it is OK to wake up the klogd. | |
670 | */ | |
89f19f04 | 671 | int runqueue_is_locked(int cpu) |
017730c1 | 672 | { |
05fa785c | 673 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
674 | } |
675 | ||
bf5c91ba IM |
676 | /* |
677 | * Debugging: various feature bits | |
678 | */ | |
f00b45c1 PZ |
679 | |
680 | #define SCHED_FEAT(name, enabled) \ | |
681 | __SCHED_FEAT_##name , | |
682 | ||
bf5c91ba | 683 | enum { |
f00b45c1 | 684 | #include "sched_features.h" |
bf5c91ba IM |
685 | }; |
686 | ||
f00b45c1 PZ |
687 | #undef SCHED_FEAT |
688 | ||
689 | #define SCHED_FEAT(name, enabled) \ | |
690 | (1UL << __SCHED_FEAT_##name) * enabled | | |
691 | ||
bf5c91ba | 692 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
693 | #include "sched_features.h" |
694 | 0; | |
695 | ||
696 | #undef SCHED_FEAT | |
697 | ||
698 | #ifdef CONFIG_SCHED_DEBUG | |
699 | #define SCHED_FEAT(name, enabled) \ | |
700 | #name , | |
701 | ||
983ed7a6 | 702 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
703 | #include "sched_features.h" |
704 | NULL | |
705 | }; | |
706 | ||
707 | #undef SCHED_FEAT | |
708 | ||
34f3a814 | 709 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 710 | { |
f00b45c1 PZ |
711 | int i; |
712 | ||
713 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
714 | if (!(sysctl_sched_features & (1UL << i))) |
715 | seq_puts(m, "NO_"); | |
716 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 717 | } |
34f3a814 | 718 | seq_puts(m, "\n"); |
f00b45c1 | 719 | |
34f3a814 | 720 | return 0; |
f00b45c1 PZ |
721 | } |
722 | ||
723 | static ssize_t | |
724 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
725 | size_t cnt, loff_t *ppos) | |
726 | { | |
727 | char buf[64]; | |
7740191c | 728 | char *cmp; |
f00b45c1 PZ |
729 | int neg = 0; |
730 | int i; | |
731 | ||
732 | if (cnt > 63) | |
733 | cnt = 63; | |
734 | ||
735 | if (copy_from_user(&buf, ubuf, cnt)) | |
736 | return -EFAULT; | |
737 | ||
738 | buf[cnt] = 0; | |
7740191c | 739 | cmp = strstrip(buf); |
f00b45c1 | 740 | |
524429c3 | 741 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
742 | neg = 1; |
743 | cmp += 3; | |
744 | } | |
745 | ||
746 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 747 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
748 | if (neg) |
749 | sysctl_sched_features &= ~(1UL << i); | |
750 | else | |
751 | sysctl_sched_features |= (1UL << i); | |
752 | break; | |
753 | } | |
754 | } | |
755 | ||
756 | if (!sched_feat_names[i]) | |
757 | return -EINVAL; | |
758 | ||
42994724 | 759 | *ppos += cnt; |
f00b45c1 PZ |
760 | |
761 | return cnt; | |
762 | } | |
763 | ||
34f3a814 LZ |
764 | static int sched_feat_open(struct inode *inode, struct file *filp) |
765 | { | |
766 | return single_open(filp, sched_feat_show, NULL); | |
767 | } | |
768 | ||
828c0950 | 769 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
770 | .open = sched_feat_open, |
771 | .write = sched_feat_write, | |
772 | .read = seq_read, | |
773 | .llseek = seq_lseek, | |
774 | .release = single_release, | |
f00b45c1 PZ |
775 | }; |
776 | ||
777 | static __init int sched_init_debug(void) | |
778 | { | |
f00b45c1 PZ |
779 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
780 | &sched_feat_fops); | |
781 | ||
782 | return 0; | |
783 | } | |
784 | late_initcall(sched_init_debug); | |
785 | ||
786 | #endif | |
787 | ||
788 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 789 | |
b82d9fdd PZ |
790 | /* |
791 | * Number of tasks to iterate in a single balance run. | |
792 | * Limited because this is done with IRQs disabled. | |
793 | */ | |
794 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
795 | ||
e9e9250b PZ |
796 | /* |
797 | * period over which we average the RT time consumption, measured | |
798 | * in ms. | |
799 | * | |
800 | * default: 1s | |
801 | */ | |
802 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
803 | ||
fa85ae24 | 804 | /* |
9f0c1e56 | 805 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
806 | * default: 1s |
807 | */ | |
9f0c1e56 | 808 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 809 | |
6892b75e IM |
810 | static __read_mostly int scheduler_running; |
811 | ||
9f0c1e56 PZ |
812 | /* |
813 | * part of the period that we allow rt tasks to run in us. | |
814 | * default: 0.95s | |
815 | */ | |
816 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 817 | |
d0b27fa7 PZ |
818 | static inline u64 global_rt_period(void) |
819 | { | |
820 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
821 | } | |
822 | ||
823 | static inline u64 global_rt_runtime(void) | |
824 | { | |
e26873bb | 825 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
826 | return RUNTIME_INF; |
827 | ||
828 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
829 | } | |
fa85ae24 | 830 | |
1da177e4 | 831 | #ifndef prepare_arch_switch |
4866cde0 NP |
832 | # define prepare_arch_switch(next) do { } while (0) |
833 | #endif | |
834 | #ifndef finish_arch_switch | |
835 | # define finish_arch_switch(prev) do { } while (0) | |
836 | #endif | |
837 | ||
051a1d1a DA |
838 | static inline int task_current(struct rq *rq, struct task_struct *p) |
839 | { | |
840 | return rq->curr == p; | |
841 | } | |
842 | ||
4866cde0 | 843 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 844 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 845 | { |
051a1d1a | 846 | return task_current(rq, p); |
4866cde0 NP |
847 | } |
848 | ||
70b97a7f | 849 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
850 | { |
851 | } | |
852 | ||
70b97a7f | 853 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 854 | { |
da04c035 IM |
855 | #ifdef CONFIG_DEBUG_SPINLOCK |
856 | /* this is a valid case when another task releases the spinlock */ | |
857 | rq->lock.owner = current; | |
858 | #endif | |
8a25d5de IM |
859 | /* |
860 | * If we are tracking spinlock dependencies then we have to | |
861 | * fix up the runqueue lock - which gets 'carried over' from | |
862 | * prev into current: | |
863 | */ | |
864 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
865 | ||
05fa785c | 866 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
867 | } |
868 | ||
869 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 870 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
871 | { |
872 | #ifdef CONFIG_SMP | |
873 | return p->oncpu; | |
874 | #else | |
051a1d1a | 875 | return task_current(rq, p); |
4866cde0 NP |
876 | #endif |
877 | } | |
878 | ||
70b97a7f | 879 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
880 | { |
881 | #ifdef CONFIG_SMP | |
882 | /* | |
883 | * We can optimise this out completely for !SMP, because the | |
884 | * SMP rebalancing from interrupt is the only thing that cares | |
885 | * here. | |
886 | */ | |
887 | next->oncpu = 1; | |
888 | #endif | |
889 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 890 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 891 | #else |
05fa785c | 892 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
893 | #endif |
894 | } | |
895 | ||
70b97a7f | 896 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
897 | { |
898 | #ifdef CONFIG_SMP | |
899 | /* | |
900 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
901 | * We must ensure this doesn't happen until the switch is completely | |
902 | * finished. | |
903 | */ | |
904 | smp_wmb(); | |
905 | prev->oncpu = 0; | |
906 | #endif | |
907 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
908 | local_irq_enable(); | |
1da177e4 | 909 | #endif |
4866cde0 NP |
910 | } |
911 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 912 | |
0970d299 | 913 | /* |
65cc8e48 PZ |
914 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
915 | * against ttwu(). | |
0970d299 PZ |
916 | */ |
917 | static inline int task_is_waking(struct task_struct *p) | |
918 | { | |
0017d735 | 919 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
920 | } |
921 | ||
b29739f9 IM |
922 | /* |
923 | * __task_rq_lock - lock the runqueue a given task resides on. | |
924 | * Must be called interrupts disabled. | |
925 | */ | |
70b97a7f | 926 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
927 | __acquires(rq->lock) |
928 | { | |
0970d299 PZ |
929 | struct rq *rq; |
930 | ||
3a5c359a | 931 | for (;;) { |
0970d299 | 932 | rq = task_rq(p); |
05fa785c | 933 | raw_spin_lock(&rq->lock); |
65cc8e48 | 934 | if (likely(rq == task_rq(p))) |
3a5c359a | 935 | return rq; |
05fa785c | 936 | raw_spin_unlock(&rq->lock); |
b29739f9 | 937 | } |
b29739f9 IM |
938 | } |
939 | ||
1da177e4 LT |
940 | /* |
941 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 942 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
943 | * explicitly disabling preemption. |
944 | */ | |
70b97a7f | 945 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
946 | __acquires(rq->lock) |
947 | { | |
70b97a7f | 948 | struct rq *rq; |
1da177e4 | 949 | |
3a5c359a AK |
950 | for (;;) { |
951 | local_irq_save(*flags); | |
952 | rq = task_rq(p); | |
05fa785c | 953 | raw_spin_lock(&rq->lock); |
65cc8e48 | 954 | if (likely(rq == task_rq(p))) |
3a5c359a | 955 | return rq; |
05fa785c | 956 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 957 | } |
1da177e4 LT |
958 | } |
959 | ||
a9957449 | 960 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
961 | __releases(rq->lock) |
962 | { | |
05fa785c | 963 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
964 | } |
965 | ||
70b97a7f | 966 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
967 | __releases(rq->lock) |
968 | { | |
05fa785c | 969 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
970 | } |
971 | ||
1da177e4 | 972 | /* |
cc2a73b5 | 973 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 974 | */ |
a9957449 | 975 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
976 | __acquires(rq->lock) |
977 | { | |
70b97a7f | 978 | struct rq *rq; |
1da177e4 LT |
979 | |
980 | local_irq_disable(); | |
981 | rq = this_rq(); | |
05fa785c | 982 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
983 | |
984 | return rq; | |
985 | } | |
986 | ||
8f4d37ec PZ |
987 | #ifdef CONFIG_SCHED_HRTICK |
988 | /* | |
989 | * Use HR-timers to deliver accurate preemption points. | |
990 | * | |
991 | * Its all a bit involved since we cannot program an hrt while holding the | |
992 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
993 | * reschedule event. | |
994 | * | |
995 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
996 | * rq->lock. | |
997 | */ | |
8f4d37ec PZ |
998 | |
999 | /* | |
1000 | * Use hrtick when: | |
1001 | * - enabled by features | |
1002 | * - hrtimer is actually high res | |
1003 | */ | |
1004 | static inline int hrtick_enabled(struct rq *rq) | |
1005 | { | |
1006 | if (!sched_feat(HRTICK)) | |
1007 | return 0; | |
ba42059f | 1008 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1009 | return 0; |
8f4d37ec PZ |
1010 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1011 | } | |
1012 | ||
8f4d37ec PZ |
1013 | static void hrtick_clear(struct rq *rq) |
1014 | { | |
1015 | if (hrtimer_active(&rq->hrtick_timer)) | |
1016 | hrtimer_cancel(&rq->hrtick_timer); | |
1017 | } | |
1018 | ||
8f4d37ec PZ |
1019 | /* |
1020 | * High-resolution timer tick. | |
1021 | * Runs from hardirq context with interrupts disabled. | |
1022 | */ | |
1023 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1024 | { | |
1025 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1026 | ||
1027 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1028 | ||
05fa785c | 1029 | raw_spin_lock(&rq->lock); |
3e51f33f | 1030 | update_rq_clock(rq); |
8f4d37ec | 1031 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1032 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1033 | |
1034 | return HRTIMER_NORESTART; | |
1035 | } | |
1036 | ||
95e904c7 | 1037 | #ifdef CONFIG_SMP |
31656519 PZ |
1038 | /* |
1039 | * called from hardirq (IPI) context | |
1040 | */ | |
1041 | static void __hrtick_start(void *arg) | |
b328ca18 | 1042 | { |
31656519 | 1043 | struct rq *rq = arg; |
b328ca18 | 1044 | |
05fa785c | 1045 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1046 | hrtimer_restart(&rq->hrtick_timer); |
1047 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1048 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1049 | } |
1050 | ||
31656519 PZ |
1051 | /* |
1052 | * Called to set the hrtick timer state. | |
1053 | * | |
1054 | * called with rq->lock held and irqs disabled | |
1055 | */ | |
1056 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1057 | { |
31656519 PZ |
1058 | struct hrtimer *timer = &rq->hrtick_timer; |
1059 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1060 | |
cc584b21 | 1061 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1062 | |
1063 | if (rq == this_rq()) { | |
1064 | hrtimer_restart(timer); | |
1065 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1066 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1067 | rq->hrtick_csd_pending = 1; |
1068 | } | |
b328ca18 PZ |
1069 | } |
1070 | ||
1071 | static int | |
1072 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1073 | { | |
1074 | int cpu = (int)(long)hcpu; | |
1075 | ||
1076 | switch (action) { | |
1077 | case CPU_UP_CANCELED: | |
1078 | case CPU_UP_CANCELED_FROZEN: | |
1079 | case CPU_DOWN_PREPARE: | |
1080 | case CPU_DOWN_PREPARE_FROZEN: | |
1081 | case CPU_DEAD: | |
1082 | case CPU_DEAD_FROZEN: | |
31656519 | 1083 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1084 | return NOTIFY_OK; |
1085 | } | |
1086 | ||
1087 | return NOTIFY_DONE; | |
1088 | } | |
1089 | ||
fa748203 | 1090 | static __init void init_hrtick(void) |
b328ca18 PZ |
1091 | { |
1092 | hotcpu_notifier(hotplug_hrtick, 0); | |
1093 | } | |
31656519 PZ |
1094 | #else |
1095 | /* | |
1096 | * Called to set the hrtick timer state. | |
1097 | * | |
1098 | * called with rq->lock held and irqs disabled | |
1099 | */ | |
1100 | static void hrtick_start(struct rq *rq, u64 delay) | |
1101 | { | |
7f1e2ca9 | 1102 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1103 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1104 | } |
b328ca18 | 1105 | |
006c75f1 | 1106 | static inline void init_hrtick(void) |
8f4d37ec | 1107 | { |
8f4d37ec | 1108 | } |
31656519 | 1109 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1110 | |
31656519 | 1111 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1112 | { |
31656519 PZ |
1113 | #ifdef CONFIG_SMP |
1114 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1115 | |
31656519 PZ |
1116 | rq->hrtick_csd.flags = 0; |
1117 | rq->hrtick_csd.func = __hrtick_start; | |
1118 | rq->hrtick_csd.info = rq; | |
1119 | #endif | |
8f4d37ec | 1120 | |
31656519 PZ |
1121 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1122 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1123 | } |
006c75f1 | 1124 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1125 | static inline void hrtick_clear(struct rq *rq) |
1126 | { | |
1127 | } | |
1128 | ||
8f4d37ec PZ |
1129 | static inline void init_rq_hrtick(struct rq *rq) |
1130 | { | |
1131 | } | |
1132 | ||
b328ca18 PZ |
1133 | static inline void init_hrtick(void) |
1134 | { | |
1135 | } | |
006c75f1 | 1136 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1137 | |
c24d20db IM |
1138 | /* |
1139 | * resched_task - mark a task 'to be rescheduled now'. | |
1140 | * | |
1141 | * On UP this means the setting of the need_resched flag, on SMP it | |
1142 | * might also involve a cross-CPU call to trigger the scheduler on | |
1143 | * the target CPU. | |
1144 | */ | |
1145 | #ifdef CONFIG_SMP | |
1146 | ||
1147 | #ifndef tsk_is_polling | |
1148 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1149 | #endif | |
1150 | ||
31656519 | 1151 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1152 | { |
1153 | int cpu; | |
1154 | ||
05fa785c | 1155 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1156 | |
5ed0cec0 | 1157 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1158 | return; |
1159 | ||
5ed0cec0 | 1160 | set_tsk_need_resched(p); |
c24d20db IM |
1161 | |
1162 | cpu = task_cpu(p); | |
1163 | if (cpu == smp_processor_id()) | |
1164 | return; | |
1165 | ||
1166 | /* NEED_RESCHED must be visible before we test polling */ | |
1167 | smp_mb(); | |
1168 | if (!tsk_is_polling(p)) | |
1169 | smp_send_reschedule(cpu); | |
1170 | } | |
1171 | ||
1172 | static void resched_cpu(int cpu) | |
1173 | { | |
1174 | struct rq *rq = cpu_rq(cpu); | |
1175 | unsigned long flags; | |
1176 | ||
05fa785c | 1177 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1178 | return; |
1179 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1180 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1181 | } |
06d8308c TG |
1182 | |
1183 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1184 | /* |
1185 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1186 | * from an idle cpu. This is good for power-savings. | |
1187 | * | |
1188 | * We don't do similar optimization for completely idle system, as | |
1189 | * selecting an idle cpu will add more delays to the timers than intended | |
1190 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1191 | */ | |
1192 | int get_nohz_timer_target(void) | |
1193 | { | |
1194 | int cpu = smp_processor_id(); | |
1195 | int i; | |
1196 | struct sched_domain *sd; | |
1197 | ||
1198 | for_each_domain(cpu, sd) { | |
1199 | for_each_cpu(i, sched_domain_span(sd)) | |
1200 | if (!idle_cpu(i)) | |
1201 | return i; | |
1202 | } | |
1203 | return cpu; | |
1204 | } | |
06d8308c TG |
1205 | /* |
1206 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1207 | * idle CPU then this timer might expire before the next timer event | |
1208 | * which is scheduled to wake up that CPU. In case of a completely | |
1209 | * idle system the next event might even be infinite time into the | |
1210 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1211 | * leaves the inner idle loop so the newly added timer is taken into | |
1212 | * account when the CPU goes back to idle and evaluates the timer | |
1213 | * wheel for the next timer event. | |
1214 | */ | |
1215 | void wake_up_idle_cpu(int cpu) | |
1216 | { | |
1217 | struct rq *rq = cpu_rq(cpu); | |
1218 | ||
1219 | if (cpu == smp_processor_id()) | |
1220 | return; | |
1221 | ||
1222 | /* | |
1223 | * This is safe, as this function is called with the timer | |
1224 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1225 | * to idle and has not yet set rq->curr to idle then it will | |
1226 | * be serialized on the timer wheel base lock and take the new | |
1227 | * timer into account automatically. | |
1228 | */ | |
1229 | if (rq->curr != rq->idle) | |
1230 | return; | |
1231 | ||
1232 | /* | |
1233 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1234 | * lockless. The worst case is that the other CPU runs the | |
1235 | * idle task through an additional NOOP schedule() | |
1236 | */ | |
5ed0cec0 | 1237 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1238 | |
1239 | /* NEED_RESCHED must be visible before we test polling */ | |
1240 | smp_mb(); | |
1241 | if (!tsk_is_polling(rq->idle)) | |
1242 | smp_send_reschedule(cpu); | |
1243 | } | |
39c0cbe2 | 1244 | |
6d6bc0ad | 1245 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1246 | |
e9e9250b PZ |
1247 | static u64 sched_avg_period(void) |
1248 | { | |
1249 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1250 | } | |
1251 | ||
1252 | static void sched_avg_update(struct rq *rq) | |
1253 | { | |
1254 | s64 period = sched_avg_period(); | |
1255 | ||
1256 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1257 | /* |
1258 | * Inline assembly required to prevent the compiler | |
1259 | * optimising this loop into a divmod call. | |
1260 | * See __iter_div_u64_rem() for another example of this. | |
1261 | */ | |
1262 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1263 | rq->age_stamp += period; |
1264 | rq->rt_avg /= 2; | |
1265 | } | |
1266 | } | |
1267 | ||
1268 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1269 | { | |
1270 | rq->rt_avg += rt_delta; | |
1271 | sched_avg_update(rq); | |
1272 | } | |
1273 | ||
6d6bc0ad | 1274 | #else /* !CONFIG_SMP */ |
31656519 | 1275 | static void resched_task(struct task_struct *p) |
c24d20db | 1276 | { |
05fa785c | 1277 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1278 | set_tsk_need_resched(p); |
c24d20db | 1279 | } |
e9e9250b PZ |
1280 | |
1281 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1282 | { | |
1283 | } | |
da2b71ed SS |
1284 | |
1285 | static void sched_avg_update(struct rq *rq) | |
1286 | { | |
1287 | } | |
6d6bc0ad | 1288 | #endif /* CONFIG_SMP */ |
c24d20db | 1289 | |
45bf76df IM |
1290 | #if BITS_PER_LONG == 32 |
1291 | # define WMULT_CONST (~0UL) | |
1292 | #else | |
1293 | # define WMULT_CONST (1UL << 32) | |
1294 | #endif | |
1295 | ||
1296 | #define WMULT_SHIFT 32 | |
1297 | ||
194081eb IM |
1298 | /* |
1299 | * Shift right and round: | |
1300 | */ | |
cf2ab469 | 1301 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1302 | |
a7be37ac PZ |
1303 | /* |
1304 | * delta *= weight / lw | |
1305 | */ | |
cb1c4fc9 | 1306 | static unsigned long |
45bf76df IM |
1307 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1308 | struct load_weight *lw) | |
1309 | { | |
1310 | u64 tmp; | |
1311 | ||
7a232e03 LJ |
1312 | if (!lw->inv_weight) { |
1313 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1314 | lw->inv_weight = 1; | |
1315 | else | |
1316 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1317 | / (lw->weight+1); | |
1318 | } | |
45bf76df IM |
1319 | |
1320 | tmp = (u64)delta_exec * weight; | |
1321 | /* | |
1322 | * Check whether we'd overflow the 64-bit multiplication: | |
1323 | */ | |
194081eb | 1324 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1325 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1326 | WMULT_SHIFT/2); |
1327 | else | |
cf2ab469 | 1328 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1329 | |
ecf691da | 1330 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1331 | } |
1332 | ||
1091985b | 1333 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1334 | { |
1335 | lw->weight += inc; | |
e89996ae | 1336 | lw->inv_weight = 0; |
45bf76df IM |
1337 | } |
1338 | ||
1091985b | 1339 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1340 | { |
1341 | lw->weight -= dec; | |
e89996ae | 1342 | lw->inv_weight = 0; |
45bf76df IM |
1343 | } |
1344 | ||
2069dd75 PZ |
1345 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1346 | { | |
1347 | lw->weight = w; | |
1348 | lw->inv_weight = 0; | |
1349 | } | |
1350 | ||
2dd73a4f PW |
1351 | /* |
1352 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1353 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1354 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1355 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1356 | * scaled version of the new time slice allocation that they receive on time |
1357 | * slice expiry etc. | |
1358 | */ | |
1359 | ||
cce7ade8 PZ |
1360 | #define WEIGHT_IDLEPRIO 3 |
1361 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1362 | |
1363 | /* | |
1364 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1365 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1366 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1367 | * that remained on nice 0. | |
1368 | * | |
1369 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1370 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1371 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1372 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1373 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1374 | */ |
1375 | static const int prio_to_weight[40] = { | |
254753dc IM |
1376 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1377 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1378 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1379 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1380 | /* 0 */ 1024, 820, 655, 526, 423, | |
1381 | /* 5 */ 335, 272, 215, 172, 137, | |
1382 | /* 10 */ 110, 87, 70, 56, 45, | |
1383 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1384 | }; |
1385 | ||
5714d2de IM |
1386 | /* |
1387 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1388 | * | |
1389 | * In cases where the weight does not change often, we can use the | |
1390 | * precalculated inverse to speed up arithmetics by turning divisions | |
1391 | * into multiplications: | |
1392 | */ | |
dd41f596 | 1393 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1394 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1395 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1396 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1397 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1398 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1399 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1400 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1401 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1402 | }; |
2dd73a4f | 1403 | |
ef12fefa BR |
1404 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1405 | enum cpuacct_stat_index { | |
1406 | CPUACCT_STAT_USER, /* ... user mode */ | |
1407 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1408 | ||
1409 | CPUACCT_STAT_NSTATS, | |
1410 | }; | |
1411 | ||
d842de87 SV |
1412 | #ifdef CONFIG_CGROUP_CPUACCT |
1413 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1414 | static void cpuacct_update_stats(struct task_struct *tsk, |
1415 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1416 | #else |
1417 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1418 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1419 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1420 | #endif |
1421 | ||
18d95a28 PZ |
1422 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1423 | { | |
1424 | update_load_add(&rq->load, load); | |
1425 | } | |
1426 | ||
1427 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1428 | { | |
1429 | update_load_sub(&rq->load, load); | |
1430 | } | |
1431 | ||
7940ca36 | 1432 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1433 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1434 | |
1435 | /* | |
1436 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1437 | * leaving it for the final time. | |
1438 | */ | |
eb755805 | 1439 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1440 | { |
1441 | struct task_group *parent, *child; | |
eb755805 | 1442 | int ret; |
c09595f6 PZ |
1443 | |
1444 | rcu_read_lock(); | |
1445 | parent = &root_task_group; | |
1446 | down: | |
eb755805 PZ |
1447 | ret = (*down)(parent, data); |
1448 | if (ret) | |
1449 | goto out_unlock; | |
c09595f6 PZ |
1450 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1451 | parent = child; | |
1452 | goto down; | |
1453 | ||
1454 | up: | |
1455 | continue; | |
1456 | } | |
eb755805 PZ |
1457 | ret = (*up)(parent, data); |
1458 | if (ret) | |
1459 | goto out_unlock; | |
c09595f6 PZ |
1460 | |
1461 | child = parent; | |
1462 | parent = parent->parent; | |
1463 | if (parent) | |
1464 | goto up; | |
eb755805 | 1465 | out_unlock: |
c09595f6 | 1466 | rcu_read_unlock(); |
eb755805 PZ |
1467 | |
1468 | return ret; | |
c09595f6 PZ |
1469 | } |
1470 | ||
eb755805 PZ |
1471 | static int tg_nop(struct task_group *tg, void *data) |
1472 | { | |
1473 | return 0; | |
c09595f6 | 1474 | } |
eb755805 PZ |
1475 | #endif |
1476 | ||
1477 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1478 | /* Used instead of source_load when we know the type == 0 */ |
1479 | static unsigned long weighted_cpuload(const int cpu) | |
1480 | { | |
1481 | return cpu_rq(cpu)->load.weight; | |
1482 | } | |
1483 | ||
1484 | /* | |
1485 | * Return a low guess at the load of a migration-source cpu weighted | |
1486 | * according to the scheduling class and "nice" value. | |
1487 | * | |
1488 | * We want to under-estimate the load of migration sources, to | |
1489 | * balance conservatively. | |
1490 | */ | |
1491 | static unsigned long source_load(int cpu, int type) | |
1492 | { | |
1493 | struct rq *rq = cpu_rq(cpu); | |
1494 | unsigned long total = weighted_cpuload(cpu); | |
1495 | ||
1496 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1497 | return total; | |
1498 | ||
1499 | return min(rq->cpu_load[type-1], total); | |
1500 | } | |
1501 | ||
1502 | /* | |
1503 | * Return a high guess at the load of a migration-target cpu weighted | |
1504 | * according to the scheduling class and "nice" value. | |
1505 | */ | |
1506 | static unsigned long target_load(int cpu, int type) | |
1507 | { | |
1508 | struct rq *rq = cpu_rq(cpu); | |
1509 | unsigned long total = weighted_cpuload(cpu); | |
1510 | ||
1511 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1512 | return total; | |
1513 | ||
1514 | return max(rq->cpu_load[type-1], total); | |
1515 | } | |
1516 | ||
ae154be1 PZ |
1517 | static unsigned long power_of(int cpu) |
1518 | { | |
e51fd5e2 | 1519 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1520 | } |
1521 | ||
eb755805 PZ |
1522 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1523 | ||
1524 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1525 | { | |
1526 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1527 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1528 | |
4cd42620 SR |
1529 | if (nr_running) |
1530 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1531 | else |
1532 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1533 | |
1534 | return rq->avg_load_per_task; | |
1535 | } | |
1536 | ||
1537 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1538 | |
c09595f6 | 1539 | /* |
c8cba857 PZ |
1540 | * Compute the cpu's hierarchical load factor for each task group. |
1541 | * This needs to be done in a top-down fashion because the load of a child | |
1542 | * group is a fraction of its parents load. | |
c09595f6 | 1543 | */ |
eb755805 | 1544 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1545 | { |
c8cba857 | 1546 | unsigned long load; |
eb755805 | 1547 | long cpu = (long)data; |
c09595f6 | 1548 | |
c8cba857 PZ |
1549 | if (!tg->parent) { |
1550 | load = cpu_rq(cpu)->load.weight; | |
1551 | } else { | |
1552 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1553 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1554 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1555 | } | |
c09595f6 | 1556 | |
c8cba857 | 1557 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1558 | |
eb755805 | 1559 | return 0; |
c09595f6 PZ |
1560 | } |
1561 | ||
eb755805 | 1562 | static void update_h_load(long cpu) |
c09595f6 | 1563 | { |
eb755805 | 1564 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1565 | } |
1566 | ||
18d95a28 PZ |
1567 | #endif |
1568 | ||
8f45e2b5 GH |
1569 | #ifdef CONFIG_PREEMPT |
1570 | ||
b78bb868 PZ |
1571 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1572 | ||
70574a99 | 1573 | /* |
8f45e2b5 GH |
1574 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1575 | * way at the expense of forcing extra atomic operations in all | |
1576 | * invocations. This assures that the double_lock is acquired using the | |
1577 | * same underlying policy as the spinlock_t on this architecture, which | |
1578 | * reduces latency compared to the unfair variant below. However, it | |
1579 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1580 | */ |
8f45e2b5 GH |
1581 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1582 | __releases(this_rq->lock) | |
1583 | __acquires(busiest->lock) | |
1584 | __acquires(this_rq->lock) | |
1585 | { | |
05fa785c | 1586 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1587 | double_rq_lock(this_rq, busiest); |
1588 | ||
1589 | return 1; | |
1590 | } | |
1591 | ||
1592 | #else | |
1593 | /* | |
1594 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1595 | * latency by eliminating extra atomic operations when the locks are | |
1596 | * already in proper order on entry. This favors lower cpu-ids and will | |
1597 | * grant the double lock to lower cpus over higher ids under contention, | |
1598 | * regardless of entry order into the function. | |
1599 | */ | |
1600 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1601 | __releases(this_rq->lock) |
1602 | __acquires(busiest->lock) | |
1603 | __acquires(this_rq->lock) | |
1604 | { | |
1605 | int ret = 0; | |
1606 | ||
05fa785c | 1607 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1608 | if (busiest < this_rq) { |
05fa785c TG |
1609 | raw_spin_unlock(&this_rq->lock); |
1610 | raw_spin_lock(&busiest->lock); | |
1611 | raw_spin_lock_nested(&this_rq->lock, | |
1612 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1613 | ret = 1; |
1614 | } else | |
05fa785c TG |
1615 | raw_spin_lock_nested(&busiest->lock, |
1616 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1617 | } |
1618 | return ret; | |
1619 | } | |
1620 | ||
8f45e2b5 GH |
1621 | #endif /* CONFIG_PREEMPT */ |
1622 | ||
1623 | /* | |
1624 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1625 | */ | |
1626 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1627 | { | |
1628 | if (unlikely(!irqs_disabled())) { | |
1629 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1630 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1631 | BUG_ON(1); |
1632 | } | |
1633 | ||
1634 | return _double_lock_balance(this_rq, busiest); | |
1635 | } | |
1636 | ||
70574a99 AD |
1637 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1638 | __releases(busiest->lock) | |
1639 | { | |
05fa785c | 1640 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1641 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1642 | } | |
1e3c88bd PZ |
1643 | |
1644 | /* | |
1645 | * double_rq_lock - safely lock two runqueues | |
1646 | * | |
1647 | * Note this does not disable interrupts like task_rq_lock, | |
1648 | * you need to do so manually before calling. | |
1649 | */ | |
1650 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1651 | __acquires(rq1->lock) | |
1652 | __acquires(rq2->lock) | |
1653 | { | |
1654 | BUG_ON(!irqs_disabled()); | |
1655 | if (rq1 == rq2) { | |
1656 | raw_spin_lock(&rq1->lock); | |
1657 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1658 | } else { | |
1659 | if (rq1 < rq2) { | |
1660 | raw_spin_lock(&rq1->lock); | |
1661 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1662 | } else { | |
1663 | raw_spin_lock(&rq2->lock); | |
1664 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1665 | } | |
1666 | } | |
1e3c88bd PZ |
1667 | } |
1668 | ||
1669 | /* | |
1670 | * double_rq_unlock - safely unlock two runqueues | |
1671 | * | |
1672 | * Note this does not restore interrupts like task_rq_unlock, | |
1673 | * you need to do so manually after calling. | |
1674 | */ | |
1675 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1676 | __releases(rq1->lock) | |
1677 | __releases(rq2->lock) | |
1678 | { | |
1679 | raw_spin_unlock(&rq1->lock); | |
1680 | if (rq1 != rq2) | |
1681 | raw_spin_unlock(&rq2->lock); | |
1682 | else | |
1683 | __release(rq2->lock); | |
1684 | } | |
1685 | ||
18d95a28 PZ |
1686 | #endif |
1687 | ||
74f5187a | 1688 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1689 | static void update_sysctl(void); |
acb4a848 | 1690 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1691 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1692 | |
cd29fe6f PZ |
1693 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1694 | { | |
1695 | set_task_rq(p, cpu); | |
1696 | #ifdef CONFIG_SMP | |
1697 | /* | |
1698 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1699 | * successfuly executed on another CPU. We must ensure that updates of | |
1700 | * per-task data have been completed by this moment. | |
1701 | */ | |
1702 | smp_wmb(); | |
1703 | task_thread_info(p)->cpu = cpu; | |
1704 | #endif | |
1705 | } | |
dce48a84 | 1706 | |
1e3c88bd | 1707 | static const struct sched_class rt_sched_class; |
dd41f596 | 1708 | |
34f971f6 | 1709 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1710 | #define for_each_class(class) \ |
1711 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1712 | |
1e3c88bd PZ |
1713 | #include "sched_stats.h" |
1714 | ||
c09595f6 | 1715 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1716 | { |
1717 | rq->nr_running++; | |
9c217245 IM |
1718 | } |
1719 | ||
c09595f6 | 1720 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1721 | { |
1722 | rq->nr_running--; | |
9c217245 IM |
1723 | } |
1724 | ||
45bf76df IM |
1725 | static void set_load_weight(struct task_struct *p) |
1726 | { | |
dd41f596 IM |
1727 | /* |
1728 | * SCHED_IDLE tasks get minimal weight: | |
1729 | */ | |
1730 | if (p->policy == SCHED_IDLE) { | |
1731 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1732 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1733 | return; | |
1734 | } | |
71f8bd46 | 1735 | |
dd41f596 IM |
1736 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1737 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1738 | } |
1739 | ||
371fd7e7 | 1740 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1741 | { |
a64692a3 | 1742 | update_rq_clock(rq); |
dd41f596 | 1743 | sched_info_queued(p); |
371fd7e7 | 1744 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1745 | p->se.on_rq = 1; |
71f8bd46 IM |
1746 | } |
1747 | ||
371fd7e7 | 1748 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1749 | { |
a64692a3 | 1750 | update_rq_clock(rq); |
46ac22ba | 1751 | sched_info_dequeued(p); |
371fd7e7 | 1752 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1753 | p->se.on_rq = 0; |
71f8bd46 IM |
1754 | } |
1755 | ||
1e3c88bd PZ |
1756 | /* |
1757 | * activate_task - move a task to the runqueue. | |
1758 | */ | |
371fd7e7 | 1759 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1760 | { |
1761 | if (task_contributes_to_load(p)) | |
1762 | rq->nr_uninterruptible--; | |
1763 | ||
371fd7e7 | 1764 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1765 | inc_nr_running(rq); |
1766 | } | |
1767 | ||
1768 | /* | |
1769 | * deactivate_task - remove a task from the runqueue. | |
1770 | */ | |
371fd7e7 | 1771 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1772 | { |
1773 | if (task_contributes_to_load(p)) | |
1774 | rq->nr_uninterruptible++; | |
1775 | ||
371fd7e7 | 1776 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1777 | dec_nr_running(rq); |
1778 | } | |
1779 | ||
b52bfee4 VP |
1780 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1781 | ||
305e6835 VP |
1782 | /* |
1783 | * There are no locks covering percpu hardirq/softirq time. | |
1784 | * They are only modified in account_system_vtime, on corresponding CPU | |
1785 | * with interrupts disabled. So, writes are safe. | |
1786 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1787 | * This may result in other CPU reading this CPU's irq time and can | |
1788 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1789 | * or new value with a side effect of accounting a slice of irq time to wrong |
1790 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1791 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1792 | */ |
b52bfee4 VP |
1793 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1794 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1795 | ||
1796 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1797 | static int sched_clock_irqtime; | |
1798 | ||
1799 | void enable_sched_clock_irqtime(void) | |
1800 | { | |
1801 | sched_clock_irqtime = 1; | |
1802 | } | |
1803 | ||
1804 | void disable_sched_clock_irqtime(void) | |
1805 | { | |
1806 | sched_clock_irqtime = 0; | |
1807 | } | |
1808 | ||
8e92c201 PZ |
1809 | #ifndef CONFIG_64BIT |
1810 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1811 | ||
1812 | static inline void irq_time_write_begin(void) | |
1813 | { | |
1814 | __this_cpu_inc(irq_time_seq.sequence); | |
1815 | smp_wmb(); | |
1816 | } | |
1817 | ||
1818 | static inline void irq_time_write_end(void) | |
1819 | { | |
1820 | smp_wmb(); | |
1821 | __this_cpu_inc(irq_time_seq.sequence); | |
1822 | } | |
1823 | ||
1824 | static inline u64 irq_time_read(int cpu) | |
1825 | { | |
1826 | u64 irq_time; | |
1827 | unsigned seq; | |
1828 | ||
1829 | do { | |
1830 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1831 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1832 | per_cpu(cpu_hardirq_time, cpu); | |
1833 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1834 | ||
1835 | return irq_time; | |
1836 | } | |
1837 | #else /* CONFIG_64BIT */ | |
1838 | static inline void irq_time_write_begin(void) | |
1839 | { | |
1840 | } | |
1841 | ||
1842 | static inline void irq_time_write_end(void) | |
1843 | { | |
1844 | } | |
1845 | ||
1846 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1847 | { |
305e6835 VP |
1848 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1849 | } | |
8e92c201 | 1850 | #endif /* CONFIG_64BIT */ |
305e6835 | 1851 | |
fe44d621 PZ |
1852 | /* |
1853 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1854 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1855 | */ | |
b52bfee4 VP |
1856 | void account_system_vtime(struct task_struct *curr) |
1857 | { | |
1858 | unsigned long flags; | |
fe44d621 | 1859 | s64 delta; |
b52bfee4 | 1860 | int cpu; |
b52bfee4 VP |
1861 | |
1862 | if (!sched_clock_irqtime) | |
1863 | return; | |
1864 | ||
1865 | local_irq_save(flags); | |
1866 | ||
b52bfee4 | 1867 | cpu = smp_processor_id(); |
fe44d621 PZ |
1868 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1869 | __this_cpu_add(irq_start_time, delta); | |
1870 | ||
8e92c201 | 1871 | irq_time_write_begin(); |
b52bfee4 VP |
1872 | /* |
1873 | * We do not account for softirq time from ksoftirqd here. | |
1874 | * We want to continue accounting softirq time to ksoftirqd thread | |
1875 | * in that case, so as not to confuse scheduler with a special task | |
1876 | * that do not consume any time, but still wants to run. | |
1877 | */ | |
1878 | if (hardirq_count()) | |
fe44d621 | 1879 | __this_cpu_add(cpu_hardirq_time, delta); |
b52bfee4 | 1880 | else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD)) |
fe44d621 | 1881 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1882 | |
8e92c201 | 1883 | irq_time_write_end(); |
b52bfee4 VP |
1884 | local_irq_restore(flags); |
1885 | } | |
b7dadc38 | 1886 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1887 | |
fe44d621 | 1888 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1889 | { |
fe44d621 PZ |
1890 | s64 irq_delta; |
1891 | ||
8e92c201 | 1892 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
1893 | |
1894 | /* | |
1895 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
1896 | * this case when a previous update_rq_clock() happened inside a | |
1897 | * {soft,}irq region. | |
1898 | * | |
1899 | * When this happens, we stop ->clock_task and only update the | |
1900 | * prev_irq_time stamp to account for the part that fit, so that a next | |
1901 | * update will consume the rest. This ensures ->clock_task is | |
1902 | * monotonic. | |
1903 | * | |
1904 | * It does however cause some slight miss-attribution of {soft,}irq | |
1905 | * time, a more accurate solution would be to update the irq_time using | |
1906 | * the current rq->clock timestamp, except that would require using | |
1907 | * atomic ops. | |
1908 | */ | |
1909 | if (irq_delta > delta) | |
1910 | irq_delta = delta; | |
1911 | ||
1912 | rq->prev_irq_time += irq_delta; | |
1913 | delta -= irq_delta; | |
1914 | rq->clock_task += delta; | |
1915 | ||
1916 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
1917 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
1918 | } |
1919 | ||
fe44d621 | 1920 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 1921 | |
fe44d621 | 1922 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 1923 | { |
fe44d621 | 1924 | rq->clock_task += delta; |
305e6835 VP |
1925 | } |
1926 | ||
fe44d621 | 1927 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 1928 | |
1e3c88bd PZ |
1929 | #include "sched_idletask.c" |
1930 | #include "sched_fair.c" | |
1931 | #include "sched_rt.c" | |
5091faa4 | 1932 | #include "sched_autogroup.c" |
34f971f6 | 1933 | #include "sched_stoptask.c" |
1e3c88bd PZ |
1934 | #ifdef CONFIG_SCHED_DEBUG |
1935 | # include "sched_debug.c" | |
1936 | #endif | |
1937 | ||
34f971f6 PZ |
1938 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
1939 | { | |
1940 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
1941 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
1942 | ||
1943 | if (stop) { | |
1944 | /* | |
1945 | * Make it appear like a SCHED_FIFO task, its something | |
1946 | * userspace knows about and won't get confused about. | |
1947 | * | |
1948 | * Also, it will make PI more or less work without too | |
1949 | * much confusion -- but then, stop work should not | |
1950 | * rely on PI working anyway. | |
1951 | */ | |
1952 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
1953 | ||
1954 | stop->sched_class = &stop_sched_class; | |
1955 | } | |
1956 | ||
1957 | cpu_rq(cpu)->stop = stop; | |
1958 | ||
1959 | if (old_stop) { | |
1960 | /* | |
1961 | * Reset it back to a normal scheduling class so that | |
1962 | * it can die in pieces. | |
1963 | */ | |
1964 | old_stop->sched_class = &rt_sched_class; | |
1965 | } | |
1966 | } | |
1967 | ||
14531189 | 1968 | /* |
dd41f596 | 1969 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1970 | */ |
14531189 IM |
1971 | static inline int __normal_prio(struct task_struct *p) |
1972 | { | |
dd41f596 | 1973 | return p->static_prio; |
14531189 IM |
1974 | } |
1975 | ||
b29739f9 IM |
1976 | /* |
1977 | * Calculate the expected normal priority: i.e. priority | |
1978 | * without taking RT-inheritance into account. Might be | |
1979 | * boosted by interactivity modifiers. Changes upon fork, | |
1980 | * setprio syscalls, and whenever the interactivity | |
1981 | * estimator recalculates. | |
1982 | */ | |
36c8b586 | 1983 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1984 | { |
1985 | int prio; | |
1986 | ||
e05606d3 | 1987 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1988 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1989 | else | |
1990 | prio = __normal_prio(p); | |
1991 | return prio; | |
1992 | } | |
1993 | ||
1994 | /* | |
1995 | * Calculate the current priority, i.e. the priority | |
1996 | * taken into account by the scheduler. This value might | |
1997 | * be boosted by RT tasks, or might be boosted by | |
1998 | * interactivity modifiers. Will be RT if the task got | |
1999 | * RT-boosted. If not then it returns p->normal_prio. | |
2000 | */ | |
36c8b586 | 2001 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2002 | { |
2003 | p->normal_prio = normal_prio(p); | |
2004 | /* | |
2005 | * If we are RT tasks or we were boosted to RT priority, | |
2006 | * keep the priority unchanged. Otherwise, update priority | |
2007 | * to the normal priority: | |
2008 | */ | |
2009 | if (!rt_prio(p->prio)) | |
2010 | return p->normal_prio; | |
2011 | return p->prio; | |
2012 | } | |
2013 | ||
1da177e4 LT |
2014 | /** |
2015 | * task_curr - is this task currently executing on a CPU? | |
2016 | * @p: the task in question. | |
2017 | */ | |
36c8b586 | 2018 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2019 | { |
2020 | return cpu_curr(task_cpu(p)) == p; | |
2021 | } | |
2022 | ||
cb469845 SR |
2023 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2024 | const struct sched_class *prev_class, | |
2025 | int oldprio, int running) | |
2026 | { | |
2027 | if (prev_class != p->sched_class) { | |
2028 | if (prev_class->switched_from) | |
2029 | prev_class->switched_from(rq, p, running); | |
2030 | p->sched_class->switched_to(rq, p, running); | |
2031 | } else | |
2032 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
2033 | } | |
2034 | ||
1e5a7405 PZ |
2035 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2036 | { | |
2037 | const struct sched_class *class; | |
2038 | ||
2039 | if (p->sched_class == rq->curr->sched_class) { | |
2040 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2041 | } else { | |
2042 | for_each_class(class) { | |
2043 | if (class == rq->curr->sched_class) | |
2044 | break; | |
2045 | if (class == p->sched_class) { | |
2046 | resched_task(rq->curr); | |
2047 | break; | |
2048 | } | |
2049 | } | |
2050 | } | |
2051 | ||
2052 | /* | |
2053 | * A queue event has occurred, and we're going to schedule. In | |
2054 | * this case, we can save a useless back to back clock update. | |
2055 | */ | |
f26f9aff | 2056 | if (rq->curr->se.on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2057 | rq->skip_clock_update = 1; |
2058 | } | |
2059 | ||
1da177e4 | 2060 | #ifdef CONFIG_SMP |
cc367732 IM |
2061 | /* |
2062 | * Is this task likely cache-hot: | |
2063 | */ | |
e7693a36 | 2064 | static int |
cc367732 IM |
2065 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2066 | { | |
2067 | s64 delta; | |
2068 | ||
e6c8fba7 PZ |
2069 | if (p->sched_class != &fair_sched_class) |
2070 | return 0; | |
2071 | ||
ef8002f6 NR |
2072 | if (unlikely(p->policy == SCHED_IDLE)) |
2073 | return 0; | |
2074 | ||
f540a608 IM |
2075 | /* |
2076 | * Buddy candidates are cache hot: | |
2077 | */ | |
f685ceac | 2078 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2079 | (&p->se == cfs_rq_of(&p->se)->next || |
2080 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2081 | return 1; |
2082 | ||
6bc1665b IM |
2083 | if (sysctl_sched_migration_cost == -1) |
2084 | return 1; | |
2085 | if (sysctl_sched_migration_cost == 0) | |
2086 | return 0; | |
2087 | ||
cc367732 IM |
2088 | delta = now - p->se.exec_start; |
2089 | ||
2090 | return delta < (s64)sysctl_sched_migration_cost; | |
2091 | } | |
2092 | ||
dd41f596 | 2093 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2094 | { |
e2912009 PZ |
2095 | #ifdef CONFIG_SCHED_DEBUG |
2096 | /* | |
2097 | * We should never call set_task_cpu() on a blocked task, | |
2098 | * ttwu() will sort out the placement. | |
2099 | */ | |
077614ee PZ |
2100 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2101 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2102 | #endif |
2103 | ||
de1d7286 | 2104 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2105 | |
0c69774e PZ |
2106 | if (task_cpu(p) != new_cpu) { |
2107 | p->se.nr_migrations++; | |
2108 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2109 | } | |
dd41f596 IM |
2110 | |
2111 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2112 | } |
2113 | ||
969c7921 | 2114 | struct migration_arg { |
36c8b586 | 2115 | struct task_struct *task; |
1da177e4 | 2116 | int dest_cpu; |
70b97a7f | 2117 | }; |
1da177e4 | 2118 | |
969c7921 TH |
2119 | static int migration_cpu_stop(void *data); |
2120 | ||
1da177e4 LT |
2121 | /* |
2122 | * The task's runqueue lock must be held. | |
2123 | * Returns true if you have to wait for migration thread. | |
2124 | */ | |
b7a2b39d | 2125 | static bool migrate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 2126 | { |
1da177e4 LT |
2127 | /* |
2128 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2129 | * the next wake-up will properly place the task. |
1da177e4 | 2130 | */ |
969c7921 | 2131 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2132 | } |
2133 | ||
2134 | /* | |
2135 | * wait_task_inactive - wait for a thread to unschedule. | |
2136 | * | |
85ba2d86 RM |
2137 | * If @match_state is nonzero, it's the @p->state value just checked and |
2138 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2139 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2140 | * we return a positive number (its total switch count). If a second call | |
2141 | * a short while later returns the same number, the caller can be sure that | |
2142 | * @p has remained unscheduled the whole time. | |
2143 | * | |
1da177e4 LT |
2144 | * The caller must ensure that the task *will* unschedule sometime soon, |
2145 | * else this function might spin for a *long* time. This function can't | |
2146 | * be called with interrupts off, or it may introduce deadlock with | |
2147 | * smp_call_function() if an IPI is sent by the same process we are | |
2148 | * waiting to become inactive. | |
2149 | */ | |
85ba2d86 | 2150 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2151 | { |
2152 | unsigned long flags; | |
dd41f596 | 2153 | int running, on_rq; |
85ba2d86 | 2154 | unsigned long ncsw; |
70b97a7f | 2155 | struct rq *rq; |
1da177e4 | 2156 | |
3a5c359a AK |
2157 | for (;;) { |
2158 | /* | |
2159 | * We do the initial early heuristics without holding | |
2160 | * any task-queue locks at all. We'll only try to get | |
2161 | * the runqueue lock when things look like they will | |
2162 | * work out! | |
2163 | */ | |
2164 | rq = task_rq(p); | |
fa490cfd | 2165 | |
3a5c359a AK |
2166 | /* |
2167 | * If the task is actively running on another CPU | |
2168 | * still, just relax and busy-wait without holding | |
2169 | * any locks. | |
2170 | * | |
2171 | * NOTE! Since we don't hold any locks, it's not | |
2172 | * even sure that "rq" stays as the right runqueue! | |
2173 | * But we don't care, since "task_running()" will | |
2174 | * return false if the runqueue has changed and p | |
2175 | * is actually now running somewhere else! | |
2176 | */ | |
85ba2d86 RM |
2177 | while (task_running(rq, p)) { |
2178 | if (match_state && unlikely(p->state != match_state)) | |
2179 | return 0; | |
3a5c359a | 2180 | cpu_relax(); |
85ba2d86 | 2181 | } |
fa490cfd | 2182 | |
3a5c359a AK |
2183 | /* |
2184 | * Ok, time to look more closely! We need the rq | |
2185 | * lock now, to be *sure*. If we're wrong, we'll | |
2186 | * just go back and repeat. | |
2187 | */ | |
2188 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2189 | trace_sched_wait_task(p); |
3a5c359a AK |
2190 | running = task_running(rq, p); |
2191 | on_rq = p->se.on_rq; | |
85ba2d86 | 2192 | ncsw = 0; |
f31e11d8 | 2193 | if (!match_state || p->state == match_state) |
93dcf55f | 2194 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2195 | task_rq_unlock(rq, &flags); |
fa490cfd | 2196 | |
85ba2d86 RM |
2197 | /* |
2198 | * If it changed from the expected state, bail out now. | |
2199 | */ | |
2200 | if (unlikely(!ncsw)) | |
2201 | break; | |
2202 | ||
3a5c359a AK |
2203 | /* |
2204 | * Was it really running after all now that we | |
2205 | * checked with the proper locks actually held? | |
2206 | * | |
2207 | * Oops. Go back and try again.. | |
2208 | */ | |
2209 | if (unlikely(running)) { | |
2210 | cpu_relax(); | |
2211 | continue; | |
2212 | } | |
fa490cfd | 2213 | |
3a5c359a AK |
2214 | /* |
2215 | * It's not enough that it's not actively running, | |
2216 | * it must be off the runqueue _entirely_, and not | |
2217 | * preempted! | |
2218 | * | |
80dd99b3 | 2219 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2220 | * running right now), it's preempted, and we should |
2221 | * yield - it could be a while. | |
2222 | */ | |
2223 | if (unlikely(on_rq)) { | |
2224 | schedule_timeout_uninterruptible(1); | |
2225 | continue; | |
2226 | } | |
fa490cfd | 2227 | |
3a5c359a AK |
2228 | /* |
2229 | * Ahh, all good. It wasn't running, and it wasn't | |
2230 | * runnable, which means that it will never become | |
2231 | * running in the future either. We're all done! | |
2232 | */ | |
2233 | break; | |
2234 | } | |
85ba2d86 RM |
2235 | |
2236 | return ncsw; | |
1da177e4 LT |
2237 | } |
2238 | ||
2239 | /*** | |
2240 | * kick_process - kick a running thread to enter/exit the kernel | |
2241 | * @p: the to-be-kicked thread | |
2242 | * | |
2243 | * Cause a process which is running on another CPU to enter | |
2244 | * kernel-mode, without any delay. (to get signals handled.) | |
2245 | * | |
2246 | * NOTE: this function doesnt have to take the runqueue lock, | |
2247 | * because all it wants to ensure is that the remote task enters | |
2248 | * the kernel. If the IPI races and the task has been migrated | |
2249 | * to another CPU then no harm is done and the purpose has been | |
2250 | * achieved as well. | |
2251 | */ | |
36c8b586 | 2252 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2253 | { |
2254 | int cpu; | |
2255 | ||
2256 | preempt_disable(); | |
2257 | cpu = task_cpu(p); | |
2258 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2259 | smp_send_reschedule(cpu); | |
2260 | preempt_enable(); | |
2261 | } | |
b43e3521 | 2262 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2263 | #endif /* CONFIG_SMP */ |
1da177e4 | 2264 | |
0793a61d TG |
2265 | /** |
2266 | * task_oncpu_function_call - call a function on the cpu on which a task runs | |
2267 | * @p: the task to evaluate | |
2268 | * @func: the function to be called | |
2269 | * @info: the function call argument | |
2270 | * | |
2271 | * Calls the function @func when the task is currently running. This might | |
2272 | * be on the current CPU, which just calls the function directly | |
2273 | */ | |
2274 | void task_oncpu_function_call(struct task_struct *p, | |
2275 | void (*func) (void *info), void *info) | |
2276 | { | |
2277 | int cpu; | |
2278 | ||
2279 | preempt_disable(); | |
2280 | cpu = task_cpu(p); | |
2281 | if (task_curr(p)) | |
2282 | smp_call_function_single(cpu, func, info, 1); | |
2283 | preempt_enable(); | |
2284 | } | |
2285 | ||
970b13ba | 2286 | #ifdef CONFIG_SMP |
30da688e ON |
2287 | /* |
2288 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2289 | */ | |
5da9a0fb PZ |
2290 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2291 | { | |
2292 | int dest_cpu; | |
2293 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2294 | ||
2295 | /* Look for allowed, online CPU in same node. */ | |
2296 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2297 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2298 | return dest_cpu; | |
2299 | ||
2300 | /* Any allowed, online CPU? */ | |
2301 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2302 | if (dest_cpu < nr_cpu_ids) | |
2303 | return dest_cpu; | |
2304 | ||
2305 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2306 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2307 | /* | |
2308 | * Don't tell them about moving exiting tasks or | |
2309 | * kernel threads (both mm NULL), since they never | |
2310 | * leave kernel. | |
2311 | */ | |
2312 | if (p->mm && printk_ratelimit()) { | |
2313 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2314 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2315 | } |
2316 | ||
2317 | return dest_cpu; | |
2318 | } | |
2319 | ||
e2912009 | 2320 | /* |
30da688e | 2321 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2322 | */ |
970b13ba | 2323 | static inline |
0017d735 | 2324 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2325 | { |
0017d735 | 2326 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2327 | |
2328 | /* | |
2329 | * In order not to call set_task_cpu() on a blocking task we need | |
2330 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2331 | * cpu. | |
2332 | * | |
2333 | * Since this is common to all placement strategies, this lives here. | |
2334 | * | |
2335 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2336 | * not worry about this generic constraint ] | |
2337 | */ | |
2338 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2339 | !cpu_online(cpu))) |
5da9a0fb | 2340 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2341 | |
2342 | return cpu; | |
970b13ba | 2343 | } |
09a40af5 MG |
2344 | |
2345 | static void update_avg(u64 *avg, u64 sample) | |
2346 | { | |
2347 | s64 diff = sample - *avg; | |
2348 | *avg += diff >> 3; | |
2349 | } | |
970b13ba PZ |
2350 | #endif |
2351 | ||
9ed3811a TH |
2352 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2353 | bool is_sync, bool is_migrate, bool is_local, | |
2354 | unsigned long en_flags) | |
2355 | { | |
2356 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2357 | if (is_sync) | |
2358 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2359 | if (is_migrate) | |
2360 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2361 | if (is_local) | |
2362 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2363 | else | |
2364 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2365 | ||
2366 | activate_task(rq, p, en_flags); | |
2367 | } | |
2368 | ||
2369 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2370 | int wake_flags, bool success) | |
2371 | { | |
2372 | trace_sched_wakeup(p, success); | |
2373 | check_preempt_curr(rq, p, wake_flags); | |
2374 | ||
2375 | p->state = TASK_RUNNING; | |
2376 | #ifdef CONFIG_SMP | |
2377 | if (p->sched_class->task_woken) | |
2378 | p->sched_class->task_woken(rq, p); | |
2379 | ||
2380 | if (unlikely(rq->idle_stamp)) { | |
2381 | u64 delta = rq->clock - rq->idle_stamp; | |
2382 | u64 max = 2*sysctl_sched_migration_cost; | |
2383 | ||
2384 | if (delta > max) | |
2385 | rq->avg_idle = max; | |
2386 | else | |
2387 | update_avg(&rq->avg_idle, delta); | |
2388 | rq->idle_stamp = 0; | |
2389 | } | |
2390 | #endif | |
21aa9af0 TH |
2391 | /* if a worker is waking up, notify workqueue */ |
2392 | if ((p->flags & PF_WQ_WORKER) && success) | |
2393 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2394 | } |
2395 | ||
2396 | /** | |
1da177e4 | 2397 | * try_to_wake_up - wake up a thread |
9ed3811a | 2398 | * @p: the thread to be awakened |
1da177e4 | 2399 | * @state: the mask of task states that can be woken |
9ed3811a | 2400 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2401 | * |
2402 | * Put it on the run-queue if it's not already there. The "current" | |
2403 | * thread is always on the run-queue (except when the actual | |
2404 | * re-schedule is in progress), and as such you're allowed to do | |
2405 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2406 | * runnable without the overhead of this. | |
2407 | * | |
9ed3811a TH |
2408 | * Returns %true if @p was woken up, %false if it was already running |
2409 | * or @state didn't match @p's state. | |
1da177e4 | 2410 | */ |
7d478721 PZ |
2411 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2412 | int wake_flags) | |
1da177e4 | 2413 | { |
cc367732 | 2414 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2415 | unsigned long flags; |
371fd7e7 | 2416 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2417 | struct rq *rq; |
1da177e4 | 2418 | |
e9c84311 | 2419 | this_cpu = get_cpu(); |
2398f2c6 | 2420 | |
04e2f174 | 2421 | smp_wmb(); |
ab3b3aa5 | 2422 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2423 | if (!(p->state & state)) |
1da177e4 LT |
2424 | goto out; |
2425 | ||
dd41f596 | 2426 | if (p->se.on_rq) |
1da177e4 LT |
2427 | goto out_running; |
2428 | ||
2429 | cpu = task_cpu(p); | |
cc367732 | 2430 | orig_cpu = cpu; |
1da177e4 LT |
2431 | |
2432 | #ifdef CONFIG_SMP | |
2433 | if (unlikely(task_running(rq, p))) | |
2434 | goto out_activate; | |
2435 | ||
e9c84311 PZ |
2436 | /* |
2437 | * In order to handle concurrent wakeups and release the rq->lock | |
2438 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2439 | * |
2440 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2441 | */ |
cc87f76a PZ |
2442 | if (task_contributes_to_load(p)) { |
2443 | if (likely(cpu_online(orig_cpu))) | |
2444 | rq->nr_uninterruptible--; | |
2445 | else | |
2446 | this_rq()->nr_uninterruptible--; | |
2447 | } | |
e9c84311 | 2448 | p->state = TASK_WAKING; |
efbbd05a | 2449 | |
371fd7e7 | 2450 | if (p->sched_class->task_waking) { |
efbbd05a | 2451 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2452 | en_flags |= ENQUEUE_WAKING; |
2453 | } | |
efbbd05a | 2454 | |
0017d735 PZ |
2455 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2456 | if (cpu != orig_cpu) | |
5d2f5a61 | 2457 | set_task_cpu(p, cpu); |
0017d735 | 2458 | __task_rq_unlock(rq); |
ab19cb23 | 2459 | |
0970d299 PZ |
2460 | rq = cpu_rq(cpu); |
2461 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2462 | |
0970d299 PZ |
2463 | /* |
2464 | * We migrated the task without holding either rq->lock, however | |
2465 | * since the task is not on the task list itself, nobody else | |
2466 | * will try and migrate the task, hence the rq should match the | |
2467 | * cpu we just moved it to. | |
2468 | */ | |
2469 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2470 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2471 | |
e7693a36 GH |
2472 | #ifdef CONFIG_SCHEDSTATS |
2473 | schedstat_inc(rq, ttwu_count); | |
2474 | if (cpu == this_cpu) | |
2475 | schedstat_inc(rq, ttwu_local); | |
2476 | else { | |
2477 | struct sched_domain *sd; | |
2478 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2479 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2480 | schedstat_inc(sd, ttwu_wake_remote); |
2481 | break; | |
2482 | } | |
2483 | } | |
2484 | } | |
6d6bc0ad | 2485 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2486 | |
1da177e4 LT |
2487 | out_activate: |
2488 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2489 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2490 | cpu == this_cpu, en_flags); | |
1da177e4 | 2491 | success = 1; |
1da177e4 | 2492 | out_running: |
9ed3811a | 2493 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2494 | out: |
2495 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2496 | put_cpu(); |
1da177e4 LT |
2497 | |
2498 | return success; | |
2499 | } | |
2500 | ||
21aa9af0 TH |
2501 | /** |
2502 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2503 | * @p: the thread to be awakened | |
2504 | * | |
2505 | * Put @p on the run-queue if it's not alredy there. The caller must | |
2506 | * ensure that this_rq() is locked, @p is bound to this_rq() and not | |
2507 | * the current task. this_rq() stays locked over invocation. | |
2508 | */ | |
2509 | static void try_to_wake_up_local(struct task_struct *p) | |
2510 | { | |
2511 | struct rq *rq = task_rq(p); | |
2512 | bool success = false; | |
2513 | ||
2514 | BUG_ON(rq != this_rq()); | |
2515 | BUG_ON(p == current); | |
2516 | lockdep_assert_held(&rq->lock); | |
2517 | ||
2518 | if (!(p->state & TASK_NORMAL)) | |
2519 | return; | |
2520 | ||
2521 | if (!p->se.on_rq) { | |
2522 | if (likely(!task_running(rq, p))) { | |
2523 | schedstat_inc(rq, ttwu_count); | |
2524 | schedstat_inc(rq, ttwu_local); | |
2525 | } | |
2526 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2527 | success = true; | |
2528 | } | |
2529 | ttwu_post_activation(p, rq, 0, success); | |
2530 | } | |
2531 | ||
50fa610a DH |
2532 | /** |
2533 | * wake_up_process - Wake up a specific process | |
2534 | * @p: The process to be woken up. | |
2535 | * | |
2536 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2537 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2538 | * running. | |
2539 | * | |
2540 | * It may be assumed that this function implies a write memory barrier before | |
2541 | * changing the task state if and only if any tasks are woken up. | |
2542 | */ | |
7ad5b3a5 | 2543 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2544 | { |
d9514f6c | 2545 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2546 | } |
1da177e4 LT |
2547 | EXPORT_SYMBOL(wake_up_process); |
2548 | ||
7ad5b3a5 | 2549 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2550 | { |
2551 | return try_to_wake_up(p, state, 0); | |
2552 | } | |
2553 | ||
1da177e4 LT |
2554 | /* |
2555 | * Perform scheduler related setup for a newly forked process p. | |
2556 | * p is forked by current. | |
dd41f596 IM |
2557 | * |
2558 | * __sched_fork() is basic setup used by init_idle() too: | |
2559 | */ | |
2560 | static void __sched_fork(struct task_struct *p) | |
2561 | { | |
dd41f596 IM |
2562 | p->se.exec_start = 0; |
2563 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2564 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2565 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2566 | |
2567 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2568 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2569 | #endif |
476d139c | 2570 | |
fa717060 | 2571 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2572 | p->se.on_rq = 0; |
4a55bd5e | 2573 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2574 | |
e107be36 AK |
2575 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2576 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2577 | #endif | |
dd41f596 IM |
2578 | } |
2579 | ||
2580 | /* | |
2581 | * fork()/clone()-time setup: | |
2582 | */ | |
2583 | void sched_fork(struct task_struct *p, int clone_flags) | |
2584 | { | |
2585 | int cpu = get_cpu(); | |
2586 | ||
2587 | __sched_fork(p); | |
06b83b5f | 2588 | /* |
0017d735 | 2589 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2590 | * nobody will actually run it, and a signal or other external |
2591 | * event cannot wake it up and insert it on the runqueue either. | |
2592 | */ | |
0017d735 | 2593 | p->state = TASK_RUNNING; |
dd41f596 | 2594 | |
b9dc29e7 MG |
2595 | /* |
2596 | * Revert to default priority/policy on fork if requested. | |
2597 | */ | |
2598 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2599 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2600 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2601 | p->normal_prio = p->static_prio; |
2602 | } | |
b9dc29e7 | 2603 | |
6c697bdf MG |
2604 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2605 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2606 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2607 | set_load_weight(p); |
2608 | } | |
2609 | ||
b9dc29e7 MG |
2610 | /* |
2611 | * We don't need the reset flag anymore after the fork. It has | |
2612 | * fulfilled its duty: | |
2613 | */ | |
2614 | p->sched_reset_on_fork = 0; | |
2615 | } | |
ca94c442 | 2616 | |
f83f9ac2 PW |
2617 | /* |
2618 | * Make sure we do not leak PI boosting priority to the child. | |
2619 | */ | |
2620 | p->prio = current->normal_prio; | |
2621 | ||
2ddbf952 HS |
2622 | if (!rt_prio(p->prio)) |
2623 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2624 | |
cd29fe6f PZ |
2625 | if (p->sched_class->task_fork) |
2626 | p->sched_class->task_fork(p); | |
2627 | ||
86951599 PZ |
2628 | /* |
2629 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2630 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2631 | * is ran before sched_fork(). | |
2632 | * | |
2633 | * Silence PROVE_RCU. | |
2634 | */ | |
2635 | rcu_read_lock(); | |
5f3edc1b | 2636 | set_task_cpu(p, cpu); |
86951599 | 2637 | rcu_read_unlock(); |
5f3edc1b | 2638 | |
52f17b6c | 2639 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2640 | if (likely(sched_info_on())) |
52f17b6c | 2641 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2642 | #endif |
d6077cb8 | 2643 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2644 | p->oncpu = 0; |
2645 | #endif | |
1da177e4 | 2646 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2647 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2648 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2649 | #endif |
806c09a7 | 2650 | #ifdef CONFIG_SMP |
917b627d | 2651 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2652 | #endif |
917b627d | 2653 | |
476d139c | 2654 | put_cpu(); |
1da177e4 LT |
2655 | } |
2656 | ||
2657 | /* | |
2658 | * wake_up_new_task - wake up a newly created task for the first time. | |
2659 | * | |
2660 | * This function will do some initial scheduler statistics housekeeping | |
2661 | * that must be done for every newly created context, then puts the task | |
2662 | * on the runqueue and wakes it. | |
2663 | */ | |
7ad5b3a5 | 2664 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2665 | { |
2666 | unsigned long flags; | |
dd41f596 | 2667 | struct rq *rq; |
c890692b | 2668 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2669 | |
2670 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2671 | rq = task_rq_lock(p, &flags); |
2672 | p->state = TASK_WAKING; | |
2673 | ||
fabf318e PZ |
2674 | /* |
2675 | * Fork balancing, do it here and not earlier because: | |
2676 | * - cpus_allowed can change in the fork path | |
2677 | * - any previously selected cpu might disappear through hotplug | |
2678 | * | |
0017d735 PZ |
2679 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2680 | * without people poking at ->cpus_allowed. | |
fabf318e | 2681 | */ |
0017d735 | 2682 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2683 | set_task_cpu(p, cpu); |
1da177e4 | 2684 | |
06b83b5f | 2685 | p->state = TASK_RUNNING; |
0017d735 PZ |
2686 | task_rq_unlock(rq, &flags); |
2687 | #endif | |
2688 | ||
2689 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2690 | activate_task(rq, p, 0); |
27a9da65 | 2691 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2692 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2693 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2694 | if (p->sched_class->task_woken) |
2695 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2696 | #endif |
dd41f596 | 2697 | task_rq_unlock(rq, &flags); |
fabf318e | 2698 | put_cpu(); |
1da177e4 LT |
2699 | } |
2700 | ||
e107be36 AK |
2701 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2702 | ||
2703 | /** | |
80dd99b3 | 2704 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2705 | * @notifier: notifier struct to register |
e107be36 AK |
2706 | */ |
2707 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2708 | { | |
2709 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2710 | } | |
2711 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2712 | ||
2713 | /** | |
2714 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2715 | * @notifier: notifier struct to unregister |
e107be36 AK |
2716 | * |
2717 | * This is safe to call from within a preemption notifier. | |
2718 | */ | |
2719 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2720 | { | |
2721 | hlist_del(¬ifier->link); | |
2722 | } | |
2723 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2724 | ||
2725 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2726 | { | |
2727 | struct preempt_notifier *notifier; | |
2728 | struct hlist_node *node; | |
2729 | ||
2730 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2731 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2732 | } | |
2733 | ||
2734 | static void | |
2735 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2736 | struct task_struct *next) | |
2737 | { | |
2738 | struct preempt_notifier *notifier; | |
2739 | struct hlist_node *node; | |
2740 | ||
2741 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2742 | notifier->ops->sched_out(notifier, next); | |
2743 | } | |
2744 | ||
6d6bc0ad | 2745 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2746 | |
2747 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2748 | { | |
2749 | } | |
2750 | ||
2751 | static void | |
2752 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2753 | struct task_struct *next) | |
2754 | { | |
2755 | } | |
2756 | ||
6d6bc0ad | 2757 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2758 | |
4866cde0 NP |
2759 | /** |
2760 | * prepare_task_switch - prepare to switch tasks | |
2761 | * @rq: the runqueue preparing to switch | |
421cee29 | 2762 | * @prev: the current task that is being switched out |
4866cde0 NP |
2763 | * @next: the task we are going to switch to. |
2764 | * | |
2765 | * This is called with the rq lock held and interrupts off. It must | |
2766 | * be paired with a subsequent finish_task_switch after the context | |
2767 | * switch. | |
2768 | * | |
2769 | * prepare_task_switch sets up locking and calls architecture specific | |
2770 | * hooks. | |
2771 | */ | |
e107be36 AK |
2772 | static inline void |
2773 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2774 | struct task_struct *next) | |
4866cde0 | 2775 | { |
e107be36 | 2776 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2777 | prepare_lock_switch(rq, next); |
2778 | prepare_arch_switch(next); | |
2779 | } | |
2780 | ||
1da177e4 LT |
2781 | /** |
2782 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2783 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2784 | * @prev: the thread we just switched away from. |
2785 | * | |
4866cde0 NP |
2786 | * finish_task_switch must be called after the context switch, paired |
2787 | * with a prepare_task_switch call before the context switch. | |
2788 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2789 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2790 | * |
2791 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2792 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2793 | * with the lock held can cause deadlocks; see schedule() for |
2794 | * details.) | |
2795 | */ | |
a9957449 | 2796 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2797 | __releases(rq->lock) |
2798 | { | |
1da177e4 | 2799 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2800 | long prev_state; |
1da177e4 LT |
2801 | |
2802 | rq->prev_mm = NULL; | |
2803 | ||
2804 | /* | |
2805 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2806 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2807 | * schedule one last time. The schedule call will never return, and |
2808 | * the scheduled task must drop that reference. | |
c394cc9f | 2809 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2810 | * still held, otherwise prev could be scheduled on another cpu, die |
2811 | * there before we look at prev->state, and then the reference would | |
2812 | * be dropped twice. | |
2813 | * Manfred Spraul <manfred@colorfullife.com> | |
2814 | */ | |
55a101f8 | 2815 | prev_state = prev->state; |
4866cde0 | 2816 | finish_arch_switch(prev); |
8381f65d JI |
2817 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2818 | local_irq_disable(); | |
2819 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2820 | perf_event_task_sched_in(current); |
8381f65d JI |
2821 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2822 | local_irq_enable(); | |
2823 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2824 | finish_lock_switch(rq, prev); |
e8fa1362 | 2825 | |
e107be36 | 2826 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2827 | if (mm) |
2828 | mmdrop(mm); | |
c394cc9f | 2829 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2830 | /* |
2831 | * Remove function-return probe instances associated with this | |
2832 | * task and put them back on the free list. | |
9761eea8 | 2833 | */ |
c6fd91f0 | 2834 | kprobe_flush_task(prev); |
1da177e4 | 2835 | put_task_struct(prev); |
c6fd91f0 | 2836 | } |
1da177e4 LT |
2837 | } |
2838 | ||
3f029d3c GH |
2839 | #ifdef CONFIG_SMP |
2840 | ||
2841 | /* assumes rq->lock is held */ | |
2842 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2843 | { | |
2844 | if (prev->sched_class->pre_schedule) | |
2845 | prev->sched_class->pre_schedule(rq, prev); | |
2846 | } | |
2847 | ||
2848 | /* rq->lock is NOT held, but preemption is disabled */ | |
2849 | static inline void post_schedule(struct rq *rq) | |
2850 | { | |
2851 | if (rq->post_schedule) { | |
2852 | unsigned long flags; | |
2853 | ||
05fa785c | 2854 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2855 | if (rq->curr->sched_class->post_schedule) |
2856 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2857 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2858 | |
2859 | rq->post_schedule = 0; | |
2860 | } | |
2861 | } | |
2862 | ||
2863 | #else | |
da19ab51 | 2864 | |
3f029d3c GH |
2865 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2866 | { | |
2867 | } | |
2868 | ||
2869 | static inline void post_schedule(struct rq *rq) | |
2870 | { | |
1da177e4 LT |
2871 | } |
2872 | ||
3f029d3c GH |
2873 | #endif |
2874 | ||
1da177e4 LT |
2875 | /** |
2876 | * schedule_tail - first thing a freshly forked thread must call. | |
2877 | * @prev: the thread we just switched away from. | |
2878 | */ | |
36c8b586 | 2879 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2880 | __releases(rq->lock) |
2881 | { | |
70b97a7f IM |
2882 | struct rq *rq = this_rq(); |
2883 | ||
4866cde0 | 2884 | finish_task_switch(rq, prev); |
da19ab51 | 2885 | |
3f029d3c GH |
2886 | /* |
2887 | * FIXME: do we need to worry about rq being invalidated by the | |
2888 | * task_switch? | |
2889 | */ | |
2890 | post_schedule(rq); | |
70b97a7f | 2891 | |
4866cde0 NP |
2892 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2893 | /* In this case, finish_task_switch does not reenable preemption */ | |
2894 | preempt_enable(); | |
2895 | #endif | |
1da177e4 | 2896 | if (current->set_child_tid) |
b488893a | 2897 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2898 | } |
2899 | ||
2900 | /* | |
2901 | * context_switch - switch to the new MM and the new | |
2902 | * thread's register state. | |
2903 | */ | |
dd41f596 | 2904 | static inline void |
70b97a7f | 2905 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2906 | struct task_struct *next) |
1da177e4 | 2907 | { |
dd41f596 | 2908 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2909 | |
e107be36 | 2910 | prepare_task_switch(rq, prev, next); |
27a9da65 | 2911 | trace_sched_switch(prev, next); |
dd41f596 IM |
2912 | mm = next->mm; |
2913 | oldmm = prev->active_mm; | |
9226d125 ZA |
2914 | /* |
2915 | * For paravirt, this is coupled with an exit in switch_to to | |
2916 | * combine the page table reload and the switch backend into | |
2917 | * one hypercall. | |
2918 | */ | |
224101ed | 2919 | arch_start_context_switch(prev); |
9226d125 | 2920 | |
31915ab4 | 2921 | if (!mm) { |
1da177e4 LT |
2922 | next->active_mm = oldmm; |
2923 | atomic_inc(&oldmm->mm_count); | |
2924 | enter_lazy_tlb(oldmm, next); | |
2925 | } else | |
2926 | switch_mm(oldmm, mm, next); | |
2927 | ||
31915ab4 | 2928 | if (!prev->mm) { |
1da177e4 | 2929 | prev->active_mm = NULL; |
1da177e4 LT |
2930 | rq->prev_mm = oldmm; |
2931 | } | |
3a5f5e48 IM |
2932 | /* |
2933 | * Since the runqueue lock will be released by the next | |
2934 | * task (which is an invalid locking op but in the case | |
2935 | * of the scheduler it's an obvious special-case), so we | |
2936 | * do an early lockdep release here: | |
2937 | */ | |
2938 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2939 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2940 | #endif |
1da177e4 LT |
2941 | |
2942 | /* Here we just switch the register state and the stack. */ | |
2943 | switch_to(prev, next, prev); | |
2944 | ||
dd41f596 IM |
2945 | barrier(); |
2946 | /* | |
2947 | * this_rq must be evaluated again because prev may have moved | |
2948 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2949 | * frame will be invalid. | |
2950 | */ | |
2951 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2952 | } |
2953 | ||
2954 | /* | |
2955 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2956 | * | |
2957 | * externally visible scheduler statistics: current number of runnable | |
2958 | * threads, current number of uninterruptible-sleeping threads, total | |
2959 | * number of context switches performed since bootup. | |
2960 | */ | |
2961 | unsigned long nr_running(void) | |
2962 | { | |
2963 | unsigned long i, sum = 0; | |
2964 | ||
2965 | for_each_online_cpu(i) | |
2966 | sum += cpu_rq(i)->nr_running; | |
2967 | ||
2968 | return sum; | |
f711f609 | 2969 | } |
1da177e4 LT |
2970 | |
2971 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2972 | { |
1da177e4 | 2973 | unsigned long i, sum = 0; |
f711f609 | 2974 | |
0a945022 | 2975 | for_each_possible_cpu(i) |
1da177e4 | 2976 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2977 | |
2978 | /* | |
1da177e4 LT |
2979 | * Since we read the counters lockless, it might be slightly |
2980 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2981 | */ |
1da177e4 LT |
2982 | if (unlikely((long)sum < 0)) |
2983 | sum = 0; | |
f711f609 | 2984 | |
1da177e4 | 2985 | return sum; |
f711f609 | 2986 | } |
f711f609 | 2987 | |
1da177e4 | 2988 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2989 | { |
cc94abfc SR |
2990 | int i; |
2991 | unsigned long long sum = 0; | |
46cb4b7c | 2992 | |
0a945022 | 2993 | for_each_possible_cpu(i) |
1da177e4 | 2994 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2995 | |
1da177e4 LT |
2996 | return sum; |
2997 | } | |
483b4ee6 | 2998 | |
1da177e4 LT |
2999 | unsigned long nr_iowait(void) |
3000 | { | |
3001 | unsigned long i, sum = 0; | |
483b4ee6 | 3002 | |
0a945022 | 3003 | for_each_possible_cpu(i) |
1da177e4 | 3004 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 3005 | |
1da177e4 LT |
3006 | return sum; |
3007 | } | |
483b4ee6 | 3008 | |
8c215bd3 | 3009 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 3010 | { |
8c215bd3 | 3011 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
3012 | return atomic_read(&this->nr_iowait); |
3013 | } | |
46cb4b7c | 3014 | |
69d25870 AV |
3015 | unsigned long this_cpu_load(void) |
3016 | { | |
3017 | struct rq *this = this_rq(); | |
3018 | return this->cpu_load[0]; | |
3019 | } | |
e790fb0b | 3020 | |
46cb4b7c | 3021 | |
dce48a84 TG |
3022 | /* Variables and functions for calc_load */ |
3023 | static atomic_long_t calc_load_tasks; | |
3024 | static unsigned long calc_load_update; | |
3025 | unsigned long avenrun[3]; | |
3026 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3027 | |
74f5187a PZ |
3028 | static long calc_load_fold_active(struct rq *this_rq) |
3029 | { | |
3030 | long nr_active, delta = 0; | |
3031 | ||
3032 | nr_active = this_rq->nr_running; | |
3033 | nr_active += (long) this_rq->nr_uninterruptible; | |
3034 | ||
3035 | if (nr_active != this_rq->calc_load_active) { | |
3036 | delta = nr_active - this_rq->calc_load_active; | |
3037 | this_rq->calc_load_active = nr_active; | |
3038 | } | |
3039 | ||
3040 | return delta; | |
3041 | } | |
3042 | ||
0f004f5a PZ |
3043 | static unsigned long |
3044 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3045 | { | |
3046 | load *= exp; | |
3047 | load += active * (FIXED_1 - exp); | |
3048 | load += 1UL << (FSHIFT - 1); | |
3049 | return load >> FSHIFT; | |
3050 | } | |
3051 | ||
74f5187a PZ |
3052 | #ifdef CONFIG_NO_HZ |
3053 | /* | |
3054 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3055 | * | |
3056 | * When making the ILB scale, we should try to pull this in as well. | |
3057 | */ | |
3058 | static atomic_long_t calc_load_tasks_idle; | |
3059 | ||
3060 | static void calc_load_account_idle(struct rq *this_rq) | |
3061 | { | |
3062 | long delta; | |
3063 | ||
3064 | delta = calc_load_fold_active(this_rq); | |
3065 | if (delta) | |
3066 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3067 | } | |
3068 | ||
3069 | static long calc_load_fold_idle(void) | |
3070 | { | |
3071 | long delta = 0; | |
3072 | ||
3073 | /* | |
3074 | * Its got a race, we don't care... | |
3075 | */ | |
3076 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3077 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3078 | ||
3079 | return delta; | |
3080 | } | |
0f004f5a PZ |
3081 | |
3082 | /** | |
3083 | * fixed_power_int - compute: x^n, in O(log n) time | |
3084 | * | |
3085 | * @x: base of the power | |
3086 | * @frac_bits: fractional bits of @x | |
3087 | * @n: power to raise @x to. | |
3088 | * | |
3089 | * By exploiting the relation between the definition of the natural power | |
3090 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3091 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3092 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3093 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3094 | * of course trivially computable in O(log_2 n), the length of our binary | |
3095 | * vector. | |
3096 | */ | |
3097 | static unsigned long | |
3098 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3099 | { | |
3100 | unsigned long result = 1UL << frac_bits; | |
3101 | ||
3102 | if (n) for (;;) { | |
3103 | if (n & 1) { | |
3104 | result *= x; | |
3105 | result += 1UL << (frac_bits - 1); | |
3106 | result >>= frac_bits; | |
3107 | } | |
3108 | n >>= 1; | |
3109 | if (!n) | |
3110 | break; | |
3111 | x *= x; | |
3112 | x += 1UL << (frac_bits - 1); | |
3113 | x >>= frac_bits; | |
3114 | } | |
3115 | ||
3116 | return result; | |
3117 | } | |
3118 | ||
3119 | /* | |
3120 | * a1 = a0 * e + a * (1 - e) | |
3121 | * | |
3122 | * a2 = a1 * e + a * (1 - e) | |
3123 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3124 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3125 | * | |
3126 | * a3 = a2 * e + a * (1 - e) | |
3127 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3128 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3129 | * | |
3130 | * ... | |
3131 | * | |
3132 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3133 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3134 | * = a0 * e^n + a * (1 - e^n) | |
3135 | * | |
3136 | * [1] application of the geometric series: | |
3137 | * | |
3138 | * n 1 - x^(n+1) | |
3139 | * S_n := \Sum x^i = ------------- | |
3140 | * i=0 1 - x | |
3141 | */ | |
3142 | static unsigned long | |
3143 | calc_load_n(unsigned long load, unsigned long exp, | |
3144 | unsigned long active, unsigned int n) | |
3145 | { | |
3146 | ||
3147 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3148 | } | |
3149 | ||
3150 | /* | |
3151 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3152 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3153 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3154 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3155 | * | |
3156 | * Once we've updated the global active value, we need to apply the exponential | |
3157 | * weights adjusted to the number of cycles missed. | |
3158 | */ | |
3159 | static void calc_global_nohz(unsigned long ticks) | |
3160 | { | |
3161 | long delta, active, n; | |
3162 | ||
3163 | if (time_before(jiffies, calc_load_update)) | |
3164 | return; | |
3165 | ||
3166 | /* | |
3167 | * If we crossed a calc_load_update boundary, make sure to fold | |
3168 | * any pending idle changes, the respective CPUs might have | |
3169 | * missed the tick driven calc_load_account_active() update | |
3170 | * due to NO_HZ. | |
3171 | */ | |
3172 | delta = calc_load_fold_idle(); | |
3173 | if (delta) | |
3174 | atomic_long_add(delta, &calc_load_tasks); | |
3175 | ||
3176 | /* | |
3177 | * If we were idle for multiple load cycles, apply them. | |
3178 | */ | |
3179 | if (ticks >= LOAD_FREQ) { | |
3180 | n = ticks / LOAD_FREQ; | |
3181 | ||
3182 | active = atomic_long_read(&calc_load_tasks); | |
3183 | active = active > 0 ? active * FIXED_1 : 0; | |
3184 | ||
3185 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3186 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3187 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3188 | ||
3189 | calc_load_update += n * LOAD_FREQ; | |
3190 | } | |
3191 | ||
3192 | /* | |
3193 | * Its possible the remainder of the above division also crosses | |
3194 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3195 | * which comes after this will take care of that. | |
3196 | * | |
3197 | * Consider us being 11 ticks before a cycle completion, and us | |
3198 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3199 | * age us 4 cycles, and the test in calc_global_load() will | |
3200 | * pick up the final one. | |
3201 | */ | |
3202 | } | |
74f5187a PZ |
3203 | #else |
3204 | static void calc_load_account_idle(struct rq *this_rq) | |
3205 | { | |
3206 | } | |
3207 | ||
3208 | static inline long calc_load_fold_idle(void) | |
3209 | { | |
3210 | return 0; | |
3211 | } | |
0f004f5a PZ |
3212 | |
3213 | static void calc_global_nohz(unsigned long ticks) | |
3214 | { | |
3215 | } | |
74f5187a PZ |
3216 | #endif |
3217 | ||
2d02494f TG |
3218 | /** |
3219 | * get_avenrun - get the load average array | |
3220 | * @loads: pointer to dest load array | |
3221 | * @offset: offset to add | |
3222 | * @shift: shift count to shift the result left | |
3223 | * | |
3224 | * These values are estimates at best, so no need for locking. | |
3225 | */ | |
3226 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3227 | { | |
3228 | loads[0] = (avenrun[0] + offset) << shift; | |
3229 | loads[1] = (avenrun[1] + offset) << shift; | |
3230 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3231 | } |
46cb4b7c | 3232 | |
46cb4b7c | 3233 | /* |
dce48a84 TG |
3234 | * calc_load - update the avenrun load estimates 10 ticks after the |
3235 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3236 | */ |
0f004f5a | 3237 | void calc_global_load(unsigned long ticks) |
7835b98b | 3238 | { |
dce48a84 | 3239 | long active; |
1da177e4 | 3240 | |
0f004f5a PZ |
3241 | calc_global_nohz(ticks); |
3242 | ||
3243 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3244 | return; |
1da177e4 | 3245 | |
dce48a84 TG |
3246 | active = atomic_long_read(&calc_load_tasks); |
3247 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3248 | |
dce48a84 TG |
3249 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3250 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3251 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3252 | |
dce48a84 TG |
3253 | calc_load_update += LOAD_FREQ; |
3254 | } | |
1da177e4 | 3255 | |
dce48a84 | 3256 | /* |
74f5187a PZ |
3257 | * Called from update_cpu_load() to periodically update this CPU's |
3258 | * active count. | |
dce48a84 TG |
3259 | */ |
3260 | static void calc_load_account_active(struct rq *this_rq) | |
3261 | { | |
74f5187a | 3262 | long delta; |
08c183f3 | 3263 | |
74f5187a PZ |
3264 | if (time_before(jiffies, this_rq->calc_load_update)) |
3265 | return; | |
783609c6 | 3266 | |
74f5187a PZ |
3267 | delta = calc_load_fold_active(this_rq); |
3268 | delta += calc_load_fold_idle(); | |
3269 | if (delta) | |
dce48a84 | 3270 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3271 | |
3272 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3273 | } |
3274 | ||
fdf3e95d VP |
3275 | /* |
3276 | * The exact cpuload at various idx values, calculated at every tick would be | |
3277 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3278 | * | |
3279 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3280 | * on nth tick when cpu may be busy, then we have: | |
3281 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3282 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3283 | * | |
3284 | * decay_load_missed() below does efficient calculation of | |
3285 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3286 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3287 | * | |
3288 | * The calculation is approximated on a 128 point scale. | |
3289 | * degrade_zero_ticks is the number of ticks after which load at any | |
3290 | * particular idx is approximated to be zero. | |
3291 | * degrade_factor is a precomputed table, a row for each load idx. | |
3292 | * Each column corresponds to degradation factor for a power of two ticks, | |
3293 | * based on 128 point scale. | |
3294 | * Example: | |
3295 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3296 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3297 | * | |
3298 | * With this power of 2 load factors, we can degrade the load n times | |
3299 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3300 | * n mult/shifts needed by the exact degradation. | |
3301 | */ | |
3302 | #define DEGRADE_SHIFT 7 | |
3303 | static const unsigned char | |
3304 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3305 | static const unsigned char | |
3306 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3307 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3308 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3309 | {96, 72, 40, 12, 1, 0, 0}, | |
3310 | {112, 98, 75, 43, 15, 1, 0}, | |
3311 | {120, 112, 98, 76, 45, 16, 2} }; | |
3312 | ||
3313 | /* | |
3314 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3315 | * would be when CPU is idle and so we just decay the old load without | |
3316 | * adding any new load. | |
3317 | */ | |
3318 | static unsigned long | |
3319 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3320 | { | |
3321 | int j = 0; | |
3322 | ||
3323 | if (!missed_updates) | |
3324 | return load; | |
3325 | ||
3326 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3327 | return 0; | |
3328 | ||
3329 | if (idx == 1) | |
3330 | return load >> missed_updates; | |
3331 | ||
3332 | while (missed_updates) { | |
3333 | if (missed_updates % 2) | |
3334 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3335 | ||
3336 | missed_updates >>= 1; | |
3337 | j++; | |
3338 | } | |
3339 | return load; | |
3340 | } | |
3341 | ||
46cb4b7c | 3342 | /* |
dd41f596 | 3343 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3344 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3345 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3346 | */ |
dd41f596 | 3347 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3348 | { |
495eca49 | 3349 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3350 | unsigned long curr_jiffies = jiffies; |
3351 | unsigned long pending_updates; | |
dd41f596 | 3352 | int i, scale; |
46cb4b7c | 3353 | |
dd41f596 | 3354 | this_rq->nr_load_updates++; |
46cb4b7c | 3355 | |
fdf3e95d VP |
3356 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3357 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3358 | return; | |
3359 | ||
3360 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3361 | this_rq->last_load_update_tick = curr_jiffies; | |
3362 | ||
dd41f596 | 3363 | /* Update our load: */ |
fdf3e95d VP |
3364 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3365 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3366 | unsigned long old_load, new_load; |
7d1e6a9b | 3367 | |
dd41f596 | 3368 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3369 | |
dd41f596 | 3370 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3371 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3372 | new_load = this_load; |
a25707f3 IM |
3373 | /* |
3374 | * Round up the averaging division if load is increasing. This | |
3375 | * prevents us from getting stuck on 9 if the load is 10, for | |
3376 | * example. | |
3377 | */ | |
3378 | if (new_load > old_load) | |
fdf3e95d VP |
3379 | new_load += scale - 1; |
3380 | ||
3381 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3382 | } |
da2b71ed SS |
3383 | |
3384 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3385 | } |
3386 | ||
3387 | static void update_cpu_load_active(struct rq *this_rq) | |
3388 | { | |
3389 | update_cpu_load(this_rq); | |
46cb4b7c | 3390 | |
74f5187a | 3391 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3392 | } |
3393 | ||
dd41f596 | 3394 | #ifdef CONFIG_SMP |
8a0be9ef | 3395 | |
46cb4b7c | 3396 | /* |
38022906 PZ |
3397 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3398 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3399 | */ |
38022906 | 3400 | void sched_exec(void) |
46cb4b7c | 3401 | { |
38022906 | 3402 | struct task_struct *p = current; |
1da177e4 | 3403 | unsigned long flags; |
70b97a7f | 3404 | struct rq *rq; |
0017d735 | 3405 | int dest_cpu; |
46cb4b7c | 3406 | |
1da177e4 | 3407 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3408 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3409 | if (dest_cpu == smp_processor_id()) | |
3410 | goto unlock; | |
38022906 | 3411 | |
46cb4b7c | 3412 | /* |
38022906 | 3413 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3414 | */ |
30da688e | 3415 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
b7a2b39d | 3416 | likely(cpu_active(dest_cpu)) && migrate_task(p, rq)) { |
969c7921 | 3417 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3418 | |
1da177e4 | 3419 | task_rq_unlock(rq, &flags); |
969c7921 | 3420 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3421 | return; |
3422 | } | |
0017d735 | 3423 | unlock: |
1da177e4 | 3424 | task_rq_unlock(rq, &flags); |
1da177e4 | 3425 | } |
dd41f596 | 3426 | |
1da177e4 LT |
3427 | #endif |
3428 | ||
1da177e4 LT |
3429 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3430 | ||
3431 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3432 | ||
3433 | /* | |
c5f8d995 | 3434 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3435 | * @p in case that task is currently running. |
c5f8d995 HS |
3436 | * |
3437 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3438 | */ |
c5f8d995 HS |
3439 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3440 | { | |
3441 | u64 ns = 0; | |
3442 | ||
3443 | if (task_current(rq, p)) { | |
3444 | update_rq_clock(rq); | |
305e6835 | 3445 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3446 | if ((s64)ns < 0) |
3447 | ns = 0; | |
3448 | } | |
3449 | ||
3450 | return ns; | |
3451 | } | |
3452 | ||
bb34d92f | 3453 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3454 | { |
1da177e4 | 3455 | unsigned long flags; |
41b86e9c | 3456 | struct rq *rq; |
bb34d92f | 3457 | u64 ns = 0; |
48f24c4d | 3458 | |
41b86e9c | 3459 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3460 | ns = do_task_delta_exec(p, rq); |
3461 | task_rq_unlock(rq, &flags); | |
1508487e | 3462 | |
c5f8d995 HS |
3463 | return ns; |
3464 | } | |
f06febc9 | 3465 | |
c5f8d995 HS |
3466 | /* |
3467 | * Return accounted runtime for the task. | |
3468 | * In case the task is currently running, return the runtime plus current's | |
3469 | * pending runtime that have not been accounted yet. | |
3470 | */ | |
3471 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3472 | { | |
3473 | unsigned long flags; | |
3474 | struct rq *rq; | |
3475 | u64 ns = 0; | |
3476 | ||
3477 | rq = task_rq_lock(p, &flags); | |
3478 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3479 | task_rq_unlock(rq, &flags); | |
3480 | ||
3481 | return ns; | |
3482 | } | |
48f24c4d | 3483 | |
c5f8d995 HS |
3484 | /* |
3485 | * Return sum_exec_runtime for the thread group. | |
3486 | * In case the task is currently running, return the sum plus current's | |
3487 | * pending runtime that have not been accounted yet. | |
3488 | * | |
3489 | * Note that the thread group might have other running tasks as well, | |
3490 | * so the return value not includes other pending runtime that other | |
3491 | * running tasks might have. | |
3492 | */ | |
3493 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3494 | { | |
3495 | struct task_cputime totals; | |
3496 | unsigned long flags; | |
3497 | struct rq *rq; | |
3498 | u64 ns; | |
3499 | ||
3500 | rq = task_rq_lock(p, &flags); | |
3501 | thread_group_cputime(p, &totals); | |
3502 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3503 | task_rq_unlock(rq, &flags); |
48f24c4d | 3504 | |
1da177e4 LT |
3505 | return ns; |
3506 | } | |
3507 | ||
1da177e4 LT |
3508 | /* |
3509 | * Account user cpu time to a process. | |
3510 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3511 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3512 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3513 | */ |
457533a7 MS |
3514 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3515 | cputime_t cputime_scaled) | |
1da177e4 LT |
3516 | { |
3517 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3518 | cputime64_t tmp; | |
3519 | ||
457533a7 | 3520 | /* Add user time to process. */ |
1da177e4 | 3521 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3522 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3523 | account_group_user_time(p, cputime); |
1da177e4 LT |
3524 | |
3525 | /* Add user time to cpustat. */ | |
3526 | tmp = cputime_to_cputime64(cputime); | |
3527 | if (TASK_NICE(p) > 0) | |
3528 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3529 | else | |
3530 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3531 | |
3532 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3533 | /* Account for user time used */ |
3534 | acct_update_integrals(p); | |
1da177e4 LT |
3535 | } |
3536 | ||
94886b84 LV |
3537 | /* |
3538 | * Account guest cpu time to a process. | |
3539 | * @p: the process that the cpu time gets accounted to | |
3540 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3541 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3542 | */ |
457533a7 MS |
3543 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3544 | cputime_t cputime_scaled) | |
94886b84 LV |
3545 | { |
3546 | cputime64_t tmp; | |
3547 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3548 | ||
3549 | tmp = cputime_to_cputime64(cputime); | |
3550 | ||
457533a7 | 3551 | /* Add guest time to process. */ |
94886b84 | 3552 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3553 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3554 | account_group_user_time(p, cputime); |
94886b84 LV |
3555 | p->gtime = cputime_add(p->gtime, cputime); |
3556 | ||
457533a7 | 3557 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3558 | if (TASK_NICE(p) > 0) { |
3559 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3560 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3561 | } else { | |
3562 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3563 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3564 | } | |
94886b84 LV |
3565 | } |
3566 | ||
1da177e4 LT |
3567 | /* |
3568 | * Account system cpu time to a process. | |
3569 | * @p: the process that the cpu time gets accounted to | |
3570 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3571 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3572 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3573 | */ |
3574 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3575 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3576 | { |
3577 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3578 | cputime64_t tmp; |
3579 | ||
983ed7a6 | 3580 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3581 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3582 | return; |
3583 | } | |
94886b84 | 3584 | |
457533a7 | 3585 | /* Add system time to process. */ |
1da177e4 | 3586 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3587 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3588 | account_group_system_time(p, cputime); |
1da177e4 LT |
3589 | |
3590 | /* Add system time to cpustat. */ | |
3591 | tmp = cputime_to_cputime64(cputime); | |
3592 | if (hardirq_count() - hardirq_offset) | |
3593 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
75e1056f | 3594 | else if (in_serving_softirq()) |
1da177e4 | 3595 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); |
1da177e4 | 3596 | else |
79741dd3 MS |
3597 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3598 | ||
ef12fefa BR |
3599 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3600 | ||
1da177e4 LT |
3601 | /* Account for system time used */ |
3602 | acct_update_integrals(p); | |
1da177e4 LT |
3603 | } |
3604 | ||
c66f08be | 3605 | /* |
1da177e4 | 3606 | * Account for involuntary wait time. |
1da177e4 | 3607 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3608 | */ |
79741dd3 | 3609 | void account_steal_time(cputime_t cputime) |
c66f08be | 3610 | { |
79741dd3 MS |
3611 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3612 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3613 | ||
3614 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3615 | } |
3616 | ||
1da177e4 | 3617 | /* |
79741dd3 MS |
3618 | * Account for idle time. |
3619 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3620 | */ |
79741dd3 | 3621 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3622 | { |
3623 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3624 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3625 | struct rq *rq = this_rq(); |
1da177e4 | 3626 | |
79741dd3 MS |
3627 | if (atomic_read(&rq->nr_iowait) > 0) |
3628 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3629 | else | |
3630 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3631 | } |
3632 | ||
79741dd3 MS |
3633 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3634 | ||
3635 | /* | |
3636 | * Account a single tick of cpu time. | |
3637 | * @p: the process that the cpu time gets accounted to | |
3638 | * @user_tick: indicates if the tick is a user or a system tick | |
3639 | */ | |
3640 | void account_process_tick(struct task_struct *p, int user_tick) | |
3641 | { | |
a42548a1 | 3642 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3643 | struct rq *rq = this_rq(); |
3644 | ||
3645 | if (user_tick) | |
a42548a1 | 3646 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3647 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3648 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3649 | one_jiffy_scaled); |
3650 | else | |
a42548a1 | 3651 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3652 | } |
3653 | ||
3654 | /* | |
3655 | * Account multiple ticks of steal time. | |
3656 | * @p: the process from which the cpu time has been stolen | |
3657 | * @ticks: number of stolen ticks | |
3658 | */ | |
3659 | void account_steal_ticks(unsigned long ticks) | |
3660 | { | |
3661 | account_steal_time(jiffies_to_cputime(ticks)); | |
3662 | } | |
3663 | ||
3664 | /* | |
3665 | * Account multiple ticks of idle time. | |
3666 | * @ticks: number of stolen ticks | |
3667 | */ | |
3668 | void account_idle_ticks(unsigned long ticks) | |
3669 | { | |
3670 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3671 | } |
3672 | ||
79741dd3 MS |
3673 | #endif |
3674 | ||
49048622 BS |
3675 | /* |
3676 | * Use precise platform statistics if available: | |
3677 | */ | |
3678 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3679 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3680 | { |
d99ca3b9 HS |
3681 | *ut = p->utime; |
3682 | *st = p->stime; | |
49048622 BS |
3683 | } |
3684 | ||
0cf55e1e | 3685 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3686 | { |
0cf55e1e HS |
3687 | struct task_cputime cputime; |
3688 | ||
3689 | thread_group_cputime(p, &cputime); | |
3690 | ||
3691 | *ut = cputime.utime; | |
3692 | *st = cputime.stime; | |
49048622 BS |
3693 | } |
3694 | #else | |
761b1d26 HS |
3695 | |
3696 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3697 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3698 | #endif |
3699 | ||
d180c5bc | 3700 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3701 | { |
d99ca3b9 | 3702 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3703 | |
3704 | /* | |
3705 | * Use CFS's precise accounting: | |
3706 | */ | |
d180c5bc | 3707 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3708 | |
3709 | if (total) { | |
e75e863d | 3710 | u64 temp = rtime; |
d180c5bc | 3711 | |
e75e863d | 3712 | temp *= utime; |
49048622 | 3713 | do_div(temp, total); |
d180c5bc HS |
3714 | utime = (cputime_t)temp; |
3715 | } else | |
3716 | utime = rtime; | |
49048622 | 3717 | |
d180c5bc HS |
3718 | /* |
3719 | * Compare with previous values, to keep monotonicity: | |
3720 | */ | |
761b1d26 | 3721 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3722 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3723 | |
d99ca3b9 HS |
3724 | *ut = p->prev_utime; |
3725 | *st = p->prev_stime; | |
49048622 BS |
3726 | } |
3727 | ||
0cf55e1e HS |
3728 | /* |
3729 | * Must be called with siglock held. | |
3730 | */ | |
3731 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3732 | { |
0cf55e1e HS |
3733 | struct signal_struct *sig = p->signal; |
3734 | struct task_cputime cputime; | |
3735 | cputime_t rtime, utime, total; | |
49048622 | 3736 | |
0cf55e1e | 3737 | thread_group_cputime(p, &cputime); |
49048622 | 3738 | |
0cf55e1e HS |
3739 | total = cputime_add(cputime.utime, cputime.stime); |
3740 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3741 | |
0cf55e1e | 3742 | if (total) { |
e75e863d | 3743 | u64 temp = rtime; |
49048622 | 3744 | |
e75e863d | 3745 | temp *= cputime.utime; |
0cf55e1e HS |
3746 | do_div(temp, total); |
3747 | utime = (cputime_t)temp; | |
3748 | } else | |
3749 | utime = rtime; | |
3750 | ||
3751 | sig->prev_utime = max(sig->prev_utime, utime); | |
3752 | sig->prev_stime = max(sig->prev_stime, | |
3753 | cputime_sub(rtime, sig->prev_utime)); | |
3754 | ||
3755 | *ut = sig->prev_utime; | |
3756 | *st = sig->prev_stime; | |
49048622 | 3757 | } |
49048622 | 3758 | #endif |
49048622 | 3759 | |
7835b98b CL |
3760 | /* |
3761 | * This function gets called by the timer code, with HZ frequency. | |
3762 | * We call it with interrupts disabled. | |
3763 | * | |
3764 | * It also gets called by the fork code, when changing the parent's | |
3765 | * timeslices. | |
3766 | */ | |
3767 | void scheduler_tick(void) | |
3768 | { | |
7835b98b CL |
3769 | int cpu = smp_processor_id(); |
3770 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3771 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3772 | |
3773 | sched_clock_tick(); | |
dd41f596 | 3774 | |
05fa785c | 3775 | raw_spin_lock(&rq->lock); |
3e51f33f | 3776 | update_rq_clock(rq); |
fdf3e95d | 3777 | update_cpu_load_active(rq); |
fa85ae24 | 3778 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3779 | raw_spin_unlock(&rq->lock); |
7835b98b | 3780 | |
e9d2b064 | 3781 | perf_event_task_tick(); |
e220d2dc | 3782 | |
e418e1c2 | 3783 | #ifdef CONFIG_SMP |
dd41f596 IM |
3784 | rq->idle_at_tick = idle_cpu(cpu); |
3785 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3786 | #endif |
1da177e4 LT |
3787 | } |
3788 | ||
132380a0 | 3789 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3790 | { |
3791 | if (in_lock_functions(addr)) { | |
3792 | addr = CALLER_ADDR2; | |
3793 | if (in_lock_functions(addr)) | |
3794 | addr = CALLER_ADDR3; | |
3795 | } | |
3796 | return addr; | |
3797 | } | |
1da177e4 | 3798 | |
7e49fcce SR |
3799 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3800 | defined(CONFIG_PREEMPT_TRACER)) | |
3801 | ||
43627582 | 3802 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3803 | { |
6cd8a4bb | 3804 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3805 | /* |
3806 | * Underflow? | |
3807 | */ | |
9a11b49a IM |
3808 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3809 | return; | |
6cd8a4bb | 3810 | #endif |
1da177e4 | 3811 | preempt_count() += val; |
6cd8a4bb | 3812 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3813 | /* |
3814 | * Spinlock count overflowing soon? | |
3815 | */ | |
33859f7f MOS |
3816 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3817 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3818 | #endif |
3819 | if (preempt_count() == val) | |
3820 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3821 | } |
3822 | EXPORT_SYMBOL(add_preempt_count); | |
3823 | ||
43627582 | 3824 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3825 | { |
6cd8a4bb | 3826 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3827 | /* |
3828 | * Underflow? | |
3829 | */ | |
01e3eb82 | 3830 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3831 | return; |
1da177e4 LT |
3832 | /* |
3833 | * Is the spinlock portion underflowing? | |
3834 | */ | |
9a11b49a IM |
3835 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3836 | !(preempt_count() & PREEMPT_MASK))) | |
3837 | return; | |
6cd8a4bb | 3838 | #endif |
9a11b49a | 3839 | |
6cd8a4bb SR |
3840 | if (preempt_count() == val) |
3841 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3842 | preempt_count() -= val; |
3843 | } | |
3844 | EXPORT_SYMBOL(sub_preempt_count); | |
3845 | ||
3846 | #endif | |
3847 | ||
3848 | /* | |
dd41f596 | 3849 | * Print scheduling while atomic bug: |
1da177e4 | 3850 | */ |
dd41f596 | 3851 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3852 | { |
838225b4 SS |
3853 | struct pt_regs *regs = get_irq_regs(); |
3854 | ||
3df0fc5b PZ |
3855 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3856 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3857 | |
dd41f596 | 3858 | debug_show_held_locks(prev); |
e21f5b15 | 3859 | print_modules(); |
dd41f596 IM |
3860 | if (irqs_disabled()) |
3861 | print_irqtrace_events(prev); | |
838225b4 SS |
3862 | |
3863 | if (regs) | |
3864 | show_regs(regs); | |
3865 | else | |
3866 | dump_stack(); | |
dd41f596 | 3867 | } |
1da177e4 | 3868 | |
dd41f596 IM |
3869 | /* |
3870 | * Various schedule()-time debugging checks and statistics: | |
3871 | */ | |
3872 | static inline void schedule_debug(struct task_struct *prev) | |
3873 | { | |
1da177e4 | 3874 | /* |
41a2d6cf | 3875 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3876 | * schedule() atomically, we ignore that path for now. |
3877 | * Otherwise, whine if we are scheduling when we should not be. | |
3878 | */ | |
3f33a7ce | 3879 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3880 | __schedule_bug(prev); |
3881 | ||
1da177e4 LT |
3882 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3883 | ||
2d72376b | 3884 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3885 | #ifdef CONFIG_SCHEDSTATS |
3886 | if (unlikely(prev->lock_depth >= 0)) { | |
fce20979 | 3887 | schedstat_inc(this_rq(), rq_sched_info.bkl_count); |
2d72376b | 3888 | schedstat_inc(prev, sched_info.bkl_count); |
b8efb561 IM |
3889 | } |
3890 | #endif | |
dd41f596 IM |
3891 | } |
3892 | ||
6cecd084 | 3893 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3894 | { |
a64692a3 MG |
3895 | if (prev->se.on_rq) |
3896 | update_rq_clock(rq); | |
6cecd084 | 3897 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3898 | } |
3899 | ||
dd41f596 IM |
3900 | /* |
3901 | * Pick up the highest-prio task: | |
3902 | */ | |
3903 | static inline struct task_struct * | |
b67802ea | 3904 | pick_next_task(struct rq *rq) |
dd41f596 | 3905 | { |
5522d5d5 | 3906 | const struct sched_class *class; |
dd41f596 | 3907 | struct task_struct *p; |
1da177e4 LT |
3908 | |
3909 | /* | |
dd41f596 IM |
3910 | * Optimization: we know that if all tasks are in |
3911 | * the fair class we can call that function directly: | |
1da177e4 | 3912 | */ |
dd41f596 | 3913 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3914 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3915 | if (likely(p)) |
3916 | return p; | |
1da177e4 LT |
3917 | } |
3918 | ||
34f971f6 | 3919 | for_each_class(class) { |
fb8d4724 | 3920 | p = class->pick_next_task(rq); |
dd41f596 IM |
3921 | if (p) |
3922 | return p; | |
dd41f596 | 3923 | } |
34f971f6 PZ |
3924 | |
3925 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3926 | } |
1da177e4 | 3927 | |
dd41f596 IM |
3928 | /* |
3929 | * schedule() is the main scheduler function. | |
3930 | */ | |
ff743345 | 3931 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3932 | { |
3933 | struct task_struct *prev, *next; | |
67ca7bde | 3934 | unsigned long *switch_count; |
dd41f596 | 3935 | struct rq *rq; |
31656519 | 3936 | int cpu; |
dd41f596 | 3937 | |
ff743345 PZ |
3938 | need_resched: |
3939 | preempt_disable(); | |
dd41f596 IM |
3940 | cpu = smp_processor_id(); |
3941 | rq = cpu_rq(cpu); | |
25502a6c | 3942 | rcu_note_context_switch(cpu); |
dd41f596 | 3943 | prev = rq->curr; |
dd41f596 IM |
3944 | |
3945 | release_kernel_lock(prev); | |
3946 | need_resched_nonpreemptible: | |
3947 | ||
3948 | schedule_debug(prev); | |
1da177e4 | 3949 | |
31656519 | 3950 | if (sched_feat(HRTICK)) |
f333fdc9 | 3951 | hrtick_clear(rq); |
8f4d37ec | 3952 | |
05fa785c | 3953 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 3954 | |
246d86b5 | 3955 | switch_count = &prev->nivcsw; |
1da177e4 | 3956 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3957 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3958 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
3959 | } else { |
3960 | /* | |
3961 | * If a worker is going to sleep, notify and | |
3962 | * ask workqueue whether it wants to wake up a | |
3963 | * task to maintain concurrency. If so, wake | |
3964 | * up the task. | |
3965 | */ | |
3966 | if (prev->flags & PF_WQ_WORKER) { | |
3967 | struct task_struct *to_wakeup; | |
3968 | ||
3969 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3970 | if (to_wakeup) | |
3971 | try_to_wake_up_local(to_wakeup); | |
3972 | } | |
371fd7e7 | 3973 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 3974 | } |
dd41f596 | 3975 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3976 | } |
3977 | ||
3f029d3c | 3978 | pre_schedule(rq, prev); |
f65eda4f | 3979 | |
dd41f596 | 3980 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3981 | idle_balance(cpu, rq); |
1da177e4 | 3982 | |
df1c99d4 | 3983 | put_prev_task(rq, prev); |
b67802ea | 3984 | next = pick_next_task(rq); |
f26f9aff MG |
3985 | clear_tsk_need_resched(prev); |
3986 | rq->skip_clock_update = 0; | |
1da177e4 | 3987 | |
1da177e4 | 3988 | if (likely(prev != next)) { |
673a90a1 | 3989 | sched_info_switch(prev, next); |
49f47433 | 3990 | perf_event_task_sched_out(prev, next); |
673a90a1 | 3991 | |
1da177e4 LT |
3992 | rq->nr_switches++; |
3993 | rq->curr = next; | |
3994 | ++*switch_count; | |
3995 | ||
dd41f596 | 3996 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3997 | /* |
246d86b5 ON |
3998 | * The context switch have flipped the stack from under us |
3999 | * and restored the local variables which were saved when | |
4000 | * this task called schedule() in the past. prev == current | |
4001 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
4002 | */ |
4003 | cpu = smp_processor_id(); | |
4004 | rq = cpu_rq(cpu); | |
1da177e4 | 4005 | } else |
05fa785c | 4006 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 4007 | |
3f029d3c | 4008 | post_schedule(rq); |
1da177e4 | 4009 | |
246d86b5 | 4010 | if (unlikely(reacquire_kernel_lock(prev))) |
1da177e4 | 4011 | goto need_resched_nonpreemptible; |
8f4d37ec | 4012 | |
1da177e4 | 4013 | preempt_enable_no_resched(); |
ff743345 | 4014 | if (need_resched()) |
1da177e4 LT |
4015 | goto need_resched; |
4016 | } | |
1da177e4 LT |
4017 | EXPORT_SYMBOL(schedule); |
4018 | ||
c08f7829 | 4019 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
4020 | /* |
4021 | * Look out! "owner" is an entirely speculative pointer | |
4022 | * access and not reliable. | |
4023 | */ | |
4024 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
4025 | { | |
4026 | unsigned int cpu; | |
4027 | struct rq *rq; | |
4028 | ||
4029 | if (!sched_feat(OWNER_SPIN)) | |
4030 | return 0; | |
4031 | ||
4032 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
4033 | /* | |
4034 | * Need to access the cpu field knowing that | |
4035 | * DEBUG_PAGEALLOC could have unmapped it if | |
4036 | * the mutex owner just released it and exited. | |
4037 | */ | |
4038 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 4039 | return 0; |
0d66bf6d PZ |
4040 | #else |
4041 | cpu = owner->cpu; | |
4042 | #endif | |
4043 | ||
4044 | /* | |
4045 | * Even if the access succeeded (likely case), | |
4046 | * the cpu field may no longer be valid. | |
4047 | */ | |
4048 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 4049 | return 0; |
0d66bf6d PZ |
4050 | |
4051 | /* | |
4052 | * We need to validate that we can do a | |
4053 | * get_cpu() and that we have the percpu area. | |
4054 | */ | |
4055 | if (!cpu_online(cpu)) | |
4b402210 | 4056 | return 0; |
0d66bf6d PZ |
4057 | |
4058 | rq = cpu_rq(cpu); | |
4059 | ||
4060 | for (;;) { | |
4061 | /* | |
4062 | * Owner changed, break to re-assess state. | |
4063 | */ | |
9d0f4dcc TC |
4064 | if (lock->owner != owner) { |
4065 | /* | |
4066 | * If the lock has switched to a different owner, | |
4067 | * we likely have heavy contention. Return 0 to quit | |
4068 | * optimistic spinning and not contend further: | |
4069 | */ | |
4070 | if (lock->owner) | |
4071 | return 0; | |
0d66bf6d | 4072 | break; |
9d0f4dcc | 4073 | } |
0d66bf6d PZ |
4074 | |
4075 | /* | |
4076 | * Is that owner really running on that cpu? | |
4077 | */ | |
4078 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
4079 | return 0; | |
4080 | ||
335d7afb | 4081 | arch_mutex_cpu_relax(); |
0d66bf6d | 4082 | } |
4b402210 | 4083 | |
0d66bf6d PZ |
4084 | return 1; |
4085 | } | |
4086 | #endif | |
4087 | ||
1da177e4 LT |
4088 | #ifdef CONFIG_PREEMPT |
4089 | /* | |
2ed6e34f | 4090 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4091 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4092 | * occur there and call schedule directly. |
4093 | */ | |
d1f74e20 | 4094 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4095 | { |
4096 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4097 | |
1da177e4 LT |
4098 | /* |
4099 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4100 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4101 | */ |
beed33a8 | 4102 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4103 | return; |
4104 | ||
3a5c359a | 4105 | do { |
d1f74e20 | 4106 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4107 | schedule(); |
d1f74e20 | 4108 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4109 | |
3a5c359a AK |
4110 | /* |
4111 | * Check again in case we missed a preemption opportunity | |
4112 | * between schedule and now. | |
4113 | */ | |
4114 | barrier(); | |
5ed0cec0 | 4115 | } while (need_resched()); |
1da177e4 | 4116 | } |
1da177e4 LT |
4117 | EXPORT_SYMBOL(preempt_schedule); |
4118 | ||
4119 | /* | |
2ed6e34f | 4120 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4121 | * off of irq context. |
4122 | * Note, that this is called and return with irqs disabled. This will | |
4123 | * protect us against recursive calling from irq. | |
4124 | */ | |
4125 | asmlinkage void __sched preempt_schedule_irq(void) | |
4126 | { | |
4127 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4128 | |
2ed6e34f | 4129 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4130 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4131 | ||
3a5c359a AK |
4132 | do { |
4133 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4134 | local_irq_enable(); |
4135 | schedule(); | |
4136 | local_irq_disable(); | |
3a5c359a | 4137 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4138 | |
3a5c359a AK |
4139 | /* |
4140 | * Check again in case we missed a preemption opportunity | |
4141 | * between schedule and now. | |
4142 | */ | |
4143 | barrier(); | |
5ed0cec0 | 4144 | } while (need_resched()); |
1da177e4 LT |
4145 | } |
4146 | ||
4147 | #endif /* CONFIG_PREEMPT */ | |
4148 | ||
63859d4f | 4149 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4150 | void *key) |
1da177e4 | 4151 | { |
63859d4f | 4152 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4153 | } |
1da177e4 LT |
4154 | EXPORT_SYMBOL(default_wake_function); |
4155 | ||
4156 | /* | |
41a2d6cf IM |
4157 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4158 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4159 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4160 | * | |
4161 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4162 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4163 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4164 | */ | |
78ddb08f | 4165 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4166 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4167 | { |
2e45874c | 4168 | wait_queue_t *curr, *next; |
1da177e4 | 4169 | |
2e45874c | 4170 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4171 | unsigned flags = curr->flags; |
4172 | ||
63859d4f | 4173 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4174 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4175 | break; |
4176 | } | |
4177 | } | |
4178 | ||
4179 | /** | |
4180 | * __wake_up - wake up threads blocked on a waitqueue. | |
4181 | * @q: the waitqueue | |
4182 | * @mode: which threads | |
4183 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4184 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4185 | * |
4186 | * It may be assumed that this function implies a write memory barrier before | |
4187 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4188 | */ |
7ad5b3a5 | 4189 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4190 | int nr_exclusive, void *key) |
1da177e4 LT |
4191 | { |
4192 | unsigned long flags; | |
4193 | ||
4194 | spin_lock_irqsave(&q->lock, flags); | |
4195 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4196 | spin_unlock_irqrestore(&q->lock, flags); | |
4197 | } | |
1da177e4 LT |
4198 | EXPORT_SYMBOL(__wake_up); |
4199 | ||
4200 | /* | |
4201 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4202 | */ | |
7ad5b3a5 | 4203 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4204 | { |
4205 | __wake_up_common(q, mode, 1, 0, NULL); | |
4206 | } | |
22c43c81 | 4207 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4208 | |
4ede816a DL |
4209 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4210 | { | |
4211 | __wake_up_common(q, mode, 1, 0, key); | |
4212 | } | |
4213 | ||
1da177e4 | 4214 | /** |
4ede816a | 4215 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4216 | * @q: the waitqueue |
4217 | * @mode: which threads | |
4218 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4219 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4220 | * |
4221 | * The sync wakeup differs that the waker knows that it will schedule | |
4222 | * away soon, so while the target thread will be woken up, it will not | |
4223 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4224 | * with each other. This can prevent needless bouncing between CPUs. | |
4225 | * | |
4226 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4227 | * |
4228 | * It may be assumed that this function implies a write memory barrier before | |
4229 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4230 | */ |
4ede816a DL |
4231 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4232 | int nr_exclusive, void *key) | |
1da177e4 LT |
4233 | { |
4234 | unsigned long flags; | |
7d478721 | 4235 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4236 | |
4237 | if (unlikely(!q)) | |
4238 | return; | |
4239 | ||
4240 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4241 | wake_flags = 0; |
1da177e4 LT |
4242 | |
4243 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4244 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4245 | spin_unlock_irqrestore(&q->lock, flags); |
4246 | } | |
4ede816a DL |
4247 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4248 | ||
4249 | /* | |
4250 | * __wake_up_sync - see __wake_up_sync_key() | |
4251 | */ | |
4252 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4253 | { | |
4254 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4255 | } | |
1da177e4 LT |
4256 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4257 | ||
65eb3dc6 KD |
4258 | /** |
4259 | * complete: - signals a single thread waiting on this completion | |
4260 | * @x: holds the state of this particular completion | |
4261 | * | |
4262 | * This will wake up a single thread waiting on this completion. Threads will be | |
4263 | * awakened in the same order in which they were queued. | |
4264 | * | |
4265 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4266 | * |
4267 | * It may be assumed that this function implies a write memory barrier before | |
4268 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4269 | */ |
b15136e9 | 4270 | void complete(struct completion *x) |
1da177e4 LT |
4271 | { |
4272 | unsigned long flags; | |
4273 | ||
4274 | spin_lock_irqsave(&x->wait.lock, flags); | |
4275 | x->done++; | |
d9514f6c | 4276 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4277 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4278 | } | |
4279 | EXPORT_SYMBOL(complete); | |
4280 | ||
65eb3dc6 KD |
4281 | /** |
4282 | * complete_all: - signals all threads waiting on this completion | |
4283 | * @x: holds the state of this particular completion | |
4284 | * | |
4285 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4286 | * |
4287 | * It may be assumed that this function implies a write memory barrier before | |
4288 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4289 | */ |
b15136e9 | 4290 | void complete_all(struct completion *x) |
1da177e4 LT |
4291 | { |
4292 | unsigned long flags; | |
4293 | ||
4294 | spin_lock_irqsave(&x->wait.lock, flags); | |
4295 | x->done += UINT_MAX/2; | |
d9514f6c | 4296 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4297 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4298 | } | |
4299 | EXPORT_SYMBOL(complete_all); | |
4300 | ||
8cbbe86d AK |
4301 | static inline long __sched |
4302 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4303 | { |
1da177e4 LT |
4304 | if (!x->done) { |
4305 | DECLARE_WAITQUEUE(wait, current); | |
4306 | ||
a93d2f17 | 4307 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4308 | do { |
94d3d824 | 4309 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4310 | timeout = -ERESTARTSYS; |
4311 | break; | |
8cbbe86d AK |
4312 | } |
4313 | __set_current_state(state); | |
1da177e4 LT |
4314 | spin_unlock_irq(&x->wait.lock); |
4315 | timeout = schedule_timeout(timeout); | |
4316 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4317 | } while (!x->done && timeout); |
1da177e4 | 4318 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4319 | if (!x->done) |
4320 | return timeout; | |
1da177e4 LT |
4321 | } |
4322 | x->done--; | |
ea71a546 | 4323 | return timeout ?: 1; |
1da177e4 | 4324 | } |
1da177e4 | 4325 | |
8cbbe86d AK |
4326 | static long __sched |
4327 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4328 | { |
1da177e4 LT |
4329 | might_sleep(); |
4330 | ||
4331 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4332 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4333 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4334 | return timeout; |
4335 | } | |
1da177e4 | 4336 | |
65eb3dc6 KD |
4337 | /** |
4338 | * wait_for_completion: - waits for completion of a task | |
4339 | * @x: holds the state of this particular completion | |
4340 | * | |
4341 | * This waits to be signaled for completion of a specific task. It is NOT | |
4342 | * interruptible and there is no timeout. | |
4343 | * | |
4344 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4345 | * and interrupt capability. Also see complete(). | |
4346 | */ | |
b15136e9 | 4347 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4348 | { |
4349 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4350 | } |
8cbbe86d | 4351 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4352 | |
65eb3dc6 KD |
4353 | /** |
4354 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4355 | * @x: holds the state of this particular completion | |
4356 | * @timeout: timeout value in jiffies | |
4357 | * | |
4358 | * This waits for either a completion of a specific task to be signaled or for a | |
4359 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4360 | * interruptible. | |
4361 | */ | |
b15136e9 | 4362 | unsigned long __sched |
8cbbe86d | 4363 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4364 | { |
8cbbe86d | 4365 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4366 | } |
8cbbe86d | 4367 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4368 | |
65eb3dc6 KD |
4369 | /** |
4370 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4371 | * @x: holds the state of this particular completion | |
4372 | * | |
4373 | * This waits for completion of a specific task to be signaled. It is | |
4374 | * interruptible. | |
4375 | */ | |
8cbbe86d | 4376 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4377 | { |
51e97990 AK |
4378 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4379 | if (t == -ERESTARTSYS) | |
4380 | return t; | |
4381 | return 0; | |
0fec171c | 4382 | } |
8cbbe86d | 4383 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4384 | |
65eb3dc6 KD |
4385 | /** |
4386 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4387 | * @x: holds the state of this particular completion | |
4388 | * @timeout: timeout value in jiffies | |
4389 | * | |
4390 | * This waits for either a completion of a specific task to be signaled or for a | |
4391 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4392 | */ | |
6bf41237 | 4393 | long __sched |
8cbbe86d AK |
4394 | wait_for_completion_interruptible_timeout(struct completion *x, |
4395 | unsigned long timeout) | |
0fec171c | 4396 | { |
8cbbe86d | 4397 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4398 | } |
8cbbe86d | 4399 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4400 | |
65eb3dc6 KD |
4401 | /** |
4402 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4403 | * @x: holds the state of this particular completion | |
4404 | * | |
4405 | * This waits to be signaled for completion of a specific task. It can be | |
4406 | * interrupted by a kill signal. | |
4407 | */ | |
009e577e MW |
4408 | int __sched wait_for_completion_killable(struct completion *x) |
4409 | { | |
4410 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4411 | if (t == -ERESTARTSYS) | |
4412 | return t; | |
4413 | return 0; | |
4414 | } | |
4415 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4416 | ||
0aa12fb4 SW |
4417 | /** |
4418 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4419 | * @x: holds the state of this particular completion | |
4420 | * @timeout: timeout value in jiffies | |
4421 | * | |
4422 | * This waits for either a completion of a specific task to be | |
4423 | * signaled or for a specified timeout to expire. It can be | |
4424 | * interrupted by a kill signal. The timeout is in jiffies. | |
4425 | */ | |
6bf41237 | 4426 | long __sched |
0aa12fb4 SW |
4427 | wait_for_completion_killable_timeout(struct completion *x, |
4428 | unsigned long timeout) | |
4429 | { | |
4430 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4431 | } | |
4432 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4433 | ||
be4de352 DC |
4434 | /** |
4435 | * try_wait_for_completion - try to decrement a completion without blocking | |
4436 | * @x: completion structure | |
4437 | * | |
4438 | * Returns: 0 if a decrement cannot be done without blocking | |
4439 | * 1 if a decrement succeeded. | |
4440 | * | |
4441 | * If a completion is being used as a counting completion, | |
4442 | * attempt to decrement the counter without blocking. This | |
4443 | * enables us to avoid waiting if the resource the completion | |
4444 | * is protecting is not available. | |
4445 | */ | |
4446 | bool try_wait_for_completion(struct completion *x) | |
4447 | { | |
7539a3b3 | 4448 | unsigned long flags; |
be4de352 DC |
4449 | int ret = 1; |
4450 | ||
7539a3b3 | 4451 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4452 | if (!x->done) |
4453 | ret = 0; | |
4454 | else | |
4455 | x->done--; | |
7539a3b3 | 4456 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4457 | return ret; |
4458 | } | |
4459 | EXPORT_SYMBOL(try_wait_for_completion); | |
4460 | ||
4461 | /** | |
4462 | * completion_done - Test to see if a completion has any waiters | |
4463 | * @x: completion structure | |
4464 | * | |
4465 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4466 | * 1 if there are no waiters. | |
4467 | * | |
4468 | */ | |
4469 | bool completion_done(struct completion *x) | |
4470 | { | |
7539a3b3 | 4471 | unsigned long flags; |
be4de352 DC |
4472 | int ret = 1; |
4473 | ||
7539a3b3 | 4474 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4475 | if (!x->done) |
4476 | ret = 0; | |
7539a3b3 | 4477 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4478 | return ret; |
4479 | } | |
4480 | EXPORT_SYMBOL(completion_done); | |
4481 | ||
8cbbe86d AK |
4482 | static long __sched |
4483 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4484 | { |
0fec171c IM |
4485 | unsigned long flags; |
4486 | wait_queue_t wait; | |
4487 | ||
4488 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4489 | |
8cbbe86d | 4490 | __set_current_state(state); |
1da177e4 | 4491 | |
8cbbe86d AK |
4492 | spin_lock_irqsave(&q->lock, flags); |
4493 | __add_wait_queue(q, &wait); | |
4494 | spin_unlock(&q->lock); | |
4495 | timeout = schedule_timeout(timeout); | |
4496 | spin_lock_irq(&q->lock); | |
4497 | __remove_wait_queue(q, &wait); | |
4498 | spin_unlock_irqrestore(&q->lock, flags); | |
4499 | ||
4500 | return timeout; | |
4501 | } | |
4502 | ||
4503 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4504 | { | |
4505 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4506 | } |
1da177e4 LT |
4507 | EXPORT_SYMBOL(interruptible_sleep_on); |
4508 | ||
0fec171c | 4509 | long __sched |
95cdf3b7 | 4510 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4511 | { |
8cbbe86d | 4512 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4513 | } |
1da177e4 LT |
4514 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4515 | ||
0fec171c | 4516 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4517 | { |
8cbbe86d | 4518 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4519 | } |
1da177e4 LT |
4520 | EXPORT_SYMBOL(sleep_on); |
4521 | ||
0fec171c | 4522 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4523 | { |
8cbbe86d | 4524 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4525 | } |
1da177e4 LT |
4526 | EXPORT_SYMBOL(sleep_on_timeout); |
4527 | ||
b29739f9 IM |
4528 | #ifdef CONFIG_RT_MUTEXES |
4529 | ||
4530 | /* | |
4531 | * rt_mutex_setprio - set the current priority of a task | |
4532 | * @p: task | |
4533 | * @prio: prio value (kernel-internal form) | |
4534 | * | |
4535 | * This function changes the 'effective' priority of a task. It does | |
4536 | * not touch ->normal_prio like __setscheduler(). | |
4537 | * | |
4538 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4539 | */ | |
36c8b586 | 4540 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4541 | { |
4542 | unsigned long flags; | |
83b699ed | 4543 | int oldprio, on_rq, running; |
70b97a7f | 4544 | struct rq *rq; |
83ab0aa0 | 4545 | const struct sched_class *prev_class; |
b29739f9 IM |
4546 | |
4547 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4548 | ||
4549 | rq = task_rq_lock(p, &flags); | |
4550 | ||
a8027073 | 4551 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4552 | oldprio = p->prio; |
83ab0aa0 | 4553 | prev_class = p->sched_class; |
dd41f596 | 4554 | on_rq = p->se.on_rq; |
051a1d1a | 4555 | running = task_current(rq, p); |
0e1f3483 | 4556 | if (on_rq) |
69be72c1 | 4557 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4558 | if (running) |
4559 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4560 | |
4561 | if (rt_prio(prio)) | |
4562 | p->sched_class = &rt_sched_class; | |
4563 | else | |
4564 | p->sched_class = &fair_sched_class; | |
4565 | ||
b29739f9 IM |
4566 | p->prio = prio; |
4567 | ||
0e1f3483 HS |
4568 | if (running) |
4569 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4570 | if (on_rq) { |
371fd7e7 | 4571 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4572 | |
4573 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4574 | } |
4575 | task_rq_unlock(rq, &flags); | |
4576 | } | |
4577 | ||
4578 | #endif | |
4579 | ||
36c8b586 | 4580 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4581 | { |
dd41f596 | 4582 | int old_prio, delta, on_rq; |
1da177e4 | 4583 | unsigned long flags; |
70b97a7f | 4584 | struct rq *rq; |
1da177e4 LT |
4585 | |
4586 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4587 | return; | |
4588 | /* | |
4589 | * We have to be careful, if called from sys_setpriority(), | |
4590 | * the task might be in the middle of scheduling on another CPU. | |
4591 | */ | |
4592 | rq = task_rq_lock(p, &flags); | |
4593 | /* | |
4594 | * The RT priorities are set via sched_setscheduler(), but we still | |
4595 | * allow the 'normal' nice value to be set - but as expected | |
4596 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4597 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4598 | */ |
e05606d3 | 4599 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4600 | p->static_prio = NICE_TO_PRIO(nice); |
4601 | goto out_unlock; | |
4602 | } | |
dd41f596 | 4603 | on_rq = p->se.on_rq; |
c09595f6 | 4604 | if (on_rq) |
69be72c1 | 4605 | dequeue_task(rq, p, 0); |
1da177e4 | 4606 | |
1da177e4 | 4607 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4608 | set_load_weight(p); |
b29739f9 IM |
4609 | old_prio = p->prio; |
4610 | p->prio = effective_prio(p); | |
4611 | delta = p->prio - old_prio; | |
1da177e4 | 4612 | |
dd41f596 | 4613 | if (on_rq) { |
371fd7e7 | 4614 | enqueue_task(rq, p, 0); |
1da177e4 | 4615 | /* |
d5f9f942 AM |
4616 | * If the task increased its priority or is running and |
4617 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4618 | */ |
d5f9f942 | 4619 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4620 | resched_task(rq->curr); |
4621 | } | |
4622 | out_unlock: | |
4623 | task_rq_unlock(rq, &flags); | |
4624 | } | |
1da177e4 LT |
4625 | EXPORT_SYMBOL(set_user_nice); |
4626 | ||
e43379f1 MM |
4627 | /* |
4628 | * can_nice - check if a task can reduce its nice value | |
4629 | * @p: task | |
4630 | * @nice: nice value | |
4631 | */ | |
36c8b586 | 4632 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4633 | { |
024f4747 MM |
4634 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4635 | int nice_rlim = 20 - nice; | |
48f24c4d | 4636 | |
78d7d407 | 4637 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4638 | capable(CAP_SYS_NICE)); |
4639 | } | |
4640 | ||
1da177e4 LT |
4641 | #ifdef __ARCH_WANT_SYS_NICE |
4642 | ||
4643 | /* | |
4644 | * sys_nice - change the priority of the current process. | |
4645 | * @increment: priority increment | |
4646 | * | |
4647 | * sys_setpriority is a more generic, but much slower function that | |
4648 | * does similar things. | |
4649 | */ | |
5add95d4 | 4650 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4651 | { |
48f24c4d | 4652 | long nice, retval; |
1da177e4 LT |
4653 | |
4654 | /* | |
4655 | * Setpriority might change our priority at the same moment. | |
4656 | * We don't have to worry. Conceptually one call occurs first | |
4657 | * and we have a single winner. | |
4658 | */ | |
e43379f1 MM |
4659 | if (increment < -40) |
4660 | increment = -40; | |
1da177e4 LT |
4661 | if (increment > 40) |
4662 | increment = 40; | |
4663 | ||
2b8f836f | 4664 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4665 | if (nice < -20) |
4666 | nice = -20; | |
4667 | if (nice > 19) | |
4668 | nice = 19; | |
4669 | ||
e43379f1 MM |
4670 | if (increment < 0 && !can_nice(current, nice)) |
4671 | return -EPERM; | |
4672 | ||
1da177e4 LT |
4673 | retval = security_task_setnice(current, nice); |
4674 | if (retval) | |
4675 | return retval; | |
4676 | ||
4677 | set_user_nice(current, nice); | |
4678 | return 0; | |
4679 | } | |
4680 | ||
4681 | #endif | |
4682 | ||
4683 | /** | |
4684 | * task_prio - return the priority value of a given task. | |
4685 | * @p: the task in question. | |
4686 | * | |
4687 | * This is the priority value as seen by users in /proc. | |
4688 | * RT tasks are offset by -200. Normal tasks are centered | |
4689 | * around 0, value goes from -16 to +15. | |
4690 | */ | |
36c8b586 | 4691 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4692 | { |
4693 | return p->prio - MAX_RT_PRIO; | |
4694 | } | |
4695 | ||
4696 | /** | |
4697 | * task_nice - return the nice value of a given task. | |
4698 | * @p: the task in question. | |
4699 | */ | |
36c8b586 | 4700 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4701 | { |
4702 | return TASK_NICE(p); | |
4703 | } | |
150d8bed | 4704 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4705 | |
4706 | /** | |
4707 | * idle_cpu - is a given cpu idle currently? | |
4708 | * @cpu: the processor in question. | |
4709 | */ | |
4710 | int idle_cpu(int cpu) | |
4711 | { | |
4712 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4713 | } | |
4714 | ||
1da177e4 LT |
4715 | /** |
4716 | * idle_task - return the idle task for a given cpu. | |
4717 | * @cpu: the processor in question. | |
4718 | */ | |
36c8b586 | 4719 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4720 | { |
4721 | return cpu_rq(cpu)->idle; | |
4722 | } | |
4723 | ||
4724 | /** | |
4725 | * find_process_by_pid - find a process with a matching PID value. | |
4726 | * @pid: the pid in question. | |
4727 | */ | |
a9957449 | 4728 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4729 | { |
228ebcbe | 4730 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4731 | } |
4732 | ||
4733 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4734 | static void |
4735 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4736 | { |
dd41f596 | 4737 | BUG_ON(p->se.on_rq); |
48f24c4d | 4738 | |
1da177e4 LT |
4739 | p->policy = policy; |
4740 | p->rt_priority = prio; | |
b29739f9 IM |
4741 | p->normal_prio = normal_prio(p); |
4742 | /* we are holding p->pi_lock already */ | |
4743 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4744 | if (rt_prio(p->prio)) |
4745 | p->sched_class = &rt_sched_class; | |
4746 | else | |
4747 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4748 | set_load_weight(p); |
1da177e4 LT |
4749 | } |
4750 | ||
c69e8d9c DH |
4751 | /* |
4752 | * check the target process has a UID that matches the current process's | |
4753 | */ | |
4754 | static bool check_same_owner(struct task_struct *p) | |
4755 | { | |
4756 | const struct cred *cred = current_cred(), *pcred; | |
4757 | bool match; | |
4758 | ||
4759 | rcu_read_lock(); | |
4760 | pcred = __task_cred(p); | |
4761 | match = (cred->euid == pcred->euid || | |
4762 | cred->euid == pcred->uid); | |
4763 | rcu_read_unlock(); | |
4764 | return match; | |
4765 | } | |
4766 | ||
961ccddd | 4767 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4768 | const struct sched_param *param, bool user) |
1da177e4 | 4769 | { |
83b699ed | 4770 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4771 | unsigned long flags; |
83ab0aa0 | 4772 | const struct sched_class *prev_class; |
70b97a7f | 4773 | struct rq *rq; |
ca94c442 | 4774 | int reset_on_fork; |
1da177e4 | 4775 | |
66e5393a SR |
4776 | /* may grab non-irq protected spin_locks */ |
4777 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4778 | recheck: |
4779 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4780 | if (policy < 0) { |
4781 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4782 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4783 | } else { |
4784 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4785 | policy &= ~SCHED_RESET_ON_FORK; | |
4786 | ||
4787 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4788 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4789 | policy != SCHED_IDLE) | |
4790 | return -EINVAL; | |
4791 | } | |
4792 | ||
1da177e4 LT |
4793 | /* |
4794 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4795 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4796 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4797 | */ |
4798 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4799 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4800 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4801 | return -EINVAL; |
e05606d3 | 4802 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4803 | return -EINVAL; |
4804 | ||
37e4ab3f OC |
4805 | /* |
4806 | * Allow unprivileged RT tasks to decrease priority: | |
4807 | */ | |
961ccddd | 4808 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4809 | if (rt_policy(policy)) { |
a44702e8 ON |
4810 | unsigned long rlim_rtprio = |
4811 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4812 | |
4813 | /* can't set/change the rt policy */ | |
4814 | if (policy != p->policy && !rlim_rtprio) | |
4815 | return -EPERM; | |
4816 | ||
4817 | /* can't increase priority */ | |
4818 | if (param->sched_priority > p->rt_priority && | |
4819 | param->sched_priority > rlim_rtprio) | |
4820 | return -EPERM; | |
4821 | } | |
dd41f596 IM |
4822 | /* |
4823 | * Like positive nice levels, dont allow tasks to | |
4824 | * move out of SCHED_IDLE either: | |
4825 | */ | |
4826 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4827 | return -EPERM; | |
5fe1d75f | 4828 | |
37e4ab3f | 4829 | /* can't change other user's priorities */ |
c69e8d9c | 4830 | if (!check_same_owner(p)) |
37e4ab3f | 4831 | return -EPERM; |
ca94c442 LP |
4832 | |
4833 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4834 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4835 | return -EPERM; | |
37e4ab3f | 4836 | } |
1da177e4 | 4837 | |
725aad24 | 4838 | if (user) { |
b0ae1981 | 4839 | retval = security_task_setscheduler(p); |
725aad24 JF |
4840 | if (retval) |
4841 | return retval; | |
4842 | } | |
4843 | ||
b29739f9 IM |
4844 | /* |
4845 | * make sure no PI-waiters arrive (or leave) while we are | |
4846 | * changing the priority of the task: | |
4847 | */ | |
1d615482 | 4848 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4849 | /* |
4850 | * To be able to change p->policy safely, the apropriate | |
4851 | * runqueue lock must be held. | |
4852 | */ | |
b29739f9 | 4853 | rq = __task_rq_lock(p); |
dc61b1d6 | 4854 | |
34f971f6 PZ |
4855 | /* |
4856 | * Changing the policy of the stop threads its a very bad idea | |
4857 | */ | |
4858 | if (p == rq->stop) { | |
4859 | __task_rq_unlock(rq); | |
4860 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4861 | return -EINVAL; | |
4862 | } | |
4863 | ||
dc61b1d6 PZ |
4864 | #ifdef CONFIG_RT_GROUP_SCHED |
4865 | if (user) { | |
4866 | /* | |
4867 | * Do not allow realtime tasks into groups that have no runtime | |
4868 | * assigned. | |
4869 | */ | |
4870 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4871 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4872 | !task_group_is_autogroup(task_group(p))) { | |
dc61b1d6 PZ |
4873 | __task_rq_unlock(rq); |
4874 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4875 | return -EPERM; | |
4876 | } | |
4877 | } | |
4878 | #endif | |
4879 | ||
1da177e4 LT |
4880 | /* recheck policy now with rq lock held */ |
4881 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4882 | policy = oldpolicy = -1; | |
b29739f9 | 4883 | __task_rq_unlock(rq); |
1d615482 | 4884 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4885 | goto recheck; |
4886 | } | |
dd41f596 | 4887 | on_rq = p->se.on_rq; |
051a1d1a | 4888 | running = task_current(rq, p); |
0e1f3483 | 4889 | if (on_rq) |
2e1cb74a | 4890 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4891 | if (running) |
4892 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4893 | |
ca94c442 LP |
4894 | p->sched_reset_on_fork = reset_on_fork; |
4895 | ||
1da177e4 | 4896 | oldprio = p->prio; |
83ab0aa0 | 4897 | prev_class = p->sched_class; |
dd41f596 | 4898 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4899 | |
0e1f3483 HS |
4900 | if (running) |
4901 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4902 | if (on_rq) { |
4903 | activate_task(rq, p, 0); | |
cb469845 SR |
4904 | |
4905 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4906 | } |
b29739f9 | 4907 | __task_rq_unlock(rq); |
1d615482 | 4908 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4909 | |
95e02ca9 TG |
4910 | rt_mutex_adjust_pi(p); |
4911 | ||
1da177e4 LT |
4912 | return 0; |
4913 | } | |
961ccddd RR |
4914 | |
4915 | /** | |
4916 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4917 | * @p: the task in question. | |
4918 | * @policy: new policy. | |
4919 | * @param: structure containing the new RT priority. | |
4920 | * | |
4921 | * NOTE that the task may be already dead. | |
4922 | */ | |
4923 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4924 | const struct sched_param *param) |
961ccddd RR |
4925 | { |
4926 | return __sched_setscheduler(p, policy, param, true); | |
4927 | } | |
1da177e4 LT |
4928 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4929 | ||
961ccddd RR |
4930 | /** |
4931 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4932 | * @p: the task in question. | |
4933 | * @policy: new policy. | |
4934 | * @param: structure containing the new RT priority. | |
4935 | * | |
4936 | * Just like sched_setscheduler, only don't bother checking if the | |
4937 | * current context has permission. For example, this is needed in | |
4938 | * stop_machine(): we create temporary high priority worker threads, | |
4939 | * but our caller might not have that capability. | |
4940 | */ | |
4941 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4942 | const struct sched_param *param) |
961ccddd RR |
4943 | { |
4944 | return __sched_setscheduler(p, policy, param, false); | |
4945 | } | |
4946 | ||
95cdf3b7 IM |
4947 | static int |
4948 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4949 | { |
1da177e4 LT |
4950 | struct sched_param lparam; |
4951 | struct task_struct *p; | |
36c8b586 | 4952 | int retval; |
1da177e4 LT |
4953 | |
4954 | if (!param || pid < 0) | |
4955 | return -EINVAL; | |
4956 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4957 | return -EFAULT; | |
5fe1d75f ON |
4958 | |
4959 | rcu_read_lock(); | |
4960 | retval = -ESRCH; | |
1da177e4 | 4961 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4962 | if (p != NULL) |
4963 | retval = sched_setscheduler(p, policy, &lparam); | |
4964 | rcu_read_unlock(); | |
36c8b586 | 4965 | |
1da177e4 LT |
4966 | return retval; |
4967 | } | |
4968 | ||
4969 | /** | |
4970 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4971 | * @pid: the pid in question. | |
4972 | * @policy: new policy. | |
4973 | * @param: structure containing the new RT priority. | |
4974 | */ | |
5add95d4 HC |
4975 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4976 | struct sched_param __user *, param) | |
1da177e4 | 4977 | { |
c21761f1 JB |
4978 | /* negative values for policy are not valid */ |
4979 | if (policy < 0) | |
4980 | return -EINVAL; | |
4981 | ||
1da177e4 LT |
4982 | return do_sched_setscheduler(pid, policy, param); |
4983 | } | |
4984 | ||
4985 | /** | |
4986 | * sys_sched_setparam - set/change the RT priority of a thread | |
4987 | * @pid: the pid in question. | |
4988 | * @param: structure containing the new RT priority. | |
4989 | */ | |
5add95d4 | 4990 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4991 | { |
4992 | return do_sched_setscheduler(pid, -1, param); | |
4993 | } | |
4994 | ||
4995 | /** | |
4996 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4997 | * @pid: the pid in question. | |
4998 | */ | |
5add95d4 | 4999 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 5000 | { |
36c8b586 | 5001 | struct task_struct *p; |
3a5c359a | 5002 | int retval; |
1da177e4 LT |
5003 | |
5004 | if (pid < 0) | |
3a5c359a | 5005 | return -EINVAL; |
1da177e4 LT |
5006 | |
5007 | retval = -ESRCH; | |
5fe85be0 | 5008 | rcu_read_lock(); |
1da177e4 LT |
5009 | p = find_process_by_pid(pid); |
5010 | if (p) { | |
5011 | retval = security_task_getscheduler(p); | |
5012 | if (!retval) | |
ca94c442 LP |
5013 | retval = p->policy |
5014 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 5015 | } |
5fe85be0 | 5016 | rcu_read_unlock(); |
1da177e4 LT |
5017 | return retval; |
5018 | } | |
5019 | ||
5020 | /** | |
ca94c442 | 5021 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5022 | * @pid: the pid in question. |
5023 | * @param: structure containing the RT priority. | |
5024 | */ | |
5add95d4 | 5025 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5026 | { |
5027 | struct sched_param lp; | |
36c8b586 | 5028 | struct task_struct *p; |
3a5c359a | 5029 | int retval; |
1da177e4 LT |
5030 | |
5031 | if (!param || pid < 0) | |
3a5c359a | 5032 | return -EINVAL; |
1da177e4 | 5033 | |
5fe85be0 | 5034 | rcu_read_lock(); |
1da177e4 LT |
5035 | p = find_process_by_pid(pid); |
5036 | retval = -ESRCH; | |
5037 | if (!p) | |
5038 | goto out_unlock; | |
5039 | ||
5040 | retval = security_task_getscheduler(p); | |
5041 | if (retval) | |
5042 | goto out_unlock; | |
5043 | ||
5044 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5045 | rcu_read_unlock(); |
1da177e4 LT |
5046 | |
5047 | /* | |
5048 | * This one might sleep, we cannot do it with a spinlock held ... | |
5049 | */ | |
5050 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5051 | ||
1da177e4 LT |
5052 | return retval; |
5053 | ||
5054 | out_unlock: | |
5fe85be0 | 5055 | rcu_read_unlock(); |
1da177e4 LT |
5056 | return retval; |
5057 | } | |
5058 | ||
96f874e2 | 5059 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5060 | { |
5a16f3d3 | 5061 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5062 | struct task_struct *p; |
5063 | int retval; | |
1da177e4 | 5064 | |
95402b38 | 5065 | get_online_cpus(); |
23f5d142 | 5066 | rcu_read_lock(); |
1da177e4 LT |
5067 | |
5068 | p = find_process_by_pid(pid); | |
5069 | if (!p) { | |
23f5d142 | 5070 | rcu_read_unlock(); |
95402b38 | 5071 | put_online_cpus(); |
1da177e4 LT |
5072 | return -ESRCH; |
5073 | } | |
5074 | ||
23f5d142 | 5075 | /* Prevent p going away */ |
1da177e4 | 5076 | get_task_struct(p); |
23f5d142 | 5077 | rcu_read_unlock(); |
1da177e4 | 5078 | |
5a16f3d3 RR |
5079 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5080 | retval = -ENOMEM; | |
5081 | goto out_put_task; | |
5082 | } | |
5083 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5084 | retval = -ENOMEM; | |
5085 | goto out_free_cpus_allowed; | |
5086 | } | |
1da177e4 | 5087 | retval = -EPERM; |
c69e8d9c | 5088 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
5089 | goto out_unlock; |
5090 | ||
b0ae1981 | 5091 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5092 | if (retval) |
5093 | goto out_unlock; | |
5094 | ||
5a16f3d3 RR |
5095 | cpuset_cpus_allowed(p, cpus_allowed); |
5096 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5097 | again: |
5a16f3d3 | 5098 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5099 | |
8707d8b8 | 5100 | if (!retval) { |
5a16f3d3 RR |
5101 | cpuset_cpus_allowed(p, cpus_allowed); |
5102 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5103 | /* |
5104 | * We must have raced with a concurrent cpuset | |
5105 | * update. Just reset the cpus_allowed to the | |
5106 | * cpuset's cpus_allowed | |
5107 | */ | |
5a16f3d3 | 5108 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5109 | goto again; |
5110 | } | |
5111 | } | |
1da177e4 | 5112 | out_unlock: |
5a16f3d3 RR |
5113 | free_cpumask_var(new_mask); |
5114 | out_free_cpus_allowed: | |
5115 | free_cpumask_var(cpus_allowed); | |
5116 | out_put_task: | |
1da177e4 | 5117 | put_task_struct(p); |
95402b38 | 5118 | put_online_cpus(); |
1da177e4 LT |
5119 | return retval; |
5120 | } | |
5121 | ||
5122 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5123 | struct cpumask *new_mask) |
1da177e4 | 5124 | { |
96f874e2 RR |
5125 | if (len < cpumask_size()) |
5126 | cpumask_clear(new_mask); | |
5127 | else if (len > cpumask_size()) | |
5128 | len = cpumask_size(); | |
5129 | ||
1da177e4 LT |
5130 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5131 | } | |
5132 | ||
5133 | /** | |
5134 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5135 | * @pid: pid of the process | |
5136 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5137 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5138 | */ | |
5add95d4 HC |
5139 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5140 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5141 | { |
5a16f3d3 | 5142 | cpumask_var_t new_mask; |
1da177e4 LT |
5143 | int retval; |
5144 | ||
5a16f3d3 RR |
5145 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5146 | return -ENOMEM; | |
1da177e4 | 5147 | |
5a16f3d3 RR |
5148 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5149 | if (retval == 0) | |
5150 | retval = sched_setaffinity(pid, new_mask); | |
5151 | free_cpumask_var(new_mask); | |
5152 | return retval; | |
1da177e4 LT |
5153 | } |
5154 | ||
96f874e2 | 5155 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5156 | { |
36c8b586 | 5157 | struct task_struct *p; |
31605683 TG |
5158 | unsigned long flags; |
5159 | struct rq *rq; | |
1da177e4 | 5160 | int retval; |
1da177e4 | 5161 | |
95402b38 | 5162 | get_online_cpus(); |
23f5d142 | 5163 | rcu_read_lock(); |
1da177e4 LT |
5164 | |
5165 | retval = -ESRCH; | |
5166 | p = find_process_by_pid(pid); | |
5167 | if (!p) | |
5168 | goto out_unlock; | |
5169 | ||
e7834f8f DQ |
5170 | retval = security_task_getscheduler(p); |
5171 | if (retval) | |
5172 | goto out_unlock; | |
5173 | ||
31605683 | 5174 | rq = task_rq_lock(p, &flags); |
96f874e2 | 5175 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 5176 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5177 | |
5178 | out_unlock: | |
23f5d142 | 5179 | rcu_read_unlock(); |
95402b38 | 5180 | put_online_cpus(); |
1da177e4 | 5181 | |
9531b62f | 5182 | return retval; |
1da177e4 LT |
5183 | } |
5184 | ||
5185 | /** | |
5186 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5187 | * @pid: pid of the process | |
5188 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5189 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5190 | */ | |
5add95d4 HC |
5191 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5192 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5193 | { |
5194 | int ret; | |
f17c8607 | 5195 | cpumask_var_t mask; |
1da177e4 | 5196 | |
84fba5ec | 5197 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5198 | return -EINVAL; |
5199 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5200 | return -EINVAL; |
5201 | ||
f17c8607 RR |
5202 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5203 | return -ENOMEM; | |
1da177e4 | 5204 | |
f17c8607 RR |
5205 | ret = sched_getaffinity(pid, mask); |
5206 | if (ret == 0) { | |
8bc037fb | 5207 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5208 | |
5209 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5210 | ret = -EFAULT; |
5211 | else | |
cd3d8031 | 5212 | ret = retlen; |
f17c8607 RR |
5213 | } |
5214 | free_cpumask_var(mask); | |
1da177e4 | 5215 | |
f17c8607 | 5216 | return ret; |
1da177e4 LT |
5217 | } |
5218 | ||
5219 | /** | |
5220 | * sys_sched_yield - yield the current processor to other threads. | |
5221 | * | |
dd41f596 IM |
5222 | * This function yields the current CPU to other tasks. If there are no |
5223 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5224 | */ |
5add95d4 | 5225 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5226 | { |
70b97a7f | 5227 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5228 | |
2d72376b | 5229 | schedstat_inc(rq, yld_count); |
4530d7ab | 5230 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5231 | |
5232 | /* | |
5233 | * Since we are going to call schedule() anyway, there's | |
5234 | * no need to preempt or enable interrupts: | |
5235 | */ | |
5236 | __release(rq->lock); | |
8a25d5de | 5237 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5238 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5239 | preempt_enable_no_resched(); |
5240 | ||
5241 | schedule(); | |
5242 | ||
5243 | return 0; | |
5244 | } | |
5245 | ||
d86ee480 PZ |
5246 | static inline int should_resched(void) |
5247 | { | |
5248 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5249 | } | |
5250 | ||
e7b38404 | 5251 | static void __cond_resched(void) |
1da177e4 | 5252 | { |
e7aaaa69 FW |
5253 | add_preempt_count(PREEMPT_ACTIVE); |
5254 | schedule(); | |
5255 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5256 | } |
5257 | ||
02b67cc3 | 5258 | int __sched _cond_resched(void) |
1da177e4 | 5259 | { |
d86ee480 | 5260 | if (should_resched()) { |
1da177e4 LT |
5261 | __cond_resched(); |
5262 | return 1; | |
5263 | } | |
5264 | return 0; | |
5265 | } | |
02b67cc3 | 5266 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5267 | |
5268 | /* | |
613afbf8 | 5269 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5270 | * call schedule, and on return reacquire the lock. |
5271 | * | |
41a2d6cf | 5272 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5273 | * operations here to prevent schedule() from being called twice (once via |
5274 | * spin_unlock(), once by hand). | |
5275 | */ | |
613afbf8 | 5276 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5277 | { |
d86ee480 | 5278 | int resched = should_resched(); |
6df3cecb JK |
5279 | int ret = 0; |
5280 | ||
f607c668 PZ |
5281 | lockdep_assert_held(lock); |
5282 | ||
95c354fe | 5283 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5284 | spin_unlock(lock); |
d86ee480 | 5285 | if (resched) |
95c354fe NP |
5286 | __cond_resched(); |
5287 | else | |
5288 | cpu_relax(); | |
6df3cecb | 5289 | ret = 1; |
1da177e4 | 5290 | spin_lock(lock); |
1da177e4 | 5291 | } |
6df3cecb | 5292 | return ret; |
1da177e4 | 5293 | } |
613afbf8 | 5294 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5295 | |
613afbf8 | 5296 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5297 | { |
5298 | BUG_ON(!in_softirq()); | |
5299 | ||
d86ee480 | 5300 | if (should_resched()) { |
98d82567 | 5301 | local_bh_enable(); |
1da177e4 LT |
5302 | __cond_resched(); |
5303 | local_bh_disable(); | |
5304 | return 1; | |
5305 | } | |
5306 | return 0; | |
5307 | } | |
613afbf8 | 5308 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5309 | |
1da177e4 LT |
5310 | /** |
5311 | * yield - yield the current processor to other threads. | |
5312 | * | |
72fd4a35 | 5313 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5314 | * thread runnable and calls sys_sched_yield(). |
5315 | */ | |
5316 | void __sched yield(void) | |
5317 | { | |
5318 | set_current_state(TASK_RUNNING); | |
5319 | sys_sched_yield(); | |
5320 | } | |
1da177e4 LT |
5321 | EXPORT_SYMBOL(yield); |
5322 | ||
5323 | /* | |
41a2d6cf | 5324 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5325 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5326 | */ |
5327 | void __sched io_schedule(void) | |
5328 | { | |
54d35f29 | 5329 | struct rq *rq = raw_rq(); |
1da177e4 | 5330 | |
0ff92245 | 5331 | delayacct_blkio_start(); |
1da177e4 | 5332 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5333 | current->in_iowait = 1; |
1da177e4 | 5334 | schedule(); |
8f0dfc34 | 5335 | current->in_iowait = 0; |
1da177e4 | 5336 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5337 | delayacct_blkio_end(); |
1da177e4 | 5338 | } |
1da177e4 LT |
5339 | EXPORT_SYMBOL(io_schedule); |
5340 | ||
5341 | long __sched io_schedule_timeout(long timeout) | |
5342 | { | |
54d35f29 | 5343 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5344 | long ret; |
5345 | ||
0ff92245 | 5346 | delayacct_blkio_start(); |
1da177e4 | 5347 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5348 | current->in_iowait = 1; |
1da177e4 | 5349 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5350 | current->in_iowait = 0; |
1da177e4 | 5351 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5352 | delayacct_blkio_end(); |
1da177e4 LT |
5353 | return ret; |
5354 | } | |
5355 | ||
5356 | /** | |
5357 | * sys_sched_get_priority_max - return maximum RT priority. | |
5358 | * @policy: scheduling class. | |
5359 | * | |
5360 | * this syscall returns the maximum rt_priority that can be used | |
5361 | * by a given scheduling class. | |
5362 | */ | |
5add95d4 | 5363 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5364 | { |
5365 | int ret = -EINVAL; | |
5366 | ||
5367 | switch (policy) { | |
5368 | case SCHED_FIFO: | |
5369 | case SCHED_RR: | |
5370 | ret = MAX_USER_RT_PRIO-1; | |
5371 | break; | |
5372 | case SCHED_NORMAL: | |
b0a9499c | 5373 | case SCHED_BATCH: |
dd41f596 | 5374 | case SCHED_IDLE: |
1da177e4 LT |
5375 | ret = 0; |
5376 | break; | |
5377 | } | |
5378 | return ret; | |
5379 | } | |
5380 | ||
5381 | /** | |
5382 | * sys_sched_get_priority_min - return minimum RT priority. | |
5383 | * @policy: scheduling class. | |
5384 | * | |
5385 | * this syscall returns the minimum rt_priority that can be used | |
5386 | * by a given scheduling class. | |
5387 | */ | |
5add95d4 | 5388 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5389 | { |
5390 | int ret = -EINVAL; | |
5391 | ||
5392 | switch (policy) { | |
5393 | case SCHED_FIFO: | |
5394 | case SCHED_RR: | |
5395 | ret = 1; | |
5396 | break; | |
5397 | case SCHED_NORMAL: | |
b0a9499c | 5398 | case SCHED_BATCH: |
dd41f596 | 5399 | case SCHED_IDLE: |
1da177e4 LT |
5400 | ret = 0; |
5401 | } | |
5402 | return ret; | |
5403 | } | |
5404 | ||
5405 | /** | |
5406 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5407 | * @pid: pid of the process. | |
5408 | * @interval: userspace pointer to the timeslice value. | |
5409 | * | |
5410 | * this syscall writes the default timeslice value of a given process | |
5411 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5412 | */ | |
17da2bd9 | 5413 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5414 | struct timespec __user *, interval) |
1da177e4 | 5415 | { |
36c8b586 | 5416 | struct task_struct *p; |
a4ec24b4 | 5417 | unsigned int time_slice; |
dba091b9 TG |
5418 | unsigned long flags; |
5419 | struct rq *rq; | |
3a5c359a | 5420 | int retval; |
1da177e4 | 5421 | struct timespec t; |
1da177e4 LT |
5422 | |
5423 | if (pid < 0) | |
3a5c359a | 5424 | return -EINVAL; |
1da177e4 LT |
5425 | |
5426 | retval = -ESRCH; | |
1a551ae7 | 5427 | rcu_read_lock(); |
1da177e4 LT |
5428 | p = find_process_by_pid(pid); |
5429 | if (!p) | |
5430 | goto out_unlock; | |
5431 | ||
5432 | retval = security_task_getscheduler(p); | |
5433 | if (retval) | |
5434 | goto out_unlock; | |
5435 | ||
dba091b9 TG |
5436 | rq = task_rq_lock(p, &flags); |
5437 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5438 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5439 | |
1a551ae7 | 5440 | rcu_read_unlock(); |
a4ec24b4 | 5441 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5442 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5443 | return retval; |
3a5c359a | 5444 | |
1da177e4 | 5445 | out_unlock: |
1a551ae7 | 5446 | rcu_read_unlock(); |
1da177e4 LT |
5447 | return retval; |
5448 | } | |
5449 | ||
7c731e0a | 5450 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5451 | |
82a1fcb9 | 5452 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5453 | { |
1da177e4 | 5454 | unsigned long free = 0; |
36c8b586 | 5455 | unsigned state; |
1da177e4 | 5456 | |
1da177e4 | 5457 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5458 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5459 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5460 | #if BITS_PER_LONG == 32 |
1da177e4 | 5461 | if (state == TASK_RUNNING) |
3df0fc5b | 5462 | printk(KERN_CONT " running "); |
1da177e4 | 5463 | else |
3df0fc5b | 5464 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5465 | #else |
5466 | if (state == TASK_RUNNING) | |
3df0fc5b | 5467 | printk(KERN_CONT " running task "); |
1da177e4 | 5468 | else |
3df0fc5b | 5469 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5470 | #endif |
5471 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5472 | free = stack_not_used(p); |
1da177e4 | 5473 | #endif |
3df0fc5b | 5474 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5475 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5476 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5477 | |
5fb5e6de | 5478 | show_stack(p, NULL); |
1da177e4 LT |
5479 | } |
5480 | ||
e59e2ae2 | 5481 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5482 | { |
36c8b586 | 5483 | struct task_struct *g, *p; |
1da177e4 | 5484 | |
4bd77321 | 5485 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5486 | printk(KERN_INFO |
5487 | " task PC stack pid father\n"); | |
1da177e4 | 5488 | #else |
3df0fc5b PZ |
5489 | printk(KERN_INFO |
5490 | " task PC stack pid father\n"); | |
1da177e4 LT |
5491 | #endif |
5492 | read_lock(&tasklist_lock); | |
5493 | do_each_thread(g, p) { | |
5494 | /* | |
5495 | * reset the NMI-timeout, listing all files on a slow | |
5496 | * console might take alot of time: | |
5497 | */ | |
5498 | touch_nmi_watchdog(); | |
39bc89fd | 5499 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5500 | sched_show_task(p); |
1da177e4 LT |
5501 | } while_each_thread(g, p); |
5502 | ||
04c9167f JF |
5503 | touch_all_softlockup_watchdogs(); |
5504 | ||
dd41f596 IM |
5505 | #ifdef CONFIG_SCHED_DEBUG |
5506 | sysrq_sched_debug_show(); | |
5507 | #endif | |
1da177e4 | 5508 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5509 | /* |
5510 | * Only show locks if all tasks are dumped: | |
5511 | */ | |
93335a21 | 5512 | if (!state_filter) |
e59e2ae2 | 5513 | debug_show_all_locks(); |
1da177e4 LT |
5514 | } |
5515 | ||
1df21055 IM |
5516 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5517 | { | |
dd41f596 | 5518 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5519 | } |
5520 | ||
f340c0d1 IM |
5521 | /** |
5522 | * init_idle - set up an idle thread for a given CPU | |
5523 | * @idle: task in question | |
5524 | * @cpu: cpu the idle task belongs to | |
5525 | * | |
5526 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5527 | * flag, to make booting more robust. | |
5528 | */ | |
5c1e1767 | 5529 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5530 | { |
70b97a7f | 5531 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5532 | unsigned long flags; |
5533 | ||
05fa785c | 5534 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5535 | |
dd41f596 | 5536 | __sched_fork(idle); |
06b83b5f | 5537 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5538 | idle->se.exec_start = sched_clock(); |
5539 | ||
96f874e2 | 5540 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5541 | /* |
5542 | * We're having a chicken and egg problem, even though we are | |
5543 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5544 | * lockdep check in task_group() will fail. | |
5545 | * | |
5546 | * Similar case to sched_fork(). / Alternatively we could | |
5547 | * use task_rq_lock() here and obtain the other rq->lock. | |
5548 | * | |
5549 | * Silence PROVE_RCU | |
5550 | */ | |
5551 | rcu_read_lock(); | |
dd41f596 | 5552 | __set_task_cpu(idle, cpu); |
6506cf6c | 5553 | rcu_read_unlock(); |
1da177e4 | 5554 | |
1da177e4 | 5555 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5556 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5557 | idle->oncpu = 1; | |
5558 | #endif | |
05fa785c | 5559 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5560 | |
5561 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5562 | #if defined(CONFIG_PREEMPT) |
5563 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5564 | #else | |
a1261f54 | 5565 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5566 | #endif |
dd41f596 IM |
5567 | /* |
5568 | * The idle tasks have their own, simple scheduling class: | |
5569 | */ | |
5570 | idle->sched_class = &idle_sched_class; | |
fb52607a | 5571 | ftrace_graph_init_task(idle); |
1da177e4 LT |
5572 | } |
5573 | ||
5574 | /* | |
5575 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5576 | * indicates which cpus entered this state. This is used | |
5577 | * in the rcu update to wait only for active cpus. For system | |
5578 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5579 | * always be CPU_BITS_NONE. |
1da177e4 | 5580 | */ |
6a7b3dc3 | 5581 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5582 | |
19978ca6 IM |
5583 | /* |
5584 | * Increase the granularity value when there are more CPUs, | |
5585 | * because with more CPUs the 'effective latency' as visible | |
5586 | * to users decreases. But the relationship is not linear, | |
5587 | * so pick a second-best guess by going with the log2 of the | |
5588 | * number of CPUs. | |
5589 | * | |
5590 | * This idea comes from the SD scheduler of Con Kolivas: | |
5591 | */ | |
acb4a848 | 5592 | static int get_update_sysctl_factor(void) |
19978ca6 | 5593 | { |
4ca3ef71 | 5594 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5595 | unsigned int factor; |
5596 | ||
5597 | switch (sysctl_sched_tunable_scaling) { | |
5598 | case SCHED_TUNABLESCALING_NONE: | |
5599 | factor = 1; | |
5600 | break; | |
5601 | case SCHED_TUNABLESCALING_LINEAR: | |
5602 | factor = cpus; | |
5603 | break; | |
5604 | case SCHED_TUNABLESCALING_LOG: | |
5605 | default: | |
5606 | factor = 1 + ilog2(cpus); | |
5607 | break; | |
5608 | } | |
19978ca6 | 5609 | |
acb4a848 CE |
5610 | return factor; |
5611 | } | |
19978ca6 | 5612 | |
acb4a848 CE |
5613 | static void update_sysctl(void) |
5614 | { | |
5615 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5616 | |
0bcdcf28 CE |
5617 | #define SET_SYSCTL(name) \ |
5618 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5619 | SET_SYSCTL(sched_min_granularity); | |
5620 | SET_SYSCTL(sched_latency); | |
5621 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5622 | #undef SET_SYSCTL |
5623 | } | |
55cd5340 | 5624 | |
0bcdcf28 CE |
5625 | static inline void sched_init_granularity(void) |
5626 | { | |
5627 | update_sysctl(); | |
19978ca6 IM |
5628 | } |
5629 | ||
1da177e4 LT |
5630 | #ifdef CONFIG_SMP |
5631 | /* | |
5632 | * This is how migration works: | |
5633 | * | |
969c7921 TH |
5634 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5635 | * stop_one_cpu(). | |
5636 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5637 | * off the CPU) | |
5638 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5639 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5640 | * it and puts it into the right queue. |
969c7921 TH |
5641 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5642 | * is done. | |
1da177e4 LT |
5643 | */ |
5644 | ||
5645 | /* | |
5646 | * Change a given task's CPU affinity. Migrate the thread to a | |
5647 | * proper CPU and schedule it away if the CPU it's executing on | |
5648 | * is removed from the allowed bitmask. | |
5649 | * | |
5650 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5651 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5652 | * call is not atomic; no spinlocks may be held. |
5653 | */ | |
96f874e2 | 5654 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5655 | { |
5656 | unsigned long flags; | |
70b97a7f | 5657 | struct rq *rq; |
969c7921 | 5658 | unsigned int dest_cpu; |
48f24c4d | 5659 | int ret = 0; |
1da177e4 | 5660 | |
65cc8e48 PZ |
5661 | /* |
5662 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5663 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5664 | */ | |
5665 | again: | |
5666 | while (task_is_waking(p)) | |
5667 | cpu_relax(); | |
1da177e4 | 5668 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5669 | if (task_is_waking(p)) { |
5670 | task_rq_unlock(rq, &flags); | |
5671 | goto again; | |
5672 | } | |
e2912009 | 5673 | |
6ad4c188 | 5674 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5675 | ret = -EINVAL; |
5676 | goto out; | |
5677 | } | |
5678 | ||
9985b0ba | 5679 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5680 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5681 | ret = -EINVAL; |
5682 | goto out; | |
5683 | } | |
5684 | ||
73fe6aae | 5685 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5686 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5687 | else { |
96f874e2 RR |
5688 | cpumask_copy(&p->cpus_allowed, new_mask); |
5689 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5690 | } |
5691 | ||
1da177e4 | 5692 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5693 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5694 | goto out; |
5695 | ||
969c7921 | 5696 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
b7a2b39d | 5697 | if (migrate_task(p, rq)) { |
969c7921 | 5698 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 LT |
5699 | /* Need help from migration thread: drop lock and wait. */ |
5700 | task_rq_unlock(rq, &flags); | |
969c7921 | 5701 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5702 | tlb_migrate_finish(p->mm); |
5703 | return 0; | |
5704 | } | |
5705 | out: | |
5706 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5707 | |
1da177e4 LT |
5708 | return ret; |
5709 | } | |
cd8ba7cd | 5710 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5711 | |
5712 | /* | |
41a2d6cf | 5713 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5714 | * this because either it can't run here any more (set_cpus_allowed() |
5715 | * away from this CPU, or CPU going down), or because we're | |
5716 | * attempting to rebalance this task on exec (sched_exec). | |
5717 | * | |
5718 | * So we race with normal scheduler movements, but that's OK, as long | |
5719 | * as the task is no longer on this CPU. | |
efc30814 KK |
5720 | * |
5721 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5722 | */ |
efc30814 | 5723 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5724 | { |
70b97a7f | 5725 | struct rq *rq_dest, *rq_src; |
e2912009 | 5726 | int ret = 0; |
1da177e4 | 5727 | |
e761b772 | 5728 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5729 | return ret; |
1da177e4 LT |
5730 | |
5731 | rq_src = cpu_rq(src_cpu); | |
5732 | rq_dest = cpu_rq(dest_cpu); | |
5733 | ||
5734 | double_rq_lock(rq_src, rq_dest); | |
5735 | /* Already moved. */ | |
5736 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5737 | goto done; |
1da177e4 | 5738 | /* Affinity changed (again). */ |
96f874e2 | 5739 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5740 | goto fail; |
1da177e4 | 5741 | |
e2912009 PZ |
5742 | /* |
5743 | * If we're not on a rq, the next wake-up will ensure we're | |
5744 | * placed properly. | |
5745 | */ | |
5746 | if (p->se.on_rq) { | |
2e1cb74a | 5747 | deactivate_task(rq_src, p, 0); |
e2912009 | 5748 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5749 | activate_task(rq_dest, p, 0); |
15afe09b | 5750 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5751 | } |
b1e38734 | 5752 | done: |
efc30814 | 5753 | ret = 1; |
b1e38734 | 5754 | fail: |
1da177e4 | 5755 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5756 | return ret; |
1da177e4 LT |
5757 | } |
5758 | ||
5759 | /* | |
969c7921 TH |
5760 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5761 | * and performs thread migration by bumping thread off CPU then | |
5762 | * 'pushing' onto another runqueue. | |
1da177e4 | 5763 | */ |
969c7921 | 5764 | static int migration_cpu_stop(void *data) |
1da177e4 | 5765 | { |
969c7921 | 5766 | struct migration_arg *arg = data; |
f7b4cddc | 5767 | |
969c7921 TH |
5768 | /* |
5769 | * The original target cpu might have gone down and we might | |
5770 | * be on another cpu but it doesn't matter. | |
5771 | */ | |
f7b4cddc | 5772 | local_irq_disable(); |
969c7921 | 5773 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5774 | local_irq_enable(); |
1da177e4 | 5775 | return 0; |
f7b4cddc ON |
5776 | } |
5777 | ||
1da177e4 | 5778 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5779 | |
054b9108 | 5780 | /* |
48c5ccae PZ |
5781 | * Ensures that the idle task is using init_mm right before its cpu goes |
5782 | * offline. | |
054b9108 | 5783 | */ |
48c5ccae | 5784 | void idle_task_exit(void) |
1da177e4 | 5785 | { |
48c5ccae | 5786 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5787 | |
48c5ccae | 5788 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5789 | |
48c5ccae PZ |
5790 | if (mm != &init_mm) |
5791 | switch_mm(mm, &init_mm, current); | |
5792 | mmdrop(mm); | |
1da177e4 LT |
5793 | } |
5794 | ||
5795 | /* | |
5796 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5797 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5798 | * for performance reasons the counter is not stricly tracking tasks to | |
5799 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5800 | * to keep the global sum constant after CPU-down: | |
5801 | */ | |
70b97a7f | 5802 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5803 | { |
6ad4c188 | 5804 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 5805 | |
1da177e4 LT |
5806 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
5807 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
5808 | } |
5809 | ||
dd41f596 | 5810 | /* |
48c5ccae | 5811 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 5812 | */ |
48c5ccae | 5813 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 5814 | { |
48c5ccae PZ |
5815 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
5816 | rq->calc_load_active = 0; | |
1da177e4 LT |
5817 | } |
5818 | ||
48f24c4d | 5819 | /* |
48c5ccae PZ |
5820 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5821 | * try_to_wake_up()->select_task_rq(). | |
5822 | * | |
5823 | * Called with rq->lock held even though we'er in stop_machine() and | |
5824 | * there's no concurrency possible, we hold the required locks anyway | |
5825 | * because of lock validation efforts. | |
1da177e4 | 5826 | */ |
48c5ccae | 5827 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5828 | { |
70b97a7f | 5829 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5830 | struct task_struct *next, *stop = rq->stop; |
5831 | int dest_cpu; | |
1da177e4 LT |
5832 | |
5833 | /* | |
48c5ccae PZ |
5834 | * Fudge the rq selection such that the below task selection loop |
5835 | * doesn't get stuck on the currently eligible stop task. | |
5836 | * | |
5837 | * We're currently inside stop_machine() and the rq is either stuck | |
5838 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5839 | * either way we should never end up calling schedule() until we're | |
5840 | * done here. | |
1da177e4 | 5841 | */ |
48c5ccae | 5842 | rq->stop = NULL; |
48f24c4d | 5843 | |
dd41f596 | 5844 | for ( ; ; ) { |
48c5ccae PZ |
5845 | /* |
5846 | * There's this thread running, bail when that's the only | |
5847 | * remaining thread. | |
5848 | */ | |
5849 | if (rq->nr_running == 1) | |
dd41f596 | 5850 | break; |
48c5ccae | 5851 | |
b67802ea | 5852 | next = pick_next_task(rq); |
48c5ccae | 5853 | BUG_ON(!next); |
79c53799 | 5854 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5855 | |
48c5ccae PZ |
5856 | /* Find suitable destination for @next, with force if needed. */ |
5857 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5858 | raw_spin_unlock(&rq->lock); | |
5859 | ||
5860 | __migrate_task(next, dead_cpu, dest_cpu); | |
5861 | ||
5862 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5863 | } |
dce48a84 | 5864 | |
48c5ccae | 5865 | rq->stop = stop; |
dce48a84 | 5866 | } |
48c5ccae | 5867 | |
1da177e4 LT |
5868 | #endif /* CONFIG_HOTPLUG_CPU */ |
5869 | ||
e692ab53 NP |
5870 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5871 | ||
5872 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5873 | { |
5874 | .procname = "sched_domain", | |
c57baf1e | 5875 | .mode = 0555, |
e0361851 | 5876 | }, |
56992309 | 5877 | {} |
e692ab53 NP |
5878 | }; |
5879 | ||
5880 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5881 | { |
5882 | .procname = "kernel", | |
c57baf1e | 5883 | .mode = 0555, |
e0361851 AD |
5884 | .child = sd_ctl_dir, |
5885 | }, | |
56992309 | 5886 | {} |
e692ab53 NP |
5887 | }; |
5888 | ||
5889 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5890 | { | |
5891 | struct ctl_table *entry = | |
5cf9f062 | 5892 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5893 | |
e692ab53 NP |
5894 | return entry; |
5895 | } | |
5896 | ||
6382bc90 MM |
5897 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5898 | { | |
cd790076 | 5899 | struct ctl_table *entry; |
6382bc90 | 5900 | |
cd790076 MM |
5901 | /* |
5902 | * In the intermediate directories, both the child directory and | |
5903 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5904 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5905 | * static strings and all have proc handlers. |
5906 | */ | |
5907 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5908 | if (entry->child) |
5909 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5910 | if (entry->proc_handler == NULL) |
5911 | kfree(entry->procname); | |
5912 | } | |
6382bc90 MM |
5913 | |
5914 | kfree(*tablep); | |
5915 | *tablep = NULL; | |
5916 | } | |
5917 | ||
e692ab53 | 5918 | static void |
e0361851 | 5919 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5920 | const char *procname, void *data, int maxlen, |
5921 | mode_t mode, proc_handler *proc_handler) | |
5922 | { | |
e692ab53 NP |
5923 | entry->procname = procname; |
5924 | entry->data = data; | |
5925 | entry->maxlen = maxlen; | |
5926 | entry->mode = mode; | |
5927 | entry->proc_handler = proc_handler; | |
5928 | } | |
5929 | ||
5930 | static struct ctl_table * | |
5931 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5932 | { | |
a5d8c348 | 5933 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5934 | |
ad1cdc1d MM |
5935 | if (table == NULL) |
5936 | return NULL; | |
5937 | ||
e0361851 | 5938 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5939 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5940 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5941 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5942 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5943 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5944 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5945 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5946 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5947 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5948 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5949 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5950 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5951 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5952 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5953 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5954 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5955 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5956 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5957 | &sd->cache_nice_tries, |
5958 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5959 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5960 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5961 | set_table_entry(&table[11], "name", sd->name, |
5962 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5963 | /* &table[12] is terminator */ | |
e692ab53 NP |
5964 | |
5965 | return table; | |
5966 | } | |
5967 | ||
9a4e7159 | 5968 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5969 | { |
5970 | struct ctl_table *entry, *table; | |
5971 | struct sched_domain *sd; | |
5972 | int domain_num = 0, i; | |
5973 | char buf[32]; | |
5974 | ||
5975 | for_each_domain(cpu, sd) | |
5976 | domain_num++; | |
5977 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5978 | if (table == NULL) |
5979 | return NULL; | |
e692ab53 NP |
5980 | |
5981 | i = 0; | |
5982 | for_each_domain(cpu, sd) { | |
5983 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5984 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5985 | entry->mode = 0555; |
e692ab53 NP |
5986 | entry->child = sd_alloc_ctl_domain_table(sd); |
5987 | entry++; | |
5988 | i++; | |
5989 | } | |
5990 | return table; | |
5991 | } | |
5992 | ||
5993 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5994 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5995 | { |
6ad4c188 | 5996 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5997 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5998 | char buf[32]; | |
5999 | ||
7378547f MM |
6000 | WARN_ON(sd_ctl_dir[0].child); |
6001 | sd_ctl_dir[0].child = entry; | |
6002 | ||
ad1cdc1d MM |
6003 | if (entry == NULL) |
6004 | return; | |
6005 | ||
6ad4c188 | 6006 | for_each_possible_cpu(i) { |
e692ab53 | 6007 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 6008 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 6009 | entry->mode = 0555; |
e692ab53 | 6010 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 6011 | entry++; |
e692ab53 | 6012 | } |
7378547f MM |
6013 | |
6014 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
6015 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
6016 | } | |
6382bc90 | 6017 | |
7378547f | 6018 | /* may be called multiple times per register */ |
6382bc90 MM |
6019 | static void unregister_sched_domain_sysctl(void) |
6020 | { | |
7378547f MM |
6021 | if (sd_sysctl_header) |
6022 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6023 | sd_sysctl_header = NULL; |
7378547f MM |
6024 | if (sd_ctl_dir[0].child) |
6025 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6026 | } |
e692ab53 | 6027 | #else |
6382bc90 MM |
6028 | static void register_sched_domain_sysctl(void) |
6029 | { | |
6030 | } | |
6031 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6032 | { |
6033 | } | |
6034 | #endif | |
6035 | ||
1f11eb6a GH |
6036 | static void set_rq_online(struct rq *rq) |
6037 | { | |
6038 | if (!rq->online) { | |
6039 | const struct sched_class *class; | |
6040 | ||
c6c4927b | 6041 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6042 | rq->online = 1; |
6043 | ||
6044 | for_each_class(class) { | |
6045 | if (class->rq_online) | |
6046 | class->rq_online(rq); | |
6047 | } | |
6048 | } | |
6049 | } | |
6050 | ||
6051 | static void set_rq_offline(struct rq *rq) | |
6052 | { | |
6053 | if (rq->online) { | |
6054 | const struct sched_class *class; | |
6055 | ||
6056 | for_each_class(class) { | |
6057 | if (class->rq_offline) | |
6058 | class->rq_offline(rq); | |
6059 | } | |
6060 | ||
c6c4927b | 6061 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6062 | rq->online = 0; |
6063 | } | |
6064 | } | |
6065 | ||
1da177e4 LT |
6066 | /* |
6067 | * migration_call - callback that gets triggered when a CPU is added. | |
6068 | * Here we can start up the necessary migration thread for the new CPU. | |
6069 | */ | |
48f24c4d IM |
6070 | static int __cpuinit |
6071 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6072 | { |
48f24c4d | 6073 | int cpu = (long)hcpu; |
1da177e4 | 6074 | unsigned long flags; |
969c7921 | 6075 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6076 | |
48c5ccae | 6077 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6078 | |
1da177e4 | 6079 | case CPU_UP_PREPARE: |
a468d389 | 6080 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6081 | break; |
48f24c4d | 6082 | |
1da177e4 | 6083 | case CPU_ONLINE: |
1f94ef59 | 6084 | /* Update our root-domain */ |
05fa785c | 6085 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6086 | if (rq->rd) { |
c6c4927b | 6087 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6088 | |
6089 | set_rq_online(rq); | |
1f94ef59 | 6090 | } |
05fa785c | 6091 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6092 | break; |
48f24c4d | 6093 | |
1da177e4 | 6094 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6095 | case CPU_DYING: |
57d885fe | 6096 | /* Update our root-domain */ |
05fa785c | 6097 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6098 | if (rq->rd) { |
c6c4927b | 6099 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6100 | set_rq_offline(rq); |
57d885fe | 6101 | } |
48c5ccae PZ |
6102 | migrate_tasks(cpu); |
6103 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6104 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6105 | |
6106 | migrate_nr_uninterruptible(rq); | |
6107 | calc_global_load_remove(rq); | |
57d885fe | 6108 | break; |
1da177e4 LT |
6109 | #endif |
6110 | } | |
6111 | return NOTIFY_OK; | |
6112 | } | |
6113 | ||
f38b0820 PM |
6114 | /* |
6115 | * Register at high priority so that task migration (migrate_all_tasks) | |
6116 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6117 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6118 | */ |
26c2143b | 6119 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6120 | .notifier_call = migration_call, |
50a323b7 | 6121 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6122 | }; |
6123 | ||
3a101d05 TH |
6124 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6125 | unsigned long action, void *hcpu) | |
6126 | { | |
6127 | switch (action & ~CPU_TASKS_FROZEN) { | |
6128 | case CPU_ONLINE: | |
6129 | case CPU_DOWN_FAILED: | |
6130 | set_cpu_active((long)hcpu, true); | |
6131 | return NOTIFY_OK; | |
6132 | default: | |
6133 | return NOTIFY_DONE; | |
6134 | } | |
6135 | } | |
6136 | ||
6137 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6138 | unsigned long action, void *hcpu) | |
6139 | { | |
6140 | switch (action & ~CPU_TASKS_FROZEN) { | |
6141 | case CPU_DOWN_PREPARE: | |
6142 | set_cpu_active((long)hcpu, false); | |
6143 | return NOTIFY_OK; | |
6144 | default: | |
6145 | return NOTIFY_DONE; | |
6146 | } | |
6147 | } | |
6148 | ||
7babe8db | 6149 | static int __init migration_init(void) |
1da177e4 LT |
6150 | { |
6151 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6152 | int err; |
48f24c4d | 6153 | |
3a101d05 | 6154 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6155 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6156 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6157 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6158 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6159 | |
3a101d05 TH |
6160 | /* Register cpu active notifiers */ |
6161 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6162 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6163 | ||
a004cd42 | 6164 | return 0; |
1da177e4 | 6165 | } |
7babe8db | 6166 | early_initcall(migration_init); |
1da177e4 LT |
6167 | #endif |
6168 | ||
6169 | #ifdef CONFIG_SMP | |
476f3534 | 6170 | |
3e9830dc | 6171 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6172 | |
f6630114 MT |
6173 | static __read_mostly int sched_domain_debug_enabled; |
6174 | ||
6175 | static int __init sched_domain_debug_setup(char *str) | |
6176 | { | |
6177 | sched_domain_debug_enabled = 1; | |
6178 | ||
6179 | return 0; | |
6180 | } | |
6181 | early_param("sched_debug", sched_domain_debug_setup); | |
6182 | ||
7c16ec58 | 6183 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6184 | struct cpumask *groupmask) |
1da177e4 | 6185 | { |
4dcf6aff | 6186 | struct sched_group *group = sd->groups; |
434d53b0 | 6187 | char str[256]; |
1da177e4 | 6188 | |
968ea6d8 | 6189 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6190 | cpumask_clear(groupmask); |
4dcf6aff IM |
6191 | |
6192 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6193 | ||
6194 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6195 | printk("does not load-balance\n"); |
4dcf6aff | 6196 | if (sd->parent) |
3df0fc5b PZ |
6197 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6198 | " has parent"); | |
4dcf6aff | 6199 | return -1; |
41c7ce9a NP |
6200 | } |
6201 | ||
3df0fc5b | 6202 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6203 | |
758b2cdc | 6204 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6205 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6206 | "CPU%d\n", cpu); | |
4dcf6aff | 6207 | } |
758b2cdc | 6208 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6209 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6210 | " CPU%d\n", cpu); | |
4dcf6aff | 6211 | } |
1da177e4 | 6212 | |
4dcf6aff | 6213 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6214 | do { |
4dcf6aff | 6215 | if (!group) { |
3df0fc5b PZ |
6216 | printk("\n"); |
6217 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6218 | break; |
6219 | } | |
6220 | ||
18a3885f | 6221 | if (!group->cpu_power) { |
3df0fc5b PZ |
6222 | printk(KERN_CONT "\n"); |
6223 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6224 | "set\n"); | |
4dcf6aff IM |
6225 | break; |
6226 | } | |
1da177e4 | 6227 | |
758b2cdc | 6228 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6229 | printk(KERN_CONT "\n"); |
6230 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6231 | break; |
6232 | } | |
1da177e4 | 6233 | |
758b2cdc | 6234 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6235 | printk(KERN_CONT "\n"); |
6236 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6237 | break; |
6238 | } | |
1da177e4 | 6239 | |
758b2cdc | 6240 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6241 | |
968ea6d8 | 6242 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6243 | |
3df0fc5b | 6244 | printk(KERN_CONT " %s", str); |
18a3885f | 6245 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6246 | printk(KERN_CONT " (cpu_power = %d)", |
6247 | group->cpu_power); | |
381512cf | 6248 | } |
1da177e4 | 6249 | |
4dcf6aff IM |
6250 | group = group->next; |
6251 | } while (group != sd->groups); | |
3df0fc5b | 6252 | printk(KERN_CONT "\n"); |
1da177e4 | 6253 | |
758b2cdc | 6254 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6255 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6256 | |
758b2cdc RR |
6257 | if (sd->parent && |
6258 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6259 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6260 | "of domain->span\n"); | |
4dcf6aff IM |
6261 | return 0; |
6262 | } | |
1da177e4 | 6263 | |
4dcf6aff IM |
6264 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6265 | { | |
d5dd3db1 | 6266 | cpumask_var_t groupmask; |
4dcf6aff | 6267 | int level = 0; |
1da177e4 | 6268 | |
f6630114 MT |
6269 | if (!sched_domain_debug_enabled) |
6270 | return; | |
6271 | ||
4dcf6aff IM |
6272 | if (!sd) { |
6273 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6274 | return; | |
6275 | } | |
1da177e4 | 6276 | |
4dcf6aff IM |
6277 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6278 | ||
d5dd3db1 | 6279 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6280 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6281 | return; | |
6282 | } | |
6283 | ||
4dcf6aff | 6284 | for (;;) { |
7c16ec58 | 6285 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6286 | break; |
1da177e4 LT |
6287 | level++; |
6288 | sd = sd->parent; | |
33859f7f | 6289 | if (!sd) |
4dcf6aff IM |
6290 | break; |
6291 | } | |
d5dd3db1 | 6292 | free_cpumask_var(groupmask); |
1da177e4 | 6293 | } |
6d6bc0ad | 6294 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6295 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6296 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6297 | |
1a20ff27 | 6298 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6299 | { |
758b2cdc | 6300 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6301 | return 1; |
6302 | ||
6303 | /* Following flags need at least 2 groups */ | |
6304 | if (sd->flags & (SD_LOAD_BALANCE | | |
6305 | SD_BALANCE_NEWIDLE | | |
6306 | SD_BALANCE_FORK | | |
89c4710e SS |
6307 | SD_BALANCE_EXEC | |
6308 | SD_SHARE_CPUPOWER | | |
6309 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6310 | if (sd->groups != sd->groups->next) |
6311 | return 0; | |
6312 | } | |
6313 | ||
6314 | /* Following flags don't use groups */ | |
c88d5910 | 6315 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6316 | return 0; |
6317 | ||
6318 | return 1; | |
6319 | } | |
6320 | ||
48f24c4d IM |
6321 | static int |
6322 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6323 | { |
6324 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6325 | ||
6326 | if (sd_degenerate(parent)) | |
6327 | return 1; | |
6328 | ||
758b2cdc | 6329 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6330 | return 0; |
6331 | ||
245af2c7 SS |
6332 | /* Flags needing groups don't count if only 1 group in parent */ |
6333 | if (parent->groups == parent->groups->next) { | |
6334 | pflags &= ~(SD_LOAD_BALANCE | | |
6335 | SD_BALANCE_NEWIDLE | | |
6336 | SD_BALANCE_FORK | | |
89c4710e SS |
6337 | SD_BALANCE_EXEC | |
6338 | SD_SHARE_CPUPOWER | | |
6339 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6340 | if (nr_node_ids == 1) |
6341 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6342 | } |
6343 | if (~cflags & pflags) | |
6344 | return 0; | |
6345 | ||
6346 | return 1; | |
6347 | } | |
6348 | ||
c6c4927b RR |
6349 | static void free_rootdomain(struct root_domain *rd) |
6350 | { | |
047106ad PZ |
6351 | synchronize_sched(); |
6352 | ||
68e74568 RR |
6353 | cpupri_cleanup(&rd->cpupri); |
6354 | ||
c6c4927b RR |
6355 | free_cpumask_var(rd->rto_mask); |
6356 | free_cpumask_var(rd->online); | |
6357 | free_cpumask_var(rd->span); | |
6358 | kfree(rd); | |
6359 | } | |
6360 | ||
57d885fe GH |
6361 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6362 | { | |
a0490fa3 | 6363 | struct root_domain *old_rd = NULL; |
57d885fe | 6364 | unsigned long flags; |
57d885fe | 6365 | |
05fa785c | 6366 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6367 | |
6368 | if (rq->rd) { | |
a0490fa3 | 6369 | old_rd = rq->rd; |
57d885fe | 6370 | |
c6c4927b | 6371 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6372 | set_rq_offline(rq); |
57d885fe | 6373 | |
c6c4927b | 6374 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6375 | |
a0490fa3 IM |
6376 | /* |
6377 | * If we dont want to free the old_rt yet then | |
6378 | * set old_rd to NULL to skip the freeing later | |
6379 | * in this function: | |
6380 | */ | |
6381 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6382 | old_rd = NULL; | |
57d885fe GH |
6383 | } |
6384 | ||
6385 | atomic_inc(&rd->refcount); | |
6386 | rq->rd = rd; | |
6387 | ||
c6c4927b | 6388 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6389 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6390 | set_rq_online(rq); |
57d885fe | 6391 | |
05fa785c | 6392 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6393 | |
6394 | if (old_rd) | |
6395 | free_rootdomain(old_rd); | |
57d885fe GH |
6396 | } |
6397 | ||
68c38fc3 | 6398 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6399 | { |
6400 | memset(rd, 0, sizeof(*rd)); | |
6401 | ||
68c38fc3 | 6402 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6403 | goto out; |
68c38fc3 | 6404 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6405 | goto free_span; |
68c38fc3 | 6406 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6407 | goto free_online; |
6e0534f2 | 6408 | |
68c38fc3 | 6409 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6410 | goto free_rto_mask; |
c6c4927b | 6411 | return 0; |
6e0534f2 | 6412 | |
68e74568 RR |
6413 | free_rto_mask: |
6414 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6415 | free_online: |
6416 | free_cpumask_var(rd->online); | |
6417 | free_span: | |
6418 | free_cpumask_var(rd->span); | |
0c910d28 | 6419 | out: |
c6c4927b | 6420 | return -ENOMEM; |
57d885fe GH |
6421 | } |
6422 | ||
6423 | static void init_defrootdomain(void) | |
6424 | { | |
68c38fc3 | 6425 | init_rootdomain(&def_root_domain); |
c6c4927b | 6426 | |
57d885fe GH |
6427 | atomic_set(&def_root_domain.refcount, 1); |
6428 | } | |
6429 | ||
dc938520 | 6430 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6431 | { |
6432 | struct root_domain *rd; | |
6433 | ||
6434 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6435 | if (!rd) | |
6436 | return NULL; | |
6437 | ||
68c38fc3 | 6438 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6439 | kfree(rd); |
6440 | return NULL; | |
6441 | } | |
57d885fe GH |
6442 | |
6443 | return rd; | |
6444 | } | |
6445 | ||
1da177e4 | 6446 | /* |
0eab9146 | 6447 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6448 | * hold the hotplug lock. |
6449 | */ | |
0eab9146 IM |
6450 | static void |
6451 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6452 | { |
70b97a7f | 6453 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6454 | struct sched_domain *tmp; |
6455 | ||
669c55e9 PZ |
6456 | for (tmp = sd; tmp; tmp = tmp->parent) |
6457 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6458 | ||
245af2c7 | 6459 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6460 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6461 | struct sched_domain *parent = tmp->parent; |
6462 | if (!parent) | |
6463 | break; | |
f29c9b1c | 6464 | |
1a848870 | 6465 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6466 | tmp->parent = parent->parent; |
1a848870 SS |
6467 | if (parent->parent) |
6468 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6469 | } else |
6470 | tmp = tmp->parent; | |
245af2c7 SS |
6471 | } |
6472 | ||
1a848870 | 6473 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6474 | sd = sd->parent; |
1a848870 SS |
6475 | if (sd) |
6476 | sd->child = NULL; | |
6477 | } | |
1da177e4 LT |
6478 | |
6479 | sched_domain_debug(sd, cpu); | |
6480 | ||
57d885fe | 6481 | rq_attach_root(rq, rd); |
674311d5 | 6482 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6483 | } |
6484 | ||
6485 | /* cpus with isolated domains */ | |
dcc30a35 | 6486 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6487 | |
6488 | /* Setup the mask of cpus configured for isolated domains */ | |
6489 | static int __init isolated_cpu_setup(char *str) | |
6490 | { | |
bdddd296 | 6491 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6492 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6493 | return 1; |
6494 | } | |
6495 | ||
8927f494 | 6496 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6497 | |
6498 | /* | |
6711cab4 SS |
6499 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6500 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6501 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6502 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6503 | * |
6504 | * init_sched_build_groups will build a circular linked list of the groups | |
6505 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6506 | * and ->cpu_power to 0. | |
6507 | */ | |
a616058b | 6508 | static void |
96f874e2 RR |
6509 | init_sched_build_groups(const struct cpumask *span, |
6510 | const struct cpumask *cpu_map, | |
6511 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6512 | struct sched_group **sg, |
96f874e2 RR |
6513 | struct cpumask *tmpmask), |
6514 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6515 | { |
6516 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6517 | int i; |
6518 | ||
96f874e2 | 6519 | cpumask_clear(covered); |
7c16ec58 | 6520 | |
abcd083a | 6521 | for_each_cpu(i, span) { |
6711cab4 | 6522 | struct sched_group *sg; |
7c16ec58 | 6523 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6524 | int j; |
6525 | ||
758b2cdc | 6526 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6527 | continue; |
6528 | ||
758b2cdc | 6529 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6530 | sg->cpu_power = 0; |
1da177e4 | 6531 | |
abcd083a | 6532 | for_each_cpu(j, span) { |
7c16ec58 | 6533 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6534 | continue; |
6535 | ||
96f874e2 | 6536 | cpumask_set_cpu(j, covered); |
758b2cdc | 6537 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6538 | } |
6539 | if (!first) | |
6540 | first = sg; | |
6541 | if (last) | |
6542 | last->next = sg; | |
6543 | last = sg; | |
6544 | } | |
6545 | last->next = first; | |
6546 | } | |
6547 | ||
9c1cfda2 | 6548 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6549 | |
9c1cfda2 | 6550 | #ifdef CONFIG_NUMA |
198e2f18 | 6551 | |
9c1cfda2 JH |
6552 | /** |
6553 | * find_next_best_node - find the next node to include in a sched_domain | |
6554 | * @node: node whose sched_domain we're building | |
6555 | * @used_nodes: nodes already in the sched_domain | |
6556 | * | |
41a2d6cf | 6557 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6558 | * finds the closest node not already in the @used_nodes map. |
6559 | * | |
6560 | * Should use nodemask_t. | |
6561 | */ | |
c5f59f08 | 6562 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6563 | { |
6564 | int i, n, val, min_val, best_node = 0; | |
6565 | ||
6566 | min_val = INT_MAX; | |
6567 | ||
076ac2af | 6568 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6569 | /* Start at @node */ |
076ac2af | 6570 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6571 | |
6572 | if (!nr_cpus_node(n)) | |
6573 | continue; | |
6574 | ||
6575 | /* Skip already used nodes */ | |
c5f59f08 | 6576 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6577 | continue; |
6578 | ||
6579 | /* Simple min distance search */ | |
6580 | val = node_distance(node, n); | |
6581 | ||
6582 | if (val < min_val) { | |
6583 | min_val = val; | |
6584 | best_node = n; | |
6585 | } | |
6586 | } | |
6587 | ||
c5f59f08 | 6588 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6589 | return best_node; |
6590 | } | |
6591 | ||
6592 | /** | |
6593 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6594 | * @node: node whose cpumask we're constructing | |
73486722 | 6595 | * @span: resulting cpumask |
9c1cfda2 | 6596 | * |
41a2d6cf | 6597 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6598 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6599 | * out optimally. | |
6600 | */ | |
96f874e2 | 6601 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6602 | { |
c5f59f08 | 6603 | nodemask_t used_nodes; |
48f24c4d | 6604 | int i; |
9c1cfda2 | 6605 | |
6ca09dfc | 6606 | cpumask_clear(span); |
c5f59f08 | 6607 | nodes_clear(used_nodes); |
9c1cfda2 | 6608 | |
6ca09dfc | 6609 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6610 | node_set(node, used_nodes); |
9c1cfda2 JH |
6611 | |
6612 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6613 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6614 | |
6ca09dfc | 6615 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6616 | } |
9c1cfda2 | 6617 | } |
6d6bc0ad | 6618 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6619 | |
5c45bf27 | 6620 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6621 | |
6c99e9ad RR |
6622 | /* |
6623 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6624 | * |
6625 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6626 | * and struct sched_domain. ) | |
6c99e9ad RR |
6627 | */ |
6628 | struct static_sched_group { | |
6629 | struct sched_group sg; | |
6630 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6631 | }; | |
6632 | ||
6633 | struct static_sched_domain { | |
6634 | struct sched_domain sd; | |
6635 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6636 | }; | |
6637 | ||
49a02c51 AH |
6638 | struct s_data { |
6639 | #ifdef CONFIG_NUMA | |
6640 | int sd_allnodes; | |
6641 | cpumask_var_t domainspan; | |
6642 | cpumask_var_t covered; | |
6643 | cpumask_var_t notcovered; | |
6644 | #endif | |
6645 | cpumask_var_t nodemask; | |
6646 | cpumask_var_t this_sibling_map; | |
6647 | cpumask_var_t this_core_map; | |
01a08546 | 6648 | cpumask_var_t this_book_map; |
49a02c51 AH |
6649 | cpumask_var_t send_covered; |
6650 | cpumask_var_t tmpmask; | |
6651 | struct sched_group **sched_group_nodes; | |
6652 | struct root_domain *rd; | |
6653 | }; | |
6654 | ||
2109b99e AH |
6655 | enum s_alloc { |
6656 | sa_sched_groups = 0, | |
6657 | sa_rootdomain, | |
6658 | sa_tmpmask, | |
6659 | sa_send_covered, | |
01a08546 | 6660 | sa_this_book_map, |
2109b99e AH |
6661 | sa_this_core_map, |
6662 | sa_this_sibling_map, | |
6663 | sa_nodemask, | |
6664 | sa_sched_group_nodes, | |
6665 | #ifdef CONFIG_NUMA | |
6666 | sa_notcovered, | |
6667 | sa_covered, | |
6668 | sa_domainspan, | |
6669 | #endif | |
6670 | sa_none, | |
6671 | }; | |
6672 | ||
9c1cfda2 | 6673 | /* |
48f24c4d | 6674 | * SMT sched-domains: |
9c1cfda2 | 6675 | */ |
1da177e4 | 6676 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6677 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6678 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6679 | |
41a2d6cf | 6680 | static int |
96f874e2 RR |
6681 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6682 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6683 | { |
6711cab4 | 6684 | if (sg) |
1871e52c | 6685 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6686 | return cpu; |
6687 | } | |
6d6bc0ad | 6688 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6689 | |
48f24c4d IM |
6690 | /* |
6691 | * multi-core sched-domains: | |
6692 | */ | |
1e9f28fa | 6693 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6694 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6695 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6696 | |
41a2d6cf | 6697 | static int |
96f874e2 RR |
6698 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6699 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6700 | { |
6711cab4 | 6701 | int group; |
f269893c | 6702 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 6703 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6704 | group = cpumask_first(mask); |
f269893c HC |
6705 | #else |
6706 | group = cpu; | |
6707 | #endif | |
6711cab4 | 6708 | if (sg) |
6c99e9ad | 6709 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6710 | return group; |
1e9f28fa | 6711 | } |
f269893c | 6712 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 6713 | |
01a08546 HC |
6714 | /* |
6715 | * book sched-domains: | |
6716 | */ | |
6717 | #ifdef CONFIG_SCHED_BOOK | |
6718 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
6719 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
6720 | ||
41a2d6cf | 6721 | static int |
01a08546 HC |
6722 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
6723 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6724 | { |
01a08546 HC |
6725 | int group = cpu; |
6726 | #ifdef CONFIG_SCHED_MC | |
6727 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
6728 | group = cpumask_first(mask); | |
6729 | #elif defined(CONFIG_SCHED_SMT) | |
6730 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
6731 | group = cpumask_first(mask); | |
6732 | #endif | |
6711cab4 | 6733 | if (sg) |
01a08546 HC |
6734 | *sg = &per_cpu(sched_group_book, group).sg; |
6735 | return group; | |
1e9f28fa | 6736 | } |
01a08546 | 6737 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 6738 | |
6c99e9ad RR |
6739 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6740 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6741 | |
41a2d6cf | 6742 | static int |
96f874e2 RR |
6743 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6744 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6745 | { |
6711cab4 | 6746 | int group; |
01a08546 HC |
6747 | #ifdef CONFIG_SCHED_BOOK |
6748 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
6749 | group = cpumask_first(mask); | |
6750 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 6751 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6752 | group = cpumask_first(mask); |
1e9f28fa | 6753 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6754 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6755 | group = cpumask_first(mask); |
1da177e4 | 6756 | #else |
6711cab4 | 6757 | group = cpu; |
1da177e4 | 6758 | #endif |
6711cab4 | 6759 | if (sg) |
6c99e9ad | 6760 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6761 | return group; |
1da177e4 LT |
6762 | } |
6763 | ||
6764 | #ifdef CONFIG_NUMA | |
1da177e4 | 6765 | /* |
9c1cfda2 JH |
6766 | * The init_sched_build_groups can't handle what we want to do with node |
6767 | * groups, so roll our own. Now each node has its own list of groups which | |
6768 | * gets dynamically allocated. | |
1da177e4 | 6769 | */ |
62ea9ceb | 6770 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6771 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6772 | |
62ea9ceb | 6773 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6774 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6775 | |
96f874e2 RR |
6776 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6777 | struct sched_group **sg, | |
6778 | struct cpumask *nodemask) | |
9c1cfda2 | 6779 | { |
6711cab4 SS |
6780 | int group; |
6781 | ||
6ca09dfc | 6782 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6783 | group = cpumask_first(nodemask); |
6711cab4 SS |
6784 | |
6785 | if (sg) | |
6c99e9ad | 6786 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6787 | return group; |
1da177e4 | 6788 | } |
6711cab4 | 6789 | |
08069033 SS |
6790 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6791 | { | |
6792 | struct sched_group *sg = group_head; | |
6793 | int j; | |
6794 | ||
6795 | if (!sg) | |
6796 | return; | |
3a5c359a | 6797 | do { |
758b2cdc | 6798 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6799 | struct sched_domain *sd; |
08069033 | 6800 | |
6c99e9ad | 6801 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6802 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6803 | /* |
6804 | * Only add "power" once for each | |
6805 | * physical package. | |
6806 | */ | |
6807 | continue; | |
6808 | } | |
08069033 | 6809 | |
18a3885f | 6810 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6811 | } |
6812 | sg = sg->next; | |
6813 | } while (sg != group_head); | |
08069033 | 6814 | } |
0601a88d AH |
6815 | |
6816 | static int build_numa_sched_groups(struct s_data *d, | |
6817 | const struct cpumask *cpu_map, int num) | |
6818 | { | |
6819 | struct sched_domain *sd; | |
6820 | struct sched_group *sg, *prev; | |
6821 | int n, j; | |
6822 | ||
6823 | cpumask_clear(d->covered); | |
6824 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6825 | if (cpumask_empty(d->nodemask)) { | |
6826 | d->sched_group_nodes[num] = NULL; | |
6827 | goto out; | |
6828 | } | |
6829 | ||
6830 | sched_domain_node_span(num, d->domainspan); | |
6831 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6832 | ||
6833 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6834 | GFP_KERNEL, num); | |
6835 | if (!sg) { | |
3df0fc5b PZ |
6836 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6837 | num); | |
0601a88d AH |
6838 | return -ENOMEM; |
6839 | } | |
6840 | d->sched_group_nodes[num] = sg; | |
6841 | ||
6842 | for_each_cpu(j, d->nodemask) { | |
6843 | sd = &per_cpu(node_domains, j).sd; | |
6844 | sd->groups = sg; | |
6845 | } | |
6846 | ||
18a3885f | 6847 | sg->cpu_power = 0; |
0601a88d AH |
6848 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6849 | sg->next = sg; | |
6850 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6851 | ||
6852 | prev = sg; | |
6853 | for (j = 0; j < nr_node_ids; j++) { | |
6854 | n = (num + j) % nr_node_ids; | |
6855 | cpumask_complement(d->notcovered, d->covered); | |
6856 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6857 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6858 | if (cpumask_empty(d->tmpmask)) | |
6859 | break; | |
6860 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6861 | if (cpumask_empty(d->tmpmask)) | |
6862 | continue; | |
6863 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6864 | GFP_KERNEL, num); | |
6865 | if (!sg) { | |
3df0fc5b PZ |
6866 | printk(KERN_WARNING |
6867 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6868 | return -ENOMEM; |
6869 | } | |
18a3885f | 6870 | sg->cpu_power = 0; |
0601a88d AH |
6871 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6872 | sg->next = prev->next; | |
6873 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6874 | prev->next = sg; | |
6875 | prev = sg; | |
6876 | } | |
6877 | out: | |
6878 | return 0; | |
6879 | } | |
6d6bc0ad | 6880 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6881 | |
a616058b | 6882 | #ifdef CONFIG_NUMA |
51888ca2 | 6883 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6884 | static void free_sched_groups(const struct cpumask *cpu_map, |
6885 | struct cpumask *nodemask) | |
51888ca2 | 6886 | { |
a616058b | 6887 | int cpu, i; |
51888ca2 | 6888 | |
abcd083a | 6889 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6890 | struct sched_group **sched_group_nodes |
6891 | = sched_group_nodes_bycpu[cpu]; | |
6892 | ||
51888ca2 SV |
6893 | if (!sched_group_nodes) |
6894 | continue; | |
6895 | ||
076ac2af | 6896 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6897 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6898 | ||
6ca09dfc | 6899 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6900 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6901 | continue; |
6902 | ||
6903 | if (sg == NULL) | |
6904 | continue; | |
6905 | sg = sg->next; | |
6906 | next_sg: | |
6907 | oldsg = sg; | |
6908 | sg = sg->next; | |
6909 | kfree(oldsg); | |
6910 | if (oldsg != sched_group_nodes[i]) | |
6911 | goto next_sg; | |
6912 | } | |
6913 | kfree(sched_group_nodes); | |
6914 | sched_group_nodes_bycpu[cpu] = NULL; | |
6915 | } | |
51888ca2 | 6916 | } |
6d6bc0ad | 6917 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6918 | static void free_sched_groups(const struct cpumask *cpu_map, |
6919 | struct cpumask *nodemask) | |
a616058b SS |
6920 | { |
6921 | } | |
6d6bc0ad | 6922 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6923 | |
89c4710e SS |
6924 | /* |
6925 | * Initialize sched groups cpu_power. | |
6926 | * | |
6927 | * cpu_power indicates the capacity of sched group, which is used while | |
6928 | * distributing the load between different sched groups in a sched domain. | |
6929 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6930 | * there are asymmetries in the topology. If there are asymmetries, group | |
6931 | * having more cpu_power will pickup more load compared to the group having | |
6932 | * less cpu_power. | |
89c4710e SS |
6933 | */ |
6934 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6935 | { | |
6936 | struct sched_domain *child; | |
6937 | struct sched_group *group; | |
f93e65c1 PZ |
6938 | long power; |
6939 | int weight; | |
89c4710e SS |
6940 | |
6941 | WARN_ON(!sd || !sd->groups); | |
6942 | ||
13318a71 | 6943 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6944 | return; |
6945 | ||
aae6d3dd SS |
6946 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
6947 | ||
89c4710e SS |
6948 | child = sd->child; |
6949 | ||
18a3885f | 6950 | sd->groups->cpu_power = 0; |
5517d86b | 6951 | |
f93e65c1 PZ |
6952 | if (!child) { |
6953 | power = SCHED_LOAD_SCALE; | |
6954 | weight = cpumask_weight(sched_domain_span(sd)); | |
6955 | /* | |
6956 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6957 | * Usually multiple threads get a better yield out of |
6958 | * that one core than a single thread would have, | |
6959 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6960 | */ |
a52bfd73 PZ |
6961 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6962 | power *= sd->smt_gain; | |
f93e65c1 | 6963 | power /= weight; |
a52bfd73 PZ |
6964 | power >>= SCHED_LOAD_SHIFT; |
6965 | } | |
18a3885f | 6966 | sd->groups->cpu_power += power; |
89c4710e SS |
6967 | return; |
6968 | } | |
6969 | ||
89c4710e | 6970 | /* |
f93e65c1 | 6971 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6972 | */ |
6973 | group = child->groups; | |
6974 | do { | |
18a3885f | 6975 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6976 | group = group->next; |
6977 | } while (group != child->groups); | |
6978 | } | |
6979 | ||
7c16ec58 MT |
6980 | /* |
6981 | * Initializers for schedule domains | |
6982 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6983 | */ | |
6984 | ||
a5d8c348 IM |
6985 | #ifdef CONFIG_SCHED_DEBUG |
6986 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6987 | #else | |
6988 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6989 | #endif | |
6990 | ||
7c16ec58 | 6991 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6992 | |
7c16ec58 MT |
6993 | #define SD_INIT_FUNC(type) \ |
6994 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6995 | { \ | |
6996 | memset(sd, 0, sizeof(*sd)); \ | |
6997 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6998 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6999 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
7000 | } |
7001 | ||
7002 | SD_INIT_FUNC(CPU) | |
7003 | #ifdef CONFIG_NUMA | |
7004 | SD_INIT_FUNC(ALLNODES) | |
7005 | SD_INIT_FUNC(NODE) | |
7006 | #endif | |
7007 | #ifdef CONFIG_SCHED_SMT | |
7008 | SD_INIT_FUNC(SIBLING) | |
7009 | #endif | |
7010 | #ifdef CONFIG_SCHED_MC | |
7011 | SD_INIT_FUNC(MC) | |
7012 | #endif | |
01a08546 HC |
7013 | #ifdef CONFIG_SCHED_BOOK |
7014 | SD_INIT_FUNC(BOOK) | |
7015 | #endif | |
7c16ec58 | 7016 | |
1d3504fc HS |
7017 | static int default_relax_domain_level = -1; |
7018 | ||
7019 | static int __init setup_relax_domain_level(char *str) | |
7020 | { | |
30e0e178 LZ |
7021 | unsigned long val; |
7022 | ||
7023 | val = simple_strtoul(str, NULL, 0); | |
7024 | if (val < SD_LV_MAX) | |
7025 | default_relax_domain_level = val; | |
7026 | ||
1d3504fc HS |
7027 | return 1; |
7028 | } | |
7029 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7030 | ||
7031 | static void set_domain_attribute(struct sched_domain *sd, | |
7032 | struct sched_domain_attr *attr) | |
7033 | { | |
7034 | int request; | |
7035 | ||
7036 | if (!attr || attr->relax_domain_level < 0) { | |
7037 | if (default_relax_domain_level < 0) | |
7038 | return; | |
7039 | else | |
7040 | request = default_relax_domain_level; | |
7041 | } else | |
7042 | request = attr->relax_domain_level; | |
7043 | if (request < sd->level) { | |
7044 | /* turn off idle balance on this domain */ | |
c88d5910 | 7045 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7046 | } else { |
7047 | /* turn on idle balance on this domain */ | |
c88d5910 | 7048 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7049 | } |
7050 | } | |
7051 | ||
2109b99e AH |
7052 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7053 | const struct cpumask *cpu_map) | |
7054 | { | |
7055 | switch (what) { | |
7056 | case sa_sched_groups: | |
7057 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
7058 | d->sched_group_nodes = NULL; | |
7059 | case sa_rootdomain: | |
7060 | free_rootdomain(d->rd); /* fall through */ | |
7061 | case sa_tmpmask: | |
7062 | free_cpumask_var(d->tmpmask); /* fall through */ | |
7063 | case sa_send_covered: | |
7064 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
7065 | case sa_this_book_map: |
7066 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
7067 | case sa_this_core_map: |
7068 | free_cpumask_var(d->this_core_map); /* fall through */ | |
7069 | case sa_this_sibling_map: | |
7070 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
7071 | case sa_nodemask: | |
7072 | free_cpumask_var(d->nodemask); /* fall through */ | |
7073 | case sa_sched_group_nodes: | |
d1b55138 | 7074 | #ifdef CONFIG_NUMA |
2109b99e AH |
7075 | kfree(d->sched_group_nodes); /* fall through */ |
7076 | case sa_notcovered: | |
7077 | free_cpumask_var(d->notcovered); /* fall through */ | |
7078 | case sa_covered: | |
7079 | free_cpumask_var(d->covered); /* fall through */ | |
7080 | case sa_domainspan: | |
7081 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 7082 | #endif |
2109b99e AH |
7083 | case sa_none: |
7084 | break; | |
7085 | } | |
7086 | } | |
3404c8d9 | 7087 | |
2109b99e AH |
7088 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7089 | const struct cpumask *cpu_map) | |
7090 | { | |
3404c8d9 | 7091 | #ifdef CONFIG_NUMA |
2109b99e AH |
7092 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
7093 | return sa_none; | |
7094 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
7095 | return sa_domainspan; | |
7096 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
7097 | return sa_covered; | |
7098 | /* Allocate the per-node list of sched groups */ | |
7099 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
7100 | sizeof(struct sched_group *), GFP_KERNEL); | |
7101 | if (!d->sched_group_nodes) { | |
3df0fc5b | 7102 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 7103 | return sa_notcovered; |
d1b55138 | 7104 | } |
2109b99e | 7105 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 7106 | #endif |
2109b99e AH |
7107 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
7108 | return sa_sched_group_nodes; | |
7109 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
7110 | return sa_nodemask; | |
7111 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
7112 | return sa_this_sibling_map; | |
01a08546 | 7113 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 7114 | return sa_this_core_map; |
01a08546 HC |
7115 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
7116 | return sa_this_book_map; | |
2109b99e AH |
7117 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
7118 | return sa_send_covered; | |
7119 | d->rd = alloc_rootdomain(); | |
7120 | if (!d->rd) { | |
3df0fc5b | 7121 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 7122 | return sa_tmpmask; |
57d885fe | 7123 | } |
2109b99e AH |
7124 | return sa_rootdomain; |
7125 | } | |
57d885fe | 7126 | |
7f4588f3 AH |
7127 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
7128 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
7129 | { | |
7130 | struct sched_domain *sd = NULL; | |
7c16ec58 | 7131 | #ifdef CONFIG_NUMA |
7f4588f3 | 7132 | struct sched_domain *parent; |
1da177e4 | 7133 | |
7f4588f3 AH |
7134 | d->sd_allnodes = 0; |
7135 | if (cpumask_weight(cpu_map) > | |
7136 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
7137 | sd = &per_cpu(allnodes_domains, i).sd; | |
7138 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 7139 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
7140 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7141 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7142 | d->sd_allnodes = 1; | |
7143 | } | |
7144 | parent = sd; | |
7145 | ||
7146 | sd = &per_cpu(node_domains, i).sd; | |
7147 | SD_INIT(sd, NODE); | |
7148 | set_domain_attribute(sd, attr); | |
7149 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
7150 | sd->parent = parent; | |
7151 | if (parent) | |
7152 | parent->child = sd; | |
7153 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 7154 | #endif |
7f4588f3 AH |
7155 | return sd; |
7156 | } | |
1da177e4 | 7157 | |
87cce662 AH |
7158 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7159 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7160 | struct sched_domain *parent, int i) | |
7161 | { | |
7162 | struct sched_domain *sd; | |
7163 | sd = &per_cpu(phys_domains, i).sd; | |
7164 | SD_INIT(sd, CPU); | |
7165 | set_domain_attribute(sd, attr); | |
7166 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
7167 | sd->parent = parent; | |
7168 | if (parent) | |
7169 | parent->child = sd; | |
7170 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7171 | return sd; | |
7172 | } | |
1da177e4 | 7173 | |
01a08546 HC |
7174 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
7175 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7176 | struct sched_domain *parent, int i) | |
7177 | { | |
7178 | struct sched_domain *sd = parent; | |
7179 | #ifdef CONFIG_SCHED_BOOK | |
7180 | sd = &per_cpu(book_domains, i).sd; | |
7181 | SD_INIT(sd, BOOK); | |
7182 | set_domain_attribute(sd, attr); | |
7183 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
7184 | sd->parent = parent; | |
7185 | parent->child = sd; | |
7186 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7187 | #endif | |
7188 | return sd; | |
7189 | } | |
7190 | ||
410c4081 AH |
7191 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7192 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7193 | struct sched_domain *parent, int i) | |
7194 | { | |
7195 | struct sched_domain *sd = parent; | |
1e9f28fa | 7196 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7197 | sd = &per_cpu(core_domains, i).sd; |
7198 | SD_INIT(sd, MC); | |
7199 | set_domain_attribute(sd, attr); | |
7200 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7201 | sd->parent = parent; | |
7202 | parent->child = sd; | |
7203 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7204 | #endif |
410c4081 AH |
7205 | return sd; |
7206 | } | |
1e9f28fa | 7207 | |
d8173535 AH |
7208 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7209 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7210 | struct sched_domain *parent, int i) | |
7211 | { | |
7212 | struct sched_domain *sd = parent; | |
1da177e4 | 7213 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7214 | sd = &per_cpu(cpu_domains, i).sd; |
7215 | SD_INIT(sd, SIBLING); | |
7216 | set_domain_attribute(sd, attr); | |
7217 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7218 | sd->parent = parent; | |
7219 | parent->child = sd; | |
7220 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7221 | #endif |
d8173535 AH |
7222 | return sd; |
7223 | } | |
1da177e4 | 7224 | |
0e8e85c9 AH |
7225 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7226 | const struct cpumask *cpu_map, int cpu) | |
7227 | { | |
7228 | switch (l) { | |
1da177e4 | 7229 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7230 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7231 | cpumask_and(d->this_sibling_map, cpu_map, | |
7232 | topology_thread_cpumask(cpu)); | |
7233 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7234 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7235 | &cpu_to_cpu_group, | |
7236 | d->send_covered, d->tmpmask); | |
7237 | break; | |
1da177e4 | 7238 | #endif |
1e9f28fa | 7239 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7240 | case SD_LV_MC: /* set up multi-core groups */ |
7241 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7242 | if (cpu == cpumask_first(d->this_core_map)) | |
7243 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7244 | &cpu_to_core_group, | |
7245 | d->send_covered, d->tmpmask); | |
7246 | break; | |
01a08546 HC |
7247 | #endif |
7248 | #ifdef CONFIG_SCHED_BOOK | |
7249 | case SD_LV_BOOK: /* set up book groups */ | |
7250 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7251 | if (cpu == cpumask_first(d->this_book_map)) | |
7252 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7253 | &cpu_to_book_group, | |
7254 | d->send_covered, d->tmpmask); | |
7255 | break; | |
1e9f28fa | 7256 | #endif |
86548096 AH |
7257 | case SD_LV_CPU: /* set up physical groups */ |
7258 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7259 | if (!cpumask_empty(d->nodemask)) | |
7260 | init_sched_build_groups(d->nodemask, cpu_map, | |
7261 | &cpu_to_phys_group, | |
7262 | d->send_covered, d->tmpmask); | |
7263 | break; | |
1da177e4 | 7264 | #ifdef CONFIG_NUMA |
de616e36 AH |
7265 | case SD_LV_ALLNODES: |
7266 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7267 | d->send_covered, d->tmpmask); | |
7268 | break; | |
7269 | #endif | |
0e8e85c9 AH |
7270 | default: |
7271 | break; | |
7c16ec58 | 7272 | } |
0e8e85c9 | 7273 | } |
9c1cfda2 | 7274 | |
2109b99e AH |
7275 | /* |
7276 | * Build sched domains for a given set of cpus and attach the sched domains | |
7277 | * to the individual cpus | |
7278 | */ | |
7279 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7280 | struct sched_domain_attr *attr) | |
7281 | { | |
7282 | enum s_alloc alloc_state = sa_none; | |
7283 | struct s_data d; | |
294b0c96 | 7284 | struct sched_domain *sd; |
2109b99e | 7285 | int i; |
7c16ec58 | 7286 | #ifdef CONFIG_NUMA |
2109b99e | 7287 | d.sd_allnodes = 0; |
7c16ec58 | 7288 | #endif |
9c1cfda2 | 7289 | |
2109b99e AH |
7290 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7291 | if (alloc_state != sa_rootdomain) | |
7292 | goto error; | |
7293 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7294 | |
1da177e4 | 7295 | /* |
1a20ff27 | 7296 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7297 | */ |
abcd083a | 7298 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7299 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7300 | cpu_map); | |
9761eea8 | 7301 | |
7f4588f3 | 7302 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7303 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7304 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7305 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7306 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7307 | } |
9c1cfda2 | 7308 | |
abcd083a | 7309 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7310 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7311 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7312 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7313 | } |
9c1cfda2 | 7314 | |
1da177e4 | 7315 | /* Set up physical groups */ |
86548096 AH |
7316 | for (i = 0; i < nr_node_ids; i++) |
7317 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7318 | |
1da177e4 LT |
7319 | #ifdef CONFIG_NUMA |
7320 | /* Set up node groups */ | |
de616e36 AH |
7321 | if (d.sd_allnodes) |
7322 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7323 | |
0601a88d AH |
7324 | for (i = 0; i < nr_node_ids; i++) |
7325 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7326 | goto error; |
1da177e4 LT |
7327 | #endif |
7328 | ||
7329 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7330 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7331 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7332 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7333 | init_sched_groups_power(i, sd); |
5c45bf27 | 7334 | } |
1da177e4 | 7335 | #endif |
1e9f28fa | 7336 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7337 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7338 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7339 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7340 | } |
7341 | #endif | |
01a08546 HC |
7342 | #ifdef CONFIG_SCHED_BOOK |
7343 | for_each_cpu(i, cpu_map) { | |
7344 | sd = &per_cpu(book_domains, i).sd; | |
7345 | init_sched_groups_power(i, sd); | |
7346 | } | |
7347 | #endif | |
1e9f28fa | 7348 | |
abcd083a | 7349 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7350 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7351 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7352 | } |
7353 | ||
9c1cfda2 | 7354 | #ifdef CONFIG_NUMA |
076ac2af | 7355 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7356 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7357 | |
49a02c51 | 7358 | if (d.sd_allnodes) { |
6711cab4 | 7359 | struct sched_group *sg; |
f712c0c7 | 7360 | |
96f874e2 | 7361 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7362 | d.tmpmask); |
f712c0c7 SS |
7363 | init_numa_sched_groups_power(sg); |
7364 | } | |
9c1cfda2 JH |
7365 | #endif |
7366 | ||
1da177e4 | 7367 | /* Attach the domains */ |
abcd083a | 7368 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7369 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7370 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7371 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7372 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7373 | #elif defined(CONFIG_SCHED_BOOK) |
7374 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7375 | #else |
6c99e9ad | 7376 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7377 | #endif |
49a02c51 | 7378 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7379 | } |
51888ca2 | 7380 | |
2109b99e AH |
7381 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7382 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7383 | return 0; | |
51888ca2 | 7384 | |
51888ca2 | 7385 | error: |
2109b99e AH |
7386 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7387 | return -ENOMEM; | |
1da177e4 | 7388 | } |
029190c5 | 7389 | |
96f874e2 | 7390 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7391 | { |
7392 | return __build_sched_domains(cpu_map, NULL); | |
7393 | } | |
7394 | ||
acc3f5d7 | 7395 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7396 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7397 | static struct sched_domain_attr *dattr_cur; |
7398 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7399 | |
7400 | /* | |
7401 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7402 | * cpumask) fails, then fallback to a single sched domain, |
7403 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7404 | */ |
4212823f | 7405 | static cpumask_var_t fallback_doms; |
029190c5 | 7406 | |
ee79d1bd HC |
7407 | /* |
7408 | * arch_update_cpu_topology lets virtualized architectures update the | |
7409 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7410 | * or 0 if it stayed the same. | |
7411 | */ | |
7412 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7413 | { |
ee79d1bd | 7414 | return 0; |
22e52b07 HC |
7415 | } |
7416 | ||
acc3f5d7 RR |
7417 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7418 | { | |
7419 | int i; | |
7420 | cpumask_var_t *doms; | |
7421 | ||
7422 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7423 | if (!doms) | |
7424 | return NULL; | |
7425 | for (i = 0; i < ndoms; i++) { | |
7426 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7427 | free_sched_domains(doms, i); | |
7428 | return NULL; | |
7429 | } | |
7430 | } | |
7431 | return doms; | |
7432 | } | |
7433 | ||
7434 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7435 | { | |
7436 | unsigned int i; | |
7437 | for (i = 0; i < ndoms; i++) | |
7438 | free_cpumask_var(doms[i]); | |
7439 | kfree(doms); | |
7440 | } | |
7441 | ||
1a20ff27 | 7442 | /* |
41a2d6cf | 7443 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7444 | * For now this just excludes isolated cpus, but could be used to |
7445 | * exclude other special cases in the future. | |
1a20ff27 | 7446 | */ |
96f874e2 | 7447 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7448 | { |
7378547f MM |
7449 | int err; |
7450 | ||
22e52b07 | 7451 | arch_update_cpu_topology(); |
029190c5 | 7452 | ndoms_cur = 1; |
acc3f5d7 | 7453 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7454 | if (!doms_cur) |
acc3f5d7 RR |
7455 | doms_cur = &fallback_doms; |
7456 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7457 | dattr_cur = NULL; |
acc3f5d7 | 7458 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7459 | register_sched_domain_sysctl(); |
7378547f MM |
7460 | |
7461 | return err; | |
1a20ff27 DG |
7462 | } |
7463 | ||
96f874e2 RR |
7464 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7465 | struct cpumask *tmpmask) | |
1da177e4 | 7466 | { |
7c16ec58 | 7467 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7468 | } |
1da177e4 | 7469 | |
1a20ff27 DG |
7470 | /* |
7471 | * Detach sched domains from a group of cpus specified in cpu_map | |
7472 | * These cpus will now be attached to the NULL domain | |
7473 | */ | |
96f874e2 | 7474 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7475 | { |
96f874e2 RR |
7476 | /* Save because hotplug lock held. */ |
7477 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7478 | int i; |
7479 | ||
abcd083a | 7480 | for_each_cpu(i, cpu_map) |
57d885fe | 7481 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7482 | synchronize_sched(); |
96f874e2 | 7483 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7484 | } |
7485 | ||
1d3504fc HS |
7486 | /* handle null as "default" */ |
7487 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7488 | struct sched_domain_attr *new, int idx_new) | |
7489 | { | |
7490 | struct sched_domain_attr tmp; | |
7491 | ||
7492 | /* fast path */ | |
7493 | if (!new && !cur) | |
7494 | return 1; | |
7495 | ||
7496 | tmp = SD_ATTR_INIT; | |
7497 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7498 | new ? (new + idx_new) : &tmp, | |
7499 | sizeof(struct sched_domain_attr)); | |
7500 | } | |
7501 | ||
029190c5 PJ |
7502 | /* |
7503 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7504 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7505 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7506 | * It destroys each deleted domain and builds each new domain. | |
7507 | * | |
acc3f5d7 | 7508 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7509 | * The masks don't intersect (don't overlap.) We should setup one |
7510 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7511 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7512 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7513 | * it as it is. | |
7514 | * | |
acc3f5d7 RR |
7515 | * The passed in 'doms_new' should be allocated using |
7516 | * alloc_sched_domains. This routine takes ownership of it and will | |
7517 | * free_sched_domains it when done with it. If the caller failed the | |
7518 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7519 | * and partition_sched_domains() will fallback to the single partition | |
7520 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7521 | * |
96f874e2 | 7522 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7523 | * ndoms_new == 0 is a special case for destroying existing domains, |
7524 | * and it will not create the default domain. | |
dfb512ec | 7525 | * |
029190c5 PJ |
7526 | * Call with hotplug lock held |
7527 | */ | |
acc3f5d7 | 7528 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7529 | struct sched_domain_attr *dattr_new) |
029190c5 | 7530 | { |
dfb512ec | 7531 | int i, j, n; |
d65bd5ec | 7532 | int new_topology; |
029190c5 | 7533 | |
712555ee | 7534 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7535 | |
7378547f MM |
7536 | /* always unregister in case we don't destroy any domains */ |
7537 | unregister_sched_domain_sysctl(); | |
7538 | ||
d65bd5ec HC |
7539 | /* Let architecture update cpu core mappings. */ |
7540 | new_topology = arch_update_cpu_topology(); | |
7541 | ||
dfb512ec | 7542 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7543 | |
7544 | /* Destroy deleted domains */ | |
7545 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7546 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7547 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7548 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7549 | goto match1; |
7550 | } | |
7551 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7552 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7553 | match1: |
7554 | ; | |
7555 | } | |
7556 | ||
e761b772 MK |
7557 | if (doms_new == NULL) { |
7558 | ndoms_cur = 0; | |
acc3f5d7 | 7559 | doms_new = &fallback_doms; |
6ad4c188 | 7560 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7561 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7562 | } |
7563 | ||
029190c5 PJ |
7564 | /* Build new domains */ |
7565 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7566 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7567 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7568 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7569 | goto match2; |
7570 | } | |
7571 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7572 | __build_sched_domains(doms_new[i], |
1d3504fc | 7573 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7574 | match2: |
7575 | ; | |
7576 | } | |
7577 | ||
7578 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7579 | if (doms_cur != &fallback_doms) |
7580 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7581 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7582 | doms_cur = doms_new; |
1d3504fc | 7583 | dattr_cur = dattr_new; |
029190c5 | 7584 | ndoms_cur = ndoms_new; |
7378547f MM |
7585 | |
7586 | register_sched_domain_sysctl(); | |
a1835615 | 7587 | |
712555ee | 7588 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7589 | } |
7590 | ||
5c45bf27 | 7591 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7592 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7593 | { |
95402b38 | 7594 | get_online_cpus(); |
dfb512ec MK |
7595 | |
7596 | /* Destroy domains first to force the rebuild */ | |
7597 | partition_sched_domains(0, NULL, NULL); | |
7598 | ||
e761b772 | 7599 | rebuild_sched_domains(); |
95402b38 | 7600 | put_online_cpus(); |
5c45bf27 SS |
7601 | } |
7602 | ||
7603 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7604 | { | |
afb8a9b7 | 7605 | unsigned int level = 0; |
5c45bf27 | 7606 | |
afb8a9b7 GS |
7607 | if (sscanf(buf, "%u", &level) != 1) |
7608 | return -EINVAL; | |
7609 | ||
7610 | /* | |
7611 | * level is always be positive so don't check for | |
7612 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7613 | * What happens on 0 or 1 byte write, | |
7614 | * need to check for count as well? | |
7615 | */ | |
7616 | ||
7617 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7618 | return -EINVAL; |
7619 | ||
7620 | if (smt) | |
afb8a9b7 | 7621 | sched_smt_power_savings = level; |
5c45bf27 | 7622 | else |
afb8a9b7 | 7623 | sched_mc_power_savings = level; |
5c45bf27 | 7624 | |
c70f22d2 | 7625 | arch_reinit_sched_domains(); |
5c45bf27 | 7626 | |
c70f22d2 | 7627 | return count; |
5c45bf27 SS |
7628 | } |
7629 | ||
5c45bf27 | 7630 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7631 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7632 | struct sysdev_class_attribute *attr, |
f718cd4a | 7633 | char *page) |
5c45bf27 SS |
7634 | { |
7635 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7636 | } | |
f718cd4a | 7637 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7638 | struct sysdev_class_attribute *attr, |
48f24c4d | 7639 | const char *buf, size_t count) |
5c45bf27 SS |
7640 | { |
7641 | return sched_power_savings_store(buf, count, 0); | |
7642 | } | |
f718cd4a AK |
7643 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7644 | sched_mc_power_savings_show, | |
7645 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7646 | #endif |
7647 | ||
7648 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7649 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7650 | struct sysdev_class_attribute *attr, |
f718cd4a | 7651 | char *page) |
5c45bf27 SS |
7652 | { |
7653 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7654 | } | |
f718cd4a | 7655 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
c9be0a36 | 7656 | struct sysdev_class_attribute *attr, |
48f24c4d | 7657 | const char *buf, size_t count) |
5c45bf27 SS |
7658 | { |
7659 | return sched_power_savings_store(buf, count, 1); | |
7660 | } | |
f718cd4a AK |
7661 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7662 | sched_smt_power_savings_show, | |
6707de00 AB |
7663 | sched_smt_power_savings_store); |
7664 | #endif | |
7665 | ||
39aac648 | 7666 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7667 | { |
7668 | int err = 0; | |
7669 | ||
7670 | #ifdef CONFIG_SCHED_SMT | |
7671 | if (smt_capable()) | |
7672 | err = sysfs_create_file(&cls->kset.kobj, | |
7673 | &attr_sched_smt_power_savings.attr); | |
7674 | #endif | |
7675 | #ifdef CONFIG_SCHED_MC | |
7676 | if (!err && mc_capable()) | |
7677 | err = sysfs_create_file(&cls->kset.kobj, | |
7678 | &attr_sched_mc_power_savings.attr); | |
7679 | #endif | |
7680 | return err; | |
7681 | } | |
6d6bc0ad | 7682 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7683 | |
1da177e4 | 7684 | /* |
3a101d05 TH |
7685 | * Update cpusets according to cpu_active mask. If cpusets are |
7686 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7687 | * around partition_sched_domains(). | |
1da177e4 | 7688 | */ |
0b2e918a TH |
7689 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7690 | void *hcpu) | |
e761b772 | 7691 | { |
3a101d05 | 7692 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7693 | case CPU_ONLINE: |
6ad4c188 | 7694 | case CPU_DOWN_FAILED: |
3a101d05 | 7695 | cpuset_update_active_cpus(); |
e761b772 | 7696 | return NOTIFY_OK; |
3a101d05 TH |
7697 | default: |
7698 | return NOTIFY_DONE; | |
7699 | } | |
7700 | } | |
e761b772 | 7701 | |
0b2e918a TH |
7702 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7703 | void *hcpu) | |
3a101d05 TH |
7704 | { |
7705 | switch (action & ~CPU_TASKS_FROZEN) { | |
7706 | case CPU_DOWN_PREPARE: | |
7707 | cpuset_update_active_cpus(); | |
7708 | return NOTIFY_OK; | |
e761b772 MK |
7709 | default: |
7710 | return NOTIFY_DONE; | |
7711 | } | |
7712 | } | |
e761b772 MK |
7713 | |
7714 | static int update_runtime(struct notifier_block *nfb, | |
7715 | unsigned long action, void *hcpu) | |
1da177e4 | 7716 | { |
7def2be1 PZ |
7717 | int cpu = (int)(long)hcpu; |
7718 | ||
1da177e4 | 7719 | switch (action) { |
1da177e4 | 7720 | case CPU_DOWN_PREPARE: |
8bb78442 | 7721 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7722 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7723 | return NOTIFY_OK; |
7724 | ||
1da177e4 | 7725 | case CPU_DOWN_FAILED: |
8bb78442 | 7726 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7727 | case CPU_ONLINE: |
8bb78442 | 7728 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7729 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7730 | return NOTIFY_OK; |
7731 | ||
1da177e4 LT |
7732 | default: |
7733 | return NOTIFY_DONE; | |
7734 | } | |
1da177e4 | 7735 | } |
1da177e4 LT |
7736 | |
7737 | void __init sched_init_smp(void) | |
7738 | { | |
dcc30a35 RR |
7739 | cpumask_var_t non_isolated_cpus; |
7740 | ||
7741 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7742 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7743 | |
434d53b0 MT |
7744 | #if defined(CONFIG_NUMA) |
7745 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7746 | GFP_KERNEL); | |
7747 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7748 | #endif | |
95402b38 | 7749 | get_online_cpus(); |
712555ee | 7750 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7751 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7752 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7753 | if (cpumask_empty(non_isolated_cpus)) | |
7754 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7755 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7756 | put_online_cpus(); |
e761b772 | 7757 | |
3a101d05 TH |
7758 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7759 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7760 | |
7761 | /* RT runtime code needs to handle some hotplug events */ | |
7762 | hotcpu_notifier(update_runtime, 0); | |
7763 | ||
b328ca18 | 7764 | init_hrtick(); |
5c1e1767 NP |
7765 | |
7766 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7767 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7768 | BUG(); |
19978ca6 | 7769 | sched_init_granularity(); |
dcc30a35 | 7770 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7771 | |
0e3900e6 | 7772 | init_sched_rt_class(); |
1da177e4 LT |
7773 | } |
7774 | #else | |
7775 | void __init sched_init_smp(void) | |
7776 | { | |
19978ca6 | 7777 | sched_init_granularity(); |
1da177e4 LT |
7778 | } |
7779 | #endif /* CONFIG_SMP */ | |
7780 | ||
cd1bb94b AB |
7781 | const_debug unsigned int sysctl_timer_migration = 1; |
7782 | ||
1da177e4 LT |
7783 | int in_sched_functions(unsigned long addr) |
7784 | { | |
1da177e4 LT |
7785 | return in_lock_functions(addr) || |
7786 | (addr >= (unsigned long)__sched_text_start | |
7787 | && addr < (unsigned long)__sched_text_end); | |
7788 | } | |
7789 | ||
a9957449 | 7790 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7791 | { |
7792 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7793 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7794 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7795 | cfs_rq->rq = rq; | |
7796 | #endif | |
67e9fb2a | 7797 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7798 | } |
7799 | ||
fa85ae24 PZ |
7800 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7801 | { | |
7802 | struct rt_prio_array *array; | |
7803 | int i; | |
7804 | ||
7805 | array = &rt_rq->active; | |
7806 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7807 | INIT_LIST_HEAD(array->queue + i); | |
7808 | __clear_bit(i, array->bitmap); | |
7809 | } | |
7810 | /* delimiter for bitsearch: */ | |
7811 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7812 | ||
052f1dc7 | 7813 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7814 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7815 | #ifdef CONFIG_SMP |
e864c499 | 7816 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7817 | #endif |
48d5e258 | 7818 | #endif |
fa85ae24 PZ |
7819 | #ifdef CONFIG_SMP |
7820 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7821 | rt_rq->overloaded = 0; |
05fa785c | 7822 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7823 | #endif |
7824 | ||
7825 | rt_rq->rt_time = 0; | |
7826 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7827 | rt_rq->rt_runtime = 0; |
0986b11b | 7828 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7829 | |
052f1dc7 | 7830 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7831 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7832 | rt_rq->rq = rq; |
7833 | #endif | |
fa85ae24 PZ |
7834 | } |
7835 | ||
6f505b16 | 7836 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 7837 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 7838 | struct sched_entity *se, int cpu, |
ec7dc8ac | 7839 | struct sched_entity *parent) |
6f505b16 | 7840 | { |
ec7dc8ac | 7841 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7842 | tg->cfs_rq[cpu] = cfs_rq; |
7843 | init_cfs_rq(cfs_rq, rq); | |
7844 | cfs_rq->tg = tg; | |
6f505b16 PZ |
7845 | |
7846 | tg->se[cpu] = se; | |
07e06b01 | 7847 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
7848 | if (!se) |
7849 | return; | |
7850 | ||
ec7dc8ac DG |
7851 | if (!parent) |
7852 | se->cfs_rq = &rq->cfs; | |
7853 | else | |
7854 | se->cfs_rq = parent->my_q; | |
7855 | ||
6f505b16 | 7856 | se->my_q = cfs_rq; |
9437178f | 7857 | update_load_set(&se->load, 0); |
ec7dc8ac | 7858 | se->parent = parent; |
6f505b16 | 7859 | } |
052f1dc7 | 7860 | #endif |
6f505b16 | 7861 | |
052f1dc7 | 7862 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 7863 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 7864 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 7865 | struct sched_rt_entity *parent) |
6f505b16 | 7866 | { |
ec7dc8ac DG |
7867 | struct rq *rq = cpu_rq(cpu); |
7868 | ||
6f505b16 PZ |
7869 | tg->rt_rq[cpu] = rt_rq; |
7870 | init_rt_rq(rt_rq, rq); | |
7871 | rt_rq->tg = tg; | |
ac086bc2 | 7872 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7873 | |
7874 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7875 | if (!rt_se) |
7876 | return; | |
7877 | ||
ec7dc8ac DG |
7878 | if (!parent) |
7879 | rt_se->rt_rq = &rq->rt; | |
7880 | else | |
7881 | rt_se->rt_rq = parent->my_q; | |
7882 | ||
6f505b16 | 7883 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7884 | rt_se->parent = parent; |
6f505b16 PZ |
7885 | INIT_LIST_HEAD(&rt_se->run_list); |
7886 | } | |
7887 | #endif | |
7888 | ||
1da177e4 LT |
7889 | void __init sched_init(void) |
7890 | { | |
dd41f596 | 7891 | int i, j; |
434d53b0 MT |
7892 | unsigned long alloc_size = 0, ptr; |
7893 | ||
7894 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7895 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7896 | #endif | |
7897 | #ifdef CONFIG_RT_GROUP_SCHED | |
7898 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7899 | #endif |
df7c8e84 | 7900 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7901 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7902 | #endif |
434d53b0 | 7903 | if (alloc_size) { |
36b7b6d4 | 7904 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7905 | |
7906 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7907 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7908 | ptr += nr_cpu_ids * sizeof(void **); |
7909 | ||
07e06b01 | 7910 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7911 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7912 | |
6d6bc0ad | 7913 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7914 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7915 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7916 | ptr += nr_cpu_ids * sizeof(void **); |
7917 | ||
07e06b01 | 7918 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7919 | ptr += nr_cpu_ids * sizeof(void **); |
7920 | ||
6d6bc0ad | 7921 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7922 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7923 | for_each_possible_cpu(i) { | |
7924 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7925 | ptr += cpumask_size(); | |
7926 | } | |
7927 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7928 | } |
dd41f596 | 7929 | |
57d885fe GH |
7930 | #ifdef CONFIG_SMP |
7931 | init_defrootdomain(); | |
7932 | #endif | |
7933 | ||
d0b27fa7 PZ |
7934 | init_rt_bandwidth(&def_rt_bandwidth, |
7935 | global_rt_period(), global_rt_runtime()); | |
7936 | ||
7937 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 7938 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7939 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7940 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7941 | |
7c941438 | 7942 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7943 | list_add(&root_task_group.list, &task_groups); |
7944 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 7945 | autogroup_init(&init_task); |
7c941438 | 7946 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7947 | |
0a945022 | 7948 | for_each_possible_cpu(i) { |
70b97a7f | 7949 | struct rq *rq; |
1da177e4 LT |
7950 | |
7951 | rq = cpu_rq(i); | |
05fa785c | 7952 | raw_spin_lock_init(&rq->lock); |
7897986b | 7953 | rq->nr_running = 0; |
dce48a84 TG |
7954 | rq->calc_load_active = 0; |
7955 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7956 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7957 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7958 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 7959 | root_task_group.shares = root_task_group_load; |
6f505b16 | 7960 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7961 | /* |
07e06b01 | 7962 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7963 | * |
7964 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7965 | * gets 100% of the cpu resources in the system. This overall | |
7966 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7967 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7968 | * based on each entity's (task or task-group's) weight |
7969 | * (se->load.weight). | |
7970 | * | |
07e06b01 | 7971 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7972 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7973 | * then A0's share of the cpu resource is: | |
7974 | * | |
0d905bca | 7975 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7976 | * |
07e06b01 YZ |
7977 | * We achieve this by letting root_task_group's tasks sit |
7978 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7979 | */ |
07e06b01 | 7980 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7981 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7982 | ||
7983 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7984 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7985 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 7986 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7987 | #endif |
1da177e4 | 7988 | |
dd41f596 IM |
7989 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7990 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7991 | |
7992 | rq->last_load_update_tick = jiffies; | |
7993 | ||
1da177e4 | 7994 | #ifdef CONFIG_SMP |
41c7ce9a | 7995 | rq->sd = NULL; |
57d885fe | 7996 | rq->rd = NULL; |
e51fd5e2 | 7997 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 7998 | rq->post_schedule = 0; |
1da177e4 | 7999 | rq->active_balance = 0; |
dd41f596 | 8000 | rq->next_balance = jiffies; |
1da177e4 | 8001 | rq->push_cpu = 0; |
0a2966b4 | 8002 | rq->cpu = i; |
1f11eb6a | 8003 | rq->online = 0; |
eae0c9df MG |
8004 | rq->idle_stamp = 0; |
8005 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 8006 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
8007 | #ifdef CONFIG_NO_HZ |
8008 | rq->nohz_balance_kick = 0; | |
8009 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
8010 | #endif | |
1da177e4 | 8011 | #endif |
8f4d37ec | 8012 | init_rq_hrtick(rq); |
1da177e4 | 8013 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
8014 | } |
8015 | ||
2dd73a4f | 8016 | set_load_weight(&init_task); |
b50f60ce | 8017 | |
e107be36 AK |
8018 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8019 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8020 | #endif | |
8021 | ||
c9819f45 | 8022 | #ifdef CONFIG_SMP |
962cf36c | 8023 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8024 | #endif |
8025 | ||
b50f60ce | 8026 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 8027 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
8028 | #endif |
8029 | ||
1da177e4 LT |
8030 | /* |
8031 | * The boot idle thread does lazy MMU switching as well: | |
8032 | */ | |
8033 | atomic_inc(&init_mm.mm_count); | |
8034 | enter_lazy_tlb(&init_mm, current); | |
8035 | ||
8036 | /* | |
8037 | * Make us the idle thread. Technically, schedule() should not be | |
8038 | * called from this thread, however somewhere below it might be, | |
8039 | * but because we are the idle thread, we just pick up running again | |
8040 | * when this runqueue becomes "idle". | |
8041 | */ | |
8042 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8043 | |
8044 | calc_load_update = jiffies + LOAD_FREQ; | |
8045 | ||
dd41f596 IM |
8046 | /* |
8047 | * During early bootup we pretend to be a normal task: | |
8048 | */ | |
8049 | current->sched_class = &fair_sched_class; | |
6892b75e | 8050 | |
6a7b3dc3 | 8051 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8052 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8053 | #ifdef CONFIG_SMP |
7d1e6a9b | 8054 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8055 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8056 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8057 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8058 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8059 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8060 | #endif |
bdddd296 RR |
8061 | /* May be allocated at isolcpus cmdline parse time */ |
8062 | if (cpu_isolated_map == NULL) | |
8063 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8064 | #endif /* SMP */ |
6a7b3dc3 | 8065 | |
6892b75e | 8066 | scheduler_running = 1; |
1da177e4 LT |
8067 | } |
8068 | ||
8069 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
8070 | static inline int preempt_count_equals(int preempt_offset) |
8071 | { | |
234da7bc | 8072 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
8073 | |
8074 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
8075 | } | |
8076 | ||
d894837f | 8077 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8078 | { |
48f24c4d | 8079 | #ifdef in_atomic |
1da177e4 LT |
8080 | static unsigned long prev_jiffy; /* ratelimiting */ |
8081 | ||
e4aafea2 FW |
8082 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8083 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8084 | return; |
8085 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8086 | return; | |
8087 | prev_jiffy = jiffies; | |
8088 | ||
3df0fc5b PZ |
8089 | printk(KERN_ERR |
8090 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8091 | file, line); | |
8092 | printk(KERN_ERR | |
8093 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8094 | in_atomic(), irqs_disabled(), | |
8095 | current->pid, current->comm); | |
aef745fc IM |
8096 | |
8097 | debug_show_held_locks(current); | |
8098 | if (irqs_disabled()) | |
8099 | print_irqtrace_events(current); | |
8100 | dump_stack(); | |
1da177e4 LT |
8101 | #endif |
8102 | } | |
8103 | EXPORT_SYMBOL(__might_sleep); | |
8104 | #endif | |
8105 | ||
8106 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8107 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8108 | { | |
8109 | int on_rq; | |
3e51f33f | 8110 | |
3a5e4dc1 AK |
8111 | on_rq = p->se.on_rq; |
8112 | if (on_rq) | |
8113 | deactivate_task(rq, p, 0); | |
8114 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8115 | if (on_rq) { | |
8116 | activate_task(rq, p, 0); | |
8117 | resched_task(rq->curr); | |
8118 | } | |
8119 | } | |
8120 | ||
1da177e4 LT |
8121 | void normalize_rt_tasks(void) |
8122 | { | |
a0f98a1c | 8123 | struct task_struct *g, *p; |
1da177e4 | 8124 | unsigned long flags; |
70b97a7f | 8125 | struct rq *rq; |
1da177e4 | 8126 | |
4cf5d77a | 8127 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8128 | do_each_thread(g, p) { |
178be793 IM |
8129 | /* |
8130 | * Only normalize user tasks: | |
8131 | */ | |
8132 | if (!p->mm) | |
8133 | continue; | |
8134 | ||
6cfb0d5d | 8135 | p->se.exec_start = 0; |
6cfb0d5d | 8136 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8137 | p->se.statistics.wait_start = 0; |
8138 | p->se.statistics.sleep_start = 0; | |
8139 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8140 | #endif |
dd41f596 IM |
8141 | |
8142 | if (!rt_task(p)) { | |
8143 | /* | |
8144 | * Renice negative nice level userspace | |
8145 | * tasks back to 0: | |
8146 | */ | |
8147 | if (TASK_NICE(p) < 0 && p->mm) | |
8148 | set_user_nice(p, 0); | |
1da177e4 | 8149 | continue; |
dd41f596 | 8150 | } |
1da177e4 | 8151 | |
1d615482 | 8152 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8153 | rq = __task_rq_lock(p); |
1da177e4 | 8154 | |
178be793 | 8155 | normalize_task(rq, p); |
3a5e4dc1 | 8156 | |
b29739f9 | 8157 | __task_rq_unlock(rq); |
1d615482 | 8158 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8159 | } while_each_thread(g, p); |
8160 | ||
4cf5d77a | 8161 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8162 | } |
8163 | ||
8164 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8165 | |
67fc4e0c | 8166 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8167 | /* |
67fc4e0c | 8168 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8169 | * |
8170 | * They can only be called when the whole system has been | |
8171 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8172 | * activity can take place. Using them for anything else would | |
8173 | * be a serious bug, and as a result, they aren't even visible | |
8174 | * under any other configuration. | |
8175 | */ | |
8176 | ||
8177 | /** | |
8178 | * curr_task - return the current task for a given cpu. | |
8179 | * @cpu: the processor in question. | |
8180 | * | |
8181 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8182 | */ | |
36c8b586 | 8183 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8184 | { |
8185 | return cpu_curr(cpu); | |
8186 | } | |
8187 | ||
67fc4e0c JW |
8188 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8189 | ||
8190 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8191 | /** |
8192 | * set_curr_task - set the current task for a given cpu. | |
8193 | * @cpu: the processor in question. | |
8194 | * @p: the task pointer to set. | |
8195 | * | |
8196 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8197 | * are serviced on a separate stack. It allows the architecture to switch the |
8198 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8199 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8200 | * and caller must save the original value of the current task (see | |
8201 | * curr_task() above) and restore that value before reenabling interrupts and | |
8202 | * re-starting the system. | |
8203 | * | |
8204 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8205 | */ | |
36c8b586 | 8206 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8207 | { |
8208 | cpu_curr(cpu) = p; | |
8209 | } | |
8210 | ||
8211 | #endif | |
29f59db3 | 8212 | |
bccbe08a PZ |
8213 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8214 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8215 | { |
8216 | int i; | |
8217 | ||
8218 | for_each_possible_cpu(i) { | |
8219 | if (tg->cfs_rq) | |
8220 | kfree(tg->cfs_rq[i]); | |
8221 | if (tg->se) | |
8222 | kfree(tg->se[i]); | |
6f505b16 PZ |
8223 | } |
8224 | ||
8225 | kfree(tg->cfs_rq); | |
8226 | kfree(tg->se); | |
6f505b16 PZ |
8227 | } |
8228 | ||
ec7dc8ac DG |
8229 | static |
8230 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8231 | { |
29f59db3 | 8232 | struct cfs_rq *cfs_rq; |
eab17229 | 8233 | struct sched_entity *se; |
9b5b7751 | 8234 | struct rq *rq; |
29f59db3 SV |
8235 | int i; |
8236 | ||
434d53b0 | 8237 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8238 | if (!tg->cfs_rq) |
8239 | goto err; | |
434d53b0 | 8240 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8241 | if (!tg->se) |
8242 | goto err; | |
052f1dc7 PZ |
8243 | |
8244 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8245 | |
8246 | for_each_possible_cpu(i) { | |
9b5b7751 | 8247 | rq = cpu_rq(i); |
29f59db3 | 8248 | |
eab17229 LZ |
8249 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8250 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8251 | if (!cfs_rq) |
8252 | goto err; | |
8253 | ||
eab17229 LZ |
8254 | se = kzalloc_node(sizeof(struct sched_entity), |
8255 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8256 | if (!se) |
dfc12eb2 | 8257 | goto err_free_rq; |
29f59db3 | 8258 | |
3d4b47b4 | 8259 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8260 | } |
8261 | ||
8262 | return 1; | |
8263 | ||
49246274 | 8264 | err_free_rq: |
dfc12eb2 | 8265 | kfree(cfs_rq); |
49246274 | 8266 | err: |
bccbe08a PZ |
8267 | return 0; |
8268 | } | |
8269 | ||
bccbe08a PZ |
8270 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8271 | { | |
3d4b47b4 PZ |
8272 | struct rq *rq = cpu_rq(cpu); |
8273 | unsigned long flags; | |
3d4b47b4 PZ |
8274 | |
8275 | /* | |
8276 | * Only empty task groups can be destroyed; so we can speculatively | |
8277 | * check on_list without danger of it being re-added. | |
8278 | */ | |
8279 | if (!tg->cfs_rq[cpu]->on_list) | |
8280 | return; | |
8281 | ||
8282 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8283 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8284 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8285 | } |
6d6bc0ad | 8286 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8287 | static inline void free_fair_sched_group(struct task_group *tg) |
8288 | { | |
8289 | } | |
8290 | ||
ec7dc8ac DG |
8291 | static inline |
8292 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8293 | { |
8294 | return 1; | |
8295 | } | |
8296 | ||
bccbe08a PZ |
8297 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8298 | { | |
8299 | } | |
6d6bc0ad | 8300 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8301 | |
8302 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8303 | static void free_rt_sched_group(struct task_group *tg) |
8304 | { | |
8305 | int i; | |
8306 | ||
d0b27fa7 PZ |
8307 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8308 | ||
bccbe08a PZ |
8309 | for_each_possible_cpu(i) { |
8310 | if (tg->rt_rq) | |
8311 | kfree(tg->rt_rq[i]); | |
8312 | if (tg->rt_se) | |
8313 | kfree(tg->rt_se[i]); | |
8314 | } | |
8315 | ||
8316 | kfree(tg->rt_rq); | |
8317 | kfree(tg->rt_se); | |
8318 | } | |
8319 | ||
ec7dc8ac DG |
8320 | static |
8321 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8322 | { |
8323 | struct rt_rq *rt_rq; | |
eab17229 | 8324 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8325 | struct rq *rq; |
8326 | int i; | |
8327 | ||
434d53b0 | 8328 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8329 | if (!tg->rt_rq) |
8330 | goto err; | |
434d53b0 | 8331 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8332 | if (!tg->rt_se) |
8333 | goto err; | |
8334 | ||
d0b27fa7 PZ |
8335 | init_rt_bandwidth(&tg->rt_bandwidth, |
8336 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8337 | |
8338 | for_each_possible_cpu(i) { | |
8339 | rq = cpu_rq(i); | |
8340 | ||
eab17229 LZ |
8341 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8342 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8343 | if (!rt_rq) |
8344 | goto err; | |
29f59db3 | 8345 | |
eab17229 LZ |
8346 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8347 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8348 | if (!rt_se) |
dfc12eb2 | 8349 | goto err_free_rq; |
29f59db3 | 8350 | |
3d4b47b4 | 8351 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8352 | } |
8353 | ||
bccbe08a PZ |
8354 | return 1; |
8355 | ||
49246274 | 8356 | err_free_rq: |
dfc12eb2 | 8357 | kfree(rt_rq); |
49246274 | 8358 | err: |
bccbe08a PZ |
8359 | return 0; |
8360 | } | |
6d6bc0ad | 8361 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8362 | static inline void free_rt_sched_group(struct task_group *tg) |
8363 | { | |
8364 | } | |
8365 | ||
ec7dc8ac DG |
8366 | static inline |
8367 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8368 | { |
8369 | return 1; | |
8370 | } | |
6d6bc0ad | 8371 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8372 | |
7c941438 | 8373 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8374 | static void free_sched_group(struct task_group *tg) |
8375 | { | |
8376 | free_fair_sched_group(tg); | |
8377 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8378 | autogroup_free(tg); |
bccbe08a PZ |
8379 | kfree(tg); |
8380 | } | |
8381 | ||
8382 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8383 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8384 | { |
8385 | struct task_group *tg; | |
8386 | unsigned long flags; | |
bccbe08a PZ |
8387 | |
8388 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8389 | if (!tg) | |
8390 | return ERR_PTR(-ENOMEM); | |
8391 | ||
ec7dc8ac | 8392 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8393 | goto err; |
8394 | ||
ec7dc8ac | 8395 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8396 | goto err; |
8397 | ||
8ed36996 | 8398 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8399 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8400 | |
8401 | WARN_ON(!parent); /* root should already exist */ | |
8402 | ||
8403 | tg->parent = parent; | |
f473aa5e | 8404 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8405 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8406 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8407 | |
9b5b7751 | 8408 | return tg; |
29f59db3 SV |
8409 | |
8410 | err: | |
6f505b16 | 8411 | free_sched_group(tg); |
29f59db3 SV |
8412 | return ERR_PTR(-ENOMEM); |
8413 | } | |
8414 | ||
9b5b7751 | 8415 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8416 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8417 | { |
29f59db3 | 8418 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8419 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8420 | } |
8421 | ||
9b5b7751 | 8422 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8423 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8424 | { |
8ed36996 | 8425 | unsigned long flags; |
9b5b7751 | 8426 | int i; |
29f59db3 | 8427 | |
3d4b47b4 PZ |
8428 | /* end participation in shares distribution */ |
8429 | for_each_possible_cpu(i) | |
bccbe08a | 8430 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8431 | |
8432 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8433 | list_del_rcu(&tg->list); |
f473aa5e | 8434 | list_del_rcu(&tg->siblings); |
8ed36996 | 8435 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8436 | |
9b5b7751 | 8437 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8438 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8439 | } |
8440 | ||
9b5b7751 | 8441 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8442 | * The caller of this function should have put the task in its new group |
8443 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8444 | * reflect its new group. | |
9b5b7751 SV |
8445 | */ |
8446 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8447 | { |
8448 | int on_rq, running; | |
8449 | unsigned long flags; | |
8450 | struct rq *rq; | |
8451 | ||
8452 | rq = task_rq_lock(tsk, &flags); | |
8453 | ||
051a1d1a | 8454 | running = task_current(rq, tsk); |
29f59db3 SV |
8455 | on_rq = tsk->se.on_rq; |
8456 | ||
0e1f3483 | 8457 | if (on_rq) |
29f59db3 | 8458 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8459 | if (unlikely(running)) |
8460 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8461 | |
810b3817 | 8462 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8463 | if (tsk->sched_class->task_move_group) |
8464 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8465 | else | |
810b3817 | 8466 | #endif |
b2b5ce02 | 8467 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8468 | |
0e1f3483 HS |
8469 | if (unlikely(running)) |
8470 | tsk->sched_class->set_curr_task(rq); | |
8471 | if (on_rq) | |
371fd7e7 | 8472 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8473 | |
29f59db3 SV |
8474 | task_rq_unlock(rq, &flags); |
8475 | } | |
7c941438 | 8476 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8477 | |
052f1dc7 | 8478 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8479 | static DEFINE_MUTEX(shares_mutex); |
8480 | ||
4cf86d77 | 8481 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8482 | { |
8483 | int i; | |
8ed36996 | 8484 | unsigned long flags; |
c61935fd | 8485 | |
ec7dc8ac DG |
8486 | /* |
8487 | * We can't change the weight of the root cgroup. | |
8488 | */ | |
8489 | if (!tg->se[0]) | |
8490 | return -EINVAL; | |
8491 | ||
18d95a28 PZ |
8492 | if (shares < MIN_SHARES) |
8493 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8494 | else if (shares > MAX_SHARES) |
8495 | shares = MAX_SHARES; | |
62fb1851 | 8496 | |
8ed36996 | 8497 | mutex_lock(&shares_mutex); |
9b5b7751 | 8498 | if (tg->shares == shares) |
5cb350ba | 8499 | goto done; |
29f59db3 | 8500 | |
9b5b7751 | 8501 | tg->shares = shares; |
c09595f6 | 8502 | for_each_possible_cpu(i) { |
9437178f PT |
8503 | struct rq *rq = cpu_rq(i); |
8504 | struct sched_entity *se; | |
8505 | ||
8506 | se = tg->se[i]; | |
8507 | /* Propagate contribution to hierarchy */ | |
8508 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8509 | for_each_sched_entity(se) | |
8510 | update_cfs_shares(group_cfs_rq(se), 0); | |
8511 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
c09595f6 | 8512 | } |
29f59db3 | 8513 | |
5cb350ba | 8514 | done: |
8ed36996 | 8515 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8516 | return 0; |
29f59db3 SV |
8517 | } |
8518 | ||
5cb350ba DG |
8519 | unsigned long sched_group_shares(struct task_group *tg) |
8520 | { | |
8521 | return tg->shares; | |
8522 | } | |
052f1dc7 | 8523 | #endif |
5cb350ba | 8524 | |
052f1dc7 | 8525 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8526 | /* |
9f0c1e56 | 8527 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8528 | */ |
9f0c1e56 PZ |
8529 | static DEFINE_MUTEX(rt_constraints_mutex); |
8530 | ||
8531 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8532 | { | |
8533 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8534 | return 1ULL << 20; |
9f0c1e56 | 8535 | |
9a7e0b18 | 8536 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8537 | } |
8538 | ||
9a7e0b18 PZ |
8539 | /* Must be called with tasklist_lock held */ |
8540 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8541 | { |
9a7e0b18 | 8542 | struct task_struct *g, *p; |
b40b2e8e | 8543 | |
9a7e0b18 PZ |
8544 | do_each_thread(g, p) { |
8545 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8546 | return 1; | |
8547 | } while_each_thread(g, p); | |
b40b2e8e | 8548 | |
9a7e0b18 PZ |
8549 | return 0; |
8550 | } | |
b40b2e8e | 8551 | |
9a7e0b18 PZ |
8552 | struct rt_schedulable_data { |
8553 | struct task_group *tg; | |
8554 | u64 rt_period; | |
8555 | u64 rt_runtime; | |
8556 | }; | |
b40b2e8e | 8557 | |
9a7e0b18 PZ |
8558 | static int tg_schedulable(struct task_group *tg, void *data) |
8559 | { | |
8560 | struct rt_schedulable_data *d = data; | |
8561 | struct task_group *child; | |
8562 | unsigned long total, sum = 0; | |
8563 | u64 period, runtime; | |
b40b2e8e | 8564 | |
9a7e0b18 PZ |
8565 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8566 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8567 | |
9a7e0b18 PZ |
8568 | if (tg == d->tg) { |
8569 | period = d->rt_period; | |
8570 | runtime = d->rt_runtime; | |
b40b2e8e | 8571 | } |
b40b2e8e | 8572 | |
4653f803 PZ |
8573 | /* |
8574 | * Cannot have more runtime than the period. | |
8575 | */ | |
8576 | if (runtime > period && runtime != RUNTIME_INF) | |
8577 | return -EINVAL; | |
6f505b16 | 8578 | |
4653f803 PZ |
8579 | /* |
8580 | * Ensure we don't starve existing RT tasks. | |
8581 | */ | |
9a7e0b18 PZ |
8582 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8583 | return -EBUSY; | |
6f505b16 | 8584 | |
9a7e0b18 | 8585 | total = to_ratio(period, runtime); |
6f505b16 | 8586 | |
4653f803 PZ |
8587 | /* |
8588 | * Nobody can have more than the global setting allows. | |
8589 | */ | |
8590 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8591 | return -EINVAL; | |
6f505b16 | 8592 | |
4653f803 PZ |
8593 | /* |
8594 | * The sum of our children's runtime should not exceed our own. | |
8595 | */ | |
9a7e0b18 PZ |
8596 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8597 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8598 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8599 | |
9a7e0b18 PZ |
8600 | if (child == d->tg) { |
8601 | period = d->rt_period; | |
8602 | runtime = d->rt_runtime; | |
8603 | } | |
6f505b16 | 8604 | |
9a7e0b18 | 8605 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8606 | } |
6f505b16 | 8607 | |
9a7e0b18 PZ |
8608 | if (sum > total) |
8609 | return -EINVAL; | |
8610 | ||
8611 | return 0; | |
6f505b16 PZ |
8612 | } |
8613 | ||
9a7e0b18 | 8614 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8615 | { |
9a7e0b18 PZ |
8616 | struct rt_schedulable_data data = { |
8617 | .tg = tg, | |
8618 | .rt_period = period, | |
8619 | .rt_runtime = runtime, | |
8620 | }; | |
8621 | ||
8622 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8623 | } |
8624 | ||
d0b27fa7 PZ |
8625 | static int tg_set_bandwidth(struct task_group *tg, |
8626 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8627 | { |
ac086bc2 | 8628 | int i, err = 0; |
9f0c1e56 | 8629 | |
9f0c1e56 | 8630 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8631 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8632 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8633 | if (err) | |
9f0c1e56 | 8634 | goto unlock; |
ac086bc2 | 8635 | |
0986b11b | 8636 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8637 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8638 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8639 | |
8640 | for_each_possible_cpu(i) { | |
8641 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8642 | ||
0986b11b | 8643 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8644 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8645 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8646 | } |
0986b11b | 8647 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8648 | unlock: |
521f1a24 | 8649 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8650 | mutex_unlock(&rt_constraints_mutex); |
8651 | ||
8652 | return err; | |
6f505b16 PZ |
8653 | } |
8654 | ||
d0b27fa7 PZ |
8655 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8656 | { | |
8657 | u64 rt_runtime, rt_period; | |
8658 | ||
8659 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8660 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8661 | if (rt_runtime_us < 0) | |
8662 | rt_runtime = RUNTIME_INF; | |
8663 | ||
8664 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8665 | } | |
8666 | ||
9f0c1e56 PZ |
8667 | long sched_group_rt_runtime(struct task_group *tg) |
8668 | { | |
8669 | u64 rt_runtime_us; | |
8670 | ||
d0b27fa7 | 8671 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8672 | return -1; |
8673 | ||
d0b27fa7 | 8674 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8675 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8676 | return rt_runtime_us; | |
8677 | } | |
d0b27fa7 PZ |
8678 | |
8679 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8680 | { | |
8681 | u64 rt_runtime, rt_period; | |
8682 | ||
8683 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8684 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8685 | ||
619b0488 R |
8686 | if (rt_period == 0) |
8687 | return -EINVAL; | |
8688 | ||
d0b27fa7 PZ |
8689 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8690 | } | |
8691 | ||
8692 | long sched_group_rt_period(struct task_group *tg) | |
8693 | { | |
8694 | u64 rt_period_us; | |
8695 | ||
8696 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8697 | do_div(rt_period_us, NSEC_PER_USEC); | |
8698 | return rt_period_us; | |
8699 | } | |
8700 | ||
8701 | static int sched_rt_global_constraints(void) | |
8702 | { | |
4653f803 | 8703 | u64 runtime, period; |
d0b27fa7 PZ |
8704 | int ret = 0; |
8705 | ||
ec5d4989 HS |
8706 | if (sysctl_sched_rt_period <= 0) |
8707 | return -EINVAL; | |
8708 | ||
4653f803 PZ |
8709 | runtime = global_rt_runtime(); |
8710 | period = global_rt_period(); | |
8711 | ||
8712 | /* | |
8713 | * Sanity check on the sysctl variables. | |
8714 | */ | |
8715 | if (runtime > period && runtime != RUNTIME_INF) | |
8716 | return -EINVAL; | |
10b612f4 | 8717 | |
d0b27fa7 | 8718 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8719 | read_lock(&tasklist_lock); |
4653f803 | 8720 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8721 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8722 | mutex_unlock(&rt_constraints_mutex); |
8723 | ||
8724 | return ret; | |
8725 | } | |
54e99124 DG |
8726 | |
8727 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8728 | { | |
8729 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8730 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8731 | return 0; | |
8732 | ||
8733 | return 1; | |
8734 | } | |
8735 | ||
6d6bc0ad | 8736 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8737 | static int sched_rt_global_constraints(void) |
8738 | { | |
ac086bc2 PZ |
8739 | unsigned long flags; |
8740 | int i; | |
8741 | ||
ec5d4989 HS |
8742 | if (sysctl_sched_rt_period <= 0) |
8743 | return -EINVAL; | |
8744 | ||
60aa605d PZ |
8745 | /* |
8746 | * There's always some RT tasks in the root group | |
8747 | * -- migration, kstopmachine etc.. | |
8748 | */ | |
8749 | if (sysctl_sched_rt_runtime == 0) | |
8750 | return -EBUSY; | |
8751 | ||
0986b11b | 8752 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8753 | for_each_possible_cpu(i) { |
8754 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8755 | ||
0986b11b | 8756 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8757 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8758 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8759 | } |
0986b11b | 8760 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8761 | |
d0b27fa7 PZ |
8762 | return 0; |
8763 | } | |
6d6bc0ad | 8764 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8765 | |
8766 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8767 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8768 | loff_t *ppos) |
8769 | { | |
8770 | int ret; | |
8771 | int old_period, old_runtime; | |
8772 | static DEFINE_MUTEX(mutex); | |
8773 | ||
8774 | mutex_lock(&mutex); | |
8775 | old_period = sysctl_sched_rt_period; | |
8776 | old_runtime = sysctl_sched_rt_runtime; | |
8777 | ||
8d65af78 | 8778 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8779 | |
8780 | if (!ret && write) { | |
8781 | ret = sched_rt_global_constraints(); | |
8782 | if (ret) { | |
8783 | sysctl_sched_rt_period = old_period; | |
8784 | sysctl_sched_rt_runtime = old_runtime; | |
8785 | } else { | |
8786 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8787 | def_rt_bandwidth.rt_period = | |
8788 | ns_to_ktime(global_rt_period()); | |
8789 | } | |
8790 | } | |
8791 | mutex_unlock(&mutex); | |
8792 | ||
8793 | return ret; | |
8794 | } | |
68318b8e | 8795 | |
052f1dc7 | 8796 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8797 | |
8798 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8799 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8800 | { |
2b01dfe3 PM |
8801 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8802 | struct task_group, css); | |
68318b8e SV |
8803 | } |
8804 | ||
8805 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8806 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8807 | { |
ec7dc8ac | 8808 | struct task_group *tg, *parent; |
68318b8e | 8809 | |
2b01dfe3 | 8810 | if (!cgrp->parent) { |
68318b8e | 8811 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8812 | return &root_task_group.css; |
68318b8e SV |
8813 | } |
8814 | ||
ec7dc8ac DG |
8815 | parent = cgroup_tg(cgrp->parent); |
8816 | tg = sched_create_group(parent); | |
68318b8e SV |
8817 | if (IS_ERR(tg)) |
8818 | return ERR_PTR(-ENOMEM); | |
8819 | ||
68318b8e SV |
8820 | return &tg->css; |
8821 | } | |
8822 | ||
41a2d6cf IM |
8823 | static void |
8824 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8825 | { |
2b01dfe3 | 8826 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8827 | |
8828 | sched_destroy_group(tg); | |
8829 | } | |
8830 | ||
41a2d6cf | 8831 | static int |
be367d09 | 8832 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8833 | { |
b68aa230 | 8834 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8835 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8836 | return -EINVAL; |
8837 | #else | |
68318b8e SV |
8838 | /* We don't support RT-tasks being in separate groups */ |
8839 | if (tsk->sched_class != &fair_sched_class) | |
8840 | return -EINVAL; | |
b68aa230 | 8841 | #endif |
be367d09 BB |
8842 | return 0; |
8843 | } | |
68318b8e | 8844 | |
be367d09 BB |
8845 | static int |
8846 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8847 | struct task_struct *tsk, bool threadgroup) | |
8848 | { | |
8849 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8850 | if (retval) | |
8851 | return retval; | |
8852 | if (threadgroup) { | |
8853 | struct task_struct *c; | |
8854 | rcu_read_lock(); | |
8855 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8856 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8857 | if (retval) { | |
8858 | rcu_read_unlock(); | |
8859 | return retval; | |
8860 | } | |
8861 | } | |
8862 | rcu_read_unlock(); | |
8863 | } | |
68318b8e SV |
8864 | return 0; |
8865 | } | |
8866 | ||
8867 | static void | |
2b01dfe3 | 8868 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8869 | struct cgroup *old_cont, struct task_struct *tsk, |
8870 | bool threadgroup) | |
68318b8e SV |
8871 | { |
8872 | sched_move_task(tsk); | |
be367d09 BB |
8873 | if (threadgroup) { |
8874 | struct task_struct *c; | |
8875 | rcu_read_lock(); | |
8876 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8877 | sched_move_task(c); | |
8878 | } | |
8879 | rcu_read_unlock(); | |
8880 | } | |
68318b8e SV |
8881 | } |
8882 | ||
052f1dc7 | 8883 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8884 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8885 | u64 shareval) |
68318b8e | 8886 | { |
2b01dfe3 | 8887 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8888 | } |
8889 | ||
f4c753b7 | 8890 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8891 | { |
2b01dfe3 | 8892 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8893 | |
8894 | return (u64) tg->shares; | |
8895 | } | |
6d6bc0ad | 8896 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8897 | |
052f1dc7 | 8898 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8899 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8900 | s64 val) |
6f505b16 | 8901 | { |
06ecb27c | 8902 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8903 | } |
8904 | ||
06ecb27c | 8905 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8906 | { |
06ecb27c | 8907 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8908 | } |
d0b27fa7 PZ |
8909 | |
8910 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8911 | u64 rt_period_us) | |
8912 | { | |
8913 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8914 | } | |
8915 | ||
8916 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8917 | { | |
8918 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8919 | } | |
6d6bc0ad | 8920 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8921 | |
fe5c7cc2 | 8922 | static struct cftype cpu_files[] = { |
052f1dc7 | 8923 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8924 | { |
8925 | .name = "shares", | |
f4c753b7 PM |
8926 | .read_u64 = cpu_shares_read_u64, |
8927 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8928 | }, |
052f1dc7 PZ |
8929 | #endif |
8930 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8931 | { |
9f0c1e56 | 8932 | .name = "rt_runtime_us", |
06ecb27c PM |
8933 | .read_s64 = cpu_rt_runtime_read, |
8934 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8935 | }, |
d0b27fa7 PZ |
8936 | { |
8937 | .name = "rt_period_us", | |
f4c753b7 PM |
8938 | .read_u64 = cpu_rt_period_read_uint, |
8939 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8940 | }, |
052f1dc7 | 8941 | #endif |
68318b8e SV |
8942 | }; |
8943 | ||
8944 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8945 | { | |
fe5c7cc2 | 8946 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8947 | } |
8948 | ||
8949 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8950 | .name = "cpu", |
8951 | .create = cpu_cgroup_create, | |
8952 | .destroy = cpu_cgroup_destroy, | |
8953 | .can_attach = cpu_cgroup_can_attach, | |
8954 | .attach = cpu_cgroup_attach, | |
8955 | .populate = cpu_cgroup_populate, | |
8956 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8957 | .early_init = 1, |
8958 | }; | |
8959 | ||
052f1dc7 | 8960 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8961 | |
8962 | #ifdef CONFIG_CGROUP_CPUACCT | |
8963 | ||
8964 | /* | |
8965 | * CPU accounting code for task groups. | |
8966 | * | |
8967 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8968 | * (balbir@in.ibm.com). | |
8969 | */ | |
8970 | ||
934352f2 | 8971 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8972 | struct cpuacct { |
8973 | struct cgroup_subsys_state css; | |
8974 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8975 | u64 __percpu *cpuusage; |
ef12fefa | 8976 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8977 | struct cpuacct *parent; |
d842de87 SV |
8978 | }; |
8979 | ||
8980 | struct cgroup_subsys cpuacct_subsys; | |
8981 | ||
8982 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8983 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8984 | { |
32cd756a | 8985 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8986 | struct cpuacct, css); |
8987 | } | |
8988 | ||
8989 | /* return cpu accounting group to which this task belongs */ | |
8990 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8991 | { | |
8992 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8993 | struct cpuacct, css); | |
8994 | } | |
8995 | ||
8996 | /* create a new cpu accounting group */ | |
8997 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8998 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8999 | { |
9000 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 9001 | int i; |
d842de87 SV |
9002 | |
9003 | if (!ca) | |
ef12fefa | 9004 | goto out; |
d842de87 SV |
9005 | |
9006 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9007 | if (!ca->cpuusage) |
9008 | goto out_free_ca; | |
9009 | ||
9010 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9011 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9012 | goto out_free_counters; | |
d842de87 | 9013 | |
934352f2 BR |
9014 | if (cgrp->parent) |
9015 | ca->parent = cgroup_ca(cgrp->parent); | |
9016 | ||
d842de87 | 9017 | return &ca->css; |
ef12fefa BR |
9018 | |
9019 | out_free_counters: | |
9020 | while (--i >= 0) | |
9021 | percpu_counter_destroy(&ca->cpustat[i]); | |
9022 | free_percpu(ca->cpuusage); | |
9023 | out_free_ca: | |
9024 | kfree(ca); | |
9025 | out: | |
9026 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9027 | } |
9028 | ||
9029 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9030 | static void |
32cd756a | 9031 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9032 | { |
32cd756a | 9033 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9034 | int i; |
d842de87 | 9035 | |
ef12fefa BR |
9036 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9037 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9038 | free_percpu(ca->cpuusage); |
9039 | kfree(ca); | |
9040 | } | |
9041 | ||
720f5498 KC |
9042 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9043 | { | |
b36128c8 | 9044 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9045 | u64 data; |
9046 | ||
9047 | #ifndef CONFIG_64BIT | |
9048 | /* | |
9049 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9050 | */ | |
05fa785c | 9051 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9052 | data = *cpuusage; |
05fa785c | 9053 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9054 | #else |
9055 | data = *cpuusage; | |
9056 | #endif | |
9057 | ||
9058 | return data; | |
9059 | } | |
9060 | ||
9061 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9062 | { | |
b36128c8 | 9063 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9064 | |
9065 | #ifndef CONFIG_64BIT | |
9066 | /* | |
9067 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9068 | */ | |
05fa785c | 9069 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9070 | *cpuusage = val; |
05fa785c | 9071 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9072 | #else |
9073 | *cpuusage = val; | |
9074 | #endif | |
9075 | } | |
9076 | ||
d842de87 | 9077 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9078 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9079 | { |
32cd756a | 9080 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9081 | u64 totalcpuusage = 0; |
9082 | int i; | |
9083 | ||
720f5498 KC |
9084 | for_each_present_cpu(i) |
9085 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9086 | |
9087 | return totalcpuusage; | |
9088 | } | |
9089 | ||
0297b803 DG |
9090 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9091 | u64 reset) | |
9092 | { | |
9093 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9094 | int err = 0; | |
9095 | int i; | |
9096 | ||
9097 | if (reset) { | |
9098 | err = -EINVAL; | |
9099 | goto out; | |
9100 | } | |
9101 | ||
720f5498 KC |
9102 | for_each_present_cpu(i) |
9103 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9104 | |
0297b803 DG |
9105 | out: |
9106 | return err; | |
9107 | } | |
9108 | ||
e9515c3c KC |
9109 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9110 | struct seq_file *m) | |
9111 | { | |
9112 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9113 | u64 percpu; | |
9114 | int i; | |
9115 | ||
9116 | for_each_present_cpu(i) { | |
9117 | percpu = cpuacct_cpuusage_read(ca, i); | |
9118 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9119 | } | |
9120 | seq_printf(m, "\n"); | |
9121 | return 0; | |
9122 | } | |
9123 | ||
ef12fefa BR |
9124 | static const char *cpuacct_stat_desc[] = { |
9125 | [CPUACCT_STAT_USER] = "user", | |
9126 | [CPUACCT_STAT_SYSTEM] = "system", | |
9127 | }; | |
9128 | ||
9129 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9130 | struct cgroup_map_cb *cb) | |
9131 | { | |
9132 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9133 | int i; | |
9134 | ||
9135 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9136 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9137 | val = cputime64_to_clock_t(val); | |
9138 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9139 | } | |
9140 | return 0; | |
9141 | } | |
9142 | ||
d842de87 SV |
9143 | static struct cftype files[] = { |
9144 | { | |
9145 | .name = "usage", | |
f4c753b7 PM |
9146 | .read_u64 = cpuusage_read, |
9147 | .write_u64 = cpuusage_write, | |
d842de87 | 9148 | }, |
e9515c3c KC |
9149 | { |
9150 | .name = "usage_percpu", | |
9151 | .read_seq_string = cpuacct_percpu_seq_read, | |
9152 | }, | |
ef12fefa BR |
9153 | { |
9154 | .name = "stat", | |
9155 | .read_map = cpuacct_stats_show, | |
9156 | }, | |
d842de87 SV |
9157 | }; |
9158 | ||
32cd756a | 9159 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9160 | { |
32cd756a | 9161 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9162 | } |
9163 | ||
9164 | /* | |
9165 | * charge this task's execution time to its accounting group. | |
9166 | * | |
9167 | * called with rq->lock held. | |
9168 | */ | |
9169 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9170 | { | |
9171 | struct cpuacct *ca; | |
934352f2 | 9172 | int cpu; |
d842de87 | 9173 | |
c40c6f85 | 9174 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9175 | return; |
9176 | ||
934352f2 | 9177 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9178 | |
9179 | rcu_read_lock(); | |
9180 | ||
d842de87 | 9181 | ca = task_ca(tsk); |
d842de87 | 9182 | |
934352f2 | 9183 | for (; ca; ca = ca->parent) { |
b36128c8 | 9184 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9185 | *cpuusage += cputime; |
9186 | } | |
a18b83b7 BR |
9187 | |
9188 | rcu_read_unlock(); | |
d842de87 SV |
9189 | } |
9190 | ||
fa535a77 AB |
9191 | /* |
9192 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9193 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9194 | * percpu_counter_add with values large enough to always overflow the | |
9195 | * per cpu batch limit causing bad SMP scalability. | |
9196 | * | |
9197 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9198 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9199 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9200 | */ | |
9201 | #ifdef CONFIG_SMP | |
9202 | #define CPUACCT_BATCH \ | |
9203 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9204 | #else | |
9205 | #define CPUACCT_BATCH 0 | |
9206 | #endif | |
9207 | ||
ef12fefa BR |
9208 | /* |
9209 | * Charge the system/user time to the task's accounting group. | |
9210 | */ | |
9211 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9212 | enum cpuacct_stat_index idx, cputime_t val) | |
9213 | { | |
9214 | struct cpuacct *ca; | |
fa535a77 | 9215 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9216 | |
9217 | if (unlikely(!cpuacct_subsys.active)) | |
9218 | return; | |
9219 | ||
9220 | rcu_read_lock(); | |
9221 | ca = task_ca(tsk); | |
9222 | ||
9223 | do { | |
fa535a77 | 9224 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9225 | ca = ca->parent; |
9226 | } while (ca); | |
9227 | rcu_read_unlock(); | |
9228 | } | |
9229 | ||
d842de87 SV |
9230 | struct cgroup_subsys cpuacct_subsys = { |
9231 | .name = "cpuacct", | |
9232 | .create = cpuacct_create, | |
9233 | .destroy = cpuacct_destroy, | |
9234 | .populate = cpuacct_populate, | |
9235 | .subsys_id = cpuacct_subsys_id, | |
9236 | }; | |
9237 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf | 9238 |