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 | ||
068c5cc5 PZ |
609 | if (p->flags & PF_EXITING) |
610 | return &root_task_group; | |
611 | ||
dc61b1d6 PZ |
612 | css = task_subsys_state_check(p, cpu_cgroup_subsys_id, |
613 | lockdep_is_held(&task_rq(p)->lock)); | |
5091faa4 MG |
614 | tg = container_of(css, struct task_group, css); |
615 | ||
616 | return autogroup_task_group(p, tg); | |
dc61b1d6 PZ |
617 | } |
618 | ||
619 | /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */ | |
620 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) | |
621 | { | |
622 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
623 | p->se.cfs_rq = task_group(p)->cfs_rq[cpu]; | |
624 | p->se.parent = task_group(p)->se[cpu]; | |
625 | #endif | |
626 | ||
627 | #ifdef CONFIG_RT_GROUP_SCHED | |
628 | p->rt.rt_rq = task_group(p)->rt_rq[cpu]; | |
629 | p->rt.parent = task_group(p)->rt_se[cpu]; | |
630 | #endif | |
631 | } | |
632 | ||
633 | #else /* CONFIG_CGROUP_SCHED */ | |
634 | ||
635 | static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } | |
636 | static inline struct task_group *task_group(struct task_struct *p) | |
637 | { | |
638 | return NULL; | |
639 | } | |
640 | ||
641 | #endif /* CONFIG_CGROUP_SCHED */ | |
642 | ||
fe44d621 | 643 | static void update_rq_clock_task(struct rq *rq, s64 delta); |
305e6835 | 644 | |
fe44d621 | 645 | static void update_rq_clock(struct rq *rq) |
3e51f33f | 646 | { |
fe44d621 | 647 | s64 delta; |
305e6835 | 648 | |
f26f9aff MG |
649 | if (rq->skip_clock_update) |
650 | return; | |
aa483808 | 651 | |
fe44d621 PZ |
652 | delta = sched_clock_cpu(cpu_of(rq)) - rq->clock; |
653 | rq->clock += delta; | |
654 | update_rq_clock_task(rq, delta); | |
3e51f33f PZ |
655 | } |
656 | ||
bf5c91ba IM |
657 | /* |
658 | * Tunables that become constants when CONFIG_SCHED_DEBUG is off: | |
659 | */ | |
660 | #ifdef CONFIG_SCHED_DEBUG | |
661 | # define const_debug __read_mostly | |
662 | #else | |
663 | # define const_debug static const | |
664 | #endif | |
665 | ||
017730c1 IM |
666 | /** |
667 | * runqueue_is_locked | |
e17b38bf | 668 | * @cpu: the processor in question. |
017730c1 IM |
669 | * |
670 | * Returns true if the current cpu runqueue is locked. | |
671 | * This interface allows printk to be called with the runqueue lock | |
672 | * held and know whether or not it is OK to wake up the klogd. | |
673 | */ | |
89f19f04 | 674 | int runqueue_is_locked(int cpu) |
017730c1 | 675 | { |
05fa785c | 676 | return raw_spin_is_locked(&cpu_rq(cpu)->lock); |
017730c1 IM |
677 | } |
678 | ||
bf5c91ba IM |
679 | /* |
680 | * Debugging: various feature bits | |
681 | */ | |
f00b45c1 PZ |
682 | |
683 | #define SCHED_FEAT(name, enabled) \ | |
684 | __SCHED_FEAT_##name , | |
685 | ||
bf5c91ba | 686 | enum { |
f00b45c1 | 687 | #include "sched_features.h" |
bf5c91ba IM |
688 | }; |
689 | ||
f00b45c1 PZ |
690 | #undef SCHED_FEAT |
691 | ||
692 | #define SCHED_FEAT(name, enabled) \ | |
693 | (1UL << __SCHED_FEAT_##name) * enabled | | |
694 | ||
bf5c91ba | 695 | const_debug unsigned int sysctl_sched_features = |
f00b45c1 PZ |
696 | #include "sched_features.h" |
697 | 0; | |
698 | ||
699 | #undef SCHED_FEAT | |
700 | ||
701 | #ifdef CONFIG_SCHED_DEBUG | |
702 | #define SCHED_FEAT(name, enabled) \ | |
703 | #name , | |
704 | ||
983ed7a6 | 705 | static __read_mostly char *sched_feat_names[] = { |
f00b45c1 PZ |
706 | #include "sched_features.h" |
707 | NULL | |
708 | }; | |
709 | ||
710 | #undef SCHED_FEAT | |
711 | ||
34f3a814 | 712 | static int sched_feat_show(struct seq_file *m, void *v) |
f00b45c1 | 713 | { |
f00b45c1 PZ |
714 | int i; |
715 | ||
716 | for (i = 0; sched_feat_names[i]; i++) { | |
34f3a814 LZ |
717 | if (!(sysctl_sched_features & (1UL << i))) |
718 | seq_puts(m, "NO_"); | |
719 | seq_printf(m, "%s ", sched_feat_names[i]); | |
f00b45c1 | 720 | } |
34f3a814 | 721 | seq_puts(m, "\n"); |
f00b45c1 | 722 | |
34f3a814 | 723 | return 0; |
f00b45c1 PZ |
724 | } |
725 | ||
726 | static ssize_t | |
727 | sched_feat_write(struct file *filp, const char __user *ubuf, | |
728 | size_t cnt, loff_t *ppos) | |
729 | { | |
730 | char buf[64]; | |
7740191c | 731 | char *cmp; |
f00b45c1 PZ |
732 | int neg = 0; |
733 | int i; | |
734 | ||
735 | if (cnt > 63) | |
736 | cnt = 63; | |
737 | ||
738 | if (copy_from_user(&buf, ubuf, cnt)) | |
739 | return -EFAULT; | |
740 | ||
741 | buf[cnt] = 0; | |
7740191c | 742 | cmp = strstrip(buf); |
f00b45c1 | 743 | |
524429c3 | 744 | if (strncmp(cmp, "NO_", 3) == 0) { |
f00b45c1 PZ |
745 | neg = 1; |
746 | cmp += 3; | |
747 | } | |
748 | ||
749 | for (i = 0; sched_feat_names[i]; i++) { | |
7740191c | 750 | if (strcmp(cmp, sched_feat_names[i]) == 0) { |
f00b45c1 PZ |
751 | if (neg) |
752 | sysctl_sched_features &= ~(1UL << i); | |
753 | else | |
754 | sysctl_sched_features |= (1UL << i); | |
755 | break; | |
756 | } | |
757 | } | |
758 | ||
759 | if (!sched_feat_names[i]) | |
760 | return -EINVAL; | |
761 | ||
42994724 | 762 | *ppos += cnt; |
f00b45c1 PZ |
763 | |
764 | return cnt; | |
765 | } | |
766 | ||
34f3a814 LZ |
767 | static int sched_feat_open(struct inode *inode, struct file *filp) |
768 | { | |
769 | return single_open(filp, sched_feat_show, NULL); | |
770 | } | |
771 | ||
828c0950 | 772 | static const struct file_operations sched_feat_fops = { |
34f3a814 LZ |
773 | .open = sched_feat_open, |
774 | .write = sched_feat_write, | |
775 | .read = seq_read, | |
776 | .llseek = seq_lseek, | |
777 | .release = single_release, | |
f00b45c1 PZ |
778 | }; |
779 | ||
780 | static __init int sched_init_debug(void) | |
781 | { | |
f00b45c1 PZ |
782 | debugfs_create_file("sched_features", 0644, NULL, NULL, |
783 | &sched_feat_fops); | |
784 | ||
785 | return 0; | |
786 | } | |
787 | late_initcall(sched_init_debug); | |
788 | ||
789 | #endif | |
790 | ||
791 | #define sched_feat(x) (sysctl_sched_features & (1UL << __SCHED_FEAT_##x)) | |
bf5c91ba | 792 | |
b82d9fdd PZ |
793 | /* |
794 | * Number of tasks to iterate in a single balance run. | |
795 | * Limited because this is done with IRQs disabled. | |
796 | */ | |
797 | const_debug unsigned int sysctl_sched_nr_migrate = 32; | |
798 | ||
e9e9250b PZ |
799 | /* |
800 | * period over which we average the RT time consumption, measured | |
801 | * in ms. | |
802 | * | |
803 | * default: 1s | |
804 | */ | |
805 | const_debug unsigned int sysctl_sched_time_avg = MSEC_PER_SEC; | |
806 | ||
fa85ae24 | 807 | /* |
9f0c1e56 | 808 | * period over which we measure -rt task cpu usage in us. |
fa85ae24 PZ |
809 | * default: 1s |
810 | */ | |
9f0c1e56 | 811 | unsigned int sysctl_sched_rt_period = 1000000; |
fa85ae24 | 812 | |
6892b75e IM |
813 | static __read_mostly int scheduler_running; |
814 | ||
9f0c1e56 PZ |
815 | /* |
816 | * part of the period that we allow rt tasks to run in us. | |
817 | * default: 0.95s | |
818 | */ | |
819 | int sysctl_sched_rt_runtime = 950000; | |
fa85ae24 | 820 | |
d0b27fa7 PZ |
821 | static inline u64 global_rt_period(void) |
822 | { | |
823 | return (u64)sysctl_sched_rt_period * NSEC_PER_USEC; | |
824 | } | |
825 | ||
826 | static inline u64 global_rt_runtime(void) | |
827 | { | |
e26873bb | 828 | if (sysctl_sched_rt_runtime < 0) |
d0b27fa7 PZ |
829 | return RUNTIME_INF; |
830 | ||
831 | return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC; | |
832 | } | |
fa85ae24 | 833 | |
1da177e4 | 834 | #ifndef prepare_arch_switch |
4866cde0 NP |
835 | # define prepare_arch_switch(next) do { } while (0) |
836 | #endif | |
837 | #ifndef finish_arch_switch | |
838 | # define finish_arch_switch(prev) do { } while (0) | |
839 | #endif | |
840 | ||
051a1d1a DA |
841 | static inline int task_current(struct rq *rq, struct task_struct *p) |
842 | { | |
843 | return rq->curr == p; | |
844 | } | |
845 | ||
4866cde0 | 846 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW |
70b97a7f | 847 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 | 848 | { |
051a1d1a | 849 | return task_current(rq, p); |
4866cde0 NP |
850 | } |
851 | ||
70b97a7f | 852 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
853 | { |
854 | } | |
855 | ||
70b97a7f | 856 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 | 857 | { |
da04c035 IM |
858 | #ifdef CONFIG_DEBUG_SPINLOCK |
859 | /* this is a valid case when another task releases the spinlock */ | |
860 | rq->lock.owner = current; | |
861 | #endif | |
8a25d5de IM |
862 | /* |
863 | * If we are tracking spinlock dependencies then we have to | |
864 | * fix up the runqueue lock - which gets 'carried over' from | |
865 | * prev into current: | |
866 | */ | |
867 | spin_acquire(&rq->lock.dep_map, 0, 0, _THIS_IP_); | |
868 | ||
05fa785c | 869 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 NP |
870 | } |
871 | ||
872 | #else /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
70b97a7f | 873 | static inline int task_running(struct rq *rq, struct task_struct *p) |
4866cde0 NP |
874 | { |
875 | #ifdef CONFIG_SMP | |
876 | return p->oncpu; | |
877 | #else | |
051a1d1a | 878 | return task_current(rq, p); |
4866cde0 NP |
879 | #endif |
880 | } | |
881 | ||
70b97a7f | 882 | static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next) |
4866cde0 NP |
883 | { |
884 | #ifdef CONFIG_SMP | |
885 | /* | |
886 | * We can optimise this out completely for !SMP, because the | |
887 | * SMP rebalancing from interrupt is the only thing that cares | |
888 | * here. | |
889 | */ | |
890 | next->oncpu = 1; | |
891 | #endif | |
892 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
05fa785c | 893 | raw_spin_unlock_irq(&rq->lock); |
4866cde0 | 894 | #else |
05fa785c | 895 | raw_spin_unlock(&rq->lock); |
4866cde0 NP |
896 | #endif |
897 | } | |
898 | ||
70b97a7f | 899 | static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) |
4866cde0 NP |
900 | { |
901 | #ifdef CONFIG_SMP | |
902 | /* | |
903 | * After ->oncpu is cleared, the task can be moved to a different CPU. | |
904 | * We must ensure this doesn't happen until the switch is completely | |
905 | * finished. | |
906 | */ | |
907 | smp_wmb(); | |
908 | prev->oncpu = 0; | |
909 | #endif | |
910 | #ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW | |
911 | local_irq_enable(); | |
1da177e4 | 912 | #endif |
4866cde0 NP |
913 | } |
914 | #endif /* __ARCH_WANT_UNLOCKED_CTXSW */ | |
1da177e4 | 915 | |
0970d299 | 916 | /* |
65cc8e48 PZ |
917 | * Check whether the task is waking, we use this to synchronize ->cpus_allowed |
918 | * against ttwu(). | |
0970d299 PZ |
919 | */ |
920 | static inline int task_is_waking(struct task_struct *p) | |
921 | { | |
0017d735 | 922 | return unlikely(p->state == TASK_WAKING); |
0970d299 PZ |
923 | } |
924 | ||
b29739f9 IM |
925 | /* |
926 | * __task_rq_lock - lock the runqueue a given task resides on. | |
927 | * Must be called interrupts disabled. | |
928 | */ | |
70b97a7f | 929 | static inline struct rq *__task_rq_lock(struct task_struct *p) |
b29739f9 IM |
930 | __acquires(rq->lock) |
931 | { | |
0970d299 PZ |
932 | struct rq *rq; |
933 | ||
3a5c359a | 934 | for (;;) { |
0970d299 | 935 | rq = task_rq(p); |
05fa785c | 936 | raw_spin_lock(&rq->lock); |
65cc8e48 | 937 | if (likely(rq == task_rq(p))) |
3a5c359a | 938 | return rq; |
05fa785c | 939 | raw_spin_unlock(&rq->lock); |
b29739f9 | 940 | } |
b29739f9 IM |
941 | } |
942 | ||
1da177e4 LT |
943 | /* |
944 | * task_rq_lock - lock the runqueue a given task resides on and disable | |
41a2d6cf | 945 | * interrupts. Note the ordering: we can safely lookup the task_rq without |
1da177e4 LT |
946 | * explicitly disabling preemption. |
947 | */ | |
70b97a7f | 948 | static struct rq *task_rq_lock(struct task_struct *p, unsigned long *flags) |
1da177e4 LT |
949 | __acquires(rq->lock) |
950 | { | |
70b97a7f | 951 | struct rq *rq; |
1da177e4 | 952 | |
3a5c359a AK |
953 | for (;;) { |
954 | local_irq_save(*flags); | |
955 | rq = task_rq(p); | |
05fa785c | 956 | raw_spin_lock(&rq->lock); |
65cc8e48 | 957 | if (likely(rq == task_rq(p))) |
3a5c359a | 958 | return rq; |
05fa785c | 959 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 | 960 | } |
1da177e4 LT |
961 | } |
962 | ||
a9957449 | 963 | static void __task_rq_unlock(struct rq *rq) |
b29739f9 IM |
964 | __releases(rq->lock) |
965 | { | |
05fa785c | 966 | raw_spin_unlock(&rq->lock); |
b29739f9 IM |
967 | } |
968 | ||
70b97a7f | 969 | static inline void task_rq_unlock(struct rq *rq, unsigned long *flags) |
1da177e4 LT |
970 | __releases(rq->lock) |
971 | { | |
05fa785c | 972 | raw_spin_unlock_irqrestore(&rq->lock, *flags); |
1da177e4 LT |
973 | } |
974 | ||
1da177e4 | 975 | /* |
cc2a73b5 | 976 | * this_rq_lock - lock this runqueue and disable interrupts. |
1da177e4 | 977 | */ |
a9957449 | 978 | static struct rq *this_rq_lock(void) |
1da177e4 LT |
979 | __acquires(rq->lock) |
980 | { | |
70b97a7f | 981 | struct rq *rq; |
1da177e4 LT |
982 | |
983 | local_irq_disable(); | |
984 | rq = this_rq(); | |
05fa785c | 985 | raw_spin_lock(&rq->lock); |
1da177e4 LT |
986 | |
987 | return rq; | |
988 | } | |
989 | ||
8f4d37ec PZ |
990 | #ifdef CONFIG_SCHED_HRTICK |
991 | /* | |
992 | * Use HR-timers to deliver accurate preemption points. | |
993 | * | |
994 | * Its all a bit involved since we cannot program an hrt while holding the | |
995 | * rq->lock. So what we do is store a state in in rq->hrtick_* and ask for a | |
996 | * reschedule event. | |
997 | * | |
998 | * When we get rescheduled we reprogram the hrtick_timer outside of the | |
999 | * rq->lock. | |
1000 | */ | |
8f4d37ec PZ |
1001 | |
1002 | /* | |
1003 | * Use hrtick when: | |
1004 | * - enabled by features | |
1005 | * - hrtimer is actually high res | |
1006 | */ | |
1007 | static inline int hrtick_enabled(struct rq *rq) | |
1008 | { | |
1009 | if (!sched_feat(HRTICK)) | |
1010 | return 0; | |
ba42059f | 1011 | if (!cpu_active(cpu_of(rq))) |
b328ca18 | 1012 | return 0; |
8f4d37ec PZ |
1013 | return hrtimer_is_hres_active(&rq->hrtick_timer); |
1014 | } | |
1015 | ||
8f4d37ec PZ |
1016 | static void hrtick_clear(struct rq *rq) |
1017 | { | |
1018 | if (hrtimer_active(&rq->hrtick_timer)) | |
1019 | hrtimer_cancel(&rq->hrtick_timer); | |
1020 | } | |
1021 | ||
8f4d37ec PZ |
1022 | /* |
1023 | * High-resolution timer tick. | |
1024 | * Runs from hardirq context with interrupts disabled. | |
1025 | */ | |
1026 | static enum hrtimer_restart hrtick(struct hrtimer *timer) | |
1027 | { | |
1028 | struct rq *rq = container_of(timer, struct rq, hrtick_timer); | |
1029 | ||
1030 | WARN_ON_ONCE(cpu_of(rq) != smp_processor_id()); | |
1031 | ||
05fa785c | 1032 | raw_spin_lock(&rq->lock); |
3e51f33f | 1033 | update_rq_clock(rq); |
8f4d37ec | 1034 | rq->curr->sched_class->task_tick(rq, rq->curr, 1); |
05fa785c | 1035 | raw_spin_unlock(&rq->lock); |
8f4d37ec PZ |
1036 | |
1037 | return HRTIMER_NORESTART; | |
1038 | } | |
1039 | ||
95e904c7 | 1040 | #ifdef CONFIG_SMP |
31656519 PZ |
1041 | /* |
1042 | * called from hardirq (IPI) context | |
1043 | */ | |
1044 | static void __hrtick_start(void *arg) | |
b328ca18 | 1045 | { |
31656519 | 1046 | struct rq *rq = arg; |
b328ca18 | 1047 | |
05fa785c | 1048 | raw_spin_lock(&rq->lock); |
31656519 PZ |
1049 | hrtimer_restart(&rq->hrtick_timer); |
1050 | rq->hrtick_csd_pending = 0; | |
05fa785c | 1051 | raw_spin_unlock(&rq->lock); |
b328ca18 PZ |
1052 | } |
1053 | ||
31656519 PZ |
1054 | /* |
1055 | * Called to set the hrtick timer state. | |
1056 | * | |
1057 | * called with rq->lock held and irqs disabled | |
1058 | */ | |
1059 | static void hrtick_start(struct rq *rq, u64 delay) | |
b328ca18 | 1060 | { |
31656519 PZ |
1061 | struct hrtimer *timer = &rq->hrtick_timer; |
1062 | ktime_t time = ktime_add_ns(timer->base->get_time(), delay); | |
b328ca18 | 1063 | |
cc584b21 | 1064 | hrtimer_set_expires(timer, time); |
31656519 PZ |
1065 | |
1066 | if (rq == this_rq()) { | |
1067 | hrtimer_restart(timer); | |
1068 | } else if (!rq->hrtick_csd_pending) { | |
6e275637 | 1069 | __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0); |
31656519 PZ |
1070 | rq->hrtick_csd_pending = 1; |
1071 | } | |
b328ca18 PZ |
1072 | } |
1073 | ||
1074 | static int | |
1075 | hotplug_hrtick(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1076 | { | |
1077 | int cpu = (int)(long)hcpu; | |
1078 | ||
1079 | switch (action) { | |
1080 | case CPU_UP_CANCELED: | |
1081 | case CPU_UP_CANCELED_FROZEN: | |
1082 | case CPU_DOWN_PREPARE: | |
1083 | case CPU_DOWN_PREPARE_FROZEN: | |
1084 | case CPU_DEAD: | |
1085 | case CPU_DEAD_FROZEN: | |
31656519 | 1086 | hrtick_clear(cpu_rq(cpu)); |
b328ca18 PZ |
1087 | return NOTIFY_OK; |
1088 | } | |
1089 | ||
1090 | return NOTIFY_DONE; | |
1091 | } | |
1092 | ||
fa748203 | 1093 | static __init void init_hrtick(void) |
b328ca18 PZ |
1094 | { |
1095 | hotcpu_notifier(hotplug_hrtick, 0); | |
1096 | } | |
31656519 PZ |
1097 | #else |
1098 | /* | |
1099 | * Called to set the hrtick timer state. | |
1100 | * | |
1101 | * called with rq->lock held and irqs disabled | |
1102 | */ | |
1103 | static void hrtick_start(struct rq *rq, u64 delay) | |
1104 | { | |
7f1e2ca9 | 1105 | __hrtimer_start_range_ns(&rq->hrtick_timer, ns_to_ktime(delay), 0, |
5c333864 | 1106 | HRTIMER_MODE_REL_PINNED, 0); |
31656519 | 1107 | } |
b328ca18 | 1108 | |
006c75f1 | 1109 | static inline void init_hrtick(void) |
8f4d37ec | 1110 | { |
8f4d37ec | 1111 | } |
31656519 | 1112 | #endif /* CONFIG_SMP */ |
8f4d37ec | 1113 | |
31656519 | 1114 | static void init_rq_hrtick(struct rq *rq) |
8f4d37ec | 1115 | { |
31656519 PZ |
1116 | #ifdef CONFIG_SMP |
1117 | rq->hrtick_csd_pending = 0; | |
8f4d37ec | 1118 | |
31656519 PZ |
1119 | rq->hrtick_csd.flags = 0; |
1120 | rq->hrtick_csd.func = __hrtick_start; | |
1121 | rq->hrtick_csd.info = rq; | |
1122 | #endif | |
8f4d37ec | 1123 | |
31656519 PZ |
1124 | hrtimer_init(&rq->hrtick_timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); |
1125 | rq->hrtick_timer.function = hrtick; | |
8f4d37ec | 1126 | } |
006c75f1 | 1127 | #else /* CONFIG_SCHED_HRTICK */ |
8f4d37ec PZ |
1128 | static inline void hrtick_clear(struct rq *rq) |
1129 | { | |
1130 | } | |
1131 | ||
8f4d37ec PZ |
1132 | static inline void init_rq_hrtick(struct rq *rq) |
1133 | { | |
1134 | } | |
1135 | ||
b328ca18 PZ |
1136 | static inline void init_hrtick(void) |
1137 | { | |
1138 | } | |
006c75f1 | 1139 | #endif /* CONFIG_SCHED_HRTICK */ |
8f4d37ec | 1140 | |
c24d20db IM |
1141 | /* |
1142 | * resched_task - mark a task 'to be rescheduled now'. | |
1143 | * | |
1144 | * On UP this means the setting of the need_resched flag, on SMP it | |
1145 | * might also involve a cross-CPU call to trigger the scheduler on | |
1146 | * the target CPU. | |
1147 | */ | |
1148 | #ifdef CONFIG_SMP | |
1149 | ||
1150 | #ifndef tsk_is_polling | |
1151 | #define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) | |
1152 | #endif | |
1153 | ||
31656519 | 1154 | static void resched_task(struct task_struct *p) |
c24d20db IM |
1155 | { |
1156 | int cpu; | |
1157 | ||
05fa785c | 1158 | assert_raw_spin_locked(&task_rq(p)->lock); |
c24d20db | 1159 | |
5ed0cec0 | 1160 | if (test_tsk_need_resched(p)) |
c24d20db IM |
1161 | return; |
1162 | ||
5ed0cec0 | 1163 | set_tsk_need_resched(p); |
c24d20db IM |
1164 | |
1165 | cpu = task_cpu(p); | |
1166 | if (cpu == smp_processor_id()) | |
1167 | return; | |
1168 | ||
1169 | /* NEED_RESCHED must be visible before we test polling */ | |
1170 | smp_mb(); | |
1171 | if (!tsk_is_polling(p)) | |
1172 | smp_send_reschedule(cpu); | |
1173 | } | |
1174 | ||
1175 | static void resched_cpu(int cpu) | |
1176 | { | |
1177 | struct rq *rq = cpu_rq(cpu); | |
1178 | unsigned long flags; | |
1179 | ||
05fa785c | 1180 | if (!raw_spin_trylock_irqsave(&rq->lock, flags)) |
c24d20db IM |
1181 | return; |
1182 | resched_task(cpu_curr(cpu)); | |
05fa785c | 1183 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
c24d20db | 1184 | } |
06d8308c TG |
1185 | |
1186 | #ifdef CONFIG_NO_HZ | |
83cd4fe2 VP |
1187 | /* |
1188 | * In the semi idle case, use the nearest busy cpu for migrating timers | |
1189 | * from an idle cpu. This is good for power-savings. | |
1190 | * | |
1191 | * We don't do similar optimization for completely idle system, as | |
1192 | * selecting an idle cpu will add more delays to the timers than intended | |
1193 | * (as that cpu's timer base may not be uptodate wrt jiffies etc). | |
1194 | */ | |
1195 | int get_nohz_timer_target(void) | |
1196 | { | |
1197 | int cpu = smp_processor_id(); | |
1198 | int i; | |
1199 | struct sched_domain *sd; | |
1200 | ||
1201 | for_each_domain(cpu, sd) { | |
1202 | for_each_cpu(i, sched_domain_span(sd)) | |
1203 | if (!idle_cpu(i)) | |
1204 | return i; | |
1205 | } | |
1206 | return cpu; | |
1207 | } | |
06d8308c TG |
1208 | /* |
1209 | * When add_timer_on() enqueues a timer into the timer wheel of an | |
1210 | * idle CPU then this timer might expire before the next timer event | |
1211 | * which is scheduled to wake up that CPU. In case of a completely | |
1212 | * idle system the next event might even be infinite time into the | |
1213 | * future. wake_up_idle_cpu() ensures that the CPU is woken up and | |
1214 | * leaves the inner idle loop so the newly added timer is taken into | |
1215 | * account when the CPU goes back to idle and evaluates the timer | |
1216 | * wheel for the next timer event. | |
1217 | */ | |
1218 | void wake_up_idle_cpu(int cpu) | |
1219 | { | |
1220 | struct rq *rq = cpu_rq(cpu); | |
1221 | ||
1222 | if (cpu == smp_processor_id()) | |
1223 | return; | |
1224 | ||
1225 | /* | |
1226 | * This is safe, as this function is called with the timer | |
1227 | * wheel base lock of (cpu) held. When the CPU is on the way | |
1228 | * to idle and has not yet set rq->curr to idle then it will | |
1229 | * be serialized on the timer wheel base lock and take the new | |
1230 | * timer into account automatically. | |
1231 | */ | |
1232 | if (rq->curr != rq->idle) | |
1233 | return; | |
1234 | ||
1235 | /* | |
1236 | * We can set TIF_RESCHED on the idle task of the other CPU | |
1237 | * lockless. The worst case is that the other CPU runs the | |
1238 | * idle task through an additional NOOP schedule() | |
1239 | */ | |
5ed0cec0 | 1240 | set_tsk_need_resched(rq->idle); |
06d8308c TG |
1241 | |
1242 | /* NEED_RESCHED must be visible before we test polling */ | |
1243 | smp_mb(); | |
1244 | if (!tsk_is_polling(rq->idle)) | |
1245 | smp_send_reschedule(cpu); | |
1246 | } | |
39c0cbe2 | 1247 | |
6d6bc0ad | 1248 | #endif /* CONFIG_NO_HZ */ |
06d8308c | 1249 | |
e9e9250b PZ |
1250 | static u64 sched_avg_period(void) |
1251 | { | |
1252 | return (u64)sysctl_sched_time_avg * NSEC_PER_MSEC / 2; | |
1253 | } | |
1254 | ||
1255 | static void sched_avg_update(struct rq *rq) | |
1256 | { | |
1257 | s64 period = sched_avg_period(); | |
1258 | ||
1259 | while ((s64)(rq->clock - rq->age_stamp) > period) { | |
0d98bb26 WD |
1260 | /* |
1261 | * Inline assembly required to prevent the compiler | |
1262 | * optimising this loop into a divmod call. | |
1263 | * See __iter_div_u64_rem() for another example of this. | |
1264 | */ | |
1265 | asm("" : "+rm" (rq->age_stamp)); | |
e9e9250b PZ |
1266 | rq->age_stamp += period; |
1267 | rq->rt_avg /= 2; | |
1268 | } | |
1269 | } | |
1270 | ||
1271 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1272 | { | |
1273 | rq->rt_avg += rt_delta; | |
1274 | sched_avg_update(rq); | |
1275 | } | |
1276 | ||
6d6bc0ad | 1277 | #else /* !CONFIG_SMP */ |
31656519 | 1278 | static void resched_task(struct task_struct *p) |
c24d20db | 1279 | { |
05fa785c | 1280 | assert_raw_spin_locked(&task_rq(p)->lock); |
31656519 | 1281 | set_tsk_need_resched(p); |
c24d20db | 1282 | } |
e9e9250b PZ |
1283 | |
1284 | static void sched_rt_avg_update(struct rq *rq, u64 rt_delta) | |
1285 | { | |
1286 | } | |
da2b71ed SS |
1287 | |
1288 | static void sched_avg_update(struct rq *rq) | |
1289 | { | |
1290 | } | |
6d6bc0ad | 1291 | #endif /* CONFIG_SMP */ |
c24d20db | 1292 | |
45bf76df IM |
1293 | #if BITS_PER_LONG == 32 |
1294 | # define WMULT_CONST (~0UL) | |
1295 | #else | |
1296 | # define WMULT_CONST (1UL << 32) | |
1297 | #endif | |
1298 | ||
1299 | #define WMULT_SHIFT 32 | |
1300 | ||
194081eb IM |
1301 | /* |
1302 | * Shift right and round: | |
1303 | */ | |
cf2ab469 | 1304 | #define SRR(x, y) (((x) + (1UL << ((y) - 1))) >> (y)) |
194081eb | 1305 | |
a7be37ac PZ |
1306 | /* |
1307 | * delta *= weight / lw | |
1308 | */ | |
cb1c4fc9 | 1309 | static unsigned long |
45bf76df IM |
1310 | calc_delta_mine(unsigned long delta_exec, unsigned long weight, |
1311 | struct load_weight *lw) | |
1312 | { | |
1313 | u64 tmp; | |
1314 | ||
7a232e03 LJ |
1315 | if (!lw->inv_weight) { |
1316 | if (BITS_PER_LONG > 32 && unlikely(lw->weight >= WMULT_CONST)) | |
1317 | lw->inv_weight = 1; | |
1318 | else | |
1319 | lw->inv_weight = 1 + (WMULT_CONST-lw->weight/2) | |
1320 | / (lw->weight+1); | |
1321 | } | |
45bf76df IM |
1322 | |
1323 | tmp = (u64)delta_exec * weight; | |
1324 | /* | |
1325 | * Check whether we'd overflow the 64-bit multiplication: | |
1326 | */ | |
194081eb | 1327 | if (unlikely(tmp > WMULT_CONST)) |
cf2ab469 | 1328 | tmp = SRR(SRR(tmp, WMULT_SHIFT/2) * lw->inv_weight, |
194081eb IM |
1329 | WMULT_SHIFT/2); |
1330 | else | |
cf2ab469 | 1331 | tmp = SRR(tmp * lw->inv_weight, WMULT_SHIFT); |
45bf76df | 1332 | |
ecf691da | 1333 | return (unsigned long)min(tmp, (u64)(unsigned long)LONG_MAX); |
45bf76df IM |
1334 | } |
1335 | ||
1091985b | 1336 | static inline void update_load_add(struct load_weight *lw, unsigned long inc) |
45bf76df IM |
1337 | { |
1338 | lw->weight += inc; | |
e89996ae | 1339 | lw->inv_weight = 0; |
45bf76df IM |
1340 | } |
1341 | ||
1091985b | 1342 | static inline void update_load_sub(struct load_weight *lw, unsigned long dec) |
45bf76df IM |
1343 | { |
1344 | lw->weight -= dec; | |
e89996ae | 1345 | lw->inv_weight = 0; |
45bf76df IM |
1346 | } |
1347 | ||
2069dd75 PZ |
1348 | static inline void update_load_set(struct load_weight *lw, unsigned long w) |
1349 | { | |
1350 | lw->weight = w; | |
1351 | lw->inv_weight = 0; | |
1352 | } | |
1353 | ||
2dd73a4f PW |
1354 | /* |
1355 | * To aid in avoiding the subversion of "niceness" due to uneven distribution | |
1356 | * of tasks with abnormal "nice" values across CPUs the contribution that | |
1357 | * each task makes to its run queue's load is weighted according to its | |
41a2d6cf | 1358 | * scheduling class and "nice" value. For SCHED_NORMAL tasks this is just a |
2dd73a4f PW |
1359 | * scaled version of the new time slice allocation that they receive on time |
1360 | * slice expiry etc. | |
1361 | */ | |
1362 | ||
cce7ade8 PZ |
1363 | #define WEIGHT_IDLEPRIO 3 |
1364 | #define WMULT_IDLEPRIO 1431655765 | |
dd41f596 IM |
1365 | |
1366 | /* | |
1367 | * Nice levels are multiplicative, with a gentle 10% change for every | |
1368 | * nice level changed. I.e. when a CPU-bound task goes from nice 0 to | |
1369 | * nice 1, it will get ~10% less CPU time than another CPU-bound task | |
1370 | * that remained on nice 0. | |
1371 | * | |
1372 | * The "10% effect" is relative and cumulative: from _any_ nice level, | |
1373 | * if you go up 1 level, it's -10% CPU usage, if you go down 1 level | |
f9153ee6 IM |
1374 | * it's +10% CPU usage. (to achieve that we use a multiplier of 1.25. |
1375 | * If a task goes up by ~10% and another task goes down by ~10% then | |
1376 | * the relative distance between them is ~25%.) | |
dd41f596 IM |
1377 | */ |
1378 | static const int prio_to_weight[40] = { | |
254753dc IM |
1379 | /* -20 */ 88761, 71755, 56483, 46273, 36291, |
1380 | /* -15 */ 29154, 23254, 18705, 14949, 11916, | |
1381 | /* -10 */ 9548, 7620, 6100, 4904, 3906, | |
1382 | /* -5 */ 3121, 2501, 1991, 1586, 1277, | |
1383 | /* 0 */ 1024, 820, 655, 526, 423, | |
1384 | /* 5 */ 335, 272, 215, 172, 137, | |
1385 | /* 10 */ 110, 87, 70, 56, 45, | |
1386 | /* 15 */ 36, 29, 23, 18, 15, | |
dd41f596 IM |
1387 | }; |
1388 | ||
5714d2de IM |
1389 | /* |
1390 | * Inverse (2^32/x) values of the prio_to_weight[] array, precalculated. | |
1391 | * | |
1392 | * In cases where the weight does not change often, we can use the | |
1393 | * precalculated inverse to speed up arithmetics by turning divisions | |
1394 | * into multiplications: | |
1395 | */ | |
dd41f596 | 1396 | static const u32 prio_to_wmult[40] = { |
254753dc IM |
1397 | /* -20 */ 48388, 59856, 76040, 92818, 118348, |
1398 | /* -15 */ 147320, 184698, 229616, 287308, 360437, | |
1399 | /* -10 */ 449829, 563644, 704093, 875809, 1099582, | |
1400 | /* -5 */ 1376151, 1717300, 2157191, 2708050, 3363326, | |
1401 | /* 0 */ 4194304, 5237765, 6557202, 8165337, 10153587, | |
1402 | /* 5 */ 12820798, 15790321, 19976592, 24970740, 31350126, | |
1403 | /* 10 */ 39045157, 49367440, 61356676, 76695844, 95443717, | |
1404 | /* 15 */ 119304647, 148102320, 186737708, 238609294, 286331153, | |
dd41f596 | 1405 | }; |
2dd73a4f | 1406 | |
ef12fefa BR |
1407 | /* Time spent by the tasks of the cpu accounting group executing in ... */ |
1408 | enum cpuacct_stat_index { | |
1409 | CPUACCT_STAT_USER, /* ... user mode */ | |
1410 | CPUACCT_STAT_SYSTEM, /* ... kernel mode */ | |
1411 | ||
1412 | CPUACCT_STAT_NSTATS, | |
1413 | }; | |
1414 | ||
d842de87 SV |
1415 | #ifdef CONFIG_CGROUP_CPUACCT |
1416 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime); | |
ef12fefa BR |
1417 | static void cpuacct_update_stats(struct task_struct *tsk, |
1418 | enum cpuacct_stat_index idx, cputime_t val); | |
d842de87 SV |
1419 | #else |
1420 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} | |
ef12fefa BR |
1421 | static inline void cpuacct_update_stats(struct task_struct *tsk, |
1422 | enum cpuacct_stat_index idx, cputime_t val) {} | |
d842de87 SV |
1423 | #endif |
1424 | ||
18d95a28 PZ |
1425 | static inline void inc_cpu_load(struct rq *rq, unsigned long load) |
1426 | { | |
1427 | update_load_add(&rq->load, load); | |
1428 | } | |
1429 | ||
1430 | static inline void dec_cpu_load(struct rq *rq, unsigned long load) | |
1431 | { | |
1432 | update_load_sub(&rq->load, load); | |
1433 | } | |
1434 | ||
7940ca36 | 1435 | #if (defined(CONFIG_SMP) && defined(CONFIG_FAIR_GROUP_SCHED)) || defined(CONFIG_RT_GROUP_SCHED) |
eb755805 | 1436 | typedef int (*tg_visitor)(struct task_group *, void *); |
c09595f6 PZ |
1437 | |
1438 | /* | |
1439 | * Iterate the full tree, calling @down when first entering a node and @up when | |
1440 | * leaving it for the final time. | |
1441 | */ | |
eb755805 | 1442 | static int walk_tg_tree(tg_visitor down, tg_visitor up, void *data) |
c09595f6 PZ |
1443 | { |
1444 | struct task_group *parent, *child; | |
eb755805 | 1445 | int ret; |
c09595f6 PZ |
1446 | |
1447 | rcu_read_lock(); | |
1448 | parent = &root_task_group; | |
1449 | down: | |
eb755805 PZ |
1450 | ret = (*down)(parent, data); |
1451 | if (ret) | |
1452 | goto out_unlock; | |
c09595f6 PZ |
1453 | list_for_each_entry_rcu(child, &parent->children, siblings) { |
1454 | parent = child; | |
1455 | goto down; | |
1456 | ||
1457 | up: | |
1458 | continue; | |
1459 | } | |
eb755805 PZ |
1460 | ret = (*up)(parent, data); |
1461 | if (ret) | |
1462 | goto out_unlock; | |
c09595f6 PZ |
1463 | |
1464 | child = parent; | |
1465 | parent = parent->parent; | |
1466 | if (parent) | |
1467 | goto up; | |
eb755805 | 1468 | out_unlock: |
c09595f6 | 1469 | rcu_read_unlock(); |
eb755805 PZ |
1470 | |
1471 | return ret; | |
c09595f6 PZ |
1472 | } |
1473 | ||
eb755805 PZ |
1474 | static int tg_nop(struct task_group *tg, void *data) |
1475 | { | |
1476 | return 0; | |
c09595f6 | 1477 | } |
eb755805 PZ |
1478 | #endif |
1479 | ||
1480 | #ifdef CONFIG_SMP | |
f5f08f39 PZ |
1481 | /* Used instead of source_load when we know the type == 0 */ |
1482 | static unsigned long weighted_cpuload(const int cpu) | |
1483 | { | |
1484 | return cpu_rq(cpu)->load.weight; | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * Return a low guess at the load of a migration-source cpu weighted | |
1489 | * according to the scheduling class and "nice" value. | |
1490 | * | |
1491 | * We want to under-estimate the load of migration sources, to | |
1492 | * balance conservatively. | |
1493 | */ | |
1494 | static unsigned long source_load(int cpu, int type) | |
1495 | { | |
1496 | struct rq *rq = cpu_rq(cpu); | |
1497 | unsigned long total = weighted_cpuload(cpu); | |
1498 | ||
1499 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1500 | return total; | |
1501 | ||
1502 | return min(rq->cpu_load[type-1], total); | |
1503 | } | |
1504 | ||
1505 | /* | |
1506 | * Return a high guess at the load of a migration-target cpu weighted | |
1507 | * according to the scheduling class and "nice" value. | |
1508 | */ | |
1509 | static unsigned long target_load(int cpu, int type) | |
1510 | { | |
1511 | struct rq *rq = cpu_rq(cpu); | |
1512 | unsigned long total = weighted_cpuload(cpu); | |
1513 | ||
1514 | if (type == 0 || !sched_feat(LB_BIAS)) | |
1515 | return total; | |
1516 | ||
1517 | return max(rq->cpu_load[type-1], total); | |
1518 | } | |
1519 | ||
ae154be1 PZ |
1520 | static unsigned long power_of(int cpu) |
1521 | { | |
e51fd5e2 | 1522 | return cpu_rq(cpu)->cpu_power; |
ae154be1 PZ |
1523 | } |
1524 | ||
eb755805 PZ |
1525 | static int task_hot(struct task_struct *p, u64 now, struct sched_domain *sd); |
1526 | ||
1527 | static unsigned long cpu_avg_load_per_task(int cpu) | |
1528 | { | |
1529 | struct rq *rq = cpu_rq(cpu); | |
af6d596f | 1530 | unsigned long nr_running = ACCESS_ONCE(rq->nr_running); |
eb755805 | 1531 | |
4cd42620 SR |
1532 | if (nr_running) |
1533 | rq->avg_load_per_task = rq->load.weight / nr_running; | |
a2d47777 BS |
1534 | else |
1535 | rq->avg_load_per_task = 0; | |
eb755805 PZ |
1536 | |
1537 | return rq->avg_load_per_task; | |
1538 | } | |
1539 | ||
1540 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
c09595f6 | 1541 | |
c09595f6 | 1542 | /* |
c8cba857 PZ |
1543 | * Compute the cpu's hierarchical load factor for each task group. |
1544 | * This needs to be done in a top-down fashion because the load of a child | |
1545 | * group is a fraction of its parents load. | |
c09595f6 | 1546 | */ |
eb755805 | 1547 | static int tg_load_down(struct task_group *tg, void *data) |
c09595f6 | 1548 | { |
c8cba857 | 1549 | unsigned long load; |
eb755805 | 1550 | long cpu = (long)data; |
c09595f6 | 1551 | |
c8cba857 PZ |
1552 | if (!tg->parent) { |
1553 | load = cpu_rq(cpu)->load.weight; | |
1554 | } else { | |
1555 | load = tg->parent->cfs_rq[cpu]->h_load; | |
2069dd75 | 1556 | load *= tg->se[cpu]->load.weight; |
c8cba857 PZ |
1557 | load /= tg->parent->cfs_rq[cpu]->load.weight + 1; |
1558 | } | |
c09595f6 | 1559 | |
c8cba857 | 1560 | tg->cfs_rq[cpu]->h_load = load; |
c09595f6 | 1561 | |
eb755805 | 1562 | return 0; |
c09595f6 PZ |
1563 | } |
1564 | ||
eb755805 | 1565 | static void update_h_load(long cpu) |
c09595f6 | 1566 | { |
eb755805 | 1567 | walk_tg_tree(tg_load_down, tg_nop, (void *)cpu); |
c09595f6 PZ |
1568 | } |
1569 | ||
18d95a28 PZ |
1570 | #endif |
1571 | ||
8f45e2b5 GH |
1572 | #ifdef CONFIG_PREEMPT |
1573 | ||
b78bb868 PZ |
1574 | static void double_rq_lock(struct rq *rq1, struct rq *rq2); |
1575 | ||
70574a99 | 1576 | /* |
8f45e2b5 GH |
1577 | * fair double_lock_balance: Safely acquires both rq->locks in a fair |
1578 | * way at the expense of forcing extra atomic operations in all | |
1579 | * invocations. This assures that the double_lock is acquired using the | |
1580 | * same underlying policy as the spinlock_t on this architecture, which | |
1581 | * reduces latency compared to the unfair variant below. However, it | |
1582 | * also adds more overhead and therefore may reduce throughput. | |
70574a99 | 1583 | */ |
8f45e2b5 GH |
1584 | static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest) |
1585 | __releases(this_rq->lock) | |
1586 | __acquires(busiest->lock) | |
1587 | __acquires(this_rq->lock) | |
1588 | { | |
05fa785c | 1589 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1590 | double_rq_lock(this_rq, busiest); |
1591 | ||
1592 | return 1; | |
1593 | } | |
1594 | ||
1595 | #else | |
1596 | /* | |
1597 | * Unfair double_lock_balance: Optimizes throughput at the expense of | |
1598 | * latency by eliminating extra atomic operations when the locks are | |
1599 | * already in proper order on entry. This favors lower cpu-ids and will | |
1600 | * grant the double lock to lower cpus over higher ids under contention, | |
1601 | * regardless of entry order into the function. | |
1602 | */ | |
1603 | static int _double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
70574a99 AD |
1604 | __releases(this_rq->lock) |
1605 | __acquires(busiest->lock) | |
1606 | __acquires(this_rq->lock) | |
1607 | { | |
1608 | int ret = 0; | |
1609 | ||
05fa785c | 1610 | if (unlikely(!raw_spin_trylock(&busiest->lock))) { |
70574a99 | 1611 | if (busiest < this_rq) { |
05fa785c TG |
1612 | raw_spin_unlock(&this_rq->lock); |
1613 | raw_spin_lock(&busiest->lock); | |
1614 | raw_spin_lock_nested(&this_rq->lock, | |
1615 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1616 | ret = 1; |
1617 | } else | |
05fa785c TG |
1618 | raw_spin_lock_nested(&busiest->lock, |
1619 | SINGLE_DEPTH_NESTING); | |
70574a99 AD |
1620 | } |
1621 | return ret; | |
1622 | } | |
1623 | ||
8f45e2b5 GH |
1624 | #endif /* CONFIG_PREEMPT */ |
1625 | ||
1626 | /* | |
1627 | * double_lock_balance - lock the busiest runqueue, this_rq is locked already. | |
1628 | */ | |
1629 | static int double_lock_balance(struct rq *this_rq, struct rq *busiest) | |
1630 | { | |
1631 | if (unlikely(!irqs_disabled())) { | |
1632 | /* printk() doesn't work good under rq->lock */ | |
05fa785c | 1633 | raw_spin_unlock(&this_rq->lock); |
8f45e2b5 GH |
1634 | BUG_ON(1); |
1635 | } | |
1636 | ||
1637 | return _double_lock_balance(this_rq, busiest); | |
1638 | } | |
1639 | ||
70574a99 AD |
1640 | static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) |
1641 | __releases(busiest->lock) | |
1642 | { | |
05fa785c | 1643 | raw_spin_unlock(&busiest->lock); |
70574a99 AD |
1644 | lock_set_subclass(&this_rq->lock.dep_map, 0, _RET_IP_); |
1645 | } | |
1e3c88bd PZ |
1646 | |
1647 | /* | |
1648 | * double_rq_lock - safely lock two runqueues | |
1649 | * | |
1650 | * Note this does not disable interrupts like task_rq_lock, | |
1651 | * you need to do so manually before calling. | |
1652 | */ | |
1653 | static void double_rq_lock(struct rq *rq1, struct rq *rq2) | |
1654 | __acquires(rq1->lock) | |
1655 | __acquires(rq2->lock) | |
1656 | { | |
1657 | BUG_ON(!irqs_disabled()); | |
1658 | if (rq1 == rq2) { | |
1659 | raw_spin_lock(&rq1->lock); | |
1660 | __acquire(rq2->lock); /* Fake it out ;) */ | |
1661 | } else { | |
1662 | if (rq1 < rq2) { | |
1663 | raw_spin_lock(&rq1->lock); | |
1664 | raw_spin_lock_nested(&rq2->lock, SINGLE_DEPTH_NESTING); | |
1665 | } else { | |
1666 | raw_spin_lock(&rq2->lock); | |
1667 | raw_spin_lock_nested(&rq1->lock, SINGLE_DEPTH_NESTING); | |
1668 | } | |
1669 | } | |
1e3c88bd PZ |
1670 | } |
1671 | ||
1672 | /* | |
1673 | * double_rq_unlock - safely unlock two runqueues | |
1674 | * | |
1675 | * Note this does not restore interrupts like task_rq_unlock, | |
1676 | * you need to do so manually after calling. | |
1677 | */ | |
1678 | static void double_rq_unlock(struct rq *rq1, struct rq *rq2) | |
1679 | __releases(rq1->lock) | |
1680 | __releases(rq2->lock) | |
1681 | { | |
1682 | raw_spin_unlock(&rq1->lock); | |
1683 | if (rq1 != rq2) | |
1684 | raw_spin_unlock(&rq2->lock); | |
1685 | else | |
1686 | __release(rq2->lock); | |
1687 | } | |
1688 | ||
18d95a28 PZ |
1689 | #endif |
1690 | ||
74f5187a | 1691 | static void calc_load_account_idle(struct rq *this_rq); |
0bcdcf28 | 1692 | static void update_sysctl(void); |
acb4a848 | 1693 | static int get_update_sysctl_factor(void); |
fdf3e95d | 1694 | static void update_cpu_load(struct rq *this_rq); |
dce48a84 | 1695 | |
cd29fe6f PZ |
1696 | static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu) |
1697 | { | |
1698 | set_task_rq(p, cpu); | |
1699 | #ifdef CONFIG_SMP | |
1700 | /* | |
1701 | * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be | |
1702 | * successfuly executed on another CPU. We must ensure that updates of | |
1703 | * per-task data have been completed by this moment. | |
1704 | */ | |
1705 | smp_wmb(); | |
1706 | task_thread_info(p)->cpu = cpu; | |
1707 | #endif | |
1708 | } | |
dce48a84 | 1709 | |
1e3c88bd | 1710 | static const struct sched_class rt_sched_class; |
dd41f596 | 1711 | |
34f971f6 | 1712 | #define sched_class_highest (&stop_sched_class) |
1f11eb6a GH |
1713 | #define for_each_class(class) \ |
1714 | for (class = sched_class_highest; class; class = class->next) | |
dd41f596 | 1715 | |
1e3c88bd PZ |
1716 | #include "sched_stats.h" |
1717 | ||
c09595f6 | 1718 | static void inc_nr_running(struct rq *rq) |
9c217245 IM |
1719 | { |
1720 | rq->nr_running++; | |
9c217245 IM |
1721 | } |
1722 | ||
c09595f6 | 1723 | static void dec_nr_running(struct rq *rq) |
9c217245 IM |
1724 | { |
1725 | rq->nr_running--; | |
9c217245 IM |
1726 | } |
1727 | ||
45bf76df IM |
1728 | static void set_load_weight(struct task_struct *p) |
1729 | { | |
dd41f596 IM |
1730 | /* |
1731 | * SCHED_IDLE tasks get minimal weight: | |
1732 | */ | |
1733 | if (p->policy == SCHED_IDLE) { | |
1734 | p->se.load.weight = WEIGHT_IDLEPRIO; | |
1735 | p->se.load.inv_weight = WMULT_IDLEPRIO; | |
1736 | return; | |
1737 | } | |
71f8bd46 | 1738 | |
dd41f596 IM |
1739 | p->se.load.weight = prio_to_weight[p->static_prio - MAX_RT_PRIO]; |
1740 | p->se.load.inv_weight = prio_to_wmult[p->static_prio - MAX_RT_PRIO]; | |
71f8bd46 IM |
1741 | } |
1742 | ||
371fd7e7 | 1743 | static void enqueue_task(struct rq *rq, struct task_struct *p, int flags) |
2087a1ad | 1744 | { |
a64692a3 | 1745 | update_rq_clock(rq); |
dd41f596 | 1746 | sched_info_queued(p); |
371fd7e7 | 1747 | p->sched_class->enqueue_task(rq, p, flags); |
dd41f596 | 1748 | p->se.on_rq = 1; |
71f8bd46 IM |
1749 | } |
1750 | ||
371fd7e7 | 1751 | static void dequeue_task(struct rq *rq, struct task_struct *p, int flags) |
71f8bd46 | 1752 | { |
a64692a3 | 1753 | update_rq_clock(rq); |
46ac22ba | 1754 | sched_info_dequeued(p); |
371fd7e7 | 1755 | p->sched_class->dequeue_task(rq, p, flags); |
dd41f596 | 1756 | p->se.on_rq = 0; |
71f8bd46 IM |
1757 | } |
1758 | ||
1e3c88bd PZ |
1759 | /* |
1760 | * activate_task - move a task to the runqueue. | |
1761 | */ | |
371fd7e7 | 1762 | static void activate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1763 | { |
1764 | if (task_contributes_to_load(p)) | |
1765 | rq->nr_uninterruptible--; | |
1766 | ||
371fd7e7 | 1767 | enqueue_task(rq, p, flags); |
1e3c88bd PZ |
1768 | inc_nr_running(rq); |
1769 | } | |
1770 | ||
1771 | /* | |
1772 | * deactivate_task - remove a task from the runqueue. | |
1773 | */ | |
371fd7e7 | 1774 | static void deactivate_task(struct rq *rq, struct task_struct *p, int flags) |
1e3c88bd PZ |
1775 | { |
1776 | if (task_contributes_to_load(p)) | |
1777 | rq->nr_uninterruptible++; | |
1778 | ||
371fd7e7 | 1779 | dequeue_task(rq, p, flags); |
1e3c88bd PZ |
1780 | dec_nr_running(rq); |
1781 | } | |
1782 | ||
b52bfee4 VP |
1783 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
1784 | ||
305e6835 VP |
1785 | /* |
1786 | * There are no locks covering percpu hardirq/softirq time. | |
1787 | * They are only modified in account_system_vtime, on corresponding CPU | |
1788 | * with interrupts disabled. So, writes are safe. | |
1789 | * They are read and saved off onto struct rq in update_rq_clock(). | |
1790 | * This may result in other CPU reading this CPU's irq time and can | |
1791 | * race with irq/account_system_vtime on this CPU. We would either get old | |
8e92c201 PZ |
1792 | * or new value with a side effect of accounting a slice of irq time to wrong |
1793 | * task when irq is in progress while we read rq->clock. That is a worthy | |
1794 | * compromise in place of having locks on each irq in account_system_time. | |
305e6835 | 1795 | */ |
b52bfee4 VP |
1796 | static DEFINE_PER_CPU(u64, cpu_hardirq_time); |
1797 | static DEFINE_PER_CPU(u64, cpu_softirq_time); | |
1798 | ||
1799 | static DEFINE_PER_CPU(u64, irq_start_time); | |
1800 | static int sched_clock_irqtime; | |
1801 | ||
1802 | void enable_sched_clock_irqtime(void) | |
1803 | { | |
1804 | sched_clock_irqtime = 1; | |
1805 | } | |
1806 | ||
1807 | void disable_sched_clock_irqtime(void) | |
1808 | { | |
1809 | sched_clock_irqtime = 0; | |
1810 | } | |
1811 | ||
8e92c201 PZ |
1812 | #ifndef CONFIG_64BIT |
1813 | static DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
1814 | ||
1815 | static inline void irq_time_write_begin(void) | |
1816 | { | |
1817 | __this_cpu_inc(irq_time_seq.sequence); | |
1818 | smp_wmb(); | |
1819 | } | |
1820 | ||
1821 | static inline void irq_time_write_end(void) | |
1822 | { | |
1823 | smp_wmb(); | |
1824 | __this_cpu_inc(irq_time_seq.sequence); | |
1825 | } | |
1826 | ||
1827 | static inline u64 irq_time_read(int cpu) | |
1828 | { | |
1829 | u64 irq_time; | |
1830 | unsigned seq; | |
1831 | ||
1832 | do { | |
1833 | seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu)); | |
1834 | irq_time = per_cpu(cpu_softirq_time, cpu) + | |
1835 | per_cpu(cpu_hardirq_time, cpu); | |
1836 | } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq)); | |
1837 | ||
1838 | return irq_time; | |
1839 | } | |
1840 | #else /* CONFIG_64BIT */ | |
1841 | static inline void irq_time_write_begin(void) | |
1842 | { | |
1843 | } | |
1844 | ||
1845 | static inline void irq_time_write_end(void) | |
1846 | { | |
1847 | } | |
1848 | ||
1849 | static inline u64 irq_time_read(int cpu) | |
305e6835 | 1850 | { |
305e6835 VP |
1851 | return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu); |
1852 | } | |
8e92c201 | 1853 | #endif /* CONFIG_64BIT */ |
305e6835 | 1854 | |
fe44d621 PZ |
1855 | /* |
1856 | * Called before incrementing preempt_count on {soft,}irq_enter | |
1857 | * and before decrementing preempt_count on {soft,}irq_exit. | |
1858 | */ | |
b52bfee4 VP |
1859 | void account_system_vtime(struct task_struct *curr) |
1860 | { | |
1861 | unsigned long flags; | |
fe44d621 | 1862 | s64 delta; |
b52bfee4 | 1863 | int cpu; |
b52bfee4 VP |
1864 | |
1865 | if (!sched_clock_irqtime) | |
1866 | return; | |
1867 | ||
1868 | local_irq_save(flags); | |
1869 | ||
b52bfee4 | 1870 | cpu = smp_processor_id(); |
fe44d621 PZ |
1871 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); |
1872 | __this_cpu_add(irq_start_time, delta); | |
1873 | ||
8e92c201 | 1874 | irq_time_write_begin(); |
b52bfee4 VP |
1875 | /* |
1876 | * We do not account for softirq time from ksoftirqd here. | |
1877 | * We want to continue accounting softirq time to ksoftirqd thread | |
1878 | * in that case, so as not to confuse scheduler with a special task | |
1879 | * that do not consume any time, but still wants to run. | |
1880 | */ | |
1881 | if (hardirq_count()) | |
fe44d621 | 1882 | __this_cpu_add(cpu_hardirq_time, delta); |
b52bfee4 | 1883 | else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD)) |
fe44d621 | 1884 | __this_cpu_add(cpu_softirq_time, delta); |
b52bfee4 | 1885 | |
8e92c201 | 1886 | irq_time_write_end(); |
b52bfee4 VP |
1887 | local_irq_restore(flags); |
1888 | } | |
b7dadc38 | 1889 | EXPORT_SYMBOL_GPL(account_system_vtime); |
b52bfee4 | 1890 | |
fe44d621 | 1891 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
aa483808 | 1892 | { |
fe44d621 PZ |
1893 | s64 irq_delta; |
1894 | ||
8e92c201 | 1895 | irq_delta = irq_time_read(cpu_of(rq)) - rq->prev_irq_time; |
fe44d621 PZ |
1896 | |
1897 | /* | |
1898 | * Since irq_time is only updated on {soft,}irq_exit, we might run into | |
1899 | * this case when a previous update_rq_clock() happened inside a | |
1900 | * {soft,}irq region. | |
1901 | * | |
1902 | * When this happens, we stop ->clock_task and only update the | |
1903 | * prev_irq_time stamp to account for the part that fit, so that a next | |
1904 | * update will consume the rest. This ensures ->clock_task is | |
1905 | * monotonic. | |
1906 | * | |
1907 | * It does however cause some slight miss-attribution of {soft,}irq | |
1908 | * time, a more accurate solution would be to update the irq_time using | |
1909 | * the current rq->clock timestamp, except that would require using | |
1910 | * atomic ops. | |
1911 | */ | |
1912 | if (irq_delta > delta) | |
1913 | irq_delta = delta; | |
1914 | ||
1915 | rq->prev_irq_time += irq_delta; | |
1916 | delta -= irq_delta; | |
1917 | rq->clock_task += delta; | |
1918 | ||
1919 | if (irq_delta && sched_feat(NONIRQ_POWER)) | |
1920 | sched_rt_avg_update(rq, irq_delta); | |
aa483808 VP |
1921 | } |
1922 | ||
fe44d621 | 1923 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ |
305e6835 | 1924 | |
fe44d621 | 1925 | static void update_rq_clock_task(struct rq *rq, s64 delta) |
305e6835 | 1926 | { |
fe44d621 | 1927 | rq->clock_task += delta; |
305e6835 VP |
1928 | } |
1929 | ||
fe44d621 | 1930 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ |
b52bfee4 | 1931 | |
1e3c88bd PZ |
1932 | #include "sched_idletask.c" |
1933 | #include "sched_fair.c" | |
1934 | #include "sched_rt.c" | |
5091faa4 | 1935 | #include "sched_autogroup.c" |
34f971f6 | 1936 | #include "sched_stoptask.c" |
1e3c88bd PZ |
1937 | #ifdef CONFIG_SCHED_DEBUG |
1938 | # include "sched_debug.c" | |
1939 | #endif | |
1940 | ||
34f971f6 PZ |
1941 | void sched_set_stop_task(int cpu, struct task_struct *stop) |
1942 | { | |
1943 | struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; | |
1944 | struct task_struct *old_stop = cpu_rq(cpu)->stop; | |
1945 | ||
1946 | if (stop) { | |
1947 | /* | |
1948 | * Make it appear like a SCHED_FIFO task, its something | |
1949 | * userspace knows about and won't get confused about. | |
1950 | * | |
1951 | * Also, it will make PI more or less work without too | |
1952 | * much confusion -- but then, stop work should not | |
1953 | * rely on PI working anyway. | |
1954 | */ | |
1955 | sched_setscheduler_nocheck(stop, SCHED_FIFO, ¶m); | |
1956 | ||
1957 | stop->sched_class = &stop_sched_class; | |
1958 | } | |
1959 | ||
1960 | cpu_rq(cpu)->stop = stop; | |
1961 | ||
1962 | if (old_stop) { | |
1963 | /* | |
1964 | * Reset it back to a normal scheduling class so that | |
1965 | * it can die in pieces. | |
1966 | */ | |
1967 | old_stop->sched_class = &rt_sched_class; | |
1968 | } | |
1969 | } | |
1970 | ||
14531189 | 1971 | /* |
dd41f596 | 1972 | * __normal_prio - return the priority that is based on the static prio |
14531189 | 1973 | */ |
14531189 IM |
1974 | static inline int __normal_prio(struct task_struct *p) |
1975 | { | |
dd41f596 | 1976 | return p->static_prio; |
14531189 IM |
1977 | } |
1978 | ||
b29739f9 IM |
1979 | /* |
1980 | * Calculate the expected normal priority: i.e. priority | |
1981 | * without taking RT-inheritance into account. Might be | |
1982 | * boosted by interactivity modifiers. Changes upon fork, | |
1983 | * setprio syscalls, and whenever the interactivity | |
1984 | * estimator recalculates. | |
1985 | */ | |
36c8b586 | 1986 | static inline int normal_prio(struct task_struct *p) |
b29739f9 IM |
1987 | { |
1988 | int prio; | |
1989 | ||
e05606d3 | 1990 | if (task_has_rt_policy(p)) |
b29739f9 IM |
1991 | prio = MAX_RT_PRIO-1 - p->rt_priority; |
1992 | else | |
1993 | prio = __normal_prio(p); | |
1994 | return prio; | |
1995 | } | |
1996 | ||
1997 | /* | |
1998 | * Calculate the current priority, i.e. the priority | |
1999 | * taken into account by the scheduler. This value might | |
2000 | * be boosted by RT tasks, or might be boosted by | |
2001 | * interactivity modifiers. Will be RT if the task got | |
2002 | * RT-boosted. If not then it returns p->normal_prio. | |
2003 | */ | |
36c8b586 | 2004 | static int effective_prio(struct task_struct *p) |
b29739f9 IM |
2005 | { |
2006 | p->normal_prio = normal_prio(p); | |
2007 | /* | |
2008 | * If we are RT tasks or we were boosted to RT priority, | |
2009 | * keep the priority unchanged. Otherwise, update priority | |
2010 | * to the normal priority: | |
2011 | */ | |
2012 | if (!rt_prio(p->prio)) | |
2013 | return p->normal_prio; | |
2014 | return p->prio; | |
2015 | } | |
2016 | ||
1da177e4 LT |
2017 | /** |
2018 | * task_curr - is this task currently executing on a CPU? | |
2019 | * @p: the task in question. | |
2020 | */ | |
36c8b586 | 2021 | inline int task_curr(const struct task_struct *p) |
1da177e4 LT |
2022 | { |
2023 | return cpu_curr(task_cpu(p)) == p; | |
2024 | } | |
2025 | ||
cb469845 SR |
2026 | static inline void check_class_changed(struct rq *rq, struct task_struct *p, |
2027 | const struct sched_class *prev_class, | |
2028 | int oldprio, int running) | |
2029 | { | |
2030 | if (prev_class != p->sched_class) { | |
2031 | if (prev_class->switched_from) | |
2032 | prev_class->switched_from(rq, p, running); | |
2033 | p->sched_class->switched_to(rq, p, running); | |
2034 | } else | |
2035 | p->sched_class->prio_changed(rq, p, oldprio, running); | |
2036 | } | |
2037 | ||
1e5a7405 PZ |
2038 | static void check_preempt_curr(struct rq *rq, struct task_struct *p, int flags) |
2039 | { | |
2040 | const struct sched_class *class; | |
2041 | ||
2042 | if (p->sched_class == rq->curr->sched_class) { | |
2043 | rq->curr->sched_class->check_preempt_curr(rq, p, flags); | |
2044 | } else { | |
2045 | for_each_class(class) { | |
2046 | if (class == rq->curr->sched_class) | |
2047 | break; | |
2048 | if (class == p->sched_class) { | |
2049 | resched_task(rq->curr); | |
2050 | break; | |
2051 | } | |
2052 | } | |
2053 | } | |
2054 | ||
2055 | /* | |
2056 | * A queue event has occurred, and we're going to schedule. In | |
2057 | * this case, we can save a useless back to back clock update. | |
2058 | */ | |
f26f9aff | 2059 | if (rq->curr->se.on_rq && test_tsk_need_resched(rq->curr)) |
1e5a7405 PZ |
2060 | rq->skip_clock_update = 1; |
2061 | } | |
2062 | ||
1da177e4 | 2063 | #ifdef CONFIG_SMP |
cc367732 IM |
2064 | /* |
2065 | * Is this task likely cache-hot: | |
2066 | */ | |
e7693a36 | 2067 | static int |
cc367732 IM |
2068 | task_hot(struct task_struct *p, u64 now, struct sched_domain *sd) |
2069 | { | |
2070 | s64 delta; | |
2071 | ||
e6c8fba7 PZ |
2072 | if (p->sched_class != &fair_sched_class) |
2073 | return 0; | |
2074 | ||
ef8002f6 NR |
2075 | if (unlikely(p->policy == SCHED_IDLE)) |
2076 | return 0; | |
2077 | ||
f540a608 IM |
2078 | /* |
2079 | * Buddy candidates are cache hot: | |
2080 | */ | |
f685ceac | 2081 | if (sched_feat(CACHE_HOT_BUDDY) && this_rq()->nr_running && |
4793241b PZ |
2082 | (&p->se == cfs_rq_of(&p->se)->next || |
2083 | &p->se == cfs_rq_of(&p->se)->last)) | |
f540a608 IM |
2084 | return 1; |
2085 | ||
6bc1665b IM |
2086 | if (sysctl_sched_migration_cost == -1) |
2087 | return 1; | |
2088 | if (sysctl_sched_migration_cost == 0) | |
2089 | return 0; | |
2090 | ||
cc367732 IM |
2091 | delta = now - p->se.exec_start; |
2092 | ||
2093 | return delta < (s64)sysctl_sched_migration_cost; | |
2094 | } | |
2095 | ||
dd41f596 | 2096 | void set_task_cpu(struct task_struct *p, unsigned int new_cpu) |
c65cc870 | 2097 | { |
e2912009 PZ |
2098 | #ifdef CONFIG_SCHED_DEBUG |
2099 | /* | |
2100 | * We should never call set_task_cpu() on a blocked task, | |
2101 | * ttwu() will sort out the placement. | |
2102 | */ | |
077614ee PZ |
2103 | WARN_ON_ONCE(p->state != TASK_RUNNING && p->state != TASK_WAKING && |
2104 | !(task_thread_info(p)->preempt_count & PREEMPT_ACTIVE)); | |
e2912009 PZ |
2105 | #endif |
2106 | ||
de1d7286 | 2107 | trace_sched_migrate_task(p, new_cpu); |
cbc34ed1 | 2108 | |
0c69774e PZ |
2109 | if (task_cpu(p) != new_cpu) { |
2110 | p->se.nr_migrations++; | |
2111 | perf_sw_event(PERF_COUNT_SW_CPU_MIGRATIONS, 1, 1, NULL, 0); | |
2112 | } | |
dd41f596 IM |
2113 | |
2114 | __set_task_cpu(p, new_cpu); | |
c65cc870 IM |
2115 | } |
2116 | ||
969c7921 | 2117 | struct migration_arg { |
36c8b586 | 2118 | struct task_struct *task; |
1da177e4 | 2119 | int dest_cpu; |
70b97a7f | 2120 | }; |
1da177e4 | 2121 | |
969c7921 TH |
2122 | static int migration_cpu_stop(void *data); |
2123 | ||
1da177e4 LT |
2124 | /* |
2125 | * The task's runqueue lock must be held. | |
2126 | * Returns true if you have to wait for migration thread. | |
2127 | */ | |
b7a2b39d | 2128 | static bool migrate_task(struct task_struct *p, struct rq *rq) |
1da177e4 | 2129 | { |
1da177e4 LT |
2130 | /* |
2131 | * If the task is not on a runqueue (and not running), then | |
e2912009 | 2132 | * the next wake-up will properly place the task. |
1da177e4 | 2133 | */ |
969c7921 | 2134 | return p->se.on_rq || task_running(rq, p); |
1da177e4 LT |
2135 | } |
2136 | ||
2137 | /* | |
2138 | * wait_task_inactive - wait for a thread to unschedule. | |
2139 | * | |
85ba2d86 RM |
2140 | * If @match_state is nonzero, it's the @p->state value just checked and |
2141 | * not expected to change. If it changes, i.e. @p might have woken up, | |
2142 | * then return zero. When we succeed in waiting for @p to be off its CPU, | |
2143 | * we return a positive number (its total switch count). If a second call | |
2144 | * a short while later returns the same number, the caller can be sure that | |
2145 | * @p has remained unscheduled the whole time. | |
2146 | * | |
1da177e4 LT |
2147 | * The caller must ensure that the task *will* unschedule sometime soon, |
2148 | * else this function might spin for a *long* time. This function can't | |
2149 | * be called with interrupts off, or it may introduce deadlock with | |
2150 | * smp_call_function() if an IPI is sent by the same process we are | |
2151 | * waiting to become inactive. | |
2152 | */ | |
85ba2d86 | 2153 | unsigned long wait_task_inactive(struct task_struct *p, long match_state) |
1da177e4 LT |
2154 | { |
2155 | unsigned long flags; | |
dd41f596 | 2156 | int running, on_rq; |
85ba2d86 | 2157 | unsigned long ncsw; |
70b97a7f | 2158 | struct rq *rq; |
1da177e4 | 2159 | |
3a5c359a AK |
2160 | for (;;) { |
2161 | /* | |
2162 | * We do the initial early heuristics without holding | |
2163 | * any task-queue locks at all. We'll only try to get | |
2164 | * the runqueue lock when things look like they will | |
2165 | * work out! | |
2166 | */ | |
2167 | rq = task_rq(p); | |
fa490cfd | 2168 | |
3a5c359a AK |
2169 | /* |
2170 | * If the task is actively running on another CPU | |
2171 | * still, just relax and busy-wait without holding | |
2172 | * any locks. | |
2173 | * | |
2174 | * NOTE! Since we don't hold any locks, it's not | |
2175 | * even sure that "rq" stays as the right runqueue! | |
2176 | * But we don't care, since "task_running()" will | |
2177 | * return false if the runqueue has changed and p | |
2178 | * is actually now running somewhere else! | |
2179 | */ | |
85ba2d86 RM |
2180 | while (task_running(rq, p)) { |
2181 | if (match_state && unlikely(p->state != match_state)) | |
2182 | return 0; | |
3a5c359a | 2183 | cpu_relax(); |
85ba2d86 | 2184 | } |
fa490cfd | 2185 | |
3a5c359a AK |
2186 | /* |
2187 | * Ok, time to look more closely! We need the rq | |
2188 | * lock now, to be *sure*. If we're wrong, we'll | |
2189 | * just go back and repeat. | |
2190 | */ | |
2191 | rq = task_rq_lock(p, &flags); | |
27a9da65 | 2192 | trace_sched_wait_task(p); |
3a5c359a AK |
2193 | running = task_running(rq, p); |
2194 | on_rq = p->se.on_rq; | |
85ba2d86 | 2195 | ncsw = 0; |
f31e11d8 | 2196 | if (!match_state || p->state == match_state) |
93dcf55f | 2197 | ncsw = p->nvcsw | LONG_MIN; /* sets MSB */ |
3a5c359a | 2198 | task_rq_unlock(rq, &flags); |
fa490cfd | 2199 | |
85ba2d86 RM |
2200 | /* |
2201 | * If it changed from the expected state, bail out now. | |
2202 | */ | |
2203 | if (unlikely(!ncsw)) | |
2204 | break; | |
2205 | ||
3a5c359a AK |
2206 | /* |
2207 | * Was it really running after all now that we | |
2208 | * checked with the proper locks actually held? | |
2209 | * | |
2210 | * Oops. Go back and try again.. | |
2211 | */ | |
2212 | if (unlikely(running)) { | |
2213 | cpu_relax(); | |
2214 | continue; | |
2215 | } | |
fa490cfd | 2216 | |
3a5c359a AK |
2217 | /* |
2218 | * It's not enough that it's not actively running, | |
2219 | * it must be off the runqueue _entirely_, and not | |
2220 | * preempted! | |
2221 | * | |
80dd99b3 | 2222 | * So if it was still runnable (but just not actively |
3a5c359a AK |
2223 | * running right now), it's preempted, and we should |
2224 | * yield - it could be a while. | |
2225 | */ | |
2226 | if (unlikely(on_rq)) { | |
2227 | schedule_timeout_uninterruptible(1); | |
2228 | continue; | |
2229 | } | |
fa490cfd | 2230 | |
3a5c359a AK |
2231 | /* |
2232 | * Ahh, all good. It wasn't running, and it wasn't | |
2233 | * runnable, which means that it will never become | |
2234 | * running in the future either. We're all done! | |
2235 | */ | |
2236 | break; | |
2237 | } | |
85ba2d86 RM |
2238 | |
2239 | return ncsw; | |
1da177e4 LT |
2240 | } |
2241 | ||
2242 | /*** | |
2243 | * kick_process - kick a running thread to enter/exit the kernel | |
2244 | * @p: the to-be-kicked thread | |
2245 | * | |
2246 | * Cause a process which is running on another CPU to enter | |
2247 | * kernel-mode, without any delay. (to get signals handled.) | |
2248 | * | |
2249 | * NOTE: this function doesnt have to take the runqueue lock, | |
2250 | * because all it wants to ensure is that the remote task enters | |
2251 | * the kernel. If the IPI races and the task has been migrated | |
2252 | * to another CPU then no harm is done and the purpose has been | |
2253 | * achieved as well. | |
2254 | */ | |
36c8b586 | 2255 | void kick_process(struct task_struct *p) |
1da177e4 LT |
2256 | { |
2257 | int cpu; | |
2258 | ||
2259 | preempt_disable(); | |
2260 | cpu = task_cpu(p); | |
2261 | if ((cpu != smp_processor_id()) && task_curr(p)) | |
2262 | smp_send_reschedule(cpu); | |
2263 | preempt_enable(); | |
2264 | } | |
b43e3521 | 2265 | EXPORT_SYMBOL_GPL(kick_process); |
476d139c | 2266 | #endif /* CONFIG_SMP */ |
1da177e4 | 2267 | |
970b13ba | 2268 | #ifdef CONFIG_SMP |
30da688e ON |
2269 | /* |
2270 | * ->cpus_allowed is protected by either TASK_WAKING or rq->lock held. | |
2271 | */ | |
5da9a0fb PZ |
2272 | static int select_fallback_rq(int cpu, struct task_struct *p) |
2273 | { | |
2274 | int dest_cpu; | |
2275 | const struct cpumask *nodemask = cpumask_of_node(cpu_to_node(cpu)); | |
2276 | ||
2277 | /* Look for allowed, online CPU in same node. */ | |
2278 | for_each_cpu_and(dest_cpu, nodemask, cpu_active_mask) | |
2279 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) | |
2280 | return dest_cpu; | |
2281 | ||
2282 | /* Any allowed, online CPU? */ | |
2283 | dest_cpu = cpumask_any_and(&p->cpus_allowed, cpu_active_mask); | |
2284 | if (dest_cpu < nr_cpu_ids) | |
2285 | return dest_cpu; | |
2286 | ||
2287 | /* No more Mr. Nice Guy. */ | |
48c5ccae PZ |
2288 | dest_cpu = cpuset_cpus_allowed_fallback(p); |
2289 | /* | |
2290 | * Don't tell them about moving exiting tasks or | |
2291 | * kernel threads (both mm NULL), since they never | |
2292 | * leave kernel. | |
2293 | */ | |
2294 | if (p->mm && printk_ratelimit()) { | |
2295 | printk(KERN_INFO "process %d (%s) no longer affine to cpu%d\n", | |
2296 | task_pid_nr(p), p->comm, cpu); | |
5da9a0fb PZ |
2297 | } |
2298 | ||
2299 | return dest_cpu; | |
2300 | } | |
2301 | ||
e2912009 | 2302 | /* |
30da688e | 2303 | * The caller (fork, wakeup) owns TASK_WAKING, ->cpus_allowed is stable. |
e2912009 | 2304 | */ |
970b13ba | 2305 | static inline |
0017d735 | 2306 | int select_task_rq(struct rq *rq, struct task_struct *p, int sd_flags, int wake_flags) |
970b13ba | 2307 | { |
0017d735 | 2308 | int cpu = p->sched_class->select_task_rq(rq, p, sd_flags, wake_flags); |
e2912009 PZ |
2309 | |
2310 | /* | |
2311 | * In order not to call set_task_cpu() on a blocking task we need | |
2312 | * to rely on ttwu() to place the task on a valid ->cpus_allowed | |
2313 | * cpu. | |
2314 | * | |
2315 | * Since this is common to all placement strategies, this lives here. | |
2316 | * | |
2317 | * [ this allows ->select_task() to simply return task_cpu(p) and | |
2318 | * not worry about this generic constraint ] | |
2319 | */ | |
2320 | if (unlikely(!cpumask_test_cpu(cpu, &p->cpus_allowed) || | |
70f11205 | 2321 | !cpu_online(cpu))) |
5da9a0fb | 2322 | cpu = select_fallback_rq(task_cpu(p), p); |
e2912009 PZ |
2323 | |
2324 | return cpu; | |
970b13ba | 2325 | } |
09a40af5 MG |
2326 | |
2327 | static void update_avg(u64 *avg, u64 sample) | |
2328 | { | |
2329 | s64 diff = sample - *avg; | |
2330 | *avg += diff >> 3; | |
2331 | } | |
970b13ba PZ |
2332 | #endif |
2333 | ||
9ed3811a TH |
2334 | static inline void ttwu_activate(struct task_struct *p, struct rq *rq, |
2335 | bool is_sync, bool is_migrate, bool is_local, | |
2336 | unsigned long en_flags) | |
2337 | { | |
2338 | schedstat_inc(p, se.statistics.nr_wakeups); | |
2339 | if (is_sync) | |
2340 | schedstat_inc(p, se.statistics.nr_wakeups_sync); | |
2341 | if (is_migrate) | |
2342 | schedstat_inc(p, se.statistics.nr_wakeups_migrate); | |
2343 | if (is_local) | |
2344 | schedstat_inc(p, se.statistics.nr_wakeups_local); | |
2345 | else | |
2346 | schedstat_inc(p, se.statistics.nr_wakeups_remote); | |
2347 | ||
2348 | activate_task(rq, p, en_flags); | |
2349 | } | |
2350 | ||
2351 | static inline void ttwu_post_activation(struct task_struct *p, struct rq *rq, | |
2352 | int wake_flags, bool success) | |
2353 | { | |
2354 | trace_sched_wakeup(p, success); | |
2355 | check_preempt_curr(rq, p, wake_flags); | |
2356 | ||
2357 | p->state = TASK_RUNNING; | |
2358 | #ifdef CONFIG_SMP | |
2359 | if (p->sched_class->task_woken) | |
2360 | p->sched_class->task_woken(rq, p); | |
2361 | ||
2362 | if (unlikely(rq->idle_stamp)) { | |
2363 | u64 delta = rq->clock - rq->idle_stamp; | |
2364 | u64 max = 2*sysctl_sched_migration_cost; | |
2365 | ||
2366 | if (delta > max) | |
2367 | rq->avg_idle = max; | |
2368 | else | |
2369 | update_avg(&rq->avg_idle, delta); | |
2370 | rq->idle_stamp = 0; | |
2371 | } | |
2372 | #endif | |
21aa9af0 TH |
2373 | /* if a worker is waking up, notify workqueue */ |
2374 | if ((p->flags & PF_WQ_WORKER) && success) | |
2375 | wq_worker_waking_up(p, cpu_of(rq)); | |
9ed3811a TH |
2376 | } |
2377 | ||
2378 | /** | |
1da177e4 | 2379 | * try_to_wake_up - wake up a thread |
9ed3811a | 2380 | * @p: the thread to be awakened |
1da177e4 | 2381 | * @state: the mask of task states that can be woken |
9ed3811a | 2382 | * @wake_flags: wake modifier flags (WF_*) |
1da177e4 LT |
2383 | * |
2384 | * Put it on the run-queue if it's not already there. The "current" | |
2385 | * thread is always on the run-queue (except when the actual | |
2386 | * re-schedule is in progress), and as such you're allowed to do | |
2387 | * the simpler "current->state = TASK_RUNNING" to mark yourself | |
2388 | * runnable without the overhead of this. | |
2389 | * | |
9ed3811a TH |
2390 | * Returns %true if @p was woken up, %false if it was already running |
2391 | * or @state didn't match @p's state. | |
1da177e4 | 2392 | */ |
7d478721 PZ |
2393 | static int try_to_wake_up(struct task_struct *p, unsigned int state, |
2394 | int wake_flags) | |
1da177e4 | 2395 | { |
cc367732 | 2396 | int cpu, orig_cpu, this_cpu, success = 0; |
1da177e4 | 2397 | unsigned long flags; |
371fd7e7 | 2398 | unsigned long en_flags = ENQUEUE_WAKEUP; |
ab3b3aa5 | 2399 | struct rq *rq; |
1da177e4 | 2400 | |
e9c84311 | 2401 | this_cpu = get_cpu(); |
2398f2c6 | 2402 | |
04e2f174 | 2403 | smp_wmb(); |
ab3b3aa5 | 2404 | rq = task_rq_lock(p, &flags); |
e9c84311 | 2405 | if (!(p->state & state)) |
1da177e4 LT |
2406 | goto out; |
2407 | ||
dd41f596 | 2408 | if (p->se.on_rq) |
1da177e4 LT |
2409 | goto out_running; |
2410 | ||
2411 | cpu = task_cpu(p); | |
cc367732 | 2412 | orig_cpu = cpu; |
1da177e4 LT |
2413 | |
2414 | #ifdef CONFIG_SMP | |
2415 | if (unlikely(task_running(rq, p))) | |
2416 | goto out_activate; | |
2417 | ||
e9c84311 PZ |
2418 | /* |
2419 | * In order to handle concurrent wakeups and release the rq->lock | |
2420 | * we put the task in TASK_WAKING state. | |
eb24073b IM |
2421 | * |
2422 | * First fix up the nr_uninterruptible count: | |
e9c84311 | 2423 | */ |
cc87f76a PZ |
2424 | if (task_contributes_to_load(p)) { |
2425 | if (likely(cpu_online(orig_cpu))) | |
2426 | rq->nr_uninterruptible--; | |
2427 | else | |
2428 | this_rq()->nr_uninterruptible--; | |
2429 | } | |
e9c84311 | 2430 | p->state = TASK_WAKING; |
efbbd05a | 2431 | |
371fd7e7 | 2432 | if (p->sched_class->task_waking) { |
efbbd05a | 2433 | p->sched_class->task_waking(rq, p); |
371fd7e7 PZ |
2434 | en_flags |= ENQUEUE_WAKING; |
2435 | } | |
efbbd05a | 2436 | |
0017d735 PZ |
2437 | cpu = select_task_rq(rq, p, SD_BALANCE_WAKE, wake_flags); |
2438 | if (cpu != orig_cpu) | |
5d2f5a61 | 2439 | set_task_cpu(p, cpu); |
0017d735 | 2440 | __task_rq_unlock(rq); |
ab19cb23 | 2441 | |
0970d299 PZ |
2442 | rq = cpu_rq(cpu); |
2443 | raw_spin_lock(&rq->lock); | |
f5dc3753 | 2444 | |
0970d299 PZ |
2445 | /* |
2446 | * We migrated the task without holding either rq->lock, however | |
2447 | * since the task is not on the task list itself, nobody else | |
2448 | * will try and migrate the task, hence the rq should match the | |
2449 | * cpu we just moved it to. | |
2450 | */ | |
2451 | WARN_ON(task_cpu(p) != cpu); | |
e9c84311 | 2452 | WARN_ON(p->state != TASK_WAKING); |
1da177e4 | 2453 | |
e7693a36 GH |
2454 | #ifdef CONFIG_SCHEDSTATS |
2455 | schedstat_inc(rq, ttwu_count); | |
2456 | if (cpu == this_cpu) | |
2457 | schedstat_inc(rq, ttwu_local); | |
2458 | else { | |
2459 | struct sched_domain *sd; | |
2460 | for_each_domain(this_cpu, sd) { | |
758b2cdc | 2461 | if (cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
e7693a36 GH |
2462 | schedstat_inc(sd, ttwu_wake_remote); |
2463 | break; | |
2464 | } | |
2465 | } | |
2466 | } | |
6d6bc0ad | 2467 | #endif /* CONFIG_SCHEDSTATS */ |
e7693a36 | 2468 | |
1da177e4 LT |
2469 | out_activate: |
2470 | #endif /* CONFIG_SMP */ | |
9ed3811a TH |
2471 | ttwu_activate(p, rq, wake_flags & WF_SYNC, orig_cpu != cpu, |
2472 | cpu == this_cpu, en_flags); | |
1da177e4 | 2473 | success = 1; |
1da177e4 | 2474 | out_running: |
9ed3811a | 2475 | ttwu_post_activation(p, rq, wake_flags, success); |
1da177e4 LT |
2476 | out: |
2477 | task_rq_unlock(rq, &flags); | |
e9c84311 | 2478 | put_cpu(); |
1da177e4 LT |
2479 | |
2480 | return success; | |
2481 | } | |
2482 | ||
21aa9af0 TH |
2483 | /** |
2484 | * try_to_wake_up_local - try to wake up a local task with rq lock held | |
2485 | * @p: the thread to be awakened | |
2486 | * | |
b595076a | 2487 | * Put @p on the run-queue if it's not already there. The caller must |
21aa9af0 TH |
2488 | * ensure that this_rq() is locked, @p is bound to this_rq() and not |
2489 | * the current task. this_rq() stays locked over invocation. | |
2490 | */ | |
2491 | static void try_to_wake_up_local(struct task_struct *p) | |
2492 | { | |
2493 | struct rq *rq = task_rq(p); | |
2494 | bool success = false; | |
2495 | ||
2496 | BUG_ON(rq != this_rq()); | |
2497 | BUG_ON(p == current); | |
2498 | lockdep_assert_held(&rq->lock); | |
2499 | ||
2500 | if (!(p->state & TASK_NORMAL)) | |
2501 | return; | |
2502 | ||
2503 | if (!p->se.on_rq) { | |
2504 | if (likely(!task_running(rq, p))) { | |
2505 | schedstat_inc(rq, ttwu_count); | |
2506 | schedstat_inc(rq, ttwu_local); | |
2507 | } | |
2508 | ttwu_activate(p, rq, false, false, true, ENQUEUE_WAKEUP); | |
2509 | success = true; | |
2510 | } | |
2511 | ttwu_post_activation(p, rq, 0, success); | |
2512 | } | |
2513 | ||
50fa610a DH |
2514 | /** |
2515 | * wake_up_process - Wake up a specific process | |
2516 | * @p: The process to be woken up. | |
2517 | * | |
2518 | * Attempt to wake up the nominated process and move it to the set of runnable | |
2519 | * processes. Returns 1 if the process was woken up, 0 if it was already | |
2520 | * running. | |
2521 | * | |
2522 | * It may be assumed that this function implies a write memory barrier before | |
2523 | * changing the task state if and only if any tasks are woken up. | |
2524 | */ | |
7ad5b3a5 | 2525 | int wake_up_process(struct task_struct *p) |
1da177e4 | 2526 | { |
d9514f6c | 2527 | return try_to_wake_up(p, TASK_ALL, 0); |
1da177e4 | 2528 | } |
1da177e4 LT |
2529 | EXPORT_SYMBOL(wake_up_process); |
2530 | ||
7ad5b3a5 | 2531 | int wake_up_state(struct task_struct *p, unsigned int state) |
1da177e4 LT |
2532 | { |
2533 | return try_to_wake_up(p, state, 0); | |
2534 | } | |
2535 | ||
1da177e4 LT |
2536 | /* |
2537 | * Perform scheduler related setup for a newly forked process p. | |
2538 | * p is forked by current. | |
dd41f596 IM |
2539 | * |
2540 | * __sched_fork() is basic setup used by init_idle() too: | |
2541 | */ | |
2542 | static void __sched_fork(struct task_struct *p) | |
2543 | { | |
dd41f596 IM |
2544 | p->se.exec_start = 0; |
2545 | p->se.sum_exec_runtime = 0; | |
f6cf891c | 2546 | p->se.prev_sum_exec_runtime = 0; |
6c594c21 | 2547 | p->se.nr_migrations = 0; |
6cfb0d5d IM |
2548 | |
2549 | #ifdef CONFIG_SCHEDSTATS | |
41acab88 | 2550 | memset(&p->se.statistics, 0, sizeof(p->se.statistics)); |
6cfb0d5d | 2551 | #endif |
476d139c | 2552 | |
fa717060 | 2553 | INIT_LIST_HEAD(&p->rt.run_list); |
dd41f596 | 2554 | p->se.on_rq = 0; |
4a55bd5e | 2555 | INIT_LIST_HEAD(&p->se.group_node); |
476d139c | 2556 | |
e107be36 AK |
2557 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2558 | INIT_HLIST_HEAD(&p->preempt_notifiers); | |
2559 | #endif | |
dd41f596 IM |
2560 | } |
2561 | ||
2562 | /* | |
2563 | * fork()/clone()-time setup: | |
2564 | */ | |
2565 | void sched_fork(struct task_struct *p, int clone_flags) | |
2566 | { | |
2567 | int cpu = get_cpu(); | |
2568 | ||
2569 | __sched_fork(p); | |
06b83b5f | 2570 | /* |
0017d735 | 2571 | * We mark the process as running here. This guarantees that |
06b83b5f PZ |
2572 | * nobody will actually run it, and a signal or other external |
2573 | * event cannot wake it up and insert it on the runqueue either. | |
2574 | */ | |
0017d735 | 2575 | p->state = TASK_RUNNING; |
dd41f596 | 2576 | |
b9dc29e7 MG |
2577 | /* |
2578 | * Revert to default priority/policy on fork if requested. | |
2579 | */ | |
2580 | if (unlikely(p->sched_reset_on_fork)) { | |
f83f9ac2 | 2581 | if (p->policy == SCHED_FIFO || p->policy == SCHED_RR) { |
b9dc29e7 | 2582 | p->policy = SCHED_NORMAL; |
f83f9ac2 PW |
2583 | p->normal_prio = p->static_prio; |
2584 | } | |
b9dc29e7 | 2585 | |
6c697bdf MG |
2586 | if (PRIO_TO_NICE(p->static_prio) < 0) { |
2587 | p->static_prio = NICE_TO_PRIO(0); | |
f83f9ac2 | 2588 | p->normal_prio = p->static_prio; |
6c697bdf MG |
2589 | set_load_weight(p); |
2590 | } | |
2591 | ||
b9dc29e7 MG |
2592 | /* |
2593 | * We don't need the reset flag anymore after the fork. It has | |
2594 | * fulfilled its duty: | |
2595 | */ | |
2596 | p->sched_reset_on_fork = 0; | |
2597 | } | |
ca94c442 | 2598 | |
f83f9ac2 PW |
2599 | /* |
2600 | * Make sure we do not leak PI boosting priority to the child. | |
2601 | */ | |
2602 | p->prio = current->normal_prio; | |
2603 | ||
2ddbf952 HS |
2604 | if (!rt_prio(p->prio)) |
2605 | p->sched_class = &fair_sched_class; | |
b29739f9 | 2606 | |
cd29fe6f PZ |
2607 | if (p->sched_class->task_fork) |
2608 | p->sched_class->task_fork(p); | |
2609 | ||
86951599 PZ |
2610 | /* |
2611 | * The child is not yet in the pid-hash so no cgroup attach races, | |
2612 | * and the cgroup is pinned to this child due to cgroup_fork() | |
2613 | * is ran before sched_fork(). | |
2614 | * | |
2615 | * Silence PROVE_RCU. | |
2616 | */ | |
2617 | rcu_read_lock(); | |
5f3edc1b | 2618 | set_task_cpu(p, cpu); |
86951599 | 2619 | rcu_read_unlock(); |
5f3edc1b | 2620 | |
52f17b6c | 2621 | #if defined(CONFIG_SCHEDSTATS) || defined(CONFIG_TASK_DELAY_ACCT) |
dd41f596 | 2622 | if (likely(sched_info_on())) |
52f17b6c | 2623 | memset(&p->sched_info, 0, sizeof(p->sched_info)); |
1da177e4 | 2624 | #endif |
d6077cb8 | 2625 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
4866cde0 NP |
2626 | p->oncpu = 0; |
2627 | #endif | |
1da177e4 | 2628 | #ifdef CONFIG_PREEMPT |
4866cde0 | 2629 | /* Want to start with kernel preemption disabled. */ |
a1261f54 | 2630 | task_thread_info(p)->preempt_count = 1; |
1da177e4 | 2631 | #endif |
806c09a7 | 2632 | #ifdef CONFIG_SMP |
917b627d | 2633 | plist_node_init(&p->pushable_tasks, MAX_PRIO); |
806c09a7 | 2634 | #endif |
917b627d | 2635 | |
476d139c | 2636 | put_cpu(); |
1da177e4 LT |
2637 | } |
2638 | ||
2639 | /* | |
2640 | * wake_up_new_task - wake up a newly created task for the first time. | |
2641 | * | |
2642 | * This function will do some initial scheduler statistics housekeeping | |
2643 | * that must be done for every newly created context, then puts the task | |
2644 | * on the runqueue and wakes it. | |
2645 | */ | |
7ad5b3a5 | 2646 | void wake_up_new_task(struct task_struct *p, unsigned long clone_flags) |
1da177e4 LT |
2647 | { |
2648 | unsigned long flags; | |
dd41f596 | 2649 | struct rq *rq; |
c890692b | 2650 | int cpu __maybe_unused = get_cpu(); |
fabf318e PZ |
2651 | |
2652 | #ifdef CONFIG_SMP | |
0017d735 PZ |
2653 | rq = task_rq_lock(p, &flags); |
2654 | p->state = TASK_WAKING; | |
2655 | ||
fabf318e PZ |
2656 | /* |
2657 | * Fork balancing, do it here and not earlier because: | |
2658 | * - cpus_allowed can change in the fork path | |
2659 | * - any previously selected cpu might disappear through hotplug | |
2660 | * | |
0017d735 PZ |
2661 | * We set TASK_WAKING so that select_task_rq() can drop rq->lock |
2662 | * without people poking at ->cpus_allowed. | |
fabf318e | 2663 | */ |
0017d735 | 2664 | cpu = select_task_rq(rq, p, SD_BALANCE_FORK, 0); |
fabf318e | 2665 | set_task_cpu(p, cpu); |
1da177e4 | 2666 | |
06b83b5f | 2667 | p->state = TASK_RUNNING; |
0017d735 PZ |
2668 | task_rq_unlock(rq, &flags); |
2669 | #endif | |
2670 | ||
2671 | rq = task_rq_lock(p, &flags); | |
cd29fe6f | 2672 | activate_task(rq, p, 0); |
27a9da65 | 2673 | trace_sched_wakeup_new(p, 1); |
a7558e01 | 2674 | check_preempt_curr(rq, p, WF_FORK); |
9a897c5a | 2675 | #ifdef CONFIG_SMP |
efbbd05a PZ |
2676 | if (p->sched_class->task_woken) |
2677 | p->sched_class->task_woken(rq, p); | |
9a897c5a | 2678 | #endif |
dd41f596 | 2679 | task_rq_unlock(rq, &flags); |
fabf318e | 2680 | put_cpu(); |
1da177e4 LT |
2681 | } |
2682 | ||
e107be36 AK |
2683 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
2684 | ||
2685 | /** | |
80dd99b3 | 2686 | * preempt_notifier_register - tell me when current is being preempted & rescheduled |
421cee29 | 2687 | * @notifier: notifier struct to register |
e107be36 AK |
2688 | */ |
2689 | void preempt_notifier_register(struct preempt_notifier *notifier) | |
2690 | { | |
2691 | hlist_add_head(¬ifier->link, ¤t->preempt_notifiers); | |
2692 | } | |
2693 | EXPORT_SYMBOL_GPL(preempt_notifier_register); | |
2694 | ||
2695 | /** | |
2696 | * preempt_notifier_unregister - no longer interested in preemption notifications | |
421cee29 | 2697 | * @notifier: notifier struct to unregister |
e107be36 AK |
2698 | * |
2699 | * This is safe to call from within a preemption notifier. | |
2700 | */ | |
2701 | void preempt_notifier_unregister(struct preempt_notifier *notifier) | |
2702 | { | |
2703 | hlist_del(¬ifier->link); | |
2704 | } | |
2705 | EXPORT_SYMBOL_GPL(preempt_notifier_unregister); | |
2706 | ||
2707 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2708 | { | |
2709 | struct preempt_notifier *notifier; | |
2710 | struct hlist_node *node; | |
2711 | ||
2712 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2713 | notifier->ops->sched_in(notifier, raw_smp_processor_id()); | |
2714 | } | |
2715 | ||
2716 | static void | |
2717 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2718 | struct task_struct *next) | |
2719 | { | |
2720 | struct preempt_notifier *notifier; | |
2721 | struct hlist_node *node; | |
2722 | ||
2723 | hlist_for_each_entry(notifier, node, &curr->preempt_notifiers, link) | |
2724 | notifier->ops->sched_out(notifier, next); | |
2725 | } | |
2726 | ||
6d6bc0ad | 2727 | #else /* !CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 AK |
2728 | |
2729 | static void fire_sched_in_preempt_notifiers(struct task_struct *curr) | |
2730 | { | |
2731 | } | |
2732 | ||
2733 | static void | |
2734 | fire_sched_out_preempt_notifiers(struct task_struct *curr, | |
2735 | struct task_struct *next) | |
2736 | { | |
2737 | } | |
2738 | ||
6d6bc0ad | 2739 | #endif /* CONFIG_PREEMPT_NOTIFIERS */ |
e107be36 | 2740 | |
4866cde0 NP |
2741 | /** |
2742 | * prepare_task_switch - prepare to switch tasks | |
2743 | * @rq: the runqueue preparing to switch | |
421cee29 | 2744 | * @prev: the current task that is being switched out |
4866cde0 NP |
2745 | * @next: the task we are going to switch to. |
2746 | * | |
2747 | * This is called with the rq lock held and interrupts off. It must | |
2748 | * be paired with a subsequent finish_task_switch after the context | |
2749 | * switch. | |
2750 | * | |
2751 | * prepare_task_switch sets up locking and calls architecture specific | |
2752 | * hooks. | |
2753 | */ | |
e107be36 AK |
2754 | static inline void |
2755 | prepare_task_switch(struct rq *rq, struct task_struct *prev, | |
2756 | struct task_struct *next) | |
4866cde0 | 2757 | { |
fe4b04fa PZ |
2758 | sched_info_switch(prev, next); |
2759 | perf_event_task_sched_out(prev, next); | |
e107be36 | 2760 | fire_sched_out_preempt_notifiers(prev, next); |
4866cde0 NP |
2761 | prepare_lock_switch(rq, next); |
2762 | prepare_arch_switch(next); | |
fe4b04fa | 2763 | trace_sched_switch(prev, next); |
4866cde0 NP |
2764 | } |
2765 | ||
1da177e4 LT |
2766 | /** |
2767 | * finish_task_switch - clean up after a task-switch | |
344babaa | 2768 | * @rq: runqueue associated with task-switch |
1da177e4 LT |
2769 | * @prev: the thread we just switched away from. |
2770 | * | |
4866cde0 NP |
2771 | * finish_task_switch must be called after the context switch, paired |
2772 | * with a prepare_task_switch call before the context switch. | |
2773 | * finish_task_switch will reconcile locking set up by prepare_task_switch, | |
2774 | * and do any other architecture-specific cleanup actions. | |
1da177e4 LT |
2775 | * |
2776 | * Note that we may have delayed dropping an mm in context_switch(). If | |
41a2d6cf | 2777 | * so, we finish that here outside of the runqueue lock. (Doing it |
1da177e4 LT |
2778 | * with the lock held can cause deadlocks; see schedule() for |
2779 | * details.) | |
2780 | */ | |
a9957449 | 2781 | static void finish_task_switch(struct rq *rq, struct task_struct *prev) |
1da177e4 LT |
2782 | __releases(rq->lock) |
2783 | { | |
1da177e4 | 2784 | struct mm_struct *mm = rq->prev_mm; |
55a101f8 | 2785 | long prev_state; |
1da177e4 LT |
2786 | |
2787 | rq->prev_mm = NULL; | |
2788 | ||
2789 | /* | |
2790 | * A task struct has one reference for the use as "current". | |
c394cc9f | 2791 | * If a task dies, then it sets TASK_DEAD in tsk->state and calls |
55a101f8 ON |
2792 | * schedule one last time. The schedule call will never return, and |
2793 | * the scheduled task must drop that reference. | |
c394cc9f | 2794 | * The test for TASK_DEAD must occur while the runqueue locks are |
1da177e4 LT |
2795 | * still held, otherwise prev could be scheduled on another cpu, die |
2796 | * there before we look at prev->state, and then the reference would | |
2797 | * be dropped twice. | |
2798 | * Manfred Spraul <manfred@colorfullife.com> | |
2799 | */ | |
55a101f8 | 2800 | prev_state = prev->state; |
4866cde0 | 2801 | finish_arch_switch(prev); |
8381f65d JI |
2802 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2803 | local_irq_disable(); | |
2804 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
49f47433 | 2805 | perf_event_task_sched_in(current); |
8381f65d JI |
2806 | #ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW |
2807 | local_irq_enable(); | |
2808 | #endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */ | |
4866cde0 | 2809 | finish_lock_switch(rq, prev); |
e8fa1362 | 2810 | |
e107be36 | 2811 | fire_sched_in_preempt_notifiers(current); |
1da177e4 LT |
2812 | if (mm) |
2813 | mmdrop(mm); | |
c394cc9f | 2814 | if (unlikely(prev_state == TASK_DEAD)) { |
c6fd91f0 | 2815 | /* |
2816 | * Remove function-return probe instances associated with this | |
2817 | * task and put them back on the free list. | |
9761eea8 | 2818 | */ |
c6fd91f0 | 2819 | kprobe_flush_task(prev); |
1da177e4 | 2820 | put_task_struct(prev); |
c6fd91f0 | 2821 | } |
1da177e4 LT |
2822 | } |
2823 | ||
3f029d3c GH |
2824 | #ifdef CONFIG_SMP |
2825 | ||
2826 | /* assumes rq->lock is held */ | |
2827 | static inline void pre_schedule(struct rq *rq, struct task_struct *prev) | |
2828 | { | |
2829 | if (prev->sched_class->pre_schedule) | |
2830 | prev->sched_class->pre_schedule(rq, prev); | |
2831 | } | |
2832 | ||
2833 | /* rq->lock is NOT held, but preemption is disabled */ | |
2834 | static inline void post_schedule(struct rq *rq) | |
2835 | { | |
2836 | if (rq->post_schedule) { | |
2837 | unsigned long flags; | |
2838 | ||
05fa785c | 2839 | raw_spin_lock_irqsave(&rq->lock, flags); |
3f029d3c GH |
2840 | if (rq->curr->sched_class->post_schedule) |
2841 | rq->curr->sched_class->post_schedule(rq); | |
05fa785c | 2842 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
3f029d3c GH |
2843 | |
2844 | rq->post_schedule = 0; | |
2845 | } | |
2846 | } | |
2847 | ||
2848 | #else | |
da19ab51 | 2849 | |
3f029d3c GH |
2850 | static inline void pre_schedule(struct rq *rq, struct task_struct *p) |
2851 | { | |
2852 | } | |
2853 | ||
2854 | static inline void post_schedule(struct rq *rq) | |
2855 | { | |
1da177e4 LT |
2856 | } |
2857 | ||
3f029d3c GH |
2858 | #endif |
2859 | ||
1da177e4 LT |
2860 | /** |
2861 | * schedule_tail - first thing a freshly forked thread must call. | |
2862 | * @prev: the thread we just switched away from. | |
2863 | */ | |
36c8b586 | 2864 | asmlinkage void schedule_tail(struct task_struct *prev) |
1da177e4 LT |
2865 | __releases(rq->lock) |
2866 | { | |
70b97a7f IM |
2867 | struct rq *rq = this_rq(); |
2868 | ||
4866cde0 | 2869 | finish_task_switch(rq, prev); |
da19ab51 | 2870 | |
3f029d3c GH |
2871 | /* |
2872 | * FIXME: do we need to worry about rq being invalidated by the | |
2873 | * task_switch? | |
2874 | */ | |
2875 | post_schedule(rq); | |
70b97a7f | 2876 | |
4866cde0 NP |
2877 | #ifdef __ARCH_WANT_UNLOCKED_CTXSW |
2878 | /* In this case, finish_task_switch does not reenable preemption */ | |
2879 | preempt_enable(); | |
2880 | #endif | |
1da177e4 | 2881 | if (current->set_child_tid) |
b488893a | 2882 | put_user(task_pid_vnr(current), current->set_child_tid); |
1da177e4 LT |
2883 | } |
2884 | ||
2885 | /* | |
2886 | * context_switch - switch to the new MM and the new | |
2887 | * thread's register state. | |
2888 | */ | |
dd41f596 | 2889 | static inline void |
70b97a7f | 2890 | context_switch(struct rq *rq, struct task_struct *prev, |
36c8b586 | 2891 | struct task_struct *next) |
1da177e4 | 2892 | { |
dd41f596 | 2893 | struct mm_struct *mm, *oldmm; |
1da177e4 | 2894 | |
e107be36 | 2895 | prepare_task_switch(rq, prev, next); |
fe4b04fa | 2896 | |
dd41f596 IM |
2897 | mm = next->mm; |
2898 | oldmm = prev->active_mm; | |
9226d125 ZA |
2899 | /* |
2900 | * For paravirt, this is coupled with an exit in switch_to to | |
2901 | * combine the page table reload and the switch backend into | |
2902 | * one hypercall. | |
2903 | */ | |
224101ed | 2904 | arch_start_context_switch(prev); |
9226d125 | 2905 | |
31915ab4 | 2906 | if (!mm) { |
1da177e4 LT |
2907 | next->active_mm = oldmm; |
2908 | atomic_inc(&oldmm->mm_count); | |
2909 | enter_lazy_tlb(oldmm, next); | |
2910 | } else | |
2911 | switch_mm(oldmm, mm, next); | |
2912 | ||
31915ab4 | 2913 | if (!prev->mm) { |
1da177e4 | 2914 | prev->active_mm = NULL; |
1da177e4 LT |
2915 | rq->prev_mm = oldmm; |
2916 | } | |
3a5f5e48 IM |
2917 | /* |
2918 | * Since the runqueue lock will be released by the next | |
2919 | * task (which is an invalid locking op but in the case | |
2920 | * of the scheduler it's an obvious special-case), so we | |
2921 | * do an early lockdep release here: | |
2922 | */ | |
2923 | #ifndef __ARCH_WANT_UNLOCKED_CTXSW | |
8a25d5de | 2924 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
3a5f5e48 | 2925 | #endif |
1da177e4 LT |
2926 | |
2927 | /* Here we just switch the register state and the stack. */ | |
2928 | switch_to(prev, next, prev); | |
2929 | ||
dd41f596 IM |
2930 | barrier(); |
2931 | /* | |
2932 | * this_rq must be evaluated again because prev may have moved | |
2933 | * CPUs since it called schedule(), thus the 'rq' on its stack | |
2934 | * frame will be invalid. | |
2935 | */ | |
2936 | finish_task_switch(this_rq(), prev); | |
1da177e4 LT |
2937 | } |
2938 | ||
2939 | /* | |
2940 | * nr_running, nr_uninterruptible and nr_context_switches: | |
2941 | * | |
2942 | * externally visible scheduler statistics: current number of runnable | |
2943 | * threads, current number of uninterruptible-sleeping threads, total | |
2944 | * number of context switches performed since bootup. | |
2945 | */ | |
2946 | unsigned long nr_running(void) | |
2947 | { | |
2948 | unsigned long i, sum = 0; | |
2949 | ||
2950 | for_each_online_cpu(i) | |
2951 | sum += cpu_rq(i)->nr_running; | |
2952 | ||
2953 | return sum; | |
f711f609 | 2954 | } |
1da177e4 LT |
2955 | |
2956 | unsigned long nr_uninterruptible(void) | |
f711f609 | 2957 | { |
1da177e4 | 2958 | unsigned long i, sum = 0; |
f711f609 | 2959 | |
0a945022 | 2960 | for_each_possible_cpu(i) |
1da177e4 | 2961 | sum += cpu_rq(i)->nr_uninterruptible; |
f711f609 GS |
2962 | |
2963 | /* | |
1da177e4 LT |
2964 | * Since we read the counters lockless, it might be slightly |
2965 | * inaccurate. Do not allow it to go below zero though: | |
f711f609 | 2966 | */ |
1da177e4 LT |
2967 | if (unlikely((long)sum < 0)) |
2968 | sum = 0; | |
f711f609 | 2969 | |
1da177e4 | 2970 | return sum; |
f711f609 | 2971 | } |
f711f609 | 2972 | |
1da177e4 | 2973 | unsigned long long nr_context_switches(void) |
46cb4b7c | 2974 | { |
cc94abfc SR |
2975 | int i; |
2976 | unsigned long long sum = 0; | |
46cb4b7c | 2977 | |
0a945022 | 2978 | for_each_possible_cpu(i) |
1da177e4 | 2979 | sum += cpu_rq(i)->nr_switches; |
46cb4b7c | 2980 | |
1da177e4 LT |
2981 | return sum; |
2982 | } | |
483b4ee6 | 2983 | |
1da177e4 LT |
2984 | unsigned long nr_iowait(void) |
2985 | { | |
2986 | unsigned long i, sum = 0; | |
483b4ee6 | 2987 | |
0a945022 | 2988 | for_each_possible_cpu(i) |
1da177e4 | 2989 | sum += atomic_read(&cpu_rq(i)->nr_iowait); |
46cb4b7c | 2990 | |
1da177e4 LT |
2991 | return sum; |
2992 | } | |
483b4ee6 | 2993 | |
8c215bd3 | 2994 | unsigned long nr_iowait_cpu(int cpu) |
69d25870 | 2995 | { |
8c215bd3 | 2996 | struct rq *this = cpu_rq(cpu); |
69d25870 AV |
2997 | return atomic_read(&this->nr_iowait); |
2998 | } | |
46cb4b7c | 2999 | |
69d25870 AV |
3000 | unsigned long this_cpu_load(void) |
3001 | { | |
3002 | struct rq *this = this_rq(); | |
3003 | return this->cpu_load[0]; | |
3004 | } | |
e790fb0b | 3005 | |
46cb4b7c | 3006 | |
dce48a84 TG |
3007 | /* Variables and functions for calc_load */ |
3008 | static atomic_long_t calc_load_tasks; | |
3009 | static unsigned long calc_load_update; | |
3010 | unsigned long avenrun[3]; | |
3011 | EXPORT_SYMBOL(avenrun); | |
46cb4b7c | 3012 | |
74f5187a PZ |
3013 | static long calc_load_fold_active(struct rq *this_rq) |
3014 | { | |
3015 | long nr_active, delta = 0; | |
3016 | ||
3017 | nr_active = this_rq->nr_running; | |
3018 | nr_active += (long) this_rq->nr_uninterruptible; | |
3019 | ||
3020 | if (nr_active != this_rq->calc_load_active) { | |
3021 | delta = nr_active - this_rq->calc_load_active; | |
3022 | this_rq->calc_load_active = nr_active; | |
3023 | } | |
3024 | ||
3025 | return delta; | |
3026 | } | |
3027 | ||
0f004f5a PZ |
3028 | static unsigned long |
3029 | calc_load(unsigned long load, unsigned long exp, unsigned long active) | |
3030 | { | |
3031 | load *= exp; | |
3032 | load += active * (FIXED_1 - exp); | |
3033 | load += 1UL << (FSHIFT - 1); | |
3034 | return load >> FSHIFT; | |
3035 | } | |
3036 | ||
74f5187a PZ |
3037 | #ifdef CONFIG_NO_HZ |
3038 | /* | |
3039 | * For NO_HZ we delay the active fold to the next LOAD_FREQ update. | |
3040 | * | |
3041 | * When making the ILB scale, we should try to pull this in as well. | |
3042 | */ | |
3043 | static atomic_long_t calc_load_tasks_idle; | |
3044 | ||
3045 | static void calc_load_account_idle(struct rq *this_rq) | |
3046 | { | |
3047 | long delta; | |
3048 | ||
3049 | delta = calc_load_fold_active(this_rq); | |
3050 | if (delta) | |
3051 | atomic_long_add(delta, &calc_load_tasks_idle); | |
3052 | } | |
3053 | ||
3054 | static long calc_load_fold_idle(void) | |
3055 | { | |
3056 | long delta = 0; | |
3057 | ||
3058 | /* | |
3059 | * Its got a race, we don't care... | |
3060 | */ | |
3061 | if (atomic_long_read(&calc_load_tasks_idle)) | |
3062 | delta = atomic_long_xchg(&calc_load_tasks_idle, 0); | |
3063 | ||
3064 | return delta; | |
3065 | } | |
0f004f5a PZ |
3066 | |
3067 | /** | |
3068 | * fixed_power_int - compute: x^n, in O(log n) time | |
3069 | * | |
3070 | * @x: base of the power | |
3071 | * @frac_bits: fractional bits of @x | |
3072 | * @n: power to raise @x to. | |
3073 | * | |
3074 | * By exploiting the relation between the definition of the natural power | |
3075 | * function: x^n := x*x*...*x (x multiplied by itself for n times), and | |
3076 | * the binary encoding of numbers used by computers: n := \Sum n_i * 2^i, | |
3077 | * (where: n_i \elem {0, 1}, the binary vector representing n), | |
3078 | * we find: x^n := x^(\Sum n_i * 2^i) := \Prod x^(n_i * 2^i), which is | |
3079 | * of course trivially computable in O(log_2 n), the length of our binary | |
3080 | * vector. | |
3081 | */ | |
3082 | static unsigned long | |
3083 | fixed_power_int(unsigned long x, unsigned int frac_bits, unsigned int n) | |
3084 | { | |
3085 | unsigned long result = 1UL << frac_bits; | |
3086 | ||
3087 | if (n) for (;;) { | |
3088 | if (n & 1) { | |
3089 | result *= x; | |
3090 | result += 1UL << (frac_bits - 1); | |
3091 | result >>= frac_bits; | |
3092 | } | |
3093 | n >>= 1; | |
3094 | if (!n) | |
3095 | break; | |
3096 | x *= x; | |
3097 | x += 1UL << (frac_bits - 1); | |
3098 | x >>= frac_bits; | |
3099 | } | |
3100 | ||
3101 | return result; | |
3102 | } | |
3103 | ||
3104 | /* | |
3105 | * a1 = a0 * e + a * (1 - e) | |
3106 | * | |
3107 | * a2 = a1 * e + a * (1 - e) | |
3108 | * = (a0 * e + a * (1 - e)) * e + a * (1 - e) | |
3109 | * = a0 * e^2 + a * (1 - e) * (1 + e) | |
3110 | * | |
3111 | * a3 = a2 * e + a * (1 - e) | |
3112 | * = (a0 * e^2 + a * (1 - e) * (1 + e)) * e + a * (1 - e) | |
3113 | * = a0 * e^3 + a * (1 - e) * (1 + e + e^2) | |
3114 | * | |
3115 | * ... | |
3116 | * | |
3117 | * an = a0 * e^n + a * (1 - e) * (1 + e + ... + e^n-1) [1] | |
3118 | * = a0 * e^n + a * (1 - e) * (1 - e^n)/(1 - e) | |
3119 | * = a0 * e^n + a * (1 - e^n) | |
3120 | * | |
3121 | * [1] application of the geometric series: | |
3122 | * | |
3123 | * n 1 - x^(n+1) | |
3124 | * S_n := \Sum x^i = ------------- | |
3125 | * i=0 1 - x | |
3126 | */ | |
3127 | static unsigned long | |
3128 | calc_load_n(unsigned long load, unsigned long exp, | |
3129 | unsigned long active, unsigned int n) | |
3130 | { | |
3131 | ||
3132 | return calc_load(load, fixed_power_int(exp, FSHIFT, n), active); | |
3133 | } | |
3134 | ||
3135 | /* | |
3136 | * NO_HZ can leave us missing all per-cpu ticks calling | |
3137 | * calc_load_account_active(), but since an idle CPU folds its delta into | |
3138 | * calc_load_tasks_idle per calc_load_account_idle(), all we need to do is fold | |
3139 | * in the pending idle delta if our idle period crossed a load cycle boundary. | |
3140 | * | |
3141 | * Once we've updated the global active value, we need to apply the exponential | |
3142 | * weights adjusted to the number of cycles missed. | |
3143 | */ | |
3144 | static void calc_global_nohz(unsigned long ticks) | |
3145 | { | |
3146 | long delta, active, n; | |
3147 | ||
3148 | if (time_before(jiffies, calc_load_update)) | |
3149 | return; | |
3150 | ||
3151 | /* | |
3152 | * If we crossed a calc_load_update boundary, make sure to fold | |
3153 | * any pending idle changes, the respective CPUs might have | |
3154 | * missed the tick driven calc_load_account_active() update | |
3155 | * due to NO_HZ. | |
3156 | */ | |
3157 | delta = calc_load_fold_idle(); | |
3158 | if (delta) | |
3159 | atomic_long_add(delta, &calc_load_tasks); | |
3160 | ||
3161 | /* | |
3162 | * If we were idle for multiple load cycles, apply them. | |
3163 | */ | |
3164 | if (ticks >= LOAD_FREQ) { | |
3165 | n = ticks / LOAD_FREQ; | |
3166 | ||
3167 | active = atomic_long_read(&calc_load_tasks); | |
3168 | active = active > 0 ? active * FIXED_1 : 0; | |
3169 | ||
3170 | avenrun[0] = calc_load_n(avenrun[0], EXP_1, active, n); | |
3171 | avenrun[1] = calc_load_n(avenrun[1], EXP_5, active, n); | |
3172 | avenrun[2] = calc_load_n(avenrun[2], EXP_15, active, n); | |
3173 | ||
3174 | calc_load_update += n * LOAD_FREQ; | |
3175 | } | |
3176 | ||
3177 | /* | |
3178 | * Its possible the remainder of the above division also crosses | |
3179 | * a LOAD_FREQ period, the regular check in calc_global_load() | |
3180 | * which comes after this will take care of that. | |
3181 | * | |
3182 | * Consider us being 11 ticks before a cycle completion, and us | |
3183 | * sleeping for 4*LOAD_FREQ + 22 ticks, then the above code will | |
3184 | * age us 4 cycles, and the test in calc_global_load() will | |
3185 | * pick up the final one. | |
3186 | */ | |
3187 | } | |
74f5187a PZ |
3188 | #else |
3189 | static void calc_load_account_idle(struct rq *this_rq) | |
3190 | { | |
3191 | } | |
3192 | ||
3193 | static inline long calc_load_fold_idle(void) | |
3194 | { | |
3195 | return 0; | |
3196 | } | |
0f004f5a PZ |
3197 | |
3198 | static void calc_global_nohz(unsigned long ticks) | |
3199 | { | |
3200 | } | |
74f5187a PZ |
3201 | #endif |
3202 | ||
2d02494f TG |
3203 | /** |
3204 | * get_avenrun - get the load average array | |
3205 | * @loads: pointer to dest load array | |
3206 | * @offset: offset to add | |
3207 | * @shift: shift count to shift the result left | |
3208 | * | |
3209 | * These values are estimates at best, so no need for locking. | |
3210 | */ | |
3211 | void get_avenrun(unsigned long *loads, unsigned long offset, int shift) | |
3212 | { | |
3213 | loads[0] = (avenrun[0] + offset) << shift; | |
3214 | loads[1] = (avenrun[1] + offset) << shift; | |
3215 | loads[2] = (avenrun[2] + offset) << shift; | |
46cb4b7c | 3216 | } |
46cb4b7c | 3217 | |
46cb4b7c | 3218 | /* |
dce48a84 TG |
3219 | * calc_load - update the avenrun load estimates 10 ticks after the |
3220 | * CPUs have updated calc_load_tasks. | |
7835b98b | 3221 | */ |
0f004f5a | 3222 | void calc_global_load(unsigned long ticks) |
7835b98b | 3223 | { |
dce48a84 | 3224 | long active; |
1da177e4 | 3225 | |
0f004f5a PZ |
3226 | calc_global_nohz(ticks); |
3227 | ||
3228 | if (time_before(jiffies, calc_load_update + 10)) | |
dce48a84 | 3229 | return; |
1da177e4 | 3230 | |
dce48a84 TG |
3231 | active = atomic_long_read(&calc_load_tasks); |
3232 | active = active > 0 ? active * FIXED_1 : 0; | |
1da177e4 | 3233 | |
dce48a84 TG |
3234 | avenrun[0] = calc_load(avenrun[0], EXP_1, active); |
3235 | avenrun[1] = calc_load(avenrun[1], EXP_5, active); | |
3236 | avenrun[2] = calc_load(avenrun[2], EXP_15, active); | |
dd41f596 | 3237 | |
dce48a84 TG |
3238 | calc_load_update += LOAD_FREQ; |
3239 | } | |
1da177e4 | 3240 | |
dce48a84 | 3241 | /* |
74f5187a PZ |
3242 | * Called from update_cpu_load() to periodically update this CPU's |
3243 | * active count. | |
dce48a84 TG |
3244 | */ |
3245 | static void calc_load_account_active(struct rq *this_rq) | |
3246 | { | |
74f5187a | 3247 | long delta; |
08c183f3 | 3248 | |
74f5187a PZ |
3249 | if (time_before(jiffies, this_rq->calc_load_update)) |
3250 | return; | |
783609c6 | 3251 | |
74f5187a PZ |
3252 | delta = calc_load_fold_active(this_rq); |
3253 | delta += calc_load_fold_idle(); | |
3254 | if (delta) | |
dce48a84 | 3255 | atomic_long_add(delta, &calc_load_tasks); |
74f5187a PZ |
3256 | |
3257 | this_rq->calc_load_update += LOAD_FREQ; | |
46cb4b7c SS |
3258 | } |
3259 | ||
fdf3e95d VP |
3260 | /* |
3261 | * The exact cpuload at various idx values, calculated at every tick would be | |
3262 | * load = (2^idx - 1) / 2^idx * load + 1 / 2^idx * cur_load | |
3263 | * | |
3264 | * If a cpu misses updates for n-1 ticks (as it was idle) and update gets called | |
3265 | * on nth tick when cpu may be busy, then we have: | |
3266 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3267 | * load = (2^idx - 1) / 2^idx) * load + 1 / 2^idx * cur_load | |
3268 | * | |
3269 | * decay_load_missed() below does efficient calculation of | |
3270 | * load = ((2^idx - 1) / 2^idx)^(n-1) * load | |
3271 | * avoiding 0..n-1 loop doing load = ((2^idx - 1) / 2^idx) * load | |
3272 | * | |
3273 | * The calculation is approximated on a 128 point scale. | |
3274 | * degrade_zero_ticks is the number of ticks after which load at any | |
3275 | * particular idx is approximated to be zero. | |
3276 | * degrade_factor is a precomputed table, a row for each load idx. | |
3277 | * Each column corresponds to degradation factor for a power of two ticks, | |
3278 | * based on 128 point scale. | |
3279 | * Example: | |
3280 | * row 2, col 3 (=12) says that the degradation at load idx 2 after | |
3281 | * 8 ticks is 12/128 (which is an approximation of exact factor 3^8/4^8). | |
3282 | * | |
3283 | * With this power of 2 load factors, we can degrade the load n times | |
3284 | * by looking at 1 bits in n and doing as many mult/shift instead of | |
3285 | * n mult/shifts needed by the exact degradation. | |
3286 | */ | |
3287 | #define DEGRADE_SHIFT 7 | |
3288 | static const unsigned char | |
3289 | degrade_zero_ticks[CPU_LOAD_IDX_MAX] = {0, 8, 32, 64, 128}; | |
3290 | static const unsigned char | |
3291 | degrade_factor[CPU_LOAD_IDX_MAX][DEGRADE_SHIFT + 1] = { | |
3292 | {0, 0, 0, 0, 0, 0, 0, 0}, | |
3293 | {64, 32, 8, 0, 0, 0, 0, 0}, | |
3294 | {96, 72, 40, 12, 1, 0, 0}, | |
3295 | {112, 98, 75, 43, 15, 1, 0}, | |
3296 | {120, 112, 98, 76, 45, 16, 2} }; | |
3297 | ||
3298 | /* | |
3299 | * Update cpu_load for any missed ticks, due to tickless idle. The backlog | |
3300 | * would be when CPU is idle and so we just decay the old load without | |
3301 | * adding any new load. | |
3302 | */ | |
3303 | static unsigned long | |
3304 | decay_load_missed(unsigned long load, unsigned long missed_updates, int idx) | |
3305 | { | |
3306 | int j = 0; | |
3307 | ||
3308 | if (!missed_updates) | |
3309 | return load; | |
3310 | ||
3311 | if (missed_updates >= degrade_zero_ticks[idx]) | |
3312 | return 0; | |
3313 | ||
3314 | if (idx == 1) | |
3315 | return load >> missed_updates; | |
3316 | ||
3317 | while (missed_updates) { | |
3318 | if (missed_updates % 2) | |
3319 | load = (load * degrade_factor[idx][j]) >> DEGRADE_SHIFT; | |
3320 | ||
3321 | missed_updates >>= 1; | |
3322 | j++; | |
3323 | } | |
3324 | return load; | |
3325 | } | |
3326 | ||
46cb4b7c | 3327 | /* |
dd41f596 | 3328 | * Update rq->cpu_load[] statistics. This function is usually called every |
fdf3e95d VP |
3329 | * scheduler tick (TICK_NSEC). With tickless idle this will not be called |
3330 | * every tick. We fix it up based on jiffies. | |
46cb4b7c | 3331 | */ |
dd41f596 | 3332 | static void update_cpu_load(struct rq *this_rq) |
46cb4b7c | 3333 | { |
495eca49 | 3334 | unsigned long this_load = this_rq->load.weight; |
fdf3e95d VP |
3335 | unsigned long curr_jiffies = jiffies; |
3336 | unsigned long pending_updates; | |
dd41f596 | 3337 | int i, scale; |
46cb4b7c | 3338 | |
dd41f596 | 3339 | this_rq->nr_load_updates++; |
46cb4b7c | 3340 | |
fdf3e95d VP |
3341 | /* Avoid repeated calls on same jiffy, when moving in and out of idle */ |
3342 | if (curr_jiffies == this_rq->last_load_update_tick) | |
3343 | return; | |
3344 | ||
3345 | pending_updates = curr_jiffies - this_rq->last_load_update_tick; | |
3346 | this_rq->last_load_update_tick = curr_jiffies; | |
3347 | ||
dd41f596 | 3348 | /* Update our load: */ |
fdf3e95d VP |
3349 | this_rq->cpu_load[0] = this_load; /* Fasttrack for idx 0 */ |
3350 | for (i = 1, scale = 2; i < CPU_LOAD_IDX_MAX; i++, scale += scale) { | |
dd41f596 | 3351 | unsigned long old_load, new_load; |
7d1e6a9b | 3352 | |
dd41f596 | 3353 | /* scale is effectively 1 << i now, and >> i divides by scale */ |
46cb4b7c | 3354 | |
dd41f596 | 3355 | old_load = this_rq->cpu_load[i]; |
fdf3e95d | 3356 | old_load = decay_load_missed(old_load, pending_updates - 1, i); |
dd41f596 | 3357 | new_load = this_load; |
a25707f3 IM |
3358 | /* |
3359 | * Round up the averaging division if load is increasing. This | |
3360 | * prevents us from getting stuck on 9 if the load is 10, for | |
3361 | * example. | |
3362 | */ | |
3363 | if (new_load > old_load) | |
fdf3e95d VP |
3364 | new_load += scale - 1; |
3365 | ||
3366 | this_rq->cpu_load[i] = (old_load * (scale - 1) + new_load) >> i; | |
dd41f596 | 3367 | } |
da2b71ed SS |
3368 | |
3369 | sched_avg_update(this_rq); | |
fdf3e95d VP |
3370 | } |
3371 | ||
3372 | static void update_cpu_load_active(struct rq *this_rq) | |
3373 | { | |
3374 | update_cpu_load(this_rq); | |
46cb4b7c | 3375 | |
74f5187a | 3376 | calc_load_account_active(this_rq); |
46cb4b7c SS |
3377 | } |
3378 | ||
dd41f596 | 3379 | #ifdef CONFIG_SMP |
8a0be9ef | 3380 | |
46cb4b7c | 3381 | /* |
38022906 PZ |
3382 | * sched_exec - execve() is a valuable balancing opportunity, because at |
3383 | * this point the task has the smallest effective memory and cache footprint. | |
46cb4b7c | 3384 | */ |
38022906 | 3385 | void sched_exec(void) |
46cb4b7c | 3386 | { |
38022906 | 3387 | struct task_struct *p = current; |
1da177e4 | 3388 | unsigned long flags; |
70b97a7f | 3389 | struct rq *rq; |
0017d735 | 3390 | int dest_cpu; |
46cb4b7c | 3391 | |
1da177e4 | 3392 | rq = task_rq_lock(p, &flags); |
0017d735 PZ |
3393 | dest_cpu = p->sched_class->select_task_rq(rq, p, SD_BALANCE_EXEC, 0); |
3394 | if (dest_cpu == smp_processor_id()) | |
3395 | goto unlock; | |
38022906 | 3396 | |
46cb4b7c | 3397 | /* |
38022906 | 3398 | * select_task_rq() can race against ->cpus_allowed |
46cb4b7c | 3399 | */ |
30da688e | 3400 | if (cpumask_test_cpu(dest_cpu, &p->cpus_allowed) && |
b7a2b39d | 3401 | likely(cpu_active(dest_cpu)) && migrate_task(p, rq)) { |
969c7921 | 3402 | struct migration_arg arg = { p, dest_cpu }; |
46cb4b7c | 3403 | |
1da177e4 | 3404 | task_rq_unlock(rq, &flags); |
969c7921 | 3405 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
3406 | return; |
3407 | } | |
0017d735 | 3408 | unlock: |
1da177e4 | 3409 | task_rq_unlock(rq, &flags); |
1da177e4 | 3410 | } |
dd41f596 | 3411 | |
1da177e4 LT |
3412 | #endif |
3413 | ||
1da177e4 LT |
3414 | DEFINE_PER_CPU(struct kernel_stat, kstat); |
3415 | ||
3416 | EXPORT_PER_CPU_SYMBOL(kstat); | |
3417 | ||
3418 | /* | |
c5f8d995 | 3419 | * Return any ns on the sched_clock that have not yet been accounted in |
f06febc9 | 3420 | * @p in case that task is currently running. |
c5f8d995 HS |
3421 | * |
3422 | * Called with task_rq_lock() held on @rq. | |
1da177e4 | 3423 | */ |
c5f8d995 HS |
3424 | static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq) |
3425 | { | |
3426 | u64 ns = 0; | |
3427 | ||
3428 | if (task_current(rq, p)) { | |
3429 | update_rq_clock(rq); | |
305e6835 | 3430 | ns = rq->clock_task - p->se.exec_start; |
c5f8d995 HS |
3431 | if ((s64)ns < 0) |
3432 | ns = 0; | |
3433 | } | |
3434 | ||
3435 | return ns; | |
3436 | } | |
3437 | ||
bb34d92f | 3438 | unsigned long long task_delta_exec(struct task_struct *p) |
1da177e4 | 3439 | { |
1da177e4 | 3440 | unsigned long flags; |
41b86e9c | 3441 | struct rq *rq; |
bb34d92f | 3442 | u64 ns = 0; |
48f24c4d | 3443 | |
41b86e9c | 3444 | rq = task_rq_lock(p, &flags); |
c5f8d995 HS |
3445 | ns = do_task_delta_exec(p, rq); |
3446 | task_rq_unlock(rq, &flags); | |
1508487e | 3447 | |
c5f8d995 HS |
3448 | return ns; |
3449 | } | |
f06febc9 | 3450 | |
c5f8d995 HS |
3451 | /* |
3452 | * Return accounted runtime for the task. | |
3453 | * In case the task is currently running, return the runtime plus current's | |
3454 | * pending runtime that have not been accounted yet. | |
3455 | */ | |
3456 | unsigned long long task_sched_runtime(struct task_struct *p) | |
3457 | { | |
3458 | unsigned long flags; | |
3459 | struct rq *rq; | |
3460 | u64 ns = 0; | |
3461 | ||
3462 | rq = task_rq_lock(p, &flags); | |
3463 | ns = p->se.sum_exec_runtime + do_task_delta_exec(p, rq); | |
3464 | task_rq_unlock(rq, &flags); | |
3465 | ||
3466 | return ns; | |
3467 | } | |
48f24c4d | 3468 | |
c5f8d995 HS |
3469 | /* |
3470 | * Return sum_exec_runtime for the thread group. | |
3471 | * In case the task is currently running, return the sum plus current's | |
3472 | * pending runtime that have not been accounted yet. | |
3473 | * | |
3474 | * Note that the thread group might have other running tasks as well, | |
3475 | * so the return value not includes other pending runtime that other | |
3476 | * running tasks might have. | |
3477 | */ | |
3478 | unsigned long long thread_group_sched_runtime(struct task_struct *p) | |
3479 | { | |
3480 | struct task_cputime totals; | |
3481 | unsigned long flags; | |
3482 | struct rq *rq; | |
3483 | u64 ns; | |
3484 | ||
3485 | rq = task_rq_lock(p, &flags); | |
3486 | thread_group_cputime(p, &totals); | |
3487 | ns = totals.sum_exec_runtime + do_task_delta_exec(p, rq); | |
41b86e9c | 3488 | task_rq_unlock(rq, &flags); |
48f24c4d | 3489 | |
1da177e4 LT |
3490 | return ns; |
3491 | } | |
3492 | ||
1da177e4 LT |
3493 | /* |
3494 | * Account user cpu time to a process. | |
3495 | * @p: the process that the cpu time gets accounted to | |
1da177e4 | 3496 | * @cputime: the cpu time spent in user space since the last update |
457533a7 | 3497 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 | 3498 | */ |
457533a7 MS |
3499 | void account_user_time(struct task_struct *p, cputime_t cputime, |
3500 | cputime_t cputime_scaled) | |
1da177e4 LT |
3501 | { |
3502 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3503 | cputime64_t tmp; | |
3504 | ||
457533a7 | 3505 | /* Add user time to process. */ |
1da177e4 | 3506 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3507 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3508 | account_group_user_time(p, cputime); |
1da177e4 LT |
3509 | |
3510 | /* Add user time to cpustat. */ | |
3511 | tmp = cputime_to_cputime64(cputime); | |
3512 | if (TASK_NICE(p) > 0) | |
3513 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3514 | else | |
3515 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
ef12fefa BR |
3516 | |
3517 | cpuacct_update_stats(p, CPUACCT_STAT_USER, cputime); | |
49b5cf34 JL |
3518 | /* Account for user time used */ |
3519 | acct_update_integrals(p); | |
1da177e4 LT |
3520 | } |
3521 | ||
94886b84 LV |
3522 | /* |
3523 | * Account guest cpu time to a process. | |
3524 | * @p: the process that the cpu time gets accounted to | |
3525 | * @cputime: the cpu time spent in virtual machine since the last update | |
457533a7 | 3526 | * @cputime_scaled: cputime scaled by cpu frequency |
94886b84 | 3527 | */ |
457533a7 MS |
3528 | static void account_guest_time(struct task_struct *p, cputime_t cputime, |
3529 | cputime_t cputime_scaled) | |
94886b84 LV |
3530 | { |
3531 | cputime64_t tmp; | |
3532 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
3533 | ||
3534 | tmp = cputime_to_cputime64(cputime); | |
3535 | ||
457533a7 | 3536 | /* Add guest time to process. */ |
94886b84 | 3537 | p->utime = cputime_add(p->utime, cputime); |
457533a7 | 3538 | p->utimescaled = cputime_add(p->utimescaled, cputime_scaled); |
f06febc9 | 3539 | account_group_user_time(p, cputime); |
94886b84 LV |
3540 | p->gtime = cputime_add(p->gtime, cputime); |
3541 | ||
457533a7 | 3542 | /* Add guest time to cpustat. */ |
ce0e7b28 RO |
3543 | if (TASK_NICE(p) > 0) { |
3544 | cpustat->nice = cputime64_add(cpustat->nice, tmp); | |
3545 | cpustat->guest_nice = cputime64_add(cpustat->guest_nice, tmp); | |
3546 | } else { | |
3547 | cpustat->user = cputime64_add(cpustat->user, tmp); | |
3548 | cpustat->guest = cputime64_add(cpustat->guest, tmp); | |
3549 | } | |
94886b84 LV |
3550 | } |
3551 | ||
1da177e4 LT |
3552 | /* |
3553 | * Account system cpu time to a process. | |
3554 | * @p: the process that the cpu time gets accounted to | |
3555 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
3556 | * @cputime: the cpu time spent in kernel space since the last update | |
457533a7 | 3557 | * @cputime_scaled: cputime scaled by cpu frequency |
1da177e4 LT |
3558 | */ |
3559 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
457533a7 | 3560 | cputime_t cputime, cputime_t cputime_scaled) |
1da177e4 LT |
3561 | { |
3562 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
1da177e4 LT |
3563 | cputime64_t tmp; |
3564 | ||
983ed7a6 | 3565 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { |
457533a7 | 3566 | account_guest_time(p, cputime, cputime_scaled); |
983ed7a6 HH |
3567 | return; |
3568 | } | |
94886b84 | 3569 | |
457533a7 | 3570 | /* Add system time to process. */ |
1da177e4 | 3571 | p->stime = cputime_add(p->stime, cputime); |
457533a7 | 3572 | p->stimescaled = cputime_add(p->stimescaled, cputime_scaled); |
f06febc9 | 3573 | account_group_system_time(p, cputime); |
1da177e4 LT |
3574 | |
3575 | /* Add system time to cpustat. */ | |
3576 | tmp = cputime_to_cputime64(cputime); | |
3577 | if (hardirq_count() - hardirq_offset) | |
3578 | cpustat->irq = cputime64_add(cpustat->irq, tmp); | |
75e1056f | 3579 | else if (in_serving_softirq()) |
1da177e4 | 3580 | cpustat->softirq = cputime64_add(cpustat->softirq, tmp); |
1da177e4 | 3581 | else |
79741dd3 MS |
3582 | cpustat->system = cputime64_add(cpustat->system, tmp); |
3583 | ||
ef12fefa BR |
3584 | cpuacct_update_stats(p, CPUACCT_STAT_SYSTEM, cputime); |
3585 | ||
1da177e4 LT |
3586 | /* Account for system time used */ |
3587 | acct_update_integrals(p); | |
1da177e4 LT |
3588 | } |
3589 | ||
c66f08be | 3590 | /* |
1da177e4 | 3591 | * Account for involuntary wait time. |
1da177e4 | 3592 | * @steal: the cpu time spent in involuntary wait |
c66f08be | 3593 | */ |
79741dd3 | 3594 | void account_steal_time(cputime_t cputime) |
c66f08be | 3595 | { |
79741dd3 MS |
3596 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; |
3597 | cputime64_t cputime64 = cputime_to_cputime64(cputime); | |
3598 | ||
3599 | cpustat->steal = cputime64_add(cpustat->steal, cputime64); | |
c66f08be MN |
3600 | } |
3601 | ||
1da177e4 | 3602 | /* |
79741dd3 MS |
3603 | * Account for idle time. |
3604 | * @cputime: the cpu time spent in idle wait | |
1da177e4 | 3605 | */ |
79741dd3 | 3606 | void account_idle_time(cputime_t cputime) |
1da177e4 LT |
3607 | { |
3608 | struct cpu_usage_stat *cpustat = &kstat_this_cpu.cpustat; | |
79741dd3 | 3609 | cputime64_t cputime64 = cputime_to_cputime64(cputime); |
70b97a7f | 3610 | struct rq *rq = this_rq(); |
1da177e4 | 3611 | |
79741dd3 MS |
3612 | if (atomic_read(&rq->nr_iowait) > 0) |
3613 | cpustat->iowait = cputime64_add(cpustat->iowait, cputime64); | |
3614 | else | |
3615 | cpustat->idle = cputime64_add(cpustat->idle, cputime64); | |
1da177e4 LT |
3616 | } |
3617 | ||
79741dd3 MS |
3618 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING |
3619 | ||
3620 | /* | |
3621 | * Account a single tick of cpu time. | |
3622 | * @p: the process that the cpu time gets accounted to | |
3623 | * @user_tick: indicates if the tick is a user or a system tick | |
3624 | */ | |
3625 | void account_process_tick(struct task_struct *p, int user_tick) | |
3626 | { | |
a42548a1 | 3627 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); |
79741dd3 MS |
3628 | struct rq *rq = this_rq(); |
3629 | ||
3630 | if (user_tick) | |
a42548a1 | 3631 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); |
f5f293a4 | 3632 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) |
a42548a1 | 3633 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, |
79741dd3 MS |
3634 | one_jiffy_scaled); |
3635 | else | |
a42548a1 | 3636 | account_idle_time(cputime_one_jiffy); |
79741dd3 MS |
3637 | } |
3638 | ||
3639 | /* | |
3640 | * Account multiple ticks of steal time. | |
3641 | * @p: the process from which the cpu time has been stolen | |
3642 | * @ticks: number of stolen ticks | |
3643 | */ | |
3644 | void account_steal_ticks(unsigned long ticks) | |
3645 | { | |
3646 | account_steal_time(jiffies_to_cputime(ticks)); | |
3647 | } | |
3648 | ||
3649 | /* | |
3650 | * Account multiple ticks of idle time. | |
3651 | * @ticks: number of stolen ticks | |
3652 | */ | |
3653 | void account_idle_ticks(unsigned long ticks) | |
3654 | { | |
3655 | account_idle_time(jiffies_to_cputime(ticks)); | |
1da177e4 LT |
3656 | } |
3657 | ||
79741dd3 MS |
3658 | #endif |
3659 | ||
49048622 BS |
3660 | /* |
3661 | * Use precise platform statistics if available: | |
3662 | */ | |
3663 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
d180c5bc | 3664 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3665 | { |
d99ca3b9 HS |
3666 | *ut = p->utime; |
3667 | *st = p->stime; | |
49048622 BS |
3668 | } |
3669 | ||
0cf55e1e | 3670 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3671 | { |
0cf55e1e HS |
3672 | struct task_cputime cputime; |
3673 | ||
3674 | thread_group_cputime(p, &cputime); | |
3675 | ||
3676 | *ut = cputime.utime; | |
3677 | *st = cputime.stime; | |
49048622 BS |
3678 | } |
3679 | #else | |
761b1d26 HS |
3680 | |
3681 | #ifndef nsecs_to_cputime | |
b7b20df9 | 3682 | # define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs) |
761b1d26 HS |
3683 | #endif |
3684 | ||
d180c5bc | 3685 | void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st) |
49048622 | 3686 | { |
d99ca3b9 | 3687 | cputime_t rtime, utime = p->utime, total = cputime_add(utime, p->stime); |
49048622 BS |
3688 | |
3689 | /* | |
3690 | * Use CFS's precise accounting: | |
3691 | */ | |
d180c5bc | 3692 | rtime = nsecs_to_cputime(p->se.sum_exec_runtime); |
49048622 BS |
3693 | |
3694 | if (total) { | |
e75e863d | 3695 | u64 temp = rtime; |
d180c5bc | 3696 | |
e75e863d | 3697 | temp *= utime; |
49048622 | 3698 | do_div(temp, total); |
d180c5bc HS |
3699 | utime = (cputime_t)temp; |
3700 | } else | |
3701 | utime = rtime; | |
49048622 | 3702 | |
d180c5bc HS |
3703 | /* |
3704 | * Compare with previous values, to keep monotonicity: | |
3705 | */ | |
761b1d26 | 3706 | p->prev_utime = max(p->prev_utime, utime); |
d99ca3b9 | 3707 | p->prev_stime = max(p->prev_stime, cputime_sub(rtime, p->prev_utime)); |
49048622 | 3708 | |
d99ca3b9 HS |
3709 | *ut = p->prev_utime; |
3710 | *st = p->prev_stime; | |
49048622 BS |
3711 | } |
3712 | ||
0cf55e1e HS |
3713 | /* |
3714 | * Must be called with siglock held. | |
3715 | */ | |
3716 | void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st) | |
49048622 | 3717 | { |
0cf55e1e HS |
3718 | struct signal_struct *sig = p->signal; |
3719 | struct task_cputime cputime; | |
3720 | cputime_t rtime, utime, total; | |
49048622 | 3721 | |
0cf55e1e | 3722 | thread_group_cputime(p, &cputime); |
49048622 | 3723 | |
0cf55e1e HS |
3724 | total = cputime_add(cputime.utime, cputime.stime); |
3725 | rtime = nsecs_to_cputime(cputime.sum_exec_runtime); | |
49048622 | 3726 | |
0cf55e1e | 3727 | if (total) { |
e75e863d | 3728 | u64 temp = rtime; |
49048622 | 3729 | |
e75e863d | 3730 | temp *= cputime.utime; |
0cf55e1e HS |
3731 | do_div(temp, total); |
3732 | utime = (cputime_t)temp; | |
3733 | } else | |
3734 | utime = rtime; | |
3735 | ||
3736 | sig->prev_utime = max(sig->prev_utime, utime); | |
3737 | sig->prev_stime = max(sig->prev_stime, | |
3738 | cputime_sub(rtime, sig->prev_utime)); | |
3739 | ||
3740 | *ut = sig->prev_utime; | |
3741 | *st = sig->prev_stime; | |
49048622 | 3742 | } |
49048622 | 3743 | #endif |
49048622 | 3744 | |
7835b98b CL |
3745 | /* |
3746 | * This function gets called by the timer code, with HZ frequency. | |
3747 | * We call it with interrupts disabled. | |
3748 | * | |
3749 | * It also gets called by the fork code, when changing the parent's | |
3750 | * timeslices. | |
3751 | */ | |
3752 | void scheduler_tick(void) | |
3753 | { | |
7835b98b CL |
3754 | int cpu = smp_processor_id(); |
3755 | struct rq *rq = cpu_rq(cpu); | |
dd41f596 | 3756 | struct task_struct *curr = rq->curr; |
3e51f33f PZ |
3757 | |
3758 | sched_clock_tick(); | |
dd41f596 | 3759 | |
05fa785c | 3760 | raw_spin_lock(&rq->lock); |
3e51f33f | 3761 | update_rq_clock(rq); |
fdf3e95d | 3762 | update_cpu_load_active(rq); |
fa85ae24 | 3763 | curr->sched_class->task_tick(rq, curr, 0); |
05fa785c | 3764 | raw_spin_unlock(&rq->lock); |
7835b98b | 3765 | |
e9d2b064 | 3766 | perf_event_task_tick(); |
e220d2dc | 3767 | |
e418e1c2 | 3768 | #ifdef CONFIG_SMP |
dd41f596 IM |
3769 | rq->idle_at_tick = idle_cpu(cpu); |
3770 | trigger_load_balance(rq, cpu); | |
e418e1c2 | 3771 | #endif |
1da177e4 LT |
3772 | } |
3773 | ||
132380a0 | 3774 | notrace unsigned long get_parent_ip(unsigned long addr) |
6cd8a4bb SR |
3775 | { |
3776 | if (in_lock_functions(addr)) { | |
3777 | addr = CALLER_ADDR2; | |
3778 | if (in_lock_functions(addr)) | |
3779 | addr = CALLER_ADDR3; | |
3780 | } | |
3781 | return addr; | |
3782 | } | |
1da177e4 | 3783 | |
7e49fcce SR |
3784 | #if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ |
3785 | defined(CONFIG_PREEMPT_TRACER)) | |
3786 | ||
43627582 | 3787 | void __kprobes add_preempt_count(int val) |
1da177e4 | 3788 | { |
6cd8a4bb | 3789 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3790 | /* |
3791 | * Underflow? | |
3792 | */ | |
9a11b49a IM |
3793 | if (DEBUG_LOCKS_WARN_ON((preempt_count() < 0))) |
3794 | return; | |
6cd8a4bb | 3795 | #endif |
1da177e4 | 3796 | preempt_count() += val; |
6cd8a4bb | 3797 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3798 | /* |
3799 | * Spinlock count overflowing soon? | |
3800 | */ | |
33859f7f MOS |
3801 | DEBUG_LOCKS_WARN_ON((preempt_count() & PREEMPT_MASK) >= |
3802 | PREEMPT_MASK - 10); | |
6cd8a4bb SR |
3803 | #endif |
3804 | if (preempt_count() == val) | |
3805 | trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3806 | } |
3807 | EXPORT_SYMBOL(add_preempt_count); | |
3808 | ||
43627582 | 3809 | void __kprobes sub_preempt_count(int val) |
1da177e4 | 3810 | { |
6cd8a4bb | 3811 | #ifdef CONFIG_DEBUG_PREEMPT |
1da177e4 LT |
3812 | /* |
3813 | * Underflow? | |
3814 | */ | |
01e3eb82 | 3815 | if (DEBUG_LOCKS_WARN_ON(val > preempt_count())) |
9a11b49a | 3816 | return; |
1da177e4 LT |
3817 | /* |
3818 | * Is the spinlock portion underflowing? | |
3819 | */ | |
9a11b49a IM |
3820 | if (DEBUG_LOCKS_WARN_ON((val < PREEMPT_MASK) && |
3821 | !(preempt_count() & PREEMPT_MASK))) | |
3822 | return; | |
6cd8a4bb | 3823 | #endif |
9a11b49a | 3824 | |
6cd8a4bb SR |
3825 | if (preempt_count() == val) |
3826 | trace_preempt_on(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); | |
1da177e4 LT |
3827 | preempt_count() -= val; |
3828 | } | |
3829 | EXPORT_SYMBOL(sub_preempt_count); | |
3830 | ||
3831 | #endif | |
3832 | ||
3833 | /* | |
dd41f596 | 3834 | * Print scheduling while atomic bug: |
1da177e4 | 3835 | */ |
dd41f596 | 3836 | static noinline void __schedule_bug(struct task_struct *prev) |
1da177e4 | 3837 | { |
838225b4 SS |
3838 | struct pt_regs *regs = get_irq_regs(); |
3839 | ||
3df0fc5b PZ |
3840 | printk(KERN_ERR "BUG: scheduling while atomic: %s/%d/0x%08x\n", |
3841 | prev->comm, prev->pid, preempt_count()); | |
838225b4 | 3842 | |
dd41f596 | 3843 | debug_show_held_locks(prev); |
e21f5b15 | 3844 | print_modules(); |
dd41f596 IM |
3845 | if (irqs_disabled()) |
3846 | print_irqtrace_events(prev); | |
838225b4 SS |
3847 | |
3848 | if (regs) | |
3849 | show_regs(regs); | |
3850 | else | |
3851 | dump_stack(); | |
dd41f596 | 3852 | } |
1da177e4 | 3853 | |
dd41f596 IM |
3854 | /* |
3855 | * Various schedule()-time debugging checks and statistics: | |
3856 | */ | |
3857 | static inline void schedule_debug(struct task_struct *prev) | |
3858 | { | |
1da177e4 | 3859 | /* |
41a2d6cf | 3860 | * Test if we are atomic. Since do_exit() needs to call into |
1da177e4 LT |
3861 | * schedule() atomically, we ignore that path for now. |
3862 | * Otherwise, whine if we are scheduling when we should not be. | |
3863 | */ | |
3f33a7ce | 3864 | if (unlikely(in_atomic_preempt_off() && !prev->exit_state)) |
dd41f596 IM |
3865 | __schedule_bug(prev); |
3866 | ||
1da177e4 LT |
3867 | profile_hit(SCHED_PROFILING, __builtin_return_address(0)); |
3868 | ||
2d72376b | 3869 | schedstat_inc(this_rq(), sched_count); |
b8efb561 IM |
3870 | #ifdef CONFIG_SCHEDSTATS |
3871 | if (unlikely(prev->lock_depth >= 0)) { | |
fce20979 | 3872 | schedstat_inc(this_rq(), rq_sched_info.bkl_count); |
2d72376b | 3873 | schedstat_inc(prev, sched_info.bkl_count); |
b8efb561 IM |
3874 | } |
3875 | #endif | |
dd41f596 IM |
3876 | } |
3877 | ||
6cecd084 | 3878 | static void put_prev_task(struct rq *rq, struct task_struct *prev) |
df1c99d4 | 3879 | { |
a64692a3 MG |
3880 | if (prev->se.on_rq) |
3881 | update_rq_clock(rq); | |
6cecd084 | 3882 | prev->sched_class->put_prev_task(rq, prev); |
df1c99d4 MG |
3883 | } |
3884 | ||
dd41f596 IM |
3885 | /* |
3886 | * Pick up the highest-prio task: | |
3887 | */ | |
3888 | static inline struct task_struct * | |
b67802ea | 3889 | pick_next_task(struct rq *rq) |
dd41f596 | 3890 | { |
5522d5d5 | 3891 | const struct sched_class *class; |
dd41f596 | 3892 | struct task_struct *p; |
1da177e4 LT |
3893 | |
3894 | /* | |
dd41f596 IM |
3895 | * Optimization: we know that if all tasks are in |
3896 | * the fair class we can call that function directly: | |
1da177e4 | 3897 | */ |
dd41f596 | 3898 | if (likely(rq->nr_running == rq->cfs.nr_running)) { |
fb8d4724 | 3899 | p = fair_sched_class.pick_next_task(rq); |
dd41f596 IM |
3900 | if (likely(p)) |
3901 | return p; | |
1da177e4 LT |
3902 | } |
3903 | ||
34f971f6 | 3904 | for_each_class(class) { |
fb8d4724 | 3905 | p = class->pick_next_task(rq); |
dd41f596 IM |
3906 | if (p) |
3907 | return p; | |
dd41f596 | 3908 | } |
34f971f6 PZ |
3909 | |
3910 | BUG(); /* the idle class will always have a runnable task */ | |
dd41f596 | 3911 | } |
1da177e4 | 3912 | |
dd41f596 IM |
3913 | /* |
3914 | * schedule() is the main scheduler function. | |
3915 | */ | |
ff743345 | 3916 | asmlinkage void __sched schedule(void) |
dd41f596 IM |
3917 | { |
3918 | struct task_struct *prev, *next; | |
67ca7bde | 3919 | unsigned long *switch_count; |
dd41f596 | 3920 | struct rq *rq; |
31656519 | 3921 | int cpu; |
dd41f596 | 3922 | |
ff743345 PZ |
3923 | need_resched: |
3924 | preempt_disable(); | |
dd41f596 IM |
3925 | cpu = smp_processor_id(); |
3926 | rq = cpu_rq(cpu); | |
25502a6c | 3927 | rcu_note_context_switch(cpu); |
dd41f596 | 3928 | prev = rq->curr; |
dd41f596 IM |
3929 | |
3930 | release_kernel_lock(prev); | |
3931 | need_resched_nonpreemptible: | |
3932 | ||
3933 | schedule_debug(prev); | |
1da177e4 | 3934 | |
31656519 | 3935 | if (sched_feat(HRTICK)) |
f333fdc9 | 3936 | hrtick_clear(rq); |
8f4d37ec | 3937 | |
05fa785c | 3938 | raw_spin_lock_irq(&rq->lock); |
1da177e4 | 3939 | |
246d86b5 | 3940 | switch_count = &prev->nivcsw; |
1da177e4 | 3941 | if (prev->state && !(preempt_count() & PREEMPT_ACTIVE)) { |
21aa9af0 | 3942 | if (unlikely(signal_pending_state(prev->state, prev))) { |
1da177e4 | 3943 | prev->state = TASK_RUNNING; |
21aa9af0 TH |
3944 | } else { |
3945 | /* | |
3946 | * If a worker is going to sleep, notify and | |
3947 | * ask workqueue whether it wants to wake up a | |
3948 | * task to maintain concurrency. If so, wake | |
3949 | * up the task. | |
3950 | */ | |
3951 | if (prev->flags & PF_WQ_WORKER) { | |
3952 | struct task_struct *to_wakeup; | |
3953 | ||
3954 | to_wakeup = wq_worker_sleeping(prev, cpu); | |
3955 | if (to_wakeup) | |
3956 | try_to_wake_up_local(to_wakeup); | |
3957 | } | |
371fd7e7 | 3958 | deactivate_task(rq, prev, DEQUEUE_SLEEP); |
21aa9af0 | 3959 | } |
dd41f596 | 3960 | switch_count = &prev->nvcsw; |
1da177e4 LT |
3961 | } |
3962 | ||
3f029d3c | 3963 | pre_schedule(rq, prev); |
f65eda4f | 3964 | |
dd41f596 | 3965 | if (unlikely(!rq->nr_running)) |
1da177e4 | 3966 | idle_balance(cpu, rq); |
1da177e4 | 3967 | |
df1c99d4 | 3968 | put_prev_task(rq, prev); |
b67802ea | 3969 | next = pick_next_task(rq); |
f26f9aff MG |
3970 | clear_tsk_need_resched(prev); |
3971 | rq->skip_clock_update = 0; | |
1da177e4 | 3972 | |
1da177e4 | 3973 | if (likely(prev != next)) { |
1da177e4 LT |
3974 | rq->nr_switches++; |
3975 | rq->curr = next; | |
3976 | ++*switch_count; | |
3977 | ||
dd41f596 | 3978 | context_switch(rq, prev, next); /* unlocks the rq */ |
8f4d37ec | 3979 | /* |
246d86b5 ON |
3980 | * The context switch have flipped the stack from under us |
3981 | * and restored the local variables which were saved when | |
3982 | * this task called schedule() in the past. prev == current | |
3983 | * is still correct, but it can be moved to another cpu/rq. | |
8f4d37ec PZ |
3984 | */ |
3985 | cpu = smp_processor_id(); | |
3986 | rq = cpu_rq(cpu); | |
1da177e4 | 3987 | } else |
05fa785c | 3988 | raw_spin_unlock_irq(&rq->lock); |
1da177e4 | 3989 | |
3f029d3c | 3990 | post_schedule(rq); |
1da177e4 | 3991 | |
246d86b5 | 3992 | if (unlikely(reacquire_kernel_lock(prev))) |
1da177e4 | 3993 | goto need_resched_nonpreemptible; |
8f4d37ec | 3994 | |
1da177e4 | 3995 | preempt_enable_no_resched(); |
ff743345 | 3996 | if (need_resched()) |
1da177e4 LT |
3997 | goto need_resched; |
3998 | } | |
1da177e4 LT |
3999 | EXPORT_SYMBOL(schedule); |
4000 | ||
c08f7829 | 4001 | #ifdef CONFIG_MUTEX_SPIN_ON_OWNER |
0d66bf6d PZ |
4002 | /* |
4003 | * Look out! "owner" is an entirely speculative pointer | |
4004 | * access and not reliable. | |
4005 | */ | |
4006 | int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner) | |
4007 | { | |
4008 | unsigned int cpu; | |
4009 | struct rq *rq; | |
4010 | ||
4011 | if (!sched_feat(OWNER_SPIN)) | |
4012 | return 0; | |
4013 | ||
4014 | #ifdef CONFIG_DEBUG_PAGEALLOC | |
4015 | /* | |
4016 | * Need to access the cpu field knowing that | |
4017 | * DEBUG_PAGEALLOC could have unmapped it if | |
4018 | * the mutex owner just released it and exited. | |
4019 | */ | |
4020 | if (probe_kernel_address(&owner->cpu, cpu)) | |
4b402210 | 4021 | return 0; |
0d66bf6d PZ |
4022 | #else |
4023 | cpu = owner->cpu; | |
4024 | #endif | |
4025 | ||
4026 | /* | |
4027 | * Even if the access succeeded (likely case), | |
4028 | * the cpu field may no longer be valid. | |
4029 | */ | |
4030 | if (cpu >= nr_cpumask_bits) | |
4b402210 | 4031 | return 0; |
0d66bf6d PZ |
4032 | |
4033 | /* | |
4034 | * We need to validate that we can do a | |
4035 | * get_cpu() and that we have the percpu area. | |
4036 | */ | |
4037 | if (!cpu_online(cpu)) | |
4b402210 | 4038 | return 0; |
0d66bf6d PZ |
4039 | |
4040 | rq = cpu_rq(cpu); | |
4041 | ||
4042 | for (;;) { | |
4043 | /* | |
4044 | * Owner changed, break to re-assess state. | |
4045 | */ | |
9d0f4dcc TC |
4046 | if (lock->owner != owner) { |
4047 | /* | |
4048 | * If the lock has switched to a different owner, | |
4049 | * we likely have heavy contention. Return 0 to quit | |
4050 | * optimistic spinning and not contend further: | |
4051 | */ | |
4052 | if (lock->owner) | |
4053 | return 0; | |
0d66bf6d | 4054 | break; |
9d0f4dcc | 4055 | } |
0d66bf6d PZ |
4056 | |
4057 | /* | |
4058 | * Is that owner really running on that cpu? | |
4059 | */ | |
4060 | if (task_thread_info(rq->curr) != owner || need_resched()) | |
4061 | return 0; | |
4062 | ||
335d7afb | 4063 | arch_mutex_cpu_relax(); |
0d66bf6d | 4064 | } |
4b402210 | 4065 | |
0d66bf6d PZ |
4066 | return 1; |
4067 | } | |
4068 | #endif | |
4069 | ||
1da177e4 LT |
4070 | #ifdef CONFIG_PREEMPT |
4071 | /* | |
2ed6e34f | 4072 | * this is the entry point to schedule() from in-kernel preemption |
41a2d6cf | 4073 | * off of preempt_enable. Kernel preemptions off return from interrupt |
1da177e4 LT |
4074 | * occur there and call schedule directly. |
4075 | */ | |
d1f74e20 | 4076 | asmlinkage void __sched notrace preempt_schedule(void) |
1da177e4 LT |
4077 | { |
4078 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4079 | |
1da177e4 LT |
4080 | /* |
4081 | * If there is a non-zero preempt_count or interrupts are disabled, | |
41a2d6cf | 4082 | * we do not want to preempt the current task. Just return.. |
1da177e4 | 4083 | */ |
beed33a8 | 4084 | if (likely(ti->preempt_count || irqs_disabled())) |
1da177e4 LT |
4085 | return; |
4086 | ||
3a5c359a | 4087 | do { |
d1f74e20 | 4088 | add_preempt_count_notrace(PREEMPT_ACTIVE); |
3a5c359a | 4089 | schedule(); |
d1f74e20 | 4090 | sub_preempt_count_notrace(PREEMPT_ACTIVE); |
1da177e4 | 4091 | |
3a5c359a AK |
4092 | /* |
4093 | * Check again in case we missed a preemption opportunity | |
4094 | * between schedule and now. | |
4095 | */ | |
4096 | barrier(); | |
5ed0cec0 | 4097 | } while (need_resched()); |
1da177e4 | 4098 | } |
1da177e4 LT |
4099 | EXPORT_SYMBOL(preempt_schedule); |
4100 | ||
4101 | /* | |
2ed6e34f | 4102 | * this is the entry point to schedule() from kernel preemption |
1da177e4 LT |
4103 | * off of irq context. |
4104 | * Note, that this is called and return with irqs disabled. This will | |
4105 | * protect us against recursive calling from irq. | |
4106 | */ | |
4107 | asmlinkage void __sched preempt_schedule_irq(void) | |
4108 | { | |
4109 | struct thread_info *ti = current_thread_info(); | |
6478d880 | 4110 | |
2ed6e34f | 4111 | /* Catch callers which need to be fixed */ |
1da177e4 LT |
4112 | BUG_ON(ti->preempt_count || !irqs_disabled()); |
4113 | ||
3a5c359a AK |
4114 | do { |
4115 | add_preempt_count(PREEMPT_ACTIVE); | |
3a5c359a AK |
4116 | local_irq_enable(); |
4117 | schedule(); | |
4118 | local_irq_disable(); | |
3a5c359a | 4119 | sub_preempt_count(PREEMPT_ACTIVE); |
1da177e4 | 4120 | |
3a5c359a AK |
4121 | /* |
4122 | * Check again in case we missed a preemption opportunity | |
4123 | * between schedule and now. | |
4124 | */ | |
4125 | barrier(); | |
5ed0cec0 | 4126 | } while (need_resched()); |
1da177e4 LT |
4127 | } |
4128 | ||
4129 | #endif /* CONFIG_PREEMPT */ | |
4130 | ||
63859d4f | 4131 | int default_wake_function(wait_queue_t *curr, unsigned mode, int wake_flags, |
95cdf3b7 | 4132 | void *key) |
1da177e4 | 4133 | { |
63859d4f | 4134 | return try_to_wake_up(curr->private, mode, wake_flags); |
1da177e4 | 4135 | } |
1da177e4 LT |
4136 | EXPORT_SYMBOL(default_wake_function); |
4137 | ||
4138 | /* | |
41a2d6cf IM |
4139 | * The core wakeup function. Non-exclusive wakeups (nr_exclusive == 0) just |
4140 | * wake everything up. If it's an exclusive wakeup (nr_exclusive == small +ve | |
1da177e4 LT |
4141 | * number) then we wake all the non-exclusive tasks and one exclusive task. |
4142 | * | |
4143 | * There are circumstances in which we can try to wake a task which has already | |
41a2d6cf | 4144 | * started to run but is not in state TASK_RUNNING. try_to_wake_up() returns |
1da177e4 LT |
4145 | * zero in this (rare) case, and we handle it by continuing to scan the queue. |
4146 | */ | |
78ddb08f | 4147 | static void __wake_up_common(wait_queue_head_t *q, unsigned int mode, |
63859d4f | 4148 | int nr_exclusive, int wake_flags, void *key) |
1da177e4 | 4149 | { |
2e45874c | 4150 | wait_queue_t *curr, *next; |
1da177e4 | 4151 | |
2e45874c | 4152 | list_for_each_entry_safe(curr, next, &q->task_list, task_list) { |
48f24c4d IM |
4153 | unsigned flags = curr->flags; |
4154 | ||
63859d4f | 4155 | if (curr->func(curr, mode, wake_flags, key) && |
48f24c4d | 4156 | (flags & WQ_FLAG_EXCLUSIVE) && !--nr_exclusive) |
1da177e4 LT |
4157 | break; |
4158 | } | |
4159 | } | |
4160 | ||
4161 | /** | |
4162 | * __wake_up - wake up threads blocked on a waitqueue. | |
4163 | * @q: the waitqueue | |
4164 | * @mode: which threads | |
4165 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
67be2dd1 | 4166 | * @key: is directly passed to the wakeup function |
50fa610a DH |
4167 | * |
4168 | * It may be assumed that this function implies a write memory barrier before | |
4169 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4170 | */ |
7ad5b3a5 | 4171 | void __wake_up(wait_queue_head_t *q, unsigned int mode, |
95cdf3b7 | 4172 | int nr_exclusive, void *key) |
1da177e4 LT |
4173 | { |
4174 | unsigned long flags; | |
4175 | ||
4176 | spin_lock_irqsave(&q->lock, flags); | |
4177 | __wake_up_common(q, mode, nr_exclusive, 0, key); | |
4178 | spin_unlock_irqrestore(&q->lock, flags); | |
4179 | } | |
1da177e4 LT |
4180 | EXPORT_SYMBOL(__wake_up); |
4181 | ||
4182 | /* | |
4183 | * Same as __wake_up but called with the spinlock in wait_queue_head_t held. | |
4184 | */ | |
7ad5b3a5 | 4185 | void __wake_up_locked(wait_queue_head_t *q, unsigned int mode) |
1da177e4 LT |
4186 | { |
4187 | __wake_up_common(q, mode, 1, 0, NULL); | |
4188 | } | |
22c43c81 | 4189 | EXPORT_SYMBOL_GPL(__wake_up_locked); |
1da177e4 | 4190 | |
4ede816a DL |
4191 | void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key) |
4192 | { | |
4193 | __wake_up_common(q, mode, 1, 0, key); | |
4194 | } | |
4195 | ||
1da177e4 | 4196 | /** |
4ede816a | 4197 | * __wake_up_sync_key - wake up threads blocked on a waitqueue. |
1da177e4 LT |
4198 | * @q: the waitqueue |
4199 | * @mode: which threads | |
4200 | * @nr_exclusive: how many wake-one or wake-many threads to wake up | |
4ede816a | 4201 | * @key: opaque value to be passed to wakeup targets |
1da177e4 LT |
4202 | * |
4203 | * The sync wakeup differs that the waker knows that it will schedule | |
4204 | * away soon, so while the target thread will be woken up, it will not | |
4205 | * be migrated to another CPU - ie. the two threads are 'synchronized' | |
4206 | * with each other. This can prevent needless bouncing between CPUs. | |
4207 | * | |
4208 | * On UP it can prevent extra preemption. | |
50fa610a DH |
4209 | * |
4210 | * It may be assumed that this function implies a write memory barrier before | |
4211 | * changing the task state if and only if any tasks are woken up. | |
1da177e4 | 4212 | */ |
4ede816a DL |
4213 | void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode, |
4214 | int nr_exclusive, void *key) | |
1da177e4 LT |
4215 | { |
4216 | unsigned long flags; | |
7d478721 | 4217 | int wake_flags = WF_SYNC; |
1da177e4 LT |
4218 | |
4219 | if (unlikely(!q)) | |
4220 | return; | |
4221 | ||
4222 | if (unlikely(!nr_exclusive)) | |
7d478721 | 4223 | wake_flags = 0; |
1da177e4 LT |
4224 | |
4225 | spin_lock_irqsave(&q->lock, flags); | |
7d478721 | 4226 | __wake_up_common(q, mode, nr_exclusive, wake_flags, key); |
1da177e4 LT |
4227 | spin_unlock_irqrestore(&q->lock, flags); |
4228 | } | |
4ede816a DL |
4229 | EXPORT_SYMBOL_GPL(__wake_up_sync_key); |
4230 | ||
4231 | /* | |
4232 | * __wake_up_sync - see __wake_up_sync_key() | |
4233 | */ | |
4234 | void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) | |
4235 | { | |
4236 | __wake_up_sync_key(q, mode, nr_exclusive, NULL); | |
4237 | } | |
1da177e4 LT |
4238 | EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ |
4239 | ||
65eb3dc6 KD |
4240 | /** |
4241 | * complete: - signals a single thread waiting on this completion | |
4242 | * @x: holds the state of this particular completion | |
4243 | * | |
4244 | * This will wake up a single thread waiting on this completion. Threads will be | |
4245 | * awakened in the same order in which they were queued. | |
4246 | * | |
4247 | * See also complete_all(), wait_for_completion() and related routines. | |
50fa610a DH |
4248 | * |
4249 | * It may be assumed that this function implies a write memory barrier before | |
4250 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4251 | */ |
b15136e9 | 4252 | void complete(struct completion *x) |
1da177e4 LT |
4253 | { |
4254 | unsigned long flags; | |
4255 | ||
4256 | spin_lock_irqsave(&x->wait.lock, flags); | |
4257 | x->done++; | |
d9514f6c | 4258 | __wake_up_common(&x->wait, TASK_NORMAL, 1, 0, NULL); |
1da177e4 LT |
4259 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4260 | } | |
4261 | EXPORT_SYMBOL(complete); | |
4262 | ||
65eb3dc6 KD |
4263 | /** |
4264 | * complete_all: - signals all threads waiting on this completion | |
4265 | * @x: holds the state of this particular completion | |
4266 | * | |
4267 | * This will wake up all threads waiting on this particular completion event. | |
50fa610a DH |
4268 | * |
4269 | * It may be assumed that this function implies a write memory barrier before | |
4270 | * changing the task state if and only if any tasks are woken up. | |
65eb3dc6 | 4271 | */ |
b15136e9 | 4272 | void complete_all(struct completion *x) |
1da177e4 LT |
4273 | { |
4274 | unsigned long flags; | |
4275 | ||
4276 | spin_lock_irqsave(&x->wait.lock, flags); | |
4277 | x->done += UINT_MAX/2; | |
d9514f6c | 4278 | __wake_up_common(&x->wait, TASK_NORMAL, 0, 0, NULL); |
1da177e4 LT |
4279 | spin_unlock_irqrestore(&x->wait.lock, flags); |
4280 | } | |
4281 | EXPORT_SYMBOL(complete_all); | |
4282 | ||
8cbbe86d AK |
4283 | static inline long __sched |
4284 | do_wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4285 | { |
1da177e4 LT |
4286 | if (!x->done) { |
4287 | DECLARE_WAITQUEUE(wait, current); | |
4288 | ||
a93d2f17 | 4289 | __add_wait_queue_tail_exclusive(&x->wait, &wait); |
1da177e4 | 4290 | do { |
94d3d824 | 4291 | if (signal_pending_state(state, current)) { |
ea71a546 ON |
4292 | timeout = -ERESTARTSYS; |
4293 | break; | |
8cbbe86d AK |
4294 | } |
4295 | __set_current_state(state); | |
1da177e4 LT |
4296 | spin_unlock_irq(&x->wait.lock); |
4297 | timeout = schedule_timeout(timeout); | |
4298 | spin_lock_irq(&x->wait.lock); | |
ea71a546 | 4299 | } while (!x->done && timeout); |
1da177e4 | 4300 | __remove_wait_queue(&x->wait, &wait); |
ea71a546 ON |
4301 | if (!x->done) |
4302 | return timeout; | |
1da177e4 LT |
4303 | } |
4304 | x->done--; | |
ea71a546 | 4305 | return timeout ?: 1; |
1da177e4 | 4306 | } |
1da177e4 | 4307 | |
8cbbe86d AK |
4308 | static long __sched |
4309 | wait_for_common(struct completion *x, long timeout, int state) | |
1da177e4 | 4310 | { |
1da177e4 LT |
4311 | might_sleep(); |
4312 | ||
4313 | spin_lock_irq(&x->wait.lock); | |
8cbbe86d | 4314 | timeout = do_wait_for_common(x, timeout, state); |
1da177e4 | 4315 | spin_unlock_irq(&x->wait.lock); |
8cbbe86d AK |
4316 | return timeout; |
4317 | } | |
1da177e4 | 4318 | |
65eb3dc6 KD |
4319 | /** |
4320 | * wait_for_completion: - waits for completion of a task | |
4321 | * @x: holds the state of this particular completion | |
4322 | * | |
4323 | * This waits to be signaled for completion of a specific task. It is NOT | |
4324 | * interruptible and there is no timeout. | |
4325 | * | |
4326 | * See also similar routines (i.e. wait_for_completion_timeout()) with timeout | |
4327 | * and interrupt capability. Also see complete(). | |
4328 | */ | |
b15136e9 | 4329 | void __sched wait_for_completion(struct completion *x) |
8cbbe86d AK |
4330 | { |
4331 | wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_UNINTERRUPTIBLE); | |
1da177e4 | 4332 | } |
8cbbe86d | 4333 | EXPORT_SYMBOL(wait_for_completion); |
1da177e4 | 4334 | |
65eb3dc6 KD |
4335 | /** |
4336 | * wait_for_completion_timeout: - waits for completion of a task (w/timeout) | |
4337 | * @x: holds the state of this particular completion | |
4338 | * @timeout: timeout value in jiffies | |
4339 | * | |
4340 | * This waits for either a completion of a specific task to be signaled or for a | |
4341 | * specified timeout to expire. The timeout is in jiffies. It is not | |
4342 | * interruptible. | |
4343 | */ | |
b15136e9 | 4344 | unsigned long __sched |
8cbbe86d | 4345 | wait_for_completion_timeout(struct completion *x, unsigned long timeout) |
1da177e4 | 4346 | { |
8cbbe86d | 4347 | return wait_for_common(x, timeout, TASK_UNINTERRUPTIBLE); |
1da177e4 | 4348 | } |
8cbbe86d | 4349 | EXPORT_SYMBOL(wait_for_completion_timeout); |
1da177e4 | 4350 | |
65eb3dc6 KD |
4351 | /** |
4352 | * wait_for_completion_interruptible: - waits for completion of a task (w/intr) | |
4353 | * @x: holds the state of this particular completion | |
4354 | * | |
4355 | * This waits for completion of a specific task to be signaled. It is | |
4356 | * interruptible. | |
4357 | */ | |
8cbbe86d | 4358 | int __sched wait_for_completion_interruptible(struct completion *x) |
0fec171c | 4359 | { |
51e97990 AK |
4360 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_INTERRUPTIBLE); |
4361 | if (t == -ERESTARTSYS) | |
4362 | return t; | |
4363 | return 0; | |
0fec171c | 4364 | } |
8cbbe86d | 4365 | EXPORT_SYMBOL(wait_for_completion_interruptible); |
1da177e4 | 4366 | |
65eb3dc6 KD |
4367 | /** |
4368 | * wait_for_completion_interruptible_timeout: - waits for completion (w/(to,intr)) | |
4369 | * @x: holds the state of this particular completion | |
4370 | * @timeout: timeout value in jiffies | |
4371 | * | |
4372 | * This waits for either a completion of a specific task to be signaled or for a | |
4373 | * specified timeout to expire. It is interruptible. The timeout is in jiffies. | |
4374 | */ | |
6bf41237 | 4375 | long __sched |
8cbbe86d AK |
4376 | wait_for_completion_interruptible_timeout(struct completion *x, |
4377 | unsigned long timeout) | |
0fec171c | 4378 | { |
8cbbe86d | 4379 | return wait_for_common(x, timeout, TASK_INTERRUPTIBLE); |
0fec171c | 4380 | } |
8cbbe86d | 4381 | EXPORT_SYMBOL(wait_for_completion_interruptible_timeout); |
1da177e4 | 4382 | |
65eb3dc6 KD |
4383 | /** |
4384 | * wait_for_completion_killable: - waits for completion of a task (killable) | |
4385 | * @x: holds the state of this particular completion | |
4386 | * | |
4387 | * This waits to be signaled for completion of a specific task. It can be | |
4388 | * interrupted by a kill signal. | |
4389 | */ | |
009e577e MW |
4390 | int __sched wait_for_completion_killable(struct completion *x) |
4391 | { | |
4392 | long t = wait_for_common(x, MAX_SCHEDULE_TIMEOUT, TASK_KILLABLE); | |
4393 | if (t == -ERESTARTSYS) | |
4394 | return t; | |
4395 | return 0; | |
4396 | } | |
4397 | EXPORT_SYMBOL(wait_for_completion_killable); | |
4398 | ||
0aa12fb4 SW |
4399 | /** |
4400 | * wait_for_completion_killable_timeout: - waits for completion of a task (w/(to,killable)) | |
4401 | * @x: holds the state of this particular completion | |
4402 | * @timeout: timeout value in jiffies | |
4403 | * | |
4404 | * This waits for either a completion of a specific task to be | |
4405 | * signaled or for a specified timeout to expire. It can be | |
4406 | * interrupted by a kill signal. The timeout is in jiffies. | |
4407 | */ | |
6bf41237 | 4408 | long __sched |
0aa12fb4 SW |
4409 | wait_for_completion_killable_timeout(struct completion *x, |
4410 | unsigned long timeout) | |
4411 | { | |
4412 | return wait_for_common(x, timeout, TASK_KILLABLE); | |
4413 | } | |
4414 | EXPORT_SYMBOL(wait_for_completion_killable_timeout); | |
4415 | ||
be4de352 DC |
4416 | /** |
4417 | * try_wait_for_completion - try to decrement a completion without blocking | |
4418 | * @x: completion structure | |
4419 | * | |
4420 | * Returns: 0 if a decrement cannot be done without blocking | |
4421 | * 1 if a decrement succeeded. | |
4422 | * | |
4423 | * If a completion is being used as a counting completion, | |
4424 | * attempt to decrement the counter without blocking. This | |
4425 | * enables us to avoid waiting if the resource the completion | |
4426 | * is protecting is not available. | |
4427 | */ | |
4428 | bool try_wait_for_completion(struct completion *x) | |
4429 | { | |
7539a3b3 | 4430 | unsigned long flags; |
be4de352 DC |
4431 | int ret = 1; |
4432 | ||
7539a3b3 | 4433 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4434 | if (!x->done) |
4435 | ret = 0; | |
4436 | else | |
4437 | x->done--; | |
7539a3b3 | 4438 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4439 | return ret; |
4440 | } | |
4441 | EXPORT_SYMBOL(try_wait_for_completion); | |
4442 | ||
4443 | /** | |
4444 | * completion_done - Test to see if a completion has any waiters | |
4445 | * @x: completion structure | |
4446 | * | |
4447 | * Returns: 0 if there are waiters (wait_for_completion() in progress) | |
4448 | * 1 if there are no waiters. | |
4449 | * | |
4450 | */ | |
4451 | bool completion_done(struct completion *x) | |
4452 | { | |
7539a3b3 | 4453 | unsigned long flags; |
be4de352 DC |
4454 | int ret = 1; |
4455 | ||
7539a3b3 | 4456 | spin_lock_irqsave(&x->wait.lock, flags); |
be4de352 DC |
4457 | if (!x->done) |
4458 | ret = 0; | |
7539a3b3 | 4459 | spin_unlock_irqrestore(&x->wait.lock, flags); |
be4de352 DC |
4460 | return ret; |
4461 | } | |
4462 | EXPORT_SYMBOL(completion_done); | |
4463 | ||
8cbbe86d AK |
4464 | static long __sched |
4465 | sleep_on_common(wait_queue_head_t *q, int state, long timeout) | |
1da177e4 | 4466 | { |
0fec171c IM |
4467 | unsigned long flags; |
4468 | wait_queue_t wait; | |
4469 | ||
4470 | init_waitqueue_entry(&wait, current); | |
1da177e4 | 4471 | |
8cbbe86d | 4472 | __set_current_state(state); |
1da177e4 | 4473 | |
8cbbe86d AK |
4474 | spin_lock_irqsave(&q->lock, flags); |
4475 | __add_wait_queue(q, &wait); | |
4476 | spin_unlock(&q->lock); | |
4477 | timeout = schedule_timeout(timeout); | |
4478 | spin_lock_irq(&q->lock); | |
4479 | __remove_wait_queue(q, &wait); | |
4480 | spin_unlock_irqrestore(&q->lock, flags); | |
4481 | ||
4482 | return timeout; | |
4483 | } | |
4484 | ||
4485 | void __sched interruptible_sleep_on(wait_queue_head_t *q) | |
4486 | { | |
4487 | sleep_on_common(q, TASK_INTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); | |
1da177e4 | 4488 | } |
1da177e4 LT |
4489 | EXPORT_SYMBOL(interruptible_sleep_on); |
4490 | ||
0fec171c | 4491 | long __sched |
95cdf3b7 | 4492 | interruptible_sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4493 | { |
8cbbe86d | 4494 | return sleep_on_common(q, TASK_INTERRUPTIBLE, timeout); |
1da177e4 | 4495 | } |
1da177e4 LT |
4496 | EXPORT_SYMBOL(interruptible_sleep_on_timeout); |
4497 | ||
0fec171c | 4498 | void __sched sleep_on(wait_queue_head_t *q) |
1da177e4 | 4499 | { |
8cbbe86d | 4500 | sleep_on_common(q, TASK_UNINTERRUPTIBLE, MAX_SCHEDULE_TIMEOUT); |
1da177e4 | 4501 | } |
1da177e4 LT |
4502 | EXPORT_SYMBOL(sleep_on); |
4503 | ||
0fec171c | 4504 | long __sched sleep_on_timeout(wait_queue_head_t *q, long timeout) |
1da177e4 | 4505 | { |
8cbbe86d | 4506 | return sleep_on_common(q, TASK_UNINTERRUPTIBLE, timeout); |
1da177e4 | 4507 | } |
1da177e4 LT |
4508 | EXPORT_SYMBOL(sleep_on_timeout); |
4509 | ||
b29739f9 IM |
4510 | #ifdef CONFIG_RT_MUTEXES |
4511 | ||
4512 | /* | |
4513 | * rt_mutex_setprio - set the current priority of a task | |
4514 | * @p: task | |
4515 | * @prio: prio value (kernel-internal form) | |
4516 | * | |
4517 | * This function changes the 'effective' priority of a task. It does | |
4518 | * not touch ->normal_prio like __setscheduler(). | |
4519 | * | |
4520 | * Used by the rt_mutex code to implement priority inheritance logic. | |
4521 | */ | |
36c8b586 | 4522 | void rt_mutex_setprio(struct task_struct *p, int prio) |
b29739f9 IM |
4523 | { |
4524 | unsigned long flags; | |
83b699ed | 4525 | int oldprio, on_rq, running; |
70b97a7f | 4526 | struct rq *rq; |
83ab0aa0 | 4527 | const struct sched_class *prev_class; |
b29739f9 IM |
4528 | |
4529 | BUG_ON(prio < 0 || prio > MAX_PRIO); | |
4530 | ||
4531 | rq = task_rq_lock(p, &flags); | |
4532 | ||
a8027073 | 4533 | trace_sched_pi_setprio(p, prio); |
d5f9f942 | 4534 | oldprio = p->prio; |
83ab0aa0 | 4535 | prev_class = p->sched_class; |
dd41f596 | 4536 | on_rq = p->se.on_rq; |
051a1d1a | 4537 | running = task_current(rq, p); |
0e1f3483 | 4538 | if (on_rq) |
69be72c1 | 4539 | dequeue_task(rq, p, 0); |
0e1f3483 HS |
4540 | if (running) |
4541 | p->sched_class->put_prev_task(rq, p); | |
dd41f596 IM |
4542 | |
4543 | if (rt_prio(prio)) | |
4544 | p->sched_class = &rt_sched_class; | |
4545 | else | |
4546 | p->sched_class = &fair_sched_class; | |
4547 | ||
b29739f9 IM |
4548 | p->prio = prio; |
4549 | ||
0e1f3483 HS |
4550 | if (running) |
4551 | p->sched_class->set_curr_task(rq); | |
dd41f596 | 4552 | if (on_rq) { |
371fd7e7 | 4553 | enqueue_task(rq, p, oldprio < prio ? ENQUEUE_HEAD : 0); |
cb469845 SR |
4554 | |
4555 | check_class_changed(rq, p, prev_class, oldprio, running); | |
b29739f9 IM |
4556 | } |
4557 | task_rq_unlock(rq, &flags); | |
4558 | } | |
4559 | ||
4560 | #endif | |
4561 | ||
36c8b586 | 4562 | void set_user_nice(struct task_struct *p, long nice) |
1da177e4 | 4563 | { |
dd41f596 | 4564 | int old_prio, delta, on_rq; |
1da177e4 | 4565 | unsigned long flags; |
70b97a7f | 4566 | struct rq *rq; |
1da177e4 LT |
4567 | |
4568 | if (TASK_NICE(p) == nice || nice < -20 || nice > 19) | |
4569 | return; | |
4570 | /* | |
4571 | * We have to be careful, if called from sys_setpriority(), | |
4572 | * the task might be in the middle of scheduling on another CPU. | |
4573 | */ | |
4574 | rq = task_rq_lock(p, &flags); | |
4575 | /* | |
4576 | * The RT priorities are set via sched_setscheduler(), but we still | |
4577 | * allow the 'normal' nice value to be set - but as expected | |
4578 | * it wont have any effect on scheduling until the task is | |
dd41f596 | 4579 | * SCHED_FIFO/SCHED_RR: |
1da177e4 | 4580 | */ |
e05606d3 | 4581 | if (task_has_rt_policy(p)) { |
1da177e4 LT |
4582 | p->static_prio = NICE_TO_PRIO(nice); |
4583 | goto out_unlock; | |
4584 | } | |
dd41f596 | 4585 | on_rq = p->se.on_rq; |
c09595f6 | 4586 | if (on_rq) |
69be72c1 | 4587 | dequeue_task(rq, p, 0); |
1da177e4 | 4588 | |
1da177e4 | 4589 | p->static_prio = NICE_TO_PRIO(nice); |
2dd73a4f | 4590 | set_load_weight(p); |
b29739f9 IM |
4591 | old_prio = p->prio; |
4592 | p->prio = effective_prio(p); | |
4593 | delta = p->prio - old_prio; | |
1da177e4 | 4594 | |
dd41f596 | 4595 | if (on_rq) { |
371fd7e7 | 4596 | enqueue_task(rq, p, 0); |
1da177e4 | 4597 | /* |
d5f9f942 AM |
4598 | * If the task increased its priority or is running and |
4599 | * lowered its priority, then reschedule its CPU: | |
1da177e4 | 4600 | */ |
d5f9f942 | 4601 | if (delta < 0 || (delta > 0 && task_running(rq, p))) |
1da177e4 LT |
4602 | resched_task(rq->curr); |
4603 | } | |
4604 | out_unlock: | |
4605 | task_rq_unlock(rq, &flags); | |
4606 | } | |
1da177e4 LT |
4607 | EXPORT_SYMBOL(set_user_nice); |
4608 | ||
e43379f1 MM |
4609 | /* |
4610 | * can_nice - check if a task can reduce its nice value | |
4611 | * @p: task | |
4612 | * @nice: nice value | |
4613 | */ | |
36c8b586 | 4614 | int can_nice(const struct task_struct *p, const int nice) |
e43379f1 | 4615 | { |
024f4747 MM |
4616 | /* convert nice value [19,-20] to rlimit style value [1,40] */ |
4617 | int nice_rlim = 20 - nice; | |
48f24c4d | 4618 | |
78d7d407 | 4619 | return (nice_rlim <= task_rlimit(p, RLIMIT_NICE) || |
e43379f1 MM |
4620 | capable(CAP_SYS_NICE)); |
4621 | } | |
4622 | ||
1da177e4 LT |
4623 | #ifdef __ARCH_WANT_SYS_NICE |
4624 | ||
4625 | /* | |
4626 | * sys_nice - change the priority of the current process. | |
4627 | * @increment: priority increment | |
4628 | * | |
4629 | * sys_setpriority is a more generic, but much slower function that | |
4630 | * does similar things. | |
4631 | */ | |
5add95d4 | 4632 | SYSCALL_DEFINE1(nice, int, increment) |
1da177e4 | 4633 | { |
48f24c4d | 4634 | long nice, retval; |
1da177e4 LT |
4635 | |
4636 | /* | |
4637 | * Setpriority might change our priority at the same moment. | |
4638 | * We don't have to worry. Conceptually one call occurs first | |
4639 | * and we have a single winner. | |
4640 | */ | |
e43379f1 MM |
4641 | if (increment < -40) |
4642 | increment = -40; | |
1da177e4 LT |
4643 | if (increment > 40) |
4644 | increment = 40; | |
4645 | ||
2b8f836f | 4646 | nice = TASK_NICE(current) + increment; |
1da177e4 LT |
4647 | if (nice < -20) |
4648 | nice = -20; | |
4649 | if (nice > 19) | |
4650 | nice = 19; | |
4651 | ||
e43379f1 MM |
4652 | if (increment < 0 && !can_nice(current, nice)) |
4653 | return -EPERM; | |
4654 | ||
1da177e4 LT |
4655 | retval = security_task_setnice(current, nice); |
4656 | if (retval) | |
4657 | return retval; | |
4658 | ||
4659 | set_user_nice(current, nice); | |
4660 | return 0; | |
4661 | } | |
4662 | ||
4663 | #endif | |
4664 | ||
4665 | /** | |
4666 | * task_prio - return the priority value of a given task. | |
4667 | * @p: the task in question. | |
4668 | * | |
4669 | * This is the priority value as seen by users in /proc. | |
4670 | * RT tasks are offset by -200. Normal tasks are centered | |
4671 | * around 0, value goes from -16 to +15. | |
4672 | */ | |
36c8b586 | 4673 | int task_prio(const struct task_struct *p) |
1da177e4 LT |
4674 | { |
4675 | return p->prio - MAX_RT_PRIO; | |
4676 | } | |
4677 | ||
4678 | /** | |
4679 | * task_nice - return the nice value of a given task. | |
4680 | * @p: the task in question. | |
4681 | */ | |
36c8b586 | 4682 | int task_nice(const struct task_struct *p) |
1da177e4 LT |
4683 | { |
4684 | return TASK_NICE(p); | |
4685 | } | |
150d8bed | 4686 | EXPORT_SYMBOL(task_nice); |
1da177e4 LT |
4687 | |
4688 | /** | |
4689 | * idle_cpu - is a given cpu idle currently? | |
4690 | * @cpu: the processor in question. | |
4691 | */ | |
4692 | int idle_cpu(int cpu) | |
4693 | { | |
4694 | return cpu_curr(cpu) == cpu_rq(cpu)->idle; | |
4695 | } | |
4696 | ||
1da177e4 LT |
4697 | /** |
4698 | * idle_task - return the idle task for a given cpu. | |
4699 | * @cpu: the processor in question. | |
4700 | */ | |
36c8b586 | 4701 | struct task_struct *idle_task(int cpu) |
1da177e4 LT |
4702 | { |
4703 | return cpu_rq(cpu)->idle; | |
4704 | } | |
4705 | ||
4706 | /** | |
4707 | * find_process_by_pid - find a process with a matching PID value. | |
4708 | * @pid: the pid in question. | |
4709 | */ | |
a9957449 | 4710 | static struct task_struct *find_process_by_pid(pid_t pid) |
1da177e4 | 4711 | { |
228ebcbe | 4712 | return pid ? find_task_by_vpid(pid) : current; |
1da177e4 LT |
4713 | } |
4714 | ||
4715 | /* Actually do priority change: must hold rq lock. */ | |
dd41f596 IM |
4716 | static void |
4717 | __setscheduler(struct rq *rq, struct task_struct *p, int policy, int prio) | |
1da177e4 | 4718 | { |
dd41f596 | 4719 | BUG_ON(p->se.on_rq); |
48f24c4d | 4720 | |
1da177e4 LT |
4721 | p->policy = policy; |
4722 | p->rt_priority = prio; | |
b29739f9 IM |
4723 | p->normal_prio = normal_prio(p); |
4724 | /* we are holding p->pi_lock already */ | |
4725 | p->prio = rt_mutex_getprio(p); | |
ffd44db5 PZ |
4726 | if (rt_prio(p->prio)) |
4727 | p->sched_class = &rt_sched_class; | |
4728 | else | |
4729 | p->sched_class = &fair_sched_class; | |
2dd73a4f | 4730 | set_load_weight(p); |
1da177e4 LT |
4731 | } |
4732 | ||
c69e8d9c DH |
4733 | /* |
4734 | * check the target process has a UID that matches the current process's | |
4735 | */ | |
4736 | static bool check_same_owner(struct task_struct *p) | |
4737 | { | |
4738 | const struct cred *cred = current_cred(), *pcred; | |
4739 | bool match; | |
4740 | ||
4741 | rcu_read_lock(); | |
4742 | pcred = __task_cred(p); | |
4743 | match = (cred->euid == pcred->euid || | |
4744 | cred->euid == pcred->uid); | |
4745 | rcu_read_unlock(); | |
4746 | return match; | |
4747 | } | |
4748 | ||
961ccddd | 4749 | static int __sched_setscheduler(struct task_struct *p, int policy, |
fe7de49f | 4750 | const struct sched_param *param, bool user) |
1da177e4 | 4751 | { |
83b699ed | 4752 | int retval, oldprio, oldpolicy = -1, on_rq, running; |
1da177e4 | 4753 | unsigned long flags; |
83ab0aa0 | 4754 | const struct sched_class *prev_class; |
70b97a7f | 4755 | struct rq *rq; |
ca94c442 | 4756 | int reset_on_fork; |
1da177e4 | 4757 | |
66e5393a SR |
4758 | /* may grab non-irq protected spin_locks */ |
4759 | BUG_ON(in_interrupt()); | |
1da177e4 LT |
4760 | recheck: |
4761 | /* double check policy once rq lock held */ | |
ca94c442 LP |
4762 | if (policy < 0) { |
4763 | reset_on_fork = p->sched_reset_on_fork; | |
1da177e4 | 4764 | policy = oldpolicy = p->policy; |
ca94c442 LP |
4765 | } else { |
4766 | reset_on_fork = !!(policy & SCHED_RESET_ON_FORK); | |
4767 | policy &= ~SCHED_RESET_ON_FORK; | |
4768 | ||
4769 | if (policy != SCHED_FIFO && policy != SCHED_RR && | |
4770 | policy != SCHED_NORMAL && policy != SCHED_BATCH && | |
4771 | policy != SCHED_IDLE) | |
4772 | return -EINVAL; | |
4773 | } | |
4774 | ||
1da177e4 LT |
4775 | /* |
4776 | * Valid priorities for SCHED_FIFO and SCHED_RR are | |
dd41f596 IM |
4777 | * 1..MAX_USER_RT_PRIO-1, valid priority for SCHED_NORMAL, |
4778 | * SCHED_BATCH and SCHED_IDLE is 0. | |
1da177e4 LT |
4779 | */ |
4780 | if (param->sched_priority < 0 || | |
95cdf3b7 | 4781 | (p->mm && param->sched_priority > MAX_USER_RT_PRIO-1) || |
d46523ea | 4782 | (!p->mm && param->sched_priority > MAX_RT_PRIO-1)) |
1da177e4 | 4783 | return -EINVAL; |
e05606d3 | 4784 | if (rt_policy(policy) != (param->sched_priority != 0)) |
1da177e4 LT |
4785 | return -EINVAL; |
4786 | ||
37e4ab3f OC |
4787 | /* |
4788 | * Allow unprivileged RT tasks to decrease priority: | |
4789 | */ | |
961ccddd | 4790 | if (user && !capable(CAP_SYS_NICE)) { |
e05606d3 | 4791 | if (rt_policy(policy)) { |
a44702e8 ON |
4792 | unsigned long rlim_rtprio = |
4793 | task_rlimit(p, RLIMIT_RTPRIO); | |
8dc3e909 ON |
4794 | |
4795 | /* can't set/change the rt policy */ | |
4796 | if (policy != p->policy && !rlim_rtprio) | |
4797 | return -EPERM; | |
4798 | ||
4799 | /* can't increase priority */ | |
4800 | if (param->sched_priority > p->rt_priority && | |
4801 | param->sched_priority > rlim_rtprio) | |
4802 | return -EPERM; | |
4803 | } | |
dd41f596 IM |
4804 | /* |
4805 | * Like positive nice levels, dont allow tasks to | |
4806 | * move out of SCHED_IDLE either: | |
4807 | */ | |
4808 | if (p->policy == SCHED_IDLE && policy != SCHED_IDLE) | |
4809 | return -EPERM; | |
5fe1d75f | 4810 | |
37e4ab3f | 4811 | /* can't change other user's priorities */ |
c69e8d9c | 4812 | if (!check_same_owner(p)) |
37e4ab3f | 4813 | return -EPERM; |
ca94c442 LP |
4814 | |
4815 | /* Normal users shall not reset the sched_reset_on_fork flag */ | |
4816 | if (p->sched_reset_on_fork && !reset_on_fork) | |
4817 | return -EPERM; | |
37e4ab3f | 4818 | } |
1da177e4 | 4819 | |
725aad24 | 4820 | if (user) { |
b0ae1981 | 4821 | retval = security_task_setscheduler(p); |
725aad24 JF |
4822 | if (retval) |
4823 | return retval; | |
4824 | } | |
4825 | ||
b29739f9 IM |
4826 | /* |
4827 | * make sure no PI-waiters arrive (or leave) while we are | |
4828 | * changing the priority of the task: | |
4829 | */ | |
1d615482 | 4830 | raw_spin_lock_irqsave(&p->pi_lock, flags); |
1da177e4 LT |
4831 | /* |
4832 | * To be able to change p->policy safely, the apropriate | |
4833 | * runqueue lock must be held. | |
4834 | */ | |
b29739f9 | 4835 | rq = __task_rq_lock(p); |
dc61b1d6 | 4836 | |
34f971f6 PZ |
4837 | /* |
4838 | * Changing the policy of the stop threads its a very bad idea | |
4839 | */ | |
4840 | if (p == rq->stop) { | |
4841 | __task_rq_unlock(rq); | |
4842 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4843 | return -EINVAL; | |
4844 | } | |
4845 | ||
dc61b1d6 PZ |
4846 | #ifdef CONFIG_RT_GROUP_SCHED |
4847 | if (user) { | |
4848 | /* | |
4849 | * Do not allow realtime tasks into groups that have no runtime | |
4850 | * assigned. | |
4851 | */ | |
4852 | if (rt_bandwidth_enabled() && rt_policy(policy) && | |
f4493771 MG |
4853 | task_group(p)->rt_bandwidth.rt_runtime == 0 && |
4854 | !task_group_is_autogroup(task_group(p))) { | |
dc61b1d6 PZ |
4855 | __task_rq_unlock(rq); |
4856 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); | |
4857 | return -EPERM; | |
4858 | } | |
4859 | } | |
4860 | #endif | |
4861 | ||
1da177e4 LT |
4862 | /* recheck policy now with rq lock held */ |
4863 | if (unlikely(oldpolicy != -1 && oldpolicy != p->policy)) { | |
4864 | policy = oldpolicy = -1; | |
b29739f9 | 4865 | __task_rq_unlock(rq); |
1d615482 | 4866 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
1da177e4 LT |
4867 | goto recheck; |
4868 | } | |
dd41f596 | 4869 | on_rq = p->se.on_rq; |
051a1d1a | 4870 | running = task_current(rq, p); |
0e1f3483 | 4871 | if (on_rq) |
2e1cb74a | 4872 | deactivate_task(rq, p, 0); |
0e1f3483 HS |
4873 | if (running) |
4874 | p->sched_class->put_prev_task(rq, p); | |
f6b53205 | 4875 | |
ca94c442 LP |
4876 | p->sched_reset_on_fork = reset_on_fork; |
4877 | ||
1da177e4 | 4878 | oldprio = p->prio; |
83ab0aa0 | 4879 | prev_class = p->sched_class; |
dd41f596 | 4880 | __setscheduler(rq, p, policy, param->sched_priority); |
f6b53205 | 4881 | |
0e1f3483 HS |
4882 | if (running) |
4883 | p->sched_class->set_curr_task(rq); | |
dd41f596 IM |
4884 | if (on_rq) { |
4885 | activate_task(rq, p, 0); | |
cb469845 SR |
4886 | |
4887 | check_class_changed(rq, p, prev_class, oldprio, running); | |
1da177e4 | 4888 | } |
b29739f9 | 4889 | __task_rq_unlock(rq); |
1d615482 | 4890 | raw_spin_unlock_irqrestore(&p->pi_lock, flags); |
b29739f9 | 4891 | |
95e02ca9 TG |
4892 | rt_mutex_adjust_pi(p); |
4893 | ||
1da177e4 LT |
4894 | return 0; |
4895 | } | |
961ccddd RR |
4896 | |
4897 | /** | |
4898 | * sched_setscheduler - change the scheduling policy and/or RT priority of a thread. | |
4899 | * @p: the task in question. | |
4900 | * @policy: new policy. | |
4901 | * @param: structure containing the new RT priority. | |
4902 | * | |
4903 | * NOTE that the task may be already dead. | |
4904 | */ | |
4905 | int sched_setscheduler(struct task_struct *p, int policy, | |
fe7de49f | 4906 | const struct sched_param *param) |
961ccddd RR |
4907 | { |
4908 | return __sched_setscheduler(p, policy, param, true); | |
4909 | } | |
1da177e4 LT |
4910 | EXPORT_SYMBOL_GPL(sched_setscheduler); |
4911 | ||
961ccddd RR |
4912 | /** |
4913 | * sched_setscheduler_nocheck - change the scheduling policy and/or RT priority of a thread from kernelspace. | |
4914 | * @p: the task in question. | |
4915 | * @policy: new policy. | |
4916 | * @param: structure containing the new RT priority. | |
4917 | * | |
4918 | * Just like sched_setscheduler, only don't bother checking if the | |
4919 | * current context has permission. For example, this is needed in | |
4920 | * stop_machine(): we create temporary high priority worker threads, | |
4921 | * but our caller might not have that capability. | |
4922 | */ | |
4923 | int sched_setscheduler_nocheck(struct task_struct *p, int policy, | |
fe7de49f | 4924 | const struct sched_param *param) |
961ccddd RR |
4925 | { |
4926 | return __sched_setscheduler(p, policy, param, false); | |
4927 | } | |
4928 | ||
95cdf3b7 IM |
4929 | static int |
4930 | do_sched_setscheduler(pid_t pid, int policy, struct sched_param __user *param) | |
1da177e4 | 4931 | { |
1da177e4 LT |
4932 | struct sched_param lparam; |
4933 | struct task_struct *p; | |
36c8b586 | 4934 | int retval; |
1da177e4 LT |
4935 | |
4936 | if (!param || pid < 0) | |
4937 | return -EINVAL; | |
4938 | if (copy_from_user(&lparam, param, sizeof(struct sched_param))) | |
4939 | return -EFAULT; | |
5fe1d75f ON |
4940 | |
4941 | rcu_read_lock(); | |
4942 | retval = -ESRCH; | |
1da177e4 | 4943 | p = find_process_by_pid(pid); |
5fe1d75f ON |
4944 | if (p != NULL) |
4945 | retval = sched_setscheduler(p, policy, &lparam); | |
4946 | rcu_read_unlock(); | |
36c8b586 | 4947 | |
1da177e4 LT |
4948 | return retval; |
4949 | } | |
4950 | ||
4951 | /** | |
4952 | * sys_sched_setscheduler - set/change the scheduler policy and RT priority | |
4953 | * @pid: the pid in question. | |
4954 | * @policy: new policy. | |
4955 | * @param: structure containing the new RT priority. | |
4956 | */ | |
5add95d4 HC |
4957 | SYSCALL_DEFINE3(sched_setscheduler, pid_t, pid, int, policy, |
4958 | struct sched_param __user *, param) | |
1da177e4 | 4959 | { |
c21761f1 JB |
4960 | /* negative values for policy are not valid */ |
4961 | if (policy < 0) | |
4962 | return -EINVAL; | |
4963 | ||
1da177e4 LT |
4964 | return do_sched_setscheduler(pid, policy, param); |
4965 | } | |
4966 | ||
4967 | /** | |
4968 | * sys_sched_setparam - set/change the RT priority of a thread | |
4969 | * @pid: the pid in question. | |
4970 | * @param: structure containing the new RT priority. | |
4971 | */ | |
5add95d4 | 4972 | SYSCALL_DEFINE2(sched_setparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
4973 | { |
4974 | return do_sched_setscheduler(pid, -1, param); | |
4975 | } | |
4976 | ||
4977 | /** | |
4978 | * sys_sched_getscheduler - get the policy (scheduling class) of a thread | |
4979 | * @pid: the pid in question. | |
4980 | */ | |
5add95d4 | 4981 | SYSCALL_DEFINE1(sched_getscheduler, pid_t, pid) |
1da177e4 | 4982 | { |
36c8b586 | 4983 | struct task_struct *p; |
3a5c359a | 4984 | int retval; |
1da177e4 LT |
4985 | |
4986 | if (pid < 0) | |
3a5c359a | 4987 | return -EINVAL; |
1da177e4 LT |
4988 | |
4989 | retval = -ESRCH; | |
5fe85be0 | 4990 | rcu_read_lock(); |
1da177e4 LT |
4991 | p = find_process_by_pid(pid); |
4992 | if (p) { | |
4993 | retval = security_task_getscheduler(p); | |
4994 | if (!retval) | |
ca94c442 LP |
4995 | retval = p->policy |
4996 | | (p->sched_reset_on_fork ? SCHED_RESET_ON_FORK : 0); | |
1da177e4 | 4997 | } |
5fe85be0 | 4998 | rcu_read_unlock(); |
1da177e4 LT |
4999 | return retval; |
5000 | } | |
5001 | ||
5002 | /** | |
ca94c442 | 5003 | * sys_sched_getparam - get the RT priority of a thread |
1da177e4 LT |
5004 | * @pid: the pid in question. |
5005 | * @param: structure containing the RT priority. | |
5006 | */ | |
5add95d4 | 5007 | SYSCALL_DEFINE2(sched_getparam, pid_t, pid, struct sched_param __user *, param) |
1da177e4 LT |
5008 | { |
5009 | struct sched_param lp; | |
36c8b586 | 5010 | struct task_struct *p; |
3a5c359a | 5011 | int retval; |
1da177e4 LT |
5012 | |
5013 | if (!param || pid < 0) | |
3a5c359a | 5014 | return -EINVAL; |
1da177e4 | 5015 | |
5fe85be0 | 5016 | rcu_read_lock(); |
1da177e4 LT |
5017 | p = find_process_by_pid(pid); |
5018 | retval = -ESRCH; | |
5019 | if (!p) | |
5020 | goto out_unlock; | |
5021 | ||
5022 | retval = security_task_getscheduler(p); | |
5023 | if (retval) | |
5024 | goto out_unlock; | |
5025 | ||
5026 | lp.sched_priority = p->rt_priority; | |
5fe85be0 | 5027 | rcu_read_unlock(); |
1da177e4 LT |
5028 | |
5029 | /* | |
5030 | * This one might sleep, we cannot do it with a spinlock held ... | |
5031 | */ | |
5032 | retval = copy_to_user(param, &lp, sizeof(*param)) ? -EFAULT : 0; | |
5033 | ||
1da177e4 LT |
5034 | return retval; |
5035 | ||
5036 | out_unlock: | |
5fe85be0 | 5037 | rcu_read_unlock(); |
1da177e4 LT |
5038 | return retval; |
5039 | } | |
5040 | ||
96f874e2 | 5041 | long sched_setaffinity(pid_t pid, const struct cpumask *in_mask) |
1da177e4 | 5042 | { |
5a16f3d3 | 5043 | cpumask_var_t cpus_allowed, new_mask; |
36c8b586 IM |
5044 | struct task_struct *p; |
5045 | int retval; | |
1da177e4 | 5046 | |
95402b38 | 5047 | get_online_cpus(); |
23f5d142 | 5048 | rcu_read_lock(); |
1da177e4 LT |
5049 | |
5050 | p = find_process_by_pid(pid); | |
5051 | if (!p) { | |
23f5d142 | 5052 | rcu_read_unlock(); |
95402b38 | 5053 | put_online_cpus(); |
1da177e4 LT |
5054 | return -ESRCH; |
5055 | } | |
5056 | ||
23f5d142 | 5057 | /* Prevent p going away */ |
1da177e4 | 5058 | get_task_struct(p); |
23f5d142 | 5059 | rcu_read_unlock(); |
1da177e4 | 5060 | |
5a16f3d3 RR |
5061 | if (!alloc_cpumask_var(&cpus_allowed, GFP_KERNEL)) { |
5062 | retval = -ENOMEM; | |
5063 | goto out_put_task; | |
5064 | } | |
5065 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) { | |
5066 | retval = -ENOMEM; | |
5067 | goto out_free_cpus_allowed; | |
5068 | } | |
1da177e4 | 5069 | retval = -EPERM; |
c69e8d9c | 5070 | if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) |
1da177e4 LT |
5071 | goto out_unlock; |
5072 | ||
b0ae1981 | 5073 | retval = security_task_setscheduler(p); |
e7834f8f DQ |
5074 | if (retval) |
5075 | goto out_unlock; | |
5076 | ||
5a16f3d3 RR |
5077 | cpuset_cpus_allowed(p, cpus_allowed); |
5078 | cpumask_and(new_mask, in_mask, cpus_allowed); | |
49246274 | 5079 | again: |
5a16f3d3 | 5080 | retval = set_cpus_allowed_ptr(p, new_mask); |
1da177e4 | 5081 | |
8707d8b8 | 5082 | if (!retval) { |
5a16f3d3 RR |
5083 | cpuset_cpus_allowed(p, cpus_allowed); |
5084 | if (!cpumask_subset(new_mask, cpus_allowed)) { | |
8707d8b8 PM |
5085 | /* |
5086 | * We must have raced with a concurrent cpuset | |
5087 | * update. Just reset the cpus_allowed to the | |
5088 | * cpuset's cpus_allowed | |
5089 | */ | |
5a16f3d3 | 5090 | cpumask_copy(new_mask, cpus_allowed); |
8707d8b8 PM |
5091 | goto again; |
5092 | } | |
5093 | } | |
1da177e4 | 5094 | out_unlock: |
5a16f3d3 RR |
5095 | free_cpumask_var(new_mask); |
5096 | out_free_cpus_allowed: | |
5097 | free_cpumask_var(cpus_allowed); | |
5098 | out_put_task: | |
1da177e4 | 5099 | put_task_struct(p); |
95402b38 | 5100 | put_online_cpus(); |
1da177e4 LT |
5101 | return retval; |
5102 | } | |
5103 | ||
5104 | static int get_user_cpu_mask(unsigned long __user *user_mask_ptr, unsigned len, | |
96f874e2 | 5105 | struct cpumask *new_mask) |
1da177e4 | 5106 | { |
96f874e2 RR |
5107 | if (len < cpumask_size()) |
5108 | cpumask_clear(new_mask); | |
5109 | else if (len > cpumask_size()) | |
5110 | len = cpumask_size(); | |
5111 | ||
1da177e4 LT |
5112 | return copy_from_user(new_mask, user_mask_ptr, len) ? -EFAULT : 0; |
5113 | } | |
5114 | ||
5115 | /** | |
5116 | * sys_sched_setaffinity - set the cpu affinity of a process | |
5117 | * @pid: pid of the process | |
5118 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5119 | * @user_mask_ptr: user-space pointer to the new cpu mask | |
5120 | */ | |
5add95d4 HC |
5121 | SYSCALL_DEFINE3(sched_setaffinity, pid_t, pid, unsigned int, len, |
5122 | unsigned long __user *, user_mask_ptr) | |
1da177e4 | 5123 | { |
5a16f3d3 | 5124 | cpumask_var_t new_mask; |
1da177e4 LT |
5125 | int retval; |
5126 | ||
5a16f3d3 RR |
5127 | if (!alloc_cpumask_var(&new_mask, GFP_KERNEL)) |
5128 | return -ENOMEM; | |
1da177e4 | 5129 | |
5a16f3d3 RR |
5130 | retval = get_user_cpu_mask(user_mask_ptr, len, new_mask); |
5131 | if (retval == 0) | |
5132 | retval = sched_setaffinity(pid, new_mask); | |
5133 | free_cpumask_var(new_mask); | |
5134 | return retval; | |
1da177e4 LT |
5135 | } |
5136 | ||
96f874e2 | 5137 | long sched_getaffinity(pid_t pid, struct cpumask *mask) |
1da177e4 | 5138 | { |
36c8b586 | 5139 | struct task_struct *p; |
31605683 TG |
5140 | unsigned long flags; |
5141 | struct rq *rq; | |
1da177e4 | 5142 | int retval; |
1da177e4 | 5143 | |
95402b38 | 5144 | get_online_cpus(); |
23f5d142 | 5145 | rcu_read_lock(); |
1da177e4 LT |
5146 | |
5147 | retval = -ESRCH; | |
5148 | p = find_process_by_pid(pid); | |
5149 | if (!p) | |
5150 | goto out_unlock; | |
5151 | ||
e7834f8f DQ |
5152 | retval = security_task_getscheduler(p); |
5153 | if (retval) | |
5154 | goto out_unlock; | |
5155 | ||
31605683 | 5156 | rq = task_rq_lock(p, &flags); |
96f874e2 | 5157 | cpumask_and(mask, &p->cpus_allowed, cpu_online_mask); |
31605683 | 5158 | task_rq_unlock(rq, &flags); |
1da177e4 LT |
5159 | |
5160 | out_unlock: | |
23f5d142 | 5161 | rcu_read_unlock(); |
95402b38 | 5162 | put_online_cpus(); |
1da177e4 | 5163 | |
9531b62f | 5164 | return retval; |
1da177e4 LT |
5165 | } |
5166 | ||
5167 | /** | |
5168 | * sys_sched_getaffinity - get the cpu affinity of a process | |
5169 | * @pid: pid of the process | |
5170 | * @len: length in bytes of the bitmask pointed to by user_mask_ptr | |
5171 | * @user_mask_ptr: user-space pointer to hold the current cpu mask | |
5172 | */ | |
5add95d4 HC |
5173 | SYSCALL_DEFINE3(sched_getaffinity, pid_t, pid, unsigned int, len, |
5174 | unsigned long __user *, user_mask_ptr) | |
1da177e4 LT |
5175 | { |
5176 | int ret; | |
f17c8607 | 5177 | cpumask_var_t mask; |
1da177e4 | 5178 | |
84fba5ec | 5179 | if ((len * BITS_PER_BYTE) < nr_cpu_ids) |
cd3d8031 KM |
5180 | return -EINVAL; |
5181 | if (len & (sizeof(unsigned long)-1)) | |
1da177e4 LT |
5182 | return -EINVAL; |
5183 | ||
f17c8607 RR |
5184 | if (!alloc_cpumask_var(&mask, GFP_KERNEL)) |
5185 | return -ENOMEM; | |
1da177e4 | 5186 | |
f17c8607 RR |
5187 | ret = sched_getaffinity(pid, mask); |
5188 | if (ret == 0) { | |
8bc037fb | 5189 | size_t retlen = min_t(size_t, len, cpumask_size()); |
cd3d8031 KM |
5190 | |
5191 | if (copy_to_user(user_mask_ptr, mask, retlen)) | |
f17c8607 RR |
5192 | ret = -EFAULT; |
5193 | else | |
cd3d8031 | 5194 | ret = retlen; |
f17c8607 RR |
5195 | } |
5196 | free_cpumask_var(mask); | |
1da177e4 | 5197 | |
f17c8607 | 5198 | return ret; |
1da177e4 LT |
5199 | } |
5200 | ||
5201 | /** | |
5202 | * sys_sched_yield - yield the current processor to other threads. | |
5203 | * | |
dd41f596 IM |
5204 | * This function yields the current CPU to other tasks. If there are no |
5205 | * other threads running on this CPU then this function will return. | |
1da177e4 | 5206 | */ |
5add95d4 | 5207 | SYSCALL_DEFINE0(sched_yield) |
1da177e4 | 5208 | { |
70b97a7f | 5209 | struct rq *rq = this_rq_lock(); |
1da177e4 | 5210 | |
2d72376b | 5211 | schedstat_inc(rq, yld_count); |
4530d7ab | 5212 | current->sched_class->yield_task(rq); |
1da177e4 LT |
5213 | |
5214 | /* | |
5215 | * Since we are going to call schedule() anyway, there's | |
5216 | * no need to preempt or enable interrupts: | |
5217 | */ | |
5218 | __release(rq->lock); | |
8a25d5de | 5219 | spin_release(&rq->lock.dep_map, 1, _THIS_IP_); |
9828ea9d | 5220 | do_raw_spin_unlock(&rq->lock); |
1da177e4 LT |
5221 | preempt_enable_no_resched(); |
5222 | ||
5223 | schedule(); | |
5224 | ||
5225 | return 0; | |
5226 | } | |
5227 | ||
d86ee480 PZ |
5228 | static inline int should_resched(void) |
5229 | { | |
5230 | return need_resched() && !(preempt_count() & PREEMPT_ACTIVE); | |
5231 | } | |
5232 | ||
e7b38404 | 5233 | static void __cond_resched(void) |
1da177e4 | 5234 | { |
e7aaaa69 FW |
5235 | add_preempt_count(PREEMPT_ACTIVE); |
5236 | schedule(); | |
5237 | sub_preempt_count(PREEMPT_ACTIVE); | |
1da177e4 LT |
5238 | } |
5239 | ||
02b67cc3 | 5240 | int __sched _cond_resched(void) |
1da177e4 | 5241 | { |
d86ee480 | 5242 | if (should_resched()) { |
1da177e4 LT |
5243 | __cond_resched(); |
5244 | return 1; | |
5245 | } | |
5246 | return 0; | |
5247 | } | |
02b67cc3 | 5248 | EXPORT_SYMBOL(_cond_resched); |
1da177e4 LT |
5249 | |
5250 | /* | |
613afbf8 | 5251 | * __cond_resched_lock() - if a reschedule is pending, drop the given lock, |
1da177e4 LT |
5252 | * call schedule, and on return reacquire the lock. |
5253 | * | |
41a2d6cf | 5254 | * This works OK both with and without CONFIG_PREEMPT. We do strange low-level |
1da177e4 LT |
5255 | * operations here to prevent schedule() from being called twice (once via |
5256 | * spin_unlock(), once by hand). | |
5257 | */ | |
613afbf8 | 5258 | int __cond_resched_lock(spinlock_t *lock) |
1da177e4 | 5259 | { |
d86ee480 | 5260 | int resched = should_resched(); |
6df3cecb JK |
5261 | int ret = 0; |
5262 | ||
f607c668 PZ |
5263 | lockdep_assert_held(lock); |
5264 | ||
95c354fe | 5265 | if (spin_needbreak(lock) || resched) { |
1da177e4 | 5266 | spin_unlock(lock); |
d86ee480 | 5267 | if (resched) |
95c354fe NP |
5268 | __cond_resched(); |
5269 | else | |
5270 | cpu_relax(); | |
6df3cecb | 5271 | ret = 1; |
1da177e4 | 5272 | spin_lock(lock); |
1da177e4 | 5273 | } |
6df3cecb | 5274 | return ret; |
1da177e4 | 5275 | } |
613afbf8 | 5276 | EXPORT_SYMBOL(__cond_resched_lock); |
1da177e4 | 5277 | |
613afbf8 | 5278 | int __sched __cond_resched_softirq(void) |
1da177e4 LT |
5279 | { |
5280 | BUG_ON(!in_softirq()); | |
5281 | ||
d86ee480 | 5282 | if (should_resched()) { |
98d82567 | 5283 | local_bh_enable(); |
1da177e4 LT |
5284 | __cond_resched(); |
5285 | local_bh_disable(); | |
5286 | return 1; | |
5287 | } | |
5288 | return 0; | |
5289 | } | |
613afbf8 | 5290 | EXPORT_SYMBOL(__cond_resched_softirq); |
1da177e4 | 5291 | |
1da177e4 LT |
5292 | /** |
5293 | * yield - yield the current processor to other threads. | |
5294 | * | |
72fd4a35 | 5295 | * This is a shortcut for kernel-space yielding - it marks the |
1da177e4 LT |
5296 | * thread runnable and calls sys_sched_yield(). |
5297 | */ | |
5298 | void __sched yield(void) | |
5299 | { | |
5300 | set_current_state(TASK_RUNNING); | |
5301 | sys_sched_yield(); | |
5302 | } | |
1da177e4 LT |
5303 | EXPORT_SYMBOL(yield); |
5304 | ||
5305 | /* | |
41a2d6cf | 5306 | * This task is about to go to sleep on IO. Increment rq->nr_iowait so |
1da177e4 | 5307 | * that process accounting knows that this is a task in IO wait state. |
1da177e4 LT |
5308 | */ |
5309 | void __sched io_schedule(void) | |
5310 | { | |
54d35f29 | 5311 | struct rq *rq = raw_rq(); |
1da177e4 | 5312 | |
0ff92245 | 5313 | delayacct_blkio_start(); |
1da177e4 | 5314 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5315 | current->in_iowait = 1; |
1da177e4 | 5316 | schedule(); |
8f0dfc34 | 5317 | current->in_iowait = 0; |
1da177e4 | 5318 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5319 | delayacct_blkio_end(); |
1da177e4 | 5320 | } |
1da177e4 LT |
5321 | EXPORT_SYMBOL(io_schedule); |
5322 | ||
5323 | long __sched io_schedule_timeout(long timeout) | |
5324 | { | |
54d35f29 | 5325 | struct rq *rq = raw_rq(); |
1da177e4 LT |
5326 | long ret; |
5327 | ||
0ff92245 | 5328 | delayacct_blkio_start(); |
1da177e4 | 5329 | atomic_inc(&rq->nr_iowait); |
8f0dfc34 | 5330 | current->in_iowait = 1; |
1da177e4 | 5331 | ret = schedule_timeout(timeout); |
8f0dfc34 | 5332 | current->in_iowait = 0; |
1da177e4 | 5333 | atomic_dec(&rq->nr_iowait); |
0ff92245 | 5334 | delayacct_blkio_end(); |
1da177e4 LT |
5335 | return ret; |
5336 | } | |
5337 | ||
5338 | /** | |
5339 | * sys_sched_get_priority_max - return maximum RT priority. | |
5340 | * @policy: scheduling class. | |
5341 | * | |
5342 | * this syscall returns the maximum rt_priority that can be used | |
5343 | * by a given scheduling class. | |
5344 | */ | |
5add95d4 | 5345 | SYSCALL_DEFINE1(sched_get_priority_max, int, policy) |
1da177e4 LT |
5346 | { |
5347 | int ret = -EINVAL; | |
5348 | ||
5349 | switch (policy) { | |
5350 | case SCHED_FIFO: | |
5351 | case SCHED_RR: | |
5352 | ret = MAX_USER_RT_PRIO-1; | |
5353 | break; | |
5354 | case SCHED_NORMAL: | |
b0a9499c | 5355 | case SCHED_BATCH: |
dd41f596 | 5356 | case SCHED_IDLE: |
1da177e4 LT |
5357 | ret = 0; |
5358 | break; | |
5359 | } | |
5360 | return ret; | |
5361 | } | |
5362 | ||
5363 | /** | |
5364 | * sys_sched_get_priority_min - return minimum RT priority. | |
5365 | * @policy: scheduling class. | |
5366 | * | |
5367 | * this syscall returns the minimum rt_priority that can be used | |
5368 | * by a given scheduling class. | |
5369 | */ | |
5add95d4 | 5370 | SYSCALL_DEFINE1(sched_get_priority_min, int, policy) |
1da177e4 LT |
5371 | { |
5372 | int ret = -EINVAL; | |
5373 | ||
5374 | switch (policy) { | |
5375 | case SCHED_FIFO: | |
5376 | case SCHED_RR: | |
5377 | ret = 1; | |
5378 | break; | |
5379 | case SCHED_NORMAL: | |
b0a9499c | 5380 | case SCHED_BATCH: |
dd41f596 | 5381 | case SCHED_IDLE: |
1da177e4 LT |
5382 | ret = 0; |
5383 | } | |
5384 | return ret; | |
5385 | } | |
5386 | ||
5387 | /** | |
5388 | * sys_sched_rr_get_interval - return the default timeslice of a process. | |
5389 | * @pid: pid of the process. | |
5390 | * @interval: userspace pointer to the timeslice value. | |
5391 | * | |
5392 | * this syscall writes the default timeslice value of a given process | |
5393 | * into the user-space timespec buffer. A value of '0' means infinity. | |
5394 | */ | |
17da2bd9 | 5395 | SYSCALL_DEFINE2(sched_rr_get_interval, pid_t, pid, |
754fe8d2 | 5396 | struct timespec __user *, interval) |
1da177e4 | 5397 | { |
36c8b586 | 5398 | struct task_struct *p; |
a4ec24b4 | 5399 | unsigned int time_slice; |
dba091b9 TG |
5400 | unsigned long flags; |
5401 | struct rq *rq; | |
3a5c359a | 5402 | int retval; |
1da177e4 | 5403 | struct timespec t; |
1da177e4 LT |
5404 | |
5405 | if (pid < 0) | |
3a5c359a | 5406 | return -EINVAL; |
1da177e4 LT |
5407 | |
5408 | retval = -ESRCH; | |
1a551ae7 | 5409 | rcu_read_lock(); |
1da177e4 LT |
5410 | p = find_process_by_pid(pid); |
5411 | if (!p) | |
5412 | goto out_unlock; | |
5413 | ||
5414 | retval = security_task_getscheduler(p); | |
5415 | if (retval) | |
5416 | goto out_unlock; | |
5417 | ||
dba091b9 TG |
5418 | rq = task_rq_lock(p, &flags); |
5419 | time_slice = p->sched_class->get_rr_interval(rq, p); | |
5420 | task_rq_unlock(rq, &flags); | |
a4ec24b4 | 5421 | |
1a551ae7 | 5422 | rcu_read_unlock(); |
a4ec24b4 | 5423 | jiffies_to_timespec(time_slice, &t); |
1da177e4 | 5424 | retval = copy_to_user(interval, &t, sizeof(t)) ? -EFAULT : 0; |
1da177e4 | 5425 | return retval; |
3a5c359a | 5426 | |
1da177e4 | 5427 | out_unlock: |
1a551ae7 | 5428 | rcu_read_unlock(); |
1da177e4 LT |
5429 | return retval; |
5430 | } | |
5431 | ||
7c731e0a | 5432 | static const char stat_nam[] = TASK_STATE_TO_CHAR_STR; |
36c8b586 | 5433 | |
82a1fcb9 | 5434 | void sched_show_task(struct task_struct *p) |
1da177e4 | 5435 | { |
1da177e4 | 5436 | unsigned long free = 0; |
36c8b586 | 5437 | unsigned state; |
1da177e4 | 5438 | |
1da177e4 | 5439 | state = p->state ? __ffs(p->state) + 1 : 0; |
28d0686c | 5440 | printk(KERN_INFO "%-15.15s %c", p->comm, |
2ed6e34f | 5441 | state < sizeof(stat_nam) - 1 ? stat_nam[state] : '?'); |
4bd77321 | 5442 | #if BITS_PER_LONG == 32 |
1da177e4 | 5443 | if (state == TASK_RUNNING) |
3df0fc5b | 5444 | printk(KERN_CONT " running "); |
1da177e4 | 5445 | else |
3df0fc5b | 5446 | printk(KERN_CONT " %08lx ", thread_saved_pc(p)); |
1da177e4 LT |
5447 | #else |
5448 | if (state == TASK_RUNNING) | |
3df0fc5b | 5449 | printk(KERN_CONT " running task "); |
1da177e4 | 5450 | else |
3df0fc5b | 5451 | printk(KERN_CONT " %016lx ", thread_saved_pc(p)); |
1da177e4 LT |
5452 | #endif |
5453 | #ifdef CONFIG_DEBUG_STACK_USAGE | |
7c9f8861 | 5454 | free = stack_not_used(p); |
1da177e4 | 5455 | #endif |
3df0fc5b | 5456 | printk(KERN_CONT "%5lu %5d %6d 0x%08lx\n", free, |
aa47b7e0 DR |
5457 | task_pid_nr(p), task_pid_nr(p->real_parent), |
5458 | (unsigned long)task_thread_info(p)->flags); | |
1da177e4 | 5459 | |
5fb5e6de | 5460 | show_stack(p, NULL); |
1da177e4 LT |
5461 | } |
5462 | ||
e59e2ae2 | 5463 | void show_state_filter(unsigned long state_filter) |
1da177e4 | 5464 | { |
36c8b586 | 5465 | struct task_struct *g, *p; |
1da177e4 | 5466 | |
4bd77321 | 5467 | #if BITS_PER_LONG == 32 |
3df0fc5b PZ |
5468 | printk(KERN_INFO |
5469 | " task PC stack pid father\n"); | |
1da177e4 | 5470 | #else |
3df0fc5b PZ |
5471 | printk(KERN_INFO |
5472 | " task PC stack pid father\n"); | |
1da177e4 LT |
5473 | #endif |
5474 | read_lock(&tasklist_lock); | |
5475 | do_each_thread(g, p) { | |
5476 | /* | |
5477 | * reset the NMI-timeout, listing all files on a slow | |
5478 | * console might take alot of time: | |
5479 | */ | |
5480 | touch_nmi_watchdog(); | |
39bc89fd | 5481 | if (!state_filter || (p->state & state_filter)) |
82a1fcb9 | 5482 | sched_show_task(p); |
1da177e4 LT |
5483 | } while_each_thread(g, p); |
5484 | ||
04c9167f JF |
5485 | touch_all_softlockup_watchdogs(); |
5486 | ||
dd41f596 IM |
5487 | #ifdef CONFIG_SCHED_DEBUG |
5488 | sysrq_sched_debug_show(); | |
5489 | #endif | |
1da177e4 | 5490 | read_unlock(&tasklist_lock); |
e59e2ae2 IM |
5491 | /* |
5492 | * Only show locks if all tasks are dumped: | |
5493 | */ | |
93335a21 | 5494 | if (!state_filter) |
e59e2ae2 | 5495 | debug_show_all_locks(); |
1da177e4 LT |
5496 | } |
5497 | ||
1df21055 IM |
5498 | void __cpuinit init_idle_bootup_task(struct task_struct *idle) |
5499 | { | |
dd41f596 | 5500 | idle->sched_class = &idle_sched_class; |
1df21055 IM |
5501 | } |
5502 | ||
f340c0d1 IM |
5503 | /** |
5504 | * init_idle - set up an idle thread for a given CPU | |
5505 | * @idle: task in question | |
5506 | * @cpu: cpu the idle task belongs to | |
5507 | * | |
5508 | * NOTE: this function does not set the idle thread's NEED_RESCHED | |
5509 | * flag, to make booting more robust. | |
5510 | */ | |
5c1e1767 | 5511 | void __cpuinit init_idle(struct task_struct *idle, int cpu) |
1da177e4 | 5512 | { |
70b97a7f | 5513 | struct rq *rq = cpu_rq(cpu); |
1da177e4 LT |
5514 | unsigned long flags; |
5515 | ||
05fa785c | 5516 | raw_spin_lock_irqsave(&rq->lock, flags); |
5cbd54ef | 5517 | |
dd41f596 | 5518 | __sched_fork(idle); |
06b83b5f | 5519 | idle->state = TASK_RUNNING; |
dd41f596 IM |
5520 | idle->se.exec_start = sched_clock(); |
5521 | ||
96f874e2 | 5522 | cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); |
6506cf6c PZ |
5523 | /* |
5524 | * We're having a chicken and egg problem, even though we are | |
5525 | * holding rq->lock, the cpu isn't yet set to this cpu so the | |
5526 | * lockdep check in task_group() will fail. | |
5527 | * | |
5528 | * Similar case to sched_fork(). / Alternatively we could | |
5529 | * use task_rq_lock() here and obtain the other rq->lock. | |
5530 | * | |
5531 | * Silence PROVE_RCU | |
5532 | */ | |
5533 | rcu_read_lock(); | |
dd41f596 | 5534 | __set_task_cpu(idle, cpu); |
6506cf6c | 5535 | rcu_read_unlock(); |
1da177e4 | 5536 | |
1da177e4 | 5537 | rq->curr = rq->idle = idle; |
4866cde0 NP |
5538 | #if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) |
5539 | idle->oncpu = 1; | |
5540 | #endif | |
05fa785c | 5541 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 LT |
5542 | |
5543 | /* Set the preempt count _outside_ the spinlocks! */ | |
8e3e076c LT |
5544 | #if defined(CONFIG_PREEMPT) |
5545 | task_thread_info(idle)->preempt_count = (idle->lock_depth >= 0); | |
5546 | #else | |
a1261f54 | 5547 | task_thread_info(idle)->preempt_count = 0; |
8e3e076c | 5548 | #endif |
dd41f596 IM |
5549 | /* |
5550 | * The idle tasks have their own, simple scheduling class: | |
5551 | */ | |
5552 | idle->sched_class = &idle_sched_class; | |
868baf07 | 5553 | ftrace_graph_init_idle_task(idle, cpu); |
1da177e4 LT |
5554 | } |
5555 | ||
5556 | /* | |
5557 | * In a system that switches off the HZ timer nohz_cpu_mask | |
5558 | * indicates which cpus entered this state. This is used | |
5559 | * in the rcu update to wait only for active cpus. For system | |
5560 | * which do not switch off the HZ timer nohz_cpu_mask should | |
6a7b3dc3 | 5561 | * always be CPU_BITS_NONE. |
1da177e4 | 5562 | */ |
6a7b3dc3 | 5563 | cpumask_var_t nohz_cpu_mask; |
1da177e4 | 5564 | |
19978ca6 IM |
5565 | /* |
5566 | * Increase the granularity value when there are more CPUs, | |
5567 | * because with more CPUs the 'effective latency' as visible | |
5568 | * to users decreases. But the relationship is not linear, | |
5569 | * so pick a second-best guess by going with the log2 of the | |
5570 | * number of CPUs. | |
5571 | * | |
5572 | * This idea comes from the SD scheduler of Con Kolivas: | |
5573 | */ | |
acb4a848 | 5574 | static int get_update_sysctl_factor(void) |
19978ca6 | 5575 | { |
4ca3ef71 | 5576 | unsigned int cpus = min_t(int, num_online_cpus(), 8); |
1983a922 CE |
5577 | unsigned int factor; |
5578 | ||
5579 | switch (sysctl_sched_tunable_scaling) { | |
5580 | case SCHED_TUNABLESCALING_NONE: | |
5581 | factor = 1; | |
5582 | break; | |
5583 | case SCHED_TUNABLESCALING_LINEAR: | |
5584 | factor = cpus; | |
5585 | break; | |
5586 | case SCHED_TUNABLESCALING_LOG: | |
5587 | default: | |
5588 | factor = 1 + ilog2(cpus); | |
5589 | break; | |
5590 | } | |
19978ca6 | 5591 | |
acb4a848 CE |
5592 | return factor; |
5593 | } | |
19978ca6 | 5594 | |
acb4a848 CE |
5595 | static void update_sysctl(void) |
5596 | { | |
5597 | unsigned int factor = get_update_sysctl_factor(); | |
19978ca6 | 5598 | |
0bcdcf28 CE |
5599 | #define SET_SYSCTL(name) \ |
5600 | (sysctl_##name = (factor) * normalized_sysctl_##name) | |
5601 | SET_SYSCTL(sched_min_granularity); | |
5602 | SET_SYSCTL(sched_latency); | |
5603 | SET_SYSCTL(sched_wakeup_granularity); | |
0bcdcf28 CE |
5604 | #undef SET_SYSCTL |
5605 | } | |
55cd5340 | 5606 | |
0bcdcf28 CE |
5607 | static inline void sched_init_granularity(void) |
5608 | { | |
5609 | update_sysctl(); | |
19978ca6 IM |
5610 | } |
5611 | ||
1da177e4 LT |
5612 | #ifdef CONFIG_SMP |
5613 | /* | |
5614 | * This is how migration works: | |
5615 | * | |
969c7921 TH |
5616 | * 1) we invoke migration_cpu_stop() on the target CPU using |
5617 | * stop_one_cpu(). | |
5618 | * 2) stopper starts to run (implicitly forcing the migrated thread | |
5619 | * off the CPU) | |
5620 | * 3) it checks whether the migrated task is still in the wrong runqueue. | |
5621 | * 4) if it's in the wrong runqueue then the migration thread removes | |
1da177e4 | 5622 | * it and puts it into the right queue. |
969c7921 TH |
5623 | * 5) stopper completes and stop_one_cpu() returns and the migration |
5624 | * is done. | |
1da177e4 LT |
5625 | */ |
5626 | ||
5627 | /* | |
5628 | * Change a given task's CPU affinity. Migrate the thread to a | |
5629 | * proper CPU and schedule it away if the CPU it's executing on | |
5630 | * is removed from the allowed bitmask. | |
5631 | * | |
5632 | * NOTE: the caller must have a valid reference to the task, the | |
41a2d6cf | 5633 | * task must not exit() & deallocate itself prematurely. The |
1da177e4 LT |
5634 | * call is not atomic; no spinlocks may be held. |
5635 | */ | |
96f874e2 | 5636 | int set_cpus_allowed_ptr(struct task_struct *p, const struct cpumask *new_mask) |
1da177e4 LT |
5637 | { |
5638 | unsigned long flags; | |
70b97a7f | 5639 | struct rq *rq; |
969c7921 | 5640 | unsigned int dest_cpu; |
48f24c4d | 5641 | int ret = 0; |
1da177e4 | 5642 | |
65cc8e48 PZ |
5643 | /* |
5644 | * Serialize against TASK_WAKING so that ttwu() and wunt() can | |
5645 | * drop the rq->lock and still rely on ->cpus_allowed. | |
5646 | */ | |
5647 | again: | |
5648 | while (task_is_waking(p)) | |
5649 | cpu_relax(); | |
1da177e4 | 5650 | rq = task_rq_lock(p, &flags); |
65cc8e48 PZ |
5651 | if (task_is_waking(p)) { |
5652 | task_rq_unlock(rq, &flags); | |
5653 | goto again; | |
5654 | } | |
e2912009 | 5655 | |
6ad4c188 | 5656 | if (!cpumask_intersects(new_mask, cpu_active_mask)) { |
1da177e4 LT |
5657 | ret = -EINVAL; |
5658 | goto out; | |
5659 | } | |
5660 | ||
9985b0ba | 5661 | if (unlikely((p->flags & PF_THREAD_BOUND) && p != current && |
96f874e2 | 5662 | !cpumask_equal(&p->cpus_allowed, new_mask))) { |
9985b0ba DR |
5663 | ret = -EINVAL; |
5664 | goto out; | |
5665 | } | |
5666 | ||
73fe6aae | 5667 | if (p->sched_class->set_cpus_allowed) |
cd8ba7cd | 5668 | p->sched_class->set_cpus_allowed(p, new_mask); |
73fe6aae | 5669 | else { |
96f874e2 RR |
5670 | cpumask_copy(&p->cpus_allowed, new_mask); |
5671 | p->rt.nr_cpus_allowed = cpumask_weight(new_mask); | |
73fe6aae GH |
5672 | } |
5673 | ||
1da177e4 | 5674 | /* Can the task run on the task's current CPU? If so, we're done */ |
96f874e2 | 5675 | if (cpumask_test_cpu(task_cpu(p), new_mask)) |
1da177e4 LT |
5676 | goto out; |
5677 | ||
969c7921 | 5678 | dest_cpu = cpumask_any_and(cpu_active_mask, new_mask); |
b7a2b39d | 5679 | if (migrate_task(p, rq)) { |
969c7921 | 5680 | struct migration_arg arg = { p, dest_cpu }; |
1da177e4 LT |
5681 | /* Need help from migration thread: drop lock and wait. */ |
5682 | task_rq_unlock(rq, &flags); | |
969c7921 | 5683 | stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg); |
1da177e4 LT |
5684 | tlb_migrate_finish(p->mm); |
5685 | return 0; | |
5686 | } | |
5687 | out: | |
5688 | task_rq_unlock(rq, &flags); | |
48f24c4d | 5689 | |
1da177e4 LT |
5690 | return ret; |
5691 | } | |
cd8ba7cd | 5692 | EXPORT_SYMBOL_GPL(set_cpus_allowed_ptr); |
1da177e4 LT |
5693 | |
5694 | /* | |
41a2d6cf | 5695 | * Move (not current) task off this cpu, onto dest cpu. We're doing |
1da177e4 LT |
5696 | * this because either it can't run here any more (set_cpus_allowed() |
5697 | * away from this CPU, or CPU going down), or because we're | |
5698 | * attempting to rebalance this task on exec (sched_exec). | |
5699 | * | |
5700 | * So we race with normal scheduler movements, but that's OK, as long | |
5701 | * as the task is no longer on this CPU. | |
efc30814 KK |
5702 | * |
5703 | * Returns non-zero if task was successfully migrated. | |
1da177e4 | 5704 | */ |
efc30814 | 5705 | static int __migrate_task(struct task_struct *p, int src_cpu, int dest_cpu) |
1da177e4 | 5706 | { |
70b97a7f | 5707 | struct rq *rq_dest, *rq_src; |
e2912009 | 5708 | int ret = 0; |
1da177e4 | 5709 | |
e761b772 | 5710 | if (unlikely(!cpu_active(dest_cpu))) |
efc30814 | 5711 | return ret; |
1da177e4 LT |
5712 | |
5713 | rq_src = cpu_rq(src_cpu); | |
5714 | rq_dest = cpu_rq(dest_cpu); | |
5715 | ||
5716 | double_rq_lock(rq_src, rq_dest); | |
5717 | /* Already moved. */ | |
5718 | if (task_cpu(p) != src_cpu) | |
b1e38734 | 5719 | goto done; |
1da177e4 | 5720 | /* Affinity changed (again). */ |
96f874e2 | 5721 | if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) |
b1e38734 | 5722 | goto fail; |
1da177e4 | 5723 | |
e2912009 PZ |
5724 | /* |
5725 | * If we're not on a rq, the next wake-up will ensure we're | |
5726 | * placed properly. | |
5727 | */ | |
5728 | if (p->se.on_rq) { | |
2e1cb74a | 5729 | deactivate_task(rq_src, p, 0); |
e2912009 | 5730 | set_task_cpu(p, dest_cpu); |
dd41f596 | 5731 | activate_task(rq_dest, p, 0); |
15afe09b | 5732 | check_preempt_curr(rq_dest, p, 0); |
1da177e4 | 5733 | } |
b1e38734 | 5734 | done: |
efc30814 | 5735 | ret = 1; |
b1e38734 | 5736 | fail: |
1da177e4 | 5737 | double_rq_unlock(rq_src, rq_dest); |
efc30814 | 5738 | return ret; |
1da177e4 LT |
5739 | } |
5740 | ||
5741 | /* | |
969c7921 TH |
5742 | * migration_cpu_stop - this will be executed by a highprio stopper thread |
5743 | * and performs thread migration by bumping thread off CPU then | |
5744 | * 'pushing' onto another runqueue. | |
1da177e4 | 5745 | */ |
969c7921 | 5746 | static int migration_cpu_stop(void *data) |
1da177e4 | 5747 | { |
969c7921 | 5748 | struct migration_arg *arg = data; |
f7b4cddc | 5749 | |
969c7921 TH |
5750 | /* |
5751 | * The original target cpu might have gone down and we might | |
5752 | * be on another cpu but it doesn't matter. | |
5753 | */ | |
f7b4cddc | 5754 | local_irq_disable(); |
969c7921 | 5755 | __migrate_task(arg->task, raw_smp_processor_id(), arg->dest_cpu); |
f7b4cddc | 5756 | local_irq_enable(); |
1da177e4 | 5757 | return 0; |
f7b4cddc ON |
5758 | } |
5759 | ||
1da177e4 | 5760 | #ifdef CONFIG_HOTPLUG_CPU |
48c5ccae | 5761 | |
054b9108 | 5762 | /* |
48c5ccae PZ |
5763 | * Ensures that the idle task is using init_mm right before its cpu goes |
5764 | * offline. | |
054b9108 | 5765 | */ |
48c5ccae | 5766 | void idle_task_exit(void) |
1da177e4 | 5767 | { |
48c5ccae | 5768 | struct mm_struct *mm = current->active_mm; |
e76bd8d9 | 5769 | |
48c5ccae | 5770 | BUG_ON(cpu_online(smp_processor_id())); |
e76bd8d9 | 5771 | |
48c5ccae PZ |
5772 | if (mm != &init_mm) |
5773 | switch_mm(mm, &init_mm, current); | |
5774 | mmdrop(mm); | |
1da177e4 LT |
5775 | } |
5776 | ||
5777 | /* | |
5778 | * While a dead CPU has no uninterruptible tasks queued at this point, | |
5779 | * it might still have a nonzero ->nr_uninterruptible counter, because | |
5780 | * for performance reasons the counter is not stricly tracking tasks to | |
5781 | * their home CPUs. So we just add the counter to another CPU's counter, | |
5782 | * to keep the global sum constant after CPU-down: | |
5783 | */ | |
70b97a7f | 5784 | static void migrate_nr_uninterruptible(struct rq *rq_src) |
1da177e4 | 5785 | { |
6ad4c188 | 5786 | struct rq *rq_dest = cpu_rq(cpumask_any(cpu_active_mask)); |
1da177e4 | 5787 | |
1da177e4 LT |
5788 | rq_dest->nr_uninterruptible += rq_src->nr_uninterruptible; |
5789 | rq_src->nr_uninterruptible = 0; | |
1da177e4 LT |
5790 | } |
5791 | ||
dd41f596 | 5792 | /* |
48c5ccae | 5793 | * remove the tasks which were accounted by rq from calc_load_tasks. |
1da177e4 | 5794 | */ |
48c5ccae | 5795 | static void calc_global_load_remove(struct rq *rq) |
1da177e4 | 5796 | { |
48c5ccae PZ |
5797 | atomic_long_sub(rq->calc_load_active, &calc_load_tasks); |
5798 | rq->calc_load_active = 0; | |
1da177e4 LT |
5799 | } |
5800 | ||
48f24c4d | 5801 | /* |
48c5ccae PZ |
5802 | * Migrate all tasks from the rq, sleeping tasks will be migrated by |
5803 | * try_to_wake_up()->select_task_rq(). | |
5804 | * | |
5805 | * Called with rq->lock held even though we'er in stop_machine() and | |
5806 | * there's no concurrency possible, we hold the required locks anyway | |
5807 | * because of lock validation efforts. | |
1da177e4 | 5808 | */ |
48c5ccae | 5809 | static void migrate_tasks(unsigned int dead_cpu) |
1da177e4 | 5810 | { |
70b97a7f | 5811 | struct rq *rq = cpu_rq(dead_cpu); |
48c5ccae PZ |
5812 | struct task_struct *next, *stop = rq->stop; |
5813 | int dest_cpu; | |
1da177e4 LT |
5814 | |
5815 | /* | |
48c5ccae PZ |
5816 | * Fudge the rq selection such that the below task selection loop |
5817 | * doesn't get stuck on the currently eligible stop task. | |
5818 | * | |
5819 | * We're currently inside stop_machine() and the rq is either stuck | |
5820 | * in the stop_machine_cpu_stop() loop, or we're executing this code, | |
5821 | * either way we should never end up calling schedule() until we're | |
5822 | * done here. | |
1da177e4 | 5823 | */ |
48c5ccae | 5824 | rq->stop = NULL; |
48f24c4d | 5825 | |
dd41f596 | 5826 | for ( ; ; ) { |
48c5ccae PZ |
5827 | /* |
5828 | * There's this thread running, bail when that's the only | |
5829 | * remaining thread. | |
5830 | */ | |
5831 | if (rq->nr_running == 1) | |
dd41f596 | 5832 | break; |
48c5ccae | 5833 | |
b67802ea | 5834 | next = pick_next_task(rq); |
48c5ccae | 5835 | BUG_ON(!next); |
79c53799 | 5836 | next->sched_class->put_prev_task(rq, next); |
e692ab53 | 5837 | |
48c5ccae PZ |
5838 | /* Find suitable destination for @next, with force if needed. */ |
5839 | dest_cpu = select_fallback_rq(dead_cpu, next); | |
5840 | raw_spin_unlock(&rq->lock); | |
5841 | ||
5842 | __migrate_task(next, dead_cpu, dest_cpu); | |
5843 | ||
5844 | raw_spin_lock(&rq->lock); | |
1da177e4 | 5845 | } |
dce48a84 | 5846 | |
48c5ccae | 5847 | rq->stop = stop; |
dce48a84 | 5848 | } |
48c5ccae | 5849 | |
1da177e4 LT |
5850 | #endif /* CONFIG_HOTPLUG_CPU */ |
5851 | ||
e692ab53 NP |
5852 | #if defined(CONFIG_SCHED_DEBUG) && defined(CONFIG_SYSCTL) |
5853 | ||
5854 | static struct ctl_table sd_ctl_dir[] = { | |
e0361851 AD |
5855 | { |
5856 | .procname = "sched_domain", | |
c57baf1e | 5857 | .mode = 0555, |
e0361851 | 5858 | }, |
56992309 | 5859 | {} |
e692ab53 NP |
5860 | }; |
5861 | ||
5862 | static struct ctl_table sd_ctl_root[] = { | |
e0361851 AD |
5863 | { |
5864 | .procname = "kernel", | |
c57baf1e | 5865 | .mode = 0555, |
e0361851 AD |
5866 | .child = sd_ctl_dir, |
5867 | }, | |
56992309 | 5868 | {} |
e692ab53 NP |
5869 | }; |
5870 | ||
5871 | static struct ctl_table *sd_alloc_ctl_entry(int n) | |
5872 | { | |
5873 | struct ctl_table *entry = | |
5cf9f062 | 5874 | kcalloc(n, sizeof(struct ctl_table), GFP_KERNEL); |
e692ab53 | 5875 | |
e692ab53 NP |
5876 | return entry; |
5877 | } | |
5878 | ||
6382bc90 MM |
5879 | static void sd_free_ctl_entry(struct ctl_table **tablep) |
5880 | { | |
cd790076 | 5881 | struct ctl_table *entry; |
6382bc90 | 5882 | |
cd790076 MM |
5883 | /* |
5884 | * In the intermediate directories, both the child directory and | |
5885 | * procname are dynamically allocated and could fail but the mode | |
41a2d6cf | 5886 | * will always be set. In the lowest directory the names are |
cd790076 MM |
5887 | * static strings and all have proc handlers. |
5888 | */ | |
5889 | for (entry = *tablep; entry->mode; entry++) { | |
6382bc90 MM |
5890 | if (entry->child) |
5891 | sd_free_ctl_entry(&entry->child); | |
cd790076 MM |
5892 | if (entry->proc_handler == NULL) |
5893 | kfree(entry->procname); | |
5894 | } | |
6382bc90 MM |
5895 | |
5896 | kfree(*tablep); | |
5897 | *tablep = NULL; | |
5898 | } | |
5899 | ||
e692ab53 | 5900 | static void |
e0361851 | 5901 | set_table_entry(struct ctl_table *entry, |
e692ab53 NP |
5902 | const char *procname, void *data, int maxlen, |
5903 | mode_t mode, proc_handler *proc_handler) | |
5904 | { | |
e692ab53 NP |
5905 | entry->procname = procname; |
5906 | entry->data = data; | |
5907 | entry->maxlen = maxlen; | |
5908 | entry->mode = mode; | |
5909 | entry->proc_handler = proc_handler; | |
5910 | } | |
5911 | ||
5912 | static struct ctl_table * | |
5913 | sd_alloc_ctl_domain_table(struct sched_domain *sd) | |
5914 | { | |
a5d8c348 | 5915 | struct ctl_table *table = sd_alloc_ctl_entry(13); |
e692ab53 | 5916 | |
ad1cdc1d MM |
5917 | if (table == NULL) |
5918 | return NULL; | |
5919 | ||
e0361851 | 5920 | set_table_entry(&table[0], "min_interval", &sd->min_interval, |
e692ab53 | 5921 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5922 | set_table_entry(&table[1], "max_interval", &sd->max_interval, |
e692ab53 | 5923 | sizeof(long), 0644, proc_doulongvec_minmax); |
e0361851 | 5924 | set_table_entry(&table[2], "busy_idx", &sd->busy_idx, |
e692ab53 | 5925 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5926 | set_table_entry(&table[3], "idle_idx", &sd->idle_idx, |
e692ab53 | 5927 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5928 | set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, |
e692ab53 | 5929 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5930 | set_table_entry(&table[5], "wake_idx", &sd->wake_idx, |
e692ab53 | 5931 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5932 | set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, |
e692ab53 | 5933 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5934 | set_table_entry(&table[7], "busy_factor", &sd->busy_factor, |
e692ab53 | 5935 | sizeof(int), 0644, proc_dointvec_minmax); |
e0361851 | 5936 | set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, |
e692ab53 | 5937 | sizeof(int), 0644, proc_dointvec_minmax); |
ace8b3d6 | 5938 | set_table_entry(&table[9], "cache_nice_tries", |
e692ab53 NP |
5939 | &sd->cache_nice_tries, |
5940 | sizeof(int), 0644, proc_dointvec_minmax); | |
ace8b3d6 | 5941 | set_table_entry(&table[10], "flags", &sd->flags, |
e692ab53 | 5942 | sizeof(int), 0644, proc_dointvec_minmax); |
a5d8c348 IM |
5943 | set_table_entry(&table[11], "name", sd->name, |
5944 | CORENAME_MAX_SIZE, 0444, proc_dostring); | |
5945 | /* &table[12] is terminator */ | |
e692ab53 NP |
5946 | |
5947 | return table; | |
5948 | } | |
5949 | ||
9a4e7159 | 5950 | static ctl_table *sd_alloc_ctl_cpu_table(int cpu) |
e692ab53 NP |
5951 | { |
5952 | struct ctl_table *entry, *table; | |
5953 | struct sched_domain *sd; | |
5954 | int domain_num = 0, i; | |
5955 | char buf[32]; | |
5956 | ||
5957 | for_each_domain(cpu, sd) | |
5958 | domain_num++; | |
5959 | entry = table = sd_alloc_ctl_entry(domain_num + 1); | |
ad1cdc1d MM |
5960 | if (table == NULL) |
5961 | return NULL; | |
e692ab53 NP |
5962 | |
5963 | i = 0; | |
5964 | for_each_domain(cpu, sd) { | |
5965 | snprintf(buf, 32, "domain%d", i); | |
e692ab53 | 5966 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5967 | entry->mode = 0555; |
e692ab53 NP |
5968 | entry->child = sd_alloc_ctl_domain_table(sd); |
5969 | entry++; | |
5970 | i++; | |
5971 | } | |
5972 | return table; | |
5973 | } | |
5974 | ||
5975 | static struct ctl_table_header *sd_sysctl_header; | |
6382bc90 | 5976 | static void register_sched_domain_sysctl(void) |
e692ab53 | 5977 | { |
6ad4c188 | 5978 | int i, cpu_num = num_possible_cpus(); |
e692ab53 NP |
5979 | struct ctl_table *entry = sd_alloc_ctl_entry(cpu_num + 1); |
5980 | char buf[32]; | |
5981 | ||
7378547f MM |
5982 | WARN_ON(sd_ctl_dir[0].child); |
5983 | sd_ctl_dir[0].child = entry; | |
5984 | ||
ad1cdc1d MM |
5985 | if (entry == NULL) |
5986 | return; | |
5987 | ||
6ad4c188 | 5988 | for_each_possible_cpu(i) { |
e692ab53 | 5989 | snprintf(buf, 32, "cpu%d", i); |
e692ab53 | 5990 | entry->procname = kstrdup(buf, GFP_KERNEL); |
c57baf1e | 5991 | entry->mode = 0555; |
e692ab53 | 5992 | entry->child = sd_alloc_ctl_cpu_table(i); |
97b6ea7b | 5993 | entry++; |
e692ab53 | 5994 | } |
7378547f MM |
5995 | |
5996 | WARN_ON(sd_sysctl_header); | |
e692ab53 NP |
5997 | sd_sysctl_header = register_sysctl_table(sd_ctl_root); |
5998 | } | |
6382bc90 | 5999 | |
7378547f | 6000 | /* may be called multiple times per register */ |
6382bc90 MM |
6001 | static void unregister_sched_domain_sysctl(void) |
6002 | { | |
7378547f MM |
6003 | if (sd_sysctl_header) |
6004 | unregister_sysctl_table(sd_sysctl_header); | |
6382bc90 | 6005 | sd_sysctl_header = NULL; |
7378547f MM |
6006 | if (sd_ctl_dir[0].child) |
6007 | sd_free_ctl_entry(&sd_ctl_dir[0].child); | |
6382bc90 | 6008 | } |
e692ab53 | 6009 | #else |
6382bc90 MM |
6010 | static void register_sched_domain_sysctl(void) |
6011 | { | |
6012 | } | |
6013 | static void unregister_sched_domain_sysctl(void) | |
e692ab53 NP |
6014 | { |
6015 | } | |
6016 | #endif | |
6017 | ||
1f11eb6a GH |
6018 | static void set_rq_online(struct rq *rq) |
6019 | { | |
6020 | if (!rq->online) { | |
6021 | const struct sched_class *class; | |
6022 | ||
c6c4927b | 6023 | cpumask_set_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6024 | rq->online = 1; |
6025 | ||
6026 | for_each_class(class) { | |
6027 | if (class->rq_online) | |
6028 | class->rq_online(rq); | |
6029 | } | |
6030 | } | |
6031 | } | |
6032 | ||
6033 | static void set_rq_offline(struct rq *rq) | |
6034 | { | |
6035 | if (rq->online) { | |
6036 | const struct sched_class *class; | |
6037 | ||
6038 | for_each_class(class) { | |
6039 | if (class->rq_offline) | |
6040 | class->rq_offline(rq); | |
6041 | } | |
6042 | ||
c6c4927b | 6043 | cpumask_clear_cpu(rq->cpu, rq->rd->online); |
1f11eb6a GH |
6044 | rq->online = 0; |
6045 | } | |
6046 | } | |
6047 | ||
1da177e4 LT |
6048 | /* |
6049 | * migration_call - callback that gets triggered when a CPU is added. | |
6050 | * Here we can start up the necessary migration thread for the new CPU. | |
6051 | */ | |
48f24c4d IM |
6052 | static int __cpuinit |
6053 | migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu) | |
1da177e4 | 6054 | { |
48f24c4d | 6055 | int cpu = (long)hcpu; |
1da177e4 | 6056 | unsigned long flags; |
969c7921 | 6057 | struct rq *rq = cpu_rq(cpu); |
1da177e4 | 6058 | |
48c5ccae | 6059 | switch (action & ~CPU_TASKS_FROZEN) { |
5be9361c | 6060 | |
1da177e4 | 6061 | case CPU_UP_PREPARE: |
a468d389 | 6062 | rq->calc_load_update = calc_load_update; |
1da177e4 | 6063 | break; |
48f24c4d | 6064 | |
1da177e4 | 6065 | case CPU_ONLINE: |
1f94ef59 | 6066 | /* Update our root-domain */ |
05fa785c | 6067 | raw_spin_lock_irqsave(&rq->lock, flags); |
1f94ef59 | 6068 | if (rq->rd) { |
c6c4927b | 6069 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a GH |
6070 | |
6071 | set_rq_online(rq); | |
1f94ef59 | 6072 | } |
05fa785c | 6073 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
1da177e4 | 6074 | break; |
48f24c4d | 6075 | |
1da177e4 | 6076 | #ifdef CONFIG_HOTPLUG_CPU |
08f503b0 | 6077 | case CPU_DYING: |
57d885fe | 6078 | /* Update our root-domain */ |
05fa785c | 6079 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe | 6080 | if (rq->rd) { |
c6c4927b | 6081 | BUG_ON(!cpumask_test_cpu(cpu, rq->rd->span)); |
1f11eb6a | 6082 | set_rq_offline(rq); |
57d885fe | 6083 | } |
48c5ccae PZ |
6084 | migrate_tasks(cpu); |
6085 | BUG_ON(rq->nr_running != 1); /* the migration thread */ | |
05fa785c | 6086 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
48c5ccae PZ |
6087 | |
6088 | migrate_nr_uninterruptible(rq); | |
6089 | calc_global_load_remove(rq); | |
57d885fe | 6090 | break; |
1da177e4 LT |
6091 | #endif |
6092 | } | |
6093 | return NOTIFY_OK; | |
6094 | } | |
6095 | ||
f38b0820 PM |
6096 | /* |
6097 | * Register at high priority so that task migration (migrate_all_tasks) | |
6098 | * happens before everything else. This has to be lower priority than | |
cdd6c482 | 6099 | * the notifier in the perf_event subsystem, though. |
1da177e4 | 6100 | */ |
26c2143b | 6101 | static struct notifier_block __cpuinitdata migration_notifier = { |
1da177e4 | 6102 | .notifier_call = migration_call, |
50a323b7 | 6103 | .priority = CPU_PRI_MIGRATION, |
1da177e4 LT |
6104 | }; |
6105 | ||
3a101d05 TH |
6106 | static int __cpuinit sched_cpu_active(struct notifier_block *nfb, |
6107 | unsigned long action, void *hcpu) | |
6108 | { | |
6109 | switch (action & ~CPU_TASKS_FROZEN) { | |
6110 | case CPU_ONLINE: | |
6111 | case CPU_DOWN_FAILED: | |
6112 | set_cpu_active((long)hcpu, true); | |
6113 | return NOTIFY_OK; | |
6114 | default: | |
6115 | return NOTIFY_DONE; | |
6116 | } | |
6117 | } | |
6118 | ||
6119 | static int __cpuinit sched_cpu_inactive(struct notifier_block *nfb, | |
6120 | unsigned long action, void *hcpu) | |
6121 | { | |
6122 | switch (action & ~CPU_TASKS_FROZEN) { | |
6123 | case CPU_DOWN_PREPARE: | |
6124 | set_cpu_active((long)hcpu, false); | |
6125 | return NOTIFY_OK; | |
6126 | default: | |
6127 | return NOTIFY_DONE; | |
6128 | } | |
6129 | } | |
6130 | ||
7babe8db | 6131 | static int __init migration_init(void) |
1da177e4 LT |
6132 | { |
6133 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 6134 | int err; |
48f24c4d | 6135 | |
3a101d05 | 6136 | /* Initialize migration for the boot CPU */ |
07dccf33 AM |
6137 | err = migration_call(&migration_notifier, CPU_UP_PREPARE, cpu); |
6138 | BUG_ON(err == NOTIFY_BAD); | |
1da177e4 LT |
6139 | migration_call(&migration_notifier, CPU_ONLINE, cpu); |
6140 | register_cpu_notifier(&migration_notifier); | |
7babe8db | 6141 | |
3a101d05 TH |
6142 | /* Register cpu active notifiers */ |
6143 | cpu_notifier(sched_cpu_active, CPU_PRI_SCHED_ACTIVE); | |
6144 | cpu_notifier(sched_cpu_inactive, CPU_PRI_SCHED_INACTIVE); | |
6145 | ||
a004cd42 | 6146 | return 0; |
1da177e4 | 6147 | } |
7babe8db | 6148 | early_initcall(migration_init); |
1da177e4 LT |
6149 | #endif |
6150 | ||
6151 | #ifdef CONFIG_SMP | |
476f3534 | 6152 | |
3e9830dc | 6153 | #ifdef CONFIG_SCHED_DEBUG |
4dcf6aff | 6154 | |
f6630114 MT |
6155 | static __read_mostly int sched_domain_debug_enabled; |
6156 | ||
6157 | static int __init sched_domain_debug_setup(char *str) | |
6158 | { | |
6159 | sched_domain_debug_enabled = 1; | |
6160 | ||
6161 | return 0; | |
6162 | } | |
6163 | early_param("sched_debug", sched_domain_debug_setup); | |
6164 | ||
7c16ec58 | 6165 | static int sched_domain_debug_one(struct sched_domain *sd, int cpu, int level, |
96f874e2 | 6166 | struct cpumask *groupmask) |
1da177e4 | 6167 | { |
4dcf6aff | 6168 | struct sched_group *group = sd->groups; |
434d53b0 | 6169 | char str[256]; |
1da177e4 | 6170 | |
968ea6d8 | 6171 | cpulist_scnprintf(str, sizeof(str), sched_domain_span(sd)); |
96f874e2 | 6172 | cpumask_clear(groupmask); |
4dcf6aff IM |
6173 | |
6174 | printk(KERN_DEBUG "%*s domain %d: ", level, "", level); | |
6175 | ||
6176 | if (!(sd->flags & SD_LOAD_BALANCE)) { | |
3df0fc5b | 6177 | printk("does not load-balance\n"); |
4dcf6aff | 6178 | if (sd->parent) |
3df0fc5b PZ |
6179 | printk(KERN_ERR "ERROR: !SD_LOAD_BALANCE domain" |
6180 | " has parent"); | |
4dcf6aff | 6181 | return -1; |
41c7ce9a NP |
6182 | } |
6183 | ||
3df0fc5b | 6184 | printk(KERN_CONT "span %s level %s\n", str, sd->name); |
4dcf6aff | 6185 | |
758b2cdc | 6186 | if (!cpumask_test_cpu(cpu, sched_domain_span(sd))) { |
3df0fc5b PZ |
6187 | printk(KERN_ERR "ERROR: domain->span does not contain " |
6188 | "CPU%d\n", cpu); | |
4dcf6aff | 6189 | } |
758b2cdc | 6190 | if (!cpumask_test_cpu(cpu, sched_group_cpus(group))) { |
3df0fc5b PZ |
6191 | printk(KERN_ERR "ERROR: domain->groups does not contain" |
6192 | " CPU%d\n", cpu); | |
4dcf6aff | 6193 | } |
1da177e4 | 6194 | |
4dcf6aff | 6195 | printk(KERN_DEBUG "%*s groups:", level + 1, ""); |
1da177e4 | 6196 | do { |
4dcf6aff | 6197 | if (!group) { |
3df0fc5b PZ |
6198 | printk("\n"); |
6199 | printk(KERN_ERR "ERROR: group is NULL\n"); | |
1da177e4 LT |
6200 | break; |
6201 | } | |
6202 | ||
18a3885f | 6203 | if (!group->cpu_power) { |
3df0fc5b PZ |
6204 | printk(KERN_CONT "\n"); |
6205 | printk(KERN_ERR "ERROR: domain->cpu_power not " | |
6206 | "set\n"); | |
4dcf6aff IM |
6207 | break; |
6208 | } | |
1da177e4 | 6209 | |
758b2cdc | 6210 | if (!cpumask_weight(sched_group_cpus(group))) { |
3df0fc5b PZ |
6211 | printk(KERN_CONT "\n"); |
6212 | printk(KERN_ERR "ERROR: empty group\n"); | |
4dcf6aff IM |
6213 | break; |
6214 | } | |
1da177e4 | 6215 | |
758b2cdc | 6216 | if (cpumask_intersects(groupmask, sched_group_cpus(group))) { |
3df0fc5b PZ |
6217 | printk(KERN_CONT "\n"); |
6218 | printk(KERN_ERR "ERROR: repeated CPUs\n"); | |
4dcf6aff IM |
6219 | break; |
6220 | } | |
1da177e4 | 6221 | |
758b2cdc | 6222 | cpumask_or(groupmask, groupmask, sched_group_cpus(group)); |
1da177e4 | 6223 | |
968ea6d8 | 6224 | cpulist_scnprintf(str, sizeof(str), sched_group_cpus(group)); |
381512cf | 6225 | |
3df0fc5b | 6226 | printk(KERN_CONT " %s", str); |
18a3885f | 6227 | if (group->cpu_power != SCHED_LOAD_SCALE) { |
3df0fc5b PZ |
6228 | printk(KERN_CONT " (cpu_power = %d)", |
6229 | group->cpu_power); | |
381512cf | 6230 | } |
1da177e4 | 6231 | |
4dcf6aff IM |
6232 | group = group->next; |
6233 | } while (group != sd->groups); | |
3df0fc5b | 6234 | printk(KERN_CONT "\n"); |
1da177e4 | 6235 | |
758b2cdc | 6236 | if (!cpumask_equal(sched_domain_span(sd), groupmask)) |
3df0fc5b | 6237 | printk(KERN_ERR "ERROR: groups don't span domain->span\n"); |
1da177e4 | 6238 | |
758b2cdc RR |
6239 | if (sd->parent && |
6240 | !cpumask_subset(groupmask, sched_domain_span(sd->parent))) | |
3df0fc5b PZ |
6241 | printk(KERN_ERR "ERROR: parent span is not a superset " |
6242 | "of domain->span\n"); | |
4dcf6aff IM |
6243 | return 0; |
6244 | } | |
1da177e4 | 6245 | |
4dcf6aff IM |
6246 | static void sched_domain_debug(struct sched_domain *sd, int cpu) |
6247 | { | |
d5dd3db1 | 6248 | cpumask_var_t groupmask; |
4dcf6aff | 6249 | int level = 0; |
1da177e4 | 6250 | |
f6630114 MT |
6251 | if (!sched_domain_debug_enabled) |
6252 | return; | |
6253 | ||
4dcf6aff IM |
6254 | if (!sd) { |
6255 | printk(KERN_DEBUG "CPU%d attaching NULL sched-domain.\n", cpu); | |
6256 | return; | |
6257 | } | |
1da177e4 | 6258 | |
4dcf6aff IM |
6259 | printk(KERN_DEBUG "CPU%d attaching sched-domain:\n", cpu); |
6260 | ||
d5dd3db1 | 6261 | if (!alloc_cpumask_var(&groupmask, GFP_KERNEL)) { |
7c16ec58 MT |
6262 | printk(KERN_DEBUG "Cannot load-balance (out of memory)\n"); |
6263 | return; | |
6264 | } | |
6265 | ||
4dcf6aff | 6266 | for (;;) { |
7c16ec58 | 6267 | if (sched_domain_debug_one(sd, cpu, level, groupmask)) |
4dcf6aff | 6268 | break; |
1da177e4 LT |
6269 | level++; |
6270 | sd = sd->parent; | |
33859f7f | 6271 | if (!sd) |
4dcf6aff IM |
6272 | break; |
6273 | } | |
d5dd3db1 | 6274 | free_cpumask_var(groupmask); |
1da177e4 | 6275 | } |
6d6bc0ad | 6276 | #else /* !CONFIG_SCHED_DEBUG */ |
48f24c4d | 6277 | # define sched_domain_debug(sd, cpu) do { } while (0) |
6d6bc0ad | 6278 | #endif /* CONFIG_SCHED_DEBUG */ |
1da177e4 | 6279 | |
1a20ff27 | 6280 | static int sd_degenerate(struct sched_domain *sd) |
245af2c7 | 6281 | { |
758b2cdc | 6282 | if (cpumask_weight(sched_domain_span(sd)) == 1) |
245af2c7 SS |
6283 | return 1; |
6284 | ||
6285 | /* Following flags need at least 2 groups */ | |
6286 | if (sd->flags & (SD_LOAD_BALANCE | | |
6287 | SD_BALANCE_NEWIDLE | | |
6288 | SD_BALANCE_FORK | | |
89c4710e SS |
6289 | SD_BALANCE_EXEC | |
6290 | SD_SHARE_CPUPOWER | | |
6291 | SD_SHARE_PKG_RESOURCES)) { | |
245af2c7 SS |
6292 | if (sd->groups != sd->groups->next) |
6293 | return 0; | |
6294 | } | |
6295 | ||
6296 | /* Following flags don't use groups */ | |
c88d5910 | 6297 | if (sd->flags & (SD_WAKE_AFFINE)) |
245af2c7 SS |
6298 | return 0; |
6299 | ||
6300 | return 1; | |
6301 | } | |
6302 | ||
48f24c4d IM |
6303 | static int |
6304 | sd_parent_degenerate(struct sched_domain *sd, struct sched_domain *parent) | |
245af2c7 SS |
6305 | { |
6306 | unsigned long cflags = sd->flags, pflags = parent->flags; | |
6307 | ||
6308 | if (sd_degenerate(parent)) | |
6309 | return 1; | |
6310 | ||
758b2cdc | 6311 | if (!cpumask_equal(sched_domain_span(sd), sched_domain_span(parent))) |
245af2c7 SS |
6312 | return 0; |
6313 | ||
245af2c7 SS |
6314 | /* Flags needing groups don't count if only 1 group in parent */ |
6315 | if (parent->groups == parent->groups->next) { | |
6316 | pflags &= ~(SD_LOAD_BALANCE | | |
6317 | SD_BALANCE_NEWIDLE | | |
6318 | SD_BALANCE_FORK | | |
89c4710e SS |
6319 | SD_BALANCE_EXEC | |
6320 | SD_SHARE_CPUPOWER | | |
6321 | SD_SHARE_PKG_RESOURCES); | |
5436499e KC |
6322 | if (nr_node_ids == 1) |
6323 | pflags &= ~SD_SERIALIZE; | |
245af2c7 SS |
6324 | } |
6325 | if (~cflags & pflags) | |
6326 | return 0; | |
6327 | ||
6328 | return 1; | |
6329 | } | |
6330 | ||
c6c4927b RR |
6331 | static void free_rootdomain(struct root_domain *rd) |
6332 | { | |
047106ad PZ |
6333 | synchronize_sched(); |
6334 | ||
68e74568 RR |
6335 | cpupri_cleanup(&rd->cpupri); |
6336 | ||
c6c4927b RR |
6337 | free_cpumask_var(rd->rto_mask); |
6338 | free_cpumask_var(rd->online); | |
6339 | free_cpumask_var(rd->span); | |
6340 | kfree(rd); | |
6341 | } | |
6342 | ||
57d885fe GH |
6343 | static void rq_attach_root(struct rq *rq, struct root_domain *rd) |
6344 | { | |
a0490fa3 | 6345 | struct root_domain *old_rd = NULL; |
57d885fe | 6346 | unsigned long flags; |
57d885fe | 6347 | |
05fa785c | 6348 | raw_spin_lock_irqsave(&rq->lock, flags); |
57d885fe GH |
6349 | |
6350 | if (rq->rd) { | |
a0490fa3 | 6351 | old_rd = rq->rd; |
57d885fe | 6352 | |
c6c4927b | 6353 | if (cpumask_test_cpu(rq->cpu, old_rd->online)) |
1f11eb6a | 6354 | set_rq_offline(rq); |
57d885fe | 6355 | |
c6c4927b | 6356 | cpumask_clear_cpu(rq->cpu, old_rd->span); |
dc938520 | 6357 | |
a0490fa3 IM |
6358 | /* |
6359 | * If we dont want to free the old_rt yet then | |
6360 | * set old_rd to NULL to skip the freeing later | |
6361 | * in this function: | |
6362 | */ | |
6363 | if (!atomic_dec_and_test(&old_rd->refcount)) | |
6364 | old_rd = NULL; | |
57d885fe GH |
6365 | } |
6366 | ||
6367 | atomic_inc(&rd->refcount); | |
6368 | rq->rd = rd; | |
6369 | ||
c6c4927b | 6370 | cpumask_set_cpu(rq->cpu, rd->span); |
00aec93d | 6371 | if (cpumask_test_cpu(rq->cpu, cpu_active_mask)) |
1f11eb6a | 6372 | set_rq_online(rq); |
57d885fe | 6373 | |
05fa785c | 6374 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
a0490fa3 IM |
6375 | |
6376 | if (old_rd) | |
6377 | free_rootdomain(old_rd); | |
57d885fe GH |
6378 | } |
6379 | ||
68c38fc3 | 6380 | static int init_rootdomain(struct root_domain *rd) |
57d885fe GH |
6381 | { |
6382 | memset(rd, 0, sizeof(*rd)); | |
6383 | ||
68c38fc3 | 6384 | if (!alloc_cpumask_var(&rd->span, GFP_KERNEL)) |
0c910d28 | 6385 | goto out; |
68c38fc3 | 6386 | if (!alloc_cpumask_var(&rd->online, GFP_KERNEL)) |
c6c4927b | 6387 | goto free_span; |
68c38fc3 | 6388 | if (!alloc_cpumask_var(&rd->rto_mask, GFP_KERNEL)) |
c6c4927b | 6389 | goto free_online; |
6e0534f2 | 6390 | |
68c38fc3 | 6391 | if (cpupri_init(&rd->cpupri) != 0) |
68e74568 | 6392 | goto free_rto_mask; |
c6c4927b | 6393 | return 0; |
6e0534f2 | 6394 | |
68e74568 RR |
6395 | free_rto_mask: |
6396 | free_cpumask_var(rd->rto_mask); | |
c6c4927b RR |
6397 | free_online: |
6398 | free_cpumask_var(rd->online); | |
6399 | free_span: | |
6400 | free_cpumask_var(rd->span); | |
0c910d28 | 6401 | out: |
c6c4927b | 6402 | return -ENOMEM; |
57d885fe GH |
6403 | } |
6404 | ||
6405 | static void init_defrootdomain(void) | |
6406 | { | |
68c38fc3 | 6407 | init_rootdomain(&def_root_domain); |
c6c4927b | 6408 | |
57d885fe GH |
6409 | atomic_set(&def_root_domain.refcount, 1); |
6410 | } | |
6411 | ||
dc938520 | 6412 | static struct root_domain *alloc_rootdomain(void) |
57d885fe GH |
6413 | { |
6414 | struct root_domain *rd; | |
6415 | ||
6416 | rd = kmalloc(sizeof(*rd), GFP_KERNEL); | |
6417 | if (!rd) | |
6418 | return NULL; | |
6419 | ||
68c38fc3 | 6420 | if (init_rootdomain(rd) != 0) { |
c6c4927b RR |
6421 | kfree(rd); |
6422 | return NULL; | |
6423 | } | |
57d885fe GH |
6424 | |
6425 | return rd; | |
6426 | } | |
6427 | ||
1da177e4 | 6428 | /* |
0eab9146 | 6429 | * Attach the domain 'sd' to 'cpu' as its base domain. Callers must |
1da177e4 LT |
6430 | * hold the hotplug lock. |
6431 | */ | |
0eab9146 IM |
6432 | static void |
6433 | cpu_attach_domain(struct sched_domain *sd, struct root_domain *rd, int cpu) | |
1da177e4 | 6434 | { |
70b97a7f | 6435 | struct rq *rq = cpu_rq(cpu); |
245af2c7 SS |
6436 | struct sched_domain *tmp; |
6437 | ||
669c55e9 PZ |
6438 | for (tmp = sd; tmp; tmp = tmp->parent) |
6439 | tmp->span_weight = cpumask_weight(sched_domain_span(tmp)); | |
6440 | ||
245af2c7 | 6441 | /* Remove the sched domains which do not contribute to scheduling. */ |
f29c9b1c | 6442 | for (tmp = sd; tmp; ) { |
245af2c7 SS |
6443 | struct sched_domain *parent = tmp->parent; |
6444 | if (!parent) | |
6445 | break; | |
f29c9b1c | 6446 | |
1a848870 | 6447 | if (sd_parent_degenerate(tmp, parent)) { |
245af2c7 | 6448 | tmp->parent = parent->parent; |
1a848870 SS |
6449 | if (parent->parent) |
6450 | parent->parent->child = tmp; | |
f29c9b1c LZ |
6451 | } else |
6452 | tmp = tmp->parent; | |
245af2c7 SS |
6453 | } |
6454 | ||
1a848870 | 6455 | if (sd && sd_degenerate(sd)) { |
245af2c7 | 6456 | sd = sd->parent; |
1a848870 SS |
6457 | if (sd) |
6458 | sd->child = NULL; | |
6459 | } | |
1da177e4 LT |
6460 | |
6461 | sched_domain_debug(sd, cpu); | |
6462 | ||
57d885fe | 6463 | rq_attach_root(rq, rd); |
674311d5 | 6464 | rcu_assign_pointer(rq->sd, sd); |
1da177e4 LT |
6465 | } |
6466 | ||
6467 | /* cpus with isolated domains */ | |
dcc30a35 | 6468 | static cpumask_var_t cpu_isolated_map; |
1da177e4 LT |
6469 | |
6470 | /* Setup the mask of cpus configured for isolated domains */ | |
6471 | static int __init isolated_cpu_setup(char *str) | |
6472 | { | |
bdddd296 | 6473 | alloc_bootmem_cpumask_var(&cpu_isolated_map); |
968ea6d8 | 6474 | cpulist_parse(str, cpu_isolated_map); |
1da177e4 LT |
6475 | return 1; |
6476 | } | |
6477 | ||
8927f494 | 6478 | __setup("isolcpus=", isolated_cpu_setup); |
1da177e4 LT |
6479 | |
6480 | /* | |
6711cab4 SS |
6481 | * init_sched_build_groups takes the cpumask we wish to span, and a pointer |
6482 | * to a function which identifies what group(along with sched group) a CPU | |
96f874e2 RR |
6483 | * belongs to. The return value of group_fn must be a >= 0 and < nr_cpu_ids |
6484 | * (due to the fact that we keep track of groups covered with a struct cpumask). | |
1da177e4 LT |
6485 | * |
6486 | * init_sched_build_groups will build a circular linked list of the groups | |
6487 | * covered by the given span, and will set each group's ->cpumask correctly, | |
6488 | * and ->cpu_power to 0. | |
6489 | */ | |
a616058b | 6490 | static void |
96f874e2 RR |
6491 | init_sched_build_groups(const struct cpumask *span, |
6492 | const struct cpumask *cpu_map, | |
6493 | int (*group_fn)(int cpu, const struct cpumask *cpu_map, | |
7c16ec58 | 6494 | struct sched_group **sg, |
96f874e2 RR |
6495 | struct cpumask *tmpmask), |
6496 | struct cpumask *covered, struct cpumask *tmpmask) | |
1da177e4 LT |
6497 | { |
6498 | struct sched_group *first = NULL, *last = NULL; | |
1da177e4 LT |
6499 | int i; |
6500 | ||
96f874e2 | 6501 | cpumask_clear(covered); |
7c16ec58 | 6502 | |
abcd083a | 6503 | for_each_cpu(i, span) { |
6711cab4 | 6504 | struct sched_group *sg; |
7c16ec58 | 6505 | int group = group_fn(i, cpu_map, &sg, tmpmask); |
1da177e4 LT |
6506 | int j; |
6507 | ||
758b2cdc | 6508 | if (cpumask_test_cpu(i, covered)) |
1da177e4 LT |
6509 | continue; |
6510 | ||
758b2cdc | 6511 | cpumask_clear(sched_group_cpus(sg)); |
18a3885f | 6512 | sg->cpu_power = 0; |
1da177e4 | 6513 | |
abcd083a | 6514 | for_each_cpu(j, span) { |
7c16ec58 | 6515 | if (group_fn(j, cpu_map, NULL, tmpmask) != group) |
1da177e4 LT |
6516 | continue; |
6517 | ||
96f874e2 | 6518 | cpumask_set_cpu(j, covered); |
758b2cdc | 6519 | cpumask_set_cpu(j, sched_group_cpus(sg)); |
1da177e4 LT |
6520 | } |
6521 | if (!first) | |
6522 | first = sg; | |
6523 | if (last) | |
6524 | last->next = sg; | |
6525 | last = sg; | |
6526 | } | |
6527 | last->next = first; | |
6528 | } | |
6529 | ||
9c1cfda2 | 6530 | #define SD_NODES_PER_DOMAIN 16 |
1da177e4 | 6531 | |
9c1cfda2 | 6532 | #ifdef CONFIG_NUMA |
198e2f18 | 6533 | |
9c1cfda2 JH |
6534 | /** |
6535 | * find_next_best_node - find the next node to include in a sched_domain | |
6536 | * @node: node whose sched_domain we're building | |
6537 | * @used_nodes: nodes already in the sched_domain | |
6538 | * | |
41a2d6cf | 6539 | * Find the next node to include in a given scheduling domain. Simply |
9c1cfda2 JH |
6540 | * finds the closest node not already in the @used_nodes map. |
6541 | * | |
6542 | * Should use nodemask_t. | |
6543 | */ | |
c5f59f08 | 6544 | static int find_next_best_node(int node, nodemask_t *used_nodes) |
9c1cfda2 JH |
6545 | { |
6546 | int i, n, val, min_val, best_node = 0; | |
6547 | ||
6548 | min_val = INT_MAX; | |
6549 | ||
076ac2af | 6550 | for (i = 0; i < nr_node_ids; i++) { |
9c1cfda2 | 6551 | /* Start at @node */ |
076ac2af | 6552 | n = (node + i) % nr_node_ids; |
9c1cfda2 JH |
6553 | |
6554 | if (!nr_cpus_node(n)) | |
6555 | continue; | |
6556 | ||
6557 | /* Skip already used nodes */ | |
c5f59f08 | 6558 | if (node_isset(n, *used_nodes)) |
9c1cfda2 JH |
6559 | continue; |
6560 | ||
6561 | /* Simple min distance search */ | |
6562 | val = node_distance(node, n); | |
6563 | ||
6564 | if (val < min_val) { | |
6565 | min_val = val; | |
6566 | best_node = n; | |
6567 | } | |
6568 | } | |
6569 | ||
c5f59f08 | 6570 | node_set(best_node, *used_nodes); |
9c1cfda2 JH |
6571 | return best_node; |
6572 | } | |
6573 | ||
6574 | /** | |
6575 | * sched_domain_node_span - get a cpumask for a node's sched_domain | |
6576 | * @node: node whose cpumask we're constructing | |
73486722 | 6577 | * @span: resulting cpumask |
9c1cfda2 | 6578 | * |
41a2d6cf | 6579 | * Given a node, construct a good cpumask for its sched_domain to span. It |
9c1cfda2 JH |
6580 | * should be one that prevents unnecessary balancing, but also spreads tasks |
6581 | * out optimally. | |
6582 | */ | |
96f874e2 | 6583 | static void sched_domain_node_span(int node, struct cpumask *span) |
9c1cfda2 | 6584 | { |
c5f59f08 | 6585 | nodemask_t used_nodes; |
48f24c4d | 6586 | int i; |
9c1cfda2 | 6587 | |
6ca09dfc | 6588 | cpumask_clear(span); |
c5f59f08 | 6589 | nodes_clear(used_nodes); |
9c1cfda2 | 6590 | |
6ca09dfc | 6591 | cpumask_or(span, span, cpumask_of_node(node)); |
c5f59f08 | 6592 | node_set(node, used_nodes); |
9c1cfda2 JH |
6593 | |
6594 | for (i = 1; i < SD_NODES_PER_DOMAIN; i++) { | |
c5f59f08 | 6595 | int next_node = find_next_best_node(node, &used_nodes); |
48f24c4d | 6596 | |
6ca09dfc | 6597 | cpumask_or(span, span, cpumask_of_node(next_node)); |
9c1cfda2 | 6598 | } |
9c1cfda2 | 6599 | } |
6d6bc0ad | 6600 | #endif /* CONFIG_NUMA */ |
9c1cfda2 | 6601 | |
5c45bf27 | 6602 | int sched_smt_power_savings = 0, sched_mc_power_savings = 0; |
48f24c4d | 6603 | |
6c99e9ad RR |
6604 | /* |
6605 | * The cpus mask in sched_group and sched_domain hangs off the end. | |
4200efd9 IM |
6606 | * |
6607 | * ( See the the comments in include/linux/sched.h:struct sched_group | |
6608 | * and struct sched_domain. ) | |
6c99e9ad RR |
6609 | */ |
6610 | struct static_sched_group { | |
6611 | struct sched_group sg; | |
6612 | DECLARE_BITMAP(cpus, CONFIG_NR_CPUS); | |
6613 | }; | |
6614 | ||
6615 | struct static_sched_domain { | |
6616 | struct sched_domain sd; | |
6617 | DECLARE_BITMAP(span, CONFIG_NR_CPUS); | |
6618 | }; | |
6619 | ||
49a02c51 AH |
6620 | struct s_data { |
6621 | #ifdef CONFIG_NUMA | |
6622 | int sd_allnodes; | |
6623 | cpumask_var_t domainspan; | |
6624 | cpumask_var_t covered; | |
6625 | cpumask_var_t notcovered; | |
6626 | #endif | |
6627 | cpumask_var_t nodemask; | |
6628 | cpumask_var_t this_sibling_map; | |
6629 | cpumask_var_t this_core_map; | |
01a08546 | 6630 | cpumask_var_t this_book_map; |
49a02c51 AH |
6631 | cpumask_var_t send_covered; |
6632 | cpumask_var_t tmpmask; | |
6633 | struct sched_group **sched_group_nodes; | |
6634 | struct root_domain *rd; | |
6635 | }; | |
6636 | ||
2109b99e AH |
6637 | enum s_alloc { |
6638 | sa_sched_groups = 0, | |
6639 | sa_rootdomain, | |
6640 | sa_tmpmask, | |
6641 | sa_send_covered, | |
01a08546 | 6642 | sa_this_book_map, |
2109b99e AH |
6643 | sa_this_core_map, |
6644 | sa_this_sibling_map, | |
6645 | sa_nodemask, | |
6646 | sa_sched_group_nodes, | |
6647 | #ifdef CONFIG_NUMA | |
6648 | sa_notcovered, | |
6649 | sa_covered, | |
6650 | sa_domainspan, | |
6651 | #endif | |
6652 | sa_none, | |
6653 | }; | |
6654 | ||
9c1cfda2 | 6655 | /* |
48f24c4d | 6656 | * SMT sched-domains: |
9c1cfda2 | 6657 | */ |
1da177e4 | 6658 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 6659 | static DEFINE_PER_CPU(struct static_sched_domain, cpu_domains); |
1871e52c | 6660 | static DEFINE_PER_CPU(struct static_sched_group, sched_groups); |
48f24c4d | 6661 | |
41a2d6cf | 6662 | static int |
96f874e2 RR |
6663 | cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map, |
6664 | struct sched_group **sg, struct cpumask *unused) | |
1da177e4 | 6665 | { |
6711cab4 | 6666 | if (sg) |
1871e52c | 6667 | *sg = &per_cpu(sched_groups, cpu).sg; |
1da177e4 LT |
6668 | return cpu; |
6669 | } | |
6d6bc0ad | 6670 | #endif /* CONFIG_SCHED_SMT */ |
1da177e4 | 6671 | |
48f24c4d IM |
6672 | /* |
6673 | * multi-core sched-domains: | |
6674 | */ | |
1e9f28fa | 6675 | #ifdef CONFIG_SCHED_MC |
6c99e9ad RR |
6676 | static DEFINE_PER_CPU(struct static_sched_domain, core_domains); |
6677 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); | |
1e9f28fa | 6678 | |
41a2d6cf | 6679 | static int |
96f874e2 RR |
6680 | cpu_to_core_group(int cpu, const struct cpumask *cpu_map, |
6681 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6682 | { |
6711cab4 | 6683 | int group; |
f269893c | 6684 | #ifdef CONFIG_SCHED_SMT |
c69fc56d | 6685 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6686 | group = cpumask_first(mask); |
f269893c HC |
6687 | #else |
6688 | group = cpu; | |
6689 | #endif | |
6711cab4 | 6690 | if (sg) |
6c99e9ad | 6691 | *sg = &per_cpu(sched_group_core, group).sg; |
6711cab4 | 6692 | return group; |
1e9f28fa | 6693 | } |
f269893c | 6694 | #endif /* CONFIG_SCHED_MC */ |
1e9f28fa | 6695 | |
01a08546 HC |
6696 | /* |
6697 | * book sched-domains: | |
6698 | */ | |
6699 | #ifdef CONFIG_SCHED_BOOK | |
6700 | static DEFINE_PER_CPU(struct static_sched_domain, book_domains); | |
6701 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_book); | |
6702 | ||
41a2d6cf | 6703 | static int |
01a08546 HC |
6704 | cpu_to_book_group(int cpu, const struct cpumask *cpu_map, |
6705 | struct sched_group **sg, struct cpumask *mask) | |
1e9f28fa | 6706 | { |
01a08546 HC |
6707 | int group = cpu; |
6708 | #ifdef CONFIG_SCHED_MC | |
6709 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); | |
6710 | group = cpumask_first(mask); | |
6711 | #elif defined(CONFIG_SCHED_SMT) | |
6712 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); | |
6713 | group = cpumask_first(mask); | |
6714 | #endif | |
6711cab4 | 6715 | if (sg) |
01a08546 HC |
6716 | *sg = &per_cpu(sched_group_book, group).sg; |
6717 | return group; | |
1e9f28fa | 6718 | } |
01a08546 | 6719 | #endif /* CONFIG_SCHED_BOOK */ |
1e9f28fa | 6720 | |
6c99e9ad RR |
6721 | static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); |
6722 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); | |
48f24c4d | 6723 | |
41a2d6cf | 6724 | static int |
96f874e2 RR |
6725 | cpu_to_phys_group(int cpu, const struct cpumask *cpu_map, |
6726 | struct sched_group **sg, struct cpumask *mask) | |
1da177e4 | 6727 | { |
6711cab4 | 6728 | int group; |
01a08546 HC |
6729 | #ifdef CONFIG_SCHED_BOOK |
6730 | cpumask_and(mask, cpu_book_mask(cpu), cpu_map); | |
6731 | group = cpumask_first(mask); | |
6732 | #elif defined(CONFIG_SCHED_MC) | |
6ca09dfc | 6733 | cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); |
96f874e2 | 6734 | group = cpumask_first(mask); |
1e9f28fa | 6735 | #elif defined(CONFIG_SCHED_SMT) |
c69fc56d | 6736 | cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); |
96f874e2 | 6737 | group = cpumask_first(mask); |
1da177e4 | 6738 | #else |
6711cab4 | 6739 | group = cpu; |
1da177e4 | 6740 | #endif |
6711cab4 | 6741 | if (sg) |
6c99e9ad | 6742 | *sg = &per_cpu(sched_group_phys, group).sg; |
6711cab4 | 6743 | return group; |
1da177e4 LT |
6744 | } |
6745 | ||
6746 | #ifdef CONFIG_NUMA | |
1da177e4 | 6747 | /* |
9c1cfda2 JH |
6748 | * The init_sched_build_groups can't handle what we want to do with node |
6749 | * groups, so roll our own. Now each node has its own list of groups which | |
6750 | * gets dynamically allocated. | |
1da177e4 | 6751 | */ |
62ea9ceb | 6752 | static DEFINE_PER_CPU(struct static_sched_domain, node_domains); |
434d53b0 | 6753 | static struct sched_group ***sched_group_nodes_bycpu; |
1da177e4 | 6754 | |
62ea9ceb | 6755 | static DEFINE_PER_CPU(struct static_sched_domain, allnodes_domains); |
6c99e9ad | 6756 | static DEFINE_PER_CPU(struct static_sched_group, sched_group_allnodes); |
9c1cfda2 | 6757 | |
96f874e2 RR |
6758 | static int cpu_to_allnodes_group(int cpu, const struct cpumask *cpu_map, |
6759 | struct sched_group **sg, | |
6760 | struct cpumask *nodemask) | |
9c1cfda2 | 6761 | { |
6711cab4 SS |
6762 | int group; |
6763 | ||
6ca09dfc | 6764 | cpumask_and(nodemask, cpumask_of_node(cpu_to_node(cpu)), cpu_map); |
96f874e2 | 6765 | group = cpumask_first(nodemask); |
6711cab4 SS |
6766 | |
6767 | if (sg) | |
6c99e9ad | 6768 | *sg = &per_cpu(sched_group_allnodes, group).sg; |
6711cab4 | 6769 | return group; |
1da177e4 | 6770 | } |
6711cab4 | 6771 | |
08069033 SS |
6772 | static void init_numa_sched_groups_power(struct sched_group *group_head) |
6773 | { | |
6774 | struct sched_group *sg = group_head; | |
6775 | int j; | |
6776 | ||
6777 | if (!sg) | |
6778 | return; | |
3a5c359a | 6779 | do { |
758b2cdc | 6780 | for_each_cpu(j, sched_group_cpus(sg)) { |
3a5c359a | 6781 | struct sched_domain *sd; |
08069033 | 6782 | |
6c99e9ad | 6783 | sd = &per_cpu(phys_domains, j).sd; |
13318a71 | 6784 | if (j != group_first_cpu(sd->groups)) { |
3a5c359a AK |
6785 | /* |
6786 | * Only add "power" once for each | |
6787 | * physical package. | |
6788 | */ | |
6789 | continue; | |
6790 | } | |
08069033 | 6791 | |
18a3885f | 6792 | sg->cpu_power += sd->groups->cpu_power; |
3a5c359a AK |
6793 | } |
6794 | sg = sg->next; | |
6795 | } while (sg != group_head); | |
08069033 | 6796 | } |
0601a88d AH |
6797 | |
6798 | static int build_numa_sched_groups(struct s_data *d, | |
6799 | const struct cpumask *cpu_map, int num) | |
6800 | { | |
6801 | struct sched_domain *sd; | |
6802 | struct sched_group *sg, *prev; | |
6803 | int n, j; | |
6804 | ||
6805 | cpumask_clear(d->covered); | |
6806 | cpumask_and(d->nodemask, cpumask_of_node(num), cpu_map); | |
6807 | if (cpumask_empty(d->nodemask)) { | |
6808 | d->sched_group_nodes[num] = NULL; | |
6809 | goto out; | |
6810 | } | |
6811 | ||
6812 | sched_domain_node_span(num, d->domainspan); | |
6813 | cpumask_and(d->domainspan, d->domainspan, cpu_map); | |
6814 | ||
6815 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6816 | GFP_KERNEL, num); | |
6817 | if (!sg) { | |
3df0fc5b PZ |
6818 | printk(KERN_WARNING "Can not alloc domain group for node %d\n", |
6819 | num); | |
0601a88d AH |
6820 | return -ENOMEM; |
6821 | } | |
6822 | d->sched_group_nodes[num] = sg; | |
6823 | ||
6824 | for_each_cpu(j, d->nodemask) { | |
6825 | sd = &per_cpu(node_domains, j).sd; | |
6826 | sd->groups = sg; | |
6827 | } | |
6828 | ||
18a3885f | 6829 | sg->cpu_power = 0; |
0601a88d AH |
6830 | cpumask_copy(sched_group_cpus(sg), d->nodemask); |
6831 | sg->next = sg; | |
6832 | cpumask_or(d->covered, d->covered, d->nodemask); | |
6833 | ||
6834 | prev = sg; | |
6835 | for (j = 0; j < nr_node_ids; j++) { | |
6836 | n = (num + j) % nr_node_ids; | |
6837 | cpumask_complement(d->notcovered, d->covered); | |
6838 | cpumask_and(d->tmpmask, d->notcovered, cpu_map); | |
6839 | cpumask_and(d->tmpmask, d->tmpmask, d->domainspan); | |
6840 | if (cpumask_empty(d->tmpmask)) | |
6841 | break; | |
6842 | cpumask_and(d->tmpmask, d->tmpmask, cpumask_of_node(n)); | |
6843 | if (cpumask_empty(d->tmpmask)) | |
6844 | continue; | |
6845 | sg = kmalloc_node(sizeof(struct sched_group) + cpumask_size(), | |
6846 | GFP_KERNEL, num); | |
6847 | if (!sg) { | |
3df0fc5b PZ |
6848 | printk(KERN_WARNING |
6849 | "Can not alloc domain group for node %d\n", j); | |
0601a88d AH |
6850 | return -ENOMEM; |
6851 | } | |
18a3885f | 6852 | sg->cpu_power = 0; |
0601a88d AH |
6853 | cpumask_copy(sched_group_cpus(sg), d->tmpmask); |
6854 | sg->next = prev->next; | |
6855 | cpumask_or(d->covered, d->covered, d->tmpmask); | |
6856 | prev->next = sg; | |
6857 | prev = sg; | |
6858 | } | |
6859 | out: | |
6860 | return 0; | |
6861 | } | |
6d6bc0ad | 6862 | #endif /* CONFIG_NUMA */ |
1da177e4 | 6863 | |
a616058b | 6864 | #ifdef CONFIG_NUMA |
51888ca2 | 6865 | /* Free memory allocated for various sched_group structures */ |
96f874e2 RR |
6866 | static void free_sched_groups(const struct cpumask *cpu_map, |
6867 | struct cpumask *nodemask) | |
51888ca2 | 6868 | { |
a616058b | 6869 | int cpu, i; |
51888ca2 | 6870 | |
abcd083a | 6871 | for_each_cpu(cpu, cpu_map) { |
51888ca2 SV |
6872 | struct sched_group **sched_group_nodes |
6873 | = sched_group_nodes_bycpu[cpu]; | |
6874 | ||
51888ca2 SV |
6875 | if (!sched_group_nodes) |
6876 | continue; | |
6877 | ||
076ac2af | 6878 | for (i = 0; i < nr_node_ids; i++) { |
51888ca2 SV |
6879 | struct sched_group *oldsg, *sg = sched_group_nodes[i]; |
6880 | ||
6ca09dfc | 6881 | cpumask_and(nodemask, cpumask_of_node(i), cpu_map); |
96f874e2 | 6882 | if (cpumask_empty(nodemask)) |
51888ca2 SV |
6883 | continue; |
6884 | ||
6885 | if (sg == NULL) | |
6886 | continue; | |
6887 | sg = sg->next; | |
6888 | next_sg: | |
6889 | oldsg = sg; | |
6890 | sg = sg->next; | |
6891 | kfree(oldsg); | |
6892 | if (oldsg != sched_group_nodes[i]) | |
6893 | goto next_sg; | |
6894 | } | |
6895 | kfree(sched_group_nodes); | |
6896 | sched_group_nodes_bycpu[cpu] = NULL; | |
6897 | } | |
51888ca2 | 6898 | } |
6d6bc0ad | 6899 | #else /* !CONFIG_NUMA */ |
96f874e2 RR |
6900 | static void free_sched_groups(const struct cpumask *cpu_map, |
6901 | struct cpumask *nodemask) | |
a616058b SS |
6902 | { |
6903 | } | |
6d6bc0ad | 6904 | #endif /* CONFIG_NUMA */ |
51888ca2 | 6905 | |
89c4710e SS |
6906 | /* |
6907 | * Initialize sched groups cpu_power. | |
6908 | * | |
6909 | * cpu_power indicates the capacity of sched group, which is used while | |
6910 | * distributing the load between different sched groups in a sched domain. | |
6911 | * Typically cpu_power for all the groups in a sched domain will be same unless | |
6912 | * there are asymmetries in the topology. If there are asymmetries, group | |
6913 | * having more cpu_power will pickup more load compared to the group having | |
6914 | * less cpu_power. | |
89c4710e SS |
6915 | */ |
6916 | static void init_sched_groups_power(int cpu, struct sched_domain *sd) | |
6917 | { | |
6918 | struct sched_domain *child; | |
6919 | struct sched_group *group; | |
f93e65c1 PZ |
6920 | long power; |
6921 | int weight; | |
89c4710e SS |
6922 | |
6923 | WARN_ON(!sd || !sd->groups); | |
6924 | ||
13318a71 | 6925 | if (cpu != group_first_cpu(sd->groups)) |
89c4710e SS |
6926 | return; |
6927 | ||
aae6d3dd SS |
6928 | sd->groups->group_weight = cpumask_weight(sched_group_cpus(sd->groups)); |
6929 | ||
89c4710e SS |
6930 | child = sd->child; |
6931 | ||
18a3885f | 6932 | sd->groups->cpu_power = 0; |
5517d86b | 6933 | |
f93e65c1 PZ |
6934 | if (!child) { |
6935 | power = SCHED_LOAD_SCALE; | |
6936 | weight = cpumask_weight(sched_domain_span(sd)); | |
6937 | /* | |
6938 | * SMT siblings share the power of a single core. | |
a52bfd73 PZ |
6939 | * Usually multiple threads get a better yield out of |
6940 | * that one core than a single thread would have, | |
6941 | * reflect that in sd->smt_gain. | |
f93e65c1 | 6942 | */ |
a52bfd73 PZ |
6943 | if ((sd->flags & SD_SHARE_CPUPOWER) && weight > 1) { |
6944 | power *= sd->smt_gain; | |
f93e65c1 | 6945 | power /= weight; |
a52bfd73 PZ |
6946 | power >>= SCHED_LOAD_SHIFT; |
6947 | } | |
18a3885f | 6948 | sd->groups->cpu_power += power; |
89c4710e SS |
6949 | return; |
6950 | } | |
6951 | ||
89c4710e | 6952 | /* |
f93e65c1 | 6953 | * Add cpu_power of each child group to this groups cpu_power. |
89c4710e SS |
6954 | */ |
6955 | group = child->groups; | |
6956 | do { | |
18a3885f | 6957 | sd->groups->cpu_power += group->cpu_power; |
89c4710e SS |
6958 | group = group->next; |
6959 | } while (group != child->groups); | |
6960 | } | |
6961 | ||
7c16ec58 MT |
6962 | /* |
6963 | * Initializers for schedule domains | |
6964 | * Non-inlined to reduce accumulated stack pressure in build_sched_domains() | |
6965 | */ | |
6966 | ||
a5d8c348 IM |
6967 | #ifdef CONFIG_SCHED_DEBUG |
6968 | # define SD_INIT_NAME(sd, type) sd->name = #type | |
6969 | #else | |
6970 | # define SD_INIT_NAME(sd, type) do { } while (0) | |
6971 | #endif | |
6972 | ||
7c16ec58 | 6973 | #define SD_INIT(sd, type) sd_init_##type(sd) |
a5d8c348 | 6974 | |
7c16ec58 MT |
6975 | #define SD_INIT_FUNC(type) \ |
6976 | static noinline void sd_init_##type(struct sched_domain *sd) \ | |
6977 | { \ | |
6978 | memset(sd, 0, sizeof(*sd)); \ | |
6979 | *sd = SD_##type##_INIT; \ | |
1d3504fc | 6980 | sd->level = SD_LV_##type; \ |
a5d8c348 | 6981 | SD_INIT_NAME(sd, type); \ |
7c16ec58 MT |
6982 | } |
6983 | ||
6984 | SD_INIT_FUNC(CPU) | |
6985 | #ifdef CONFIG_NUMA | |
6986 | SD_INIT_FUNC(ALLNODES) | |
6987 | SD_INIT_FUNC(NODE) | |
6988 | #endif | |
6989 | #ifdef CONFIG_SCHED_SMT | |
6990 | SD_INIT_FUNC(SIBLING) | |
6991 | #endif | |
6992 | #ifdef CONFIG_SCHED_MC | |
6993 | SD_INIT_FUNC(MC) | |
6994 | #endif | |
01a08546 HC |
6995 | #ifdef CONFIG_SCHED_BOOK |
6996 | SD_INIT_FUNC(BOOK) | |
6997 | #endif | |
7c16ec58 | 6998 | |
1d3504fc HS |
6999 | static int default_relax_domain_level = -1; |
7000 | ||
7001 | static int __init setup_relax_domain_level(char *str) | |
7002 | { | |
30e0e178 LZ |
7003 | unsigned long val; |
7004 | ||
7005 | val = simple_strtoul(str, NULL, 0); | |
7006 | if (val < SD_LV_MAX) | |
7007 | default_relax_domain_level = val; | |
7008 | ||
1d3504fc HS |
7009 | return 1; |
7010 | } | |
7011 | __setup("relax_domain_level=", setup_relax_domain_level); | |
7012 | ||
7013 | static void set_domain_attribute(struct sched_domain *sd, | |
7014 | struct sched_domain_attr *attr) | |
7015 | { | |
7016 | int request; | |
7017 | ||
7018 | if (!attr || attr->relax_domain_level < 0) { | |
7019 | if (default_relax_domain_level < 0) | |
7020 | return; | |
7021 | else | |
7022 | request = default_relax_domain_level; | |
7023 | } else | |
7024 | request = attr->relax_domain_level; | |
7025 | if (request < sd->level) { | |
7026 | /* turn off idle balance on this domain */ | |
c88d5910 | 7027 | sd->flags &= ~(SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7028 | } else { |
7029 | /* turn on idle balance on this domain */ | |
c88d5910 | 7030 | sd->flags |= (SD_BALANCE_WAKE|SD_BALANCE_NEWIDLE); |
1d3504fc HS |
7031 | } |
7032 | } | |
7033 | ||
2109b99e AH |
7034 | static void __free_domain_allocs(struct s_data *d, enum s_alloc what, |
7035 | const struct cpumask *cpu_map) | |
7036 | { | |
7037 | switch (what) { | |
7038 | case sa_sched_groups: | |
7039 | free_sched_groups(cpu_map, d->tmpmask); /* fall through */ | |
7040 | d->sched_group_nodes = NULL; | |
7041 | case sa_rootdomain: | |
7042 | free_rootdomain(d->rd); /* fall through */ | |
7043 | case sa_tmpmask: | |
7044 | free_cpumask_var(d->tmpmask); /* fall through */ | |
7045 | case sa_send_covered: | |
7046 | free_cpumask_var(d->send_covered); /* fall through */ | |
01a08546 HC |
7047 | case sa_this_book_map: |
7048 | free_cpumask_var(d->this_book_map); /* fall through */ | |
2109b99e AH |
7049 | case sa_this_core_map: |
7050 | free_cpumask_var(d->this_core_map); /* fall through */ | |
7051 | case sa_this_sibling_map: | |
7052 | free_cpumask_var(d->this_sibling_map); /* fall through */ | |
7053 | case sa_nodemask: | |
7054 | free_cpumask_var(d->nodemask); /* fall through */ | |
7055 | case sa_sched_group_nodes: | |
d1b55138 | 7056 | #ifdef CONFIG_NUMA |
2109b99e AH |
7057 | kfree(d->sched_group_nodes); /* fall through */ |
7058 | case sa_notcovered: | |
7059 | free_cpumask_var(d->notcovered); /* fall through */ | |
7060 | case sa_covered: | |
7061 | free_cpumask_var(d->covered); /* fall through */ | |
7062 | case sa_domainspan: | |
7063 | free_cpumask_var(d->domainspan); /* fall through */ | |
3404c8d9 | 7064 | #endif |
2109b99e AH |
7065 | case sa_none: |
7066 | break; | |
7067 | } | |
7068 | } | |
3404c8d9 | 7069 | |
2109b99e AH |
7070 | static enum s_alloc __visit_domain_allocation_hell(struct s_data *d, |
7071 | const struct cpumask *cpu_map) | |
7072 | { | |
3404c8d9 | 7073 | #ifdef CONFIG_NUMA |
2109b99e AH |
7074 | if (!alloc_cpumask_var(&d->domainspan, GFP_KERNEL)) |
7075 | return sa_none; | |
7076 | if (!alloc_cpumask_var(&d->covered, GFP_KERNEL)) | |
7077 | return sa_domainspan; | |
7078 | if (!alloc_cpumask_var(&d->notcovered, GFP_KERNEL)) | |
7079 | return sa_covered; | |
7080 | /* Allocate the per-node list of sched groups */ | |
7081 | d->sched_group_nodes = kcalloc(nr_node_ids, | |
7082 | sizeof(struct sched_group *), GFP_KERNEL); | |
7083 | if (!d->sched_group_nodes) { | |
3df0fc5b | 7084 | printk(KERN_WARNING "Can not alloc sched group node list\n"); |
2109b99e | 7085 | return sa_notcovered; |
d1b55138 | 7086 | } |
2109b99e | 7087 | sched_group_nodes_bycpu[cpumask_first(cpu_map)] = d->sched_group_nodes; |
d1b55138 | 7088 | #endif |
2109b99e AH |
7089 | if (!alloc_cpumask_var(&d->nodemask, GFP_KERNEL)) |
7090 | return sa_sched_group_nodes; | |
7091 | if (!alloc_cpumask_var(&d->this_sibling_map, GFP_KERNEL)) | |
7092 | return sa_nodemask; | |
7093 | if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) | |
7094 | return sa_this_sibling_map; | |
01a08546 | 7095 | if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL)) |
2109b99e | 7096 | return sa_this_core_map; |
01a08546 HC |
7097 | if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) |
7098 | return sa_this_book_map; | |
2109b99e AH |
7099 | if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) |
7100 | return sa_send_covered; | |
7101 | d->rd = alloc_rootdomain(); | |
7102 | if (!d->rd) { | |
3df0fc5b | 7103 | printk(KERN_WARNING "Cannot alloc root domain\n"); |
2109b99e | 7104 | return sa_tmpmask; |
57d885fe | 7105 | } |
2109b99e AH |
7106 | return sa_rootdomain; |
7107 | } | |
57d885fe | 7108 | |
7f4588f3 AH |
7109 | static struct sched_domain *__build_numa_sched_domains(struct s_data *d, |
7110 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, int i) | |
7111 | { | |
7112 | struct sched_domain *sd = NULL; | |
7c16ec58 | 7113 | #ifdef CONFIG_NUMA |
7f4588f3 | 7114 | struct sched_domain *parent; |
1da177e4 | 7115 | |
7f4588f3 AH |
7116 | d->sd_allnodes = 0; |
7117 | if (cpumask_weight(cpu_map) > | |
7118 | SD_NODES_PER_DOMAIN * cpumask_weight(d->nodemask)) { | |
7119 | sd = &per_cpu(allnodes_domains, i).sd; | |
7120 | SD_INIT(sd, ALLNODES); | |
1d3504fc | 7121 | set_domain_attribute(sd, attr); |
7f4588f3 AH |
7122 | cpumask_copy(sched_domain_span(sd), cpu_map); |
7123 | cpu_to_allnodes_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7124 | d->sd_allnodes = 1; | |
7125 | } | |
7126 | parent = sd; | |
7127 | ||
7128 | sd = &per_cpu(node_domains, i).sd; | |
7129 | SD_INIT(sd, NODE); | |
7130 | set_domain_attribute(sd, attr); | |
7131 | sched_domain_node_span(cpu_to_node(i), sched_domain_span(sd)); | |
7132 | sd->parent = parent; | |
7133 | if (parent) | |
7134 | parent->child = sd; | |
7135 | cpumask_and(sched_domain_span(sd), sched_domain_span(sd), cpu_map); | |
1da177e4 | 7136 | #endif |
7f4588f3 AH |
7137 | return sd; |
7138 | } | |
1da177e4 | 7139 | |
87cce662 AH |
7140 | static struct sched_domain *__build_cpu_sched_domain(struct s_data *d, |
7141 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7142 | struct sched_domain *parent, int i) | |
7143 | { | |
7144 | struct sched_domain *sd; | |
7145 | sd = &per_cpu(phys_domains, i).sd; | |
7146 | SD_INIT(sd, CPU); | |
7147 | set_domain_attribute(sd, attr); | |
7148 | cpumask_copy(sched_domain_span(sd), d->nodemask); | |
7149 | sd->parent = parent; | |
7150 | if (parent) | |
7151 | parent->child = sd; | |
7152 | cpu_to_phys_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7153 | return sd; | |
7154 | } | |
1da177e4 | 7155 | |
01a08546 HC |
7156 | static struct sched_domain *__build_book_sched_domain(struct s_data *d, |
7157 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7158 | struct sched_domain *parent, int i) | |
7159 | { | |
7160 | struct sched_domain *sd = parent; | |
7161 | #ifdef CONFIG_SCHED_BOOK | |
7162 | sd = &per_cpu(book_domains, i).sd; | |
7163 | SD_INIT(sd, BOOK); | |
7164 | set_domain_attribute(sd, attr); | |
7165 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i)); | |
7166 | sd->parent = parent; | |
7167 | parent->child = sd; | |
7168 | cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask); | |
7169 | #endif | |
7170 | return sd; | |
7171 | } | |
7172 | ||
410c4081 AH |
7173 | static struct sched_domain *__build_mc_sched_domain(struct s_data *d, |
7174 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7175 | struct sched_domain *parent, int i) | |
7176 | { | |
7177 | struct sched_domain *sd = parent; | |
1e9f28fa | 7178 | #ifdef CONFIG_SCHED_MC |
410c4081 AH |
7179 | sd = &per_cpu(core_domains, i).sd; |
7180 | SD_INIT(sd, MC); | |
7181 | set_domain_attribute(sd, attr); | |
7182 | cpumask_and(sched_domain_span(sd), cpu_map, cpu_coregroup_mask(i)); | |
7183 | sd->parent = parent; | |
7184 | parent->child = sd; | |
7185 | cpu_to_core_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1e9f28fa | 7186 | #endif |
410c4081 AH |
7187 | return sd; |
7188 | } | |
1e9f28fa | 7189 | |
d8173535 AH |
7190 | static struct sched_domain *__build_smt_sched_domain(struct s_data *d, |
7191 | const struct cpumask *cpu_map, struct sched_domain_attr *attr, | |
7192 | struct sched_domain *parent, int i) | |
7193 | { | |
7194 | struct sched_domain *sd = parent; | |
1da177e4 | 7195 | #ifdef CONFIG_SCHED_SMT |
d8173535 AH |
7196 | sd = &per_cpu(cpu_domains, i).sd; |
7197 | SD_INIT(sd, SIBLING); | |
7198 | set_domain_attribute(sd, attr); | |
7199 | cpumask_and(sched_domain_span(sd), cpu_map, topology_thread_cpumask(i)); | |
7200 | sd->parent = parent; | |
7201 | parent->child = sd; | |
7202 | cpu_to_cpu_group(i, cpu_map, &sd->groups, d->tmpmask); | |
1da177e4 | 7203 | #endif |
d8173535 AH |
7204 | return sd; |
7205 | } | |
1da177e4 | 7206 | |
0e8e85c9 AH |
7207 | static void build_sched_groups(struct s_data *d, enum sched_domain_level l, |
7208 | const struct cpumask *cpu_map, int cpu) | |
7209 | { | |
7210 | switch (l) { | |
1da177e4 | 7211 | #ifdef CONFIG_SCHED_SMT |
0e8e85c9 AH |
7212 | case SD_LV_SIBLING: /* set up CPU (sibling) groups */ |
7213 | cpumask_and(d->this_sibling_map, cpu_map, | |
7214 | topology_thread_cpumask(cpu)); | |
7215 | if (cpu == cpumask_first(d->this_sibling_map)) | |
7216 | init_sched_build_groups(d->this_sibling_map, cpu_map, | |
7217 | &cpu_to_cpu_group, | |
7218 | d->send_covered, d->tmpmask); | |
7219 | break; | |
1da177e4 | 7220 | #endif |
1e9f28fa | 7221 | #ifdef CONFIG_SCHED_MC |
a2af04cd AH |
7222 | case SD_LV_MC: /* set up multi-core groups */ |
7223 | cpumask_and(d->this_core_map, cpu_map, cpu_coregroup_mask(cpu)); | |
7224 | if (cpu == cpumask_first(d->this_core_map)) | |
7225 | init_sched_build_groups(d->this_core_map, cpu_map, | |
7226 | &cpu_to_core_group, | |
7227 | d->send_covered, d->tmpmask); | |
7228 | break; | |
01a08546 HC |
7229 | #endif |
7230 | #ifdef CONFIG_SCHED_BOOK | |
7231 | case SD_LV_BOOK: /* set up book groups */ | |
7232 | cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu)); | |
7233 | if (cpu == cpumask_first(d->this_book_map)) | |
7234 | init_sched_build_groups(d->this_book_map, cpu_map, | |
7235 | &cpu_to_book_group, | |
7236 | d->send_covered, d->tmpmask); | |
7237 | break; | |
1e9f28fa | 7238 | #endif |
86548096 AH |
7239 | case SD_LV_CPU: /* set up physical groups */ |
7240 | cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); | |
7241 | if (!cpumask_empty(d->nodemask)) | |
7242 | init_sched_build_groups(d->nodemask, cpu_map, | |
7243 | &cpu_to_phys_group, | |
7244 | d->send_covered, d->tmpmask); | |
7245 | break; | |
1da177e4 | 7246 | #ifdef CONFIG_NUMA |
de616e36 AH |
7247 | case SD_LV_ALLNODES: |
7248 | init_sched_build_groups(cpu_map, cpu_map, &cpu_to_allnodes_group, | |
7249 | d->send_covered, d->tmpmask); | |
7250 | break; | |
7251 | #endif | |
0e8e85c9 AH |
7252 | default: |
7253 | break; | |
7c16ec58 | 7254 | } |
0e8e85c9 | 7255 | } |
9c1cfda2 | 7256 | |
2109b99e AH |
7257 | /* |
7258 | * Build sched domains for a given set of cpus and attach the sched domains | |
7259 | * to the individual cpus | |
7260 | */ | |
7261 | static int __build_sched_domains(const struct cpumask *cpu_map, | |
7262 | struct sched_domain_attr *attr) | |
7263 | { | |
7264 | enum s_alloc alloc_state = sa_none; | |
7265 | struct s_data d; | |
294b0c96 | 7266 | struct sched_domain *sd; |
2109b99e | 7267 | int i; |
7c16ec58 | 7268 | #ifdef CONFIG_NUMA |
2109b99e | 7269 | d.sd_allnodes = 0; |
7c16ec58 | 7270 | #endif |
9c1cfda2 | 7271 | |
2109b99e AH |
7272 | alloc_state = __visit_domain_allocation_hell(&d, cpu_map); |
7273 | if (alloc_state != sa_rootdomain) | |
7274 | goto error; | |
7275 | alloc_state = sa_sched_groups; | |
9c1cfda2 | 7276 | |
1da177e4 | 7277 | /* |
1a20ff27 | 7278 | * Set up domains for cpus specified by the cpu_map. |
1da177e4 | 7279 | */ |
abcd083a | 7280 | for_each_cpu(i, cpu_map) { |
49a02c51 AH |
7281 | cpumask_and(d.nodemask, cpumask_of_node(cpu_to_node(i)), |
7282 | cpu_map); | |
9761eea8 | 7283 | |
7f4588f3 | 7284 | sd = __build_numa_sched_domains(&d, cpu_map, attr, i); |
87cce662 | 7285 | sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); |
01a08546 | 7286 | sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i); |
410c4081 | 7287 | sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); |
d8173535 | 7288 | sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); |
1da177e4 | 7289 | } |
9c1cfda2 | 7290 | |
abcd083a | 7291 | for_each_cpu(i, cpu_map) { |
0e8e85c9 | 7292 | build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); |
01a08546 | 7293 | build_sched_groups(&d, SD_LV_BOOK, cpu_map, i); |
a2af04cd | 7294 | build_sched_groups(&d, SD_LV_MC, cpu_map, i); |
1da177e4 | 7295 | } |
9c1cfda2 | 7296 | |
1da177e4 | 7297 | /* Set up physical groups */ |
86548096 AH |
7298 | for (i = 0; i < nr_node_ids; i++) |
7299 | build_sched_groups(&d, SD_LV_CPU, cpu_map, i); | |
9c1cfda2 | 7300 | |
1da177e4 LT |
7301 | #ifdef CONFIG_NUMA |
7302 | /* Set up node groups */ | |
de616e36 AH |
7303 | if (d.sd_allnodes) |
7304 | build_sched_groups(&d, SD_LV_ALLNODES, cpu_map, 0); | |
9c1cfda2 | 7305 | |
0601a88d AH |
7306 | for (i = 0; i < nr_node_ids; i++) |
7307 | if (build_numa_sched_groups(&d, cpu_map, i)) | |
51888ca2 | 7308 | goto error; |
1da177e4 LT |
7309 | #endif |
7310 | ||
7311 | /* Calculate CPU power for physical packages and nodes */ | |
5c45bf27 | 7312 | #ifdef CONFIG_SCHED_SMT |
abcd083a | 7313 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7314 | sd = &per_cpu(cpu_domains, i).sd; |
89c4710e | 7315 | init_sched_groups_power(i, sd); |
5c45bf27 | 7316 | } |
1da177e4 | 7317 | #endif |
1e9f28fa | 7318 | #ifdef CONFIG_SCHED_MC |
abcd083a | 7319 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7320 | sd = &per_cpu(core_domains, i).sd; |
89c4710e | 7321 | init_sched_groups_power(i, sd); |
5c45bf27 SS |
7322 | } |
7323 | #endif | |
01a08546 HC |
7324 | #ifdef CONFIG_SCHED_BOOK |
7325 | for_each_cpu(i, cpu_map) { | |
7326 | sd = &per_cpu(book_domains, i).sd; | |
7327 | init_sched_groups_power(i, sd); | |
7328 | } | |
7329 | #endif | |
1e9f28fa | 7330 | |
abcd083a | 7331 | for_each_cpu(i, cpu_map) { |
294b0c96 | 7332 | sd = &per_cpu(phys_domains, i).sd; |
89c4710e | 7333 | init_sched_groups_power(i, sd); |
1da177e4 LT |
7334 | } |
7335 | ||
9c1cfda2 | 7336 | #ifdef CONFIG_NUMA |
076ac2af | 7337 | for (i = 0; i < nr_node_ids; i++) |
49a02c51 | 7338 | init_numa_sched_groups_power(d.sched_group_nodes[i]); |
9c1cfda2 | 7339 | |
49a02c51 | 7340 | if (d.sd_allnodes) { |
6711cab4 | 7341 | struct sched_group *sg; |
f712c0c7 | 7342 | |
96f874e2 | 7343 | cpu_to_allnodes_group(cpumask_first(cpu_map), cpu_map, &sg, |
49a02c51 | 7344 | d.tmpmask); |
f712c0c7 SS |
7345 | init_numa_sched_groups_power(sg); |
7346 | } | |
9c1cfda2 JH |
7347 | #endif |
7348 | ||
1da177e4 | 7349 | /* Attach the domains */ |
abcd083a | 7350 | for_each_cpu(i, cpu_map) { |
1da177e4 | 7351 | #ifdef CONFIG_SCHED_SMT |
6c99e9ad | 7352 | sd = &per_cpu(cpu_domains, i).sd; |
1e9f28fa | 7353 | #elif defined(CONFIG_SCHED_MC) |
6c99e9ad | 7354 | sd = &per_cpu(core_domains, i).sd; |
01a08546 HC |
7355 | #elif defined(CONFIG_SCHED_BOOK) |
7356 | sd = &per_cpu(book_domains, i).sd; | |
1da177e4 | 7357 | #else |
6c99e9ad | 7358 | sd = &per_cpu(phys_domains, i).sd; |
1da177e4 | 7359 | #endif |
49a02c51 | 7360 | cpu_attach_domain(sd, d.rd, i); |
1da177e4 | 7361 | } |
51888ca2 | 7362 | |
2109b99e AH |
7363 | d.sched_group_nodes = NULL; /* don't free this we still need it */ |
7364 | __free_domain_allocs(&d, sa_tmpmask, cpu_map); | |
7365 | return 0; | |
51888ca2 | 7366 | |
51888ca2 | 7367 | error: |
2109b99e AH |
7368 | __free_domain_allocs(&d, alloc_state, cpu_map); |
7369 | return -ENOMEM; | |
1da177e4 | 7370 | } |
029190c5 | 7371 | |
96f874e2 | 7372 | static int build_sched_domains(const struct cpumask *cpu_map) |
1d3504fc HS |
7373 | { |
7374 | return __build_sched_domains(cpu_map, NULL); | |
7375 | } | |
7376 | ||
acc3f5d7 | 7377 | static cpumask_var_t *doms_cur; /* current sched domains */ |
029190c5 | 7378 | static int ndoms_cur; /* number of sched domains in 'doms_cur' */ |
4285f594 IM |
7379 | static struct sched_domain_attr *dattr_cur; |
7380 | /* attribues of custom domains in 'doms_cur' */ | |
029190c5 PJ |
7381 | |
7382 | /* | |
7383 | * Special case: If a kmalloc of a doms_cur partition (array of | |
4212823f RR |
7384 | * cpumask) fails, then fallback to a single sched domain, |
7385 | * as determined by the single cpumask fallback_doms. | |
029190c5 | 7386 | */ |
4212823f | 7387 | static cpumask_var_t fallback_doms; |
029190c5 | 7388 | |
ee79d1bd HC |
7389 | /* |
7390 | * arch_update_cpu_topology lets virtualized architectures update the | |
7391 | * cpu core maps. It is supposed to return 1 if the topology changed | |
7392 | * or 0 if it stayed the same. | |
7393 | */ | |
7394 | int __attribute__((weak)) arch_update_cpu_topology(void) | |
22e52b07 | 7395 | { |
ee79d1bd | 7396 | return 0; |
22e52b07 HC |
7397 | } |
7398 | ||
acc3f5d7 RR |
7399 | cpumask_var_t *alloc_sched_domains(unsigned int ndoms) |
7400 | { | |
7401 | int i; | |
7402 | cpumask_var_t *doms; | |
7403 | ||
7404 | doms = kmalloc(sizeof(*doms) * ndoms, GFP_KERNEL); | |
7405 | if (!doms) | |
7406 | return NULL; | |
7407 | for (i = 0; i < ndoms; i++) { | |
7408 | if (!alloc_cpumask_var(&doms[i], GFP_KERNEL)) { | |
7409 | free_sched_domains(doms, i); | |
7410 | return NULL; | |
7411 | } | |
7412 | } | |
7413 | return doms; | |
7414 | } | |
7415 | ||
7416 | void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms) | |
7417 | { | |
7418 | unsigned int i; | |
7419 | for (i = 0; i < ndoms; i++) | |
7420 | free_cpumask_var(doms[i]); | |
7421 | kfree(doms); | |
7422 | } | |
7423 | ||
1a20ff27 | 7424 | /* |
41a2d6cf | 7425 | * Set up scheduler domains and groups. Callers must hold the hotplug lock. |
029190c5 PJ |
7426 | * For now this just excludes isolated cpus, but could be used to |
7427 | * exclude other special cases in the future. | |
1a20ff27 | 7428 | */ |
96f874e2 | 7429 | static int arch_init_sched_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7430 | { |
7378547f MM |
7431 | int err; |
7432 | ||
22e52b07 | 7433 | arch_update_cpu_topology(); |
029190c5 | 7434 | ndoms_cur = 1; |
acc3f5d7 | 7435 | doms_cur = alloc_sched_domains(ndoms_cur); |
029190c5 | 7436 | if (!doms_cur) |
acc3f5d7 RR |
7437 | doms_cur = &fallback_doms; |
7438 | cpumask_andnot(doms_cur[0], cpu_map, cpu_isolated_map); | |
1d3504fc | 7439 | dattr_cur = NULL; |
acc3f5d7 | 7440 | err = build_sched_domains(doms_cur[0]); |
6382bc90 | 7441 | register_sched_domain_sysctl(); |
7378547f MM |
7442 | |
7443 | return err; | |
1a20ff27 DG |
7444 | } |
7445 | ||
96f874e2 RR |
7446 | static void arch_destroy_sched_domains(const struct cpumask *cpu_map, |
7447 | struct cpumask *tmpmask) | |
1da177e4 | 7448 | { |
7c16ec58 | 7449 | free_sched_groups(cpu_map, tmpmask); |
9c1cfda2 | 7450 | } |
1da177e4 | 7451 | |
1a20ff27 DG |
7452 | /* |
7453 | * Detach sched domains from a group of cpus specified in cpu_map | |
7454 | * These cpus will now be attached to the NULL domain | |
7455 | */ | |
96f874e2 | 7456 | static void detach_destroy_domains(const struct cpumask *cpu_map) |
1a20ff27 | 7457 | { |
96f874e2 RR |
7458 | /* Save because hotplug lock held. */ |
7459 | static DECLARE_BITMAP(tmpmask, CONFIG_NR_CPUS); | |
1a20ff27 DG |
7460 | int i; |
7461 | ||
abcd083a | 7462 | for_each_cpu(i, cpu_map) |
57d885fe | 7463 | cpu_attach_domain(NULL, &def_root_domain, i); |
1a20ff27 | 7464 | synchronize_sched(); |
96f874e2 | 7465 | arch_destroy_sched_domains(cpu_map, to_cpumask(tmpmask)); |
1a20ff27 DG |
7466 | } |
7467 | ||
1d3504fc HS |
7468 | /* handle null as "default" */ |
7469 | static int dattrs_equal(struct sched_domain_attr *cur, int idx_cur, | |
7470 | struct sched_domain_attr *new, int idx_new) | |
7471 | { | |
7472 | struct sched_domain_attr tmp; | |
7473 | ||
7474 | /* fast path */ | |
7475 | if (!new && !cur) | |
7476 | return 1; | |
7477 | ||
7478 | tmp = SD_ATTR_INIT; | |
7479 | return !memcmp(cur ? (cur + idx_cur) : &tmp, | |
7480 | new ? (new + idx_new) : &tmp, | |
7481 | sizeof(struct sched_domain_attr)); | |
7482 | } | |
7483 | ||
029190c5 PJ |
7484 | /* |
7485 | * Partition sched domains as specified by the 'ndoms_new' | |
41a2d6cf | 7486 | * cpumasks in the array doms_new[] of cpumasks. This compares |
029190c5 PJ |
7487 | * doms_new[] to the current sched domain partitioning, doms_cur[]. |
7488 | * It destroys each deleted domain and builds each new domain. | |
7489 | * | |
acc3f5d7 | 7490 | * 'doms_new' is an array of cpumask_var_t's of length 'ndoms_new'. |
41a2d6cf IM |
7491 | * The masks don't intersect (don't overlap.) We should setup one |
7492 | * sched domain for each mask. CPUs not in any of the cpumasks will | |
7493 | * not be load balanced. If the same cpumask appears both in the | |
029190c5 PJ |
7494 | * current 'doms_cur' domains and in the new 'doms_new', we can leave |
7495 | * it as it is. | |
7496 | * | |
acc3f5d7 RR |
7497 | * The passed in 'doms_new' should be allocated using |
7498 | * alloc_sched_domains. This routine takes ownership of it and will | |
7499 | * free_sched_domains it when done with it. If the caller failed the | |
7500 | * alloc call, then it can pass in doms_new == NULL && ndoms_new == 1, | |
7501 | * and partition_sched_domains() will fallback to the single partition | |
7502 | * 'fallback_doms', it also forces the domains to be rebuilt. | |
029190c5 | 7503 | * |
96f874e2 | 7504 | * If doms_new == NULL it will be replaced with cpu_online_mask. |
700018e0 LZ |
7505 | * ndoms_new == 0 is a special case for destroying existing domains, |
7506 | * and it will not create the default domain. | |
dfb512ec | 7507 | * |
029190c5 PJ |
7508 | * Call with hotplug lock held |
7509 | */ | |
acc3f5d7 | 7510 | void partition_sched_domains(int ndoms_new, cpumask_var_t doms_new[], |
1d3504fc | 7511 | struct sched_domain_attr *dattr_new) |
029190c5 | 7512 | { |
dfb512ec | 7513 | int i, j, n; |
d65bd5ec | 7514 | int new_topology; |
029190c5 | 7515 | |
712555ee | 7516 | mutex_lock(&sched_domains_mutex); |
a1835615 | 7517 | |
7378547f MM |
7518 | /* always unregister in case we don't destroy any domains */ |
7519 | unregister_sched_domain_sysctl(); | |
7520 | ||
d65bd5ec HC |
7521 | /* Let architecture update cpu core mappings. */ |
7522 | new_topology = arch_update_cpu_topology(); | |
7523 | ||
dfb512ec | 7524 | n = doms_new ? ndoms_new : 0; |
029190c5 PJ |
7525 | |
7526 | /* Destroy deleted domains */ | |
7527 | for (i = 0; i < ndoms_cur; i++) { | |
d65bd5ec | 7528 | for (j = 0; j < n && !new_topology; j++) { |
acc3f5d7 | 7529 | if (cpumask_equal(doms_cur[i], doms_new[j]) |
1d3504fc | 7530 | && dattrs_equal(dattr_cur, i, dattr_new, j)) |
029190c5 PJ |
7531 | goto match1; |
7532 | } | |
7533 | /* no match - a current sched domain not in new doms_new[] */ | |
acc3f5d7 | 7534 | detach_destroy_domains(doms_cur[i]); |
029190c5 PJ |
7535 | match1: |
7536 | ; | |
7537 | } | |
7538 | ||
e761b772 MK |
7539 | if (doms_new == NULL) { |
7540 | ndoms_cur = 0; | |
acc3f5d7 | 7541 | doms_new = &fallback_doms; |
6ad4c188 | 7542 | cpumask_andnot(doms_new[0], cpu_active_mask, cpu_isolated_map); |
faa2f98f | 7543 | WARN_ON_ONCE(dattr_new); |
e761b772 MK |
7544 | } |
7545 | ||
029190c5 PJ |
7546 | /* Build new domains */ |
7547 | for (i = 0; i < ndoms_new; i++) { | |
d65bd5ec | 7548 | for (j = 0; j < ndoms_cur && !new_topology; j++) { |
acc3f5d7 | 7549 | if (cpumask_equal(doms_new[i], doms_cur[j]) |
1d3504fc | 7550 | && dattrs_equal(dattr_new, i, dattr_cur, j)) |
029190c5 PJ |
7551 | goto match2; |
7552 | } | |
7553 | /* no match - add a new doms_new */ | |
acc3f5d7 | 7554 | __build_sched_domains(doms_new[i], |
1d3504fc | 7555 | dattr_new ? dattr_new + i : NULL); |
029190c5 PJ |
7556 | match2: |
7557 | ; | |
7558 | } | |
7559 | ||
7560 | /* Remember the new sched domains */ | |
acc3f5d7 RR |
7561 | if (doms_cur != &fallback_doms) |
7562 | free_sched_domains(doms_cur, ndoms_cur); | |
1d3504fc | 7563 | kfree(dattr_cur); /* kfree(NULL) is safe */ |
029190c5 | 7564 | doms_cur = doms_new; |
1d3504fc | 7565 | dattr_cur = dattr_new; |
029190c5 | 7566 | ndoms_cur = ndoms_new; |
7378547f MM |
7567 | |
7568 | register_sched_domain_sysctl(); | |
a1835615 | 7569 | |
712555ee | 7570 | mutex_unlock(&sched_domains_mutex); |
029190c5 PJ |
7571 | } |
7572 | ||
5c45bf27 | 7573 | #if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) |
c70f22d2 | 7574 | static void arch_reinit_sched_domains(void) |
5c45bf27 | 7575 | { |
95402b38 | 7576 | get_online_cpus(); |
dfb512ec MK |
7577 | |
7578 | /* Destroy domains first to force the rebuild */ | |
7579 | partition_sched_domains(0, NULL, NULL); | |
7580 | ||
e761b772 | 7581 | rebuild_sched_domains(); |
95402b38 | 7582 | put_online_cpus(); |
5c45bf27 SS |
7583 | } |
7584 | ||
7585 | static ssize_t sched_power_savings_store(const char *buf, size_t count, int smt) | |
7586 | { | |
afb8a9b7 | 7587 | unsigned int level = 0; |
5c45bf27 | 7588 | |
afb8a9b7 GS |
7589 | if (sscanf(buf, "%u", &level) != 1) |
7590 | return -EINVAL; | |
7591 | ||
7592 | /* | |
7593 | * level is always be positive so don't check for | |
7594 | * level < POWERSAVINGS_BALANCE_NONE which is 0 | |
7595 | * What happens on 0 or 1 byte write, | |
7596 | * need to check for count as well? | |
7597 | */ | |
7598 | ||
7599 | if (level >= MAX_POWERSAVINGS_BALANCE_LEVELS) | |
5c45bf27 SS |
7600 | return -EINVAL; |
7601 | ||
7602 | if (smt) | |
afb8a9b7 | 7603 | sched_smt_power_savings = level; |
5c45bf27 | 7604 | else |
afb8a9b7 | 7605 | sched_mc_power_savings = level; |
5c45bf27 | 7606 | |
c70f22d2 | 7607 | arch_reinit_sched_domains(); |
5c45bf27 | 7608 | |
c70f22d2 | 7609 | return count; |
5c45bf27 SS |
7610 | } |
7611 | ||
5c45bf27 | 7612 | #ifdef CONFIG_SCHED_MC |
f718cd4a | 7613 | static ssize_t sched_mc_power_savings_show(struct sysdev_class *class, |
c9be0a36 | 7614 | struct sysdev_class_attribute *attr, |
f718cd4a | 7615 | char *page) |
5c45bf27 SS |
7616 | { |
7617 | return sprintf(page, "%u\n", sched_mc_power_savings); | |
7618 | } | |
f718cd4a | 7619 | static ssize_t sched_mc_power_savings_store(struct sysdev_class *class, |
c9be0a36 | 7620 | struct sysdev_class_attribute *attr, |
48f24c4d | 7621 | const char *buf, size_t count) |
5c45bf27 SS |
7622 | { |
7623 | return sched_power_savings_store(buf, count, 0); | |
7624 | } | |
f718cd4a AK |
7625 | static SYSDEV_CLASS_ATTR(sched_mc_power_savings, 0644, |
7626 | sched_mc_power_savings_show, | |
7627 | sched_mc_power_savings_store); | |
5c45bf27 SS |
7628 | #endif |
7629 | ||
7630 | #ifdef CONFIG_SCHED_SMT | |
f718cd4a | 7631 | static ssize_t sched_smt_power_savings_show(struct sysdev_class *dev, |
c9be0a36 | 7632 | struct sysdev_class_attribute *attr, |
f718cd4a | 7633 | char *page) |
5c45bf27 SS |
7634 | { |
7635 | return sprintf(page, "%u\n", sched_smt_power_savings); | |
7636 | } | |
f718cd4a | 7637 | static ssize_t sched_smt_power_savings_store(struct sysdev_class *dev, |
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, 1); | |
7642 | } | |
f718cd4a AK |
7643 | static SYSDEV_CLASS_ATTR(sched_smt_power_savings, 0644, |
7644 | sched_smt_power_savings_show, | |
6707de00 AB |
7645 | sched_smt_power_savings_store); |
7646 | #endif | |
7647 | ||
39aac648 | 7648 | int __init sched_create_sysfs_power_savings_entries(struct sysdev_class *cls) |
6707de00 AB |
7649 | { |
7650 | int err = 0; | |
7651 | ||
7652 | #ifdef CONFIG_SCHED_SMT | |
7653 | if (smt_capable()) | |
7654 | err = sysfs_create_file(&cls->kset.kobj, | |
7655 | &attr_sched_smt_power_savings.attr); | |
7656 | #endif | |
7657 | #ifdef CONFIG_SCHED_MC | |
7658 | if (!err && mc_capable()) | |
7659 | err = sysfs_create_file(&cls->kset.kobj, | |
7660 | &attr_sched_mc_power_savings.attr); | |
7661 | #endif | |
7662 | return err; | |
7663 | } | |
6d6bc0ad | 7664 | #endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */ |
5c45bf27 | 7665 | |
1da177e4 | 7666 | /* |
3a101d05 TH |
7667 | * Update cpusets according to cpu_active mask. If cpusets are |
7668 | * disabled, cpuset_update_active_cpus() becomes a simple wrapper | |
7669 | * around partition_sched_domains(). | |
1da177e4 | 7670 | */ |
0b2e918a TH |
7671 | static int cpuset_cpu_active(struct notifier_block *nfb, unsigned long action, |
7672 | void *hcpu) | |
e761b772 | 7673 | { |
3a101d05 | 7674 | switch (action & ~CPU_TASKS_FROZEN) { |
e761b772 | 7675 | case CPU_ONLINE: |
6ad4c188 | 7676 | case CPU_DOWN_FAILED: |
3a101d05 | 7677 | cpuset_update_active_cpus(); |
e761b772 | 7678 | return NOTIFY_OK; |
3a101d05 TH |
7679 | default: |
7680 | return NOTIFY_DONE; | |
7681 | } | |
7682 | } | |
e761b772 | 7683 | |
0b2e918a TH |
7684 | static int cpuset_cpu_inactive(struct notifier_block *nfb, unsigned long action, |
7685 | void *hcpu) | |
3a101d05 TH |
7686 | { |
7687 | switch (action & ~CPU_TASKS_FROZEN) { | |
7688 | case CPU_DOWN_PREPARE: | |
7689 | cpuset_update_active_cpus(); | |
7690 | return NOTIFY_OK; | |
e761b772 MK |
7691 | default: |
7692 | return NOTIFY_DONE; | |
7693 | } | |
7694 | } | |
e761b772 MK |
7695 | |
7696 | static int update_runtime(struct notifier_block *nfb, | |
7697 | unsigned long action, void *hcpu) | |
1da177e4 | 7698 | { |
7def2be1 PZ |
7699 | int cpu = (int)(long)hcpu; |
7700 | ||
1da177e4 | 7701 | switch (action) { |
1da177e4 | 7702 | case CPU_DOWN_PREPARE: |
8bb78442 | 7703 | case CPU_DOWN_PREPARE_FROZEN: |
7def2be1 | 7704 | disable_runtime(cpu_rq(cpu)); |
1da177e4 LT |
7705 | return NOTIFY_OK; |
7706 | ||
1da177e4 | 7707 | case CPU_DOWN_FAILED: |
8bb78442 | 7708 | case CPU_DOWN_FAILED_FROZEN: |
1da177e4 | 7709 | case CPU_ONLINE: |
8bb78442 | 7710 | case CPU_ONLINE_FROZEN: |
7def2be1 | 7711 | enable_runtime(cpu_rq(cpu)); |
e761b772 MK |
7712 | return NOTIFY_OK; |
7713 | ||
1da177e4 LT |
7714 | default: |
7715 | return NOTIFY_DONE; | |
7716 | } | |
1da177e4 | 7717 | } |
1da177e4 LT |
7718 | |
7719 | void __init sched_init_smp(void) | |
7720 | { | |
dcc30a35 RR |
7721 | cpumask_var_t non_isolated_cpus; |
7722 | ||
7723 | alloc_cpumask_var(&non_isolated_cpus, GFP_KERNEL); | |
cb5fd13f | 7724 | alloc_cpumask_var(&fallback_doms, GFP_KERNEL); |
5c1e1767 | 7725 | |
434d53b0 MT |
7726 | #if defined(CONFIG_NUMA) |
7727 | sched_group_nodes_bycpu = kzalloc(nr_cpu_ids * sizeof(void **), | |
7728 | GFP_KERNEL); | |
7729 | BUG_ON(sched_group_nodes_bycpu == NULL); | |
7730 | #endif | |
95402b38 | 7731 | get_online_cpus(); |
712555ee | 7732 | mutex_lock(&sched_domains_mutex); |
6ad4c188 | 7733 | arch_init_sched_domains(cpu_active_mask); |
dcc30a35 RR |
7734 | cpumask_andnot(non_isolated_cpus, cpu_possible_mask, cpu_isolated_map); |
7735 | if (cpumask_empty(non_isolated_cpus)) | |
7736 | cpumask_set_cpu(smp_processor_id(), non_isolated_cpus); | |
712555ee | 7737 | mutex_unlock(&sched_domains_mutex); |
95402b38 | 7738 | put_online_cpus(); |
e761b772 | 7739 | |
3a101d05 TH |
7740 | hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); |
7741 | hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); | |
e761b772 MK |
7742 | |
7743 | /* RT runtime code needs to handle some hotplug events */ | |
7744 | hotcpu_notifier(update_runtime, 0); | |
7745 | ||
b328ca18 | 7746 | init_hrtick(); |
5c1e1767 NP |
7747 | |
7748 | /* Move init over to a non-isolated CPU */ | |
dcc30a35 | 7749 | if (set_cpus_allowed_ptr(current, non_isolated_cpus) < 0) |
5c1e1767 | 7750 | BUG(); |
19978ca6 | 7751 | sched_init_granularity(); |
dcc30a35 | 7752 | free_cpumask_var(non_isolated_cpus); |
4212823f | 7753 | |
0e3900e6 | 7754 | init_sched_rt_class(); |
1da177e4 LT |
7755 | } |
7756 | #else | |
7757 | void __init sched_init_smp(void) | |
7758 | { | |
19978ca6 | 7759 | sched_init_granularity(); |
1da177e4 LT |
7760 | } |
7761 | #endif /* CONFIG_SMP */ | |
7762 | ||
cd1bb94b AB |
7763 | const_debug unsigned int sysctl_timer_migration = 1; |
7764 | ||
1da177e4 LT |
7765 | int in_sched_functions(unsigned long addr) |
7766 | { | |
1da177e4 LT |
7767 | return in_lock_functions(addr) || |
7768 | (addr >= (unsigned long)__sched_text_start | |
7769 | && addr < (unsigned long)__sched_text_end); | |
7770 | } | |
7771 | ||
a9957449 | 7772 | static void init_cfs_rq(struct cfs_rq *cfs_rq, struct rq *rq) |
dd41f596 IM |
7773 | { |
7774 | cfs_rq->tasks_timeline = RB_ROOT; | |
4a55bd5e | 7775 | INIT_LIST_HEAD(&cfs_rq->tasks); |
dd41f596 IM |
7776 | #ifdef CONFIG_FAIR_GROUP_SCHED |
7777 | cfs_rq->rq = rq; | |
7778 | #endif | |
67e9fb2a | 7779 | cfs_rq->min_vruntime = (u64)(-(1LL << 20)); |
dd41f596 IM |
7780 | } |
7781 | ||
fa85ae24 PZ |
7782 | static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq) |
7783 | { | |
7784 | struct rt_prio_array *array; | |
7785 | int i; | |
7786 | ||
7787 | array = &rt_rq->active; | |
7788 | for (i = 0; i < MAX_RT_PRIO; i++) { | |
7789 | INIT_LIST_HEAD(array->queue + i); | |
7790 | __clear_bit(i, array->bitmap); | |
7791 | } | |
7792 | /* delimiter for bitsearch: */ | |
7793 | __set_bit(MAX_RT_PRIO, array->bitmap); | |
7794 | ||
052f1dc7 | 7795 | #if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED |
e864c499 | 7796 | rt_rq->highest_prio.curr = MAX_RT_PRIO; |
398a153b | 7797 | #ifdef CONFIG_SMP |
e864c499 | 7798 | rt_rq->highest_prio.next = MAX_RT_PRIO; |
48d5e258 | 7799 | #endif |
48d5e258 | 7800 | #endif |
fa85ae24 PZ |
7801 | #ifdef CONFIG_SMP |
7802 | rt_rq->rt_nr_migratory = 0; | |
fa85ae24 | 7803 | rt_rq->overloaded = 0; |
05fa785c | 7804 | plist_head_init_raw(&rt_rq->pushable_tasks, &rq->lock); |
fa85ae24 PZ |
7805 | #endif |
7806 | ||
7807 | rt_rq->rt_time = 0; | |
7808 | rt_rq->rt_throttled = 0; | |
ac086bc2 | 7809 | rt_rq->rt_runtime = 0; |
0986b11b | 7810 | raw_spin_lock_init(&rt_rq->rt_runtime_lock); |
6f505b16 | 7811 | |
052f1dc7 | 7812 | #ifdef CONFIG_RT_GROUP_SCHED |
23b0fdfc | 7813 | rt_rq->rt_nr_boosted = 0; |
6f505b16 PZ |
7814 | rt_rq->rq = rq; |
7815 | #endif | |
fa85ae24 PZ |
7816 | } |
7817 | ||
6f505b16 | 7818 | #ifdef CONFIG_FAIR_GROUP_SCHED |
ec7dc8ac | 7819 | static void init_tg_cfs_entry(struct task_group *tg, struct cfs_rq *cfs_rq, |
3d4b47b4 | 7820 | struct sched_entity *se, int cpu, |
ec7dc8ac | 7821 | struct sched_entity *parent) |
6f505b16 | 7822 | { |
ec7dc8ac | 7823 | struct rq *rq = cpu_rq(cpu); |
6f505b16 PZ |
7824 | tg->cfs_rq[cpu] = cfs_rq; |
7825 | init_cfs_rq(cfs_rq, rq); | |
7826 | cfs_rq->tg = tg; | |
6f505b16 PZ |
7827 | |
7828 | tg->se[cpu] = se; | |
07e06b01 | 7829 | /* se could be NULL for root_task_group */ |
354d60c2 DG |
7830 | if (!se) |
7831 | return; | |
7832 | ||
ec7dc8ac DG |
7833 | if (!parent) |
7834 | se->cfs_rq = &rq->cfs; | |
7835 | else | |
7836 | se->cfs_rq = parent->my_q; | |
7837 | ||
6f505b16 | 7838 | se->my_q = cfs_rq; |
9437178f | 7839 | update_load_set(&se->load, 0); |
ec7dc8ac | 7840 | se->parent = parent; |
6f505b16 | 7841 | } |
052f1dc7 | 7842 | #endif |
6f505b16 | 7843 | |
052f1dc7 | 7844 | #ifdef CONFIG_RT_GROUP_SCHED |
ec7dc8ac | 7845 | static void init_tg_rt_entry(struct task_group *tg, struct rt_rq *rt_rq, |
3d4b47b4 | 7846 | struct sched_rt_entity *rt_se, int cpu, |
ec7dc8ac | 7847 | struct sched_rt_entity *parent) |
6f505b16 | 7848 | { |
ec7dc8ac DG |
7849 | struct rq *rq = cpu_rq(cpu); |
7850 | ||
6f505b16 PZ |
7851 | tg->rt_rq[cpu] = rt_rq; |
7852 | init_rt_rq(rt_rq, rq); | |
7853 | rt_rq->tg = tg; | |
ac086bc2 | 7854 | rt_rq->rt_runtime = tg->rt_bandwidth.rt_runtime; |
6f505b16 PZ |
7855 | |
7856 | tg->rt_se[cpu] = rt_se; | |
354d60c2 DG |
7857 | if (!rt_se) |
7858 | return; | |
7859 | ||
ec7dc8ac DG |
7860 | if (!parent) |
7861 | rt_se->rt_rq = &rq->rt; | |
7862 | else | |
7863 | rt_se->rt_rq = parent->my_q; | |
7864 | ||
6f505b16 | 7865 | rt_se->my_q = rt_rq; |
ec7dc8ac | 7866 | rt_se->parent = parent; |
6f505b16 PZ |
7867 | INIT_LIST_HEAD(&rt_se->run_list); |
7868 | } | |
7869 | #endif | |
7870 | ||
1da177e4 LT |
7871 | void __init sched_init(void) |
7872 | { | |
dd41f596 | 7873 | int i, j; |
434d53b0 MT |
7874 | unsigned long alloc_size = 0, ptr; |
7875 | ||
7876 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
7877 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
7878 | #endif | |
7879 | #ifdef CONFIG_RT_GROUP_SCHED | |
7880 | alloc_size += 2 * nr_cpu_ids * sizeof(void **); | |
eff766a6 | 7881 | #endif |
df7c8e84 | 7882 | #ifdef CONFIG_CPUMASK_OFFSTACK |
8c083f08 | 7883 | alloc_size += num_possible_cpus() * cpumask_size(); |
434d53b0 | 7884 | #endif |
434d53b0 | 7885 | if (alloc_size) { |
36b7b6d4 | 7886 | ptr = (unsigned long)kzalloc(alloc_size, GFP_NOWAIT); |
434d53b0 MT |
7887 | |
7888 | #ifdef CONFIG_FAIR_GROUP_SCHED | |
07e06b01 | 7889 | root_task_group.se = (struct sched_entity **)ptr; |
434d53b0 MT |
7890 | ptr += nr_cpu_ids * sizeof(void **); |
7891 | ||
07e06b01 | 7892 | root_task_group.cfs_rq = (struct cfs_rq **)ptr; |
434d53b0 | 7893 | ptr += nr_cpu_ids * sizeof(void **); |
eff766a6 | 7894 | |
6d6bc0ad | 7895 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
434d53b0 | 7896 | #ifdef CONFIG_RT_GROUP_SCHED |
07e06b01 | 7897 | root_task_group.rt_se = (struct sched_rt_entity **)ptr; |
434d53b0 MT |
7898 | ptr += nr_cpu_ids * sizeof(void **); |
7899 | ||
07e06b01 | 7900 | root_task_group.rt_rq = (struct rt_rq **)ptr; |
eff766a6 PZ |
7901 | ptr += nr_cpu_ids * sizeof(void **); |
7902 | ||
6d6bc0ad | 7903 | #endif /* CONFIG_RT_GROUP_SCHED */ |
df7c8e84 RR |
7904 | #ifdef CONFIG_CPUMASK_OFFSTACK |
7905 | for_each_possible_cpu(i) { | |
7906 | per_cpu(load_balance_tmpmask, i) = (void *)ptr; | |
7907 | ptr += cpumask_size(); | |
7908 | } | |
7909 | #endif /* CONFIG_CPUMASK_OFFSTACK */ | |
434d53b0 | 7910 | } |
dd41f596 | 7911 | |
57d885fe GH |
7912 | #ifdef CONFIG_SMP |
7913 | init_defrootdomain(); | |
7914 | #endif | |
7915 | ||
d0b27fa7 PZ |
7916 | init_rt_bandwidth(&def_rt_bandwidth, |
7917 | global_rt_period(), global_rt_runtime()); | |
7918 | ||
7919 | #ifdef CONFIG_RT_GROUP_SCHED | |
07e06b01 | 7920 | init_rt_bandwidth(&root_task_group.rt_bandwidth, |
d0b27fa7 | 7921 | global_rt_period(), global_rt_runtime()); |
6d6bc0ad | 7922 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 | 7923 | |
7c941438 | 7924 | #ifdef CONFIG_CGROUP_SCHED |
07e06b01 YZ |
7925 | list_add(&root_task_group.list, &task_groups); |
7926 | INIT_LIST_HEAD(&root_task_group.children); | |
5091faa4 | 7927 | autogroup_init(&init_task); |
7c941438 | 7928 | #endif /* CONFIG_CGROUP_SCHED */ |
6f505b16 | 7929 | |
0a945022 | 7930 | for_each_possible_cpu(i) { |
70b97a7f | 7931 | struct rq *rq; |
1da177e4 LT |
7932 | |
7933 | rq = cpu_rq(i); | |
05fa785c | 7934 | raw_spin_lock_init(&rq->lock); |
7897986b | 7935 | rq->nr_running = 0; |
dce48a84 TG |
7936 | rq->calc_load_active = 0; |
7937 | rq->calc_load_update = jiffies + LOAD_FREQ; | |
dd41f596 | 7938 | init_cfs_rq(&rq->cfs, rq); |
6f505b16 | 7939 | init_rt_rq(&rq->rt, rq); |
dd41f596 | 7940 | #ifdef CONFIG_FAIR_GROUP_SCHED |
07e06b01 | 7941 | root_task_group.shares = root_task_group_load; |
6f505b16 | 7942 | INIT_LIST_HEAD(&rq->leaf_cfs_rq_list); |
354d60c2 | 7943 | /* |
07e06b01 | 7944 | * How much cpu bandwidth does root_task_group get? |
354d60c2 DG |
7945 | * |
7946 | * In case of task-groups formed thr' the cgroup filesystem, it | |
7947 | * gets 100% of the cpu resources in the system. This overall | |
7948 | * system cpu resource is divided among the tasks of | |
07e06b01 | 7949 | * root_task_group and its child task-groups in a fair manner, |
354d60c2 DG |
7950 | * based on each entity's (task or task-group's) weight |
7951 | * (se->load.weight). | |
7952 | * | |
07e06b01 | 7953 | * In other words, if root_task_group has 10 tasks of weight |
354d60c2 DG |
7954 | * 1024) and two child groups A0 and A1 (of weight 1024 each), |
7955 | * then A0's share of the cpu resource is: | |
7956 | * | |
0d905bca | 7957 | * A0's bandwidth = 1024 / (10*1024 + 1024 + 1024) = 8.33% |
354d60c2 | 7958 | * |
07e06b01 YZ |
7959 | * We achieve this by letting root_task_group's tasks sit |
7960 | * directly in rq->cfs (i.e root_task_group->se[] = NULL). | |
354d60c2 | 7961 | */ |
07e06b01 | 7962 | init_tg_cfs_entry(&root_task_group, &rq->cfs, NULL, i, NULL); |
354d60c2 DG |
7963 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
7964 | ||
7965 | rq->rt.rt_runtime = def_rt_bandwidth.rt_runtime; | |
052f1dc7 | 7966 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 7967 | INIT_LIST_HEAD(&rq->leaf_rt_rq_list); |
07e06b01 | 7968 | init_tg_rt_entry(&root_task_group, &rq->rt, NULL, i, NULL); |
dd41f596 | 7969 | #endif |
1da177e4 | 7970 | |
dd41f596 IM |
7971 | for (j = 0; j < CPU_LOAD_IDX_MAX; j++) |
7972 | rq->cpu_load[j] = 0; | |
fdf3e95d VP |
7973 | |
7974 | rq->last_load_update_tick = jiffies; | |
7975 | ||
1da177e4 | 7976 | #ifdef CONFIG_SMP |
41c7ce9a | 7977 | rq->sd = NULL; |
57d885fe | 7978 | rq->rd = NULL; |
e51fd5e2 | 7979 | rq->cpu_power = SCHED_LOAD_SCALE; |
3f029d3c | 7980 | rq->post_schedule = 0; |
1da177e4 | 7981 | rq->active_balance = 0; |
dd41f596 | 7982 | rq->next_balance = jiffies; |
1da177e4 | 7983 | rq->push_cpu = 0; |
0a2966b4 | 7984 | rq->cpu = i; |
1f11eb6a | 7985 | rq->online = 0; |
eae0c9df MG |
7986 | rq->idle_stamp = 0; |
7987 | rq->avg_idle = 2*sysctl_sched_migration_cost; | |
dc938520 | 7988 | rq_attach_root(rq, &def_root_domain); |
83cd4fe2 VP |
7989 | #ifdef CONFIG_NO_HZ |
7990 | rq->nohz_balance_kick = 0; | |
7991 | init_sched_softirq_csd(&per_cpu(remote_sched_softirq_cb, i)); | |
7992 | #endif | |
1da177e4 | 7993 | #endif |
8f4d37ec | 7994 | init_rq_hrtick(rq); |
1da177e4 | 7995 | atomic_set(&rq->nr_iowait, 0); |
1da177e4 LT |
7996 | } |
7997 | ||
2dd73a4f | 7998 | set_load_weight(&init_task); |
b50f60ce | 7999 | |
e107be36 AK |
8000 | #ifdef CONFIG_PREEMPT_NOTIFIERS |
8001 | INIT_HLIST_HEAD(&init_task.preempt_notifiers); | |
8002 | #endif | |
8003 | ||
c9819f45 | 8004 | #ifdef CONFIG_SMP |
962cf36c | 8005 | open_softirq(SCHED_SOFTIRQ, run_rebalance_domains); |
c9819f45 CL |
8006 | #endif |
8007 | ||
b50f60ce | 8008 | #ifdef CONFIG_RT_MUTEXES |
1d615482 | 8009 | plist_head_init_raw(&init_task.pi_waiters, &init_task.pi_lock); |
b50f60ce HC |
8010 | #endif |
8011 | ||
1da177e4 LT |
8012 | /* |
8013 | * The boot idle thread does lazy MMU switching as well: | |
8014 | */ | |
8015 | atomic_inc(&init_mm.mm_count); | |
8016 | enter_lazy_tlb(&init_mm, current); | |
8017 | ||
8018 | /* | |
8019 | * Make us the idle thread. Technically, schedule() should not be | |
8020 | * called from this thread, however somewhere below it might be, | |
8021 | * but because we are the idle thread, we just pick up running again | |
8022 | * when this runqueue becomes "idle". | |
8023 | */ | |
8024 | init_idle(current, smp_processor_id()); | |
dce48a84 TG |
8025 | |
8026 | calc_load_update = jiffies + LOAD_FREQ; | |
8027 | ||
dd41f596 IM |
8028 | /* |
8029 | * During early bootup we pretend to be a normal task: | |
8030 | */ | |
8031 | current->sched_class = &fair_sched_class; | |
6892b75e | 8032 | |
6a7b3dc3 | 8033 | /* Allocate the nohz_cpu_mask if CONFIG_CPUMASK_OFFSTACK */ |
49557e62 | 8034 | zalloc_cpumask_var(&nohz_cpu_mask, GFP_NOWAIT); |
bf4d83f6 | 8035 | #ifdef CONFIG_SMP |
7d1e6a9b | 8036 | #ifdef CONFIG_NO_HZ |
83cd4fe2 VP |
8037 | zalloc_cpumask_var(&nohz.idle_cpus_mask, GFP_NOWAIT); |
8038 | alloc_cpumask_var(&nohz.grp_idle_mask, GFP_NOWAIT); | |
8039 | atomic_set(&nohz.load_balancer, nr_cpu_ids); | |
8040 | atomic_set(&nohz.first_pick_cpu, nr_cpu_ids); | |
8041 | atomic_set(&nohz.second_pick_cpu, nr_cpu_ids); | |
7d1e6a9b | 8042 | #endif |
bdddd296 RR |
8043 | /* May be allocated at isolcpus cmdline parse time */ |
8044 | if (cpu_isolated_map == NULL) | |
8045 | zalloc_cpumask_var(&cpu_isolated_map, GFP_NOWAIT); | |
bf4d83f6 | 8046 | #endif /* SMP */ |
6a7b3dc3 | 8047 | |
6892b75e | 8048 | scheduler_running = 1; |
1da177e4 LT |
8049 | } |
8050 | ||
8051 | #ifdef CONFIG_DEBUG_SPINLOCK_SLEEP | |
e4aafea2 FW |
8052 | static inline int preempt_count_equals(int preempt_offset) |
8053 | { | |
234da7bc | 8054 | int nested = (preempt_count() & ~PREEMPT_ACTIVE) + rcu_preempt_depth(); |
e4aafea2 FW |
8055 | |
8056 | return (nested == PREEMPT_INATOMIC_BASE + preempt_offset); | |
8057 | } | |
8058 | ||
d894837f | 8059 | void __might_sleep(const char *file, int line, int preempt_offset) |
1da177e4 | 8060 | { |
48f24c4d | 8061 | #ifdef in_atomic |
1da177e4 LT |
8062 | static unsigned long prev_jiffy; /* ratelimiting */ |
8063 | ||
e4aafea2 FW |
8064 | if ((preempt_count_equals(preempt_offset) && !irqs_disabled()) || |
8065 | system_state != SYSTEM_RUNNING || oops_in_progress) | |
aef745fc IM |
8066 | return; |
8067 | if (time_before(jiffies, prev_jiffy + HZ) && prev_jiffy) | |
8068 | return; | |
8069 | prev_jiffy = jiffies; | |
8070 | ||
3df0fc5b PZ |
8071 | printk(KERN_ERR |
8072 | "BUG: sleeping function called from invalid context at %s:%d\n", | |
8073 | file, line); | |
8074 | printk(KERN_ERR | |
8075 | "in_atomic(): %d, irqs_disabled(): %d, pid: %d, name: %s\n", | |
8076 | in_atomic(), irqs_disabled(), | |
8077 | current->pid, current->comm); | |
aef745fc IM |
8078 | |
8079 | debug_show_held_locks(current); | |
8080 | if (irqs_disabled()) | |
8081 | print_irqtrace_events(current); | |
8082 | dump_stack(); | |
1da177e4 LT |
8083 | #endif |
8084 | } | |
8085 | EXPORT_SYMBOL(__might_sleep); | |
8086 | #endif | |
8087 | ||
8088 | #ifdef CONFIG_MAGIC_SYSRQ | |
3a5e4dc1 AK |
8089 | static void normalize_task(struct rq *rq, struct task_struct *p) |
8090 | { | |
8091 | int on_rq; | |
3e51f33f | 8092 | |
3a5e4dc1 AK |
8093 | on_rq = p->se.on_rq; |
8094 | if (on_rq) | |
8095 | deactivate_task(rq, p, 0); | |
8096 | __setscheduler(rq, p, SCHED_NORMAL, 0); | |
8097 | if (on_rq) { | |
8098 | activate_task(rq, p, 0); | |
8099 | resched_task(rq->curr); | |
8100 | } | |
8101 | } | |
8102 | ||
1da177e4 LT |
8103 | void normalize_rt_tasks(void) |
8104 | { | |
a0f98a1c | 8105 | struct task_struct *g, *p; |
1da177e4 | 8106 | unsigned long flags; |
70b97a7f | 8107 | struct rq *rq; |
1da177e4 | 8108 | |
4cf5d77a | 8109 | read_lock_irqsave(&tasklist_lock, flags); |
a0f98a1c | 8110 | do_each_thread(g, p) { |
178be793 IM |
8111 | /* |
8112 | * Only normalize user tasks: | |
8113 | */ | |
8114 | if (!p->mm) | |
8115 | continue; | |
8116 | ||
6cfb0d5d | 8117 | p->se.exec_start = 0; |
6cfb0d5d | 8118 | #ifdef CONFIG_SCHEDSTATS |
41acab88 LDM |
8119 | p->se.statistics.wait_start = 0; |
8120 | p->se.statistics.sleep_start = 0; | |
8121 | p->se.statistics.block_start = 0; | |
6cfb0d5d | 8122 | #endif |
dd41f596 IM |
8123 | |
8124 | if (!rt_task(p)) { | |
8125 | /* | |
8126 | * Renice negative nice level userspace | |
8127 | * tasks back to 0: | |
8128 | */ | |
8129 | if (TASK_NICE(p) < 0 && p->mm) | |
8130 | set_user_nice(p, 0); | |
1da177e4 | 8131 | continue; |
dd41f596 | 8132 | } |
1da177e4 | 8133 | |
1d615482 | 8134 | raw_spin_lock(&p->pi_lock); |
b29739f9 | 8135 | rq = __task_rq_lock(p); |
1da177e4 | 8136 | |
178be793 | 8137 | normalize_task(rq, p); |
3a5e4dc1 | 8138 | |
b29739f9 | 8139 | __task_rq_unlock(rq); |
1d615482 | 8140 | raw_spin_unlock(&p->pi_lock); |
a0f98a1c IM |
8141 | } while_each_thread(g, p); |
8142 | ||
4cf5d77a | 8143 | read_unlock_irqrestore(&tasklist_lock, flags); |
1da177e4 LT |
8144 | } |
8145 | ||
8146 | #endif /* CONFIG_MAGIC_SYSRQ */ | |
1df5c10a | 8147 | |
67fc4e0c | 8148 | #if defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) |
1df5c10a | 8149 | /* |
67fc4e0c | 8150 | * These functions are only useful for the IA64 MCA handling, or kdb. |
1df5c10a LT |
8151 | * |
8152 | * They can only be called when the whole system has been | |
8153 | * stopped - every CPU needs to be quiescent, and no scheduling | |
8154 | * activity can take place. Using them for anything else would | |
8155 | * be a serious bug, and as a result, they aren't even visible | |
8156 | * under any other configuration. | |
8157 | */ | |
8158 | ||
8159 | /** | |
8160 | * curr_task - return the current task for a given cpu. | |
8161 | * @cpu: the processor in question. | |
8162 | * | |
8163 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8164 | */ | |
36c8b586 | 8165 | struct task_struct *curr_task(int cpu) |
1df5c10a LT |
8166 | { |
8167 | return cpu_curr(cpu); | |
8168 | } | |
8169 | ||
67fc4e0c JW |
8170 | #endif /* defined(CONFIG_IA64) || defined(CONFIG_KGDB_KDB) */ |
8171 | ||
8172 | #ifdef CONFIG_IA64 | |
1df5c10a LT |
8173 | /** |
8174 | * set_curr_task - set the current task for a given cpu. | |
8175 | * @cpu: the processor in question. | |
8176 | * @p: the task pointer to set. | |
8177 | * | |
8178 | * Description: This function must only be used when non-maskable interrupts | |
41a2d6cf IM |
8179 | * are serviced on a separate stack. It allows the architecture to switch the |
8180 | * notion of the current task on a cpu in a non-blocking manner. This function | |
1df5c10a LT |
8181 | * must be called with all CPU's synchronized, and interrupts disabled, the |
8182 | * and caller must save the original value of the current task (see | |
8183 | * curr_task() above) and restore that value before reenabling interrupts and | |
8184 | * re-starting the system. | |
8185 | * | |
8186 | * ONLY VALID WHEN THE WHOLE SYSTEM IS STOPPED! | |
8187 | */ | |
36c8b586 | 8188 | void set_curr_task(int cpu, struct task_struct *p) |
1df5c10a LT |
8189 | { |
8190 | cpu_curr(cpu) = p; | |
8191 | } | |
8192 | ||
8193 | #endif | |
29f59db3 | 8194 | |
bccbe08a PZ |
8195 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8196 | static void free_fair_sched_group(struct task_group *tg) | |
6f505b16 PZ |
8197 | { |
8198 | int i; | |
8199 | ||
8200 | for_each_possible_cpu(i) { | |
8201 | if (tg->cfs_rq) | |
8202 | kfree(tg->cfs_rq[i]); | |
8203 | if (tg->se) | |
8204 | kfree(tg->se[i]); | |
6f505b16 PZ |
8205 | } |
8206 | ||
8207 | kfree(tg->cfs_rq); | |
8208 | kfree(tg->se); | |
6f505b16 PZ |
8209 | } |
8210 | ||
ec7dc8ac DG |
8211 | static |
8212 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
29f59db3 | 8213 | { |
29f59db3 | 8214 | struct cfs_rq *cfs_rq; |
eab17229 | 8215 | struct sched_entity *se; |
9b5b7751 | 8216 | struct rq *rq; |
29f59db3 SV |
8217 | int i; |
8218 | ||
434d53b0 | 8219 | tg->cfs_rq = kzalloc(sizeof(cfs_rq) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8220 | if (!tg->cfs_rq) |
8221 | goto err; | |
434d53b0 | 8222 | tg->se = kzalloc(sizeof(se) * nr_cpu_ids, GFP_KERNEL); |
29f59db3 SV |
8223 | if (!tg->se) |
8224 | goto err; | |
052f1dc7 PZ |
8225 | |
8226 | tg->shares = NICE_0_LOAD; | |
29f59db3 SV |
8227 | |
8228 | for_each_possible_cpu(i) { | |
9b5b7751 | 8229 | rq = cpu_rq(i); |
29f59db3 | 8230 | |
eab17229 LZ |
8231 | cfs_rq = kzalloc_node(sizeof(struct cfs_rq), |
8232 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 SV |
8233 | if (!cfs_rq) |
8234 | goto err; | |
8235 | ||
eab17229 LZ |
8236 | se = kzalloc_node(sizeof(struct sched_entity), |
8237 | GFP_KERNEL, cpu_to_node(i)); | |
29f59db3 | 8238 | if (!se) |
dfc12eb2 | 8239 | goto err_free_rq; |
29f59db3 | 8240 | |
3d4b47b4 | 8241 | init_tg_cfs_entry(tg, cfs_rq, se, i, parent->se[i]); |
bccbe08a PZ |
8242 | } |
8243 | ||
8244 | return 1; | |
8245 | ||
49246274 | 8246 | err_free_rq: |
dfc12eb2 | 8247 | kfree(cfs_rq); |
49246274 | 8248 | err: |
bccbe08a PZ |
8249 | return 0; |
8250 | } | |
8251 | ||
bccbe08a PZ |
8252 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8253 | { | |
3d4b47b4 PZ |
8254 | struct rq *rq = cpu_rq(cpu); |
8255 | unsigned long flags; | |
3d4b47b4 PZ |
8256 | |
8257 | /* | |
8258 | * Only empty task groups can be destroyed; so we can speculatively | |
8259 | * check on_list without danger of it being re-added. | |
8260 | */ | |
8261 | if (!tg->cfs_rq[cpu]->on_list) | |
8262 | return; | |
8263 | ||
8264 | raw_spin_lock_irqsave(&rq->lock, flags); | |
822bc180 | 8265 | list_del_leaf_cfs_rq(tg->cfs_rq[cpu]); |
3d4b47b4 | 8266 | raw_spin_unlock_irqrestore(&rq->lock, flags); |
bccbe08a | 8267 | } |
6d6bc0ad | 8268 | #else /* !CONFG_FAIR_GROUP_SCHED */ |
bccbe08a PZ |
8269 | static inline void free_fair_sched_group(struct task_group *tg) |
8270 | { | |
8271 | } | |
8272 | ||
ec7dc8ac DG |
8273 | static inline |
8274 | int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8275 | { |
8276 | return 1; | |
8277 | } | |
8278 | ||
bccbe08a PZ |
8279 | static inline void unregister_fair_sched_group(struct task_group *tg, int cpu) |
8280 | { | |
8281 | } | |
6d6bc0ad | 8282 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
052f1dc7 PZ |
8283 | |
8284 | #ifdef CONFIG_RT_GROUP_SCHED | |
bccbe08a PZ |
8285 | static void free_rt_sched_group(struct task_group *tg) |
8286 | { | |
8287 | int i; | |
8288 | ||
d0b27fa7 PZ |
8289 | destroy_rt_bandwidth(&tg->rt_bandwidth); |
8290 | ||
bccbe08a PZ |
8291 | for_each_possible_cpu(i) { |
8292 | if (tg->rt_rq) | |
8293 | kfree(tg->rt_rq[i]); | |
8294 | if (tg->rt_se) | |
8295 | kfree(tg->rt_se[i]); | |
8296 | } | |
8297 | ||
8298 | kfree(tg->rt_rq); | |
8299 | kfree(tg->rt_se); | |
8300 | } | |
8301 | ||
ec7dc8ac DG |
8302 | static |
8303 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8304 | { |
8305 | struct rt_rq *rt_rq; | |
eab17229 | 8306 | struct sched_rt_entity *rt_se; |
bccbe08a PZ |
8307 | struct rq *rq; |
8308 | int i; | |
8309 | ||
434d53b0 | 8310 | tg->rt_rq = kzalloc(sizeof(rt_rq) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8311 | if (!tg->rt_rq) |
8312 | goto err; | |
434d53b0 | 8313 | tg->rt_se = kzalloc(sizeof(rt_se) * nr_cpu_ids, GFP_KERNEL); |
bccbe08a PZ |
8314 | if (!tg->rt_se) |
8315 | goto err; | |
8316 | ||
d0b27fa7 PZ |
8317 | init_rt_bandwidth(&tg->rt_bandwidth, |
8318 | ktime_to_ns(def_rt_bandwidth.rt_period), 0); | |
bccbe08a PZ |
8319 | |
8320 | for_each_possible_cpu(i) { | |
8321 | rq = cpu_rq(i); | |
8322 | ||
eab17229 LZ |
8323 | rt_rq = kzalloc_node(sizeof(struct rt_rq), |
8324 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 PZ |
8325 | if (!rt_rq) |
8326 | goto err; | |
29f59db3 | 8327 | |
eab17229 LZ |
8328 | rt_se = kzalloc_node(sizeof(struct sched_rt_entity), |
8329 | GFP_KERNEL, cpu_to_node(i)); | |
6f505b16 | 8330 | if (!rt_se) |
dfc12eb2 | 8331 | goto err_free_rq; |
29f59db3 | 8332 | |
3d4b47b4 | 8333 | init_tg_rt_entry(tg, rt_rq, rt_se, i, parent->rt_se[i]); |
29f59db3 SV |
8334 | } |
8335 | ||
bccbe08a PZ |
8336 | return 1; |
8337 | ||
49246274 | 8338 | err_free_rq: |
dfc12eb2 | 8339 | kfree(rt_rq); |
49246274 | 8340 | err: |
bccbe08a PZ |
8341 | return 0; |
8342 | } | |
6d6bc0ad | 8343 | #else /* !CONFIG_RT_GROUP_SCHED */ |
bccbe08a PZ |
8344 | static inline void free_rt_sched_group(struct task_group *tg) |
8345 | { | |
8346 | } | |
8347 | ||
ec7dc8ac DG |
8348 | static inline |
8349 | int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent) | |
bccbe08a PZ |
8350 | { |
8351 | return 1; | |
8352 | } | |
6d6bc0ad | 8353 | #endif /* CONFIG_RT_GROUP_SCHED */ |
bccbe08a | 8354 | |
7c941438 | 8355 | #ifdef CONFIG_CGROUP_SCHED |
bccbe08a PZ |
8356 | static void free_sched_group(struct task_group *tg) |
8357 | { | |
8358 | free_fair_sched_group(tg); | |
8359 | free_rt_sched_group(tg); | |
e9aa1dd1 | 8360 | autogroup_free(tg); |
bccbe08a PZ |
8361 | kfree(tg); |
8362 | } | |
8363 | ||
8364 | /* allocate runqueue etc for a new task group */ | |
ec7dc8ac | 8365 | struct task_group *sched_create_group(struct task_group *parent) |
bccbe08a PZ |
8366 | { |
8367 | struct task_group *tg; | |
8368 | unsigned long flags; | |
bccbe08a PZ |
8369 | |
8370 | tg = kzalloc(sizeof(*tg), GFP_KERNEL); | |
8371 | if (!tg) | |
8372 | return ERR_PTR(-ENOMEM); | |
8373 | ||
ec7dc8ac | 8374 | if (!alloc_fair_sched_group(tg, parent)) |
bccbe08a PZ |
8375 | goto err; |
8376 | ||
ec7dc8ac | 8377 | if (!alloc_rt_sched_group(tg, parent)) |
bccbe08a PZ |
8378 | goto err; |
8379 | ||
8ed36996 | 8380 | spin_lock_irqsave(&task_group_lock, flags); |
6f505b16 | 8381 | list_add_rcu(&tg->list, &task_groups); |
f473aa5e PZ |
8382 | |
8383 | WARN_ON(!parent); /* root should already exist */ | |
8384 | ||
8385 | tg->parent = parent; | |
f473aa5e | 8386 | INIT_LIST_HEAD(&tg->children); |
09f2724a | 8387 | list_add_rcu(&tg->siblings, &parent->children); |
8ed36996 | 8388 | spin_unlock_irqrestore(&task_group_lock, flags); |
29f59db3 | 8389 | |
9b5b7751 | 8390 | return tg; |
29f59db3 SV |
8391 | |
8392 | err: | |
6f505b16 | 8393 | free_sched_group(tg); |
29f59db3 SV |
8394 | return ERR_PTR(-ENOMEM); |
8395 | } | |
8396 | ||
9b5b7751 | 8397 | /* rcu callback to free various structures associated with a task group */ |
6f505b16 | 8398 | static void free_sched_group_rcu(struct rcu_head *rhp) |
29f59db3 | 8399 | { |
29f59db3 | 8400 | /* now it should be safe to free those cfs_rqs */ |
6f505b16 | 8401 | free_sched_group(container_of(rhp, struct task_group, rcu)); |
29f59db3 SV |
8402 | } |
8403 | ||
9b5b7751 | 8404 | /* Destroy runqueue etc associated with a task group */ |
4cf86d77 | 8405 | void sched_destroy_group(struct task_group *tg) |
29f59db3 | 8406 | { |
8ed36996 | 8407 | unsigned long flags; |
9b5b7751 | 8408 | int i; |
29f59db3 | 8409 | |
3d4b47b4 PZ |
8410 | /* end participation in shares distribution */ |
8411 | for_each_possible_cpu(i) | |
bccbe08a | 8412 | unregister_fair_sched_group(tg, i); |
3d4b47b4 PZ |
8413 | |
8414 | spin_lock_irqsave(&task_group_lock, flags); | |
6f505b16 | 8415 | list_del_rcu(&tg->list); |
f473aa5e | 8416 | list_del_rcu(&tg->siblings); |
8ed36996 | 8417 | spin_unlock_irqrestore(&task_group_lock, flags); |
9b5b7751 | 8418 | |
9b5b7751 | 8419 | /* wait for possible concurrent references to cfs_rqs complete */ |
6f505b16 | 8420 | call_rcu(&tg->rcu, free_sched_group_rcu); |
29f59db3 SV |
8421 | } |
8422 | ||
9b5b7751 | 8423 | /* change task's runqueue when it moves between groups. |
3a252015 IM |
8424 | * The caller of this function should have put the task in its new group |
8425 | * by now. This function just updates tsk->se.cfs_rq and tsk->se.parent to | |
8426 | * reflect its new group. | |
9b5b7751 SV |
8427 | */ |
8428 | void sched_move_task(struct task_struct *tsk) | |
29f59db3 SV |
8429 | { |
8430 | int on_rq, running; | |
8431 | unsigned long flags; | |
8432 | struct rq *rq; | |
8433 | ||
8434 | rq = task_rq_lock(tsk, &flags); | |
8435 | ||
051a1d1a | 8436 | running = task_current(rq, tsk); |
29f59db3 SV |
8437 | on_rq = tsk->se.on_rq; |
8438 | ||
0e1f3483 | 8439 | if (on_rq) |
29f59db3 | 8440 | dequeue_task(rq, tsk, 0); |
0e1f3483 HS |
8441 | if (unlikely(running)) |
8442 | tsk->sched_class->put_prev_task(rq, tsk); | |
29f59db3 | 8443 | |
810b3817 | 8444 | #ifdef CONFIG_FAIR_GROUP_SCHED |
b2b5ce02 PZ |
8445 | if (tsk->sched_class->task_move_group) |
8446 | tsk->sched_class->task_move_group(tsk, on_rq); | |
8447 | else | |
810b3817 | 8448 | #endif |
b2b5ce02 | 8449 | set_task_rq(tsk, task_cpu(tsk)); |
810b3817 | 8450 | |
0e1f3483 HS |
8451 | if (unlikely(running)) |
8452 | tsk->sched_class->set_curr_task(rq); | |
8453 | if (on_rq) | |
371fd7e7 | 8454 | enqueue_task(rq, tsk, 0); |
29f59db3 | 8455 | |
29f59db3 SV |
8456 | task_rq_unlock(rq, &flags); |
8457 | } | |
7c941438 | 8458 | #endif /* CONFIG_CGROUP_SCHED */ |
29f59db3 | 8459 | |
052f1dc7 | 8460 | #ifdef CONFIG_FAIR_GROUP_SCHED |
8ed36996 PZ |
8461 | static DEFINE_MUTEX(shares_mutex); |
8462 | ||
4cf86d77 | 8463 | int sched_group_set_shares(struct task_group *tg, unsigned long shares) |
29f59db3 SV |
8464 | { |
8465 | int i; | |
8ed36996 | 8466 | unsigned long flags; |
c61935fd | 8467 | |
ec7dc8ac DG |
8468 | /* |
8469 | * We can't change the weight of the root cgroup. | |
8470 | */ | |
8471 | if (!tg->se[0]) | |
8472 | return -EINVAL; | |
8473 | ||
18d95a28 PZ |
8474 | if (shares < MIN_SHARES) |
8475 | shares = MIN_SHARES; | |
cb4ad1ff MX |
8476 | else if (shares > MAX_SHARES) |
8477 | shares = MAX_SHARES; | |
62fb1851 | 8478 | |
8ed36996 | 8479 | mutex_lock(&shares_mutex); |
9b5b7751 | 8480 | if (tg->shares == shares) |
5cb350ba | 8481 | goto done; |
29f59db3 | 8482 | |
9b5b7751 | 8483 | tg->shares = shares; |
c09595f6 | 8484 | for_each_possible_cpu(i) { |
9437178f PT |
8485 | struct rq *rq = cpu_rq(i); |
8486 | struct sched_entity *se; | |
8487 | ||
8488 | se = tg->se[i]; | |
8489 | /* Propagate contribution to hierarchy */ | |
8490 | raw_spin_lock_irqsave(&rq->lock, flags); | |
8491 | for_each_sched_entity(se) | |
8492 | update_cfs_shares(group_cfs_rq(se), 0); | |
8493 | raw_spin_unlock_irqrestore(&rq->lock, flags); | |
c09595f6 | 8494 | } |
29f59db3 | 8495 | |
5cb350ba | 8496 | done: |
8ed36996 | 8497 | mutex_unlock(&shares_mutex); |
9b5b7751 | 8498 | return 0; |
29f59db3 SV |
8499 | } |
8500 | ||
5cb350ba DG |
8501 | unsigned long sched_group_shares(struct task_group *tg) |
8502 | { | |
8503 | return tg->shares; | |
8504 | } | |
052f1dc7 | 8505 | #endif |
5cb350ba | 8506 | |
052f1dc7 | 8507 | #ifdef CONFIG_RT_GROUP_SCHED |
6f505b16 | 8508 | /* |
9f0c1e56 | 8509 | * Ensure that the real time constraints are schedulable. |
6f505b16 | 8510 | */ |
9f0c1e56 PZ |
8511 | static DEFINE_MUTEX(rt_constraints_mutex); |
8512 | ||
8513 | static unsigned long to_ratio(u64 period, u64 runtime) | |
8514 | { | |
8515 | if (runtime == RUNTIME_INF) | |
9a7e0b18 | 8516 | return 1ULL << 20; |
9f0c1e56 | 8517 | |
9a7e0b18 | 8518 | return div64_u64(runtime << 20, period); |
9f0c1e56 PZ |
8519 | } |
8520 | ||
9a7e0b18 PZ |
8521 | /* Must be called with tasklist_lock held */ |
8522 | static inline int tg_has_rt_tasks(struct task_group *tg) | |
b40b2e8e | 8523 | { |
9a7e0b18 | 8524 | struct task_struct *g, *p; |
b40b2e8e | 8525 | |
9a7e0b18 PZ |
8526 | do_each_thread(g, p) { |
8527 | if (rt_task(p) && rt_rq_of_se(&p->rt)->tg == tg) | |
8528 | return 1; | |
8529 | } while_each_thread(g, p); | |
b40b2e8e | 8530 | |
9a7e0b18 PZ |
8531 | return 0; |
8532 | } | |
b40b2e8e | 8533 | |
9a7e0b18 PZ |
8534 | struct rt_schedulable_data { |
8535 | struct task_group *tg; | |
8536 | u64 rt_period; | |
8537 | u64 rt_runtime; | |
8538 | }; | |
b40b2e8e | 8539 | |
9a7e0b18 PZ |
8540 | static int tg_schedulable(struct task_group *tg, void *data) |
8541 | { | |
8542 | struct rt_schedulable_data *d = data; | |
8543 | struct task_group *child; | |
8544 | unsigned long total, sum = 0; | |
8545 | u64 period, runtime; | |
b40b2e8e | 8546 | |
9a7e0b18 PZ |
8547 | period = ktime_to_ns(tg->rt_bandwidth.rt_period); |
8548 | runtime = tg->rt_bandwidth.rt_runtime; | |
b40b2e8e | 8549 | |
9a7e0b18 PZ |
8550 | if (tg == d->tg) { |
8551 | period = d->rt_period; | |
8552 | runtime = d->rt_runtime; | |
b40b2e8e | 8553 | } |
b40b2e8e | 8554 | |
4653f803 PZ |
8555 | /* |
8556 | * Cannot have more runtime than the period. | |
8557 | */ | |
8558 | if (runtime > period && runtime != RUNTIME_INF) | |
8559 | return -EINVAL; | |
6f505b16 | 8560 | |
4653f803 PZ |
8561 | /* |
8562 | * Ensure we don't starve existing RT tasks. | |
8563 | */ | |
9a7e0b18 PZ |
8564 | if (rt_bandwidth_enabled() && !runtime && tg_has_rt_tasks(tg)) |
8565 | return -EBUSY; | |
6f505b16 | 8566 | |
9a7e0b18 | 8567 | total = to_ratio(period, runtime); |
6f505b16 | 8568 | |
4653f803 PZ |
8569 | /* |
8570 | * Nobody can have more than the global setting allows. | |
8571 | */ | |
8572 | if (total > to_ratio(global_rt_period(), global_rt_runtime())) | |
8573 | return -EINVAL; | |
6f505b16 | 8574 | |
4653f803 PZ |
8575 | /* |
8576 | * The sum of our children's runtime should not exceed our own. | |
8577 | */ | |
9a7e0b18 PZ |
8578 | list_for_each_entry_rcu(child, &tg->children, siblings) { |
8579 | period = ktime_to_ns(child->rt_bandwidth.rt_period); | |
8580 | runtime = child->rt_bandwidth.rt_runtime; | |
6f505b16 | 8581 | |
9a7e0b18 PZ |
8582 | if (child == d->tg) { |
8583 | period = d->rt_period; | |
8584 | runtime = d->rt_runtime; | |
8585 | } | |
6f505b16 | 8586 | |
9a7e0b18 | 8587 | sum += to_ratio(period, runtime); |
9f0c1e56 | 8588 | } |
6f505b16 | 8589 | |
9a7e0b18 PZ |
8590 | if (sum > total) |
8591 | return -EINVAL; | |
8592 | ||
8593 | return 0; | |
6f505b16 PZ |
8594 | } |
8595 | ||
9a7e0b18 | 8596 | static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime) |
521f1a24 | 8597 | { |
9a7e0b18 PZ |
8598 | struct rt_schedulable_data data = { |
8599 | .tg = tg, | |
8600 | .rt_period = period, | |
8601 | .rt_runtime = runtime, | |
8602 | }; | |
8603 | ||
8604 | return walk_tg_tree(tg_schedulable, tg_nop, &data); | |
521f1a24 DG |
8605 | } |
8606 | ||
d0b27fa7 PZ |
8607 | static int tg_set_bandwidth(struct task_group *tg, |
8608 | u64 rt_period, u64 rt_runtime) | |
6f505b16 | 8609 | { |
ac086bc2 | 8610 | int i, err = 0; |
9f0c1e56 | 8611 | |
9f0c1e56 | 8612 | mutex_lock(&rt_constraints_mutex); |
521f1a24 | 8613 | read_lock(&tasklist_lock); |
9a7e0b18 PZ |
8614 | err = __rt_schedulable(tg, rt_period, rt_runtime); |
8615 | if (err) | |
9f0c1e56 | 8616 | goto unlock; |
ac086bc2 | 8617 | |
0986b11b | 8618 | raw_spin_lock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
d0b27fa7 PZ |
8619 | tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period); |
8620 | tg->rt_bandwidth.rt_runtime = rt_runtime; | |
ac086bc2 PZ |
8621 | |
8622 | for_each_possible_cpu(i) { | |
8623 | struct rt_rq *rt_rq = tg->rt_rq[i]; | |
8624 | ||
0986b11b | 8625 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8626 | rt_rq->rt_runtime = rt_runtime; |
0986b11b | 8627 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8628 | } |
0986b11b | 8629 | raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); |
49246274 | 8630 | unlock: |
521f1a24 | 8631 | read_unlock(&tasklist_lock); |
9f0c1e56 PZ |
8632 | mutex_unlock(&rt_constraints_mutex); |
8633 | ||
8634 | return err; | |
6f505b16 PZ |
8635 | } |
8636 | ||
d0b27fa7 PZ |
8637 | int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us) |
8638 | { | |
8639 | u64 rt_runtime, rt_period; | |
8640 | ||
8641 | rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8642 | rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC; | |
8643 | if (rt_runtime_us < 0) | |
8644 | rt_runtime = RUNTIME_INF; | |
8645 | ||
8646 | return tg_set_bandwidth(tg, rt_period, rt_runtime); | |
8647 | } | |
8648 | ||
9f0c1e56 PZ |
8649 | long sched_group_rt_runtime(struct task_group *tg) |
8650 | { | |
8651 | u64 rt_runtime_us; | |
8652 | ||
d0b27fa7 | 8653 | if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF) |
9f0c1e56 PZ |
8654 | return -1; |
8655 | ||
d0b27fa7 | 8656 | rt_runtime_us = tg->rt_bandwidth.rt_runtime; |
9f0c1e56 PZ |
8657 | do_div(rt_runtime_us, NSEC_PER_USEC); |
8658 | return rt_runtime_us; | |
8659 | } | |
d0b27fa7 PZ |
8660 | |
8661 | int sched_group_set_rt_period(struct task_group *tg, long rt_period_us) | |
8662 | { | |
8663 | u64 rt_runtime, rt_period; | |
8664 | ||
8665 | rt_period = (u64)rt_period_us * NSEC_PER_USEC; | |
8666 | rt_runtime = tg->rt_bandwidth.rt_runtime; | |
8667 | ||
619b0488 R |
8668 | if (rt_period == 0) |
8669 | return -EINVAL; | |
8670 | ||
d0b27fa7 PZ |
8671 | return tg_set_bandwidth(tg, rt_period, rt_runtime); |
8672 | } | |
8673 | ||
8674 | long sched_group_rt_period(struct task_group *tg) | |
8675 | { | |
8676 | u64 rt_period_us; | |
8677 | ||
8678 | rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period); | |
8679 | do_div(rt_period_us, NSEC_PER_USEC); | |
8680 | return rt_period_us; | |
8681 | } | |
8682 | ||
8683 | static int sched_rt_global_constraints(void) | |
8684 | { | |
4653f803 | 8685 | u64 runtime, period; |
d0b27fa7 PZ |
8686 | int ret = 0; |
8687 | ||
ec5d4989 HS |
8688 | if (sysctl_sched_rt_period <= 0) |
8689 | return -EINVAL; | |
8690 | ||
4653f803 PZ |
8691 | runtime = global_rt_runtime(); |
8692 | period = global_rt_period(); | |
8693 | ||
8694 | /* | |
8695 | * Sanity check on the sysctl variables. | |
8696 | */ | |
8697 | if (runtime > period && runtime != RUNTIME_INF) | |
8698 | return -EINVAL; | |
10b612f4 | 8699 | |
d0b27fa7 | 8700 | mutex_lock(&rt_constraints_mutex); |
9a7e0b18 | 8701 | read_lock(&tasklist_lock); |
4653f803 | 8702 | ret = __rt_schedulable(NULL, 0, 0); |
9a7e0b18 | 8703 | read_unlock(&tasklist_lock); |
d0b27fa7 PZ |
8704 | mutex_unlock(&rt_constraints_mutex); |
8705 | ||
8706 | return ret; | |
8707 | } | |
54e99124 DG |
8708 | |
8709 | int sched_rt_can_attach(struct task_group *tg, struct task_struct *tsk) | |
8710 | { | |
8711 | /* Don't accept realtime tasks when there is no way for them to run */ | |
8712 | if (rt_task(tsk) && tg->rt_bandwidth.rt_runtime == 0) | |
8713 | return 0; | |
8714 | ||
8715 | return 1; | |
8716 | } | |
8717 | ||
6d6bc0ad | 8718 | #else /* !CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8719 | static int sched_rt_global_constraints(void) |
8720 | { | |
ac086bc2 PZ |
8721 | unsigned long flags; |
8722 | int i; | |
8723 | ||
ec5d4989 HS |
8724 | if (sysctl_sched_rt_period <= 0) |
8725 | return -EINVAL; | |
8726 | ||
60aa605d PZ |
8727 | /* |
8728 | * There's always some RT tasks in the root group | |
8729 | * -- migration, kstopmachine etc.. | |
8730 | */ | |
8731 | if (sysctl_sched_rt_runtime == 0) | |
8732 | return -EBUSY; | |
8733 | ||
0986b11b | 8734 | raw_spin_lock_irqsave(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 PZ |
8735 | for_each_possible_cpu(i) { |
8736 | struct rt_rq *rt_rq = &cpu_rq(i)->rt; | |
8737 | ||
0986b11b | 8738 | raw_spin_lock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8739 | rt_rq->rt_runtime = global_rt_runtime(); |
0986b11b | 8740 | raw_spin_unlock(&rt_rq->rt_runtime_lock); |
ac086bc2 | 8741 | } |
0986b11b | 8742 | raw_spin_unlock_irqrestore(&def_rt_bandwidth.rt_runtime_lock, flags); |
ac086bc2 | 8743 | |
d0b27fa7 PZ |
8744 | return 0; |
8745 | } | |
6d6bc0ad | 8746 | #endif /* CONFIG_RT_GROUP_SCHED */ |
d0b27fa7 PZ |
8747 | |
8748 | int sched_rt_handler(struct ctl_table *table, int write, | |
8d65af78 | 8749 | void __user *buffer, size_t *lenp, |
d0b27fa7 PZ |
8750 | loff_t *ppos) |
8751 | { | |
8752 | int ret; | |
8753 | int old_period, old_runtime; | |
8754 | static DEFINE_MUTEX(mutex); | |
8755 | ||
8756 | mutex_lock(&mutex); | |
8757 | old_period = sysctl_sched_rt_period; | |
8758 | old_runtime = sysctl_sched_rt_runtime; | |
8759 | ||
8d65af78 | 8760 | ret = proc_dointvec(table, write, buffer, lenp, ppos); |
d0b27fa7 PZ |
8761 | |
8762 | if (!ret && write) { | |
8763 | ret = sched_rt_global_constraints(); | |
8764 | if (ret) { | |
8765 | sysctl_sched_rt_period = old_period; | |
8766 | sysctl_sched_rt_runtime = old_runtime; | |
8767 | } else { | |
8768 | def_rt_bandwidth.rt_runtime = global_rt_runtime(); | |
8769 | def_rt_bandwidth.rt_period = | |
8770 | ns_to_ktime(global_rt_period()); | |
8771 | } | |
8772 | } | |
8773 | mutex_unlock(&mutex); | |
8774 | ||
8775 | return ret; | |
8776 | } | |
68318b8e | 8777 | |
052f1dc7 | 8778 | #ifdef CONFIG_CGROUP_SCHED |
68318b8e SV |
8779 | |
8780 | /* return corresponding task_group object of a cgroup */ | |
2b01dfe3 | 8781 | static inline struct task_group *cgroup_tg(struct cgroup *cgrp) |
68318b8e | 8782 | { |
2b01dfe3 PM |
8783 | return container_of(cgroup_subsys_state(cgrp, cpu_cgroup_subsys_id), |
8784 | struct task_group, css); | |
68318b8e SV |
8785 | } |
8786 | ||
8787 | static struct cgroup_subsys_state * | |
2b01dfe3 | 8788 | cpu_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cgrp) |
68318b8e | 8789 | { |
ec7dc8ac | 8790 | struct task_group *tg, *parent; |
68318b8e | 8791 | |
2b01dfe3 | 8792 | if (!cgrp->parent) { |
68318b8e | 8793 | /* This is early initialization for the top cgroup */ |
07e06b01 | 8794 | return &root_task_group.css; |
68318b8e SV |
8795 | } |
8796 | ||
ec7dc8ac DG |
8797 | parent = cgroup_tg(cgrp->parent); |
8798 | tg = sched_create_group(parent); | |
68318b8e SV |
8799 | if (IS_ERR(tg)) |
8800 | return ERR_PTR(-ENOMEM); | |
8801 | ||
68318b8e SV |
8802 | return &tg->css; |
8803 | } | |
8804 | ||
41a2d6cf IM |
8805 | static void |
8806 | cpu_cgroup_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) | |
68318b8e | 8807 | { |
2b01dfe3 | 8808 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8809 | |
8810 | sched_destroy_group(tg); | |
8811 | } | |
8812 | ||
41a2d6cf | 8813 | static int |
be367d09 | 8814 | cpu_cgroup_can_attach_task(struct cgroup *cgrp, struct task_struct *tsk) |
68318b8e | 8815 | { |
b68aa230 | 8816 | #ifdef CONFIG_RT_GROUP_SCHED |
54e99124 | 8817 | if (!sched_rt_can_attach(cgroup_tg(cgrp), tsk)) |
b68aa230 PZ |
8818 | return -EINVAL; |
8819 | #else | |
68318b8e SV |
8820 | /* We don't support RT-tasks being in separate groups */ |
8821 | if (tsk->sched_class != &fair_sched_class) | |
8822 | return -EINVAL; | |
b68aa230 | 8823 | #endif |
be367d09 BB |
8824 | return 0; |
8825 | } | |
68318b8e | 8826 | |
be367d09 BB |
8827 | static int |
8828 | cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, | |
8829 | struct task_struct *tsk, bool threadgroup) | |
8830 | { | |
8831 | int retval = cpu_cgroup_can_attach_task(cgrp, tsk); | |
8832 | if (retval) | |
8833 | return retval; | |
8834 | if (threadgroup) { | |
8835 | struct task_struct *c; | |
8836 | rcu_read_lock(); | |
8837 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8838 | retval = cpu_cgroup_can_attach_task(cgrp, c); | |
8839 | if (retval) { | |
8840 | rcu_read_unlock(); | |
8841 | return retval; | |
8842 | } | |
8843 | } | |
8844 | rcu_read_unlock(); | |
8845 | } | |
68318b8e SV |
8846 | return 0; |
8847 | } | |
8848 | ||
8849 | static void | |
2b01dfe3 | 8850 | cpu_cgroup_attach(struct cgroup_subsys *ss, struct cgroup *cgrp, |
be367d09 BB |
8851 | struct cgroup *old_cont, struct task_struct *tsk, |
8852 | bool threadgroup) | |
68318b8e SV |
8853 | { |
8854 | sched_move_task(tsk); | |
be367d09 BB |
8855 | if (threadgroup) { |
8856 | struct task_struct *c; | |
8857 | rcu_read_lock(); | |
8858 | list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { | |
8859 | sched_move_task(c); | |
8860 | } | |
8861 | rcu_read_unlock(); | |
8862 | } | |
68318b8e SV |
8863 | } |
8864 | ||
068c5cc5 PZ |
8865 | static void |
8866 | cpu_cgroup_exit(struct cgroup_subsys *ss, struct task_struct *task) | |
8867 | { | |
8868 | /* | |
8869 | * cgroup_exit() is called in the copy_process() failure path. | |
8870 | * Ignore this case since the task hasn't ran yet, this avoids | |
8871 | * trying to poke a half freed task state from generic code. | |
8872 | */ | |
8873 | if (!(task->flags & PF_EXITING)) | |
8874 | return; | |
8875 | ||
8876 | sched_move_task(task); | |
8877 | } | |
8878 | ||
052f1dc7 | 8879 | #ifdef CONFIG_FAIR_GROUP_SCHED |
f4c753b7 | 8880 | static int cpu_shares_write_u64(struct cgroup *cgrp, struct cftype *cftype, |
2b01dfe3 | 8881 | u64 shareval) |
68318b8e | 8882 | { |
2b01dfe3 | 8883 | return sched_group_set_shares(cgroup_tg(cgrp), shareval); |
68318b8e SV |
8884 | } |
8885 | ||
f4c753b7 | 8886 | static u64 cpu_shares_read_u64(struct cgroup *cgrp, struct cftype *cft) |
68318b8e | 8887 | { |
2b01dfe3 | 8888 | struct task_group *tg = cgroup_tg(cgrp); |
68318b8e SV |
8889 | |
8890 | return (u64) tg->shares; | |
8891 | } | |
6d6bc0ad | 8892 | #endif /* CONFIG_FAIR_GROUP_SCHED */ |
68318b8e | 8893 | |
052f1dc7 | 8894 | #ifdef CONFIG_RT_GROUP_SCHED |
0c70814c | 8895 | static int cpu_rt_runtime_write(struct cgroup *cgrp, struct cftype *cft, |
06ecb27c | 8896 | s64 val) |
6f505b16 | 8897 | { |
06ecb27c | 8898 | return sched_group_set_rt_runtime(cgroup_tg(cgrp), val); |
6f505b16 PZ |
8899 | } |
8900 | ||
06ecb27c | 8901 | static s64 cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft) |
6f505b16 | 8902 | { |
06ecb27c | 8903 | return sched_group_rt_runtime(cgroup_tg(cgrp)); |
6f505b16 | 8904 | } |
d0b27fa7 PZ |
8905 | |
8906 | static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype, | |
8907 | u64 rt_period_us) | |
8908 | { | |
8909 | return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us); | |
8910 | } | |
8911 | ||
8912 | static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft) | |
8913 | { | |
8914 | return sched_group_rt_period(cgroup_tg(cgrp)); | |
8915 | } | |
6d6bc0ad | 8916 | #endif /* CONFIG_RT_GROUP_SCHED */ |
6f505b16 | 8917 | |
fe5c7cc2 | 8918 | static struct cftype cpu_files[] = { |
052f1dc7 | 8919 | #ifdef CONFIG_FAIR_GROUP_SCHED |
fe5c7cc2 PM |
8920 | { |
8921 | .name = "shares", | |
f4c753b7 PM |
8922 | .read_u64 = cpu_shares_read_u64, |
8923 | .write_u64 = cpu_shares_write_u64, | |
fe5c7cc2 | 8924 | }, |
052f1dc7 PZ |
8925 | #endif |
8926 | #ifdef CONFIG_RT_GROUP_SCHED | |
6f505b16 | 8927 | { |
9f0c1e56 | 8928 | .name = "rt_runtime_us", |
06ecb27c PM |
8929 | .read_s64 = cpu_rt_runtime_read, |
8930 | .write_s64 = cpu_rt_runtime_write, | |
6f505b16 | 8931 | }, |
d0b27fa7 PZ |
8932 | { |
8933 | .name = "rt_period_us", | |
f4c753b7 PM |
8934 | .read_u64 = cpu_rt_period_read_uint, |
8935 | .write_u64 = cpu_rt_period_write_uint, | |
d0b27fa7 | 8936 | }, |
052f1dc7 | 8937 | #endif |
68318b8e SV |
8938 | }; |
8939 | ||
8940 | static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont) | |
8941 | { | |
fe5c7cc2 | 8942 | return cgroup_add_files(cont, ss, cpu_files, ARRAY_SIZE(cpu_files)); |
68318b8e SV |
8943 | } |
8944 | ||
8945 | struct cgroup_subsys cpu_cgroup_subsys = { | |
38605cae IM |
8946 | .name = "cpu", |
8947 | .create = cpu_cgroup_create, | |
8948 | .destroy = cpu_cgroup_destroy, | |
8949 | .can_attach = cpu_cgroup_can_attach, | |
8950 | .attach = cpu_cgroup_attach, | |
068c5cc5 | 8951 | .exit = cpu_cgroup_exit, |
38605cae IM |
8952 | .populate = cpu_cgroup_populate, |
8953 | .subsys_id = cpu_cgroup_subsys_id, | |
68318b8e SV |
8954 | .early_init = 1, |
8955 | }; | |
8956 | ||
052f1dc7 | 8957 | #endif /* CONFIG_CGROUP_SCHED */ |
d842de87 SV |
8958 | |
8959 | #ifdef CONFIG_CGROUP_CPUACCT | |
8960 | ||
8961 | /* | |
8962 | * CPU accounting code for task groups. | |
8963 | * | |
8964 | * Based on the work by Paul Menage (menage@google.com) and Balbir Singh | |
8965 | * (balbir@in.ibm.com). | |
8966 | */ | |
8967 | ||
934352f2 | 8968 | /* track cpu usage of a group of tasks and its child groups */ |
d842de87 SV |
8969 | struct cpuacct { |
8970 | struct cgroup_subsys_state css; | |
8971 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43cf38eb | 8972 | u64 __percpu *cpuusage; |
ef12fefa | 8973 | struct percpu_counter cpustat[CPUACCT_STAT_NSTATS]; |
934352f2 | 8974 | struct cpuacct *parent; |
d842de87 SV |
8975 | }; |
8976 | ||
8977 | struct cgroup_subsys cpuacct_subsys; | |
8978 | ||
8979 | /* return cpu accounting group corresponding to this container */ | |
32cd756a | 8980 | static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) |
d842de87 | 8981 | { |
32cd756a | 8982 | return container_of(cgroup_subsys_state(cgrp, cpuacct_subsys_id), |
d842de87 SV |
8983 | struct cpuacct, css); |
8984 | } | |
8985 | ||
8986 | /* return cpu accounting group to which this task belongs */ | |
8987 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
8988 | { | |
8989 | return container_of(task_subsys_state(tsk, cpuacct_subsys_id), | |
8990 | struct cpuacct, css); | |
8991 | } | |
8992 | ||
8993 | /* create a new cpu accounting group */ | |
8994 | static struct cgroup_subsys_state *cpuacct_create( | |
32cd756a | 8995 | struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 SV |
8996 | { |
8997 | struct cpuacct *ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
ef12fefa | 8998 | int i; |
d842de87 SV |
8999 | |
9000 | if (!ca) | |
ef12fefa | 9001 | goto out; |
d842de87 SV |
9002 | |
9003 | ca->cpuusage = alloc_percpu(u64); | |
ef12fefa BR |
9004 | if (!ca->cpuusage) |
9005 | goto out_free_ca; | |
9006 | ||
9007 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) | |
9008 | if (percpu_counter_init(&ca->cpustat[i], 0)) | |
9009 | goto out_free_counters; | |
d842de87 | 9010 | |
934352f2 BR |
9011 | if (cgrp->parent) |
9012 | ca->parent = cgroup_ca(cgrp->parent); | |
9013 | ||
d842de87 | 9014 | return &ca->css; |
ef12fefa BR |
9015 | |
9016 | out_free_counters: | |
9017 | while (--i >= 0) | |
9018 | percpu_counter_destroy(&ca->cpustat[i]); | |
9019 | free_percpu(ca->cpuusage); | |
9020 | out_free_ca: | |
9021 | kfree(ca); | |
9022 | out: | |
9023 | return ERR_PTR(-ENOMEM); | |
d842de87 SV |
9024 | } |
9025 | ||
9026 | /* destroy an existing cpu accounting group */ | |
41a2d6cf | 9027 | static void |
32cd756a | 9028 | cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9029 | { |
32cd756a | 9030 | struct cpuacct *ca = cgroup_ca(cgrp); |
ef12fefa | 9031 | int i; |
d842de87 | 9032 | |
ef12fefa BR |
9033 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) |
9034 | percpu_counter_destroy(&ca->cpustat[i]); | |
d842de87 SV |
9035 | free_percpu(ca->cpuusage); |
9036 | kfree(ca); | |
9037 | } | |
9038 | ||
720f5498 KC |
9039 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) |
9040 | { | |
b36128c8 | 9041 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9042 | u64 data; |
9043 | ||
9044 | #ifndef CONFIG_64BIT | |
9045 | /* | |
9046 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
9047 | */ | |
05fa785c | 9048 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9049 | data = *cpuusage; |
05fa785c | 9050 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9051 | #else |
9052 | data = *cpuusage; | |
9053 | #endif | |
9054 | ||
9055 | return data; | |
9056 | } | |
9057 | ||
9058 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
9059 | { | |
b36128c8 | 9060 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
720f5498 KC |
9061 | |
9062 | #ifndef CONFIG_64BIT | |
9063 | /* | |
9064 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
9065 | */ | |
05fa785c | 9066 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); |
720f5498 | 9067 | *cpuusage = val; |
05fa785c | 9068 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); |
720f5498 KC |
9069 | #else |
9070 | *cpuusage = val; | |
9071 | #endif | |
9072 | } | |
9073 | ||
d842de87 | 9074 | /* return total cpu usage (in nanoseconds) of a group */ |
32cd756a | 9075 | static u64 cpuusage_read(struct cgroup *cgrp, struct cftype *cft) |
d842de87 | 9076 | { |
32cd756a | 9077 | struct cpuacct *ca = cgroup_ca(cgrp); |
d842de87 SV |
9078 | u64 totalcpuusage = 0; |
9079 | int i; | |
9080 | ||
720f5498 KC |
9081 | for_each_present_cpu(i) |
9082 | totalcpuusage += cpuacct_cpuusage_read(ca, i); | |
d842de87 SV |
9083 | |
9084 | return totalcpuusage; | |
9085 | } | |
9086 | ||
0297b803 DG |
9087 | static int cpuusage_write(struct cgroup *cgrp, struct cftype *cftype, |
9088 | u64 reset) | |
9089 | { | |
9090 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9091 | int err = 0; | |
9092 | int i; | |
9093 | ||
9094 | if (reset) { | |
9095 | err = -EINVAL; | |
9096 | goto out; | |
9097 | } | |
9098 | ||
720f5498 KC |
9099 | for_each_present_cpu(i) |
9100 | cpuacct_cpuusage_write(ca, i, 0); | |
0297b803 | 9101 | |
0297b803 DG |
9102 | out: |
9103 | return err; | |
9104 | } | |
9105 | ||
e9515c3c KC |
9106 | static int cpuacct_percpu_seq_read(struct cgroup *cgroup, struct cftype *cft, |
9107 | struct seq_file *m) | |
9108 | { | |
9109 | struct cpuacct *ca = cgroup_ca(cgroup); | |
9110 | u64 percpu; | |
9111 | int i; | |
9112 | ||
9113 | for_each_present_cpu(i) { | |
9114 | percpu = cpuacct_cpuusage_read(ca, i); | |
9115 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
9116 | } | |
9117 | seq_printf(m, "\n"); | |
9118 | return 0; | |
9119 | } | |
9120 | ||
ef12fefa BR |
9121 | static const char *cpuacct_stat_desc[] = { |
9122 | [CPUACCT_STAT_USER] = "user", | |
9123 | [CPUACCT_STAT_SYSTEM] = "system", | |
9124 | }; | |
9125 | ||
9126 | static int cpuacct_stats_show(struct cgroup *cgrp, struct cftype *cft, | |
9127 | struct cgroup_map_cb *cb) | |
9128 | { | |
9129 | struct cpuacct *ca = cgroup_ca(cgrp); | |
9130 | int i; | |
9131 | ||
9132 | for (i = 0; i < CPUACCT_STAT_NSTATS; i++) { | |
9133 | s64 val = percpu_counter_read(&ca->cpustat[i]); | |
9134 | val = cputime64_to_clock_t(val); | |
9135 | cb->fill(cb, cpuacct_stat_desc[i], val); | |
9136 | } | |
9137 | return 0; | |
9138 | } | |
9139 | ||
d842de87 SV |
9140 | static struct cftype files[] = { |
9141 | { | |
9142 | .name = "usage", | |
f4c753b7 PM |
9143 | .read_u64 = cpuusage_read, |
9144 | .write_u64 = cpuusage_write, | |
d842de87 | 9145 | }, |
e9515c3c KC |
9146 | { |
9147 | .name = "usage_percpu", | |
9148 | .read_seq_string = cpuacct_percpu_seq_read, | |
9149 | }, | |
ef12fefa BR |
9150 | { |
9151 | .name = "stat", | |
9152 | .read_map = cpuacct_stats_show, | |
9153 | }, | |
d842de87 SV |
9154 | }; |
9155 | ||
32cd756a | 9156 | static int cpuacct_populate(struct cgroup_subsys *ss, struct cgroup *cgrp) |
d842de87 | 9157 | { |
32cd756a | 9158 | return cgroup_add_files(cgrp, ss, files, ARRAY_SIZE(files)); |
d842de87 SV |
9159 | } |
9160 | ||
9161 | /* | |
9162 | * charge this task's execution time to its accounting group. | |
9163 | * | |
9164 | * called with rq->lock held. | |
9165 | */ | |
9166 | static void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
9167 | { | |
9168 | struct cpuacct *ca; | |
934352f2 | 9169 | int cpu; |
d842de87 | 9170 | |
c40c6f85 | 9171 | if (unlikely(!cpuacct_subsys.active)) |
d842de87 SV |
9172 | return; |
9173 | ||
934352f2 | 9174 | cpu = task_cpu(tsk); |
a18b83b7 BR |
9175 | |
9176 | rcu_read_lock(); | |
9177 | ||
d842de87 | 9178 | ca = task_ca(tsk); |
d842de87 | 9179 | |
934352f2 | 9180 | for (; ca; ca = ca->parent) { |
b36128c8 | 9181 | u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); |
d842de87 SV |
9182 | *cpuusage += cputime; |
9183 | } | |
a18b83b7 BR |
9184 | |
9185 | rcu_read_unlock(); | |
d842de87 SV |
9186 | } |
9187 | ||
fa535a77 AB |
9188 | /* |
9189 | * When CONFIG_VIRT_CPU_ACCOUNTING is enabled one jiffy can be very large | |
9190 | * in cputime_t units. As a result, cpuacct_update_stats calls | |
9191 | * percpu_counter_add with values large enough to always overflow the | |
9192 | * per cpu batch limit causing bad SMP scalability. | |
9193 | * | |
9194 | * To fix this we scale percpu_counter_batch by cputime_one_jiffy so we | |
9195 | * batch the same amount of time with CONFIG_VIRT_CPU_ACCOUNTING disabled | |
9196 | * and enabled. We cap it at INT_MAX which is the largest allowed batch value. | |
9197 | */ | |
9198 | #ifdef CONFIG_SMP | |
9199 | #define CPUACCT_BATCH \ | |
9200 | min_t(long, percpu_counter_batch * cputime_one_jiffy, INT_MAX) | |
9201 | #else | |
9202 | #define CPUACCT_BATCH 0 | |
9203 | #endif | |
9204 | ||
ef12fefa BR |
9205 | /* |
9206 | * Charge the system/user time to the task's accounting group. | |
9207 | */ | |
9208 | static void cpuacct_update_stats(struct task_struct *tsk, | |
9209 | enum cpuacct_stat_index idx, cputime_t val) | |
9210 | { | |
9211 | struct cpuacct *ca; | |
fa535a77 | 9212 | int batch = CPUACCT_BATCH; |
ef12fefa BR |
9213 | |
9214 | if (unlikely(!cpuacct_subsys.active)) | |
9215 | return; | |
9216 | ||
9217 | rcu_read_lock(); | |
9218 | ca = task_ca(tsk); | |
9219 | ||
9220 | do { | |
fa535a77 | 9221 | __percpu_counter_add(&ca->cpustat[idx], val, batch); |
ef12fefa BR |
9222 | ca = ca->parent; |
9223 | } while (ca); | |
9224 | rcu_read_unlock(); | |
9225 | } | |
9226 | ||
d842de87 SV |
9227 | struct cgroup_subsys cpuacct_subsys = { |
9228 | .name = "cpuacct", | |
9229 | .create = cpuacct_create, | |
9230 | .destroy = cpuacct_destroy, | |
9231 | .populate = cpuacct_populate, | |
9232 | .subsys_id = cpuacct_subsys_id, | |
9233 | }; | |
9234 | #endif /* CONFIG_CGROUP_CPUACCT */ | |
03b042bf | 9235 |